jackkuo commited on
Commit
86482b4
·
verified ·
1 Parent(s): 066c2f6

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +50 -0
  2. 0NE2T4oBgHgl3EQfMwbp/content/2301.03729v1.pdf +3 -0
  3. 1dAzT4oBgHgl3EQft_1-/content/tmp_files/2301.01684v1.pdf.txt +1506 -0
  4. 1dAzT4oBgHgl3EQft_1-/content/tmp_files/load_file.txt +0 -0
  5. 2tAyT4oBgHgl3EQfovik/content/2301.00513v1.pdf +3 -0
  6. 2tAyT4oBgHgl3EQfovik/vector_store/index.faiss +3 -0
  7. 3NFKT4oBgHgl3EQf8S4-/content/tmp_files/2301.11948v1.pdf.txt +2362 -0
  8. 3NFKT4oBgHgl3EQf8S4-/content/tmp_files/load_file.txt +0 -0
  9. 4NE4T4oBgHgl3EQfbQzc/content/tmp_files/2301.05072v1.pdf.txt +688 -0
  10. 4NE4T4oBgHgl3EQfbQzc/content/tmp_files/load_file.txt +344 -0
  11. 4dE1T4oBgHgl3EQfmQQK/content/2301.03294v1.pdf +3 -0
  12. 4dE1T4oBgHgl3EQfmQQK/vector_store/index.faiss +3 -0
  13. 4dE1T4oBgHgl3EQfmQQK/vector_store/index.pkl +3 -0
  14. 4tFJT4oBgHgl3EQfkCwt/content/2301.11577v1.pdf +3 -0
  15. 4tFJT4oBgHgl3EQfkCwt/vector_store/index.faiss +3 -0
  16. 4tFJT4oBgHgl3EQfkCwt/vector_store/index.pkl +3 -0
  17. 6dAzT4oBgHgl3EQf-P4i/content/2301.01930v1.pdf +3 -0
  18. 6dAzT4oBgHgl3EQf-P4i/vector_store/index.pkl +3 -0
  19. 79E2T4oBgHgl3EQfPgaW/content/tmp_files/load_file.txt +0 -0
  20. 7NAyT4oBgHgl3EQf2vm0/content/2301.00757v1.pdf +3 -0
  21. 7NAyT4oBgHgl3EQf2vm0/vector_store/index.faiss +3 -0
  22. 7NAyT4oBgHgl3EQf2vm0/vector_store/index.pkl +3 -0
  23. 9dFAT4oBgHgl3EQfpx0O/vector_store/index.pkl +3 -0
  24. 9dFRT4oBgHgl3EQfqjdh/vector_store/index.faiss +3 -0
  25. ANFQT4oBgHgl3EQfMDYd/content/2301.13266v1.pdf +3 -0
  26. ANFQT4oBgHgl3EQfMDYd/vector_store/index.pkl +3 -0
  27. B9FAT4oBgHgl3EQfsh6e/vector_store/index.faiss +3 -0
  28. C9A0T4oBgHgl3EQfAf_H/content/2301.01964v1.pdf +3 -0
  29. C9A0T4oBgHgl3EQfAf_H/vector_store/index.pkl +3 -0
  30. C9AzT4oBgHgl3EQfTvzN/vector_store/index.pkl +3 -0
  31. C9FQT4oBgHgl3EQf_jdA/content/tmp_files/2301.13458v1.pdf.txt +1107 -0
  32. C9FQT4oBgHgl3EQf_jdA/content/tmp_files/load_file.txt +0 -0
  33. C9FRT4oBgHgl3EQfwjj-/vector_store/index.faiss +3 -0
  34. D9AyT4oBgHgl3EQf4vqB/content/2301.00792v1.pdf +3 -0
  35. D9AyT4oBgHgl3EQf4vqB/vector_store/index.faiss +3 -0
  36. D9AyT4oBgHgl3EQf4vqB/vector_store/index.pkl +3 -0
  37. DdE2T4oBgHgl3EQf9gki/content/2301.04228v1.pdf +3 -0
  38. DdE2T4oBgHgl3EQf9gki/vector_store/index.faiss +3 -0
  39. EtE5T4oBgHgl3EQfUw_x/content/tmp_files/2301.05547v1.pdf.txt +2479 -0
  40. EtE5T4oBgHgl3EQfUw_x/content/tmp_files/load_file.txt +0 -0
  41. HNE1T4oBgHgl3EQf_Qa5/vector_store/index.faiss +3 -0
  42. HNE1T4oBgHgl3EQf_Qa5/vector_store/index.pkl +3 -0
  43. ItFAT4oBgHgl3EQfux4k/content/2301.08671v1.pdf +3 -0
  44. J9E3T4oBgHgl3EQfvQsy/content/tmp_files/2301.04691v1.pdf.txt +2333 -0
  45. J9E3T4oBgHgl3EQfvQsy/content/tmp_files/load_file.txt +0 -0
  46. JtE1T4oBgHgl3EQfGQNW/content/tmp_files/2301.02911v1.pdf.txt +821 -0
  47. JtE1T4oBgHgl3EQfGQNW/content/tmp_files/load_file.txt +0 -0
  48. K9FRT4oBgHgl3EQf1Th-/content/tmp_files/2301.13656v1.pdf.txt +0 -0
  49. K9FRT4oBgHgl3EQf1Th-/content/tmp_files/load_file.txt +0 -0
  50. KtFRT4oBgHgl3EQf1Dhl/content/tmp_files/2301.13655v1.pdf.txt +4881 -0
.gitattributes CHANGED
@@ -2369,3 +2369,53 @@ e9E5T4oBgHgl3EQfhA8F/content/2301.05637v1.pdf filter=lfs diff=lfs merge=lfs -tex
2369
  V9AyT4oBgHgl3EQfV_fO/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2370
  1NFIT4oBgHgl3EQf3ivh/content/2301.11382v1.pdf filter=lfs diff=lfs merge=lfs -text
2371
  0NFIT4oBgHgl3EQf2ysh/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2369
  V9AyT4oBgHgl3EQfV_fO/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2370
  1NFIT4oBgHgl3EQf3ivh/content/2301.11382v1.pdf filter=lfs diff=lfs merge=lfs -text
2371
  0NFIT4oBgHgl3EQf2ysh/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2372
+ 7NAyT4oBgHgl3EQf2vm0/content/2301.00757v1.pdf filter=lfs diff=lfs merge=lfs -text
2373
+ 2tAyT4oBgHgl3EQfovik/content/2301.00513v1.pdf filter=lfs diff=lfs merge=lfs -text
2374
+ ItFAT4oBgHgl3EQfux4k/content/2301.08671v1.pdf filter=lfs diff=lfs merge=lfs -text
2375
+ aNE5T4oBgHgl3EQfDg77/content/2301.05407v1.pdf filter=lfs diff=lfs merge=lfs -text
2376
+ XNFLT4oBgHgl3EQfTi_1/content/2301.12046v1.pdf filter=lfs diff=lfs merge=lfs -text
2377
+ dNE3T4oBgHgl3EQfGgn0/content/2301.04316v1.pdf filter=lfs diff=lfs merge=lfs -text
2378
+ l9A0T4oBgHgl3EQfJf8H/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2379
+ b9FKT4oBgHgl3EQfpS4j/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2380
+ RtA0T4oBgHgl3EQfDv-e/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2381
+ UtE5T4oBgHgl3EQfBA5d/content/2301.05383v1.pdf filter=lfs diff=lfs merge=lfs -text
2382
+ DdE2T4oBgHgl3EQf9gki/content/2301.04228v1.pdf filter=lfs diff=lfs merge=lfs -text
2383
+ DdE2T4oBgHgl3EQf9gki/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2384
+ dNE3T4oBgHgl3EQfGgn0/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2385
+ 9dFRT4oBgHgl3EQfqjdh/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2386
+ C9FRT4oBgHgl3EQfwjj-/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2387
+ D9AyT4oBgHgl3EQf4vqB/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2388
+ z9FJT4oBgHgl3EQfjSwc/content/2301.11573v1.pdf filter=lfs diff=lfs merge=lfs -text
2389
+ 7NAyT4oBgHgl3EQf2vm0/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2390
+ xNFRT4oBgHgl3EQfhTfg/content/2301.13583v1.pdf filter=lfs diff=lfs merge=lfs -text
2391
+ VdFOT4oBgHgl3EQf6jSQ/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2392
+ NdFJT4oBgHgl3EQfzy31/content/2301.11645v1.pdf filter=lfs diff=lfs merge=lfs -text
2393
+ b9FKT4oBgHgl3EQfpS4j/content/2301.11869v1.pdf filter=lfs diff=lfs merge=lfs -text
2394
+ UtE5T4oBgHgl3EQfBA5d/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2395
+ 2tAyT4oBgHgl3EQfovik/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2396
+ etFRT4oBgHgl3EQfVTd7/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2397
+ 4dE1T4oBgHgl3EQfmQQK/content/2301.03294v1.pdf filter=lfs diff=lfs merge=lfs -text
2398
+ vdAzT4oBgHgl3EQfdPzu/content/2301.01419v1.pdf filter=lfs diff=lfs merge=lfs -text
2399
+ C9A0T4oBgHgl3EQfAf_H/content/2301.01964v1.pdf filter=lfs diff=lfs merge=lfs -text
2400
+ HNE1T4oBgHgl3EQf_Qa5/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2401
+ xNFRT4oBgHgl3EQfhTfg/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2402
+ D9AyT4oBgHgl3EQf4vqB/content/2301.00792v1.pdf filter=lfs diff=lfs merge=lfs -text
2403
+ vdAzT4oBgHgl3EQfdPzu/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2404
+ etFRT4oBgHgl3EQfVTd7/content/2301.13539v1.pdf filter=lfs diff=lfs merge=lfs -text
2405
+ NdFJT4oBgHgl3EQfzy31/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2406
+ ANFQT4oBgHgl3EQfMDYd/content/2301.13266v1.pdf filter=lfs diff=lfs merge=lfs -text
2407
+ 4tFJT4oBgHgl3EQfkCwt/content/2301.11577v1.pdf filter=lfs diff=lfs merge=lfs -text
2408
+ P9E2T4oBgHgl3EQfVwcE/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2409
+ aNE5T4oBgHgl3EQfDg77/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2410
+ B9FAT4oBgHgl3EQfsh6e/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2411
+ 6dAzT4oBgHgl3EQf-P4i/content/2301.01930v1.pdf filter=lfs diff=lfs merge=lfs -text
2412
+ 4dE1T4oBgHgl3EQfmQQK/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2413
+ 4tFJT4oBgHgl3EQfkCwt/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2414
+ 0NE2T4oBgHgl3EQfMwbp/content/2301.03729v1.pdf filter=lfs diff=lfs merge=lfs -text
2415
+ odE1T4oBgHgl3EQfhwTG/content/2301.03245v1.pdf filter=lfs diff=lfs merge=lfs -text
2416
+ zdE0T4oBgHgl3EQf_ALP/content/2301.02822v1.pdf filter=lfs diff=lfs merge=lfs -text
2417
+ tdE5T4oBgHgl3EQfmg_h/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2418
+ T9FLT4oBgHgl3EQfQi86/content/2301.12033v1.pdf filter=lfs diff=lfs merge=lfs -text
2419
+ lNA0T4oBgHgl3EQfJP8k/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2420
+ odE1T4oBgHgl3EQfhwTG/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
2421
+ VdFOT4oBgHgl3EQf6jSQ/content/2301.12959v1.pdf filter=lfs diff=lfs merge=lfs -text
0NE2T4oBgHgl3EQfMwbp/content/2301.03729v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:79f52ecc02a4165e708d7090861598905099d849998debaa670b39ab20be615c
3
+ size 22312331
1dAzT4oBgHgl3EQft_1-/content/tmp_files/2301.01684v1.pdf.txt ADDED
@@ -0,0 +1,1506 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Astronomy & Astrophysics manuscript no. main
2
+ ©ESO 2023
3
+ January 5, 2023
4
+ Letter to the Editor
5
+ A kinematically-detected planet candidate in a transition disk
6
+ J. Stadler1, 2 , M. Benisty1, 2, A. Izquierdo3, 4, S. Facchini5, R. Teague6, N. Kurtovic7, P. Pinilla8, J. Bae9,
7
+ M. Ansdell10, R. Loomis11, S. Mayama12, L. M. Perez13, L. Testi3
8
+ (Affiliations can be found after the references)
9
+ Received 7 November 2022 / Accepted 2 January 2023
10
+ ABSTRACT
11
+ Context. Transition disks are protoplanetary disks with inner cavities possibly cleared by massive companions. They are prime targets to observe
12
+ at high resolution to map their velocity structure and probe companion-disk interactions.
13
+ Aims. We present Atacama Large (sub-)Millimeter Array (ALMA) Band 6 dust and gas observations of the transition disk around
14
+ RXJ1604.3–2130 A, known to feature nearly symmetric shadows in scattered light, and aim to search for non-Keplerian features.
15
+ Methods. We study the 12CO line channel maps and moment maps of the line of sight velocity and peak intensity. We fit a Keplerian model of the
16
+ channel-by-channel emission to study line profile differences, and produce deprojected radial profiles for all velocity components.
17
+ Results. The 12CO emission is detected out to R =∼1.8′′ (265 au). It shows a cavity inwards of 0.39′′ (∼56 au) and within the dust continuum
18
+ ring (at ∼0.56′′, i.e., 81 au). Azimuthal brightness variations in the 12CO line and dust continuum are broadly aligned with the shadows detected in
19
+ scattered light observations. We find a strong localized non-Keplerian feature towards the west within the continuum ring (at R = 41 ± 10 au and
20
+ PA = 280 ± 2◦). It accounts for ∆vφ/vkep ∼ 0.4, or ∆vz/vkep ∼ 0.04, if the perturbation is in the rotational or vertical direction. A tightly wound
21
+ spiral is also detected and extending over 300◦ in azimuth, possibly connected to the localized non-Keplerian feature. Finally, a bending of the
22
+ iso-velocity contours within the gas cavity indicates a highly perturbed inner region, possibly related to the presence of a misaligned inner disk.
23
+ Conclusions. While broadly aligned with the scattered light shadows, the localized non-Keplerian feature cannot be solely due to changes in
24
+ temperature. Instead, we interpret the kinematical feature as tracing a massive companion located at the edge of the dust continuum ring. We
25
+ speculate that the spiral is caused by buoyancy resonances driven by planet-disk-interactions. However, this potential planet at ∼41 au cannot
26
+ explain the gas-depleted cavity, the low accretion rate and the misaligned inner disk, suggesting the presence of another companion closer-in.
27
+ Key words. planet formation – circumstellar disks
28
+ 1. Introduction
29
+ Planet formation appears to be a robust and efficient process, oc-
30
+ curring both around single and multiple stellar systems (Kostov
31
+ et al. 2016) in protoplanetary disks. The advent of high resolu-
32
+ tion imaging facilities demonstrated that nearly all bright and
33
+ extended disks show substructures, in particular in the small
34
+ (micron-sized) and large (mm-sized) dust tracers seen through
35
+ scattered and thermal light, respectively (e.g., Andrews et al.
36
+ 2018; Long et al. 2018; Rich et al. 2022; Benisty et al. 2022;
37
+ Bae et al. 2022a). Such high resolution studies applied to the
38
+ gas tracers allow to probe overall physical conditions in the disk,
39
+ such as its temperature structure, its surface height (Rich et al.
40
+ 2021; Law et al. 2021), and pressure variations (Teague et al.
41
+ 2018a,b; Rosotti et al. 2020). Studies of the disk density and the
42
+ velocity structure reveal a great complexity, including localized
43
+ non-Keplerian features that can be attributed to embedded mas-
44
+ sive protoplanets (e.g., Pinte et al. 2022; Wölfer et al. 2022).
45
+ Such perturbations from smooth density and velocity distribu-
46
+ tions can directly constrain planet formation, as it is expected
47
+ to leave clear signatures on the disk structure (e.g., Perez et al.
48
+ 2015; Yun et al. 2019). For example, the mapping of spiral wakes
49
+ (Calcino et al. 2022), the detection of so-called ’Doppler flips’
50
+ (change of sign in the non-Keplerian feature; e.g., Casassus &
51
+ Pérez 2019; Norfolk et al. 2022), of meridional flows within
52
+ dust-depleted gaps (Teague et al. 2019a), as well as of a veloc-
53
+ ity perturbation associated with a circumplanetary disk candi-
54
+ date (Bae et al. 2022b) enable to zoom onto the processes of
55
+ planet-disk interaction. While most localized kinematical per-
56
+ turbations are analyzed empirically, statistical methods to quan-
57
+ tify their significance have been developed and led to the de-
58
+ tection of localized signatures possibly associated with unseen
59
+ planets (Izquierdo et al. 2021, 2022). Prime targets to search for
60
+ protoplanets still embedded in their birth environment are the
61
+ so-called transition disks. As in PDS70 (Keppler et al. 2019) or
62
+ AB Aur (Tang et al. 2017), these disks host a dust-depleted cav-
63
+ ity that has possibly been cleared by massive companions (Zhu
64
+ et al. 2011).
65
+ In this Letter, we focus on RXJ1604.3-2130 A (d=144.6 pc,
66
+ 1.24 M⊙, Gaia Collaboration et al. 2022; Manara et al. 2020, re-
67
+ spectively), hereafter J1604, one of the brightest protoplanetary
68
+ disks of the Upper Scorpius Association in the millimeter (mm)
69
+ regime (Barenfeld et al. 2016), that exhibits a prominent cav-
70
+ ity in the dust continuum and CO line emission (Zhang et al.
71
+ 2014; Dong et al. 2017; van der Marel et al. 2021). J1604 has
72
+ a stellar companion located at ∼2300 au, itself a binary with
73
+ separation 13 au (Köhler et al. 2000). The outer disk of J1604
74
+ was resolved with the Atacama Large (sub-)Millimeter Array
75
+ (ALMA) (Mayama et al. 2018) and the Spectro-Polarimetric
76
+ High-contrast Exoplanet REsearch instrument (SPHERE) on the
77
+ Very Large Telescope (VLT) (Pinilla et al. 2015), indicating a
78
+ nearly face-on geometry. Complementary observations are in-
79
+ dicative of a misaligned inner disk with respect to the outer disk.
80
+ Its variable light curve is that of an irregular dipper (Ansdell et al.
81
+ 2020), infrared scattered light observations show the presence of
82
+ two shadows with variable morphology on timescales possibly
83
+ shorter than a day (Pinilla et al. 2018), and ALMA 12CO (J=3–
84
+ 2) line observations show deviations from Keplerian rotation in
85
+ Article number, page 1 of 13
86
+ arXiv:2301.01684v1 [astro-ph.EP] 4 Jan 2023
87
+
88
+ A&A proofs: manuscript no. main
89
+ Fig. 1: ALMA observations of J1604. Panel (a) 231 GHz dust continuum, black solid contours drawn at [5, 15, 25, 35, 45]σ, the image is plotted
90
+ with a power-law scaling of γ = 0.6. (b)12CO peak brightness temperature map computed from I0 using the Planck law with black solid contours
91
+ drawn at [5, 10, 20, 40, 60, 65, 70] σ, pixel below 5σ are masked. (c) Peak intensity residuals after subtracting an azimuthally-averaged radial
92
+ profile from the data, where we adjusted the colour scale such that residuals smaller than 1σ are white. The beam sizes are shown in the lower left
93
+ corner and the position of the star is marked by a green cross. In (b) & (c), we overlaid the continuum contours in white and black, respectively.
94
+ the cavity (Mayama et al. 2018). The position of the scattered
95
+ light shadows are suggestive of a large misalignment (∼70-90◦).
96
+ Measurements of the projected rotational velocity (v sini) indi-
97
+ cate that the star is aligned with the inner disk, thus misaligned
98
+ with the outer disk (Sicilia-Aguilar et al. 2020).
99
+ In this work, we present new ALMA observations of J1604
100
+ and focus on the kinematics of the 12CO (J=2–1) line. In the
101
+ following, Sect. 2 presents the observations, Sect. 3 and 4 our
102
+ methodology and results, respectively. Section 5 provides a dis-
103
+ cussion of the results and Sect. 6, our conclusions.
104
+ 2. Observations, calibration and imaging
105
+ We present new ALMA Band 6 observations (2018.1.01255.S;
106
+ PI: Benisty) with five executions spread over two years obtained
107
+ on 2019 April 4, July 30 and 31st, and 2021 April 29, Septem-
108
+ ber 27. The spectral set-up was designed for continuum detec-
109
+ tion, but includes the 12CO J=2-1 line. The data were combined
110
+ with archival data from program 2015.1.00964.S (PI Oberg; see
111
+ Tab. A.1). The data calibration and imaging were performed
112
+ following the procedure of Andrews et al. (2018), with CASA
113
+ v.5.6.1 (McMullin et al. 2007), and is detailed in Appendix
114
+ A. The synthesized beam of the 12CO line and dust continuum
115
+ images are 0.18′′ x 0.15′′ (102◦) and 0.060′′ x 0.039′′ (- 78◦), re-
116
+ spectively. The rms in a line-free channel was measured to be
117
+ 1.1 mJy beam−1 (4.3 K) for CO and 10 µJy beam−1 for the dust
118
+ continuum. Figure 1 shows the dust continuum map (left), that
119
+ displays a cavity and a bright dust ring peaking at R ∼0.56′′
120
+ (∼81 au), and the 12CO peak brightness temperature map (cen-
121
+ ter) that indicates a smaller cavity in gas, with a peak at R ∼0.39′′
122
+ (∼56 au). A selection of channel maps can be found in Fig. A.1.
123
+ 3. Methodology
124
+ Channel maps model.
125
+ To model the disk line intensity and
126
+ kinematics, we use the discminer package of Izquierdo et al.
127
+ (2021). The code uses parametric prescriptions for the line peak
128
+ intensity, line width, rotational velocity and disk emission height
129
+ to produce channel maps and emcee (Foreman-Mackey et al.
130
+ 2013) to maximise a χ2 log-likelihood function of the difference
131
+ between the model and input intensity for each pixel in a channel
132
+ map. To prescribe the model intensity, we use a generalized bell
133
+ kernel, function of the disk cylindrical coordinates (R, z):
134
+ Im(R, z; vch) = Ip(R, z)
135
+
136
+ 1 +
137
+ �����
138
+ vch − v
139
+ Lw(R, z)
140
+ �����
141
+ 2Ls�−1
142
+ ,
143
+ (1)
144
+ where Ip is the peak intensity, Lw is half the line width at half
145
+ power, hereafter ’the line width’, and Ls the line slope. vch is the
146
+ channel velocity at which the intensity is computed and v the
147
+ observed Keplerian line-of-sight velocity. As the disk is nearly
148
+ face-on, the code is unable to infer an emission height, and we
149
+ therefore assume a flat emission surface. We additionally fix the
150
+ inclination i of the disk to the one inferred from the dust con-
151
+ tinuum (i = 6.0◦; Dong et al. 2017) to break the degeneracy of
152
+ M⋆ · sin i. The fitting procedure and the MCMC search are ex-
153
+ plained in detail in appendix B, where the functional form of
154
+ each model parameter together with its best-fit parameters are
155
+ summarized in Table B.1. We compare selected channel maps to
156
+ best-fit model using these parameters in Fig. A.1.
157
+ Moment
158
+ maps.
159
+ The moment maps are computed with
160
+ bettermoments (Teague & Foreman-Mackey 2018). Since the
161
+ 12CO line emission is optically thick, we fitted the following line
162
+ profile to both data and model channel maps:
163
+ I(v) = I0 · 1 − exp (−τ (v))
164
+ 1 − exp(−τ0)
165
+ with τ = τ0 exp
166
+ �−(v − v0)2
167
+ ∆V2
168
+
169
+ ,
170
+ (2)
171
+ where I0 is the peak intensity of the line and the optical depth
172
+ τ (v) varies like a Gaussian with v0 the line centroid, τ0 the peak
173
+ optical depth and ∆V the width of the line (where the full width
174
+ at half maximum FWHM = 2
175
+
176
+ ln2∆V), as used in Teague et al.
177
+ (2022). In Fig. 1 (b), we show I0 for 12CO in units of brightness
178
+ temperature. The corresponding v0-maps for the data and model
179
+ are displayed in Fig. 2 (a) & (b), respectively. The moment maps
180
+ Article number, page 2 of 13
181
+
182
+ Peak Brigthness Temp. (K)
183
+ Residual (lo - <lo>) (K)
184
+ I (mJy/beam)
185
+ 0.2
186
+ -15
187
+ -10
188
+ -5
189
+ 0.01 0.05 0.1
190
+ 0.4
191
+ 0
192
+ 5
193
+ 10
194
+ 20
195
+ 30
196
+ 40
197
+ 50
198
+ 60
199
+ 70
200
+ 10
201
+ 15
202
+ 0
203
+ 2.0
204
+ 12CO (2-1)
205
+ 231 GHz Continuum
206
+ (a)
207
+ (b)
208
+ (c)
209
+ 1.5
210
+ 1.0
211
+ (arcsec)
212
+ 0.5
213
+ 0.0
214
+ ffset
215
+ -0.5
216
+ -1.0
217
+ -1.5
218
+ -2.0.
219
+ 5
220
+ 2
221
+ 0
222
+ -2
223
+ 2
224
+ 0
225
+ -2
226
+ 2
227
+ 0
228
+ -2
229
+ 1
230
+ Offset (arcsec)
231
+ Offset (arcsec)
232
+ Offset (arcsec)Stadler et al.: A kinematically-detected planet candidate in a transition disk
233
+ Fig. 2: Line of sight velocity maps for data v0 (a) and discminer model vmod (b). (c) Velocity residual map after subtracting v0-vmod, where the
234
+ dust continuum is overlaid in solid contours with equal levels as in Fig. 1. The innermost region was masked during the fit by one beam size in
235
+ radius, shown as the grey shaded ellipse. The insets in subplots (a) & (c) zoom into the innermost region of the disk to highlight the non-Keplerian
236
+ velocities. Contours are drawn at vsys = (4.62 ± 0.60) km s−1 in steps of 0.1 km s−1 and from -60 to 60 m s−1 in steps of 10 m s−1 , respectively. All
237
+ maps show the synthesized beam for CO (black) and the continuum (white) in the lower left corner and are masked where the CO peak intensity
238
+ falls below a 5σ level for panel (a) and where R > Rout for the rest.
239
+ for ∆V and τ0, as well the error of the line centroid fitting δv0
240
+ can be found in Fig. A.4.
241
+ 4. Results
242
+ 4.1. Dust and gas radial and azimuthal brightness profiles
243
+ Figure 1 shows the 1.3 mm dust continuum together with the
244
+ peak brightness temperature map I0 of the 12CO (J=2-1) line
245
+ emission. Both dust and gas tracers show a cavity, and the 12CO
246
+ (J=2-1) line emission extends inward of the dust continuum as
247
+ expected if the continuum ring results from dust trapping (e.g.,
248
+ Facchini et al. 2018b, see Fig. A.2). We note that the 12CO cavity
249
+ appears asymmetric with respect to the position of the star and
250
+ that we observe a gap-like feature in the 12CO peak intensity map
251
+ at R ∼1.2′′, apparent as a plateau of I0 ≈ 31 mJy beam−1 stretch-
252
+ ing over ∆R ≈ 0.1′′ (Fig. A.2). Interestingly, the disk shows sig-
253
+ nificant azimuthal intensity variations (34% at R=0.56′′; 19% at
254
+ 0.39′′ for continuum and gas, respectively, see also Fig. A.3).
255
+ Figure 1 (c) shows the residuals obtained after subtracting an
256
+ azimuthally-averaged radial profile from the 12CO peak bright-
257
+ ness temperature map. Azimuthal variations are clearly apparent
258
+ within the dust cavity, with residual values of > 10σ. The fainter
259
+ regions, distributed along the east-west direction, are broadly
260
+ aligned with fainter regions seen in the continuum (see contours
261
+ of Fig. 1, b and Fig. A.3) and with the shadows reported in scat-
262
+ tered light (Pinilla et al. 2018, ; Kurtovic et al. in prep).
263
+ 4.2. Kinematical features
264
+ 4.2.1. Localized velocity residuals
265
+ The centroid residual map in Fig. 2 (c) shows a prominent, local-
266
+ ized non-Keplerian velocity feature of δv ≈ 60 m s−1 , between
267
+ ∼0.35′′ and 0.55′′ (i.e., 50-80 au), that is, at the edge of the dust
268
+ continuum ring, and oriented at PA ≈ (270 ± 15)◦. To assess its
269
+ significance we follow the Variance Peak method from Izquierdo
270
+ et al. (2021). First, the centroid velocity residuals are folded and
271
+ Fig. 3: Folded velocity residuals (left) and detected clusters of peak
272
+ velocities (right) in the disk reference frame. Green wedges in the right
273
+ plot mark the significant clusters. The position of the localized velocity
274
+ perturbation inferred from these clusters is marked with a magenta point
275
+ with error bars. The gray region (one beam size in radius) indicates the
276
+ masked area, and the black dashed lines, the FWHM of the dust ring.
277
+ subtracted along the disk minor axis to remove axisymmetric
278
+ features. Second, a 2D scan is performed to search for peak
279
+ velocity residuals and obtain their locations in the folded map.
280
+ Using these detected points, a K-means clustering algorithm
281
+ searches for coherent velocity perturbations within predefined
282
+ radial and azimuthal bins (MacQueen 1967; Pedregosa et al.
283
+ 2011). We considered seven radial and ten azimuthal bins, which
284
+ corresponds to a width of roughly one beam size, to identify
285
+ clusters. The algorithm now subdivides the input residual points
286
+ such that the center of each cluster is the closest to all points
287
+ in the cluster, by iteratively minimizing the sum of squared dis-
288
+ tances from the data points to the center of the cluster. This leads
289
+ Article number, page 3 of 13
290
+
291
+ Vo (km/s)
292
+ Vmodel (km/s)
293
+ Vo - Vmodel (m/s)
294
+ 4.2
295
+ 4.4
296
+ 4.6
297
+ 4.8
298
+ 5.0
299
+ 5.2
300
+ 4.2
301
+ 4.4
302
+ 4.6
303
+ 4.8
304
+ 5.0
305
+ -60
306
+ -40
307
+ -20
308
+ 5.2
309
+ 0
310
+ 20
311
+ 40
312
+ 60
313
+ 2.0 FT
314
+ (a)
315
+ (b)
316
+ (c)
317
+ 0.5
318
+ 0.25
319
+ 1.5
320
+ 0.00
321
+ 0.0
322
+ 1.0
323
+ -0.25
324
+ -0.5
325
+ (arcsec)
326
+ 0.5
327
+ 0.25
328
+ 0.000.25
329
+ 0.5
330
+ 0.0
331
+ 0.5
332
+ 0.0
333
+ Offset
334
+ -0.5
335
+ -1.0
336
+ -1.5
337
+ -2.0 E
338
+ 2n
339
+ 6
340
+ 2n
341
+ 2
342
+ -2
343
+ 2
344
+ 0
345
+ -2
346
+ 2
347
+ 0
348
+ 0
349
+ -2
350
+ -1
351
+ -1
352
+ Offset (arcsec)
353
+ Offset (arcsec)
354
+ Offset (arcsec)96
355
+ 0A&A proofs: manuscript no. main
356
+ to irregularly spaced bin boundaries, since the cluster centers are
357
+ near to the densest accumulations of points.
358
+ In Figure 3, we show the folded velocity residual map to-
359
+ gether with the detected peak velocity residuals (grey points).
360
+ The location of the detected peak velocity residuals in azimuth
361
+ and radius, within identified clusters, can be found in Fig. B.1.
362
+ Clusters with high significance (those with peak velocity resid-
363
+ uals larger than 3 times the variance in other clusters) are lo-
364
+ cated within one radial and azimuthal bin shown in Fig. 3 as
365
+ green shaded annuli and wedges, respectively. Taking the centers
366
+ of the selected clusters allows to identify a localized perturba-
367
+ tion at 0.28′′ ± 0.07′′ (R = 41 ± 10 au) and PA = 280◦ ± 2◦. The
368
+ reported errors are the standard deviation of the peak residual
369
+ point (R, φ)-locations within the selected clusters. The detec-
370
+ tion yields a cluster significance of 5.4 σφ in azimuth and 5.3 σR
371
+ in radius, where σ represents the standard deviation of back-
372
+ ground cluster variances with a mean of σφ = 0.034 km2s−2 and
373
+ σR = 0.018 km2s−2 (see black crosses in Fig. B.1). We note that a
374
+ localized signature is robustly detected regardless of the amount
375
+ of clusters defined, which we tested using 7-12 azimuthal or 5-
376
+ 9 radial clusters. We reported the clusters associated with the
377
+ highest significance. Additionally, we note that there are other
378
+ detections with lower significance at 0.65′′ (94 au). This means
379
+ that the radial extent of the prominent perturbation is roughly
380
+ 0.40′′ (58 au) and the global peak of the folded velocity resid-
381
+ uals is at 0.39′′ (56 au) (as can be seen in the middle panel of
382
+ Fig. B.1). This analysis confirms the presence of a significant lo-
383
+ calized non-Keplerian feature as identified visually in Fig. 2 (c),
384
+ within the continuum ring.
385
+ 4.2.2. Spiral feature
386
+ Figure 2 (c) also shows an extended arc-like positive resid-
387
+ ual feature, beyond the dust continuum emission and cover-
388
+ ing nearly 300◦ in azimuth, more evident in the polar depro-
389
+ jection of the velocity residual map (Fig. 4). To assess if this
390
+ feature is a coherent structure, we use the FilFinder pack-
391
+ age (Koch & Rosolowsky 2015) implemented in discminer
392
+ between 0.30′′ and 1.25′′ (43-180 au) (see Fig. A.5). As indi-
393
+ cated by the coherent filaments, the strong localized positive ve-
394
+ locity residual discussed in Sect. 4.2.1 seems to be the starting
395
+ point of a spiral tracing outwards up to roughly 1.1′′ (159 au).
396
+ In Fig. 4, we overlay an Archimedean (linear) spiral, prescribed
397
+ by rspiral = a + b φspiral, using {a, b} = {0.48, 0.12}. Computing
398
+ the pitch angles tan(β) = −(dr/dφ)/r, we obtain values ranging
399
+ from 14◦ to 6◦ over the spiral extent.
400
+ 4.2.3. A possible warp in the 12CO cavity
401
+ An additional feature clearly evident from the velocity maps is
402
+ the highly perturbed inner disk regions. As seen in the inset of
403
+ the v0 line centroid map in Fig. 2 (a), the iso-velocity lines show
404
+ strong bending in the inner region (∼3 beam-sizes in diameter),
405
+ indicative of non-Keplerian velocities. This is likely tracing a
406
+ warp or a misaligned inner disk, as reported in Mayama et al.
407
+ (2018); Pinilla et al. (2018); Sicilia-Aguilar et al. (2020); Ans-
408
+ dell et al. (2020) to explain the scattered light shadows and vari-
409
+ able photometry of J1604. Higher angular resolution deep gas
410
+ observations are however needed to assess its morphology and
411
+ kinematics.
412
+ Fig. 4: Polar projection of the velocity residual map. The black solid
413
+ line shows a linear spiral trace. The grey region indicates the masked
414
+ area and the black dashed lines, the FWHM of the dust ring. The y-axis
415
+ extends further than 360◦ to enhance the visibility of the spiral.
416
+ 4.3. Deprojected velocity components
417
+ To understand the contributions from vφ, vr, vz, we produce three
418
+ centroid residual maps for each velocity component, after de-
419
+ projection (Eq. C.1) assuming that all the velocities are either
420
+ azimuthal, radial or vertical (see Fig. C.2; Teague et al. 2022).
421
+ The localized residual feature at the edge of the dust ring ap-
422
+ pears to trace variations in the vertical vz or in the rotational
423
+ vφ motions, or a combination of both. Radial perturbations can
424
+ be ruled out, since it is located close to the disk red-shifted
425
+ major axis (PAdisk=258◦), where vr,proj ≈ 0. Assuming purely
426
+ rotational velocities, it corresponds to perturbations as high as
427
+ δvφ ≈ 600 m s−1 (∼ 0.4 · vkepler), due to the low disk inclination.
428
+ As seen in Fig. 2 (c), the spiral-like velocity residual feature
429
+ does not change sign around the disk major/minor axes, which
430
+ would occur for rotational vφ or radial vr velocity perturbations,
431
+ respectively (see Eq. C.1). We are likely seeing vertical pertur-
432
+ bations, which we are most sensitive to in a nearly face on disk.
433
+ Figure C.1 shows the deprojected and azimuthally aver-
434
+ aged radial profiles of each velocity component determined
435
+ with eddy. For vφ, we observe super-Keplerian rotation from
436
+ R∼0.35′′-0.70′′ (51-101 au), peaking at 0.45′′ (65 au) right be-
437
+ yond the dust continuum. The rotational velocities then sharply
438
+ drop to being sub-Keplerian in the inner disk regions. However,
439
+ we stress that the azimuthally averaged velocities at the radial
440
+ location of the strong localized perturbation (R∼ 0.3 − 0.6′′,43-
441
+ 87 au) are likely affected by the feature. We tentatively observe
442
+ radial inflow inward of the CO intensity peak but with very large
443
+ uncertainties on vr. Finally, we mostly detect downward vertical
444
+ motion of the disk within R∼1.25′′ (181 au).
445
+ 5. Discussion
446
+ 5.1. Origin of v0 residuals
447
+ In this paper, we report the detection of two main non-Keplerian
448
+ features, in addition to highly perturbed gas velocities in the gas
449
+ cavity, that are: (1) a localized positive residual near the edge
450
+ of the dust ring, and (2) an extended spiral-like feature, possibly
451
+ starting from (1). A variety of velocity residual features were
452
+ detected in other systems, with a diverse range of inclinations
453
+ (e.g., Wölfer et al. 2022). In the case of TW Hya, a similarly
454
+ face-on disk, the detected perturbations are ∼40 m s−1 (Teague
455
+ et al. 2022), lower than what is derived for (1) that can account
456
+ Article number, page 4 of 13
457
+
458
+ 0.0
459
+ 0.2
460
+ 0.4
461
+ 0.6
462
+ 0.8
463
+ 1.0
464
+ 1.2
465
+ 1.4
466
+ 1.6
467
+ 1.8
468
+ 270
469
+ major beam size
470
+ 60
471
+ dust ring FWHM
472
+ 50
473
+ linear spiral
474
+ 40
475
+ 180
476
+ 30
477
+ Postion Angle (degree)
478
+ 20
479
+ 90
480
+ 10
481
+ 0
482
+ -10
483
+ 0
484
+ -20
485
+ disk rotation
486
+ -30
487
+ 270
488
+ -40
489
+ -50
490
+ 60
491
+ 180
492
+ :
493
+ 0.0
494
+ 0.2
495
+ 0.4
496
+ 0.6
497
+ 0.8
498
+ 1.0
499
+ 1.2
500
+ 1.4
501
+ 1.6
502
+ 1.8
503
+ Radius (arcsec)Stadler et al.: A kinematically-detected planet candidate in a transition disk
504
+ for 40% of the local Keplerian velocity assuming that the per-
505
+ turbation is purely due to rotational velocities. This is also larger
506
+ in the magnitude of deviation than the Doppler Flip reported in
507
+ the HD 100546 transition disk (Casassus et al. 2022). These ve-
508
+ locity residuals are often interpreted as tracing planet-disk inter-
509
+ actions from massive companions (Pinte et al. 2022) that poten-
510
+ tially carve out gaps. It is thus worth noting, that our inferred
511
+ planet location (R = 41 ± 10 au) is close to the gap location in
512
+ 13CO (J=2-1) at 37 au reported in van der Marel et al. (2021).
513
+ Comparison with simulations (Rabago & Zhu 2021; Izquierdo
514
+ et al. 2022) or semi-analytical prescriptions (Bollati et al. 2021)
515
+ allows to estimate a possible planet mass from the velocity de-
516
+ viations. To this end, we consider Eq. 14 from Yun et al. (2019)
517
+ that relates the change in rotational velocity δvφ to the planet
518
+ mass Mp through two-dimensional hydrodynamic simulations.
519
+ Since δV is the difference between the super- and sub-
520
+ Keplerian peak, we consider the peak of the super-Keplerian
521
+ motion (vφ/vkep)/vkep (see Fig. C.1) as a lower limit for the
522
+ dimensionless amplitude of the perturbed rotational velocity
523
+ (δmin
524
+ V =0.06), and its double (δmax
525
+ V
526
+ = 0.12) as a upper limit. As-
527
+ suming (H/R)p = 0.1 at the planet location (R = 41 au), we
528
+ estimate a planet mass to roughly be between Mp ≈ (1.6 −
529
+ 2.9)Mjup (α/10−3)0.5.
530
+ The extended spiral-like feature appears to be related to the
531
+ significant localized velocity residual. Due to its low pitch an-
532
+ gles and the consistent positive velocity residuals, we speculate
533
+ that the spiral is caused by buoyancy resonances excited by a
534
+ massive planet located within the dust ring. Indeed, in contrast
535
+ with Lindblad spirals, buoyancy spirals are shown to exhibit
536
+ a tightly wound morphology with predominantly vertical mo-
537
+ tions (Bae et al. 2021). Such a spiral has also been suggested in
538
+ TW Hya (Teague et al. 2019b), which is similar in its radial ex-
539
+ tent as the one reported here. Interestingly, Wölfer et al. (2022)
540
+ reports a tentative arc-feature in J1604, at R ∼ 1.0′′ ranging from
541
+ PA≈ 160 − 200◦, probed by the kinematics of the 12CO (J=3-2)
542
+ line emission, and partly coinciding with the spiral-like feature
543
+ that we detect. Additional observations in optically thin tracers,
544
+ would be very useful to assess its nature.
545
+ 5.2. Kinematic perturbations due to shadows
546
+ The localized residual feature seems to roughly align with the
547
+ shadows detected in scattered light. Comparing Figs. 1 (c) and
548
+ 2 (c), positive and negative v0 residuals broadly align with cold
549
+ and hot regions in the brightness temperature of 12CO, respec-
550
+ tively (see also Fig. A.6). In particular, the orientation of the sig-
551
+ nificant localized velocity perturbation coincides with the west-
552
+ ern shadow. Such a shadow can cool down the disk material and
553
+ possibly induce a local drop in pressure support and therefore
554
+ impact the gas velocity. In this section, we estimate whether
555
+ the detected velocity perturbations could be caused by azimuthal
556
+ variations in temperature. We relate the azimuthal change in tem-
557
+ perature ∆Tφ to variations in rotational velocity ∆vφ by solving
558
+ for the Navier-Stokes-equation in cylindrical coordinates. Fol-
559
+ lowing the derivation in the appendix D, we obtain
560
+ ∆vφ
561
+ vkep
562
+
563
+ �H
564
+ R
565
+ �2 ∆Tφ
566
+ T ,
567
+ (3)
568
+ with H/R the disk aspect ratio, T the 12CO brightness temper-
569
+ ature tracing the gas temperature (as 12CO is optically thick)
570
+ and vkep the Keplerian velocity. Evaluating this equation for a
571
+ large H/R = 0.2 along a radially averaged annulus centered
572
+ at R = (0.39 ± 0.07)′′, where we experience the strongest az-
573
+ imuthal changes in the 12CO brightness temperature of up to
574
+ ∆T ≈ 15K over T = 60K, we estimate the change in azimuthal
575
+ velocity to be a mere δvφ/vkep ≈ 1%. Hence, the shadows cannot
576
+ be solely responsible for the localized velocity residual feature.
577
+ In addition, as the shadows are nearly symmetric, we would ex-
578
+ pect a similar feature opposite along the east direction.
579
+ Montesinos et al. (2016) investigated the development of spi-
580
+ rals due to pressure gradients caused by temperature differences
581
+ between obscured and illuminated regions. In their simulations
582
+ symmetric shadows always form two-armed spirals, however,
583
+ they only develop for massive (0.25 M⋆) and/or strongly illu-
584
+ minated disks (100 L⊙) which does not seem to be the case for
585
+ J1604 (∼ 0.02 M⊙ & 0.7 L⊙, Manara et al. 2020), where we also
586
+ only observe one spiral. Additionally, we would also expect such
587
+ spiral features to appear in the brightness temperature residuals
588
+ (Fig. 1 c). It is therefore unlikely that shadows are responsible
589
+ for the extended spiral-like velocity residual feature.
590
+ 5.3. A warped / misaligned inner disk?
591
+ The bending of the iso-velocity curves that we observe in the in-
592
+ set of Fig. 2 (a), is reminiscent of a warped or misaligned inner
593
+ disk (Juhász & Facchini 2017; Facchini et al. 2018a). However,
594
+ as the inner disk is unresolved in our observations, the warp mor-
595
+ phology can’t be derived. We note that radial inflows are also de-
596
+ generate in appearance with warps as shown by Rosenfeld et al.
597
+ (2014), and that our observations can not be conclusive on the
598
+ origin of the disturbed kinematics in the innermost disk. We at-
599
+ tempted to infer varying position angle or inclination with radius
600
+ by fitting the innermost disk only (R≤0.5′′) with eddy and con-
601
+ sidering a fixed stellar mass but did not find to any significant
602
+ variations compared to our best-fit values. We therefore con-
603
+ strain the warp to be confined within one beam size (∼0.18′′, i.e.,
604
+ 26 au) from the center. We note that we obtain a 5 % higher dy-
605
+ namical mass of the system when fitting for M⋆ while masking
606
+ the innermost beam size in radius, an effect predicted by hydro-
607
+ dynamical simulations of warps (Young et al. 2022).
608
+ While our observations cannot provide a full picture of the
609
+ system due to a limited angular resolution, the very low mass ac-
610
+ cretion rate and near infrared excess (Sicilia-Aguilar et al. 2020),
611
+ as well as the gas cavity in 12CO with non-Keplerian veloci-
612
+ ties suggest the presence of an additional, very massive (pos-
613
+ sibly stellar) companion within the inner ∼0.25′′ (∼35 au). Such
614
+ a companion would need to be on an inclined (nearly polar) orbit
615
+ to misalign the inner disk (Zhu 2019) which would then lead to
616
+ the shadows (Nealon et al. 2019) and variable extinction events
617
+ in the light curves (Ansdell et al. 2020; Sicilia-Aguilar et al.
618
+ 2020). It would however not explain the strong localized veloc-
619
+ ity residual feature that we report, which we speculate traces a
620
+ planetary-mass object located at the edge of the dust continuum.
621
+ Detailed modeling of the system is thus needed to assess the
622
+ need for an additional companion. An interesting comparison
623
+ is the CS Cha spectro-binary system (separation ∼7 au) which
624
+ shows similar dust continuum and gas emission at similarly low
625
+ inclination but no departure from Keplerian rotation in the 12CO
626
+ kinematics (Kurtovic et al. 2022).
627
+ 6. Conclusions
628
+ In this letter, we present new ALMA observations of the 1.3 mm
629
+ dust continuum and the 12CO (J=2-1) line emission from the
630
+ transition disk around RXJ1604.3–2130 A. The dust continuum
631
+ shows a large cavity enclosing a smaller 12CO cavity. Azimuthal
632
+ Article number, page 5 of 13
633
+
634
+ A&A proofs: manuscript no. main
635
+ brightness variations in the 12CO line and dust continuum are
636
+ broadly aligned with shadows detected in scattered light (Pinilla
637
+ et al. 2018). Using the discminer package (Izquierdo et al.
638
+ 2021), we model the channel-by-channel line emission and cal-
639
+ culate the line-of-sight velocity maps. We report the detection
640
+ of a coherent, localized non-Keplerian feature at R = 41 ± 10 au
641
+ and PA = 280◦ ± 2◦, that is within the continuum ring. While
642
+ broadly aligned with the scattered light shadows, the localized
643
+ non-Keplerian feature cannot be due to changes in temperature.
644
+ Instead, we interpret the kinematical perturbation as tracing the
645
+ presence of a massive companion of Mp ≈ (1.6 − 2.9) Mjup. We
646
+ also detect a tightly wound spiral that extends over 300◦ in az-
647
+ imuth, possibly connected to the localized feature and caused by
648
+ buoyancy resonances driven by planet-disk-interactions. Bend-
649
+ ing of the iso-velocity contours within the gas cavity indicates
650
+ a highly perturbed inner region, possibly related to the pres-
651
+ ence of a misaligned inner disk. However, as the putative planet
652
+ at ∼41 au cannot explain the gas cavity, the low accretion rate
653
+ and the misaligned inner disk, we speculate that another mas-
654
+ sive companion, likely on an inclined orbit, shapes the inner
655
+ ∼0.25′′(∼35 au).
656
+ Acknowledgements. We would like to thank the anonymous referee for the
657
+ constructive feedback, as well as Clement Baruteau, Kees Dullemond, Guil-
658
+ laume Laibe and Andrew Winter for helpful discussions. This Letter makes
659
+ use of the following ALMA data: ADS/JAO.ALMA#2017.A.01255.S and
660
+ ADS/JAO.ALMA#2015.1.00964.S. ALMA is a partnership of ESO (represent-
661
+ ing its member states), NSF (USA), and NINS (Japan), together with NRC
662
+ (Canada), NSC and ASIAA (Taiwan), and KASI (Republic of Korea), in co-
663
+ operation with the Republic of Chile. The Joint ALMA Observatory is oper-
664
+ ated by ESO, AUI/NRAO, and NAOJ. This project has received funding from
665
+ the European Research Council (ERC) under the European Union’s Horizon
666
+ 2020 research and innovation programme (PROTOPLANETS, grant agreement
667
+ No. 101002188). Software: CARTA (Comrie et al. 2021), CASA (McMullin
668
+ et al. 2007), Discminer (Izquierdo et al. 2021), Eddy (Teague 2019a), FilFinder
669
+ (Koch & Rosolowsky 2015), GoFish (Teague 2019b), Matplotlib (Hunter 2007),
670
+ Numpy (van der Walt et al. 2011), Scipy (Virtanen et al. 2020).
671
+ References
672
+ Andrews, S. M., Huang, J., Pérez, L. M., et al. 2018, ApJ, 869, L41
673
+ Ansdell, M., Gaidos, E., Hedges, C., et al. 2020, MNRAS, 492, 572
674
+ Bae, J., Isella, A., Zhu, Z., et al. 2022a, arXiv e-prints, arXiv:2210.13314
675
+ Bae, J., Teague, R., Andrews, S. M., et al. 2022b, ApJ, 934, L20
676
+ Bae, J., Teague, R., & Zhu, Z. 2021, ApJ, 912, 56
677
+ Barenfeld, S. A., Carpenter, J. M., Ricci, L., & Isella, A. 2016, ApJ, 827, 142
678
+ Benisty,
679
+ M.,
680
+ Dominik,
681
+ C.,
682
+ Follette,
683
+ K.,
684
+ et
685
+ al.
686
+ 2022,
687
+ arXiv
688
+ e-prints,
689
+ arXiv:2203.09991
690
+ Bollati, F., Lodato, G., Price, D. J., & Pinte, C. 2021, MNRAS, 504, 5444
691
+ Calcino, J., Hilder, T., Price, D. J., et al. 2022, ApJ, 929, L25
692
+ Casassus, S., Cárcamo, M., Hales, A., Weber, P., & Dent, B. 2022, ApJ, 933, L4
693
+ Casassus, S. & Pérez, S. 2019, ApJ, 883, L41
694
+ Comrie, A., Wang, K.-S., Hsu, S.-C., et al. 2021, CARTA: The Cube Analysis
695
+ and Rendering Tool for Astronomy, Zenodo
696
+ Czekala, I., Loomis, R. A., Teague, R., et al. 2021, ApJS, 257, 2
697
+ Dong, R., van der Marel, N., Hashimoto, J., et al. 2017, ApJ, 836, 201
698
+ Facchini, S., Juhász, A., & Lodato, G. 2018a, MNRAS, 473, 4459
699
+ Facchini, S., Pinilla, P., van Dishoeck, E. F., & de Juan Ovelar, M. 2018b, A&A,
700
+ 612, A104
701
+ Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J. 2013, PASP, 125,
702
+ 306
703
+ Gaia Collaboration, Vallenari, A., Brown, A. G. A., et al. 2022, arXiv e-prints,
704
+ arXiv:2208.00211
705
+ Hunter, J. D. 2007, Computing in Science Engineering, 9, 90
706
+ Izquierdo, A. F., Facchini, S., Rosotti, G. P., van Dishoeck, E. F., & Testi, L.
707
+ 2022, ApJ, 928, 2
708
+ Izquierdo, A. F., Testi, L., Facchini, S., Rosotti, G. P., & van Dishoeck, E. F.
709
+ 2021, A&A, 650, A179
710
+ Jorsater, S. & van Moorsel, G. A. 1995, AJ, 110, 2037
711
+ Juhász, A. & Facchini, S. 2017, MNRAS, 466, 4053
712
+ Keppler, M., Teague, R., Bae, J., et al. 2019, A&A, 625, A118
713
+ Koch, E. W. & Rosolowsky, E. W. 2015, MNRAS, 452, 3435
714
+ Köhler, R., Kunkel, M., Leinert, C., & Zinnecker, H. 2000, A&A, 356, 541
715
+ Kostov, V. B., Orosz, J. A., Welsh, W. F., et al. 2016, ApJ, 827, 86
716
+ Kurtovic, N. T., Pinilla, P., Penzlin, A. B. T., et al. 2022, A&A, 664, A151
717
+ Law, C. J., Teague, R., Loomis, R. A., et al. 2021, ApJS, 257, 4
718
+ Long, F., Pinilla, P., Herczeg, G. J., et al. 2018, ApJ, 869, 17
719
+ MacQueen, J. 1967, in 5th Berkeley Symp. Math. Statist. Probability, 281–297
720
+ Manara, C. F., Natta, A., Rosotti, G. P., et al. 2020, A&A, 639, A58
721
+ Mayama, S., Akiyama, E., Pani´c, O., et al. 2018, ApJ, 868, L3
722
+ McMullin, J. P., Waters, B., Schiebel, D., Young, W., & Golap, K. 2007, in As-
723
+ tronomical Society of the Pacific Conference Series, Vol. 376, Astronomical
724
+ Data Analysis Software and Systems XVI, ed. R. A. Shaw, F. Hill, & D. J.
725
+ Bell, 127
726
+ Montesinos, M., Perez, S., Casassus, S., et al. 2016, ApJ, 823, L8
727
+ Nealon, R., Pinte, C., Alexander, R., Mentiplay, D., & Dipierro, G. 2019, MN-
728
+ RAS, 484, 4951
729
+ Norfolk, B. J., Pinte, C., Calcino, J., et al. 2022, ApJ, 936, L4
730
+ Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, the Journal of machine
731
+ Learning research, 12, 2825
732
+ Perez, S., Dunhill, A., Casassus, S., et al. 2015, ApJ, 811, L5
733
+ Pinilla, P., Benisty, M., de Boer, J., et al. 2018, ApJ, 868, 85
734
+ Pinilla, P., de Boer, J., Benisty, M., et al. 2015, A&A, 584, L4
735
+ Pinte, C., Teague, R., Flaherty, K., et al. 2022, arXiv e-prints, arXiv:2203.09528
736
+ Rabago, I. & Zhu, Z. 2021, MNRAS, 502, 5325
737
+ Rich, E. A., Monnier, J. D., Aarnio, A., et al. 2022, AJ, 164, 109
738
+ Rich, E. A., Teague, R., Monnier, J. D., et al. 2021, ApJ, 913, 138
739
+ Rosenfeld, K. A., Chiang, E., & Andrews, S. M. 2014, ApJ, 782, 62
740
+ Rosotti, G. P., Benisty, M., Juhász, A., et al. 2020, MNRAS, 491, 1335
741
+ Sicilia-Aguilar, A., Manara, C. F., de Boer, J., et al. 2020, A&A, 633, A37
742
+ Tang, Y.-W., Guilloteau, S., Dutrey, A., et al. 2017, ApJ, 840, 32
743
+ Teague, R. 2019a, The Journal of Open Source Software, 4, 1220
744
+ Teague, R. 2019b, The Journal of Open Source Software, 4, 1632
745
+ Teague, R., Bae, J., Andrews, S. M., et al. 2022, ApJ, 936, 163
746
+ Teague, R., Bae, J., & Bergin, E. A. 2019a, Nature, 574, 378
747
+ Teague, R., Bae, J., Bergin, E. A., Birnstiel, T., & Foreman-Mackey, D. 2018a,
748
+ ApJ, 860, L12
749
+ Teague, R., Bae, J., Birnstiel, T., & Bergin, E. A. 2018b, ApJ, 868, 113
750
+ Teague, R., Bae, J., Huang, J., & Bergin, E. A. 2019b, ApJ, 884, L56
751
+ Teague, R. & Foreman-Mackey, D. 2018, Bettermoments: A Robust Method To
752
+ Measure Line Centroids, Zenodo
753
+ van der Marel, N., Birnstiel, T., Garufi, A., et al. 2021, AJ, 161, 33
754
+ van der Walt, S., Colbert, S. C., & Varoquaux, G. 2011, Computing in Science
755
+ and Engineering, 13, 22
756
+ Virtanen, P., Gommers, R., Burovski, E., et al. 2020, scipy/scipy: SciPy 1.5.3
757
+ Wölfer, L., Facchini, S., van der Marel, N., et al. 2022, arXiv e-prints,
758
+ arXiv:2208.09494
759
+ Young, A. K., Alexander, R., Rosotti, G., & Pinte, C. 2022, MNRAS, 513, 487
760
+ Yun, H.-G., Kim, W.-T., Bae, J., & Han, C. 2019, ApJ, 884, 142
761
+ Zhang, K., Isella, A., Carpenter, J. M., & Blake, G. A. 2014, ApJ, 791, 42
762
+ Zhu, Z. 2019, MNRAS, 483, 4221
763
+ Zhu, Z., Nelson, R. P., Hartmann, L., Espaillat, C., & Calvet, N. 2011, ApJ, 729,
764
+ 47
765
+ 1 Laboratoire Lagrange, Université Côte d’Azur, CNRS, Observatoire
766
+ de la Côte d’Azur, 06304 Nice, France;
767
+ e-mail: [email protected]
768
+ 2 Univ. Grenoble Alpes, CNRS, IPAG, 38000 Grenoble, France
769
+ 3 European Southern Observatory, Karl-Schwarzschild-Str. 2, D-85748
770
+ Garching bei München, Germany
771
+ 4 Leiden Observatory, Leiden University, P.O. Box 9513, NL-2300 RA
772
+ Leiden, The Netherlands
773
+ 5 Universitá degli Studi di Milano, via Celoria 16, 20133 Milano, Italy
774
+ 6 Department of Earth, Atmospheric, and Planetary Sciences, Mas-
775
+ sachusetts Institute of Technology, Cambridge, MA 02139, USA
776
+ 7 Max Planck Institute for Astronomy, Königstuhl 17, 69117, Heidel-
777
+ berg, Germany
778
+ 8 Mullard Space Science Laboratory, University College London,
779
+ Holmbury St Mary, Dorking, Surrey RH5 6NT, UK
780
+ 9 Department of Astronomy, University of Florida, Gainesville, FL
781
+ 32611, USA
782
+ 10 NASA Headquarters, 300 E Street SW, Washington, DC 20546,
783
+ USA
784
+ 11 National Radio Astronomy Observatory, Charlottesville, VA 22903,
785
+ USA
786
+ 12 The Graduate University for Advanced Studies, SOKENDAI,
787
+ Shonan Village, Hayama, Kanagawa 240-0193, Japan
788
+ 13 Departamento de Astronomía, Universidad de Chile, Camino El
789
+ Observatorio 1515, Las Condes, Santiago, Chile
790
+ Article number, page 6 of 13
791
+
792
+ Stadler et al.: A kinematically-detected planet candidate in a transition disk
793
+ Appendix A: Observations, calibration and imaging
794
+ Table A.1: Summary of the ALMA Band 6 observations of J1604 presented in this paper.
795
+ ID
796
+ EB Code
797
+ Date
798
+ Baselines
799
+ Frequency
800
+ Exp. Time
801
+ PI
802
+ [m]
803
+ [GHz]
804
+ [min]
805
+ 2015.1.00964.S
806
+ X412
807
+ 2016 Jul 2
808
+ 15-704
809
+ 217.2-233.4
810
+ 8.87
811
+ Oberg
812
+ 2017.A.01255.S
813
+ Xb18
814
+ 2019 Sep 4
815
+ 38-3638
816
+ 213.0-230.6
817
+ 14.49
818
+ Benisty
819
+ X2fe5
820
+ 2021 Apr 29
821
+ 15-1263
822
+ 14.47
823
+ X4583
824
+ 2019 Jul 30
825
+ 92-8548
826
+ 29.33
827
+ X5f6e
828
+ 2019 Jul 31
829
+ 92-8548
830
+ 29.33
831
+ X104fc
832
+ 2021 Sep 27
833
+ 70-14362
834
+ 29.33
835
+ To self-calibrate our observations, we proceeded as follows. We first flagged the channels containing the line to produce a
836
+ continuum dataset. We centered the individual execution blocks (EBs) by fitting the continuum visibilities with a ring model,
837
+ allowing for a different center and amplitude, enabling us to recover for the phase-shift and amplitude re-scaling to apply to the EBs
838
+ before combining them. To determine a good initial model for the self-calibration, we used multi-scale cleaning with the tclean
839
+ task using a threshold of 2 times the rms noise level of the image. Using the tasks gaincal and applycal, we corrected for phase
840
+ offsets between spectral windows, and between polarizations considering a solution interval of the scan length (solint=inf).
841
+ Executions obtained in 2019 were concatenated and self calibrated together, and similarly for those obtained in 2021. In addition to
842
+ the first found of self calibration, two additional iterations of phase self-calibration were done with solution intervals of 300s and
843
+ 180s for the 2019 data, and only one for the 2021 data, with a solution intervals of 360s. For both datasets, a round of amplitude
844
+ self-calibration was applied with solint=inf. The solutions were then applied to the gas data. While these two epochs will be
845
+ analyzed separately for the continuum in a forthcoming paper (Kurtovic et al.), to analyze the gas data, we concatenated them
846
+ after checking that the data do not show significant variation between the two epochs. We imaged the resulting visibilities with
847
+ the tclean task using the multi-scale CLEAN algorithm with scales of 0, 1, 3 and 6 times the beam FWHM, and an elliptic CLEAN
848
+ mask encompassing the disk emission. The 12CO (2-1) molecular line observations are imaged with a robust value of 1.0, a channel
849
+ width of 0.1 km s−1 and masked by 4.0 σ threshold. The data was tapered to 0.′′1 and we used the ’JvM correction’ (Jorsater & van
850
+ Moorsel 1995; Czekala et al. 2021).
851
+ Fig. A.1: Gallery of selected channel maps. Panels show the 12CO data (top row) and best-fit model (middle row) channel maps, together with
852
+ intensity residuals in Kelvin for each channel (bottom row), where in the latter the colorbar has been adjusted such that residuals smaller than
853
+ 1σ are white. The beam size is depicted in the lower left corner of each channel. For reference the best-fit systemic velocity was found to be
854
+ vsys = 4.62 km s−1 and the channel spacing is 100 m s−1 .
855
+ Article number, page 7 of 13
856
+
857
+ _ 4.21 km/s
858
+ 4.41 km/s
859
+ 4.61 km/s
860
+ 4.81 km/s
861
+ 5.01 km/s
862
+ 5.21 km/s
863
+ Data
864
+ 250
865
+ Offset [au]
866
+ 0
867
+ -250
868
+ .
869
+ 0
870
+ _4.21 km/s
871
+ 4.41 km/s
872
+ 4.61 km/s
873
+ 4.81 km/s
874
+ 5.01 km/s
875
+ 5.21 km/s
876
+ Model
877
+ 250
878
+ [au]
879
+ Offset
880
+ 0
881
+ 75
882
+ Intensity [K]
883
+ 56
884
+ 37
885
+ -250
886
+ 18
887
+ :
888
+ 4.21 km/s
889
+ 4.41 km/s
890
+ 5.01km/s
891
+ _5.21 km/s
892
+ 4.61km/s
893
+ 4.81 km/s
894
+ Residual
895
+ 250
896
+ [au]
897
+ Offset
898
+ 20
899
+ Residuals [K]
900
+ 10
901
+ 0
902
+ -250
903
+ -10
904
+ -20
905
+ -250
906
+ 250
907
+ Offset [au]A&A proofs: manuscript no. main
908
+ Fig. A.2: Radial profile of the surface brightness for different tracers. Profiles are normalized to the peak of the emission for the 231 GHz
909
+ continuum, CO peak flux, both for the data and discminer model, as well as for the SPHERE scattered light observation. Shaded regions show
910
+ the standard deviation of each azimuthal average. The lines in the lower right corner show the major beam size (resolution) for each profile in the
911
+ corresponding colour.
912
+ Fig. A.3: Azimuthal profiles of the surface brightness, normalized to the peak of the emission. Profiles extracted at an annulus with a width of
913
+ approximately one corresponding beam size centered at 0.56′′ and 0.39′′ for the 231 GHz continuum and the CO peak flux, which both show
914
+ significant azimuthal intensity variations of 34% and 19%, respectively. Shaded regions show the standard deviation of each radial average.
915
+ Fig. A.4: Additional moment maps of the centroid fitting. Panels show the line width ∆V (a), the peak optical depth τ0 (b) and the error of the
916
+ centroid fitting δv0. Note that for τ0 < 1 one can assume the line profile to be well presented by a Gaussian, while for τ0 > 5 the line profile
917
+ has a saturated cored, i.e. a very flat top (see Eq. 2). The beam size is depicted in the lower left corner and only regions where I0 > 5σ with
918
+ σ = 1.1 mJy beam−1 are shown.
919
+ Article number, page 8 of 13
920
+
921
+ CO peak intensity
922
+ 100
923
+ model peak intensity
924
+ Normalised Surface Brightness
925
+ dust continuum
926
+ scattered light
927
+ 10-
928
+ 0.00
929
+ 0.25
930
+ 0.50
931
+ 0.75
932
+ 1.00
933
+ 1.25
934
+ 1.50
935
+ 1.75
936
+ Radius (arcsec)1.2
937
+ 12CO at R=(0.39±0.07)"
938
+ Normalised Surface Brightness
939
+ Continuum at R=(0.56±0.02)"
940
+ .0
941
+ 0.8
942
+ 0.7
943
+ 0.6
944
+ -180 -150 -120
945
+ -90
946
+ -60
947
+ -30
948
+ 0
949
+ 30
950
+ 60
951
+ 90
952
+ 120
953
+ 150
954
+ 180
955
+ Position Angle (deg)△V (m/s)
956
+ 6vo (m/s)
957
+ To
958
+ 100
959
+ 200
960
+ 300
961
+ 400
962
+ 500
963
+ 46810121416
964
+ ¥18
965
+ 20
966
+ 2
967
+ 4
968
+ 6
969
+ 8
970
+ 10
971
+ ¥12
972
+ 14
973
+ 16
974
+ 18
975
+ 20
976
+ 0
977
+ 2
978
+ 2.0 F
979
+ (a)
980
+ (b)
981
+ (c)
982
+ 1.5
983
+ 1.0
984
+ 0.5
985
+ 0.0
986
+ Offset (
987
+ -0.5
988
+ -1.0
989
+ -1.5
990
+ D
991
+ -2.0
992
+ 2
993
+ 0
994
+ 0
995
+ 2
996
+ 0
997
+ 2
998
+ 2
999
+ Offset (arcsec)
1000
+ Offset (arcsec)
1001
+ Offset (arcsec)Stadler et al.: A kinematically-detected planet candidate in a transition disk
1002
+ Fig. A.5: Polar map of the velocity residuals. Same as Fig. 4, but now overlaid by filamentary structures found by FilFinder. The red and blue
1003
+ lines overplotted are the medial axes of the filamentary structures found by the algorithm. To trace the apparent spiral in the residuals, we restricted
1004
+ the algorithm to search for filaments in the radial locations r = [0.3, 1.25]′′. For the filamentary detection, we assume a smoothing size of one
1005
+ synthesized beam size and a minimum size of 500 pixels for a filament to be considered.
1006
+ Fig. A.6: Polar contour map of the centroid residuals. Upper panel shows residuals inside the cavity, middle panel between cavity and outer edge
1007
+ of dust ring and lower panel the outer disk. The radial spacing between each contour is ∼1.8 au and the opacity of the lines increase with radius.
1008
+ We like to emphasis that the bump between PA≈ −(110 − 70)◦ in the middle panel makes most of the residuals points we detect with the Peak
1009
+ Variance method of discminer, which can readily be seen in the left panel of Fig. B.1.
1010
+ In Fig. A.7, we show a comparison of velocity residual maps for additional cubes, imaged using different imaging parameters
1011
+ to assess the robustness of our detections. We compare residual maps for the same cube as used in the main text, but without JvM-
1012
+ correction, and for a cube imaged with a different tapering (0.15" instead of 0.10"). Best-fit Keplerian models were subtracted from
1013
+ each of the cubes. As evident from the comparison of the residual maps, the detection of non-Keplerian features reported are robust
1014
+ irrespectively of the imaging parameters. However, the detailed morphology of the velocity residual peaks changes with imaging
1015
+ Article number, page 9 of 13
1016
+
1017
+ Location (0.09i< R ≤ 0.29)"
1018
+ Continuum shadows
1019
+ 400
1020
+ Centroid residual (m/s)
1021
+ 200
1022
+ 0
1023
+ 200
1024
+ -400
1025
+ -270
1026
+ -240
1027
+ -210
1028
+ -180
1029
+ -150
1030
+ -120
1031
+ -90
1032
+ 09-
1033
+ -30
1034
+ 0
1035
+ 30
1036
+ 60
1037
+ 60
1038
+ FLocation (0.29i< R ≤ 0.62)"
1039
+ Centroid residual (m/s)
1040
+ 40
1041
+ 20
1042
+ 0
1043
+ -20
1044
+ -40
1045
+ -60
1046
+ -270
1047
+ -240
1048
+ -210
1049
+ -180
1050
+ -150
1051
+ -120
1052
+ -90
1053
+ -60
1054
+ -30
1055
+ 0
1056
+ 30
1057
+ 60
1058
+ 40 FLocation (0.62 < R ≤ 1.38)"
1059
+ Centroid residual (m/s)
1060
+ 20
1061
+ 0
1062
+ -20
1063
+ -40
1064
+ -270
1065
+ -240
1066
+ -210
1067
+ -180
1068
+ -150
1069
+ -120
1070
+ -90
1071
+ -60
1072
+ -30
1073
+ 0
1074
+ 30
1075
+ 60
1076
+ Position Angle (deg)0.0
1077
+ 0.2
1078
+ 0.4
1079
+ 0.6
1080
+ 0.8
1081
+ 1.0
1082
+ 1.2
1083
+ 1.4
1084
+ 1.6
1085
+ 1.8
1086
+ 270
1087
+ major beam size
1088
+ 60
1089
+ dust ring FWHM
1090
+ 50
1091
+ linear spiral
1092
+ 40
1093
+ 180
1094
+ 30
1095
+ Postion Angle (degree)
1096
+ 20
1097
+ 90
1098
+ 10
1099
+ 0
1100
+ 0
1101
+ -20
1102
+ -30
1103
+ 270
1104
+ -40
1105
+ -50
1106
+ 60
1107
+ 180
1108
+ 0.0
1109
+ 0.2
1110
+ 0.4
1111
+ 0.6
1112
+ 0.8
1113
+ 1.0
1114
+ 1.2
1115
+ 1.4
1116
+ 1.6
1117
+ 1.8
1118
+ Radius (arcsec)A&A proofs: manuscript no. main
1119
+ Fig. A.7: Comparison of the velocity residual maps for different imaging parameters. Left corresponds to the cube used in the main text, while the
1120
+ middle panel corresponds to the same cube without JvM-correction. The right panel shows the residual maps for a non-JvM corrected cube with a
1121
+ different taper (0.15′′). In all pannels the dust continuum is overlaid in solid contours with equal levels as in Fig. 1. Best-fit Keplerian models were
1122
+ subtracted from each of the cubes. The detection of the non-Keplerian features is quite robust irrespectively of the imaging procedure.
1123
+ parameters, and as a consequence, the value of the inferred planet location from the discminer analysis. We find that the best-fit
1124
+ discminer models to the non-JvM corrected cubes are similar within 3% w.r.t. the best-fit parameters listed in B, with the exception
1125
+ of the line slope and some of the peak intensity parameters, that vary up to 15%. While estimating the systematics due to imaging
1126
+ parameters is beyond the scope of this letter, Fig. A.7 provides the evidence that the detection of non Keplerian features is robust.
1127
+ We measure the rms in a line-free channel to be 2.6 mJy beam−1 and 2.9 mJy beam−1 for the non-JvM corrected cubes with a taper
1128
+ of 0.10 and 0.15, respectively.
1129
+ Appendix B: Model best fit parameters
1130
+ Attribute
1131
+ Prescription
1132
+ Best-fit parameters
1133
+ Centre offset
1134
+ xc, yc
1135
+ xc = −2.66+0.05
1136
+ −0.06 au
1137
+ yc = −0.07 ± 0.03 au
1138
+ Position angle
1139
+ PA
1140
+ PA = 258.75+0.06
1141
+ −0.05 deg
1142
+ -
1143
+ Systemic velocity
1144
+ vsys
1145
+ vsys = 4617.2+0.3
1146
+ −0.4 m s−1
1147
+ -
1148
+ Rotation velocity
1149
+ vkep =
1150
+
1151
+ GM⋆
1152
+ R
1153
+ M⋆ = 1.220 ± 0.001 M⊙
1154
+ -
1155
+ Ip = Ip0 (R/Rbreak)p0
1156
+ R ≤ Rbreak
1157
+ Ip0 = 9.388+0.003
1158
+ −0.005 mJy pixel−1
1159
+ p0 = 1.497+0.004
1160
+ −0.005
1161
+ Peak intensity
1162
+ Ip = Ip0 (R/Rbreak)p1
1163
+ Rbreak < R ≤ Rout
1164
+ Rbreak = 56.78+0.06
1165
+ −0.05 au
1166
+ p1 = −0.789 ± 0.001
1167
+ Ip = 0
1168
+ R < Rout
1169
+ Rout = 267.2 ± 0.1 au
1170
+ -
1171
+ Line width
1172
+ Lw = Lw0(R/D0)p
1173
+ Lw0 = 0.4097 ± 0.0004 km s−1
1174
+ p = −0.592+0.001
1175
+ −0.002
1176
+ Line slope
1177
+ Ls = Ls0(R/D0)p
1178
+ Ls0 = 4.569+0.008
1179
+ −0.009
1180
+ p = −0.454+0.005
1181
+ −0.008
1182
+ Table B.1: Table of attributes of the discminer model for the 12CO intensity channel maps of the disk around J1604. PA is the
1183
+ position angle of the semi-major axis of the disc on the red-shifted side, R the cylindrical radius and D0 = 100 au a normalization
1184
+ constant for the line properties. The (down-sampled) pixel size of the model is 8.8 au.
1185
+ For the initial emcee run, we use literature values for the position angle and stellar mass (PA=260◦, M⋆ = 1.24 M⊙, Dong et al.
1186
+ 2017; Manara et al. 2020, respectively). The initial values of the other parameters were found by comparing the overall morphology
1187
+ between the data and a prototype model. We performed the MCMC search with 150 walkers which evolved for 2000 steps for an
1188
+ initial burn-in stage. We proceeded in two steps. First, we masked the disk region inward of the dust continuum and only fitted
1189
+ the outer disk (R > 90 au) to get a robust estimate of the stellar mass and avoid confusion of the code with strongly non-Keplerian
1190
+ velocity features in the inner regions. In this run, we interestingly find a strong offset from the disk center in x-direction of −8.0 au. In
1191
+ a second step, we fixed the stellar mass and now fitted for the whole disk, masking an inner region corresponding to one major beam
1192
+ Article number, page 10 of 13
1193
+
1194
+ main cube
1195
+ non-JvM
1196
+ non-JvM
1197
+ taper 0.10
1198
+ taper 0.10
1199
+ taper 0.15Stadler et al.: A kinematically-detected planet candidate in a transition disk
1200
+ size (26 au) in radius where effects of beam smearing are strongest. We run 150 walkers for 20000 steps till we reach convergence,
1201
+ resembled by a nearly normal distribution of the walkers. The variance and median of the parameters walkers remain effectively
1202
+ unchanged after ∼ 7000 steps. The best-fit parameters are the median of the posterior distributions and given errors are the 16 and
1203
+ 84 percentiles in the last 5000 steps of the 20000 step run, summarized in Table B.1.
1204
+ Fig. B.1: Location of the folded peak velocity residuals. The detected points are shown in azimuth (left) and radius (middle) obtained with the
1205
+ Peak Variance method. Colours indicate the 7 different radial clusters specified, where blue peak residual points are within detected significant
1206
+ radial cluster. The black crosses are the velocity variances of the clusters plotted at the (R, φ)-location of each cluster center. The centers of the
1207
+ accepted clusters (those with peak velocity residuals larger then three times the variance in other clusters) in radius and azimuth are marked with
1208
+ black vertical lines in both panels. The right hand plot shows the normal distribution of the peak residual points in a histogram. Note that outliers
1209
+ of the distribution are related to the localized perturbation. The maximum value of all peak folded centroid residuals is at 0.39′′ (57 au), its mean
1210
+ value is 39 m/s and 1σv = 20 m/s (not to be mistaken with the cluster variances).
1211
+ Appendix C: Decomposition and deprojection of velocity components
1212
+ To determine the rotation curves of each velocity component, we use the code eddy (Teague 2019a). We follow the method presented
1213
+ in Teague et al. (2018b) that uses a Gaussian process to determine the azimuthal vφ and radial vr velocity components along a given
1214
+ annulus. To this end, we divide the disk into concentric annuli with a radial width of 1/4 of the synthesized beam (∼ 0.05′′) ranging
1215
+ from 0.18′′ to 1.85′′ and extract the velocities over 20 iterations to minimize their standard deviation. To obtain the vertical velocity
1216
+ component vz, we use the measured azimuthally averaged profiles of vφ and vr and extend them to produce 2D maps, considering
1217
+ the projection of these components along the line of sight:
1218
+ vφ, proj = vφ cos (φ) sin (|i|),
1219
+ vr, proj = vr sin (φ) sin (i),
1220
+ vz, proj = −vz cos (i),
1221
+ (C.1)
1222
+ where φ is the polar angle in the disk frame (such that φ= 0 corresponds to the red-shifted major axis) and i the inclination of the
1223
+ disk. In the case of J1604, the disk rotates clockwise which corresponds to a negative inclination in the above definition. We then
1224
+ subtract these maps together with the systemic velocity vsys from the line of sight velocity v0-map (Fig. 2,a) to obtain a map of the
1225
+ vertical velocity component vz, proj:
1226
+ vz, proj = v0 − vsys − vφ, proj − vr, proj,
1227
+ (C.2)
1228
+ The radial profile of vz is obtained by deprojecting and azimuthally averaging its 2D velocity map. The radial profiles of the
1229
+ deprojected velocity components can be found in Fig. C.1.
1230
+ Article number, page 11 of 13
1231
+
1232
+ 0
1233
+ 0.40
1234
+ acc. cluster centers
1235
+ 0.10
1236
+ .
1237
+ 80
1238
+ 80
1239
+ 2
1240
+ 0.35
1241
+ 3
1242
+ Cluster Velocity
1243
+ I Residual (m/s)
1244
+ 70
1245
+ 70
1246
+ (s/w)
1247
+ 0.08
1248
+ 5
1249
+ 0.30
1250
+ 6
1251
+ Residual
1252
+ 60
1253
+ 1g
1254
+ 60
1255
+ 0.25
1256
+ 0.06
1257
+ y Variance (
1258
+ 50
1259
+ 50
1260
+ Folded Centroid
1261
+ .
1262
+ Centroid
1263
+ 0.20
1264
+ X
1265
+ S
1266
+ :
1267
+ X
1268
+ 40
1269
+ 40
1270
+ 0.15
1271
+ 0.04
1272
+ (km²/s2)
1273
+ Folded (
1274
+ C
1275
+ .
1276
+ 30
1277
+ X
1278
+ 30
1279
+ 0.10
1280
+ 8
1281
+ 0.02
1282
+ 20
1283
+ 20
1284
+ 0.05
1285
+ X
1286
+ X
1287
+ :
1288
+ X
1289
+ LX.
1290
+ X
1291
+ X
1292
+ X
1293
+ XK
1294
+ 0.00
1295
+ 0.00
1296
+ -80
1297
+ 0.0
1298
+ -40
1299
+ 0
1300
+ 40
1301
+ 80
1302
+ 0.5
1303
+ 1.0
1304
+ 1.5
1305
+ Azimuth (degree)
1306
+ Radius (arcsec)A&A proofs: manuscript no. main
1307
+ Fig. C.1: Azimuthally averaged and deprojected azimuthal, radial and vertical velocity components. The radial width of each annulus is 1/4
1308
+ synthesized beam size. The error bars are given by the standard deviation for each velocity component averaged over the 20 iterations used.
1309
+ Fig. C.2: J1604 deprojected velocity components. It is assumed that all velocities are either azimuthal (left column), radial (central column) or
1310
+ vertical (right column). For the azimuthal and radial components wedges along the minor and major axis have been masked as the observations
1311
+ are insensitive to these components (see Eq. C.1).In each panel the synthesised beam is shown in the lower left corner.
1312
+ Article number, page 12 of 13
1313
+
1314
+ 6
1315
+ dust ring FWHM
1316
+ 12CO lo peak
1317
+ (s/u>y)
1318
+ Vkepl
1319
+ 2
1320
+ 0.1
1321
+ (V - Vkepi)/Vkepl
1322
+ 0.0
1323
+ -0.1
1324
+ 0.2
1325
+ -0.3
1326
+ 100
1327
+ 50
1328
+ (s/w)
1329
+ 0
1330
+ -50
1331
+ -100
1332
+ 20
1333
+ (s/w)
1334
+ 0
1335
+ -20
1336
+ 0.25
1337
+ 0.50
1338
+ 0.75
1339
+ 1.00
1340
+ 1.25
1341
+ 1.50
1342
+ 1.75
1343
+ Radius (arcsec)Vμ (m/s)
1344
+ Vr (m/s)
1345
+ Vz (m/s)
1346
+ -400
1347
+ -200
1348
+ 0
1349
+ 200
1350
+ 400
1351
+ 600
1352
+ 600
1353
+ -400
1354
+ -200
1355
+ 0
1356
+ 200
1357
+ 400
1358
+ 600
1359
+ -60
1360
+ -40
1361
+ -20
1362
+ 0
1363
+ 20
1364
+ 40
1365
+ 600
1366
+ 60
1367
+ ← slower
1368
+ faster
1369
+ ←inwards
1370
+ outwards
1371
+ ← downwards
1372
+ upwards→
1373
+ 2.0
1374
+ (b)
1375
+ (a)
1376
+ (c)
1377
+ 1.5
1378
+ 1.0
1379
+ 0.5
1380
+ 0.0
1381
+ -0.5
1382
+ -1.0
1383
+ -1.5
1384
+ -2.0
1385
+ 2
1386
+ 1
1387
+ 0
1388
+ -2
1389
+ 2
1390
+ 0
1391
+ -1
1392
+ -2
1393
+ 2
1394
+ 1
1395
+ 0
1396
+ -1
1397
+ -2
1398
+ Offset (arcsec)
1399
+ Offset (arcsec)
1400
+ Offset (arcsec)Stadler et al.: A kinematically-detected planet candidate in a transition disk
1401
+ Appendix D: Derivation of ∆vφ in dependence of azimuthal temperature variations ∆T
1402
+ To relate the change in the brightness temperature ∆T of 12CO to variations in the rotational velocity vφ we solve the Navier-Stokes-
1403
+ equation in cylindrical coordinates in the φ-direction:
1404
+ ρg
1405
+
1406
+ R
1407
+ ∂vφ
1408
+ ∂φ = 1
1409
+ R
1410
+ ∂p
1411
+ ∂φ + µ
1412
+ � ∂
1413
+ ∂R
1414
+ � 1
1415
+ R
1416
+
1417
+ ∂R(Rvφ)
1418
+ � 1
1419
+ R2
1420
+ ∂2vφ
1421
+ ∂φ2
1422
+
1423
+ ,
1424
+ (D.1)
1425
+ with the cylindrical radius R, the gas density ρg and the mean molecular weight µ. In a first step, we assume that the radial variations
1426
+ within the chosen annulus are negligible and insert the disk gas pressure in the vertically isothermal assumption p = ρgc2
1427
+ s = ρg
1428
+ kBT
1429
+ µmp ,
1430
+ with the Boltzmann constant kB and the proton mass mp. In the second step, we further assume the gas density to be constant along
1431
+ the annulus ρg = const. and re-arrange the equation.
1432
+ ρg
1433
+
1434
+ R
1435
+ ∂vφ
1436
+ ∂φ = 1
1437
+ R
1438
+
1439
+ ∂φ
1440
+
1441
+ ρg
1442
+ kBT
1443
+ µmp
1444
+
1445
+ + µ
1446
+ R2
1447
+ ∂2vφ
1448
+ ∂φ2
1449
+ ∂vφ
1450
+ ∂φ −
1451
+ µ
1452
+ vφRρg
1453
+ ∂2vφ
1454
+ ∂φ2 =
1455
+ kB
1456
+ vφµmp
1457
+ ∂T
1458
+ ∂φ
1459
+ ∂vφ
1460
+ ∂φ −
1461
+ 0.4µ
1462
+
1463
+
1464
+ 2πΣg
1465
+ ∂2vφ
1466
+ ∂φ2 =
1467
+ kB
1468
+ vφµmp
1469
+ ∂T
1470
+ ∂φ
1471
+ (D.2)
1472
+ In the last step, we inserted the gas midplane density ρg = Σg/(
1473
+
1474
+ 2πH) = Σg/(
1475
+
1476
+ 2π 0.2R) assuming a disk aspect ratio of H/R = 0.2.
1477
+ Assuming that vφ ≈ vkep and Σg ≈ 1g/cm2 (see Fig. 3 of Dong et al. 2017) at the location of the annulus at R ∼ 0.4′′ (58 au), we can
1478
+ assess the order of magnitude of the second term on the left hand side of the equation which is only on the order of 10−6. Therefore,
1479
+ we neglect the second order derivative and further identify the sound speed cs:
1480
+ ∆vφ ≈
1481
+ kB
1482
+ vkep µmp
1483
+ T
1484
+ T ∆Tφ ≈ c2
1485
+ s
1486
+ vkep
1487
+ ∆Tφ
1488
+ T
1489
+ ���� ÷ vkep
1490
+ ∆vφ
1491
+ vkep
1492
+
1493
+ � cs
1494
+ vkep
1495
+ �2 ∆Tφ
1496
+ T
1497
+
1498
+ �H
1499
+ R
1500
+ �2 ∆Tφ
1501
+ T
1502
+ (D.3)
1503
+ The last equation now connects the fractional azimuthal temperature variation ∆Tφ/T to the rotational velocity deviation relative to
1504
+ Keplerian ∆vφ/vkep.
1505
+ Article number, page 13 of 13
1506
+
1dAzT4oBgHgl3EQft_1-/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
2tAyT4oBgHgl3EQfovik/content/2301.00513v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f1e160a4d2f22be06ba1e9b18a1393ab462ce96b280ee3ce3199e6a9c595d74c
3
+ size 544634
2tAyT4oBgHgl3EQfovik/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:169f307601eb7b0a66ecc03ba15e01850d6c848583a7844022d1d755f63086d8
3
+ size 5767213
3NFKT4oBgHgl3EQf8S4-/content/tmp_files/2301.11948v1.pdf.txt ADDED
@@ -0,0 +1,2362 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Pseudo-Goldstone modes and dynamical gap generation from order-by-thermal-disorder
2
+ Subhankar Khatua,1, 2 Michel J. P. Gingras,2 and Jeffrey G. Rau1
3
+ 1Department of Physics, University of Windsor, 401 Sunset Avenue, Windsor, Ontario, N9B 3P4, Canada
4
+ 2Department of Physics and Astronomy, University of Waterloo, Waterloo, Ontario, N2L 3G1, Canada
5
+ (Dated: January 31, 2023)
6
+ Accidental ground state degeneracies – those not a consequence of global symmetries of the Hamiltonian
7
+ – are inevitably lifted by fluctuations, often leading to long-range order, a phenomenon known as “order-by-
8
+ disorder” (ObD). The detection and characterization of ObD in real materials currently lacks clear, qualitative
9
+ signatures that distinguish ObD from conventional energetic selection. We show that for order-by-thermal-
10
+ disorder (ObTD) such a signature exists: a characteristic temperature dependence of the fluctuation-induced
11
+ pseudo-Goldstone gap. We demonstrate this in a minimal two-dimensional model that exhibits ObTD, the fer-
12
+ romagnetic Heisenberg-compass model on a square lattice. Using spin-dynamics simulations and self-consistent
13
+ mean-field calculations, we determine the pseudo-Goldstone gap, ∆, and show that at low temperatures it scales
14
+ as the square root of temperature,
15
+
16
+ T. We establish that a power-law temperature dependence of the gap is a
17
+ general consequence of ObTD, showing that all key features of this physics can be captured in a simple model
18
+ of a particle moving in an effective potential generated by the fluctuation-induced free energy.
19
+ Strongly competing interactions, or frustration, enhance
20
+ quantum and thermal fluctuations, and undermine the devel-
21
+ opment of conventional magnetic order. The latter can even be
22
+ prevented entirely down to zero temperature, leading to classi-
23
+ cal [1–3] or quantum spin liquids [4–10]. However, additional
24
+ perturbative interactions can relieve the frustration and favor
25
+ the development of long-range order (LRO). Accordingly, the
26
+ majority of spin liquid candidates ultimately evade fate as a
27
+ spin liquid [8, 11]. The ability of these interactions, incon-
28
+ sequential without frustration, to dictate the ground state and
29
+ low-temperature properties of a system is at the root of the
30
+ plethora of exotic phenomena displayed by highly-frustrated
31
+ magnetic materials [10, 12–18].
32
+ This relief of frustration is not always complete. Instead
33
+ of an extensively degenerate manifold, a system can possess
34
+ a sub-extensive accidental ground state degeneracy, unpro-
35
+ tected by symmetry. Classically, this degeneracy can be ro-
36
+ bust to a range of realistic interactions including symmetry-
37
+ allowed two-spin exchange [19]. Here, the role of fluctuations
38
+ is dramatically changed: instead of being detrimental, they
39
+ can lift the classical degeneracy and stabilize order – this is the
40
+ celebrated phenomenon of order-by-disorder (ObD) [20–22].
41
+ While numerous theoretical models have been proposed [20–
42
+ 33], there is a paucity of real materials that unambiguously
43
+ harbor ObD [19, 34–37]. The standard strategy for exper-
44
+ imental confirmation is indirect, relying on parametrizing a
45
+ theoretical model of the material, establishing ObD within
46
+ that model, and then validating its predictions for the ordered
47
+ state experimentally.
48
+ While this program has been applied somewhat success-
49
+ fully to a handful of materials [19, 34–37], the inability
50
+ to evince ObD directly, without relying on detailed mod-
51
+ elling, highlights something lacking in our understanding
52
+ of ObD. Clear qualitative, model-independent signatures are
53
+ needed; for example, experimental observation of characteris-
54
+ tic power-laws in heat capacity or transport can diagnose the
55
+ character of low-energy excitations, such as exchange statis-
56
+ tics, dimensionality or their dispersion relations [9, 11, 38,
57
+ 39]. Does the presence of ObD exhibit a “smoking-gun” ex-
58
+ perimental signature? This can be difficult or subtle to discern.
59
+ For ObD from quantum fluctuations [21], the formation of an
60
+ ObD spin-wave gap is generally not distinguishable from one
61
+ induced energetically by multi-spin interactions [40–42].
62
+ In this Letter, we identify a clear signature of order-by-
63
+ thermal-disorder (ObTD): a dynamically generated gap grow-
64
+ ing as the square root of temperature.
65
+ We investigate this
66
+ gapped “pseudo-Goldstone” (PG) mode [44–46] in a minimal
67
+ 2D classical spin model exhibiting ObTD, the ferromagnetic
68
+ Heisenberg-compass model on a square lattice, belonging to
69
+ a class of models relevant to Mott insulators with strong spin-
70
+ orbit coupling [47–55]. Through spin-dynamics simulations,
71
+ we determine the PG gap, ∆, and show it varies with tempera-
72
+ ture as ∆ ∝
73
+
74
+ T, in quantitative agreement with self-consistent
75
+ mean-field theory (SCMFT). This mode is well-defined, with
76
+ the linewidth, Γ, due to thermal broadening, Γ ∝ T 2 ≪ ∆. We
77
+ further demonstrate that our key results can be captured by an
78
+ effective description of a particle moving in a potential gener-
79
+ ated by the ���uctuation-induced free energy. Using this picture,
80
+ we argue that the temperature dependence of the PG gap,
81
+
82
+ T
83
+ (T) for type-I (II) PG modes [56], is universal, applicable to
84
+ any system exhibiting ObTD. Finally, due to the low dimen-
85
+ sionality [57], ObTD faces a subtle competition against poten-
86
+ tially infrared-divergent fluctuations [58, 59]. While ObTD
87
+ ultimately prevails, and true LRO develops, the magnetization
88
+ displays logarithmic corrections at low temperature, a rem-
89
+ nant of the diverging infrared fluctuations.
90
+ Model.—
91
+ We
92
+ consider
93
+ the
94
+ classical
95
+ ferromagnetic
96
+ Heisenberg-compass model on a square lattice
97
+ H =
98
+
99
+ r
100
+
101
+ −J
102
+
103
+ δ=ˆx,ˆy
104
+ Sr · Sr+δ − K
105
+
106
+ S x
107
+ rS x
108
+ r+ˆx + S y
109
+ rS y
110
+ r+ˆy
111
+ ��
112
+ ,
113
+ (1)
114
+ where Sr ≡ (S x
115
+ r, S y
116
+ r, S z
117
+ r) is a unit vector at site r, and δ =
118
+ ˆx, ˆy denote the nearest-neighbor bond directions. We consider
119
+ ferromagnetic Heisenberg and compass interactions with J >
120
+ 0, K > 0 (see SM [60] for a discussion of other signs) and
121
+ with J the unit of energy, setting J ≡ ℏ ≡ kB ≡ 1 throughout.
122
+ For K = 0, the model [Eq.(1)] is the well-known Heisen-
123
+ berg ferromagnet with uniform ferromagnetic ground states
124
+ of arbitrary direction, Sr = ˆn, related by global spin-rotation
125
+ arXiv:2301.11948v1 [cond-mat.str-el] 27 Jan 2023
126
+
127
+ 2
128
+ −π/4
129
+ 0
130
+ π/4
131
+ φ
132
+ 0.0
133
+ 0.2
134
+ 0.4
135
+ 0.6
136
+ 0.8
137
+ P(φ)
138
+ (a)
139
+ L = 14
140
+ L = 10
141
+ L = 6
142
+ [00]
143
+ [π0]
144
+ [ππ]
145
+ [00]
146
+ [0π]
147
+ 0
148
+ 5
149
+ 10
150
+ 15
151
+ 20
152
+ 25
153
+ ω
154
+ (b)
155
+ [00]
156
+ [π0]
157
+ [ππ]
158
+ [0π]
159
+ kx
160
+ ky
161
+ 0
162
+ 2
163
+ 4
164
+ 0
165
+ 200
166
+ 0
167
+ 1
168
+ 2
169
+ 3
170
+ 4
171
+ 5
172
+ 0.0
173
+ 0.2
174
+ 0.4
175
+ 0.6
176
+ ω
177
+ S(0,ω) [arb.]
178
+ (c)
179
+ T = 0.040
180
+ T = 0.032
181
+ T = 0.024
182
+ T = 0.016
183
+ T = 0.008
184
+ [00]
185
+ [π0]
186
+ [ππ]
187
+ [00]
188
+ [0π]
189
+ 0
190
+ 5
191
+ 10
192
+ 15
193
+ 20
194
+ 25
195
+ ω
196
+ (d)
197
+ 0
198
+ 5
199
+ Spin-dynamics
200
+ LSWT
201
+ SCMFT
202
+ FIG. 1. (a) Probability distribution, P(φ), of the angle, φ, characterizing the direction of the net magnetization obtained using MC simulations
203
+ with K = 5 at T = 0.4 for several system sizes, L. Due to C4 symmetry, P(φ) is shown for φ ∈ [−π/4, π/4]. (b) Dynamical structure factor,
204
+ S(k, ω) obtained from spin-dynamics simulations for L = 100 with K = 5 at T = 0.4 along a path through the Brillouin zone (see left inset).
205
+ Overall intensity is arbitrary. (Right inset) Spectrum near [00] showing the PG gap [43]. (c) Dynamical structure factor at k = 0, S(0, ω),
206
+ obtained from spin-dynamics simulations for L = 40 at various temperatures with K = 5. Overall intensity is arbitrary. (d) Excitation spectrum
207
+ along the same path as in panel-(b) from the LSWT, SCMFT, and spin-dynamics simulations with K = 5 for L = 100 at T = 0.4. The
208
+ spin-dynamics spectrum tracks the frequencies of maximum of S(k, ω). The inset highlights a small region near [00], showing the PG mode.
209
+ symmetry. For K > 0, this symmetry is absent and H in
210
+ Eq. (1) is minimized by any uniform magnetization in the
211
+ ˆx − ˆy plane. These ground states are characterized by an an-
212
+ gle φ ∈ [0, 2π) with Sr = cos φ ˆx + sin φ ˆy. Unlike the pure
213
+ Heisenberg ferromagnet, these are only accidentally degener-
214
+ ate, as the continuous in-plane spin rotations connecting them
215
+ do not preserve the anisotropic compass term. However, a dis-
216
+ crete C4 symmetry about the ˆz axis and C2 symmetries about
217
+ the ˆx and ˆy axes still remain.
218
+ Simulations.— We first show that this model exhibits ObTD
219
+ via Monte Carlo (MC) simulations on a lattice with N = L2
220
+ sites. To expose the state selection, we construct a proba-
221
+ bility distribution for magnetization direction, encoded in φ,
222
+ P(φ), using a sample of thermalized states (see SM [60]). As
223
+ shown in Fig. 1(a), P(φ) exhibits maxima at φ = 0, π/2, π,
224
+ 3π/2, corresponding to ferromagnetic ground states with ˆn
225
+ along the ±ˆx, ±ˆy directions. At low temperatures, fluctua-
226
+ tions thus select four discrete ground states via ObTD from a
227
+ one-parameter manifold of states.
228
+ We now consider the classical dynamics to examine the as-
229
+ sociated PG mode. The equation of motion for the classical
230
+ spins is the Landau-Lifshitz equation [61], dSi/dt = Br × Sr,
231
+ describing precession about the exchange field, Br, produced
232
+ by neighboring spins
233
+ Br ≡ −
234
+
235
+ δ=±ˆx,±ˆy
236
+
237
+ JSr+δ + KS δ
238
+ r+δδ
239
+
240
+ .
241
+ (2)
242
+ Starting with states drawn via MC sampling at temperature T,
243
+ we numerically integrate the Landau-Lifshitz equations, and
244
+ compute the dynamical structure factor, S(k, ω) = ⟨|Sk(ω)|2⟩,
245
+ where Sk(ω) is the Fourier transform of spins, and ⟨· · ·⟩ de-
246
+ notes averaging over the initial states [60]. Results for S(k, ω)
247
+ at a representative T and K [60] are shown in Fig. 1(b),
248
+ exhibiting sharp spin-waves with a nearly gapless mode at
249
+ k = 0. Closer examination reveals a well-defined gap, as
250
+ highlighted in the top right inset of Fig. 1(b) – this is the PG
251
+ gap.
252
+ To determine the PG gap quantitatively, we consider a cut
253
+ of the structure factor at k = 0, i.e., S(0, ω). As the PG
254
+ gap is much smaller than the bandwidth of the spectrum [see
255
+ Fig. 1(b)], a significantly higher frequency resolution is re-
256
+ quired to accurately compute the gap [60], so a much longer
257
+ integration time window is necessary. Cuts, S(0, ω), for sev-
258
+ eral temperatures are presented in Fig. 1(c), with the peak lo-
259
+
260
+ 3
261
+ 0.000
262
+ 0.005
263
+ 0.010
264
+ 0.015
265
+ 0.020
266
+ 0.025
267
+ 0.030
268
+ 0.035
269
+ 0.040
270
+ T
271
+ 0.0
272
+ 0.1
273
+ 0.2
274
+ 0.3
275
+ 0.4
276
+ 0.5
277
+
278
+ 0.00
279
+ 0.25
280
+ 0.50
281
+ 0.75
282
+ 1.00
283
+ T
284
+ 0.0
285
+ 0.2
286
+ 0.4
287
+ Γ
288
+ Spin-dynamics
289
+ SCMFT
290
+ SCMFT asymptotic limit
291
+ FIG. 2. Pseudo-Goldstone gap, ∆, as a function of temperature from
292
+ spin-dynamics simulations with K = 5. The data is well-described
293
+ by the fit ∆ = 2.46242
294
+
295
+ T − 3.21907 T 3/2. The SCMFT gap agrees
296
+ with it quantitatively and provides the asymptotic T → 0 scaling,
297
+ 2.46147
298
+
299
+ T. (Inset) Linewidth of the PG mode, Γ, as a function of
300
+ temperature from spin-dynamics simulations. It is well described by
301
+ the fit, Γ = 0.709286 T 2 − 0.329751 T 3. All data have been extrapo-
302
+ lated in the system size to the thermodynamic limit [60].
303
+ cation indicating the PG gap (see SM [60]). The temperature
304
+ dependence of ∆ is shown in Fig. 2. The leading contribution
305
+ to the PG gap scales as the square root of temperature, van-
306
+ ishing as T → 0, and is well-described by the fit ∆ ∼ 2.46
307
+
308
+ T.
309
+ The thermal broadening of the spectrum induces a finite
310
+ width to all excitations, including the PG mode. The PG mode
311
+ linewidth, Γ, can be obtained from the full-width at half max-
312
+ imum of S(0, ω) [see Fig. 1(c)] as a function of temperature.
313
+ The inset in Fig. 2 shows that Γ ∝ T 2 at low temperatures
314
+ (see SM [60]). Since Γ ≪ ∆ as T → 0, this PG mode is
315
+ well-defined.
316
+ Spin-wave analysis.— The simulations have revealed that
317
+ the system has LRO and hosts a PG excitation, where the
318
+ PG gap and linewidth scale with temperature as
319
+
320
+ T and
321
+ T 2, respectively.
322
+ To understand how these scaling laws
323
+ arise, we consider a spin-wave analysis about the ordered
324
+ state [62]. Since tackling spin-wave interactions is difficult
325
+ within a purely classical approach [63–65], we follow the
326
+ more widely used and computationally convenient quantum
327
+ spin-wave analysis [66–68], taking the classical limit only at
328
+ the end.
329
+ We first discuss the spectrum and state selection due to
330
+ ObTD in linear spin-wave theory (LSWT). Expanding about a
331
+ classical ground state (parametrized by φ) using the Holstein-
332
+ Primakoff (HP) transformation [62], we obtain to O(S )
333
+ H2 =
334
+
335
+ k
336
+
337
+ Aka†
338
+ kak + 1
339
+ 2!
340
+
341
+ Bka†
342
+ ka†
343
+ −k + H.c.
344
+ ��
345
+ ,
346
+ (3)
347
+ where ak denotes the bosonic annihilation operator at wave
348
+ vector k, and Ak and Bk depend on φ, J, and K (see SM [60]).
349
+ H2 in Eq. (3) can be diagonalized by a Bogoliubov transfor-
350
+ mation [62], giving spin-wave energies ωk =
351
+
352
+ A2
353
+ k − B2
354
+ k. As
355
+ the spectrum depends on the ground state angle φ, fluctuations
356
+ can lift the accidental classical degeneracy. To examine state
357
+ selection due to ObTD, we search for the ground states where
358
+ the free energy is minimal. Starting with the quantum free en-
359
+ ergy Fqu = 1
360
+ 2
361
+
362
+ k ωk + T �
363
+ k ln
364
+
365
+ 1 − e−ωk/T�
366
+ , the classical limit
367
+ T ≫ ωk yields F = T �
368
+ k ln ωk [69]. This classical free en-
369
+ ergy has four minima at φ = 0, π/2, π, 3π/2 – establishing
370
+ selection by ObTD, in agreement with the MC results.
371
+ Within LSWT, quantum and classical calculations give the
372
+ same spectrum, ωk [22]. This spectrum, calculated about φ =
373
+ 0, exhibits a gapless mode at k = 0 as shown in Fig. 1(d). To
374
+ obtain a PG gap, spin-wave interactions must be included, as
375
+ we next discuss.
376
+ Interacting spin waves.— Performing the HP expansion to
377
+ next order in 1/S , the LSWT Hamiltonian [Eq. (3)] is aug-
378
+ mented by interaction terms. Three-boson interactions are ab-
379
+ sent due to a C2 symmetry about the ordering direction, leav-
380
+ ing only terms quartic in the bosons at O(S 0) (see SM [60]).
381
+ To treat this interacting problem, we adopt a mean-field ap-
382
+ proach [66, 67], decoupling the quartic terms into products
383
+ of quadratic terms and thermal averages of two-boson oper-
384
+ ators. Following this procedure, the new effective quadratic
385
+ Hamiltonian mirrors Eq. (3), but with Ak and Bk replaced with
386
+ (Ak + δAk) and (Bk + δBk). These corrections are
387
+ δAk = 1
388
+ N
389
+
390
+ q
391
+
392
+ Vk,q,0⟨a†
393
+ qaq⟩ + 1
394
+ 2
395
+
396
+ Dq,−q,k⟨a†
397
+ qa†
398
+ −q⟩ + c.c.
399
+ ��
400
+ ,
401
+ δBk = 1
402
+ N
403
+
404
+ q
405
+
406
+ Dk,−k,q⟨a†
407
+ qaq⟩ + 1
408
+ 2Vq,−q,k−q⟨aqa−q⟩
409
+
410
+ ,
411
+ (4)
412
+ where Vk1,k2,k3 and Dk1,k2,k3 are the coefficients for the 2-2
413
+ and 3-1 magnon scattering terms at O(S 0) [60], and ⟨· · ·⟩ is
414
+ a thermal average. When these averages are computed using
415
+ LSWT [Eq. (3)], the corrections [Eq. (4)] reproduce leading
416
+ order perturbation theory [70, 71]. However, because of the
417
+ gapless mode, these individual δAk and δBk diverge in the
418
+ classical limit and perturbation theory breaks down [60].
419
+ To resolve these divergences, we perform the averages in
420
+ Eq. (4) using SCMFT, obtaining a renormalized spectrum, Ωk
421
+ (see SM [60]). Explicitly, ⟨a†
422
+ qaq⟩ and ⟨a†
423
+ qa†
424
+ −q⟩ are, classically,
425
+ computed self-consistently (until convergence) using Eq. (4)
426
+ and
427
+ ⟨a†
428
+ kak⟩ = T(Ak + δAk)
429
+ Ω2
430
+ k
431
+ ,
432
+ ⟨aka−k�� = −T(Bk + δBk)
433
+ Ω2
434
+ k
435
+ ,
436
+ (5)
437
+ where Ωk =
438
+
439
+ (Ak + δAk)2 − (Bk + δBk)2 and ⟨aka−k⟩ =
440
+ ⟨a†
441
+ ka†
442
+ −k⟩.
443
+ The SCMFT spectrum Ωk, plotted in Fig. 1(d), exhibits a
444
+ clear gap at k = 0. The PG mode, gapless in LSWT, has now
445
+ become gapped due to magnon-magnon interactions. Excel-
446
+ lent agreement between the spectra from SCMFT and spin-
447
+ dynamics simulations is observed across the full Brillouin
448
+ zone [see Fig. 1(d)]. The temperature dependences of ∆ from
449
+ the two approaches in Fig. 2 agree quantitatively, with identi-
450
+ cal
451
+
452
+ T scaling as T → 0. This is a key result of this work,
453
+ establishing a clear spectral signature of ObTD.
454
+
455
+ 4
456
+ While the SCMFT is successful in describing the excita-
457
+ tion energies, it does not address thermal broadening, since
458
+ δAk and δBk are real, giving an infinite magnon lifetime.
459
+ To obtain a finite linewidth, perturbation theory must be car-
460
+ ried out to higher order. We expect that δA0 ≡ δAk=0 and
461
+ δB0 ≡ δBk=0, interpreted as contributions to the magnon self-
462
+ energy [60], can be expanded in T as δA0 = a1T + a2T 2 + · · ·
463
+ and δB0 = b1T + b2T 2 + · · · . Since |A0| = |B0|, reflecting
464
+ the gapless LSWT spectrum, and a1, b1 [the O(T) corrections
465
+ in Eq. (4)] are real; any imaginary part, and thus finite life-
466
+ time, must arise from a2 or b2. Expanding Ω0 ≡ Ωk=0 in T
467
+ yields Im Ω0 ≈ (Im a2) T 2 + · · · (see SM [60]). The real part,
468
+ Re Ω0, maintains its leading
469
+
470
+ T dependence (providing the
471
+ PG gap) while Im Ω0, giving the linewidth, has a leading T 2
472
+ dependence, consistent with the simulation results (see inset
473
+ of Fig. 2).
474
+ Effective description.— We now present an effective de-
475
+ scription capturing the key aspects of the PG mode in a
476
+ significantly simpler language and with broader applicabil-
477
+ ity, adapting an approach formulated for order-by-quantum-
478
+ disorder (ObQD) [72].
479
+ We consider small uniform devi-
480
+ ations from a classical ground state (say φ
481
+ =
482
+ 0) with
483
+ Sr
484
+ ≈ (
485
+
486
+ 1 − φ2 − θ2, φ, θ), accurate to quadratic order in φ
487
+ and θ, where φ is the soft mode and θ its conjugate momentum.
488
+ For small φ and θ, φ ≈ 1
489
+ N
490
+
491
+ r S y
492
+ r and θ ≈ 1
493
+ N
494
+
495
+ r S z
496
+ r, with Pois-
497
+ son bracket {φ, θ} = 1/N. For this configuration, we define an
498
+ effective free energy Feff(θ, φ) = Ecl(θ) − TS (φ), where Ecl(θ)
499
+ is the classical cost of nonzero θ and S (φ) = − �
500
+ k ln ωk(φ)
501
+ is the entropy. For small θ and φ, Feff can be expanded as
502
+ Feff ≈ 1
503
+ 2N
504
+
505
+ Cθθ2 + Cφφ2�
506
+ , where Cθ = (∂2Feff/∂θ2)/N = 2K
507
+ and Cφ = (∂2Feff/∂φ2)/N. Taking Feff as an effective Hamil-
508
+ tonian, the equations of motion [73] for θ and φ are
509
+ ∂φ
510
+ ∂t = + 1
511
+ N
512
+ ∂Feff
513
+ ∂θ
514
+ = +Cθθ,
515
+ ∂θ
516
+ ∂t = − 1
517
+ N
518
+ ∂Feff
519
+ ∂φ
520
+ = −Cφφ,
521
+ (6)
522
+ describing a harmonic oscillator. We identify the PG gap as
523
+ its frequency, ∆ = �CθCφ. Remarkably, the
524
+
525
+ T dependence
526
+ of the PG gap is recovered, since Cφ is O(T) and Cθ is O(1).
527
+ The curvature Cφ can be calculated within LSWT, yielding a
528
+ frequency 2.46147
529
+
530
+ T for K = 5 – exactly the PG gap found
531
+ in SCMFT as T → 0 and in agreement with the spin-dynamics
532
+ simulations (see Fig. 2).
533
+ While formulated for the Heisenberg-compass model, this
534
+ line of argument can be deployed to obtain the PG gap
535
+ for any spin model exhibiting ObTD. A proof of this state-
536
+ ment, following the strategy of Ref. [72], will be reported
537
+ elsewhere [74].
538
+ For type-I PG modes (ω ∝ |k|, as in
539
+ the Heisenberg-compass model) ∆ ∝
540
+
541
+ T, while for type-II
542
+ modes (ω ∝ |k|2), both Cφ, Cφ are O(T) and thus ∆ ∝ T.
543
+ Consequences of MWH divergence.— The ability to obtain
544
+ the PG gap from LSWT presents a puzzle: the perturbative
545
+ corrections δA0 and δB0 diverge logarithmically with system
546
+ size [57], just as in the MWH theorem [58, 59]. How then
547
+ do the curvatures of Feff avoid these singularities and give the
548
+ correct scaling? An analysis of the infrared divergences [60]
549
+ shows that while δA0 and δB0 are singular, δA0 + δB0, which
550
+ determines the leading contribution to the PG gap, is finite,
551
+ 0.00
552
+ 0.05
553
+ 0.10
554
+ 0.15
555
+ 0.20
556
+ 0.25
557
+ 0.30
558
+ T
559
+ −0.16
560
+ −0.15
561
+ −0.14
562
+ −0.13
563
+ −0.12
564
+ ∂M/∂T
565
+ 0.0
566
+ 0.1
567
+ 0.2
568
+ 0.3
569
+ T
570
+ 0.96
571
+ 0.98
572
+ 1.00
573
+ M
574
+ SCMFT
575
+ Monte Carlo
576
+ FIG. 3. Derivative of magnetization with respect to temperature,
577
+ ∂M/∂T, as a function of temperature for L = 60, K = 5 using MC
578
+ simulation and SCMFT. MC data is well-described by a fit motivated
579
+ by SCMFT [60], −0.09815 − 0.03563T + 0.01485 ln T. A similar
580
+ fit to SCMFT data yields −0.09631 − 0.01494T + 0.01491 ln T. The
581
+ inset shows M as a function of temperature for the same parameters.
582
+ MC error bars on M are smaller than the symbol size.
583
+ and reproduces the result from Eq. (6). However, divergences
584
+ in higher order terms do not cancel, and must be cured self-
585
+ consistently [60].
586
+ While these divergences are mostly benign for the PG gap,
587
+ they appear more dramatically in other quantities, like the
588
+ magnetization, M = 1 − 1
589
+ N
590
+
591
+ k⟨a†
592
+ kak⟩. Here, the thermal popu-
593
+ lation, ⟨a†
594
+ kak⟩ diverges in LSWT, rendering SCMFT necessary
595
+ to obtain meaningful results. In SCMFT, the PG gap provides
596
+ an infrared cutoff ℓ ∼ 1/∆ ∝ 1/
597
+
598
+ T, giving a logarithmic con-
599
+ tribution to M scaling as ∝ Tln T as T → 0 [60]. The presence
600
+ of this term can be diagnosed from ∂M/∂T, which exhibits a
601
+ logarithmic singularity as T → 0 for both the MC simulations
602
+ and SCMFT (see Fig. 3).
603
+ Outlook.— Our analysis of the PG gap will provide a deeper
604
+ understanding of real materials exhibiting ObD. The existence
605
+ of PG modes has been used to diagnose ObD, for example
606
+ in the compounds Fe2Ca3(GeO4)3 [34], Sr2Cu3O4Cl2 [35]
607
+ and Er2Ti2O7 [36, 41, 75].
608
+ In such materials, the ObQD
609
+ gap likely dominates the ObTD-induced gap discussed in this
610
+ work. However, in systems where the effect of ObQD is weak
611
+ or the degrees of freedom are sufficiently classical, ObTD
612
+ can resurface as the leading selection effect. For example,
613
+ our results may shed light on the rapidly growing family of
614
+ two-dimensional van der Waals (vdW) ferromagnets [76–78]
615
+ where the ObQD gap is expected to be small and thus the gap
616
+ induced by thermal fluctuations may be more significant. Ad-
617
+ ditionally, while reaching the classical thermal regime is chal-
618
+ lenging in magnetic materials (due to small spin length S ),
619
+ it may be more accessible in other platforms such as those
620
+ involving lattice vibrations [79, 80], dipole-coupled nanocon-
621
+ fined molecular rotors [81–84] or artificial mesoscale mag-
622
+ netic crystals [85–88]. Whether ObTD can be realized in such
623
+ topical systems, and how to detect the temperature dependent
624
+ PG gap, are open questions; our approach provides a theoret-
625
+
626
+ 5
627
+ ical framework and guidance for future experimental studies
628
+ in this promising area of research.
629
+ ACKNOWLEDGMENTS
630
+ We thank Itamar Aharony, Kristian Tyn Kai Chung, Alex
631
+ Hickey, Daniel Lozano-Gómez, and Darren Pereira for use-
632
+ ful discussions. We acknowledge the use of computational
633
+ resources provided by Digital Research Alliance of Canada.
634
+ This research was funded by the NSERC of Canada (MJPG,
635
+ JGR) and the Canada Research Chair Program (MJPG, Tier
636
+ I).
637
+ [1] P. W. Anderson, “Ordering and antiferromagnetism in ferrites,”
638
+ Phys. Rev. 102, 1008–1013 (1956).
639
+ [2] Jacques Villain, “Insulating spin glasses,” Zeitschrift für Physik
640
+ B Condensed Matter 33, 31–42 (1979).
641
+ [3] R. Moessner and J. T. Chalker, “Low-temperature properties of
642
+ classical geometrically frustrated antiferromagnets,” Phys. Rev.
643
+ B 58, 12049–12062 (1998).
644
+ [4] B. Canals and C. Lacroix, “Pyrochlore antiferromagnet: A
645
+ three-dimensional quantum spin liquid,” Phys. Rev. Lett. 80,
646
+ 2933–2936 (1998).
647
+ [5] Alexei Kitaev, “Anyons in an exactly solved model and be-
648
+ yond,” Annals of Physics 321, 2–111 (2006).
649
+ [6] M. J. P. Gingras and P. A. McClarty, “Quantum spin ice: a
650
+ search for gapless quantum spin liquids in pyrochlore magnets,”
651
+ Reports on Progress in Physics 77, 056501 (2014).
652
+ [7] Lucile Savary and Leon Balents, “Quantum spin liquids: a re-
653
+ view,” Reports on Progress in Physics 80, 016502 (2017).
654
+ [8] Takashi Imai and Young S. Lee, “Do quantum spin liquids ex-
655
+ ist?” Physics Today 69, 30–36 (2016).
656
+ [9] J. Knolle and R. Moessner, “A field guide to spin liquids,” An-
657
+ nual Review of Condensed Matter Physics 10, 451–472 (2019).
658
+ [10] Leon Balents, “Spin liquids in frustrated magnets,” Nature 464,
659
+ 199–208 (2010).
660
+ [11] Jinsheng Wen, Shun-Li Yu, Shiyan Li, Weiqiang Yu, and Jian-
661
+ Xin Li, “Experimental identification of quantum spin liquids,”
662
+ npj Quantum Materials 4, 12 (2019).
663
+ [12] C. Lacroix, P. Mendels,
664
+ and F. Mila (eds.), Introduction
665
+ to Frustrated Magnetism:
666
+ Materials, Experiments, Theory
667
+ (Springer Berlin, 2011).
668
+ [13] M. Udagawa and L. Jaubert (eds.), Spin Ice (Springer Cham,
669
+ 2021).
670
+ [14] Jason S. Gardner, Michel J. P. Gingras, and John E. Greedan,
671
+ “Magnetic pyrochlore oxides,” Rev. Mod. Phys. 82, 53–107
672
+ (2010).
673
+ [15] Alannah M. Hallas, Jonathan Gaudet, and Bruce D. Gaulin,
674
+ “Experimental insights into ground-state selection of quantum
675
+ XY pyrochlores,” Annual Review of Condensed Matter Physics
676
+ 9, 105–124 (2018).
677
+ [16] Jeffrey G. Rau and Michel J.P. Gingras, “Frustrated quantum
678
+ rare-earth pyrochlores,” Annual Review of Condensed Matter
679
+ Physics 10, 357–386 (2019).
680
+ [17] Simon Trebst and Ciarán Hickey, “Kitaev materials,” Physics
681
+ Reports 950, 1–37 (2022).
682
+ [18] H. Takagi, T. Takayama, G. Jackeli, G. Khaliullin, and S. E.
683
+ Nagler, “Concept and realization of Kitaev quantum spin liq-
684
+ uids,” Nature Reviews Physics 1, 264–280 (2019).
685
+ [19] Lucile Savary, Kate A. Ross, Bruce D. Gaulin, Jacob P. C. Ruff,
686
+ and Leon Balents, “Order by quantum disorder in Er2Ti2O7,”
687
+ Phys. Rev. Lett. 109, 167201 (2012).
688
+ [20] J. Villain, R. Bidaux, J.-P. Carton, and R. Conte, “Order as an
689
+ effect of disorder,” J. Phys. France 41, 1263–1272 (1980).
690
+ [21] E. F. Shender, “Antiferromagnetic garnets with fluctuationally
691
+ interacting sublattices,” Sov. Phys. JETP 56, 178 (1982).
692
+ [22] Christopher L. Henley, “Ordering due to disorder in a frus-
693
+ trated vector antiferromagnet,” Phys. Rev. Lett. 62, 2056–2059
694
+ (1989).
695
+ [23] Jack R. Tessman, “Magnetic anisotropy at 0°k,” Phys. Rev. 96,
696
+ 1192–1195 (1954).
697
+ [24] Sona Prakash and Christopher L. Henley, “Ordering due to dis-
698
+ order in dipolar magnets on two-dimensional lattices,” Phys.
699
+ Rev. B 42, 6574–6589 (1990).
700
+ [25] Kenn Kubo and Tatsuya Kishi, “Ordering due to quantum fluc-
701
+ tuations in the frustrated Heisenberg model,” Journal of the
702
+ Physical Society of Japan 60, 567–572 (1991).
703
+ [26] Andrey Chubukov, “Order from disorder in a kagomé antiferro-
704
+ magnet,” Phys. Rev. Lett. 69, 832–835 (1992).
705
+ [27] Jan N. Reimers and A. J. Berlinsky, “Order by disorder in the
706
+ classical Heisenberg kagomé antiferromagnet,” Phys. Rev. B
707
+ 48, 9539–9554 (1993).
708
+ [28] S. T. Bramwell, M. J. P. Gingras, and J. N. Reimers, “Order by
709
+ disorder in an anisotropic pyrochlore lattice antiferromagnet,”
710
+ J. Appl. Phys. 75, 5523–5525 (1994).
711
+ [29] Christopher L. Henley, “Selection by quantum fluctuations of
712
+ dipolar order in a diamond lattice,” Phys. Rev. Lett. 73, 2788–
713
+ 2788 (1994).
714
+ [30] J. D. M. Champion, M. J. Harris, P. C. W. Holdsworth, A. S.
715
+ Wills, G. Balakrishnan, S. T. Bramwell, E. ˇCižmár, T. Fen-
716
+ nell, J. S. Gardner, J. Lago, D. F. McMorrow, M. Orendáˇc,
717
+ A. Orendáˇcová, D. McK. Paul, R. I. Smith, M. T. F. Telling,
718
+ and A. Wildes, “Er2Ti2O7 : Evidence of quantum order by dis-
719
+ order in a frustrated antiferromagnet,” Phys. Rev. B 68, 020401
720
+ (2003).
721
+ [31] G. Baskaran, Diptiman Sen, and R. Shankar, “Spin-S Kitaev
722
+ model: Classical ground states, order from disorder, and exact
723
+ correlation functions,” Phys. Rev. B 78, 115116 (2008).
724
+ [32] Paul A. McClarty, Pawel Stasiak,
725
+ and Michel J. P. Gin-
726
+ gras, “Order-by-disorder in the XY pyrochlore antiferromag-
727
+ net,” Phys. Rev. B 89, 024425 (2014).
728
+ [33] Bimla Danu, Gautam Nambiar, and R. Ganesh, “Extended de-
729
+ generacy and order by disorder in the square lattice J1 − J2 − J3
730
+ model,” Phys. Rev. B 94, 094438 (2016).
731
+ [34] Th. Brueckel, B. Dorner, A. G. Gukasov, V. P. Plakhty,
732
+ W. Prandl, E. F. Shender, and O. P. Smirnow, “Dynamical in-
733
+ teraction of antiferromagnetic subsystems: a neutron scattering
734
+ study of the spinwave spectrum of the garnet Fe2Ca3(GeO4)3,”
735
+ Zeitschrift für Physik B Condensed Matter 72, 477–485 (1988).
736
+ [35] Y. J. Kim, A. Aharony, R. J. Birgeneau, F. C. Chou, O. Entin-
737
+ Wohlman, R. W. Erwin, M. Greven, A. B. Harris, M. A. Kast-
738
+
739
+ 6
740
+ ner, I. Ya. Korenblit, Y. S. Lee, and G. Shirane, “Ordering due
741
+ to quantum fluctuations in Sr2Cu3O4Cl2,” Phys. Rev. Lett. 83,
742
+ 852–855 (1999).
743
+ [36] K. A. Ross, Y. Qiu, J. R. D. Copley, H. A. Dabkowska, and
744
+ B. D. Gaulin, “Order by disorder spin wave gap in the XY
745
+ pyrochlore magnet Er2Ti2O7,” Phys. Rev. Lett. 112, 057201
746
+ (2014).
747
+ [37] C. L. Sarkis, J. G. Rau, L. D. Sanjeewa, M. Powell, J. Kolis,
748
+ J. Marbey, S. Hill, J. A. Rodriguez-Rivera, H. S. Nair, D. R.
749
+ Yahne, S. Säubert, M. J. P. Gingras, and K. A. Ross, “Unrav-
750
+ elling competing microscopic interactions at a phase boundary:
751
+ A single-crystal study of the metastable antiferromagnetic py-
752
+ rochlore Yb2Ge2O7,” Phys. Rev. B 102, 134418 (2020).
753
+ [38] N. W. Ashcroft and N. D. Mermin, Solid State Physics (Saun-
754
+ ders College Publishing, Fort Worth, 1976).
755
+ [39] Xiao-Gang Wen, Quantum Field Theory of Many-Body Sys-
756
+ tems: From the Origin of Sound to an Origin of Light and Elec-
757
+ trons (Oxford University Press, 2007).
758
+ [40] P. A. McClarty, S. H. Curnoe, and M. J. P. Gingras, “Energetic
759
+ selection of ordered states in a model of the Er2Ti2O7 frustrated
760
+ pyrochlore XY antiferromagnet,” Journal of Physics: Confer-
761
+ ence Series 145, 012032 (2009).
762
+ [41] Sylvain Petit, Julien Robert, Solène Guitteny, Pierre Bonville,
763
+ Claudia Decorse, Jacques Ollivier, Hannu Mutka, Michel J. P.
764
+ Gingras, and Isabelle Mirebeau, “Order by disorder or ener-
765
+ getic selection of the ground state in the XY pyrochlore anti-
766
+ ferromagnet Er2Ti2O7: An inelastic neutron scattering study,”
767
+ Phys. Rev. B 90, 060410 (2014).
768
+ [42] Jeffrey G. Rau, Sylvain Petit, and Michel J. P. Gingras, “Order
769
+ by virtual crystal field fluctuations in pyrochlore XY antiferro-
770
+ magnets,” Phys. Rev. B 93, 184408 (2016).
771
+ [43] The lower intensity at the band minimum compared to its vicin-
772
+ ity is a numerical artifact. Due to the finite frequency resolution
773
+ of the spin-dynamics simulations, the intensity maximum of the
774
+ PG mode may have fallen between frequency grid points, re-
775
+ sulting in an (apparent) lower intensity at the band minimum.
776
+ [44] Steven Weinberg, “Approximate symmetries and pseudo-
777
+ Goldstone bosons,” Phys. Rev. Lett. 29, 1698–1701 (1972).
778
+ [45] C.P. Burgess, “Goldstone and pseudo-Goldstone bosons in nu-
779
+ clear, particle and condensed-matter physics,” Physics Reports
780
+ 330, 193–261 (2000).
781
+ [46] Muneto Nitta and Daisuke A. Takahashi, “Quasi-Nambu-
782
+ Goldstone modes in nonrelativistic systems,” Phys. Rev. D 91,
783
+ 025018 (2015).
784
+ [47] Zohar Nussinov and Jeroen van den Brink, “Compass models:
785
+ Theory and physical motivations,” Rev. Mod. Phys. 87, 1–59
786
+ (2015).
787
+ [48] Julien Dorier, Federico Becca, and Frédéric Mila, “Quantum
788
+ compass model on the square lattice,” Phys. Rev. B 72, 024448
789
+ (2005).
790
+ [49] G. Jackeli and G. Khaliullin, “Mott insulators in the strong spin-
791
+ orbit coupling limit: From Heisenberg to a quantum compass
792
+ and Kitaev models,” Phys. Rev. Lett. 102, 017205 (2009).
793
+ [50] F. Trousselet, A. M. Ole´s,
794
+ and P. Horsch, “Compass-
795
+ Heisenberg model on the square lattice —spin order and ele-
796
+ mentary excitations,” Europhysics Letters 91, 40005 (2010).
797
+ [51] Fabien Trousselet, Andrzej M. Ole´s, and Peter Horsch, “Mag-
798
+ netic properties of nanoscale compass-Heisenberg planar clus-
799
+ ters,” Phys. Rev. B 86, 134412 (2012).
800
+ [52] S. Boseggia, R. Springell, H. C. Walker, H. M. Rønnow, Ch.
801
+ Rüegg, H. Okabe, M. Isobe, R. S. Perry, S. P. Collins, and D. F.
802
+ McMorrow, “Robustness of basal-plane antiferromagnetic or-
803
+ der and the Jeff=1/2 state in single-layer iridate spin-orbit Mott
804
+ insulators,” Phys. Rev. Lett. 110, 117207 (2013).
805
+ [53] Vamshi M. Katukuri, Viktor Yushankhai, Liudmila Siurakshina,
806
+ Jeroen van den Brink, Liviu Hozoi, and Ioannis Rousochatza-
807
+ kis, “Mechanism of basal-plane antiferromagnetism in the spin-
808
+ orbit driven iridate Ba2IrO4,” Phys. Rev. X 4, 021051 (2014).
809
+ [54] Artem A. Vladimirov, Dieter Ihle,
810
+ and Nikolay M. Plakida,
811
+ “Magnetic order in the two-dimensional compass-Heisenberg
812
+ model,” Eur. Phys. J. B 88, 148 (2015).
813
+ [55] Long Zhang, Fa Wang, and Dung-Hai Lee, “Compass impurity
814
+ model of Tb substitution in Sr2IrO4,” Phys. Rev. B 94, 161118
815
+ (2016).
816
+ [56] Haruki Watanabe, “Counting rules of Nambu–Goldstone
817
+ modes,” Annual Review of Condensed Matter Physics 11, 169–
818
+ 187 (2020).
819
+ [57] Sidney Coleman, “There are no Goldstone bosons in two di-
820
+ mensions,” Communications in Mathematical Physics 31, 259–
821
+ 264 (1973).
822
+ [58] N. D. Mermin and H. Wagner, “Absence of ferromagnetism
823
+ or antiferromagnetism in one- or two-dimensional isotropic
824
+ Heisenberg models,” Phys. Rev. Lett. 17, 1133–1136 (1966).
825
+ [59] P. C. Hohenberg, “Existence of long-range order in one and two
826
+ dimensions,” Phys. Rev. 158, 383–386 (1967).
827
+ [60] See Supplemental Material at (...) for details about Monte Carlo
828
+ and spin-dynamics simulations and finite size scaling analysis,
829
+ as well as details of the linear and non-linear spin wave theory,
830
+ including of cancellation of divergences in the PG gap calcula-
831
+ tion and derivation of logarithmic correction to the magnetiza-
832
+ tion. It also includes Refs. [62, 66, 67, 70–72, 89–97].
833
+ [61] L. Landau and E. Lifshitz, “3 - On the theory of the dispersion
834
+ of magnetic permeability in ferromagnetic bodies,” in Perspec-
835
+ tives in Theoretical Physics, edited by L. P. Pitaevksi (Perga-
836
+ mon, Amsterdam, 1992) pp. 51–65, reprinted from Physikalis-
837
+ che Zeitschrift der Sowjetunion 8, Part 2, 153, 1935.
838
+ [62] A. Auerbach, Interacting Electrons and Quantum Magnetism,
839
+ Graduate Texts in Contemporary Physics (Springer New York,
840
+ 1998).
841
+ [63] P. C. Martin, E. D. Siggia, and H. A. Rose, “Statistical dynam-
842
+ ics of classical systems,” Phys. Rev. A 8, 423–437 (1973).
843
+ [64] Uli Deker and Fritz Haake, “Fluctuation-dissipation theorems
844
+ for classical processes,” Phys. Rev. A 11, 2043–2056 (1975).
845
+ [65] C. P. Enz and L. Garrido, “Perturbation theory for classical ther-
846
+ modynamic Green’s functions,” Phys. Rev. A 14, 1258–1268
847
+ (1976).
848
+ [66] H. Bruus and K. Flensberg, Many-Body Quantum Theory in
849
+ Condensed Matter Physics: An Introduction, Oxford Graduate
850
+ Texts (Oxford University Press, Oxford, 2004).
851
+ [67] J.P. Blaizot and G. Ripka, Quantum Theory of Finite Systems
852
+ (MIT Press, Cambridge, 1986).
853
+ [68] G. D. Mahan, Many Particle Physics, Third Edition (Springer
854
+ New York, 2000).
855
+ [69] Mehran Kardar, Statistical Physics of Particles (Cambridge
856
+ University Press, 2007).
857
+ [70] P. D. Loly, “The Heisenberg ferromagnet in the selfconsistently
858
+ renormalized spin wave approximation,” Journal of Physics C:
859
+ Solid State Physics 4, 1365–1377 (1971).
860
+ [71] A. V. Chubukov, S. Sachdev, and T. Senthil, “Large-S expan-
861
+ sion for quantum antiferromagnets on a triangular lattice,” Jour-
862
+ nal of Physics: Condensed Matter 6, 8891–8902 (1994).
863
+ [72] Jeffrey G. Rau, Paul A. McClarty,
864
+ and Roderich Moess-
865
+ ner, “Pseudo-Goldstone gaps and order-by-quantum disorder in
866
+ frustrated magnets,” Phys. Rev. Lett. 121, 237201 (2018).
867
+ [73] H. Goldstein, C. P. Poole, and J. L. Safko, Classical Mechanics,
868
+ Third Edition (Pearson, 2002).
869
+ [74] Subhankar Khatua, Michel J. P. Gingras, and Jeffrey G. Rau,
870
+ (unpublished).
871
+
872
+ 7
873
+ [75] E. Lhotel, J. Robert, E. Ressouche, F. Damay, I. Mirebeau,
874
+ J. Ollivier, H. Mutka, P. Dalmas de Réotier, A. Yaouanc,
875
+ C. Marin, C. Decorse, and S. Petit, “Field-induced phase di-
876
+ agram of the XY pyrochlore antiferromagnet Er2Ti2O7,” Phys.
877
+ Rev. B 95, 134426 (2017).
878
+ [76] Cheng Gong, Lin Li, Zhenglu Li, Huiwen Ji, Alex Stern, Yang
879
+ Xia, Ting Cao, Wei Bao, Chenzhe Wang, Yuan Wang, Z. Q.
880
+ Qiu, R. J. Cava, Steven G. Louie, Jing Xia, and Xiang Zhang,
881
+ “Discovery of intrinsic ferromagnetism in two-dimensional van
882
+ der Waals crystals,” Nature 546, 265–269 (2017).
883
+ [77] Lebing Chen, Jae-Ho Chung, Matthew B. Stone, Alexander I.
884
+ Kolesnikov, Barry Winn, V. Ovidiu Garlea, Douglas L. Aber-
885
+ nathy, Bin Gao, Mathias Augustin, Elton J. G. Santos,
886
+ and
887
+ Pengcheng Dai, “Magnetic field effect on topological spin ex-
888
+ citations in CrI3,” Phys. Rev. X 11, 031047 (2021).
889
+ [78] S. E. Nikitin, B. Fåk, K. W. Krämer, T. Fennell, B. Normand,
890
+ A. M. Läuchli, and Ch. Rüegg, “Thermal evolution of Dirac
891
+ magnons in the honeycomb ferromagnet CrBr3,” Phys. Rev.
892
+ Lett. 129, 127201 (2022).
893
+ [79] Yilong Han, Yair Shokef, Ahmed M. Alsayed, Peter Yunker,
894
+ Tom C. Lubensky, and Arjun G. Yodh, “Geometric frustration
895
+ in buckled colloidal monolayers,” Nature 456, 898–903 (2008).
896
+ [80] Yair Shokef, Anton Souslov, and T. C. Lubensky, “Order by
897
+ disorder in the antiferromagnetic Ising model on an elastic tri-
898
+ angular lattice,” Proceedings of the National Academy of Sci-
899
+ ences 108, 11804–11809 (2011).
900
+ [81] Jerzy Cioslowski and Asiri Nanayakkara, “Endohedral ful-
901
+ lerites: A new class of ferroelectric materials,” Phys. Rev. Lett.
902
+ 69, 2871–2873 (1992).
903
+ [82] Shinobu Aoyagi, Norihisa Hoshino, Tomoyuki Akutagawa,
904
+ Yuki Sado, Ryo Kitaura, Hisanori Shinohara, Kunihisa Sugi-
905
+ moto, Rui Zhang, and Yasujiro Murata, “A cubic dipole lat-
906
+ tice of water molecules trapped inside carbon cages,” Chemical
907
+ Communications 50, 524–526 (2014).
908
+ [83] B. P. Gorshunov, V. I. Torgashev, E. S. Zhukova, V. G.
909
+ Thomas, M. A. Belyanchikov, C. Kadlec, F. Kadlec, M. Savi-
910
+ nov, T. Ostapchuk, J. Petzelt, J. Prokleška, P. V. Tomas, E. V.
911
+ Pestrjakov, D. A. Fursenko, G. S. Shakurov, A. S. Prokhorov,
912
+ V. S. Gorelik, L. S. Kadyrov, V. V. Uskov, R. K. Kremer, and
913
+ M. Dressel, “Incipient ferroelectricity of water molecules con-
914
+ fined to nano-channels of beryl,” Nature Communications 7,
915
+ 12842 (2016).
916
+ [84] M. A. Belyanchikov, M. Savinov, Z. V. Bedran, P. Bednyakov,
917
+ P. Proschek, J. Prokleska, V. A. Abalmasov, J. Petzelt, E. S.
918
+ Zhukova, V. G. Thomas, A. Dudka, A. Zhugayevych, A. S.
919
+ Prokhorov, V. B. Anzin, R. K. Kremer, J. K. H. Fischer,
920
+ P. Lunkenheimer, A. Loidl, E. Uykur, M. Dressel, and B. Gor-
921
+ shunov, “Dielectric ordering of water molecules arranged in a
922
+ dipolar lattice,” Nature Communications 11, 3927 (2020).
923
+ [85] Oleksandr V. Dobrovolskiy, Oleksandr V. Pylypovskyi, Luka
924
+ Skoric, Amalio Fernández-Pacheco, Arjen Van Den Berg, Sam
925
+ Ladak,
926
+ and Michael Huth, “Complex-shaped 3D nanoarchi-
927
+ tectures for magnetism and superconductivity,” in Curvilinear
928
+ Micromagnetism: From Fundamentals to Applications, edited
929
+ by Denys Makarov and Denis D. Sheka (Springer International
930
+ Publishing, Cham, 2022) pp. 215–268.
931
+ [86] Lukas Keller, Mohanad K. I. Al Mamoori, Jonathan Pieper,
932
+ Christian Gspan, Irina Stockem, Christian Schröder, Sven
933
+ Barth, Robert Winkler, Harald Plank, Merlin Pohlit, Jens
934
+ Müller, and Michael Huth, “Direct-write of free-form building
935
+ blocks for artificial magnetic 3D lattices,” Scientific Reports 8,
936
+ 6160 (2018).
937
+ [87] Andrew May, Matthew Hunt, Arjen Van Den Berg, Alaa He-
938
+ jazi, and Sam Ladak, “Realisation of a frustrated 3D magnetic
939
+ nanowire lattice,” Communications Physics 2, 13 (2019).
940
+ [88] Peter Fischer, Dédalo Sanz-Hernández, Robert Streubel, and
941
+ Amalio Fernández-Pacheco, “Launching a new dimension with
942
+ 3D magnetic nanostructures,” APL Materials 8, 010701 (2020).
943
+ [89] J. D. Alzate-Cardona, D. Sabogal-Suárez, R. F. L. Evans, and
944
+ E. Restrepo-Parra, “Optimal phase space sampling for Monte
945
+ Carlo simulations of Heisenberg spin systems,” Journal of
946
+ Physics: Condensed Matter 31, 095802 (2019).
947
+ [90] Michael Creutz, “Overrelaxation and Monte Carlo simulation,”
948
+ Phys. Rev. D 36, 515–519 (1987).
949
+ [91] M. E. J. Newman and G. T. Barkema, Monte Carlo methods in
950
+ statistical physics (Clarendon Press, Oxford, 1999).
951
+ [92] J. R. Dormand and P. J. Prince, “A family of embedded Runge-
952
+ Kutta formulae,” Journal of Computational and Applied Math-
953
+ ematics 6, 19–26 (1980).
954
+ [93] K. Ahnert and M. Mulansky, “Odeint – solving ordinary differ-
955
+ ential equations in C++,” AIP Conference Proceedings 1389,
956
+ 1586–1589 (2011).
957
+ [94] K. Ahnert and M. Mulansky, “Boost C++ Library: Odeint,”
958
+ (2012).
959
+ [95] J. C. Bowman and M. Roberts, “FFTW++: A fast Fourier trans-
960
+ form C++ header class for the FFTW3 library,” (2010).
961
+ [96] M. E. Zhitomirsky and A. L. Chernyshev, “Colloquium: Spon-
962
+ taneous magnon decays,” Rev. Mod. Phys. 85, 219–242 (2013).
963
+ [97] P. M. Chaikin and T. C. Lubensky, Principles of Condensed
964
+ Matter Physics (Cambridge University Press, 1995).
965
+
966
+ 8
967
+ Supplemental Material for “Pseudo-Goldstone modes and dynamical gap generation from
968
+ order-by-thermal-disorder”
969
+ Subhankar Khatua,1, 2 Michel J. P. Gingras,2 and Jeffrey G. Rau1
970
+ 1Department of Physics, University of Windsor, 401 Sunset Avenue, Windsor, Ontario, N9B 3P4, Canada
971
+ 2Department of Physics and Astronomy, University of Waterloo, Waterloo, Ontario, N2L 3G1, Canada
972
+ (Dated: January 27, 2023)
973
+ I.
974
+ DETAILS OF MONTE CARLO SIMULATIONS
975
+ A.
976
+ Details of Monte Carlo procedure
977
+ The Monte Carlo (MC) simulations described in the main
978
+ text are based on adaptive single-site Metropolis moves [1],
979
+ combined with over-relaxation moves [2]. A typical single-
980
+ site Metropolis move involves randomly selecting a spin and
981
+ changing its orientation to a random direction.
982
+ However,
983
+ at low temperature, most such moves result in configura-
984
+ tions that are of much higher energies and thus rejected [3].
985
+ Therefore, we follow an adaptive approach that selects a
986
+ spin randomly and changes its orientation to a Gaussian
987
+ distributed random direction within a solid-angle of certain
988
+ width. The solid-angle-width changes adaptively to ensure
989
+ that the update-acceptance rate remains close to 50% at each
990
+ temperature (see Ref. [1] for details).
991
+ The over-relaxation
992
+ move rotates a randomly selected spin by a random angle
993
+ about its local exchange field. This move is energy-conserving
994
+ and thus always accepted. We define a Monte Carlo sweep
995
+ at a certain temperature as a combination of N (total num-
996
+ ber of spins) random successive adaptive single-site Metropo-
997
+ lis moves with each followed by five random over-relaxation
998
+ moves. All the simulation results discussed in the main text
999
+ have been obtained by considering periodic boundary condi-
1000
+ tions on the square lattice of size L by L and N = L2 spins. As
1001
+ in the main text, we use units such that J = ℏ = kB = 1.
1002
+ B.
1003
+ Simulation details of the order parameter distribution
1004
+ Starting from a random spin configuration at high tempera-
1005
+ ture, T = 10 (larger than K) where the system is in the param-
1006
+ agnetic phase, we slowly cool down in steps of size δT = 0.1
1007
+ to a final temperature T = 0.4 (much smaller than K). At
1008
+ each temperature, we perform 105 MC sweeps to equilibrate
1009
+ the system. Finally, at T = 0.4, after equilibration, we record
1010
+ the net magnetization-per-spin over 106 MC samples, leaving
1011
+ five MC sweeps in between two consecutive measurements.
1012
+ From the net magnetization per spin, M = (Mx, My, Mz), we
1013
+ calculate ϕ = arctan(My/Mx), computing a distribution for ϕ.
1014
+ Since the ferromagnetic Heisenberg-compass model has a C4
1015
+ rotation symmetry in the ˆx− ˆy plane, we symmetrize the distri-
1016
+ bution by shifting the data by π/2, π, and 3π/2 , i.e., add π/2,
1017
+ π, and 3π/2 to each entry of the dataset. We have plotted the
1018
+ final dataset as a probability density, P(ϕ) for ϕ ∈ [−π/4, π/4]
1019
+ with 50 bins for three different system sizes, N = 62, 102,
1020
+ and 142 in Fig. 1(a) in the main text. We have chosen a large
1021
+ value for K, i.e., K = 5, for all simulations in order to obtain
1022
+ a strong selection effect at accessible system sizes. For the
1023
+ gross spectral features, the largest system size considered for
1024
+ spin-dynamics simulations was N = 1002, while for detailed
1025
+ features, such as the temperature dependence of the pseudo-
1026
+ Goldstone (PG) gap, up to N = 402 was used. Had smaller
1027
+ values of K been used, all the MC simulations, as well as
1028
+ spin-dynamics simulations, would have had to be performed
1029
+ for much larger system sizes to obtain results that converge
1030
+ when system size is extrapolated to the thermodynamic limit
1031
+ (N → ∞).
1032
+ C.
1033
+ Simulation details of magnetization and its derivative with
1034
+ respect to temperature
1035
+ Independently at each temperature T, 5 × 105 MC sweeps
1036
+ are performed on a perfectly aligned ferromagnetic spin con-
1037
+ figuration along ˆx for equilibration, followed by 3 × 106 suc-
1038
+ cessive MC sweeps to measure the net magnetization along
1039
+ ˆx (M), energy (E), and their product (EM). Their product is
1040
+ recorded in order to calculate the derivative of the magnetiza-
1041
+ tion with respect to temperature, given by
1042
+ ∂M
1043
+ ∂T ≡ ⟨EM⟩ − ⟨E⟩⟨M⟩
1044
+ T 2
1045
+ ,
1046
+ (S1)
1047
+ where ⟨x⟩ is the MC thermal average of quantity x. To es-
1048
+ timate the statistical errors on static quantities, the 3 × 106
1049
+ measurements are divided into 30 blocks, and then resampled
1050
+ using the standard bootstrap method [3]. Typically, O(103)
1051
+ bootstrap samples were generated from these blocks to esti-
1052
+ mate the statistical errors. In Fig. 3 of the main text, the error
1053
+ bars shown correspond to twice the standard deviation esti-
1054
+ mated via bootstrap.
1055
+ II.
1056
+ DETAILS OF SPIN-DYNAMICS SIMULATIONS
1057
+ Numerical integrations of the Landau-Lifshitz equations
1058
+ have been done using an adaptive step size RK5(4) Dormand-
1059
+ Prince integrator [4] from the Boost-Odeint C++ library [5,
1060
+ 6]. The initial spin configurations for the numerical integra-
1061
+ tion are generated from MC simulations described in Sec. I.
1062
+ To obtain the results shown in Fig. 1(b) in the main text,
1063
+ we perform 5 × 105 equilibration MC sweeps on a perfectly
1064
+ aligned ferromagnetic configuration along ˆx at T = 0.4. Start-
1065
+ ing from the final state, we perform another 15 × 103 MC
1066
+ sweeps for 350 independent parallel runs to generate well-
1067
+ equilibrated configurations at T = 0.4. Next, we feed each
1068
+
1069
+ 9
1070
+ 2
1071
+ 0.000
1072
+ 0.001
1073
+ 0.002
1074
+ 0.003
1075
+ 0.004
1076
+ 1/L2
1077
+ 0.20
1078
+ 0.25
1079
+ 0.30
1080
+ 0.35
1081
+ 0.40
1082
+ 0.45
1083
+ 0.50
1084
+
1085
+ T = 0.040
1086
+ T = 0.032
1087
+ T = 0.024
1088
+ T = 0.016
1089
+ T = 0.008
1090
+ FIG. S1.
1091
+ Finite size scaling of the PG gap obtained from spin-
1092
+ dynamics simulations for several temperatures (K = 5).
1093
+ 0.0000 0.0002 0.0004 0.0006 0.0008 0.0010 0.0012
1094
+ 1/L2
1095
+ 0.20
1096
+ 0.25
1097
+ 0.30
1098
+ 0.35
1099
+ 0.40
1100
+ 0.45
1101
+ 0.50
1102
+
1103
+ T = 0.040
1104
+ T = 0.032
1105
+ T = 0.024
1106
+ T = 0.016
1107
+ T = 0.008
1108
+ FIG. S2. Finite size scaling of the PG gap obtained using SCMFT
1109
+ for several temperatures (K = 5).
1110
+ of these 350 configurations into the Dormand-Prince integra-
1111
+ tor as an initial state and integrate to a final time, tmax = 50.
1112
+ The error tolerance of the integrator is set to 10−8, such that
1113
+ the energy-per-spin and individual spin lengths are conserved
1114
+ to at least one part in 107 and 1010, respectively. In each of
1115
+ these independent 350 integrations, we calculate the Fourier
1116
+ transform of the spin configurations in space and time, S(k, ω)
1117
+ using FFTW++ [7] and then compute the dynamical structure
1118
+ 0.00
1119
+ 0.01
1120
+ 0.02
1121
+ 0.03
1122
+ 0.04
1123
+ 0.05
1124
+ 1/L
1125
+ 0.0
1126
+ 0.1
1127
+ 0.2
1128
+ 0.3
1129
+ 0.4
1130
+ 0.5
1131
+ Γ
1132
+ T = 1.000
1133
+ T = 0.800
1134
+ T = 0.600
1135
+ T = 0.400
1136
+ T = 0.200
1137
+ T = 0.016
1138
+ FIG. S3. Finite size scaling of the PG linewidth obtained from spin-
1139
+ dynamics simulations for several temperatures (K = 5).
1140
+ factor, S(k, ω) = |S(k, ω)|2, finally taking an average of the
1141
+ structure factors found from the 350 initial configurations to
1142
+ obtain Fig. 1(b).
1143
+ The results in Fig. 1(c) in the main text are obtained
1144
+ as follows: The system is initialized in a perfectly aligned
1145
+ ferromagnetic configuration along ˆx at T
1146
+ = 0.0004, and
1147
+ slowly warmed up in steps of δT = 0.0004 to a temper-
1148
+ ature T
1149
+ = 0.04.
1150
+ At each temperature, we perform 105
1151
+ equilibration MC sweeps, generating a configuration at T =
1152
+ 0.008, 0.016, 0.024, 0.032, 0.04.
1153
+ At each of these tempera-
1154
+ tures, we then apply the following procedure. Starting from
1155
+ a given spin configuration, say at T = 0.008, we generate
1156
+ a total of 2 × 103 configurations independently by perform-
1157
+ ing 105 MC sweeps. Each of these configurations is fed into
1158
+ the Dormand-Prince integrator independently to integrate to
1159
+ a final time, tmax = 2500. Note that tmax here is taken to be
1160
+ much larger than the tmax = 50 value used to obtain the re-
1161
+ sults shown in Fig. 1(b). As discussed in the main text, to
1162
+ determine the PG gap, ∆, and linewidth, Γ, a much higher
1163
+ frequency resolution is needed and thus the total integration
1164
+ time must be larger. The error tolerance of the integrator is
1165
+ set to 10−10, such that the energy-per-spin and spin-length are
1166
+ conserved to at least one part in 108 and 1010, respectively.
1167
+ After the time evolution, we compute the Fourier transform of
1168
+ the spin configurations in space and time using FFTW++ and
1169
+ then compute the dynamical structure factor, S(k, ω). Finally,
1170
+ we perform an average over the 2 × 103 initial spin config-
1171
+ urations to obtain the average dynamical structure factor at
1172
+ T = 0.008. In Fig. 1(c), we show only a cut of the average dy-
1173
+ namical structure factor at the zone center, S(0, ω). To clearly
1174
+ visualize S(0, ω) at several different temperatures in a single
1175
+ plot, we stagger them on the y-axis with a constant spacing
1176
+ between the S(0, ω) data at two consecutive temperatures.
1177
+ To obtain the results shown in Fig. 2 of the main text,
1178
+ we proceed as follows: At each temperature, we follow the
1179
+ same method as described for Fig. 1(c) in the previous para-
1180
+ graph and compute S(0, ω) for several system sizes, L =
1181
+ 20, 24, 28, 32, 36, and 40. To find the gap and linewidth for
1182
+ each system size, we fit each data to a Gaussian (a Gaussian
1183
+ lineshape fits the data in the range T ≤ 0.04 best, compared to,
1184
+ e.g., a Lorentzian). The center of the Gaussian is used to de-
1185
+ fine the PG gap and the full-width at half maximum (FWHM)
1186
+ of the Gaussian, i.e., 2.355σ (standard deviation), is taken as
1187
+ the PG linewidth. Then, finite size L-dependent PG gaps and
1188
+ linewidths are then extrapolated in system size (L → ∞) to
1189
+ obtain the corresponding values in the thermodynamic limit.
1190
+ Finite size scaling of the PG gap is shown in Fig. S1. The finite
1191
+ size scaling of the PG gap obtained using the self-consistent
1192
+ mean-field theory (SCMFT) is shown in Fig. S2 (See Sec. III E
1193
+ for details).
1194
+ At very low temperatures, e.g., T ≤ 0.04, where S(0, ω)
1195
+ falls very sharply away from the center of the peak, a Gaus-
1196
+ sian lineshape is a natural choice. However, as temperature
1197
+ increases further, S(0, ω) shows more pronounced tails and a
1198
+ Lorentzian lineshape was found to provide a better descrip-
1199
+ tion of the data. Finite size scaling of the PG linewidth is
1200
+ shown in Fig. S3. At T = 0.016, the PG linewidths for differ-
1201
+ ent system sizes are found by fitting to a Gaussian while for
1202
+
1203
+ 10
1204
+ 3
1205
+ the remaining temperatures in Fig. S3, the PG linewidths are
1206
+ found from fitting to Lorentzian (via the FWHM of the corre-
1207
+ sponding Lorentzian). Finite size scaling reveals that at very
1208
+ low temperatures, the PG linewidth scales almost linearly with
1209
+ 1/L with the scaling becoming quadratic in 1/L as tempera-
1210
+ ture increases (see Fig. S3).
1211
+ III.
1212
+ SPIN WAVE THEORY
1213
+ Here, we elaborate on the formalism for interacting spin
1214
+ waves in the ferromagnetic Heisenberg-compass model on the
1215
+ square lattice. We consider the Heisenberg-compass model
1216
+ Hamiltonian
1217
+ H = −
1218
+
1219
+
1220
+
1221
+ JSr · Sr+δ + KS δ
1222
+ rS δ
1223
+ r+δ
1224
+
1225
+
1226
+
1227
+
1228
+ S⊺
1229
+ r JδSr+δ,
1230
+ (S2)
1231
+ where δ = ˆx, ˆy denotes the nearest-neighbour (horizontal and
1232
+ vertical) bond directions. For J > 0 and K > 0, the classical
1233
+ ground state is ferromagnetic and has an accidental degener-
1234
+ acy parametrized by an angle ϕ
1235
+ Sr = S (cos ϕ ˆx + sin ϕ ˆy).
1236
+ (S3)
1237
+ For small |K| and K < 0, one finds only a (symmetry-
1238
+ enforced) discrete degeneracy, with Sr = ±S ˆz. For large |K|
1239
+ and K < 0, the ground state is described by an XY-stripe phase
1240
+ parametrized by a single angle whose two extreme limits are
1241
+ X-stripe phase (i.e., all spins are lying along the ˆx axis, ar-
1242
+ ranging themselves antiferromagnetically along the ˆx axis and
1243
+ ferromagnetically along the ˆy axis) and Y-stripe phase (i.e., all
1244
+ spins are lying along the ˆy axis, arranging themselves antifer-
1245
+ romagnetically along the ˆy axis and ferromagnetically along
1246
+ the ˆx axis). The phases for J < 0 can be obtained by map-
1247
+ ping Sr → (−1)rSr which alternates on the two sublattices.
1248
+ We note that the dynamics however differ between J > 0 and
1249
+ J < 0, since the sign change on one sublattice is not a canoni-
1250
+ cal transformation.
1251
+ Returning to the J > 0, K > 0 case, we define a frame
1252
+ aligned with the ground state with angle ϕ
1253
+ ˆex = − sin ϕ ˆx + cos ϕ ˆy,
1254
+ ˆey = ˆz,
1255
+ ˆez = cos ϕ ˆx + sin ϕ ˆy,
1256
+ as well as ˆe± ≡ (ˆex ± iˆey)/
1257
+
1258
+ 2 and ˆe0 ≡
1259
+ ˆez.
1260
+ We then
1261
+ have the local exchanges Jµν
1262
+ δ
1263
+ = ˆe⊺
1264
+ µ Jδˆeν. The Fourier trans-
1265
+ forms of the exchange matrices, Jµν
1266
+ δ , are defined as J k ≡
1267
+
1268
+ δ 2 cos (k · δ)J δ where the fact that −δ and δ are equivalent
1269
+ has been used. Explicitly, these are given by
1270
+ J+−
1271
+ k
1272
+ = −
1273
+
1274
+ 2J + Ksin2ϕ
1275
+
1276
+ cos kx −
1277
+
1278
+ 2J + Kcos2ϕ
1279
+
1280
+ cos ky,
1281
+ J00
1282
+ k = −2
1283
+
1284
+ J + Kcos2ϕ
1285
+
1286
+ cos kx − 2
1287
+
1288
+ J + Ksin2ϕ
1289
+
1290
+ cos ky,
1291
+ J++
1292
+ k
1293
+ = −
1294
+
1295
+ Ksin2ϕ
1296
+
1297
+ cos kx −
1298
+
1299
+ Kcos2ϕ
1300
+
1301
+ cos ky,
1302
+ J0±
1303
+ k
1304
+ = − K
1305
+
1306
+ 2
1307
+ sin (2ϕ)
1308
+
1309
+ cos ky − cos kx
1310
+
1311
+ ,
1312
+ with J−+
1313
+ k
1314
+ = [J+−
1315
+ k ]∗, J−−
1316
+ k
1317
+ = [J++
1318
+ k ]∗ and J0±
1319
+ k
1320
+ = J±0
1321
+ k . Note
1322
+ that J00
1323
+ 0
1324
+ = −2(2J + K). For one of the four ground states
1325
+ selected by order-by-thermal-disorder (ObTD), e.g. ϕ = 0,
1326
+ these Jµν
1327
+ k are given by
1328
+ J+−
1329
+ k
1330
+ = −2J cos kx − (2J + K) cos ky,
1331
+ J00
1332
+ k = −2 (J + K) cos kx − 2J cos ky,
1333
+ J++
1334
+ k
1335
+ = −K cos ky,
1336
+ where J0±
1337
+ k
1338
+ = 0. Performing the usual Holstein-Primakoff
1339
+ expansion [8] to O(S 0) on this model yields [9]
1340
+ H ≈ E0 + H2 + �H4,2−2 + H4,3−1 + H4,1−3
1341
+ � + · · · ,
1342
+ (S4)
1343
+ where we have defined the constant classical part E0
1344
+ =
1345
+ −NS 2(2J + K) [at O(S 2)] and
1346
+ H2 =
1347
+
1348
+ k
1349
+
1350
+ Aka†
1351
+ kak + 1
1352
+ 2!
1353
+
1354
+ Bka†
1355
+ ka†
1356
+ −k + B∗
1357
+ ka−kak
1358
+ ��
1359
+ ,
1360
+ (S5a)
1361
+ H4,2−2 = 1
1362
+ N
1363
+
1364
+ kk′q
1365
+ 1
1366
+ (2!)2 Vk,k′,qa†
1367
+ k+qa†
1368
+ k′−qak′ak,
1369
+ (S5b)
1370
+ H4,3−1 = 1
1371
+ N
1372
+
1373
+ kk′q
1374
+ 1
1375
+ 3!Dk,k′,qa†
1376
+ ka†
1377
+ k′a†
1378
+ qak+k′+q = H†
1379
+ 4,1−3.
1380
+ (S5c)
1381
+ This incldues the quadratic parts [at O(S )] in H2 as well as
1382
+ the quartic parts [at O(S 0)] in H4 ≡ H4,2−2 + H4,3−1 + H4,1−3.
1383
+ The quartic part has been decomposed into a 2 − 2 scattering
1384
+ term, H4,2−2, and anomalous 3−1 and 1−3 terms, H4,3−1 and
1385
+ H4,1−3. Since J0,±
1386
+ k
1387
+ = 0, there are no three boson terms in H
1388
+ [Eq. (S4)]. In terms of the local exchanges, the coefficents in
1389
+ H2 and H4 are given explicitly by
1390
+ Ak = S
1391
+
1392
+ J+−
1393
+ k
1394
+ − J00
1395
+ 0
1396
+
1397
+ ,
1398
+ Bk = S J++
1399
+ k ,
1400
+ Vk,k′,q = 1
1401
+ 2
1402
+
1403
+ J00
1404
+ k′−k−q + J00
1405
+ −q + J00
1406
+ +q + J00
1407
+ k−k′+q
1408
+
1409
+ − 1
1410
+ 2
1411
+
1412
+ J+−
1413
+ k
1414
+ + J+−
1415
+ k′ + J+−
1416
+ k′−q + J+−
1417
+ k+q
1418
+
1419
+ ,
1420
+ Dk,k′,q = −1
1421
+ 2
1422
+
1423
+ J++
1424
+ k
1425
+ + J++
1426
+ k′ + J++
1427
+ q
1428
+
1429
+ .
1430
+ By construction, these coefficients must satisfy the symmetry
1431
+ relations
1432
+ Ak = A∗
1433
+ k,
1434
+ Bk = B−k,
1435
+ Vk,k′,q = Vk′,k,−q = Vk,k′,k′−k−q = Vk′,k,k−k′+q = V∗
1436
+ k+q,k′−q,−q,
1437
+ Dk,k′,q = Dk,q,k′ = Dk′,k,q = Dk′,q,k = Dq,k,k′ = Dq,k′,k.
1438
+ A.
1439
+ Non-Interacting Spin-Waves
1440
+ Consider first only the quadratic (non-interacting magnon)
1441
+ portion of H,
1442
+ H2 =
1443
+
1444
+ k
1445
+
1446
+ Aka†
1447
+ kak + 1
1448
+ 2!
1449
+
1450
+ Bka†
1451
+ ka†
1452
+ −k + H.c.
1453
+ ��
1454
+ .
1455
+ (S6)
1456
+
1457
+ 11
1458
+ 4
1459
+ This can be diagonalized by the usual Bogoliubov transforma-
1460
+ tion [8]. Defining the matrix
1461
+ Mk ≡
1462
+
1463
+ Ak Bk
1464
+ B∗
1465
+ k Ak
1466
+
1467
+ ,
1468
+ (S7)
1469
+ the spin-wave spectrum is obtained by diagonalization of
1470
+ σzMk, where σz is a (block) Pauli matrix. One finds the pos-
1471
+ itive frequency mode
1472
+ ωk =
1473
+
1474
+ A2
1475
+ k − |Bk|2 > 0.
1476
+ For the ferromagnetic Heisenberg-compass model, Ak and Bk
1477
+ are given by
1478
+ Ak = −S
1479
+ ��
1480
+ 2J + Ksin2ϕ
1481
+
1482
+ cos kx +
1483
+
1484
+ 2J + Kcos2ϕ
1485
+
1486
+ cos ky −
1487
+ 2(2J + K)
1488
+
1489
+ Bk = −S
1490
+ ��
1491
+ Ksin2ϕ
1492
+
1493
+ cos kx +
1494
+
1495
+ Kcos2ϕ
1496
+
1497
+ cos ky
1498
+
1499
+ .
1500
+ Note that A0 = KS and B0 = −KS , yielding a zero energy
1501
+ mode at k = 0, with ω0 = 0 and with both Ak and Bk real.
1502
+ The eigenvector of σzMk associated with the positive mode
1503
+ can be written as (uk, vk) where
1504
+ uk =
1505
+
1506
+ ωk + Ak
1507
+ 2ωk
1508
+ ,
1509
+ vk = −
1510
+ Bk
1511
+ √2ωk(ωk + Ak)
1512
+ ,
1513
+ which we have defined so that u2
1514
+ k − v2
1515
+ k = 1. Note that since
1516
+ both Ak and Bk are inversion even, we have u−k = uk, v−k = vk
1517
+ and ωk = ω−k. Since both Ak and Bk are real, we find that uk
1518
+ and vk are real as well. The diagonalized boson operators are
1519
+ defined via
1520
+ ak = ukγk + vkγ†
1521
+ −k,
1522
+ a†
1523
+ k = vkγ−k + ukγ†
1524
+ k.
1525
+ Expectation values of bilinears of these bosons can be written
1526
+ in terms of uk and vk. Noting that at temperature T these are
1527
+ ⟨γ†
1528
+ kγk⟩ = nB(ωk),
1529
+ ⟨γkγ†
1530
+ k⟩ = 1 + nB(ωk),
1531
+ where nB(ω) = [exp(ω/T)−1]−1 is the boson thermal occupa-
1532
+ tion number. The above thermal expectations for the original
1533
+ a-bosons are given by
1534
+ ⟨a†
1535
+ kak⟩ = nB(ωk)u2
1536
+ k + [1 + nB(ωk)] v2
1537
+ k,
1538
+ ⟨aka−k⟩ = ⟨a†
1539
+ −ka†
1540
+ k⟩
1541
+ ∗ = [1 + 2nB(ωk)] ukvk.
1542
+ In the classical limit, where T ≫ ωk, we have nB(ωk) ≈
1543
+ T/ωk ≫ 1. The expectations then become
1544
+ ⟨a†
1545
+ kak⟩ = T
1546
+ ωk
1547
+
1548
+ u2
1549
+ k + v2
1550
+ k
1551
+
1552
+ = T
1553
+ ωk
1554
+ � Ak
1555
+ ωk
1556
+
1557
+ ,
1558
+ (S8a)
1559
+ ⟨aka−k⟩ = ⟨a†
1560
+ −ka†
1561
+ k⟩ = 2T
1562
+ ωk
1563
+ ukvk = − T
1564
+ ωk
1565
+ � Bk
1566
+ ωk
1567
+
1568
+ .
1569
+ (S8b)
1570
+ Finally, the ordered moment (selected by ObTD), M ≡
1571
+ 1
1572
+ N
1573
+
1574
+ r⟨Sr⟩ ≡ M ˆx, can be expressed in terms of these boson
1575
+ averages as
1576
+ M = S − 1
1577
+ N
1578
+
1579
+ k
1580
+ ⟨a†
1581
+ kak⟩ ≡ S
1582
+ 1 − T
1583
+ S N
1584
+
1585
+ k
1586
+ Ak
1587
+ ω2
1588
+ k
1589
+  .
1590
+ (S9)
1591
+ B.
1592
+ Interacting Spin-Waves
1593
+ To consider the effects of the quartic parts of H in Eq. S4,
1594
+ H4,2−2, H4,3−1 and H4,1−3, we adopt a mean-field like ap-
1595
+ proach, replacing each possible contraction of operators with
1596
+ averages with respect to the quadratic, or “free” part, H2 [10,
1597
+ 11]. This procedure is equivalent to leading order perturbation
1598
+ theory in the interactions [12, 13]. For example, consider the
1599
+ scattering term
1600
+ a†
1601
+ k+qa†
1602
+ k′−qak′ak ≈ ⟨a†
1603
+ k+qak′⟩a†
1604
+ k′−qak + ⟨a†
1605
+ k′−qak⟩a†
1606
+ k+qak′
1607
+ + ⟨a†
1608
+ k+qak⟩a†
1609
+ k′−qak′ + ⟨a†
1610
+ k′−qak′⟩a†
1611
+ k+qak
1612
+ + ⟨a†
1613
+ k+qa†
1614
+ k′−q⟩ak′ak + ⟨ak′ak⟩a†
1615
+ k+qa†
1616
+ k′−q.
1617
+ Using that the expectation values satisfy ⟨a†
1618
+ kak′⟩ ∝ δk,k′ and
1619
+ ⟨akak′⟩ ∝ δk,−k′ one finds
1620
+ a†
1621
+ k+qa†
1622
+ k′−qak′ak ≈
1623
+
1624
+ δq,0 + δk+q,k′
1625
+ � �
1626
+ ⟨a†
1627
+ k′ak′⟩a†
1628
+ kak + ⟨a†
1629
+ kak⟩a†
1630
+ k′ak′
1631
+
1632
+ + δk,−k′
1633
+
1634
+ ⟨a†
1635
+ k+qa†
1636
+ −k−q⟩a−kak + ⟨a−kak⟩a†
1637
+ k+qa†
1638
+ −k−q
1639
+
1640
+ .
1641
+ Combing this decomposition with the interaction vertex, as
1642
+ specified in Eq. (S5b), gives the expression
1643
+ H4,2−2 ≈
1644
+
1645
+ k
1646
+ 
1647
+ 1
1648
+ N
1649
+
1650
+ q
1651
+ Vk,q,0⟨a†
1652
+ qaq⟩
1653
+  a†
1654
+ kak
1655
+ +1
1656
+ 2
1657
+
1658
+ k
1659
+ 
1660
+ 
1661
+ 1
1662
+ 2N
1663
+
1664
+ q
1665
+ Vq,−q,k−q⟨aqa−q⟩
1666
+  a†
1667
+ ka†
1668
+ −k + H.c.
1669
+  ,
1670
+ where Vk,k′,k′−k = Vk,k′,0 and Vk,k′,0 = Vk′,k,0 has been used
1671
+ to simplify the normal term, and shifting the momentum has
1672
+ been used to simplify the anomalous terms. The quartic terms
1673
+ thus appear as corrections to the Ak and Bk quadratic terms.
1674
+ Next, consider the same manipulations for the anomalous
1675
+ boson terms, starting with
1676
+ a†
1677
+ ka†
1678
+ k′a†
1679
+ qak+k′+q ≈ ⟨a†
1680
+ ka†
1681
+ k′⟩a†
1682
+ qak+k′+q + a†
1683
+ ka†
1684
+ k′⟨a†
1685
+ qak+k′+q⟩
1686
+ + ⟨a†
1687
+ ka†
1688
+ q⟩a†
1689
+ k′ak+k′+q + a†
1690
+ ka†
1691
+ q⟨a†
1692
+ k′ak+k′+q⟩
1693
+ + ⟨a†
1694
+ kak+k′+q⟩a†
1695
+ k′a†
1696
+ q + a†
1697
+ kak+k′+q⟨a†
1698
+ k′a†
1699
+ q⟩.
1700
+ Using the fact that the expectations in this last equation are
1701
+ diagonal in k (or skew-diagonal) [as in Eq. (S8)], we find
1702
+ a†
1703
+ ka†
1704
+ k′a†
1705
+ qak+k′+q ≈ δk,−k′
1706
+
1707
+ ⟨a†
1708
+ ka†
1709
+ −k⟩a†
1710
+ qaq + a†
1711
+ ka†
1712
+ −k⟨a†
1713
+ qaq⟩
1714
+
1715
+ + δk,−q
1716
+
1717
+ ⟨a†
1718
+ ka†
1719
+ −k⟩a†
1720
+ k′ak′ + a†
1721
+ ka†
1722
+ −k⟨a†
1723
+ k′ak′⟩
1724
+
1725
+ + δk′,−q
1726
+
1727
+ ⟨a†
1728
+ kak⟩a†
1729
+ k′a†
1730
+ −k′ + a†
1731
+ kak⟨a†
1732
+ k′a†
1733
+ −k′⟩
1734
+
1735
+ .
1736
+ Combining this decomposition with the anomalous interac-
1737
+ tion vertex, Dk,k′,q from Eq. (S5c), and using the permutation
1738
+ symmetry of its arguments, we find
1739
+ H4,3−1 ≈
1740
+
1741
+ k
1742
+ 
1743
+ 1
1744
+ 2N
1745
+
1746
+ q
1747
+ Dq,−q,k⟨a†
1748
+ qa†
1749
+ −q⟩
1750
+  a†
1751
+ kak
1752
+ +1
1753
+ 2
1754
+
1755
+ k
1756
+ 
1757
+ 1
1758
+ N
1759
+
1760
+ q
1761
+ Dk,−k,q⟨a†
1762
+ qaq⟩
1763
+  a†
1764
+ ka†
1765
+ −k.
1766
+
1767
+ 12
1768
+ 5
1769
+ These terms thus also appear as corrections to the Ak and Bk
1770
+ in the quadratic part of the Hamiltonian. Note that the Her-
1771
+ mitian conjugate term of this H4,3−1 also contributes, with its
1772
+ contribution read off from the expression above.
1773
+ H4,1−3 ≈
1774
+
1775
+ k
1776
+ 
1777
+ 1
1778
+ 2N
1779
+
1780
+ q
1781
+ D∗
1782
+ q,−q,k⟨aqa−q⟩
1783
+  a†
1784
+ kak
1785
+ +1
1786
+ 2
1787
+
1788
+ k
1789
+ 
1790
+ 1
1791
+ N
1792
+
1793
+ q
1794
+ D∗
1795
+ k,−k,q⟨a†
1796
+ qaq⟩
1797
+  a−kak.
1798
+ Finally, we can summarize all of these contributions as cor-
1799
+ rections δAk and δBk to the original Ak and Bk of quadratic
1800
+ H2 origin and write
1801
+ δAk = 1
1802
+ N
1803
+
1804
+ q
1805
+
1806
+ Vk,q,0⟨a†
1807
+ qaq⟩ + 1
1808
+ 2
1809
+
1810
+ Dq,−q,k⟨a†
1811
+ qa†
1812
+ −q⟩ + c.c.
1813
+ ��
1814
+ ,
1815
+ (S10a)
1816
+ δBk = 1
1817
+ N
1818
+
1819
+ q
1820
+
1821
+ Dk,−k,q⟨a†
1822
+ qaq⟩ + 1
1823
+ 2Vq,−q,k−q⟨aqa−q⟩
1824
+
1825
+ .
1826
+ (S10b)
1827
+ In terms of these corrections, the renormalized spectrum is
1828
+ given by
1829
+ Ωk ≡
1830
+
1831
+ (Ak + δAk)2 − (Bk + δBk)2.
1832
+ (S11)
1833
+ These corrections can be evaluated using the bare, free av-
1834
+ erages from Eq. (S8), though this approach leads to diver-
1835
+ gences (see Sec. III F). Alternatively, they can be evaluated
1836
+ self-consistently, with the averages in Eq. (S8) computed us-
1837
+ ing (Ak + δAk), (Bk + δBk) and Ωk instead of Ak, Bk and ωk,
1838
+ which cures the divergences.
1839
+ C.
1840
+ Pseudo-Goldstone gap
1841
+ The effects of the interactions on the pseudo-Goldstone
1842
+ mode can now be examined. The energy of the k = 0 mode is
1843
+ given by
1844
+ ∆ ≡ Ω0 =
1845
+
1846
+ 2KS (δA0 + δB0) + δA2
1847
+ 0 − δB2
1848
+ 0.
1849
+ (S12)
1850
+ For small corrections δA0, δB0, ∆ above can be approxi-
1851
+ mated by (the leading term)
1852
+ ∆ ≈
1853
+
1854
+ 2KS
1855
+
1856
+ δA0 + δB0.
1857
+ (S13)
1858
+ In the quantum limit where T ≪ ωk, the corrections δAk,
1859
+ δBk are O(S 0) and thus the gap scales as ∆ ∝
1860
+
1861
+ S .
1862
+ In
1863
+ the classical limit where T ≫ ωk the corrections scale as
1864
+ δAk, δBk ∼ O(T/S ) and thus the gap scales as ∆ ∝
1865
+
1866
+ T, inde-
1867
+ pendent of S .
1868
+ D.
1869
+ Pseudo-Goldstone Linewidth
1870
+ To estimate the scaling of the pseudo-Goldstone mode
1871
+ linewidth with temperature, we consider the magnon self-
1872
+ energy [11] at k = 0 near ω = 0, which takes the form
1873
+ Σ(0, 0) ≡
1874
+
1875
+ δA0 δB0
1876
+ δB∗
1877
+ 0 δA∗
1878
+ 0
1879
+
1880
+ ,
1881
+ where δA0 and δB0 are corrections due to magnon-magnon
1882
+ interactions. Perturbatively, we expect that
1883
+ δA0 = a1T + a2T 2 + · · · ,
1884
+ (S14a)
1885
+ δB0 = b1T + b2T 2 + · · · ,
1886
+ (S14b)
1887
+ where the O(T) corrections (computed in this work) encoded
1888
+ in a1, b1 are both real. The quasi-normal modes, correspond-
1889
+ ing to the locations of poles of the magnon Green’s func-
1890
+ tion [11, 14], are determined from eigenvalues of σzMeff
1891
+ 0
1892
+ where
1893
+ Meff
1894
+ 0 =
1895
+
1896
+ A0 + δA0
1897
+ −A0 + δB0
1898
+ −A0 + δB∗
1899
+ 0
1900
+ A0 + δA∗
1901
+ 0
1902
+
1903
+ .
1904
+ Up to and including terms of O(T 2), the quasi-normal mode
1905
+ frequency is thus given by
1906
+ Re Ω0 ≈
1907
+
1908
+ 2A0(a1 + b1)
1909
+
1910
+ T
1911
+ +
1912
+ 
1913
+ a2
1914
+ 1 − b2
1915
+ 1 + 2A0(Re a2 + Re b2)
1916
+ 4A0(a1 + b1)
1917
+  T 3/2 + · · · ,
1918
+ Im Ω0 ≈ (Im a2)T 2 + · · · .
1919
+ We thus see that the linewidth, determined by Im Ω0, is ex-
1920
+ pected to scale as T 2.
1921
+ E.
1922
+ Self-Consistent Mean-Field Theory (SCMFT)
1923
+ To include the effects of the magnon-magnon interactions
1924
+ self-consistently, we define the “mean-fields”
1925
+ nk ≡ ⟨a†
1926
+ kak⟩,
1927
+ dk ≡ ⟨a†
1928
+ ka†
1929
+ −k⟩.
1930
+ (S15)
1931
+ Using Eq. (S10), new values of nk and dk can then be com-
1932
+ puted by iteratively updating Ak and Bk to
1933
+ A′
1934
+ k = Ak + 1
1935
+ N
1936
+
1937
+ q
1938
+
1939
+ Vk,q,0nq + 1
1940
+ 2
1941
+
1942
+ Dq,−q,kdq + D∗
1943
+ q,−q,kd∗
1944
+ q
1945
+ ��
1946
+ ,
1947
+ B′
1948
+ k = Bk + 1
1949
+ N
1950
+
1951
+ q
1952
+
1953
+ Dk,−k,qnq + 1
1954
+ 2Vq,−q,k−qd∗
1955
+ q
1956
+
1957
+ ,
1958
+ which, using Eq. (S8), results in updated values of nk and dk.
1959
+ This process is repeated until the variables nk and dk have con-
1960
+ verged to the desired precision across the full Brillouin zone.
1961
+ For the calculations reported here, and in the main text, con-
1962
+ vergence was considered reached when the sum of all absolute
1963
+ values of the changes in nk and dk in Eq. (S15) over the Bril-
1964
+ louin zone between iterations was less than 10−10. To launch
1965
+ the iterative process, the mean-fields, nk and dk for each k,
1966
+ are initially set to a value of 1/2, though the precise choice of
1967
+ initial value was not found to affect the final results. Follow-
1968
+ ing this approach, we calculate the PG gap for several system
1969
+
1970
+ 13
1971
+ 6
1972
+ sizes, using a discrete sum of the Brillouin zone with N = L2
1973
+ points. We then extrapolate the gap in the system size to ob-
1974
+ tain the result in the thermodynamic limit (N → ∞). The
1975
+ finite size scaling of the PG gap using SCMFT is shown in
1976
+ Fig. S2.
1977
+ F.
1978
+ Cancellation of divergences in the pseudo-Goldstone gap
1979
+ Since the non-interacting LSWT spectrum is gapless, we
1980
+ must be mindful of infrared divergent contributions to δA0 and
1981
+ δB0. Let us first address this issue in the simplest context, bare
1982
+ perturbation theory in the quartic interactions.
1983
+ We focus on the classical limit where ωk ≪ T, but similar
1984
+ considerations apply in the full quantum case at finite tem-
1985
+ perature; since ωk → 0 as k → 0, there is always a regime
1986
+ in k near the zone center where the frequency is small rela-
1987
+ tive to temperature, even in the quantum limit. Consider the
1988
+ corrections, Eq. (S10), in the thermodynamic limit (N → ∞),
1989
+ replacing the discrete sums with integrals. At k = 0, this gives
1990
+ [using Eq. (S8)]
1991
+ δA0 =
1992
+
1993
+ d2q
1994
+ (2π)2
1995
+ T
1996
+ ω2q
1997
+
1998
+ V0,q,0Aq − Dq,−q,0Bq
1999
+
2000
+ ,
2001
+ (S16a)
2002
+ δB0 =
2003
+
2004
+ d2q
2005
+ (2π)2
2006
+ T
2007
+ ω2q
2008
+
2009
+ D0,0,qAq − 1
2010
+ 2Vq,−q,−qBq
2011
+
2012
+ ,
2013
+ (S16b)
2014
+ where the integral is over the Brillouin zone −π ≤ qx, qy ≤
2015
+ π (the lattice spacing has been set to one). At small q, the
2016
+ spectrum is approximately linear in q with
2017
+ ωq = S
2018
+
2019
+ 2K
2020
+
2021
+ J|q|2 + Kq2y
2022
+
2023
+ + O(|q|2)
2024
+ and thus the factor T/ω2
2025
+ q ∝ 1/|q|2 is singular as |q| → 0. The
2026
+ numerators of the integrals in Eq. (S16) remain finite in this
2027
+ limit, with
2028
+ V0,q,0Aq − Dq,−q,0Bq = −S K2
2029
+ 2
2030
+ + O(|q|2),
2031
+ D0,0,qAq − 1
2032
+ 2Vq,−q,−qBq = +S K2
2033
+ 2
2034
+ + O(|q|2).
2035
+ One therefore finds that both δA0 and δB0 are logarithmically
2036
+ divergent. Explicitly, integrating over a region 2π/L < |q| <
2037
+ Λ ≪ π
2038
+
2039
+ 2π/L<|q|<Λ
2040
+ d2q
2041
+ (2π)2
2042
+ 1
2043
+ ω2q
2044
+ =
2045
+ 1
2046
+ 4πS 2K √J(J + K)
2047
+ ln
2048
+ �LΛ
2049
+
2050
+
2051
+ .
2052
+ (S17)
2053
+ Since the upper cutoff is chosen to satisfy Λ ≪ π, the diver-
2054
+ gent contributions to δA0 and δB0 take the form
2055
+ δA0 = −
2056
+ TK ln L
2057
+ 8πS √J(J + K)
2058
+ + (reg.),
2059
+ (S18a)
2060
+ δB0 = +
2061
+ TK ln L
2062
+ 8πS √J(J + K)
2063
+ + (reg.),
2064
+ (S18b)
2065
+ where (reg.) stands for terms that remain finite as L → ∞. In-
2066
+ terestingly, while δA0 and δB0 are each ln L divergent, the sum
2067
+ (δA0 + δB0) which appears in the expression for the pseudo-
2068
+ Goldstone gap [Eq. (S13)], ∆, is finite. This can be made more
2069
+ explicit by carrying out the same expansions for (δA0 + δB0),
2070
+ δA0 + δB0 =
2071
+
2072
+ d2q
2073
+ (2π)2
2074
+ T
2075
+ ω2q
2076
+
2077
+ 2K2S (q2
2078
+ x − q2
2079
+ y) + O(|q|4)
2080
+
2081
+ . (S19)
2082
+ The O(|q|2) term in the ω2
2083
+ q denominator is thus com-
2084
+ pensated by a corresponding O(|q|2) in the numerator of
2085
+ Eq. (S19).
2086
+ However, note that this cancellation only oc-
2087
+ curs at leading order in δA0, δB0. The complete expression
2088
+
2089
+ (A0 + δA0)2 − (B0 + δB0)2, which incorporates higher-order
2090
+ contributions, remains logarithmically divergent. Similarly,
2091
+ the leading corrections from bare perturbation theory to Ωk at
2092
+ non-zero k are also divergent.
2093
+ Since the bare perturbation theory diverges, except for the
2094
+ leading temperature dependence of the PG gap at q = 0, in
2095
+ order to obtain the full temperature dependence of the inter-
2096
+ action corrections to Ωq, we proceed with a self-consistent
2097
+ approach. This way, the equation for the corrections δA0 and
2098
+ δB0 become
2099
+ δA0 =
2100
+
2101
+ d2q
2102
+ (2π)2
2103
+ T
2104
+ Ω2q
2105
+
2106
+ V0,q,0(Aq + δAq) − Dq,−q,0(Bq + δBq)
2107
+
2108
+ ,
2109
+ δB0 =
2110
+
2111
+ d2q
2112
+ (2π)2
2113
+ T
2114
+ Ω2q
2115
+
2116
+ D0,0,q(Aq + δAq) − 1
2117
+ 2Vq,−q,−q(Bq + δBq)
2118
+
2119
+ ,
2120
+ where the renormalized spectrum Ωq arises from evaluating
2121
+ the averages in Eq. (S8) self-consistently.
2122
+ In such a self-
2123
+ consistent mean-field theory, the spectrum Ωq “already” con-
2124
+ tains a finite gap at q = 0. The gap acts as an effective infrared
2125
+ cutoff rendering the integrals in Eq. (S17) finite. The disap-
2126
+ pearance of the divergence then manifests itself in the cancel-
2127
+ lation of the leading (self-consistent) dependence on the gap
2128
+ ∆.
2129
+ To see this explicitly, consider the self-consistent spectrum
2130
+ which, for small q, takes the form
2131
+ Ωq =
2132
+
2133
+ 2KS 2 �
2134
+ J|q|2 + Kq2y
2135
+
2136
+ + ∆2 + O(|q|2).
2137
+ For sufficiently small ∆, the integration region can be divided
2138
+ into two parts: |q| ≳ k0 and |q| ≲ k0 such that
2139
+ Ωq ≈
2140
+ 
2141
+ ∆,
2142
+ |q| ≲ k0,
2143
+ S
2144
+
2145
+ 2K
2146
+
2147
+ J|q|2 + Kq2y
2148
+
2149
+ ,
2150
+ |q| ≳ k0.
2151
+ Roughly, the boundary separating these regions scales as
2152
+ k0 ∼ ∆
2153
+ S K ∝
2154
+
2155
+ T
2156
+ (S21)
2157
+ when K ≳ J. Alternatively, k0 is the wave-vector at which
2158
+ the bare spectrum ωk becomes comparable to the interaction
2159
+ induced gap, ∆. The primary change to the spectrum, and thus
2160
+ to δA0 and δB0, occurs for |q| < k0. Carrying out the integra-
2161
+ tion in Eq. (S17) over the region responsible for its divergent
2162
+ contributions, we find they are rendered finite. Explicitly,
2163
+
2164
+ k0<|q|<Λ
2165
+ d2q
2166
+ (2π)2
2167
+ 1
2168
+ Ω2q
2169
+
2170
+ ln (KS Λ/∆)
2171
+ 4πS 2K √J(J + K)
2172
+ .
2173
+
2174
+ 14
2175
+ 7
2176
+ Given that ∆ ∝
2177
+
2178
+ T, this contribution to the integral now
2179
+ scales as − ln T. Thus the divergence has been cured in the in-
2180
+ dividual corrections δA0 and δB0. We note that the ln(Λ/∆) ∼
2181
+ − ln T terms cancel in the sum (δA0 + δB0) which controls the
2182
+ leading contribution to the gap [similarly to Eq. (S13)] and
2183
+ the result from bare perturbation theory is recovered. In this
2184
+ way, bare perturbation theory for the asymptotic
2185
+
2186
+ T scaling
2187
+ of the pseudo-Goldstone gap is well-defined and divergence
2188
+ free, and matches the results from the SCMFT calculations.
2189
+ Note that the region 0 < |q| < k0 gives only a finite contribu-
2190
+ tion that goes as
2191
+
2192
+ 0<|q|<k0
2193
+ d2q
2194
+ (2π)2
2195
+ 1
2196
+ Ω2q
2197
+
2198
+ k2
2199
+ 0
2200
+ 4π∆2 ∼ const. ,
2201
+ since k0 ∝ ∆.
2202
+ G.
2203
+ Logarithmic corrections to magnetization
2204
+ While the contributions ∝ ln T cancel in the leading parts
2205
+ of ∆, they reappear explicitly in static quantities such as the
2206
+ magnetization. In the classical limit, the net magnetization
2207
+ [Eq. (S9)] is given by
2208
+ M = S − T
2209
+
2210
+ d2k
2211
+ (2π)2
2212
+ Ak
2213
+ ω2
2214
+ k
2215
+ ,
2216
+ (S22)
2217
+ in the thermodynamic limit (N → ∞). Like the bare cor-
2218
+ rections δA0 and δB0 in Eq. (S16), in LSWT M has a (log-
2219
+ arithmic) infrared divergence, given that ωk scales linearly
2220
+ with k at small |k|, while Ak tends to the constant A0 = KS .
2221
+ Na¨ıvely, this would indicate the destruction of the order, as
2222
+ happens when there is a true symmetry-protected Goldstone
2223
+ mode [15].
2224
+ To resolve this divergence, we must include the dynami-
2225
+ cally generated pseudo-Goldstone gap. The expression for M
2226
+ in the SCMFT can be obtained from Eq. (S22), via the re-
2227
+ placements ωk → Ωk and Ak → Ak + δAk. Explicitly,
2228
+ M = S − T
2229
+
2230
+ d2k
2231
+ (2π)2
2232
+ Ak + δAk
2233
+ Ω2
2234
+ k
2235
+ ,
2236
+ (S23)
2237
+ where, again, Ωk is the self-consistent spectrum with pseudo-
2238
+ Goldstone gap ∆. Following the same strategy as in Sec. III F,
2239
+ we approximate the self-consistent spectrum, Ωk, over the
2240
+ three pertinent regions of reciprocal space
2241
+ Ωk ≈
2242
+ 
2243
+ ∆,
2244
+ 0 < |k| ≲ k0
2245
+ S
2246
+
2247
+ 2K
2248
+
2249
+ J|k|2 + Kk2y
2250
+
2251
+ ,
2252
+ k0 ≲ |k| ≲ Λ
2253
+ ωk.
2254
+ |k| ≳ Λ
2255
+ (S24)
2256
+ The integral defining M is then split into three parts
2257
+
2258
+ d2k
2259
+ (2π)2 =
2260
+
2261
+ 0<|k|<k0
2262
+ d2k
2263
+ (2π)2 +
2264
+
2265
+ k0<|k|<Λ
2266
+ d2k
2267
+ (2π)2 +
2268
+
2269
+ |k|>Λ
2270
+ d2k
2271
+ (2π)2 .
2272
+ The first and second integrals depend on temperature through
2273
+ ∆ ∝
2274
+
2275
+ T and k0 ∼ ∆/(KS ) [see Eq. (S21)]. The last integral
2276
+ is over wave-vectors large enough such that the interaction
2277
+ corrections are minor and, therefore, this contribution to the
2278
+ integral has no additional temperature dependence. The cor-
2279
+ rection to M from this last (third) term is thus ∝ T.
2280
+ For the region |k| ≲ Λ, we approximate Ak + δAk ≈ KS ,
2281
+ leaving the two contributions
2282
+
2283
+ 0<|k|<k0
2284
+ d2k
2285
+ (2π)2
2286
+ KS T
2287
+ ∆2
2288
+ =
2289
+ T
2290
+ 4πKS ,
2291
+
2292
+ k0<|k|<Λ
2293
+ d2k
2294
+ (2π)2
2295
+ KS T
2296
+ 2KS 2 �
2297
+ J|k|2 + Kk2y
2298
+ � = T ln (KS Λ/∆)
2299
+ 4πS √J(J + K)
2300
+ .
2301
+ The renormalized spectrum has thus yielded two additional
2302
+ contributions to M: one that adds an additional part ∝ T and
2303
+ another new, and perhaps most interesting, part, ∝ T ln T, and
2304
+ where we used ∆ ∝
2305
+
2306
+ T.
2307
+ To summarize, the magnetization at low temperatures takes
2308
+ the form
2309
+ M = S − c1T − c2T ln T + · · · ,
2310
+ (S25)
2311
+ where c1 and c2 are temperature independent constants. The
2312
+ logarithmic T ln T dependence arises from the temperature
2313
+ dependence of the pseudo-Goldstone gap, ∆. As T → 0,
2314
+ both of these temperature dependent terms go to zero (as it
2315
+ should classically) and the system becomes fully polarized
2316
+ with M → S . Hence true long-range order is induced by
2317
+ ObTD, with the lurking infrared divergences tamed by the
2318
+ ObTD-induced PG gap, ∆.
2319
+ [1] J. D. Alzate-Cardona, D. Sabogal-Su´arez, R. F. L. Evans, and
2320
+ E. Restrepo-Parra, “Optimal phase space sampling for Monte
2321
+ Carlo simulations of Heisenberg spin systems,” Journal of
2322
+ Physics: Condensed Matter 31, 095802 (2019).
2323
+ [2] Michael Creutz, “Overrelaxation and Monte Carlo simulation,”
2324
+ Phys. Rev. D 36, 515–519 (1987).
2325
+ [3] M. E. J. Newman and G. T. Barkema, Monte Carlo methods in
2326
+ statistical physics (Clarendon Press, Oxford, 1999).
2327
+ [4] J. R. Dormand and P. J. Prince, “A family of embedded Runge-
2328
+ Kutta formulae,” Journal of Computational and Applied Math-
2329
+ ematics 6, 19–26 (1980).
2330
+ [5] K. Ahnert and M. Mulansky, “Odeint – solving ordinary differ-
2331
+ ential equations in C++,” AIP Conference Proceedings 1389,
2332
+ 1586–1589 (2011).
2333
+ [6] K. Ahnert and M. Mulansky, “Boost C++ Library: Odeint,”
2334
+ (2012).
2335
+ [7] J. C. Bowman and M. Roberts, “FFTW++: A fast Fourier trans-
2336
+ form C++ header class for the FFTW3 library,” (2010).
2337
+ [8] A. Auerbach, Interacting Electrons and Quantum Magnetism,
2338
+ Graduate Texts in Contemporary Physics (Springer New York,
2339
+ 1998).
2340
+ [9] Jeffrey G. Rau, Paul A. McClarty,
2341
+ and Roderich Moess-
2342
+
2343
+ 15
2344
+ 8
2345
+ ner, “Pseudo-Goldstone gaps and order-by-quantum disorder in
2346
+ frustrated magnets,” Phys. Rev. Lett. 121, 237201 (2018).
2347
+ [10] H. Bruus and K. Flensberg, Many-Body Quantum Theory in
2348
+ Condensed Matter Physics: An Introduction, Oxford Graduate
2349
+ Texts (Oxford University Press, Oxford, 2004).
2350
+ [11] J.P. Blaizot and G. Ripka, Quantum Theory of Finite Systems
2351
+ (MIT Press, Cambridge, 1986).
2352
+ [12] P. D. Loly, “The Heisenberg ferromagnet in the selfconsistently
2353
+ renormalized spin wave approximation,” Journal of Physics C:
2354
+ Solid State Physics 4, 1365–1377 (1971).
2355
+ [13] A. V. Chubukov, S. Sachdev, and T. Senthil, “Large-S expan-
2356
+ sion for quantum antiferromagnets on a triangular lattice,” Jour-
2357
+ nal of Physics: Condensed Matter 6, 8891–8902 (1994).
2358
+ [14] M. E. Zhitomirsky and A. L. Chernyshev, “Colloquium: Spon-
2359
+ taneous magnon decays,” Rev. Mod. Phys. 85, 219–242 (2013).
2360
+ [15] P. M. Chaikin and T. C. Lubensky, Principles of Condensed
2361
+ Matter Physics (Cambridge University Press, 1995).
2362
+
3NFKT4oBgHgl3EQf8S4-/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
4NE4T4oBgHgl3EQfbQzc/content/tmp_files/2301.05072v1.pdf.txt ADDED
@@ -0,0 +1,688 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.05072v1 [cond-mat.stat-mech] 12 Jan 2023
2
+ Chemical kinetics and stochastic differential equations
3
+ Chiara Pezzotti and Massimiliano Giona1, ∗
4
+ 1Dipartimento di Ingegneria Chimica, Materiali,
5
+ Ambiente La Sapienza Universit`a di Roma
6
+ Via Eudossiana 18, 00184 Roma, Italy
7
+ (Dated: January 13, 2023)
8
+ Abstract
9
+ We propose a general stochastic formalism for describing the evolution of chemical reactions
10
+ involving a finite number of molecules. This approach is consistent with the statistical analysis
11
+ based on the Chemical Master Equation, and provides the formal setting for the existing algorithmic
12
+ approaches (Gillespie algorithm). Some practical advantages of this formulation are addressed, and
13
+ several examples are discussed pointing out the connection with quantum transitions (radiative
14
+ interactions).
15
+ ∗ corresponding author:[email protected]
16
+ 1
17
+
18
+ All the chemical physical processes involve, in an atomistic perspective, a stochastic de-
19
+ scription of the events, be them reactive or associated with a change of phase (for instance
20
+ adsorption) [1]. Nonetheless, in the overwhelming majority of the cases of practical and labo-
21
+ ratory interest, the number of molecules involved is so large to justify a mean field approach,
22
+ essentially based on the Boltzmannian hypothesis of molecular chaos (the “stosszahlansatz”)
23
+ [2]. The mean field formulation represents the backbone of the classical theory of chemical
24
+ reaction kinetics [3, 4].
25
+ It is well known that, in all the cases where the number of molecule is small (and this
26
+ occurs in subcellular biochemical reactions, in nanoscale systems, or in the growth kinetics of
27
+ microorganisms [5–7]), the effects of fluctuations become significant, motivating a stochastic
28
+ description of chemical kinetic processes, involving the number of molecules present in the
29
+ system, thus explicitly accounting for due to their finite number [8–11]. The statistical theory
30
+ of chemical kinetics in these conditions is grounded on the Chemical Master Equation (CME)
31
+ [12, 13], expressing the evolution equation for the probabilities p(N, t) of all the possible
32
+ number-configurations N(t) = (N1(t), . . . , Ns(t)), where Nh(t) is the number of molecules of
33
+ the h-th reacting species at time t, h = 1, . . . , s. However, apart from a handful of simple
34
+ cases, for which the CME can be solved analytically [14], numerical methods should be
35
+ applied to it in order to compute mean values and higher-order moments. But also this
36
+ choice reveals itself to be unfeasible in most of the situations of practical and theoretical
37
+ interests, due to the extremely large number of configurations involved, making the multi-
38
+ index matrix p(N, t) so huge to exceed reasonable computational facilities.
39
+ In order to solve this problem, Gillespie proposed an algorithmic solution to the numeri-
40
+ cal simulation of stochastic reacting systems, based on the Markovian nature of the reactive
41
+ events [15, 16]. The original Gillespie algorithm has been extended and improved over time,
42
+ providing a variety of slightly different computational alternatives. A common denominator
43
+ of the first family of the Gillespie algorithms (namely those based on the direct method, the
44
+ first reaction method or their derivates [17–19]) is to associate to every time step the occur-
45
+ rence of just one reaction. This formulation comes directly from the assumption that, if the
46
+ time step is small enough, the probability that more than one reaction will occur is negligi-
47
+ ble. While correct, this choice brings to significant computational costs for complex reaction
48
+ schemes. This problem has been highlighted several times, from the Gillespie group itself,
49
+ as stiffness in stochastic chemical reacting systems [20]. A brilliant way to overcome this
50
+ 2
51
+
52
+ limit originates the famous tau-leaping method, which, unfortunately, requires to check that
53
+ the propensity functions remain almost constant at each iteration and can be applied just
54
+ if this condition is verified [21, 22]. The algorithmic solution associated with the formalism
55
+ here introduced combines the accuracy of the first SSA with the computational advantages
56
+ of the τ-leaping method.
57
+ There is, moreover, a missing link between the CME theory and the Gillespie algorithm,
58
+ consisting in the straight mathematical formulation of the stochastic differential equations
59
+ associated with a chemical reacting system, the statistical description of which would cor-
60
+ respond to the CME. To clarify this issue, consider the conceptually analogous problem of
61
+ particle diffusion over the real line, the statistical description of which is expressed by the
62
+ parabolic equation ∂p(x, t)/∂t = D ∂2p(x, t)/∂x2, for the probability density p(x, t) of finding
63
+ a particle at position x at time t. Setting xn = x(n ∆t), an algorithm describing this process
64
+ can be simply expressed by the discrete evolution equation xn+1 = xn +
65
+
66
+ 2 D ∆t rn+1, where
67
+ rh, h = 1, 2, . . . represent independent random variables sampled from a normal distribution
68
+ (with zero mean, and unit variance) [23]. This represents an efficient algorithmic solution
69
+ of the problem, whenever the time resolution ∆t is small enough. Nevertheless, the mere
70
+ algorithmic approach cannot be considered physically satisfactory, in a comprehensive for-
71
+ mulation of transport theory embedded in a continuous space-time (in which both position x
72
+ and time t are real valued). In point of fact, only with the mathematical formulation due to
73
+ K. Ito of stochastic differential equations driven by the increments dw(t) of a Wiener process
74
+ (Langevin equations) [24], namely dx(t) =
75
+
76
+ 2 D dw(t) the theory of diffusive motion has
77
+ found a proper mathematical physical setting.
78
+ A similar situation applies to the case of stochastic models of chemical reaction kinetics,
79
+ and the present Letter is aimed at filling this gap.
80
+ The basic idea is that any reactive
81
+ process corresponds to a system of elementary events (the single reaction) possessing a
82
+ Markovian transitional structure, and, consequently, amenable to a description by means
83
+ of the increments of counting processes (Poisson processes, in the Markovian case). This
84
+ topic has been also pointed out in [25] in terms of Poisson measures, although the latter
85
+ formulation is much less simple and physically intuitive than the approach proposed in the
86
+ present Letter.
87
+ To begin with, consider the simple case of a first-order chemical reaction A
88
+ k1
89
+
90
+ k−1 B (for
91
+ instance, an isomerization). This model is perfectly analogous to the radiative transition
92
+ 3
93
+
94
+ FIG. 1. Schematic representation of the analogy between a two-level quantum system and a first-
95
+ order chemical kinetics, such as an isomerization.
96
+ of a molecule possessing two energy states, due to emission and adsorption of an energy
97
+ quantum (figure 1). Let NA(0) + NB(0) = Ng the total number of molecules at time t = 0.
98
+ The state of the system is characterized by the state functions σh(t), h = 1, . . . , Ng for each
99
+ molecule, attaining values {0, 1}, and such that σh(t) = 0 if the energy state at time t is E0
100
+ (or equivalently if the molecule finds itself in the state A), and σh(t) = 1 in the opposite
101
+ case (energy state E1, or isomeric state B).
102
+ Let {χ(1)
103
+ h (t, k1), χ(2)
104
+ h (t, k−1)}Ng
105
+ h=1 be two systems of independent Poisson processes, char-
106
+ acterized by the transition rates k1, and k−1, respectively. The evolution of σh(t) can be
107
+ expressed via the stochastic differential equation
108
+ dσh(t)
109
+ dt
110
+ = (1 − σh(t)) dχ(1)
111
+ h (t, k1)
112
+ dt
113
+ − σh(t) dχ(2)
114
+ h (t, k−1)
115
+ dt
116
+ (1)
117
+ h = 1, . . . , Ng, where dχ(t, λ)/dt is the distributional derivative of the Poisson process
118
+ χ(t, λ), corresponding to a sequence of unit impulsive functions at the transition instants
119
+ t∗
120
+ i , i = 1, 2, . . . , 0 < t∗
121
+ i < t∗
122
+ i+1, where for ε > 0, limε→0
123
+ � t∗
124
+ i +ε
125
+ t∗
126
+ i −ε dχ(t, λ) = 1. Summing over
127
+ h = 1, . . . Ng, and observing that NA(t) = �Ng
128
+ h=1 (1 − σh(t)), NB(t) = �Ng
129
+ h=1 σh(t), we have
130
+ dNB(t)
131
+ dt
132
+ =
133
+ NA(t)
134
+
135
+ h=1
136
+ dχ(1)
137
+ h (t, k1)
138
+ dt
139
+
140
+ NB(t)
141
+
142
+ h=1
143
+ dχ(2)
144
+ h (t, k−1)
145
+ dt
146
+ (2)
147
+ 4
148
+
149
+ (b)(a)OPand dNA(t)/dt = −dNB(t)/dt, representing the evolution equation for NA(t) and NB(t),
150
+ attaining integer values. The stochastic evolution of the number of molecules NA(t), NB(t)
151
+ is thus expressed as a differential equation with respect to the continuous physical time
152
+ t ∈ R+, over the increments of a Poisson process. Intepreted in a mean-field way, if ctot is
153
+ the overall concentration of the reactants at time t = 0, then the concentrations cα(t) at
154
+ time t can be recovered from eq. (2) as
155
+ cα(t) = ctot
156
+ Nα(t)
157
+ Ng
158
+ ,
159
+ α = A, B
160
+ (3)
161
+ representing the calibration relation connecting the stochastic description in terms of num-
162
+ ber of molecules Nα(t) and the concentrations cα(t), α = A, B entering the mean-field
163
+ description.
164
+ The analytical formulation of a stochastic differential equation for chemical kinetics,
165
+ expressed in terms of the number of molecules of the chemical species involved, rather than
166
+ an algorithm defined for discretized times, permits to develop a variety of different numerical
167
+ strategies, that naturally perform a modified tau-leaping procedure, as the occurrence of
168
+ several distinct reactive events in any elementary time step ∆t is intrinsically accounted for.
169
+ This can be easily seen by considering the simple reaction defined by the evolution equation
170
+ (2).
171
+ In terms of increments, eq.
172
+ (2) can be written as dNB(t) = �NA(t)
173
+ h=1 dχ(1)(t, k1) −
174
+ �NB(t)
175
+ h=1 dχ(2)(t, k−1). If ∆t is the chosen time step, it follows from this formulation, a simple
176
+ numerical approximation for eq. (2), namely,
177
+ ∆NB(t) = NB(t + ∆t) − NB(t) =
178
+ NA(t)
179
+
180
+ h=1
181
+ ξ(1)
182
+ h (k1 ∆t) −
183
+ NB(t)
184
+
185
+ h=1
186
+ ξ(2)
187
+ h (k−1 ∆t)
188
+ (4)
189
+ where ξ(1)(k1 ∆t), ξ(2)
190
+ h (k−1 ∆t) h = 1, 2, . . . , are two families of independent binary random
191
+ variables, where
192
+ ξ(α)
193
+ h (p) =
194
+
195
+
196
+
197
+ 1
198
+ with probability p
199
+ 0
200
+ otherwise
201
+ (5)
202
+ α = 1, 2, h = 1, 2, . . . . The time step ∆t, can be chosen in eq. (4) from the condition
203
+ K ∆t < 1 ,
204
+ K = max{k1, k−1}
205
+ (6)
206
+ In practice, we choose ∆t = 0.1/K. As can be observed, the choice of ∆t is limited by the
207
+ intrinsic rates of the process. The advantage of deriving different algorithmic schemes for
208
+ 5
209
+
210
+ solving numerically the stochastic kinetic equations becomes more evident in dealing with
211
+ bimolecular reactions (addressed below). Due to the intrinsic limitations of this commu-
212
+ nication, a thorouh discussion of this issue is postponed to a future more extensive article
213
+ [26].
214
+ The same approach can be extended to include amongst the elementary events not only
215
+ the reactive steps, but also feeding conditions, thus representing the evolution of chemically
216
+ reacting systems with a finite number of molecules in a perfectly stirred open reactor. This is
217
+ the case of the tank-loading problem, in which a tracer is injected in an open vessel assumed
218
+ perfectly mixed, for which, in the absence of chemical reactions, the mean field equation for
219
+ the concentration of the tracer reads
220
+ dc(t)
221
+ dt
222
+ = D (c0 − c(t))
223
+ (7)
224
+ where c0 is the inlet concentration and D the dilution rate (reciprocal of the mean retention
225
+ time), and c(0) = 0. Fixing Ng so that c(t) = c0 N(t)/Ng, the corresponding stochastic
226
+ differential equation for the integer N(t) involves, also in this case, two families of counting
227
+ processes, one for the loading at constant concentration c0, and the other for tracer discharge
228
+ in the outlet stream, characterized by the same transition rate D,
229
+ dN(t)
230
+ dt
231
+ =
232
+ Ng
233
+
234
+ h=1
235
+ dχ(1)
236
+ h (t, D)
237
+ dt
238
+
239
+ N(t)
240
+
241
+ k=1
242
+ dχ(2)
243
+ h (t, D)
244
+ dt
245
+ (8)
246
+ starting from N(0) = 0. Figure 2 depicts several realizations of the tank-loading process,
247
+ obtained by discretizing eq. (8) with a time step ∆t = 10−3. Despite the simplicity of the
248
+ process, this example permits to highlight the role of Ng, that can be referred to as the granu-
249
+ larity number, and the way stochastic models of chemical reactions can be fruitfully applied.
250
+ Indeed, there is a two-fold use of the stochastic formulation of chemical kinetic schemes.
251
+ The first refers to a chemical reacting system involving a small number of molecules, and
252
+ in this case Ng represents the effective number of molecules present in the system. The
253
+ other is to use stochastic algorithms for simulating reacting systems in an alternative (and
254
+ sometimes more efficient way) with respect to the solution of the corresponding mean-field
255
+ equations. In the latter case, the granularity number Ng represents essentially a computa-
256
+ tional parameter, tuning the intensity of the fluctuations. Two choices are then possible:
257
+ (i) it can be chosen large enough, in order to obtain from a single realization of the process
258
+ an accurate approximation of the mean-field behavior, or (ii) it can be chosen small enough
259
+ 6
260
+
261
+ 0
262
+ 0.4
263
+ 0.8
264
+ 1.2
265
+ 0
266
+ 2
267
+ 4
268
+ 6
269
+ 8
270
+ 10
271
+ c(t)
272
+ t
273
+ 0
274
+ 0.4
275
+ 0.8
276
+ 1.2
277
+ 0
278
+ 2
279
+ 4
280
+ 6
281
+ 8
282
+ 10
283
+ c(t)
284
+ t
285
+ 0
286
+ 0.4
287
+ 0.8
288
+ 1.2
289
+ 0
290
+ 2
291
+ 4
292
+ 6
293
+ 8
294
+ 10
295
+ c(t)
296
+ t
297
+ (a)
298
+ (b)
299
+ (c)
300
+ FIG. 2. c(t) = N(t)/Ng vs t from a single realization of the tank-loading process eq. (8) with
301
+ D = 1, c0 = 1 a.u.. Panel (a): Ng = 30, panel (b) Ng = 100, panel (c) Ng = 1000. The solid
302
+ horizontal lines represent the steady-state value c∗ = 1.
303
+ in order, to deal with extremely fast simulations of a single realization of the process, that
304
+ could be averaged over a statistically significant number of realizations in due time. These
305
+ two choices are depicted in figure 2 (panel c), choosing Ng = 103, and in figure 3 panel (a)
306
+ obtained for Ng = 30. Of course, the latter approach is valid as long as the low-granularity
307
+ (low values of Ng) does not influence the qualitative properties of the kinetics.
308
+ The second (computational) use of stochastic simulations of chemical kinetics requires a
309
+ further discussion. At a first sight, it may appear that any stochastic simulation would be
310
+ computationally less efficient than the solution of the corresponding mean-field equations.
311
+ This is certainly true for classical chemical reaction schemes in a perfectly mixed system,
312
+ 7
313
+
314
+ 0
315
+ 0.2
316
+ 0.4
317
+ 0.6
318
+ 0.8
319
+ 1
320
+ 0
321
+ 2
322
+ 4
323
+ 6
324
+ 8
325
+ 10
326
+ <c>(t)
327
+ t
328
+ 0
329
+ 0.05
330
+ 0.1
331
+ 0.15
332
+ 0.2
333
+ 0
334
+ 2
335
+ 4
336
+ 6
337
+ 8
338
+ 10
339
+ a
340
+ b
341
+ σc(t)
342
+ t
343
+ (a)
344
+ (b)
345
+ FIG. 3. Panel (a): ⟨c⟩(t) vs t at Ng = 30 (symbols) averaged over [106/Ng] realizations of the
346
+ tank-loading process with D = 1, c0 = 1 a.u. Here, [·] indicates the integer part of its argument.
347
+ The solid line represents the mean-field result ⟨c⟩(t) = 1 − e−t. Panel (b): Variance σc(t) vs t for
348
+ the tank-loading process. Symbols are the results of stochastic simulations of eq. (8) averaged
349
+ over [106/Ng] realizations, lines the solutions of eq. (10). Line (a) refers to Ng = 30, line (b) to
350
+ Ng = 100.
351
+ for which the mean-field model reduces to a system of ordinary differential equations for
352
+ the concentrations of the reactants. But there are kinetic problems e.g., associated with the
353
+ growth of microorganisms and eukaryotic cell lines in bioreactors (these growth phenom-
354
+ ena, are indeed amenable to a description in terms of equivalent chemical reactions), the
355
+ mean-field model of which is expressed in the form of higher-dimensional nonlinear integro-
356
+ differential equations . For this class of problems, the use of stochastic simulations is the
357
+ 8
358
+
359
+ most efficient, if not the only way to achieve a quantitative description of the process, in
360
+ those cases where the number np of internal parameters describing the physiological state
361
+ of an eukaryotic cell becomes large enough, np ≥ 3. This issue is addressed in detail in [27].
362
+ This case is altogether similar to some transport problems, such as Taylor-Aris dispersion for
363
+ high P´eclet numbers or the analysis of microfluidic separation processes (DLD devices) for
364
+ which the stochastic simulation of particle motion is far more efficient that the corresponding
365
+ solution of the corresponding mean-field model expressed in the form of advection-diffusion
366
+ equations [28, 29].
367
+ To complete the analysis of the tank-loading problem, the associated CME reads
368
+ dp(n, t)
369
+ dt
370
+ = D Ng [p(n − 1, t) ηn−1 − p(n, t)] + D [(n + 1) p(n + 1, t) − n p(n, t)]
371
+ (9)
372
+ where ηh = 1 for h ≥ 0 and ηh = 0 otherwise. It follows that ⟨c⟩(t) = c0
373
+ �∞
374
+ n=1 n p(n, t)/Ng
375
+ satisfies identically the mean-field equation (due to the linearity of the problem), while the
376
+ variance σc(t), with σ2
377
+ c(t) = c2
378
+ 0
379
+ �∞
380
+ n=1 n2 p(n, t)/N2
381
+ g − (c0
382
+ �∞
383
+ n=1 n p(n, t)/Ng)2, satisfies the
384
+ equation
385
+ dσ2
386
+ c
387
+ dt = −2 D σ2
388
+ c + D
389
+ � 1
390
+ Ng
391
+ + ⟨c⟩
392
+ Ng
393
+
394
+ (10)
395
+ Figure 3 panel (b) compares the results of stochastic simulations against the solutions of eq.
396
+ (10) for two values of Ng.
397
+ The above approach can be extended to any system of nonlinear reaction schemes involv-
398
+ ing unimolecular and bimolecular reaction, and in the presence of slow/fast kinetics. The
399
+ structure of the reaction mechanism can be arbitrarily complicated without adding any fur-
400
+ ther complexity (other than purely notational) in the formulation of the stochastic evolution
401
+ expressed in terms of number of molecules. The only practical issue, is that the number
402
+ of different families of stochastic processes grows with the number of elementary reactive
403
+ processes considered. For instance, in the case of the subtrate-inhibited Michaelin-Menten
404
+ kinetics
405
+ E + S
406
+ k1
407
+
408
+ k−1 ES
409
+ ES
410
+ k2
411
+ → E + P
412
+ (11)
413
+ ES + S
414
+ k3
415
+
416
+ k−3 ESS
417
+ there are five reactive processes (five channels in the language of the Gillespie algorithm)
418
+ and consequently five families of counting processes {χ(h)
419
+ ih (t, ·)}, h = 1, . . . , 5, should be
420
+ 9
421
+
422
+ introduced, so that the formulation of the discrete stochastic dynamics reads
423
+ dNS(t)
424
+ dt
425
+ = −
426
+ NS(t)
427
+
428
+ i=1
429
+ dχ(1)
430
+ i (t, �k1 NE(t))
431
+ dt
432
+ +
433
+ NES(t)
434
+
435
+ j=1
436
+ dχ(2)
437
+ j (t, k−1)
438
+ dt
439
+ dNE(t)
440
+ dt
441
+ = −
442
+ NS(t)
443
+
444
+ i=1
445
+ dχ(1)
446
+ i (t, �k1 NE(t))
447
+ dt
448
+ +
449
+ NES(t)
450
+
451
+ j=1
452
+ dχ(2)
453
+ j (t, k−1)
454
+ dt
455
+ +
456
+ NES(t)
457
+
458
+ h=1
459
+ dχ(3)
460
+ h (t, k2)
461
+ dt
462
+ dNES(t)
463
+ dt
464
+ =
465
+ NS(t)
466
+
467
+ i=1
468
+ dχ(1)
469
+ i (t, �k1 NE(t))
470
+ dt
471
+
472
+ NES(t)
473
+
474
+ j=1
475
+ dχ(2)
476
+ j (t, k−1)
477
+ dt
478
+
479
+ NES(t)
480
+
481
+ h=1
482
+ dχ(3)
483
+ h (t, k2)
484
+ dt
485
+
486
+ NS(t)
487
+
488
+ k=1
489
+ dχ(4)
490
+ k (t, �k3 NES(t))
491
+ dt
492
+ +
493
+ NESS(t)
494
+
495
+ l=1
496
+ dχ(5)
497
+ l (t, k−3)
498
+ dt
499
+ (12)
500
+ dNESS(t)
501
+ dt
502
+ =
503
+ NS(t)
504
+
505
+ k=1
506
+ dχ(4)
507
+ k (t, �k3 NES(t))
508
+ dt
509
+
510
+ NESS(t)
511
+
512
+ l=1
513
+ dχ(5)
514
+ l (t, k−3)
515
+ dt
516
+ dNP(t)
517
+ dt
518
+ =
519
+ NES(t)
520
+
521
+ h=1
522
+ dχ(3)
523
+ h (t, k2)
524
+ dt
525
+ equipped with the initial conditions cS(0) = cS,0, cE(0) = cE,0, cES(0) = cESS(0) = cP(0) =
526
+ 0. Observe that for the bimolecular steps we have used a number-dependent rate coefficient.
527
+ This is just one possibility, out of other fully equivalent alternatives, of defining bimolecular
528
+ reacting processes, and out of tem a numerical algorithm for solving them. This issue, and
529
+ its computational implications will be addressed elsewhere [26]. The granularity number Ng
530
+ can be fixed, so that
531
+ NS(0) = [cS,0 Ng] ,
532
+ NE,0 = [cE,0 Ng]
533
+ (13)
534
+ where [ξ] indicates the integer part of ξ, thus defining the relation betwen Nα(t) and cα(t),
535
+ namely cα(t) = Nα(t)/Ng, α = S, E, ES, ESS, P. This implies also that the effective rate
536
+ parameters entering the discrete stochastic evolution equation (12), and associated with the
537
+ two bimolecular reactive steps, are given by �k1 = k1/Ng, and �k3 = k3/Ng.
538
+ Consider the case k−1 = k2 = k3 = k−3 = 1, cS,0 = 4, cE,0 = 0.1. In this case the
539
+ quasi steady-state approximation of the cES-cS curve (representing the slow manifold of the
540
+ kinetics takes the expression
541
+ cES =
542
+ cE,0 cS
543
+ KM + cS + β c2
544
+ S
545
+ ,
546
+ KM = k−1 + k2
547
+ k1
548
+ ,
549
+ β = k−3
550
+ k3
551
+ (14)
552
+ Figure 4 depicts the cES-cS graph obtained from a single realization of the stochastic process
553
+ eq. (11) at several values of k1 so as to modify the Michaelis-Menten constant KM for a
554
+ value Ng = 106 of the granularity number.
555
+ 10
556
+
557
+ 0
558
+ 0.02
559
+ 0.04
560
+ 0.06
561
+ 0.08
562
+ 0
563
+ 1
564
+ 2
565
+ 3
566
+ 4
567
+ cES
568
+ cS
569
+ FIG. 4. cES vs cS plot of the substrate-inhibited enzymatic kinetics discussed in the main text.
570
+ Symbols (in color) are the results of stochastic simulations of a single realization of the process eq.
571
+ (11), (black) solid lines the graph of the quasi steady-state approximation. The arrow indicates
572
+ increasing values of KM, i.e. decreasing values of k1 = 20, 6, 2.
573
+ Apart from the initial transient giving rise to an overshot in the values of cES near
574
+ cS ≃ cS,0, the dynamics rapidly collapses towards the slow manifold and the stochastic
575
+ simulations at high Ng-value provide a reliable description of the mean-field behavior starting
576
+ from a single stochastic realization.
577
+ To conclude, we want to point out some advantages and extensions of the present ap-
578
+ proach:
579
+ • it shows a direct analogy between chemical reaction kinetics, radiative processes and
580
+ stochastic formulation of open quantum systems, thus, paving the way for a unified
581
+ treatment of the interpaly between these phenomena, that is particularly important
582
+ in the field of photochemistry, and in the foundation of statistical physics [30, 31];
583
+ • it can be easily extended to semi-Markov transition. This is indeed the case of the
584
+ growth kinetics of eukaryotic microorganisms, the physiological state of which can
585
+ be parametrized with respect to internal (hidden) parameters such as the age, the
586
+ cytoplasmatic content, etc.;
587
+ • it can be easily extended to include transport phenomena. In point of fact, the oc-
588
+ currence of Markovian or semi-Markovian transitions in modeling chemical kinetics is
589
+ 11
590
+
591
+ analogous to the transitions occurring in the direction of motion (Poisson-Kac pro-
592
+ cesses, L´evy flights, Extended Poisson-Kac processes) or in the velocity (linearized
593
+ Boltzmannian schemes) [32–34].
594
+ • it is closely related to the formulation of stochastic differential equations for the ther-
595
+ malization of athermal system [35], in which the classical mesoscopic description of
596
+ thermal fluctuations, using the increments of a Wiener process, is replaced by a dy-
597
+ namic model involving the increments of a counting process.
598
+ Due to the limitations of a Letter, all these issues will be addressed in forthcoming works. But
599
+ apart for these extensions and improvements, the proposed formulation indicates that the
600
+ stochastic theory of chemical reactions can be built upon a simple and consistent mathemat-
601
+ ical formalism describing the elementary reactive events as Markovian or semi-Markovian
602
+ counting processes [36], that perfectly fits with the description of molecular non reactive
603
+ events (molecular collisions), providing an unifying stochastic formalism of elementary (clas-
604
+ sical and quantum) molecular events.
605
+ [1] P. L. Krapivsky, S. Redner, E. Ben-Naim, A Kinetic View to Statistical Physics, Cambridge
606
+ University Press, Cambridge (2010).
607
+ [2] L. Boltzmann, Weitere Studien ¨uber das W¨armeglichgenicht unter Gas-molek¨ulen, Sitzungs-
608
+ berichte Akademie der Wissenschaften 66 (1872) 275-370.
609
+ [3] G. B. Marin, G. S. Yablonsky, D. Constales, Kinetics of chemical reactions: decoding com-
610
+ plexity, John Wiley & Sons, New York, (2019).
611
+ [4] O. Levenspiel, Chemical Reaction Engineering, J. Wiley & Sons (1998).
612
+ [5] Z. Wang, Z. Hou, H. Xin, Internal noise stochastic resonance of synthetic gene network,
613
+ Chemical Physics Letters, 401 (1-3) (2005) 307-311.
614
+ [6] M. Perc, M. Gosak, and M. Marhl, From stochasticity to determinism in the collective dy-
615
+ namics of diffusively coupled cells, Chemical Physics Letters, 421 (1-3) (2006) 106–110.
616
+ [7] G. Lente, A binomial stochastic kinetic approach to the michaelis–menten mechanism, Chem-
617
+ ical Physics Letters, 568 (2013) 167–169.
618
+ 12
619
+
620
+ [8] D. A. McQuarrie, Stochastic approach to chemical kinetics, Journal of Applied Probability, 4
621
+ (3) (1967) 413-478.
622
+ [9] D. T. Gillespie, Stochastic simulation of chemical kinetics, Annual Review of Physical Chem-
623
+ istry 58 (1) (2007) 35-55.
624
+ [10] M. Delbr¨uck, Statistical fluctuations in autocatalytic reactions, The Journal of Chemical
625
+ Physics 8 (1) (1940) 120-124.
626
+ [11] A. F. Bartholomay, A stochastic approach to statistical kinetics with application to enzyme
627
+ kinetics, Biochemistry 1 (2) (1962) 223-230.
628
+ [12] D. T. Gillespie, A rigorous derivation of the chemical master equation, Physica A 188 (1-3)
629
+ (1992) 404-425.
630
+ [13] J. Keizer, On the necessity of using the master equation to describe the chemical reaction
631
+ X + A ⇋ B + X, Chemical Physics Letters, 10 (4) (1971) 371–374.
632
+ [14] B. J. Gaynor, R. G. Gilbert, K. D. King, Solution of the master equation for unimolecular
633
+ reactions, Chemical Physics Letters, 55 (1) (1978) 40-43.
634
+ [15] D. T. Gillespie, A general method for numerically simulating the stochastic time evolution of
635
+ coupled chemical reactions, Journal of Computational Physics 22 (4) (1976) 403-434.
636
+ [16] D. T. Gillespie, Exact stochastic simulation of coupled chemical reactions, The Journal of
637
+ Physical Chemistry 81 (25) (1977) 2340-2361.
638
+ [17] M. A. Gibson, J. Bruck, Efficient exact stochastic simulation of chemical systems with many
639
+ species and many channels, The Journal of Physical Chemistry A 104 (9) (2000) 1876-1889.
640
+ [18] L. Lok , R. Brent, Automatic generation of cellular reaction networks with Moleculizer, Nature
641
+ Biotechnology 23 (2005) 131–36
642
+ [19] Y. Cao, H. Li,L. R. Petzold, Efficient formulation of the stochastic simulation algorithm for
643
+ chemically reacting systems, The Journal of Chemical Physics 121 (2004) 4059–67.
644
+ [20] M. Rathinam, L. R. Petzold, Y. Cao, D. T. Gillespie, Stiffness in stochastic chemically reacting
645
+ systems: The implicit tau-leaping method, The Journal of Chemical Physics, 119 (24) (2003)
646
+ 12784-12794.
647
+ [21] C. Yang, D. T. Gillespie, L. R. Petzold, Efficient step size selection for the tau-leaping simu-
648
+ lation method, The Journal of Chemical Physics 124 (4) (2006) 044109.
649
+ [22] C. Yang, D. T. Gillespie, L. R. Petzold, Adaptive explicit-implicit tau-leaping method with
650
+ automatic tau selection, The Journal of Chemical Physics 126 (22) (2007) 224101.
651
+ 13
652
+
653
+ [23] D. C. Venerus and H. C. ¨Ottinger, A modern Course in Transport Phenomena, Cambridge
654
+ University Press, Cambridge (2018).
655
+ [24] K. Ito and H. P. McKean Jr., Diffusion Processes and their Sample Paths, Springer, Berlin
656
+ (1974).
657
+ [25] F. Campillo, M. Chebbi, S. Toumi, Stochastic modeling for biotechnologies Anaerobic model
658
+ AM2b, Revue Africaine de la de la Recherche en Informatique et Math´ematiques Appliqu´es,
659
+ INRIA 28 (2018 - 2019), Mathematics for Biology and the Environment 13-23.
660
+ [26] C. Pezzotti, M. Giona, Stochastic chemical reactions: from algorithmic approaches to stochas-
661
+ tic differential models, in preparation (2022).
662
+ [27] C. Pezzotti, G. Procopio, A. Brasiello, M. Giona, Stochastic simulations of bioreactors in
663
+ the presence of biomass heterogeneity and structured eukaryotic populations, in preparation
664
+ (2022).
665
+ [28] R. Aris, ”On the dispersion of a solute in a fluid flowing through a tube, Proceedings of the
666
+ Royal Society of London A (235) (1956) 67-77.
667
+ [29] S. Cerbelli, M. Giona, F. Garofalo, Quantifying dispersion of finite-sized particles in determin-
668
+ istic lateral displacement microflow separators through Brenner’s macrotransport paradigm,
669
+ Microfluidics and nanofluidics 15 (2013) 431-449.
670
+ [30] C. Pezzotti and M. Giona, Particle-photon radiative interactions and thermalization, in preper-
671
+ ation (2022).
672
+ [31] H.-P. Breuer, F. Petruccione, The Theory of Open Quantum Systems, Clarendon Press, Ox-
673
+ ford (2002).
674
+ [32] M. Giona, A. Brasiello, S. Crescitelli, Stochastic foundations of undulatory transport phe-
675
+ nomena: Generalized Poisson–Kac processes—Part I basic theory, Journal of Physics A (50)
676
+ (2017) 335002.
677
+ [33] M. Giona, A. Cairoli, R. Klages, Extended Poisson-Kac theory: A unifying framework for
678
+ stochastic processes with finite propagation velocity, Physical Review X (12) (2022) 021004.
679
+ [34] K.-I. Sato, L´evy processes and infinitely divisible distributions, Cambridge University Press,
680
+ Cambridge (1999).
681
+ [35] K. Kanazawa, Statistical Mechanics for Athermal Fluctuation, Springer Nature, Singapore
682
+ (2017).
683
+ [36] D. Cocco, M. Giona, Generalized Counting Processes in a Stochastic Environment. Mathe-
684
+ 14
685
+
686
+ matics 9 (2021) 25-73.
687
+ 15
688
+
4NE4T4oBgHgl3EQfbQzc/content/tmp_files/load_file.txt ADDED
@@ -0,0 +1,344 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf,len=343
2
+ page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
3
+ page_content='05072v1 [cond-mat.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
4
+ page_content='stat-mech] 12 Jan 2023 Chemical kinetics and stochastic differential equations Chiara Pezzotti and Massimiliano Giona1, ∗ 1Dipartimento di Ingegneria Chimica, Materiali, Ambiente La Sapienza Universit`a di Roma Via Eudossiana 18, 00184 Roma, Italy (Dated: January 13, 2023) Abstract We propose a general stochastic formalism for describing the evolution of chemical reactions involving a finite number of molecules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
5
+ page_content=' This approach is consistent with the statistical analysis based on the Chemical Master Equation, and provides the formal setting for the existing algorithmic approaches (Gillespie algorithm).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
6
+ page_content=' Some practical advantages of this formulation are addressed, and several examples are discussed pointing out the connection with quantum transitions (radiative interactions).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
7
+ page_content=' ∗ corresponding author:massimiliano.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
8
+ page_content='giona@uniroma1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
9
+ page_content='it 1 All the chemical physical processes involve, in an atomistic perspective, a stochastic de- scription of the events, be them reactive or associated with a change of phase (for instance adsorption) [1].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
10
+ page_content=' Nonetheless, in the overwhelming majority of the cases of practical and labo- ratory interest, the number of molecules involved is so large to justify a mean field approach, essentially based on the Boltzmannian hypothesis of molecular chaos (the “stosszahlansatz”) [2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
11
+ page_content=' The mean field formulation represents the backbone of the classical theory of chemical reaction kinetics [3, 4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
12
+ page_content=' It is well known that, in all the cases where the number of molecule is small (and this occurs in subcellular biochemical reactions, in nanoscale systems, or in the growth kinetics of microorganisms [5–7]), the effects of fluctuations become significant, motivating a stochastic description of chemical kinetic processes, involving the number of molecules present in the system, thus explicitly accounting for due to their finite number [8–11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
13
+ page_content=' The statistical theory of chemical kinetics in these conditions is grounded on the Chemical Master Equation (CME) [12, 13], expressing the evolution equation for the probabilities p(N, t) of all the possible number-configurations N(t) = (N1(t), .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
14
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
15
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
16
+ page_content=' , Ns(t)), where Nh(t) is the number of molecules of the h-th reacting species at time t, h = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
17
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
18
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
19
+ page_content=' , s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
20
+ page_content=' However, apart from a handful of simple cases, for which the CME can be solved analytically [14], numerical methods should be applied to it in order to compute mean values and higher-order moments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
21
+ page_content=' But also this choice reveals itself to be unfeasible in most of the situations of practical and theoretical interests, due to the extremely large number of configurations involved, making the multi- index matrix p(N, t) so huge to exceed reasonable computational facilities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
22
+ page_content=' In order to solve this problem, Gillespie proposed an algorithmic solution to the numeri- cal simulation of stochastic reacting systems, based on the Markovian nature of the reactive events [15, 16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
23
+ page_content=' The original Gillespie algorithm has been extended and improved over time, providing a variety of slightly different computational alternatives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
24
+ page_content=' A common denominator of the first family of the Gillespie algorithms (namely those based on the direct method, the first reaction method or their derivates [17–19]) is to associate to every time step the occur- rence of just one reaction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
25
+ page_content=' This formulation comes directly from the assumption that, if the time step is small enough, the probability that more than one reaction will occur is negligi- ble.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
26
+ page_content=' While correct, this choice brings to significant computational costs for complex reaction schemes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
27
+ page_content=' This problem has been highlighted several times, from the Gillespie group itself, as stiffness in stochastic chemical reacting systems [20].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
28
+ page_content=' A brilliant way to overcome this 2 limit originates the famous tau-leaping method, which, unfortunately, requires to check that the propensity functions remain almost constant at each iteration and can be applied just if this condition is verified [21, 22].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
29
+ page_content=' The algorithmic solution associated with the formalism here introduced combines the accuracy of the first SSA with the computational advantages of the τ-leaping method.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
30
+ page_content=' There is, moreover, a missing link between the CME theory and the Gillespie algorithm, consisting in the straight mathematical formulation of the stochastic differential equations associated with a chemical reacting system, the statistical description of which would cor- respond to the CME.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
31
+ page_content=' To clarify this issue, consider the conceptually analogous problem of particle diffusion over the real line, the statistical description of which is expressed by the parabolic equation ∂p(x, t)/∂t = D ∂2p(x, t)/∂x2, for the probability density p(x, t) of finding a particle at position x at time t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
32
+ page_content=' Setting xn = x(n ∆t), an algorithm describing this process can be simply expressed by the discrete evolution equation xn+1 = xn + √ 2 D ∆t rn+1, where rh, h = 1, 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
33
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
34
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
35
+ page_content=' represent independent random variables sampled from a normal distribution (with zero mean, and unit variance) [23].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
36
+ page_content=' This represents an efficient algorithmic solution of the problem, whenever the time resolution ∆t is small enough.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
37
+ page_content=' Nevertheless, the mere algorithmic approach cannot be considered physically satisfactory, in a comprehensive for- mulation of transport theory embedded in a continuous space-time (in which both position x and time t are real valued).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
38
+ page_content=' In point of fact, only with the mathematical formulation due to K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
39
+ page_content=' Ito of stochastic differential equations driven by the increments dw(t) of a Wiener process (Langevin equations) [24], namely dx(t) = √ 2 D dw(t) the theory of diffusive motion has found a proper mathematical physical setting.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
40
+ page_content=' A similar situation applies to the case of stochastic models of chemical reaction kinetics, and the present Letter is aimed at filling this gap.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
41
+ page_content=' The basic idea is that any reactive process corresponds to a system of elementary events (the single reaction) possessing a Markovian transitional structure, and, consequently, amenable to a description by means of the increments of counting processes (Poisson processes, in the Markovian case).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
42
+ page_content=' This topic has been also pointed out in [25] in terms of Poisson measures, although the latter formulation is much less simple and physically intuitive than the approach proposed in the present Letter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
43
+ page_content=' To begin with, consider the simple case of a first-order chemical reaction A k1 ⇋ k−1 B (for instance, an isomerization).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
44
+ page_content=' This model is perfectly analogous to the radiative transition 3 FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
45
+ page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
46
+ page_content=' Schematic representation of the analogy between a two-level quantum system and a first- order chemical kinetics, such as an isomerization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
47
+ page_content=' of a molecule possessing two energy states, due to emission and adsorption of an energy quantum (figure 1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
48
+ page_content=' Let NA(0) + NB(0) = Ng the total number of molecules at time t = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
49
+ page_content=' The state of the system is characterized by the state functions σh(t), h = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
50
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
51
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
52
+ page_content=' , Ng for each molecule, attaining values {0, 1}, and such that σh(t) = 0 if the energy state at time t is E0 (or equivalently if the molecule finds itself in the state A), and σh(t) = 1 in the opposite case (energy state E1, or isomeric state B).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
53
+ page_content=' Let {χ(1) h (t, k1), χ(2) h (t, k−1)}Ng h=1 be two systems of independent Poisson processes, char- acterized by the transition rates k1, and k−1, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
54
+ page_content=' The evolution of σh(t) can be expressed via the stochastic differential equation dσh(t) dt = (1 − σh(t)) dχ(1) h (t, k1) dt − σh(t) dχ(2) h (t, k−1) dt (1) h = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
55
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
56
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
57
+ page_content=' , Ng, where dχ(t, λ)/dt is the distributional derivative of the Poisson process χ(t, λ), corresponding to a sequence of unit impulsive functions at the transition instants t∗ i , i = 1, 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
58
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
59
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
60
+ page_content=' , 0 < t∗ i < t∗ i+1, where for ε > 0, limε→0 � t∗ i +ε t∗ i −ε dχ(t, λ) = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
61
+ page_content=' Summing over h = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
62
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
63
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
64
+ page_content=' Ng, and observing that NA(t) = �Ng h=1 (1 − σh(t)), NB(t) = �Ng h=1 σh(t), we have dNB(t) dt = NA(t) � h=1 dχ(1) h (t, k1) dt − NB(t) � h=1 dχ(2) h (t, k−1) dt (2) 4 (b)(a)OPand dNA(t)/dt = −dNB(t)/dt, representing the evolution equation for NA(t) and NB(t), attaining integer values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
65
+ page_content=' The stochastic evolution of the number of molecules NA(t), NB(t) is thus expressed as a differential equation with respect to the continuous physical time t ∈ R+, over the increments of a Poisson process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
66
+ page_content=' Intepreted in a mean-field way, if ctot is the overall concentration of the reactants at time t = 0, then the concentrations cα(t) at time t can be recovered from eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
67
+ page_content=' (2) as cα(t) = ctot Nα(t) Ng , α = A, B (3) representing the calibration relation connecting the stochastic description in terms of num- ber of molecules Nα(t) and the concentrations cα(t), α = A, B entering the mean-field description.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
68
+ page_content=' The analytical formulation of a stochastic differential equation for chemical kinetics, expressed in terms of the number of molecules of the chemical species involved, rather than an algorithm defined for discretized times, permits to develop a variety of different numerical strategies, that naturally perform a modified tau-leaping procedure, as the occurrence of several distinct reactive events in any elementary time step ∆t is intrinsically accounted for.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
69
+ page_content=' This can be easily seen by considering the simple reaction defined by the evolution equation (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
70
+ page_content=' In terms of increments, eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
71
+ page_content=' (2) can be written as dNB(t) = �NA(t) h=1 dχ(1)(t, k1) − �NB(t) h=1 dχ(2)(t, k−1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
72
+ page_content=' If ∆t is the chosen time step, it follows from this formulation, a simple numerical approximation for eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
73
+ page_content=' (2), namely, ∆NB(t) = NB(t + ∆t) − NB(t) = NA(t) � h=1 ξ(1) h (k1 ∆t) − NB(t) � h=1 ξ(2) h (k−1 ∆t) (4) where ξ(1)(k1 ∆t), ξ(2) h (k−1 ∆t) h = 1, 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
74
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
75
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
76
+ page_content=' , are two families of independent binary random variables, where ξ(α) h (p) = \uf8f1 \uf8f2 \uf8f3 1 with probability p 0 otherwise (5) α = 1, 2, h = 1, 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
77
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
78
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
79
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
80
+ page_content=' The time step ∆t, can be chosen in eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
81
+ page_content=' (4) from the condition K ∆t < 1 , K = max{k1, k−1} (6) In practice, we choose ∆t = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
82
+ page_content='1/K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
83
+ page_content=' As can be observed, the choice of ∆t is limited by the intrinsic rates of the process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
84
+ page_content=' The advantage of deriving different algorithmic schemes for 5 solving numerically the stochastic kinetic equations becomes more evident in dealing with bimolecular reactions (addressed below).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
85
+ page_content=' Due to the intrinsic limitations of this commu- nication, a thorouh discussion of this issue is postponed to a future more extensive article [26].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
86
+ page_content=' The same approach can be extended to include amongst the elementary events not only the reactive steps, but also feeding conditions, thus representing the evolution of chemically reacting systems with a finite number of molecules in a perfectly stirred open reactor.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
87
+ page_content=' This is the case of the tank-loading problem, in which a tracer is injected in an open vessel assumed perfectly mixed, for which, in the absence of chemical reactions, the mean field equation for the concentration of the tracer reads dc(t) dt = D (c0 − c(t)) (7) where c0 is the inlet concentration and D the dilution rate (reciprocal of the mean retention time), and c(0) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
88
+ page_content=' Fixing Ng so that c(t) = c0 N(t)/Ng, the corresponding stochastic differential equation for the integer N(t) involves, also in this case, two families of counting processes, one for the loading at constant concentration c0, and the other for tracer discharge in the outlet stream, characterized by the same transition rate D, dN(t) dt = Ng � h=1 dχ(1) h (t, D) dt − N(t) � k=1 dχ(2) h (t, D) dt (8) starting from N(0) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
89
+ page_content=' Figure 2 depicts several realizations of the tank-loading process, obtained by discretizing eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
90
+ page_content=' (8) with a time step ∆t = 10−3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
91
+ page_content=' Despite the simplicity of the process, this example permits to highlight the role of Ng, that can be referred to as the granu- larity number, and the way stochastic models of chemical reactions can be fruitfully applied.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
92
+ page_content=' Indeed, there is a two-fold use of the stochastic formulation of chemical kinetic schemes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
93
+ page_content=' The first refers to a chemical reacting system involving a small number of molecules, and in this case Ng represents the effective number of molecules present in the system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
94
+ page_content=' The other is to use stochastic algorithms for simulating reacting systems in an alternative (and sometimes more efficient way) with respect to the solution of the corresponding mean-field equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
95
+ page_content=' In the latter case, the granularity number Ng represents essentially a computa- tional parameter, tuning the intensity of the fluctuations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
96
+ page_content=' Two choices are then possible: (i) it can be chosen large enough, in order to obtain from a single realization of the process an accurate approximation of the mean-field behavior, or (ii) it can be chosen small enough 6 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
97
+ page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
98
+ page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
99
+ page_content='2 0 2 4 6 8 10 c(t) t 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
100
+ page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
101
+ page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
102
+ page_content='2 0 2 4 6 8 10 c(t) t 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
103
+ page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
104
+ page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
105
+ page_content='2 0 2 4 6 8 10 c(t) t (a) (b) (c) FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
106
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
107
+ page_content=' c(t) = N(t)/Ng vs t from a single realization of the tank-loading process eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
108
+ page_content=' (8) with D = 1, c0 = 1 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
109
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
110
+ page_content='. Panel (a): Ng = 30, panel (b) Ng = 100, panel (c) Ng = 1000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
111
+ page_content=' The solid horizontal lines represent the steady-state value c∗ = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
112
+ page_content=' in order, to deal with extremely fast simulations of a single realization of the process, that could be averaged over a statistically significant number of realizations in due time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
113
+ page_content=' These two choices are depicted in figure 2 (panel c), choosing Ng = 103, and in figure 3 panel (a) obtained for Ng = 30.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
114
+ page_content=' Of course, the latter approach is valid as long as the low-granularity (low values of Ng) does not influence the qualitative properties of the kinetics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
115
+ page_content=' The second (computational) use of stochastic simulations of chemical kinetics requires a further discussion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
116
+ page_content=' At a first sight, it may appear that any stochastic simulation would be computationally less efficient than the solution of the corresponding mean-field equations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
117
+ page_content=' This is certainly true for classical chemical reaction schemes in a perfectly mixed system, 7 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
118
+ page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
119
+ page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
120
+ page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
121
+ page_content='8 1 0 2 4 6 8 10 <c>(t) t 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
122
+ page_content='05 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
123
+ page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
124
+ page_content='15 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
125
+ page_content='2 0 2 4 6 8 10 a b σc(t) t (a) (b) FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
126
+ page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
127
+ page_content=' Panel (a): ⟨c⟩(t) vs t at Ng = 30 (symbols) averaged over [106/Ng] realizations of the tank-loading process with D = 1, c0 = 1 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
128
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
129
+ page_content=' Here, [·] indicates the integer part of its argument.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
130
+ page_content=' The solid line represents the mean-field result ⟨c⟩(t) = 1 − e−t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
131
+ page_content=' Panel (b): Variance σc(t) vs t for the tank-loading process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
132
+ page_content=' Symbols are the results of stochastic simulations of eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
133
+ page_content=' (8) averaged over [106/Ng] realizations, lines the solutions of eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
134
+ page_content=' (10).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
135
+ page_content=' Line (a) refers to Ng = 30, line (b) to Ng = 100.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
136
+ page_content=' for which the mean-field model reduces to a system of ordinary differential equations for the concentrations of the reactants.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
137
+ page_content=' But there are kinetic problems e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
138
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
139
+ page_content=', associated with the growth of microorganisms and eukaryotic cell lines in bioreactors (these growth phenom- ena, are indeed amenable to a description in terms of equivalent chemical reactions), the mean-field model of which is expressed in the form of higher-dimensional nonlinear integro- differential equations .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
140
+ page_content=' For this class of problems, the use of stochastic simulations is the 8 most efficient, if not the only way to achieve a quantitative description of the process, in those cases where the number np of internal parameters describing the physiological state of an eukaryotic cell becomes large enough, np ≥ 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
141
+ page_content=' This issue is addressed in detail in [27].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
142
+ page_content=' This case is altogether similar to some transport problems, such as Taylor-Aris dispersion for high P´eclet numbers or the analysis of microfluidic separation processes (DLD devices) for which the stochastic simulation of particle motion is far more efficient that the corresponding solution of the corresponding mean-field model expressed in the form of advection-diffusion equations [28, 29].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
143
+ page_content=' To complete the analysis of the tank-loading problem, the associated CME reads dp(n, t) dt = D Ng [p(n − 1, t) ηn−1 − p(n, t)] + D [(n + 1) p(n + 1, t) − n p(n, t)] (9) where ηh = 1 for h ≥ 0 and ηh = 0 otherwise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
144
+ page_content=' It follows that ⟨c⟩(t) = c0 �∞ n=1 n p(n, t)/Ng satisfies identically the mean-field equation (due to the linearity of the problem), while the variance σc(t), with σ2 c(t) = c2 0 �∞ n=1 n2 p(n, t)/N2 g − (c0 �∞ n=1 n p(n, t)/Ng)2, satisfies the equation dσ2 c dt = −2 D σ2 c + D � 1 Ng + ⟨c⟩ Ng � (10) Figure 3 panel (b) compares the results of stochastic simulations against the solutions of eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
145
+ page_content=' (10) for two values of Ng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
146
+ page_content=' The above approach can be extended to any system of nonlinear reaction schemes involv- ing unimolecular and bimolecular reaction, and in the presence of slow/fast kinetics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
147
+ page_content=' The structure of the reaction mechanism can be arbitrarily complicated without adding any fur- ther complexity (other than purely notational) in the formulation of the stochastic evolution expressed in terms of number of molecules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
148
+ page_content=' The only practical issue, is that the number of different families of stochastic processes grows with the number of elementary reactive processes considered.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
149
+ page_content=' For instance, in the case of the subtrate-inhibited Michaelin-Menten kinetics E + S k1 ⇋ k−1 ES ES k2 → E + P (11) ES + S k3 ⇋ k−3 ESS there are five reactive processes (five channels in the language of the Gillespie algorithm) and consequently five families of counting processes {χ(h) ih (t, ·)}, h = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
150
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
151
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
152
+ page_content=' ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
153
+ page_content=' 5,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
154
+ page_content=' should be 9 introduced,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
155
+ page_content=' so that the formulation of the discrete stochastic dynamics reads dNS(t) dt = − NS(t) � i=1 dχ(1) i (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
156
+ page_content=' �k1 NE(t)) dt + NES(t) � j=1 dχ(2) j (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
157
+ page_content=' k−1) dt dNE(t) dt = − NS(t) � i=1 dχ(1) i (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
158
+ page_content=' �k1 NE(t)) dt + NES(t) � j=1 dχ(2) j (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
159
+ page_content=' k−1) dt + NES(t) � h=1 dχ(3) h (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
160
+ page_content=' k2) dt dNES(t) dt = NS(t) � i=1 dχ(1) i (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
161
+ page_content=' �k1 NE(t)) dt − NES(t) � j=1 dχ(2) j (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
162
+ page_content=' k−1) dt − NES(t) � h=1 dχ(3) h (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
163
+ page_content=' k2) dt − NS(t) � k=1 dχ(4) k (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
164
+ page_content=' �k3 NES(t)) dt + NESS(t) � l=1 dχ(5) l (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
165
+ page_content=' k−3) dt (12) dNESS(t) dt = NS(t) � k=1 dχ(4) k (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
166
+ page_content=' �k3 NES(t)) dt − NESS(t) � l=1 dχ(5) l (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
167
+ page_content=' k−3) dt dNP(t) dt = NES(t) � h=1 dχ(3) h (t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
168
+ page_content=' k2) dt equipped with the initial conditions cS(0) = cS,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
169
+ page_content='0,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
170
+ page_content=' cE(0) = cE,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
171
+ page_content='0,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
172
+ page_content=' cES(0) = cESS(0) = cP(0) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
173
+ page_content=' Observe that for the bimolecular steps we have used a number-dependent rate coefficient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
174
+ page_content=' This is just one possibility, out of other fully equivalent alternatives, of defining bimolecular reacting processes, and out of tem a numerical algorithm for solving them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
175
+ page_content=' This issue, and its computational implications will be addressed elsewhere [26].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
176
+ page_content=' The granularity number Ng can be fixed, so that NS(0) = [cS,0 Ng] , NE,0 = [cE,0 Ng] (13) where [ξ] indicates the integer part of ξ, thus defining the relation betwen Nα(t) and cα(t), namely cα(t) = Nα(t)/Ng, α = S, E, ES, ESS, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
177
+ page_content=' This implies also that the effective rate parameters entering the discrete stochastic evolution equation (12), and associated with the two bimolecular reactive steps, are given by �k1 = k1/Ng, and �k3 = k3/Ng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
178
+ page_content=' Consider the case k−1 = k2 = k3 = k−3 = 1, cS,0 = 4, cE,0 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
179
+ page_content='1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
180
+ page_content=' In this case the quasi steady-state approximation of the cES-cS curve (representing the slow manifold of the kinetics takes the expression cES = cE,0 cS KM + cS + β c2 S , KM = k−1 + k2 k1 , β = k−3 k3 (14) Figure 4 depicts the cES-cS graph obtained from a single realization of the stochastic process eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
181
+ page_content=' (11) at several values of k1 so as to modify the Michaelis-Menten constant KM for a value Ng = 106 of the granularity number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
182
+ page_content=' 10 0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
183
+ page_content='02 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
184
+ page_content='04 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
185
+ page_content='06 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
186
+ page_content='08 0 1 2 3 4 cES cS FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
187
+ page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
188
+ page_content=' cES vs cS plot of the substrate-inhibited enzymatic kinetics discussed in the main text.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
189
+ page_content=' Symbols (in color) are the results of stochastic simulations of a single realization of the process eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
190
+ page_content=' (11), (black) solid lines the graph of the quasi steady-state approximation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
191
+ page_content=' The arrow indicates increasing values of KM, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
192
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
193
+ page_content=' decreasing values of k1 = 20, 6, 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
194
+ page_content=' Apart from the initial transient giving rise to an overshot in the values of cES near cS ≃ cS,0, the dynamics rapidly collapses towards the slow manifold and the stochastic simulations at high Ng-value provide a reliable description of the mean-field behavior starting from a single stochastic realization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
195
+ page_content=' To conclude, we want to point out some advantages and extensions of the present ap- proach: it shows a direct analogy between chemical reaction kinetics, radiative processes and stochastic formulation of open quantum systems, thus, paving the way for a unified treatment of the interpaly between these phenomena, that is particularly important in the field of photochemistry, and in the foundation of statistical physics [30, 31];' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
196
+ page_content=' it can be easily extended to semi-Markov transition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
197
+ page_content=' This is indeed the case of the growth kinetics of eukaryotic microorganisms, the physiological state of which can be parametrized with respect to internal (hidden) parameters such as the age, the cytoplasmatic content, etc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
198
+ page_content=' ;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
199
+ page_content=' it can be easily extended to include transport phenomena.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
200
+ page_content=' In point of fact, the oc- currence of Markovian or semi-Markovian transitions in modeling chemical kinetics is 11 analogous to the transitions occurring in the direction of motion (Poisson-Kac pro- cesses, L´evy flights, Extended Poisson-Kac processes) or in the velocity (linearized Boltzmannian schemes) [32–34].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
201
+ page_content=' it is closely related to the formulation of stochastic differential equations for the ther- malization of athermal system [35], in which the classical mesoscopic description of thermal fluctuations, using the increments of a Wiener process, is replaced by a dy- namic model involving the increments of a counting process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
202
+ page_content=' Due to the limitations of a Letter, all these issues will be addressed in forthcoming works.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
203
+ page_content=' But apart for these extensions and improvements,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
204
+ page_content=' the proposed formulation indicates that the stochastic theory of chemical reactions can be built upon a simple and consistent mathemat- ical formalism describing the elementary reactive events as Markovian or semi-Markovian counting processes [36],' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
205
+ page_content=' that perfectly fits with the description of molecular non reactive events (molecular collisions),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
206
+ page_content=' providing an unifying stochastic formalism of elementary (clas- sical and quantum) molecular events.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
207
+ page_content=' [1] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
208
+ page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
209
+ page_content=' Krapivsky, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
210
+ page_content=' Redner, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
211
+ page_content=' Ben-Naim, A Kinetic View to Statistical Physics, Cambridge University Press, Cambridge (2010).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
212
+ page_content=' [2] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
213
+ page_content=' Boltzmann, Weitere Studien ¨uber das W¨armeglichgenicht unter Gas-molek¨ulen, Sitzungs- berichte Akademie der Wissenschaften 66 (1872) 275-370.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
214
+ page_content=' [3] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
215
+ page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
216
+ page_content=' Marin, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
217
+ page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
218
+ page_content=' Yablonsky, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
219
+ page_content=' Constales, Kinetics of chemical reactions: decoding com- plexity, John Wiley & Sons, New York, (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
220
+ page_content=' [4] O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
221
+ page_content=' Levenspiel, Chemical Reaction Engineering, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
222
+ page_content=' Wiley & Sons (1998).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
223
+ page_content=' [5] Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
224
+ page_content=' Wang, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
225
+ page_content=' Hou, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
226
+ page_content=' Xin, Internal noise stochastic resonance of synthetic gene network, Chemical Physics Letters, 401 (1-3) (2005) 307-311.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
227
+ page_content=' [6] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
228
+ page_content=' Perc, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
229
+ page_content=' Gosak, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
230
+ page_content=' Marhl, From stochasticity to determinism in the collective dy- namics of diffusively coupled cells, Chemical Physics Letters, 421 (1-3) (2006) 106–110.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
231
+ page_content=' [7] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
232
+ page_content=' Lente, A binomial stochastic kinetic approach to the michaelis–menten mechanism, Chem- ical Physics Letters, 568 (2013) 167–169.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
233
+ page_content=' 12 [8] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
234
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
235
+ page_content=' McQuarrie, Stochastic approach to chemical kinetics, Journal of Applied Probability, 4 (3) (1967) 413-478.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
236
+ page_content=' [9] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
237
+ page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
238
+ page_content=' Gillespie, Stochastic simulation of chemical kinetics, Annual Review of Physical Chem- istry 58 (1) (2007) 35-55.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
239
+ page_content=' [10] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
240
+ page_content=' Delbr¨uck, Statistical fluctuations in autocatalytic reactions, The Journal of Chemical Physics 8 (1) (1940) 120-124.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
241
+ page_content=' [11] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
242
+ page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
243
+ page_content=' Bartholomay, A stochastic approach to statistical kinetics with application to enzyme kinetics, Biochemistry 1 (2) (1962) 223-230.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
244
+ page_content=' [12] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
245
+ page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
246
+ page_content=' Gillespie, A rigorous derivation of the chemical master equation, Physica A 188 (1-3) (1992) 404-425.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
247
+ page_content=' [13] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
248
+ page_content=' Keizer, On the necessity of using the master equation to describe the chemical reaction X + A ⇋ B + X, Chemical Physics Letters, 10 (4) (1971) 371–374.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
249
+ page_content=' [14] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
250
+ page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
251
+ page_content=' Gaynor, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
252
+ page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
253
+ page_content=' Gilbert, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
254
+ page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
255
+ page_content=' King, Solution of the master equation for unimolecular reactions, Chemical Physics Letters, 55 (1) (1978) 40-43.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
256
+ page_content=' [15] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
257
+ page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
258
+ page_content=' Gillespie, A general method for numerically simulating the stochastic time evolution of coupled chemical reactions, Journal of Computational Physics 22 (4) (1976) 403-434.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
259
+ page_content=' [16] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
260
+ page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
261
+ page_content=' Gillespie, Exact stochastic simulation of coupled chemical reactions, The Journal of Physical Chemistry 81 (25) (1977) 2340-2361.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
262
+ page_content=' [17] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
263
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
264
+ page_content=' Gibson, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
265
+ page_content=' Bruck, Efficient exact stochastic simulation of chemical systems with many species and many channels, The Journal of Physical Chemistry A 104 (9) (2000) 1876-1889.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
266
+ page_content=' [18] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
267
+ page_content=' Lok , R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
268
+ page_content=' Brent, Automatic generation of cellular reaction networks with Moleculizer, Nature Biotechnology 23 (2005) 131–36 [19] Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
269
+ page_content=' Cao, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
270
+ page_content=' Li,L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
271
+ page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
272
+ page_content=' Petzold, Efficient formulation of the stochastic simulation algorithm for chemically reacting systems, The Journal of Chemical Physics 121 (2004) 4059–67.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
273
+ page_content=' [20] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
274
+ page_content=' Rathinam, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
275
+ page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
276
+ page_content=' Petzold, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
277
+ page_content=' Cao, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
278
+ page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
279
+ page_content=' Gillespie, Stiffness in stochastic chemically reacting systems: The implicit tau-leaping method, The Journal of Chemical Physics, 119 (24) (2003) 12784-12794.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
280
+ page_content=' [21] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
281
+ page_content=' Yang, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
282
+ page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
283
+ page_content=' Gillespie, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
284
+ page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
285
+ page_content=' Petzold, Efficient step size selection for the tau-leaping simu- lation method, The Journal of Chemical Physics 124 (4) (2006) 044109.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
286
+ page_content=' [22] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
287
+ page_content=' Yang, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
288
+ page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
289
+ page_content=' Gillespie, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
290
+ page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
291
+ page_content=' Petzold, Adaptive explicit-implicit tau-leaping method with automatic tau selection, The Journal of Chemical Physics 126 (22) (2007) 224101.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
292
+ page_content=' 13 [23] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
293
+ page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
294
+ page_content=' Venerus and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
295
+ page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
296
+ page_content=' ¨Ottinger, A modern Course in Transport Phenomena, Cambridge University Press, Cambridge (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
297
+ page_content=' [24] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
298
+ page_content=' Ito and H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
299
+ page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
300
+ page_content=' McKean Jr.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
301
+ page_content=', Diffusion Processes and their Sample Paths, Springer, Berlin (1974).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
302
+ page_content=' [25] F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
303
+ page_content=' Campillo, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
304
+ page_content=' Chebbi, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
305
+ page_content=' Toumi, Stochastic modeling for biotechnologies Anaerobic model AM2b, Revue Africaine de la de la Recherche en Informatique et Math´ematiques Appliqu´es, INRIA 28 (2018 - 2019), Mathematics for Biology and the Environment 13-23.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
306
+ page_content=' [26] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
307
+ page_content=' Pezzotti, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
308
+ page_content=' Giona, Stochastic chemical reactions: from algorithmic approaches to stochas- tic differential models, in preparation (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
309
+ page_content=' [27] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
310
+ page_content=' Pezzotti, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
311
+ page_content=' Procopio, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
312
+ page_content=' Brasiello, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
313
+ page_content=' Giona, Stochastic simulations of bioreactors in the presence of biomass heterogeneity and structured eukaryotic populations, in preparation (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
314
+ page_content=' [28] R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
315
+ page_content=' Aris, ”On the dispersion of a solute in a fluid flowing through a tube, Proceedings of the Royal Society of London A (235) (1956) 67-77.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
316
+ page_content=' [29] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
317
+ page_content=' Cerbelli, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
318
+ page_content=' Giona, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
319
+ page_content=' Garofalo, Quantifying dispersion of finite-sized particles in determin- istic lateral displacement microflow separators through Brenner’s macrotransport paradigm, Microfluidics and nanofluidics 15 (2013) 431-449.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
320
+ page_content=' [30] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
321
+ page_content=' Pezzotti and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
322
+ page_content=' Giona, Particle-photon radiative interactions and thermalization, in preper- ation (2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
323
+ page_content=' [31] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
324
+ page_content='-P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
325
+ page_content=' Breuer, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
326
+ page_content=' Petruccione, The Theory of Open Quantum Systems, Clarendon Press, Ox- ford (2002).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
327
+ page_content=' [32] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
328
+ page_content=' Giona, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
329
+ page_content=' Brasiello, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
330
+ page_content=' Crescitelli, Stochastic foundations of undulatory transport phe- nomena: Generalized Poisson–Kac processes—Part I basic theory, Journal of Physics A (50) (2017) 335002.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
331
+ page_content=' [33] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
332
+ page_content=' Giona, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
333
+ page_content=' Cairoli, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
334
+ page_content=' Klages, Extended Poisson-Kac theory: A unifying framework for stochastic processes with finite propagation velocity, Physical Review X (12) (2022) 021004.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
335
+ page_content=' [34] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
336
+ page_content='-I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
337
+ page_content=' Sato, L´evy processes and infinitely divisible distributions, Cambridge University Press, Cambridge (1999).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
338
+ page_content=' [35] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
339
+ page_content=' Kanazawa, Statistical Mechanics for Athermal Fluctuation, Springer Nature, Singapore (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
340
+ page_content=' [36] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
341
+ page_content=' Cocco, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
342
+ page_content=' Giona, Generalized Counting Processes in a Stochastic Environment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
343
+ page_content=' Mathe- 14 matics 9 (2021) 25-73.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
344
+ page_content=' 15' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/4NE4T4oBgHgl3EQfbQzc/content/2301.05072v1.pdf'}
4dE1T4oBgHgl3EQfmQQK/content/2301.03294v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:47d4b0d29f700f9d98ef98273ffbb55dbfa3baaec534a0f52eeca84885709be3
3
+ size 143997
4dE1T4oBgHgl3EQfmQQK/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb923a5f4d50604a55fe841aafb276d677a23634a0b9cc992c6a89fedc4c3038
3
+ size 1835053
4dE1T4oBgHgl3EQfmQQK/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b398367ce5296935cfa179832ec371080587295dbdd6987019b799025b564836
3
+ size 64185
4tFJT4oBgHgl3EQfkCwt/content/2301.11577v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:21652d68e533882a5f04bd0431b1f57a1ebf529f9875d525c24de7a8879b1882
3
+ size 432790
4tFJT4oBgHgl3EQfkCwt/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:19ae3edb74421dfb699d1c36d40bef1ce5af2b48c3d811332919d186f37a36b8
3
+ size 1966125
4tFJT4oBgHgl3EQfkCwt/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0b9aa0d2e1f84aaceec16a656ffd50a51910312ca549851d3d9e22284cfa41c9
3
+ size 82341
6dAzT4oBgHgl3EQf-P4i/content/2301.01930v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:300a10add628570ecd83f4edf00138d3ac98774bf8baa8ba5a959fc01d84a7cb
3
+ size 3227275
6dAzT4oBgHgl3EQf-P4i/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:720858774ac48a4c95232460c7571bbe9a8d2c816b4524c7298aa98ac68d7a4c
3
+ size 153947
79E2T4oBgHgl3EQfPgaW/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
7NAyT4oBgHgl3EQf2vm0/content/2301.00757v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6db7262ed7dc2aa0f46d3a971237bd09ee549e1f4f9f4877451a33ccf4ff5b59
3
+ size 2232183
7NAyT4oBgHgl3EQf2vm0/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:34c48973a8899abd5329bc00ccc558fdf1bde514c67a03c097c4f135f677618c
3
+ size 3407917
7NAyT4oBgHgl3EQf2vm0/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5f0be2f484c5543d4847eb1d705951df4840ac587b3ddbf524ea9f5f7f2e2cac
3
+ size 142311
9dFAT4oBgHgl3EQfpx0O/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:45f14d75b3dbc8057feb562d2f64521bce31122f81baf7ad86739ab99ec5f18e
3
+ size 160256
9dFRT4oBgHgl3EQfqjdh/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9738d94488a58de813c9f2ddde6d6f0097930decbc76f1f89cd2580585400795
3
+ size 17956909
ANFQT4oBgHgl3EQfMDYd/content/2301.13266v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c768dec977730cfb5d3903fb0d1c35e1d274a47888d4f3804d5e55ba957d5905
3
+ size 330246
ANFQT4oBgHgl3EQfMDYd/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d0b979ee27318feb2357a22fe683fc6f12ec51399c0540cde79d90ebe7095057
3
+ size 155712
B9FAT4oBgHgl3EQfsh6e/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9f848ab8330711c2280767532827bc8a40b020e0d88223f5feb433ecb11c971e
3
+ size 10485805
C9A0T4oBgHgl3EQfAf_H/content/2301.01964v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fb18ee4d7634eb4ed637f8c3deae8a6d3cb5b16d5974f921302691e2e6789459
3
+ size 1750578
C9A0T4oBgHgl3EQfAf_H/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c5755a466c4d409922ceb92366d780ae828029efc5f7edf0763e9c6674ad7a5b
3
+ size 151909
C9AzT4oBgHgl3EQfTvzN/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:611337e9be380c4f7b6ef3e53dfcb4cc5c6af57b64b41de0e900fffa3f8f7ad0
3
+ size 321156
C9FQT4oBgHgl3EQf_jdA/content/tmp_files/2301.13458v1.pdf.txt ADDED
@@ -0,0 +1,1107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ MNRAS 000, 000–000 (0000)
2
+ Preprint 1 February 2023
3
+ Compiled using MNRAS LATEX style file v3.0
4
+ Tracing of Magnetic field with gradients: Sub-Sonic Turbulence
5
+ K. W. Ho, 1 ★ A. Lazarian, 1,2 †
6
+ 1Department of Astronomy, University of Wisconsin-Madison, Madison, WI, 53706, USA
7
+ 2Centro de Investigación en Astronomía, Universidad Bernardo O’Higgins, Santiago, General Gana 1760, 8370993, Chile
8
+ Accepted 2023 January 12. Received 2023 January 12; in original form 2022 March 21
9
+ ABSTRACT
10
+ Recent development of the velocity gradient technique shows the capability of the technique in the way of tracing magnetic
11
+ fields morphology in diffuse interstellar gas and molecular clouds. In this paper, we perform the numerical systemic study of the
12
+ performance of velocity and synchrotron gradient for a wide range of magnetization in the sub-sonic environment. Addressing
13
+ the studies of magnetic field in atomic hydrogen, we also study the formation of velocity caustics in the spectroscopic channel
14
+ maps in the presence of the thermal broadening. We show that the velocity caustics can be recovered when applied to the Cold
15
+ Neutral Medium (CNM) and the Gradient Technique (GT) can reliably trace magnetic fields there. Finally, we discuss the changes
16
+ of the anisotropy of observed structure functions when we apply to the analysis the procedures developed within the framework
17
+ of GT studies.
18
+ Key words: ISM: structure – ISM: atoms – ISM: clouds – ISM: magnetic fields
19
+ 1 INTRODUCTION
20
+ Magnetic fields are very important for key astrophysical processes in
21
+ interstellar media (ISM) such as the formation of stars (see McKee
22
+ & Ostriker 2007; Mac Low & Klessen 2004), the propagation and
23
+ acceleration of cosmic rays (see Jokipii 1966; Yan & Lazarian 2008),
24
+ the regulation of heat and mass transfer between different ISM phases
25
+ (see Draine 2009 for the list of the different ISM phases). Polarized
26
+ radiation arising from the presence of the magnetic field also inter-
27
+ feres with the sygnal of the enigmatic CMB B-modes arising from
28
+ gravity waves in the early Universe. (Zaldarriaga & Seljak 1997;
29
+ Caldwell et al. 2017; Kandel et al. 2017). Therefore, it is essential to
30
+ have a reliable way to study the properties of magnetic fields in those
31
+ process.
32
+ The traditional way to study the Plane of Sky (POS) magnetic fields
33
+ is using polarimetry measurements (Planck Collaboration
34
+ 2018;
35
+ Lazarian 2002). It is widely used from radio to optical wavelengths
36
+ to trace the magnetic field morphology at various scales in the ISM.
37
+ Recently, a new promising technique has been proposed, the veloc-
38
+ ity gradient technique (VGT), which is capable of tracing magnetic
39
+ field using spectroscopic data (Yuen & Lazarian 2017a; Lazarian et
40
+ al. 2018; Hu et al. 2019; Ho & Lazarian 2021). The technique makes
41
+ use of the fact that magnetic fields make turbulence anisotropic, with
42
+ turbulent eddies being elongated along the magnetic field (See Beres-
43
+ nyak & Lazarian (2019) for a monograph). As a result, the turbulence
44
+ induces the fluid motion mostly perpendicular to the direction sur-
45
+ rounding magnetic eddies. It is important that the magnetic field
46
+ direction is the local direction of magnetic field in the vicinity of
47
+ turbulent eddies. This follows directly from the theory of turbulent
48
+ reconnection that predicts that magnetic fields of the eddies reconnect
49
+ ★ E-mail: [email protected]
50
+ † E-mail: alazarian@facstaff.wisc.edu
51
+ over one eddy turnover time (Lazarian & Vishniac (1999), hereafter
52
+ LV99). This property of magnetic turbulence is central for magnetic
53
+ field tracing with both velocity gradients as well as other types of
54
+ gradients, e.g. synchrotron intensity gradients (Lazarian et al. 2017),
55
+ synchrotron polarization gradients (Lazarian et al. 2018).
56
+ The VGT has been numerically tested for a wide range of column
57
+ densities from diffuse transparent gas to molecular self-absorbing
58
+ dense gas (Yuen & Lazarian 2017a; Lazarian & Yuen 2018a; Hu
59
+ et al. 2019; Hu & Lazarian 2021). The technique was shown to
60
+ be able to provide both the orientations of the magnetic field as
61
+ well as a measure of media magnetization (Lazarian et al. 2018).
62
+ A VGT survey was conducted recently to study the morphology of
63
+ a few nearby molecular cloud (Hu et al. 2019). The result showed
64
+ consistency with the Planck polarization measurement and indicate
65
+ the capability of the VGT on tracing magnetic field in different ISM
66
+ region.
67
+ While the earlier VGT study mainly focused on the supersonic
68
+ spectroscopic data, the same idea of tracing magnetic with gradients
69
+ can be employed with different types of astrophysical data. For in-
70
+ stance, Lazarian et al. (2017) showed gradient can also be applied
71
+ to trace magnetic field with synchrotron intensity gradients (SIGs)
72
+ maps. The corresponding emission comes from subsonic warm/hot
73
+ media. The lack of shock wave in sub sonic environment is beneficial
74
+ for magnetic field tracing.
75
+ Tracing of magnetic field in subsonic media is also important
76
+ within the VGT. The velocity gradients can be obtained in this setting
77
+ using velocity centroids which are not sensitive to thermal broaden-
78
+ ing. If the channel maps are applied to subsonic data, first of all, one
79
+ can use heavier species as spectroscopic tracers. For such species,
80
+ the thermal broadening is suppressed and caustics produced in chan-
81
+ nel maps are prominent. In addition, the newly introduced Velocity
82
+ Decomposition Algorithm (VDA) Yuen et al. (2021) opens ways of
83
+ exploring velocity caustics in the presence of the thermal broadening.
84
+ © 0000 The Authors
85
+ arXiv:2301.13458v1 [astro-ph.GA] 31 Jan 2023
86
+
87
+ 2
88
+ Ho & Lazarian
89
+ Therefore this study explores the ability of magnetic field tracing
90
+ using both the VGT and the SIGs for subsonic medium. Several
91
+ concerns arise on the application of Gradient Technique (GT) in the
92
+ sub-sonic environment. First, multi-phase media study (see Yuen et
93
+ al. (2021)) shows that thermal broadening is a crucial factor that
94
+ smooths out the structure in the subsonic spectroscopic data. It may
95
+ potentially weaken the ability of the VGT to trace the magnetic field.
96
+ Second, Ho & Lazarian (2021) found out that the intermittency of fast
97
+ mode could also play an important role in affecting the VGT analysis.
98
+ In the case that the fast mode dominate the energetics of a particular
99
+ region they induce there the rotation of the velocity gradient direction
100
+ from parallel to perpendicular to the magnetic field. This, however,
101
+ does not happen with the SIGs, for which the gradients induced by
102
+ fast and Alfven modes are parallel.
103
+ In Ho & Lazarian (2021) we proposed a new technique, Gradient
104
+ of Gradient Amplitude (GGA), which improves the magnetic field
105
+ tracing by gradients. However, an in-depth study is required to ana-
106
+ lyze the applicability of GGA in sub-sonic regime versus the change
107
+ of Alfven Mach number.
108
+ Below, we perform a new study of the GT in the sub-sonic environ-
109
+ ment to answer the concerns above and evaluated the performance
110
+ of the GT in a low Ms regime. In what follows, we would cover the
111
+ theory in section 2 and our numerical setup in section 3. Then we
112
+ would discuss the result of the alignment measure of the gradient in
113
+ the ideal observable measure and the velocity gradient in the pres-
114
+ ence of thermal broadening in multi-phase media in section 4. We
115
+ further extend the study of GGA in section 5. We then discuss the the
116
+ Correlation Function Analysis (Hereafter CFA) alignment in section
117
+ 6. At last, we would discuss our work in section 7 and summarize the
118
+ paper in section 8.
119
+ 2 GRADIENT TECHNIQUE
120
+ 2.1 Theoretical Considerations
121
+ The most important component of Magnetohydrodynamic (MHD)
122
+ turbulence is the cascade of Alfvenic motions. Therefore, below we
123
+ will focus on the properties of Alfven modes.
124
+ The modern theory of MHD turbulence originates from the work
125
+ of Goldreich & Sridhar 1995 (henceforth GS95) that described the
126
+ scaling of transAlfvenic incompressible turbulence in what is now
127
+ known to be the strong MHD turbulence regime. The description
128
+ was, however in the frame of the mean magnetic field, which, as it
129
+ was shown by the later studies, the GS95 statitical scalings are not
130
+ applicable.
131
+ Further advances were related to understanding of the importance
132
+ of the local system of reference as well as the generalization of the
133
+ theory for the sub-Alfvenic regime in Lazarian & Vishniac 1999
134
+ (henceforth LV99). There also the regime of weak turbulence was
135
+ quantified (see also Galtier et al. (2000)).
136
+ The local system of reference is the system of reference in respect
137
+ to which the turbulent motions should be considered. Its importance
138
+ is easiest to see considering magnetic eddies. Due to fast turbulent
139
+ reconnection the eddies aligned with the magnetic field direction
140
+ in their vicinity can reconnect and perform a turnover within one
141
+ eddy turnover time (LV99). This happens on the eddy turnover scale
142
+ ∼ 𝑙⊥/𝑣𝑙,
143
+ where 𝑙⊥, 𝑣𝑙 are the size of eddy perpendicular to the
144
+ local magnetic field direction and the eddy’s velocity at the scale l.
145
+ Incidentally, this mixing results in inducing an Alfven perturbation
146
+ with the same period, i.e. 𝑙⊥/𝑣𝑙 ∼ 𝑙∥/𝑉𝐴, where 𝑉𝐴 is the Alfven
147
+ velocity. The latter corresponds to the condition termed "critical
148
+ balance" in GS95. However, unlike the origianal GS95 claim, the
149
+ critical balance is only in the system of reference aligned with the
150
+ local direction of the magnetic field, i.e. with the direction of the
151
+ magnetic field in the direct vicinity of the eddy. The local system of
152
+ reference is absolutely critical for the GT. It is only because of the
153
+ localized alignment that the gradients of velocity and magnetic field
154
+ can trace 3D magnetic field.
155
+ The numerical study in Cho & Vishniac 2000; Maron & Goldreich
156
+ 2000 established numerically the vital importance of the local system
157
+ of reference for the description of MHD turbulence. The subsequent
158
+ studies in Lithwick & Goldreich (2001) as well as Cho & Lazarian
159
+ (2002, 2003); Kowal et al. (2009), extended the the theory to the
160
+ compressible case. This theory of MHD compressible turbulence
161
+ (see the monograph by Beresnyak & Lazarian (2019)) is at the basis
162
+ of the GT.
163
+ It is important to note that the motions perpendicular to the lo-
164
+ cal magnetic field have the form of Alfvenic eddies and they ex-
165
+ hibit Komlogorov scaling 𝑣𝑙 ∼ 𝑙1/3
166
+ ⊥ . Therefore the gradients scale
167
+ as 𝑣𝑙/𝑙⊥ ∼ 𝑙−2/3
168
+
169
+ , meaning that the gradients at the smallest re-
170
+ solved scales are the most important (see Lazarian et al. (2020) for
171
+ the analytical theory of gradient measurements). These gradients are
172
+ perpendicular to the magnetic field and their direction should be
173
+ turned 90 degrees to get the magnetic field tracing. It is important
174
+ that the amplitude of the gradients increases with the decrease of the
175
+ scale. Therefore, the gradients measured at the smallest scales are
176
+ the most prominent. These gradients, similar to aligned grains (see
177
+ (Andersson et al. 2015)), sample the 3D magnetic field along the
178
+ line of sight. Due to this effect, the large scale gradients, e.g. arising
179
+ from galactic shear, are not important for the analysis of the high
180
+ resolution data.
181
+ 2.2 Velocity and magnetic gradients
182
+ 2.2.1 General outlook
183
+ The 3D velocity fluctuation are not directly available from the obser-
184
+ vations. Instead, the gradients of velocity centroids and the gradients
185
+ of intensity fluctuations measured within thin channel maps 1 can be
186
+ used as proxies of the velocity gradients. In both cases, the gradients
187
+ are measured for turbulent volume extended by L > 𝐿𝑖𝑛 𝑗 along the
188
+ LOS, and this entails additional complications, where L, 𝐿𝑖𝑛 𝑗is the
189
+ LOS depth and the injection scale. While eddies stay aligned with
190
+ respect to the local magnetic field, the direction of the local magnetic
191
+ field is expected to change along the LOS. Thus, the contribution of
192
+ 3D velocity gradient are also summed up along the line of sight.
193
+ The spectrum of observed fluctuations changes due to the averag-
194
+ ing effect along the LOS. It is easy to show that the 2D spectrum
195
+ of the turbulence obtained by projecting the fluctuations from 3D
196
+ has the same spectral index of -11/3 2. The relation between the
197
+ spectral slope of the correlation function and the slope of the turbu-
198
+ lence power spectrum in 2D in this situation is −11/3 + 2 = −5/3,
199
+ where 2 is the dimensionality of the space. Therefore, the 2D velocity
200
+ fluctuations arise from the 3D Kolmogorove-type turbulence scale
201
+ as 𝑙5/6
202
+ 2𝐷 with the gradient anisotropy scaling as 𝑙−1/3
203
+ 2𝐷 . It is important
204
+ that the amplitude of the gradients increases with the decrease of the
205
+ 1
206
+ For a channel maps with channel width Δ𝑣, the thin channel map means
207
+ its Δ𝑣 ≤
208
+ √︃
209
+ 𝛿𝑣2
210
+ 𝑅, where 𝛿𝑣𝑅 is the velocity dispersion.
211
+ 2 Starting from 1D spectrum 𝑃1𝐷 with spectral index -5/3, we can get back
212
+ 3D spectral index of −11/3 by considering the dimensional analysis of 𝑃3𝐷
213
+ = 𝑃1𝐷𝑘−2
214
+ MNRAS 000, 000–000 (0000)
215
+
216
+ 3
217
+ scale. Therefore, the gradients measured at the smallest scales are
218
+ the most prominent. These gradients, similar to aligned grains (see
219
+ (Andersson et al. 2015)), sample the 3D magnetic field along the
220
+ line of sight. Due to this effect, the large scale gradients, e.g. arising
221
+ from galactic shear, are not important for the analysis of the high
222
+ resolution data.
223
+ The slow modes follow the scaling of the Alfven modes (Goldreich
224
+ & Sridhar 1995; Lithwick & Goldreich 2001; Cho & Lazarian 2002,
225
+ 2003) and therefore induce the same type of gradients as Alfvenic
226
+ modes. while fast modes are different (Cho & Lazarian 2002, 2003;
227
+ Kowal et al. 2009; Ho & Lazarian 2021)). It follows from the the-
228
+ ory in (Lazarian & Pogosyan (2012), hereafter LP12) that gradients
229
+ of synchrotron emission arising from fast modes are also aligned
230
+ perpendicular magnetic field direction, while the anisotropies of the
231
+ gradients of velocity caustics and velocity centroids are different
232
+ (Kandel et al. 2017, 2018). It is possible to show (Lazarian et al.
233
+ 2018) that the corresponding gradients are perpendicular to those
234
+ created by Alfven and slow modes. Therefore, the contribution of the
235
+ fast modes can decrease the accuracy of the GT. We are dealing with
236
+ their contribution in this paper.
237
+ 2.2.2 VGT for molecular clouds and diffuse HI
238
+ The magnetic field tracing with velocity gradients in molecular
239
+ clouds can be tested successfully with isothermal numerical sim-
240
+ ulations (see Hu et al. (2019)). This is due to efficient cooling of
241
+ the molecular clouds, which is different from HI gas (See Field et
242
+ al. (1969); Wolfire et al. (1995, 2003)). The HI gas is stabilized by
243
+ the thermal equilibrium between the heating and cooling and forms
244
+ two stable phases: the warm and cold phases. Other than the two
245
+ phases, the thermally unstable phase also plays a vital role in the
246
+ atomic hydrogen environment due to the consequence of strong tur-
247
+ bulence. Due to the presence of magnetized turbulence in the atomic
248
+ hydrogen it is a promising medium of applying the VGT. In such an
249
+ environment, the VGT has already demonstrated the reliable tracing
250
+ of the magnetic field (Yuen & Lazarian 2017a; Hu et al. 2019).
251
+ The turbulence is subsonic in most volume of galactic HI, which
252
+ corresponds to the warm phase.(Saury et al. 2014; Marchal, Mar-
253
+ tin & Gong 2021) The Velocity Decomposition Algorithm (VDA)
254
+ developed in Yuen et al. (2021) allows to identify velocity caustics
255
+ produced in this phase.
256
+ 2.3 Velocity caustics
257
+ The concept of velocity caustics is first proposed by Lazarian &
258
+ Pogosyan (2000) and further facilitated by Yuen et al. (2021). Veloc-
259
+ ity caustics describes the effect of pure turbulent velocity fluctuation
260
+ and how they come into the thin channel map. One ideal picture would
261
+ be, even though considering a incompressible magnetized turbulent
262
+ fluid with no density fluctuation, we can still observe a channel map
263
+ with anisotropic fluctuation arising from the turbulence. Those fluc-
264
+ tuations are often referred to as the velocity contribution and different
265
+ statistical tools (for example, VGT) could utilize the information to
266
+ trace magnetic field. However, the fluid contains compressibility and
267
+ density contamination caused by thermal broadening effect, making
268
+ the fluctuation of channel map contains the contribution from both
269
+ density and velocity part. Nonetheless, the density effect on sub-sonic
270
+ media is sub-dominate and can be removed by using the algorithm
271
+ proposed by Yuen et al. (2021).
272
+ 2.3.1 Synchrotron emission
273
+ Measurements of polarized synchrotron radiation and Faraday ro-
274
+ tation (see Beck & Wielebinski (2013); Oppermann et al. (2015);
275
+ Fletcher et al. (2011); Lenc et al. (2016); Van Eck et al. (2017) )
276
+ provide an important insight into the magnetic structure of the Milky
277
+ Way and the neighboring galaxies. Synchrotron radiation fluctuation
278
+ carries the statistical information of MHD turbulence. Serial studies
279
+ discussed how to apply gradient onto measurable quantities, such as
280
+ synchrotron intensity and synchrotron polarization (See Lazarian et
281
+ al. (2017); Lazarian & Yuen (2018a)). In this paper we focus on the
282
+ gradient on synchrotron intensity map as it is a observable that we
283
+ deal with.
284
+ For the power-law distribution of electrons 𝑁(𝐸)𝐸 ∼ 𝐸 𝛼𝑑𝐸, the
285
+ synchrotron emissivity is
286
+ 𝐼𝑠𝑦𝑛𝑐(X) ∝
287
+
288
+ 𝑑𝑧𝐵𝛾
289
+ 𝑃𝑂𝑆(X, z)
290
+ (1)
291
+ where 𝐵𝛾
292
+ 𝑃𝑂𝑆 =
293
+ √︃
294
+ 𝐵2𝑥 + 𝐵2𝑦 corresponds to the magnetic field com-
295
+ ponent perpendicular to the line of sight, X is the plane of sky vector
296
+ defined in x and y direction, z the line of sight axis and, 𝐵𝑥, 𝐵𝑦 the
297
+ 3D magnetic field in x and y direction. The fractional power of the
298
+ index 𝛾 = (𝛼 + 1)/2 was a impediment for quantitative synchrotorn
299
+ statistical studies. However, LP12 showed that the correlation func-
300
+ tions and spectra of the 𝐵𝛾
301
+ ⊥ could express as 𝛼 = 3, which gives 𝛾
302
+ and therefore the dependence of synchrotron intensity on the squared
303
+ magnetic field strength.
304
+ 2.4 Application of Gradient in Sub-Sonic Environment
305
+ Below we will discuss two important examples to which we will apply
306
+ the GT. Those are the centroid map and the synchrotron intensity
307
+ map. We will perform a systematic study of the GT by changing
308
+ the magnetization of the numerical data used to produce synthetic
309
+ observations. Other than that, we would also like to study the behavior
310
+ of GT in the HI spectroscopic velocity channel maps due to the recent
311
+ debate of the velocity caustics effect in the channel map (See section
312
+ 4.3 and 7.2 for more information).
313
+ 3 NUMERICAL SIMULATION AND MEASURES
314
+ EMPLOYED
315
+ 3.1 Simulation Setup
316
+ The numerical data that we analyzed in this work are obtained by 3D
317
+ MHD simulations using the single-fluid, operator-split, staggered-
318
+ grid MHD Eulerian code ZEUS-MP/ 3D (Hayes et al. 2006) to set up
319
+ a 3D, uniform, and isothermal turbulent medium. Periodic boundary
320
+ conditions are applied to emulate a part of the interstellar cloud.
321
+ Solenoidal turbulence injections are employed. To extend our study
322
+ from super sonic regime to sub sonic regime, we simulate two sets
323
+ of ensemble in each regime. Two sets of simulations employ various
324
+ Alfvenic Mach numbers 𝑀𝐴 = 𝑉𝐿/𝑉𝐴 with Sonic Mach Number
325
+ 𝑀𝑆 = 𝑉𝐿/𝑉𝑆 at about 6 and 0.5 where 𝑉𝐿 represents the injection
326
+ velocity, 𝑉𝐴 the Alfven velocities, 𝑉𝑠 the sonic velocity. For the
327
+ generation of turbulence, the turbulence is injected solenoidally for
328
+ all the simulations using the Fourier-space method. Turbulent energy
329
+ is injected at the large scale ( k=2 ) and dissipated by the viscosity
330
+ at small scale. We adjust the strength of the injection such that the
331
+ cubes reach desired 𝑀𝑠 value. All of the cubes are listed in Table
332
+ 1. However, limited by the turbulence scaling (Please see LV99), we
333
+ MNRAS 000, 000–000 (0000)
334
+
335
+ 4
336
+ Ho & Lazarian
337
+ Subsonic
338
+ Supersonic
339
+ Model
340
+ 𝑀𝑆
341
+ 𝑀𝐴
342
+ Model
343
+ 𝑀𝑆
344
+ 𝑀𝐴
345
+ H1S
346
+ 0.67
347
+ 0.13
348
+ H1
349
+ 7.31
350
+ 0.22
351
+ H2S
352
+ 0.64
353
+ 0.38
354
+ H2
355
+ 6.10
356
+ 0.42
357
+ H3S
358
+ 0.62
359
+ 0.64
360
+ H3
361
+ 6.47
362
+ 0.61
363
+ H4S
364
+ 0.61
365
+ 0.90
366
+ H4
367
+ 6.14
368
+ 0.82
369
+ H5S
370
+ 0.61
371
+ 1.17
372
+ H5
373
+ 6.03
374
+ 1.01
375
+ H6
376
+ 6.02
377
+ 1.21
378
+ Table 1. Simulation parameters where 𝑀𝑆, 𝑀𝐴 represents the sonic Mach
379
+ number and Alfvenic Mach number. For all simulations, the resolution is set
380
+ to 7923. 𝑀𝑆, 𝑀𝐴 are the sonic Mach number and the Alfvenic Mach number.
381
+ devote most of our research to the sub-Alfvenic and trans-Alfvenic
382
+ case in this study.
383
+ 3.2 Plane of sky magnetic field
384
+ We trace the plane of sky (POS) magnetic field orientation with
385
+ polarization. We shall assume a constant-emissivity dust grain align-
386
+ ment process. As a comparison to gradient, we generate polarization
387
+ maps by projecting our data cubes along the z-axis. We construct an
388
+ synthetic Stokes parameters Q, U.
389
+ By assuming that the constant emissivity and the dust followed the
390
+ gas, which the dust uniformly aligned with respect to the magnetic
391
+ field, the Stokes parameter 𝑄(X),𝑈(X) can than be expressed as a
392
+ function of angle 𝜃 at plane of sky magnetic field by tan(𝑥, 𝑦) =
393
+ 𝐵𝑦(𝑥, 𝑦)/𝐵𝑥(𝑥, 𝑦) :
394
+ 𝑄(X, 𝑧) ∝
395
+
396
+ 𝑑𝑧 𝜌(X, 𝑧)𝑐𝑜𝑠(2𝜃(X, 𝑧))
397
+ 𝑈(X, 𝑧) ∝
398
+
399
+ 𝑑𝑧 𝜌(X, 𝑧)𝑠𝑖𝑛(2𝜃(X, 𝑧)),
400
+ (2)
401
+ where 𝜌 is the density, X is the plane of sky vector defined in x and y
402
+ direction, z the line of sight axis and, 𝐵𝑥, 𝐵𝑦 the 3D magnetic field
403
+ in x and y direction. The dust polarized intensity 𝐼𝑃 =
404
+ √︁
405
+ 𝑄2 + 𝑈2and
406
+ angle 𝜃 𝑝 = 0.5𝑎𝑡𝑎𝑛2(𝑈/𝑄) are then defined correspondingly.
407
+ 3.3 Synchrotron intensity map
408
+ For our present paper, we follow the approach in LP12 that amplitudes
409
+ of Stokes parameters are scaled up with respect to the cosmic-ray
410
+ index and the spatial variations of the Stokes parameters are similar
411
+ to the case of cosmic-ray index 𝛾 = 2 .
412
+ 3.4 Alignment Measure (AM) and sub-block averaging
413
+ To quantify how good two vector fields are aligned, we employ
414
+ the alignment measure that is introduced in analogy with the grain
415
+ alignment studies (see Lazarian 2002):
416
+ 𝐴𝑀 = 2⟨cos2 𝜃𝑟⟩ − 1,
417
+ (3)
418
+ as discussed for the VGT in González-Casanova & Lazarian 2017;
419
+ Yuen & Lazarian 2017a). The range of AM is [−1, 1] measuring
420
+ the relative alignment between the 90𝑜-rotated gradients and mag-
421
+ netic fields, where 𝜃𝑟 is the relative angle between the two vectors.
422
+ A perfect alignment gives 𝐴𝑀 = 1, whereas random orientations
423
+ generate 𝐴𝑀 = 0 and a perfect perpendicular alignment case refers
424
+ to 𝐴𝑀 = −1 . In what follows we use 𝐴𝑀 to quantify the alignments
425
+ of VGT in respect to magnetic field.
426
+ We adopt the sub-block averaging introduced in Yuen & Lazarian
427
+ (2017a). The use of sub-block averaging comes from the fact that
428
+ the orientation of turbulent eddies with respect to the local magnetic
429
+ field is a statistical concept. In real space the individual gradient
430
+ vectors are not necessarily required to have any relation to the local
431
+ magnetic field direction. Yuen & Lazarian (2017a) reported that the
432
+ velocity gradient orientations in a sub-region–or sub-block–would
433
+ form a Gaussian distribution in which the peak of the Gaussian fit
434
+ reflects the statistical most probable magnetic field orientation in this
435
+ sub–block. As the area of the sampled region increases, the precision
436
+ of the magnetic field traced through the use of Gaussian block fit
437
+ becomes more and more accurate. We will discuss it more in section
438
+ 5.
439
+ 4 RESULTS
440
+ For observational tracing of the magnetic field, it is essential to know
441
+ what to expect in terms of AM dependence on magnetization when
442
+ we employ the gradient method in the ideal synthetic environment.
443
+ We investigate how the change in Alfvenic Mach number 𝑀𝐴 would
444
+ alter the tracing power of Gradient Technique (GT) with two types
445
+ of data: spectroscopic maps and synchrotron intensity map.
446
+ 4.1 Gradients of Synchrotron Intensity
447
+ The synchrotron intensity gradient (SIG) results are presented in the
448
+ left panel of Figure 1. We adopt the sub-block averaging approach,
449
+ and the results are computed using the block size of 722. To compare
450
+ the change of tracing power of GT in different hydro-dynamical
451
+ regimes, we include the result of supersonic simulation(𝑀𝑆 ∼ 6)
452
+ with similar coverage of 𝑀𝐴 as a reference. The setting of block size
453
+ is the same as the sub-sonic regime.
454
+ Throughout the change of 𝑀𝐴, the tracing power of SIG shows a
455
+ different trend in different hydro-dynamical regimes. The result of
456
+ sub-sonic environments (Blue curve) shows that the tracing power of
457
+ SIG is insensitive to the change of magnetization. The AM maintains
458
+ at about 0.8 with a mild drop in 𝑀𝐴 ∼ 0.4 case. For the case
459
+ of supersonic, we observe a steady downtrend of 𝐴𝑀 in the sub-
460
+ Alfvenic regime. The 𝐴𝑀 starts at ∼ 0.58 at 𝑀𝐴 ∼ 0.2 and drops
461
+ gradually to 0.38 at 𝑀𝐴 ∼ 0.8. The declining trend disappears at
462
+ the trans-Alfvenic and super-Alfvenic regime, which the AM steady
463
+ at around 0.38. Besides, we notice that the AM of SIG in sub-sonic
464
+ ensembles always higher than supersonic ensembles.
465
+ 4.2 Result of Gradient in Centroid
466
+ For the benchmark of Velocity centroid gradient (VCG) in the sub-
467
+ sonic environment, Figure. 1 showed the change of AM of centroid
468
+ as a function of 𝑀𝐴 in the right panel. The sub-block setting is the
469
+ same as SIG. As a reference, we also add the change of AM for the
470
+ supersonic environment in orange color. We observe that the AM of
471
+ VGT behaves as a monotonic function of 𝑀𝐴 in the sub-Alfvenic
472
+ regime for both hydro-dynamical regimes. The 𝐴𝑀 declines when
473
+ 𝑀𝐴 increased. The 𝐴𝑀 continues the declining trend throughout
474
+ from sub-Alfvenic to trans-Alfvenic regimes. However, similar to the
475
+ SIG result for supersonic ensembles, the 𝐴𝑀 of VGT for supersonic
476
+ ensembles becomes stable at about 0.4 at the transition from trans-
477
+ Alfvenic to the super-Alfvenic regime,
478
+ A tendency of well alignment between VGT and magnetic field in
479
+ the sub-sonic case is observed here. The AM of sub-sonic set always
480
+ better than supersonic case with the AM improvement of about 0.2
481
+ throughout the change of 𝑀𝐴 from 0.2 to 1.2.
482
+ MNRAS 000, 000–000 (0000)
483
+
484
+ 5
485
+ Figure 1. Left panel: Result of Synchrotron intensity gradient . Right panel: Result of Centroid Gradient. Both block size used = 72. X-axis: Alfvenic Mach
486
+ Number 𝑀𝐴, y-axis: AM. The blue lines represent the AM of sub-sonic ensembles and orange lines represent the change of AM of super-sonic ensembles.
487
+ Figure 2. The Comparison of intensity structure under the influence of thermal broadening. Simulation used in the figure: H4S. Warmer color means denser
488
+ pixels and coolers means pixels with lower density. The blue and red arrow represents the magnetic field direction and gradient direction within the sub-block
489
+ (block size = 662). The bottom right shows the alignment measure value between magnetic field and gradient for each maps.
490
+ 4.3 Velocity Channel gradient in the multi-phase Interstellar
491
+ medium
492
+ The Velocity Channel Gradients provide another way to study the
493
+ magnetic field’s morphology in the interstellar medium Lazarian &
494
+ Yuen (2018a). The intensity fluctuation is strongly affected by its
495
+ width and the thermal properties of the medium. Hu et al. (2019)
496
+ demonstrated the reliable performance of VChGs in tracing the mag-
497
+ netic field directions in super-sonic molecular clouds. However, con-
498
+ cerns of the thermal broadening effect were raised in a sub-sonic
499
+ environment, which the effect could smooth out the velocity caustics
500
+ in the channel maps (Clark et al. 2019). In the extreme case, when
501
+ the thermal width larger than the velocity dispersion width, the fine
502
+ structure of the channel map would be washed out. In addition, this
503
+ can makes it similar to the intensity map. However, the physics of
504
+ the interstellar medium is complicated and involves external physical
505
+ processes, especially for the HI medium. Thermal instability plays a
506
+ crucial role in shaping the proprieties of the HI medium, resulting
507
+ in the multi-phase interstellar medium. In multi-phase media, the
508
+ numerical study found that the warm phase gas occupies most of
509
+ the medium with about 5000K. On the other hand, the cold phase
510
+ medium cools down to about 100K and occupied about 10% space (
511
+ Heiles & Troland 2003; Kritsuk et al. 2017; Ho , Yuen & Lazarian
512
+ 2022).
513
+ Since two-phase media has a dramatic difference in temperature,
514
+ MNRAS 000, 000–000 (0000)
515
+
516
+ Synchrotron Intensity
517
+ Centroid
518
+ 1.0
519
+ Sub-Sonic
520
+ Super-Sonic
521
+ 0.8
522
+ 0.6-
523
+ AM
524
+ 0.4 -
525
+ 0.2
526
+ 0.0 -
527
+ 0.2
528
+ 0.4
529
+ 0.6
530
+ 0.8
531
+ 1.0
532
+ 1.2
533
+ 0.2
534
+ 0.4
535
+ 0.6
536
+ 0.8
537
+ 1.0
538
+ 1.2
539
+ MA
540
+ MABroadening with warm gas only
541
+ No Broadening
542
+ Broadening with cold & warm gas
543
+ M
544
+ = 0.68
545
+ AM = 0.96
546
+ 0.94
547
+ Broadening like
548
+ Velocity castics like6
549
+ Ho & Lazarian
550
+ the influence of broadening effect on the intensity structure in the
551
+ channel map behaves entirely differently. The velocity profile of
552
+ warm phase gas will greatly be extended because of its tempera-
553
+ ture and its fine structure in the channel map being affected. As a
554
+ result, when we look at the transition of fine structure in channel
555
+ map when switching different velocity channel, the caustics created
556
+ in channel maps by turbulence in the warm phase gas will lose their
557
+ contrast due to thermal broadening. A new technique, namely, the
558
+ Velocity Decomposition Technique (VDA) can deal with the effect
559
+ of thermal broadening and focus on the velocity caustics (Yuen et
560
+ al. 2021). In what follows, we another way of how the dynamics of
561
+ warm gas can be revealed in the multi-phase medium.
562
+ If the multi-phase media is a unified turbulent system, dynamics
563
+ between cold gas and warm gas are coupled (Yuen et al. 2022).
564
+ The cold phase gas forms clumps that moving with the surrounding
565
+ warm gas. It suggests the dynamical information of warm phase gas
566
+ will imprint in the cold phase that is not much affected by thermal
567
+ broadening. We expect this effect to be important in multi-phase
568
+ galactic HI.
569
+ To explore and verify this effect, we adopt a post-processing analy-
570
+ sis to make synthetic observation of a multi-phase environment with
571
+ broadening based on our sub-sonic ensembles simulation set. In our
572
+ synthetic observation , we randomly select 15% of pixels and label
573
+ them as a cold phase gas tracer. We label the rest of the pixels as
574
+ warm phase gas. We then transform the Position Position Position
575
+ data cube (PPP) to Position Position Velocity (PPV) cube. We cal-
576
+ culate a PPV cube accounting for a broadening effect. To do so, we
577
+ convolved each pixel with its temperature profile. To simplify our set
578
+ up, we set the temperature of warm gas as 5000K and 100K for cold
579
+ gas. The idea of the post-processing synthetic observation is inspired
580
+ by Yuen et al. (2021). As noticed in Lazarian & Pogosyan (2000), the
581
+ fluctuation of channel maps can be divided into those arising from
582
+ density and velocity. It is demonstrated in Yuen et al. (2021) that,
583
+ without changing the density value, one can vary the sound speed to
584
+ change the fraction of density and velocity contributions in a channel
585
+ map. We should stress that the isothermal simulation could not cap-
586
+ ture the full physics in multiphase ISM. However, Yuen et al. (2021)
587
+ demonstrated that the contribution of CNM and WNM in channel
588
+ map could also be separated into the density and velocity part with
589
+ the difference of different thermal profile. As a result, we can apply
590
+ two thermal profiles to the gas to try to simulate the behaviour of
591
+ CNM and WNM in a channel map.
592
+ Figure 2 demonstrates the center channel Map of synthetic obser-
593
+ vation from one of our simulation cubes(Right). As a reference, the
594
+ figure also includes two comparison plots of the same Channel Map
595
+ but one with a broadening effect with only warm phase (Left) and
596
+ another one without broadening(Mid). This two picture represents
597
+ two different regimes. In the sub-sonic regime, the morphology of
598
+ the channel map without broadening shows a reference of intensity
599
+ fluctuations caused by velocity caustics. Because of the existence of
600
+ the velocity caustics effect, the channel map structure without broad-
601
+ ening effect would demonstrate an intensity structure, which filling
602
+ with thin and long filaments. Those intensity filaments caused by
603
+ caustics within the thin channel map are elongated along the mag-
604
+ netic field, as described in LY18. On the contrary, the morphology of
605
+ the channel map dominated by the broadening effect is different. In
606
+ particular, the intensity fluctuation in the channel map is washed out
607
+ because of the wide thermal velocity profiles. Therefore, the intensity
608
+ structure in the channel map has a high similarity with the intensity
609
+ maps. The similarity of the effects of thermal broadening and the
610
+ increase of the thickness of the channel maps is discussed in LP00.
611
+ The situation is changed if we observe the intensity of emission in
612
+ Figure 3. Result of Channel Gradient considering the effect of thermal broad-
613
+ ening. Block size used = 66. X-axis: Alfvenic Mach Number 𝑀𝐴, y-axis: AM
614
+ thin channel maps arising from the mixture of warm and cold gas.
615
+ There, the thin and long filamentary structures are clearly seen. This
616
+ suggests that the main structure of velocity caustics is preserved in
617
+ the the presence of multi-phase media with cold and warm gas mixed
618
+ together.
619
+ Figure 3 shows a scatter and line plot of AM of VGChT using
620
+ channel map of multi-phase synthetic simulation with respect to 𝑀𝐴
621
+ using the gradient recipe same as the Figure 1. The plot includes the
622
+ 𝐴𝑀 obtained in the channel maps with and without broadening. The
623
+ AM for multi-phase simulation starting with 𝐴𝑀 ∼ 1.0 in 𝑀𝐴 ∼ 0.2
624
+ with slowly decline to 𝐴𝑀 ∼ 0.88 in 𝑀𝐴 ∼ 1.2. In contrast to the
625
+ broadening regime, the AM curve for multi-phase simulation is very
626
+ close to the velocity caustics regime in the sub-Alfvenic simulation
627
+ with a small difference of AM. This discrepancy becomes broader
628
+ as we transfer to the trans-Alfvenic environment.
629
+ 5 IMPROVING AM IN SUB-SONIC MAP USING GGA
630
+ TECHNIQUE
631
+ Ho & Lazarian (2021) identified the effect of intermittency of fast
632
+ mode in low-plasma 𝛽 media. Therefore, the concentration of fast
633
+ modes in selected regions would alter the anisotropy of the distri-
634
+ bution of velocity centroids compared to the neighboring regions
635
+ dominated by Alfvenic modes (see Kandel et al. (2018)).
636
+ This effect would be reflected in the observed centroid gradients
637
+ to abruptly change 90 degrees in the fast mode dominated regions.
638
+ We refer those gradients as orthogonal gradient. Ho & Lazarian
639
+ (2021) introduced new data sets, namely, gradient amplitude map,
640
+ and demonstrated that using these data sets one could suppress the
641
+ orthogonal gradient effect. As a result, the new gradient technique,
642
+ Gradient of Gradient Amplitudes (hereafter GGA), could improve
643
+ the alignment measure. The performance of GGA in ideal case (En-
644
+ MNRAS 000, 000–000 (0000)
645
+
646
+ Channel map with the thermal broadening effect
647
+ 1.00
648
+ 0.95
649
+ 0.90
650
+ 0.85
651
+ AM
652
+ 0.80
653
+ 0.75
654
+ No broadening
655
+ Cold gas included
656
+ 0.70
657
+ Warm gas only
658
+ 0.2
659
+ 0.4
660
+ 0.6
661
+ 0.8
662
+ 1.0
663
+ 1.2
664
+ MA7
665
+ Figure 4. The AM of GGA versus the block size using synchrotron intensity.
666
+ The line with different colors represent the performance of GGA with certain
667
+ strength of white noise added. As a reference, the dotted line with red color
668
+ illustrate the performance of gradient with the noise amplitude of 1𝜎. The
669
+ x-axis showed in log scale for demonstrating the performance of technique in
670
+ small block size.
671
+ Simulation used: H1S
672
+ Block size covered: [11,18,22,33,36,44,66,72,99,132,198,396]
673
+ vironment without noise) could provide prefect alignment (AM∼ 1)
674
+ with the use of block size larger than 502.
675
+ However, we noticed that the performance of GGA could strongly
676
+ depends on the level of noise. The performance of GGA will declin
677
+ rapidly with the increase of noise. To demonstrate the effect of GGA
678
+ in the presence of noise, we add white noise with the amplitude
679
+ relative to the standard deviation of the observable measures and see
680
+ how the 𝐴𝑀 of GGA is varied as a function of noise amplitude.
681
+ Figure 4 shows the 𝐴𝑀 of GGA in centoid maps versus block size
682
+ with white noise added of the amplitude 0.05 𝜎 and 0.1 𝜎. As a
683
+ reference, we also added the AM of GGA without noise. Also, we
684
+ include the AM of gradient with noise of 0.1 𝜎 for a comparison.
685
+ For the computation of GGA, we first define the gradient amplitude
686
+ map (GA), which mechanistically defined as 𝐺𝐴 = �
687
+ 𝑖 𝐴2
688
+ 𝑖 , where 𝐴𝑖
689
+ is gradient component in direction i. For the gradient technique, 𝐴𝑖
690
+ can be computed though the Sobel kernel. The GGA would then be
691
+ the output of the Sobel kernel of GA.
692
+ One can see from the figure, the performance of GGA drops rapidly
693
+ with mild noise added. Compare to ideal case, the AM of GGA falls
694
+ from ∼ 0.9 to ∼ 0.6 in the small block size For noise amplitude of
695
+ 0.05𝜎. The performance gap narrows down with the larger block size
696
+ but block size of ≥ 1202 is required to match the performance of ideal
697
+ case. The advantage of GGA over ordinary gradient decreases for the
698
+ case of noise amplitude 0.1 𝜎. We can see that the performance of
699
+ GGA is very sensitive to the noise level if we use a smaller block
700
+ size.
701
+ To restore the performance of GGA, we employ the Gaussian
702
+ smoothing of 𝜎 = 2 pixel as proposed in Lazarian et al. (2017) and
703
+ tested in Lazarian & Yuen (2018a). According to Lazarian et al.
704
+ Figure 5. The comparison of GGA before and after the smoothing technique
705
+ using the synchrotron intensity map. As a reference, a blue line is added for
706
+ representing the idea case.
707
+ Simulation used: H1S
708
+ Block size covered: [11,18,22,33,36,44,66,72,99,132,198,396]
709
+ (2017), the kernel size we picked here would preserve most of the
710
+ small-scale structures while efficiently suppressing the noise in the
711
+ synthetic map globally. By adding the noise and also the smoothing
712
+ kernel, we can then test whether in noisy observations we can still
713
+ use the GGA as a tool to trace magnetic field. Figure 5 shows the
714
+ result of GGA verus block size with noise added of amplitude 0.1𝜎
715
+ and smoothing. The setup is the same as Figure 4. We can see that the
716
+ application of the smoothing technique shows that the performance of
717
+ GGA can be improved. The drop of AM from 0.5 decrease to 0.8 in the
718
+ small block size while the performance gap between smoothing and
719
+ ideal case become negligible in the block size of 602. The smoothing
720
+ technique could relax the noise level requirement of the GGA.
721
+ 6 CFA IN GRADIENT AMPLITUDE MAP
722
+ Other than gradient, Correlation Function Analysis(CFA) is another
723
+ technique of tracing magnetic field direction by utilizing observable
724
+ measure information (Esquivel & Lazarian 2005; Kandel et al. 2017;
725
+ Hernández-Padilla et al. 2020). CFA was suggested to study magnetic
726
+ field statistically and it is based on the theoretical understanding of
727
+ properties of observed fluctuations (see LP12). For the (2 order)
728
+ correlation function 𝐶𝐹𝐶 of a velocity centroid map 𝐶, it is defined
729
+ as
730
+ 𝐶𝐹𝐶 (R) =< 𝐶(r)𝐶(r + R) >,
731
+ (4)
732
+ where 𝑟, 𝑅 are the vector quantities on 2D maps and separation
733
+ distance from r. The output of 2D correlation map 𝐶𝐹𝐶 (R) can
734
+ be interpreted as the fluctuations between different distance R. If
735
+ the fluctuations are isotropic, the shape of contour line will be cir-
736
+ cular. In opposite, the shape turns to elliptical when the fluctuation
737
+ MNRAS 000, 000–000 (0000)
738
+
739
+ 1.0
740
+ 0.9
741
+ 0.8
742
+ AM
743
+ 0.7
744
+ 0.6-
745
+ GGAwithout noise
746
+ GGA,noise=0.05o
747
+ 0.5
748
+ GGA.noise = O.1 o
749
+ Gradient.noise=O.lo
750
+ 0.4
751
+ 101
752
+ 102
753
+ BlocksizeSynchrotron intensity with noise = o.1
754
+ 1.0
755
+ 0.9
756
+ 0.8 -
757
+ AM
758
+ 0.7
759
+ 0.6
760
+ no noise
761
+ 0.5 -
762
+ noisewithoutsmoothing
763
+ noisewithsmoothing
764
+ 0.4
765
+ 101
766
+ 102
767
+ Blocksize8
768
+ Ho & Lazarian
769
+ is anisotropic. Therefore, the magnetic field direction could be ob-
770
+ tained from the elongated direction of elliptical shape structure after
771
+ the observational map processed by the CFA analysis (Esquivel &
772
+ Lazarian 2005). The elongation depends on the relative importance
773
+ of the three basic MHD modes in turbulence (Kandel et al. 2017). It
774
+ was applied to both observation and simulation data in Yuen et. al
775
+ (2019). However, the study showed that the tracing power of the CFA
776
+ is weaker and the technique is less stable than the gradient technique.
777
+ In this section, we explore the behavior of CFA with the gradient
778
+ amplitude maps.
779
+ A detailed study was conducted to compare the performance be-
780
+ tween gradient and other magnetic field tracing method, including
781
+ CFA (Yuen et. al 2019). One of the issues of CFA showed from Yuen
782
+ et. al (2019) is that the performance of CFA is not stable for the ve-
783
+ locity centroid map. The anisotropy is changed when one selects a
784
+ different block size (For example, figure 15 in Yuen et. al (2019)).
785
+ This change of anisotropy could change 90 degrees by switching the
786
+ block size while the mean field’s direction stays the same throughout
787
+ the region. We repeated this study and extended it to the comparison
788
+ between observable map and gradient amplitude processed map.
789
+ Figure 6 shows how the shape of anisotropy of both maps is
790
+ changed when one selects a different size of a averaging block.
791
+ For sub-Alfvenic simulations like H3S, the mean magnetic field
792
+ strength and direction remain the same throughout the region. For
793
+ CFA, showed from the top side of the figure, we get the same conclu-
794
+ sion as in Yuen et. al (2019). While switching to the small size block
795
+ region, the resolution problem can not only distort the shape of the
796
+ anisotropies in different scales but also destroy the prominent ellipti-
797
+ cal shape. The shape of the elliptical structure is being destroyed for
798
+ the block size is smaller than 120. Also, the direction of anisotropy
799
+ changes when the block size changes.
800
+ However, the situation improves dramatically with the application
801
+ of the GA technique. For the procedure of processing GA-CFA, it is
802
+ same as the computation of the CFA from Yuen et. al (2019) but
803
+ switching the input map to the gradient amplitude map. The bottom
804
+ side of the figure shows the elliptical shape of CFA can be recovered
805
+ after the GA technique. Nonetheless, the anisotropy stays towards
806
+ horizontal direction throughout different block size. From the figure,
807
+ We noticed that there are differences between anisotropy direction
808
+ and magnetic field in block size of 302 but the anisotropy aligns with
809
+ the magnetic field once increase the block size to 602. On the other
810
+ hand, one should mention that the size of the elliptical structure
811
+ is smaller and more elongated compared to the normal CFA. The
812
+ ellipse’s shape exists on a small scale, about 20 to 60 pixels for GA-
813
+ CFA, while it is about 40 to 60 pixels for the CFA. This is due to the
814
+ map process after the gradient amplitude, the morphology of the map
815
+ becomes more filamentary. The size of the filamentary structure is
816
+ more prominent on a small scale in the CFA analysis. So, to improve
817
+ the tracing power of GA-CFA, we have to measure the direction of
818
+ anisotropy on a smaller scale.
819
+ As the performance of CFA improved after combining with the
820
+ GA technique, we then test the improvement of the new GA-CFA
821
+ technique compared to gradient and GGA. We repeat the test showed
822
+ in Figure 1 and extend it to both GGA and GA-CFA technique.
823
+ Inspired by the result from figure 4 and figure 6, we observed a block
824
+ size of 722 would be a common "sweet spot" for both technique
825
+ between the resolution required and the alignment improvement. We
826
+ then pick the sub-block size of 722 for the comparison. The algorithm
827
+ of determining the anisotropy direction of the CFA technique is the
828
+ same as mentioned in the Yuen et. al (2019). For direct comparison
829
+ with Yuen et. al (2019), we also adopt the same pixel distance of
830
+ 10 pixels from the center of the elliptical structure for anisotropy
831
+ contour detection.
832
+ Figure 7 shows the results. One can see a significant advan-
833
+ tage of GGA compared to the other two in the figure in terms of
834
+ the AM. For the performance of GGA in both
835
+ synthetic obser-
836
+ vation maps, the AM decreases according to the Alfvenic Mach
837
+ number. The performance drop is mild for GGA for the amount of
838
+ Δ𝐴𝑀 = 𝐴𝑀𝑀𝐴=0.13 − 𝐴𝑀𝑀𝐴=1.17 ∼ 0.1 when 𝑀𝐴 change from
839
+ sub-Alfvenic to super-Alfvenic. The performance of the GA-CFA
840
+ line between the gradient and GGA but closer to GGA in most of
841
+ the cases but with a small effect of fluctuations. Compared to the
842
+ gradient, GA-CFA has a noticeable better performance, which AM
843
+ improves by about 0.1 for most cases. This shows the performance
844
+ of CFA can be improved by unitizing the Gradient amplitude tech-
845
+ nique. The synergy of the gradients and the GA-CFA approach will
846
+ be explored elsewhere.
847
+ 7 DISCUSSIONS
848
+ 7.1 Connection to earlier gradient studies
849
+ The gradient research opens a new avenue of studying magnetic
850
+ fields and turbulence properties and it is based on of the modern un-
851
+ derstanding of MHD turbulence. Starting from the velocity centroids
852
+ gradient in González-Casanova & Lazarian (2017), studies employed
853
+ later the gradient to different observable maps, such as synchrotron
854
+ intensity/polarization (Lazarian et al. 2017), channel maps (Lazarian
855
+ & Yuen 2018a). This enabled to trace the magnetic field in different
856
+ media from the molecular cloud on the scale of 0.1 pc to the galaxy
857
+ clusters in the scale of 10kpc (see Hu et al. (2020, 2021)). The appli-
858
+ cability of gradient techniques covers two different hydrodynamics
859
+ regimes to both sub-sonic to supersonic regimes. Meanwhile, the
860
+ relationship between gradient and fundamental properties(such as
861
+ 𝑀𝑆, 𝑀𝐴, and MHD modes) of MHD turbulence is being discovered.
862
+ The gradient behavior could change 90 degrees in the particular re-
863
+ gion, for instance, shock or fast mode dominated region. In those
864
+ regions, the direction of rotated gradient vectors would change from
865
+ parallel to perpendicular to the magnetic field, which identifies as an
866
+ orthogonal gradient region. Those orthogonal gradients could lower
867
+ the tracing performance of gradient techniques.
868
+ This paper revisits the performance of gradient techniques in the
869
+ sub-sonic turbulent environment. We study the change of perfor-
870
+ mance of different gradient techniques in sub-sonic medium regard-
871
+ ing Alfvenic number systematically. In addition, we extend the study
872
+ of gradient amplitude. This new technique showed a good perfor-
873
+ mance in removing the distortion in the VGT-produced magnetic
874
+ field maps arising from the effects of the intermittent regions dom-
875
+ inated by fast MHD mode. This GGA technique was demonstrated
876
+ to be capable of removing distortions caused by the fast mode. We
877
+ noticed that GGA amplifies the small scale fluctuation that aligned
878
+ with magnetic field, which suppresses the dominance of fast mode.
879
+ However, its performance is influenced by the noise and the relia-
880
+ bility of GGA could drop rapidly in the presence of the noise. We
881
+ showed in section 5 that the GGA performance in the presence of
882
+ noise could be improved by employing suitable Gaussian filtering.
883
+ This enables the new technique to be applied to realistic observation
884
+ data.
885
+ Furthermore, we explore a new way of combing the gradient am-
886
+ plitude maps with the CFA technique in section 6. The classical CFA
887
+ technique has its limitations while applied to the small block size
888
+ region. This results in the requirement of block size > 1002 and
889
+ MNRAS 000, 000–000 (0000)
890
+
891
+ 9
892
+ [t]
893
+ Figure 6. Variation of correlation function anisotropy shapes with respect to block size. The subplot located at the right panel is the magnified view of the centre
894
+ part of the plot. The blue arrow at the upper left plot shows the direction of the plane of sky magnetic field.
895
+ Top panel: CFA
896
+ Bottom panel: GA-CFA
897
+ [t]
898
+ Figure 7. Left panel: Result of Synchrotron intensity map betwen the gradient, GGA and GA-CFA. Right panel: Result of Centroid map betwen the gradient,
899
+ GGA and GA-CFA. X-axis: Alfvenic Mach Number 𝑀𝐴, y-axis: AM.
900
+ MNRAS 000, 000–000 (0000)
901
+
902
+ Block size = 30
903
+ Block size = 60
904
+ Block size = 120
905
+ Block size = 480
906
+ BposSynchrotron Intensity
907
+ Centroid
908
+ 1.0
909
+ 0.8
910
+ 0.6
911
+ AM
912
+ 0.4
913
+ 0.2 -
914
+ Gradient
915
+ GA-CFA
916
+ GGA
917
+ 0.0
918
+ 0.2
919
+ 0.4
920
+ 0.6
921
+ 0.8
922
+ 1.0
923
+ 1.2
924
+ 0.2
925
+ 0.4
926
+ 0.6
927
+ 0.8
928
+ 1.0
929
+ 1.2
930
+ MA
931
+ MA10
932
+ Ho & Lazarian
933
+ limits the abilities of the CFA in tracing magnetic field. The new
934
+ combined GA-CFA technique minimizes the block size e.g. to 302
935
+ without decreasing the performance of the technique. This extends
936
+ the applicability of the CFA technique and makes it competitive to
937
+ the gradient technique.
938
+ 7.2 Intensity structure and velocity caustics in Channel Map
939
+ The theory of describing the fluctuations of intensity within spectro-
940
+ scopic data that arise from turbulence was formulated in (Lazarian
941
+ & Pogosyan 2000). There the concept of velocity caustics has been
942
+ proposed to describe the effect of turbulent velocities eddies made to
943
+ the channel map. This provided the basis for the technique of tracing
944
+ magnetic fields using velocity channel gradients.
945
+ However, the density also affect fluctuations in channel map fluc-
946
+ tuations. Several HI studies have been discussed on the influence of
947
+ thermal broadening of warm phase made to channel map and the
948
+ importance of cold phase media. The applicability of Lazarian &
949
+ Pogosyan (2000) to galactic HI was questioned in (Clark et al. 2019).
950
+ A rebuttal to these arguments was given by Yuen et. al (2019) and
951
+ the applicability of the LP00-based approach was demonstrated in
952
+ Yuen et al. (2021) where the Velocity Decomposition Algorithm was
953
+ introduced to deal with density fluctuations in subsonic flow. The
954
+ later observational study by Yuen et al. (2022) reported the velocity
955
+ caustics could be fully restored after applying the algorithm.
956
+ Our study in section 4.3 showed provides another argument in
957
+ favor of the applicability of the LP00 theory to multi-phase media.
958
+ We showed that if the phases of the media move together in the
959
+ galactic disk, they can be viewed as a unified turbulent system, and
960
+ our result from figure 3 suggests that most of the information of
961
+ velocity anisotropy can be preserved without the VDA.
962
+ 8 SUMMARY
963
+ This paper extends our studies the Gradient Technique (GT) in the
964
+ sub-sonic environment. Our main results are:
965
+ 1. The alignment between gradient and POS magnetic field is
966
+ better in the subsonic regimes compared to the supersonic one.
967
+ 2. In the multi-phase media, the morphology of filamentary struc-
968
+ ture in the channel map and the statistical anisotropy of thin channel
969
+ intensity fluctuations is preserved in the presence of thermal broad-
970
+ ening if the phases are moving together.
971
+ 3. We extended the study of GGA introduced in the Ho & Lazar-
972
+ ian (2021). We examined the applicability of GGA in the synthetic
973
+ observation map with noise added. The performance of GGA is sen-
974
+ sitive to noise, but the employment of the Gaussian kernel alleviates
975
+ the noise effect.
976
+ 4. We demonstrated that the gradient amplitude maps can be suc-
977
+ cessfully combined with Correlation Function Analysis (CFA). In
978
+ this case the anisotropy can is prominent in small block of the order
979
+ of 302. This makes the new technique competitive with the gradient
980
+ technique.
981
+ MNRAS 000, 000–000 (0000)
982
+
983
+ 11
984
+ ACKNOWLEDGEMENTS
985
+ We acknowledge Ka Ho Yuen and Yue Hu for the fruit-
986
+ ful
987
+ discussions.
988
+ We
989
+ acknowledge
990
+ the
991
+ support
992
+ the
993
+ NASA
994
+ ATP
995
+ AAH7546
996
+ and
997
+ NASA
998
+ TCAN
999
+ 144AAG1967
1000
+ grants.
1001
+ SOFTWARE
1002
+ Julia-v1.2.0/Julia-v1.8.2, Jupyter/miniconda3, LazTech-VGT (Yuen
1003
+ & Lazarian 2017a) : https://github.com/kyuen2/LazTech-VGT
1004
+ DATA AVAILABILITY
1005
+ The data underlying this article will be shared on reasonable request
1006
+ to the corresponding author.
1007
+ REFERENCES
1008
+ Andersson, B., G., Lazarian, A. & Vaillancourt, John E.,
1009
+ 2015, Annual
1010
+ Review of Astronomy and Astrophysics, 53, 501-539
1011
+ Armstrong, J. W., Rickett, B. J., & Spangler, S. R. 1995,The Astrophysical
1012
+ Journal, 443, 209
1013
+ Beck, R., & Wielebinski, R. 2013, Planets, Stars and Stellar Systems. Volume
1014
+ 5: Galactic Structure and Stellar Populations, 5, 641
1015
+ Beresnyak, A., & Lazarian, A. (ed.) 2019, Turbulence in Magnetohydrody-
1016
+ namics (Berlin: De Gruyter)
1017
+ Burkhart, B., Lazarian, A., Ossenkopf, V., & Stutzki, J. 2013, ApJ, 771, 123
1018
+ Caldwell, A., Dvali, G., Majorovits, B., et al. 2017, Physical Review Letters,
1019
+ 118, 091801
1020
+ Chepurnov, A., & Lazarian, A. 2009, ApJ, 693, 1074
1021
+ Chepurnov, A., & Lazarian, A. 2010, The Astrophysical Journal, Volume
1022
+ 710, Issue 1, pp. 853-858 (2010)., 710, 853
1023
+ Cho, J., & Vishniac, E., T. 2000, ApJ, 539, 1, 273
1024
+ Cho, J., & Lazarian, A. 2002, PhRvL, 88, 245001
1025
+ Cho, J., & Lazarian, A. 2003, MNRAS, 345, 325
1026
+ Clark, S., E., Hill, J., C., Peek, J., E., G., Putman, M., E., & Babler, B., L.
1027
+ 2015, Physical Review Letters, 115, 241302
1028
+ Clark, S., E., Peek, J., E., G., & Miville-Deschenes, M.-A. 2019, ApJ, 874, 2
1029
+ Draine, B., T. 2009, Cosmic Dust - Near and Far, 414, 453
1030
+ Esquivel, A, Lazarian, A., 2005, ApJ, 631, 320
1031
+ Federrath, C., Roman-Duval, J., Klessen, R. S., Schmidt, W., & Mac Low,
1032
+ M.-M. 2010, A&A, 512, A81
1033
+ Field, G. B., Goldsmith, D. W., & Habing, H. J., 1969, ApJ, 155, L149
1034
+ Fletcher, A., Beck, R., Shukurov, A., Berkhuijsen, E. M., & Horellou, C.
1035
+ 2011, MNRAS, 412, 2396
1036
+ Goldreich, P., & Sridhar, S. 1995, ApJ, 438, 763
1037
+ González-Casanova, D. F., & Lazarian, A. 2017, ApJ, 835, 41
1038
+ Galtier, S., Nazarenko, S.V., Newel, A.C., & Pouquet, A., J. Plasma Phys.,
1039
+ 63, 447
1040
+ Hayes, J. C. , Norman, M. L., Fiedler, R. A. et al. 2006., ApJS, 165, 188
1041
+ Haud, U. 2000, A&A, 364, 83
1042
+ Hernández-Padilla, D., Esquivel, A., Lazarian, A., et al. 2020, ApJ, 901, 11.
1043
+ doi:10.3847/1538-4357/abad9e
1044
+ Heiles, C., Troland, T. H., 2003, ApJ, 586, 1067
1045
+ Hill, A. S., Benjamin, R. A., Kowal, G., et al. 2008, ApJ, 686, 363
1046
+ Ho, K. W. & Lazarian, A, 2021, ApJ, accepted
1047
+ Ho, K. W., Yuen, K. H., & Lazarian, A,
1048
+ 2022, MNRAS, submitted,
1049
+ arXiv:2111.06845
1050
+ Hu, Y. , Yuen, K. H., Lazarian, V. et al. 2019, Nature Astronomy, 3, 8, 776-782
1051
+ Hu, Y., Lazarian, A. & Li, Y. 2020, ApJ, 901 , 2, 162
1052
+ Hu, Y. & Lazarian, A. 2020, MNRAS, 502 , 2, 1768-1784
1053
+ Hu, Y., Lazarian, A. & Wang, Q. D., 2021, MNRAS, 511 , 1, 829 - 842
1054
+ Jokipii, J., R. 1966, ApJ, 146, 480
1055
+ Kalberla, P. M. W., & Haud, U. 2019, A&A, 627, A112
1056
+ Kandel, D., Lazarian, A., & Pogosyan, D. 2017, MNRAS, 464, 3617
1057
+ Kandel, D., Lazarian, A., & Pogosyan, D. 2018, MNRAS, 478, 530K
1058
+ Kleiner, S., C., & Dickman, R., L. 1985, ApJ, 295, 466
1059
+ Kritsuk, A., G., Ustyugov, S., D., & Norman, M., L. 2017, New Journal of
1060
+ Physics, 19, 065003
1061
+ Kritsuk, A., G., Flauger, R., & Ustyugov, S., D. 2018, Physical Review Letters,
1062
+ 121, 021104
1063
+ Kowal, G., Lazarian, A., & Beresnyak, A. 2007, ApJ, 658, 423
1064
+ Kowal, G., Lazarian, A., & Vishniac, A., L. et. al 2007, ApJ, 700, 63
1065
+ Lazarian, A., & Vishniac, E. T. 1999, ApJ, 517, 700
1066
+ Lazarian, A. 2007, Journal of Quantitative Spectroscopy and Radiative Trans-
1067
+ fer, 106, 225
1068
+ Lazarian, A. 2009, Space Sci. Rev., 143, 357
1069
+ Lazarian, A., & Pogosyan, D. 2000, ApJ, 537, 720
1070
+ Lazarian, A., & Pogosyan, D. 2004, ApJ, 616, 943
1071
+ Lazarian, A., & Pogosyan, D. 2012, ApJ, 747, 5L
1072
+ Lazarian, A., Yuen, K. H., Hyeseung L. , Cho, J., 2017 ApJ, 842, 30
1073
+ Lazarian, A., & Yuen, K. H. 2018, ApJ, 853, 96
1074
+ Lazarian, A., Yuen, K., H., Ho, K., W., et al. 2018, ApJ, 865, 46
1075
+ Lazarian, A., Yuen, K., H.,& Pogosyan, D., 2022, ApJ, 935, 77 arXiv preprint
1076
+ arXiv:2002.07996
1077
+ Lenc, E., Gaensler, B., M., Sun, X., H., et al. 2016, ApJ, 830, 38
1078
+ Li, G.-X., & Burkert, A. 2016, MNRAS, 461, 3027
1079
+ Lithwick, Y. , & Goldreich, P. 2001, ApJ, 562, 279
1080
+ Maron, J. & Goldreich, P. 2000, ApJ, 554,1175
1081
+ Mac Low, M., M., & Klessen, R., S. 2004, Reviews of Modern Physics, 76,
1082
+ 125
1083
+ Marchal, A. Martin, Peter G. , & Gong, M. 2021, ApJ, 921, 11
1084
+ McKee, C., F., & Ostriker, E., C. 2007, ARA&A, 45, 565
1085
+ Miesch, M., S., Scalo, J., & Bally, J. 1999, ApJ, 524, 895
1086
+ Münch, G., & Wheelon, A., D. 1958, Physics of Fluids, 1, 462
1087
+ Oppermann, N., Junklewitz, H., Greiner, M., et al. 2015, A&A, 575, A118
1088
+ O’dell, C., R., & Castaneda, H., O. 1987, ApJ, 317, 686
1089
+ Planck collaboration, 2020, A& A, A3, 50
1090
+ Saury, E., Miville-Deschênes, M., A., Hennebelle et al. 2014, A&A, 567,
1091
+ A16
1092
+ Stanimirović, S., & Lazarian, A. 2001, ApJ, 551, L53
1093
+ Van Eck, C., L., Haverkorn, M., Alves, M., I., R., et al. 2017, A&A, 597, A98
1094
+ Wolfire, M., G., Hollenbach, D., McKee, C., F., Tielens, A., G., G., M., &
1095
+ Bakes, E., L., O. 1995, ApJ, 443, 152
1096
+ Wolfire, M., G., D., McKee, Hollenbach, C., F. & Tielens, A., G., G., M.,
1097
+ 2003 , ApJ, 587, 278
1098
+ Yan, H., & Lazarian, A. 2008, ApJ, 673, 942-953
1099
+ Yuen, K., H., & Lazarian, A. 2017, ApJ, 837, L24
1100
+ Yuen, K., H., Chen, J., Hu, Y. et al 2018, ApJ, 865, 54
1101
+ Yuen, K., H., Hu, Y., Lazarian, A., & Pogosyan, D. 2019, arXiv:1904.03173
1102
+ Yuen, K., H., HO, K., W, & Lazarian, A. 2021, ApJ, 887 (2), 258
1103
+ Yuen, K. H., HO, K. W, & Law, C. Y. et al. 2022, , submitted to nature
1104
+ communications, arXiv:2204.13760
1105
+ Zaldarriaga, M., & Seljak, U. 1997, Phys. Rev. D, 55, 1830
1106
+ MNRAS 000, 000–000 (0000)
1107
+
C9FQT4oBgHgl3EQf_jdA/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
C9FRT4oBgHgl3EQfwjj-/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d2740d7d53ffe8b3d1f01b4995526f0481143296a23b1e856b48db2fa26672aa
3
+ size 5308461
D9AyT4oBgHgl3EQf4vqB/content/2301.00792v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1a19841495a473518ce568e1604b23e761af31f955895a2f438f3c1ef5f0c095
3
+ size 985013
D9AyT4oBgHgl3EQf4vqB/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f735552d9709cd3aee78eb231e4f76d4149614d9c235d3773c94624cbf48bd3a
3
+ size 1572909
D9AyT4oBgHgl3EQf4vqB/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:90ae4609fc3646a47daa7e8ab433755f90d05d173004a7b9b981bd14ef0b51b0
3
+ size 59620
DdE2T4oBgHgl3EQf9gki/content/2301.04228v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f835b33bd1abd1a051836d60b724388570df1acc0af4cd52cefc82886d723b06
3
+ size 1892656
DdE2T4oBgHgl3EQf9gki/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1af7f92cbc6925d57f80a9d787dca4db743c343493908781137fd5bda18a6959
3
+ size 4653101
EtE5T4oBgHgl3EQfUw_x/content/tmp_files/2301.05547v1.pdf.txt ADDED
@@ -0,0 +1,2479 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Springer Nature 2021 LATEX template
2
+ Resilient Model Predictive Control of
3
+ Distributed Systems Under Attack Using
4
+ Local Attack Identification
5
+ Sarah Braun1*, Sebastian Albrecht1 and Sergio Lucia2
6
+ 1*Siemens AG, Otto-Hahn-Ring 6, 81739 M¨unchen, Germany.
7
+ 2TU Dortmund University, August-Schmidt-Straße, 44227
8
+ Dortmund, State.
9
+ *Corresponding author(s). E-mail(s): [email protected];
10
+ Contributing authors: [email protected];
11
12
+ Abstract
13
+ With the growing share of renewable energy sources, the uncertainty
14
+ in power supply is increasing. In addition to the inherent fluctuations
15
+ in the renewables, this is due to the threat of deliberate malicious
16
+ attacks, which may become more prevalent with a growing number
17
+ of distributed generation units. Also in other safety-critical technology
18
+ sectors, control systems are becoming more and more decentralized,
19
+ causing the targets for attackers and thus the risk of attacks to
20
+ increase. It is thus essential that distributed controllers are robust
21
+ toward these uncertainties and able to react quickly to disturbances
22
+ of any kind. To this end, we present novel methods for model-based
23
+ identification of attacks and combine them with distributed model pre-
24
+ dictive control to obtain a resilient framework for adaptively robust
25
+ control. The methodology is specially designed for distributed setups
26
+ with limited local information due to privacy and security reasons. To
27
+ demonstrate the efficiency of the method, we introduce a mathematical
28
+ model for physically coupled microgrids under the uncertain influence
29
+ of renewable generation and adversarial attacks, and perform numeri-
30
+ cal experiments, applying the proposed method for microgrid control.
31
+ Keywords: Attack Identification, Robust Nonlinear Control, Distributed
32
+ Model Predictive Control, Microgrids Under Attack
33
+ 1
34
+ arXiv:2301.05547v1 [cs.SY] 13 Jan 2023
35
+
36
+ Springer Nature 2021 LATEX template
37
+ 2
38
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
39
+ 1 Introduction
40
+ Due to the energy transition, power generation is facing a technological
41
+ change toward increasingly distributed generation, primarily from renewable
42
+ energy sources. Also in other technology areas such as industrial production
43
+ or the transport sector, advancing automation and digitization are creating
44
+ an increasing need for distributed control methods that can be applied to
45
+ safety-critical systems in real time. When designing such methods, it is impor-
46
+ tant to take into account that distributed systems with many components can
47
+ increase flexibility, but at the same time provide many targets for malicious
48
+ attacks. Therefore, distributed control methods should be designed robustly
49
+ and securely, and complemented with appropriate tools to increase the sys-
50
+ tem’s resilience to any type of disruption, which is particularly challenging in
51
+ the event of unpredictable, adversarial attacks.
52
+ Model predictive control (MPC) is one of the most popular control methods
53
+ for dynamic systems in various fields of application as it applies to multivari-
54
+ able systems and allows to include constraints and cost functions in a natural
55
+ way. Based on updated measurements, it repeatedly computes optimal inputs
56
+ to the system at each sampling time. Distributed MPC (DMPC) methods, see
57
+ [1] for an overview and [2] for security-related DMPC, are designed for large
58
+ systems of coupled subsystems and locally apply MPC in each subsystem. In
59
+ contrast to fully decentralized approaches where the neighbors’ dynamic evo-
60
+ lution is unknown to every subsystem, DMPC schemes involve some exchange
61
+ of information among neighbors. In [3], e.g., subsystems provide each other
62
+ with corridors in which future values of their coupling variables lie. Given
63
+ such information about the uncertainty range, robust MPC can be applied
64
+ to explicitly take uncertain influences into account when computing optimal
65
+ inputs. Robust MPC schemes typically build upon tube-based ideas as in [4] or
66
+ multi-stage approaches [5]. It has been demonstrated in several works [6, 7, 8]
67
+ that robust (D)MPC cannot only be applied for robustness against uncertain
68
+ parameters or neighboring couplings, but also against adversarial attacks.
69
+ While robust MPC can reduce the impact of disruptions if the uncertainty
70
+ ranges are known, appropriate security measures for unknown attacks require
71
+ that their presence and points of attack are recognized in the first place. In
72
+ this context, Pasqualetti et al. [9] introduce attack detection and identification
73
+ (ADI) as the tasks of revealing the presence of an attack and localizing all
74
+ attacked system components. For both linear and nonlinear dynamics, there
75
+ are many methods to detect and identify attacks or, closely related, unin-
76
+ tentional system faults. For a broad overview of physics- and control-based
77
+ approaches we refer to the survey in [10]. Some works like [9, 11, 12] design
78
+ unknown-input observers and employ one observer per attack scenario for iden-
79
+ tification, resulting in a combinatorial complexity. Moreover, works on fault
80
+ identification [11] often assume that all possible faults are known, which is an
81
+ invalid assumption for adversarial attacks. In distributed ADI, each subsys-
82
+ tem employs its own estimator to detect and identify local perturbations, be
83
+ it based on observer systems as in [11, 12, 13] or sparse optimization problems
84
+
85
+ Springer Nature 2021 LATEX template
86
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
87
+ 3
88
+ as in [14]. To represent the influence of other subsystems, the local problems
89
+ typically involve measurements of the neighboring couplings transmitted by
90
+ the neighbors [11] or approximated by adaptive local estimators [13].
91
+ In recent years, several approaches that intertwine the handling of attacks
92
+ with (robust) DMPC have been published. In [6], e.g., a DMPC-based strategy
93
+ is presented by which systems reach resilient consensus even if some agents are
94
+ malicious and transmit disturbed state values to their neighbors. An attack
95
+ identification method using Bayesian inference is introduced in [15] and com-
96
+ bined with DMPC to solve robust chance-constrained problems. The approach
97
+ involves testing a series of hypotheses about the attack set and requires full enu-
98
+ meration of all possible attack scenarios. To avoid the resulting combinatorial
99
+ complexity, we combined a DMPC scheme from [3] with our optimization-
100
+ based global ADI method from [16] and proposed an adaptively robust DMPC
101
+ method in [17] for targeted robust control against previously identified attack.
102
+ The contribution of this work, which is an extension of [18], consists in two
103
+ novel approaches for distributed attack identification, a DMPC scheme embed-
104
+ ding these ADI methods for adaptively robust control, and a numerical case
105
+ study to illustrate the proposed resilient control framework using an example
106
+ of interconnected microgrids under attack. The new methods for model-based
107
+ distributed ADI are derived in Section 3 (significantly more detailed compared
108
+ to [18] and including one completely new method). They involve a targeted
109
+ exchange of information between neighbors and solve sparse optimization prob-
110
+ lems to locally identify an attack. The identified insights are used by the DMPC
111
+ framework for adaptively robust control presented in Section 4 (considerably
112
+ exceeding the summarized version in [18]) to initiate suitable preparatory
113
+ measures against previously identified attacks. Unlike the related technique
114
+ introduced in [17], it involves one of the new distributed ADI techniques pre-
115
+ sented in this paper. Finally, we introduce here a more detailed numerical case
116
+ study (in comparison to [18]) with a nonlinear dynamic model for tertiary
117
+ control of interconnected microgrids under attack in Section 5 and perform
118
+ numerical experiments with several attack scenarios in Section 6, illustrating
119
+ the great potential of our resilient control framework for attacked microgrids
120
+ with uncertain renewable generation.
121
+ 2 Problem Formulation
122
+ We consider nonlinear dynamic systems with states x ∈ X ⊆ Rnx, inputs
123
+ u ∈ U ⊆ Rnu, outputs y ∈ Y ⊆ Rny, and uncertain parameters w ∈ W ⊆ Rnw
124
+ that behave according to discrete-time dynamics of the form
125
+ xk+1 = f
126
+
127
+ xk, uk + ak, wk�
128
+ ,
129
+ yk+1 = c
130
+
131
+ xk+1�
132
+ ,
133
+ (1)
134
+ with nonlinear functions f : X×Rnu ×W → X and c : X → Y that are assumed
135
+ to be sufficiently smooth. The system is exposed to the threat of potential
136
+
137
+ Springer Nature 2021 LATEX template
138
+ 4
139
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
140
+ attacks, which are modeled by attack inputs a ∈ A(u) ⊆ Rnu unknown to the
141
+ controller. We consider arbitrary attack vectors a and make no assumptions
142
+ about the set A(u) of possible attacks. While the attack model is additive
143
+ in the input, an attack a affects the states and outputs of the system in a
144
+ nonlinear, nonadditive way.
145
+ The system is partitioned into a set D of subsystems I with local states
146
+ xI ∈ XI ⊆ RnxI , local control inputs uI ∈ UI ⊆ RnuI , local attack inputs
147
+ aI ∈ AI(u) ⊆ RnaI , local outputs yI ∈ YI ⊆ RnyI , and uncertain parameters
148
+ wI ∈ WI ⊆ RnwI . A distributed version of the dynamic system in (1) with
149
+ local dynamic functions fI and local output functions cI is formulated as
150
+ xk+1
151
+ I
152
+ = fI
153
+
154
+ xk
155
+ I, uk
156
+ I + ak
157
+ I, �zk
158
+ NI, wk
159
+ I
160
+
161
+ ,
162
+ zk+1
163
+ I
164
+ = hI
165
+
166
+ xk+1
167
+ I
168
+
169
+ ,
170
+ yk+1
171
+ I
172
+ = cI
173
+
174
+ xk+1
175
+ I
176
+
177
+ ,
178
+ (2)
179
+ where the physical interconnection of subsystems is modeled through coupling
180
+ variables zI ∈ ZI ⊆ RnzI that are related to the local states xI through local
181
+ coupling functions hI : XI → ZI. Since the dynamic evolution of the neigh-
182
+ boring coupling variables zNI(t) during some time interval t ∈ [tk, tk+1] is not
183
+ determined by subsystem I, distributed models typically approximate zNI(t)
184
+ using some information provided by the neighbors. Here, we apply a parame-
185
+ terization scheme proposed in [19] and represent zI(t) on [tk, tk+1] as the linear
186
+ combination
187
+ zI(t) =
188
+ �n
189
+
190
+ j=1
191
+ zk,j
192
+ I βk
193
+ j (t)
194
+ of �n basis functions βk
195
+ 1, . . . , βk
196
+ �n
197
+ :
198
+ [tk, tk+1)
199
+
200
+ R. The coupling coeffi-
201
+ cients zk,j
202
+ I
203
+ are exchanged among neighbors and �zk
204
+ I denotes the coefficient
205
+ matrix �zk
206
+ I := (zk,1
207
+ I
208
+ , . . . , zk,�n
209
+ I
210
+ ) ∈ �ZI ⊆ RnzI ×�n. For a simplified notation, we
211
+ introduce the chained local coupling function ζI := hI◦fI and the chained local
212
+ output function ηI := cI ◦ fI. Similarly, the dense output coupling function
213
+ �ζI : XI × Rnu × �ZNI × WI → �ZI maps to the space �ZI of coupling coefficients.
214
+ Based on the local coupling functions ζI, so-called nominal coupling
215
+ values ¯zk
216
+ I can be determined for the undisturbed case of no attack:
217
+ ¯zk+1
218
+ I
219
+ := ζI
220
+
221
+ xk
222
+ I, uk
223
+ I,�¯z
224
+ k
225
+ NI, 0
226
+
227
+ .
228
+ (3)
229
+ This nominal value is attained if no local attack is applied to the system, i.e.,
230
+ ak
231
+ I = 0, no model uncertainty is present, i.e., wk
232
+ I = 0, and all neighboring
233
+ subsystems also behave according to their nominal values, i.e., �zk
234
+ NI = �¯z
235
+ k
236
+ NI. For
237
+ all methods presented in this paper we assume:
238
+
239
+ Springer Nature 2021 LATEX template
240
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
241
+ 5
242
+ Assumption 1 At each time k, each subsystems I ∈ D transmits the predicted
243
+ nominal values �¯zk
244
+ I , . . . ,�¯zk+Np−1
245
+ I
246
+ of its coupling coefficients with prediction horizon
247
+ Np ∈ N to its neighbors.
248
+ Given this exchange of information among neighbors, the above definition
249
+ in (3) allows for a distributed calculation of the nominal values in a receding
250
+ horizon fashion, where the local values computed and transmitted by subsys-
251
+ tem I at time k are used by its neighbors to update their predictions one time
252
+ step later. The definition further requires suitable initial values �¯z
253
+ 0
254
+ I to be avail-
255
+ able. For simplicity, we assume the system to be in a steady state x0 at time 0
256
+ and take ¯z0,j
257
+ I
258
+ = hI(x0
259
+ I) for all j ∈ {1, . . . , �n}.
260
+ Finally, each subsystem is subject to a set of local constraints
261
+ gI
262
+
263
+ xk
264
+ I, uk
265
+ I + ak
266
+ I, �zk
267
+ NI, wk
268
+ I
269
+
270
+ ≤ 0
271
+ (4)
272
+ for some nonlinear function gI : XI × RnuI × �ZNI × WI → RngI that must be
273
+ satisfied at all times.
274
+ 3 Distributed Attack Identification Based on
275
+ Sparse Optimization
276
+ The goal of this section is to propose a distributed ADI method that, in con-
277
+ trast to global methods, does not involve a central authority which has access
278
+ to a global model of the system. Instead, we formulate a bank of local problems
279
+ that allow each subsystem to identify a suspicion a∗
280
+ I about a potential local
281
+ attack aI based on locally available model knowledge and, possibly, interaction
282
+ with its neighboring subsystems. In contrast to the centralized ADI method
283
+ we presented in [16], no local model knowledge is published globally.
284
+ Before that, we briefly recall the distributed method for the detection of
285
+ attacks that has already been presented in [16]. It is based on each subsystem I
286
+ monitoring the deviations ∆zk+1
287
+ I
288
+ := zk+1
289
+ I
290
+ − ¯zk+1
291
+ I
292
+ in its local coupling variables
293
+ from the respective nominal values ¯zk+1
294
+ I
295
+ . As the nominal values ¯zk+1
296
+ I
297
+ defined
298
+ in (3) are attained in the undisturbed case, a deviation from them indicates
299
+ a disturbance at time k. Using a detection threshold τD ∈ R>0, the method
300
+ detects an attack if ∥∆zk+1
301
+ I
302
+ ∥∞ > τD for any I, i.e., if a distinct deviation is
303
+ observed in any subsystem. To ensure that only significant attacks are revealed
304
+ rather than small model inaccuracies or measurement noise, one can assume
305
+ a probability distribution of the uncertainty and define τD accordingly as in,
306
+ e.g., [11]. Even if subsystem I detects an attack by observing a clear deviation
307
+ ∥∆zk+1
308
+ I
309
+ ∥∞ > τD, it does not necessarily have to be caused by an attack ak
310
+ I ̸= 0
311
+ in I, but can just as well be caused by neighboring subsystems deviating from
312
+ their nominal couplings �¯z
313
+ k
314
+ NI. Identifying the root of the disturbance and thus
315
+ locating the attack is the task of attack identification.
316
+
317
+ Springer Nature 2021 LATEX template
318
+ 6
319
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
320
+ I
321
+ L
322
+ K
323
+ Local ADI
324
+ Local ADI
325
+ Local ADI
326
+ involving
327
+ problem (5)
328
+ �¯z
329
+ k
330
+ I, ∆�zk
331
+ I
332
+ �¯z
333
+ k
334
+ K, ∆�zk
335
+ K
336
+ �¯z
337
+ k
338
+ K, ∆�zk
339
+ K
340
+ �¯z
341
+ k
342
+ L, ∆�zk
343
+ L
344
+ Fig. 1: If neighboring subsystems in a distributed system exchange suitable
345
+ information about their local coupling variables, each subsystem can employ
346
+ a local ADI method to identify suspicions about unknown local attack inputs.
347
+ In this paper, also the identification of attacks is addressed in a distributed
348
+ manner. Depending on the amount and type of information that neighbors
349
+ are willing to share, we derive two different versions of local identification
350
+ problems. Clearly, the more specific the transmitted information describes
351
+ the neighbors’ behavior, the more precisely a local attack or even an attack
352
+ on neighboring subsystems can be identified. Therefore, the design of a local
353
+ identification problem needs to suitably balance the required amount of infor-
354
+ mation and the significance of the obtained suspicions. For the first local
355
+ identification problem that we establish, we propose that in addition to the
356
+ exchange of nominal values �¯z
357
+ k
358
+ I according to Assumption 1, also the deviations
359
+ ∆�zk
360
+ I in the coupling coefficients are repeatedly transmitted to neighboring sub-
361
+ systems. This exchange is performed at each step k when an attack is detected
362
+ and is illustrated in Figure 1. Assuming that each subsystem can locally mea-
363
+ sure the impact onto its output variables yk+1
364
+ I
365
+ ∈ YI ⊆ RnyI , we formulate a
366
+ local attack identification problem to identify local attacks ak
367
+ I as
368
+ min
369
+ aI
370
+ ∥aI∥1
371
+ s.t.
372
+ ���yk+1
373
+ I
374
+ − ηI
375
+
376
+ xk
377
+ I, uk
378
+ I + aI,�¯z
379
+ k
380
+ NI + ∆�zk
381
+ NI, 0
382
+ ����
383
+ 2 ≤ εI.
384
+ (5)
385
+ A solution of problem (5), which has already been proposed in [18], identifies
386
+ a local suspicion a∗
387
+ I for some subsystem I, which is ℓ1-norm sparsest among all
388
+ possible attack vectors in RnuI that explain the observed output yk+1
389
+ I
390
+ accord-
391
+ ing to the local model with output function ηI up to a predefined tolerance
392
+ εI ∈ R≥0, neglecting possible parametric uncertainties wk
393
+ I . While the opti-
394
+ mization variable aI ∈ RnuI represents the unknown attack to be identified,
395
+ the local state xk
396
+ I, input uk
397
+ I, and output yk+1
398
+ I
399
+ are measured or known from
400
+
401
+ Springer Nature 2021 LATEX template
402
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
403
+ 7
404
+ local control computations, and the values �¯z
405
+ k
406
+ NI and ∆�zk
407
+ NI, and thus the actual
408
+ neighboring coupling values �zk
409
+ NI = �¯z
410
+ k
411
+ NI + ∆�zk
412
+ NI, are transmitted by neighbors.
413
+ Computing a sparse suspicion to identify the attack is common in related work
414
+ on attack identification, e.g., [9, 14] and is justified by the observation that
415
+ attackers typically have limited resources and are thus confined to impairing
416
+ only few control components. Some approaches formulate related optimization
417
+ problems using an ℓ0-“norm” cost term ∥aI∥0 to count the number of attacked
418
+ inputs, but solving them requires solution methods from mixed integer pro-
419
+ gramming and is NP-hard [9]. To reduce the computational complexity and
420
+ to obtain a numerically more tractable problem, the ℓ0-“norm” is typically
421
+ relaxed by the ℓ1-norm, see also [16, 20].
422
+ If the neighboring subsystems in NI agree to provide I with even more
423
+ information, subsystem I can apply another version of local identification
424
+ problem, which allows to draw not only conclusions about a potential local
425
+ attack ak
426
+ I, but even about attack inputs ak
427
+ NI in the neighborhood of I. Since
428
+ distributed methods are often applied when sensitive local information must
429
+ not be made publicly available, we assume that neighbors still seek to keep
430
+ their analytical model knowledge private and are only willing to reveal suitable
431
+ numerical derivative information evaluated at the current iterate. We pursued
432
+ a similar approach for the centralized ADI method presented in [16], involv-
433
+ ing the exchange of locally computed sensitivity matrices. To motivate which
434
+ kind of sensitivity information about the dynamic behavior of its neighbors
435
+ subsystem I requires, we approximate the neighboring influence onto the local
436
+ output yI by a first-order Taylor expansion of ηI(xk
437
+ I, uk
438
+ I + ak
439
+ I, �zk
440
+ NI, 0) in the
441
+ �zNI-argument around the nominal value �¯z
442
+ k
443
+ NI. To this end, we define a local
444
+ sensitivity function Sz
445
+ INI : RnuI → RnyI ×nzNI , which maps each given attack
446
+ input aI ∈ RnuI to the Jacobian
447
+ Sz
448
+ INI (aI) := ∂ηI
449
+ ∂�zNI
450
+
451
+ xk
452
+ I, uk
453
+ I + aI,�¯z
454
+ k
455
+ NI, 0
456
+
457
+ ,
458
+ that expresses the first-order dependence of the local output function ηI on
459
+ the neighboring coupling variables �zNI. It can be evaluated locally by I and
460
+ allows to approximate the local output variables yk+1
461
+ I
462
+ according to Taylor’s
463
+ theorem, e.g., [21, §7] as
464
+ yk+1
465
+ I
466
+ = ηI
467
+
468
+ xk
469
+ I, uk
470
+ I + ak
471
+ I,�¯z
472
+ k
473
+ NI, 0
474
+
475
+ + Sz
476
+ INI
477
+
478
+ ak
479
+ I
480
+
481
+ ∆�zk
482
+ NI + Rlin
483
+ I
484
+ + Rw
485
+ I .
486
+ (6)
487
+ Here, the remainder term of the Taylor expansion is denoted by Rlin
488
+ I
489
+ and can be
490
+ estimated similar to the upper bound proven in [16]. The term Rw
491
+ I represents
492
+ a model error which occurs as all uncertain parameters wk
493
+ I are considered zero
494
+ in (6) and due to the fact that the distributed model in (2) only approximates
495
+ the global dynamics in (1).
496
+
497
+ Springer Nature 2021 LATEX template
498
+ 8
499
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
500
+ At this point, the additional sensitivity information provided by the neigh-
501
+ bors NI of I comes into play. Denoting the coupling coefficients of the
502
+ neighbors’ neighbors by �zNNI , we introduce two types of sensitivity matrices as
503
+ �Sa
504
+ NI := ∂�ζNI
505
+ ∂aNI
506
+
507
+ xk
508
+ NI, uk
509
+ NI,�¯z
510
+ k
511
+ NNI , 0
512
+
513
+ and
514
+ �Sz
515
+ NI := ∂�ζNI
516
+ ∂�zNNI
517
+
518
+ xk
519
+ NI, uk
520
+ NI,�¯z
521
+ k
522
+ NNI , 0
523
+
524
+ .
525
+ The function �ζNI denotes the dense coupling function of all neighbors in NI,
526
+ which maps to the space �ZNI of coupling coefficients �zNI and is obtained
527
+ by combining the local dense coupling functions �ζL for all L ∈ NI. Hence,
528
+ the sensitivity matrices �Sa
529
+ NI and �Sz
530
+ NI represent first-order approximations of
531
+ how disturbances in uNI and �zNNI affect the coupling coefficients �zNI. If the
532
+ neighbors in NI provide subsystems I with this information, the deviation
533
+ ∆�zk
534
+ NI of neighboring couplings �zk
535
+ NI from their transmitted nominal values �¯z
536
+ k
537
+ NI
538
+ can be expressed as
539
+ ∆�zk
540
+ NI = �Sa
541
+ NIak
542
+ NI + �Sz
543
+ NI∆�zk
544
+ NNI + Rlin
545
+ NI + Rw
546
+ NI.
547
+ (7)
548
+ The model error Rw
549
+ NI is caused by the uncertain influence of the parameters
550
+ wk
551
+ NI and the linearization error Rlin
552
+ NI denotes the Taylor remainder term when
553
+ expanding the neighbors’ coupling function �ζNI around �¯z
554
+ k
555
+ NNI . The represen-
556
+ tation in (7) gives subsystem I more detailed insights into why its neighbors’
557
+ coupling values �zk
558
+ NI differ from the nominal values �¯z
559
+ k
560
+ NI. More precisely, it
561
+ allows subsystem I to distinguish whether the deviation is caused by an attack
562
+ ak
563
+ NI that the neighbors are exposed to or whether they pass on the disturbing
564
+ effect of any of their neighbors. In order to figure out which source of distur-
565
+ bance applies, subsystem I solves the following local identification problem
566
+ with optimization variables aI, aNI, and ∆�zNNI :
567
+ min
568
+ aI,aNI ,∆�zNNI
569
+ ∥aI∥1 + ∥aNI∥1 +
570
+ ���∆�zNNI
571
+ ���
572
+ 1
573
+ s.t.
574
+ ���yk+1
575
+ I
576
+ − ηI
577
+
578
+ xk
579
+ I, uk
580
+ I + aI,�¯z
581
+ k
582
+ NI, 0
583
+
584
+ + Sz
585
+ INI(aI)
586
+
587
+ �Sa
588
+ NIaNI + �Sz
589
+ NI∆�zNNI
590
+ � ���
591
+ 2 ≤ εI.
592
+ (8)
593
+ An optimal solution (a∗
594
+ I, a∗
595
+ NI, ∆�z∗
596
+ NNI ) of problem (8) is sparsest with respect
597
+ to the ℓ1-norm among all feasible points satisfying the constraints, which are
598
+ obtained by combining (6) and (7) and neglecting all error terms. Similar
599
+ to problem (5), the constraints are relaxed by some tolerance εI ∈ R≥0 to
600
+ account for model inaccuracies. Besides the local quantities uk
601
+ I, yk+1
602
+ I
603
+ , and
604
+ xk
605
+ I, which are known, measured, or estimated by the local control scheme,
606
+ problem (8) also involves the nominal coefficients �¯z
607
+ k
608
+ NI, which are assumed
609
+ to be exchanged among neighboring subsystems according to Assumption 1.
610
+ Instead of the coupling deviations ∆�zk
611
+ NI, the exchange of which is illustrated
612
+ in Figure 1 and taken for granted by the first local identification problem
613
+
614
+ Springer Nature 2021 LATEX template
615
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
616
+ 9
617
+ Algorithm 1 Distributed Attack Detection and Identification Based on Sparse
618
+ Optimization
619
+ Input: local dynamic model for each subsystem I ∈ D as in (2),
620
+ version ∈ {1, 2}
621
+ 1: detected = false, a∗
622
+ I = 0 for all I
623
+ ▷ initialization
624
+ 2: for I ∈ D do
625
+ ▷ distributed attack detection
626
+ 3:
627
+ measure zI, determine ∆zI
628
+ 4:
629
+ if ∥∆zI∥∞ > τD then
630
+ 5:
631
+ detected = true
632
+ 6:
633
+ break
634
+ 7:
635
+ end if
636
+ 8: end for
637
+ 9: if detected then
638
+ ▷ distributed attack identification
639
+ 10:
640
+ for I ∈ D do
641
+ 11:
642
+ if version == 1 then
643
+ 12:
644
+ obtain coupling deviation ∆�zNI from neighbors
645
+ 13:
646
+ solve local identification problem (5) to obtain a∗
647
+ I
648
+ 14:
649
+ else
650
+ 15:
651
+ obtain sensitivity information �Sa
652
+ NI, �Sz
653
+ NI from neighbors
654
+ 16:
655
+ solve local identification problem (8) to obtain a∗
656
+ I
657
+ 17:
658
+ end if
659
+ 18:
660
+ end for
661
+ 19: end if
662
+ 20: return detected, a∗
663
+ I for all I
664
+ (5), the new distributed ADI approach requires all neighbors to provide the
665
+ sensitivity matrices �Sa
666
+ NI and �Sz
667
+ NI. The third sensitivity matrix Sz
668
+ INI (aI) that
669
+ is contained in the constraints of problem (8), in contrast, is computed locally
670
+ by subsystem I in dependence on the optimization variable aI.
671
+ Now that two different formulations of local identification problems have
672
+ been presented, we briefly explain how a complete distributed ADI method is
673
+ obtained from the local optimizations problem (5) or (8), respectively, summa-
674
+ rized as Algorithm 1. The distributed detection scheme is based on monitoring
675
+ the coupling variables and raises an alarm if an abnormal deviation ∆zI > τD
676
+ is observed in any subsystem I. Then, the identification procedure is initiated
677
+ and neighboring subsystems exchange the necessary information to set up the
678
+ identification problem (5) or (8), depending on which version is applied, and
679
+ compute a solution to obtain a suspicion a∗
680
+ I of the local attack. If problem (8)
681
+ is considered, the solution also suggests suspicions a∗
682
+ NI and ∆�z∗
683
+ NNI about the
684
+ disturbing activities in the neighborhood.
685
+ Since the problem formulations in (5) and (8) show some similarities to the
686
+ global identification problem of our publication [16], some of the theoretical
687
+ considerations in [16] can be adopted with only minor changes. E.g., an upper
688
+ bound on the remainder term of the Taylor expansion can be obtained for the
689
+
690
+ Springer Nature 2021 LATEX template
691
+ 10
692
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
693
+ linearization error Rlin
694
+ I
695
+ in (6), when adapting the reasoning of [16] to the fact
696
+ that here the expansion is only applied in the �zNI-argument but not the input.
697
+ The major difference between the identification problems for global versus
698
+ distributed ADI is, however, that the constraints in problem (5) and (8) are
699
+ nonlinear, whereas a linear problem is considered in [16]. As a consequence, the
700
+ theoretical results from [20] on relaxing the ℓ0-“norm” cost term in compressed
701
+ sensing problems by the ℓ1-norm are not applicable here since Candes and Tao
702
+ restrict their considerations to linear constraints. In fact, there is a body of
703
+ research on nonlinear compressed sensing, e.g., [22, 23], the results of which
704
+ can be useful to prove rigorous guarantees for the distributed ADI method
705
+ presented in this section. However, a precise elaboration of such proofs is out
706
+ of scope for this paper and a promising direction for future work.
707
+ 4 Resilient Distributed MPC
708
+ While methods for attack identification are a very powerful tool to localize
709
+ a priori unknown attacks and thus improve the resilience of control systems
710
+ under malicious disturbances, they cannot prevent future attacks or reduce
711
+ their impact. On the other hand, robust control schemes can limit the impact
712
+ of a perturbation by ensuring that no constraints are violated, but require
713
+ information about the value range in which possible disturbances will lie, which
714
+ is typically not available for unknown adversarial attacks. We combine the
715
+ advantages of both approaches by embedding the proposed ADI method into
716
+ a DMPC setup, thus utilizing the identified insights about the attacker toward
717
+ targeted robust DMPC. To this end, we first describe an existing approach for
718
+ robust DMPC in Section 4.1, and enhance it with Algorithm 1 to obtain an
719
+ adaptively robust DMPC scheme in Section 4.2 that computes robust control
720
+ inputs against previously identified attacks in a distributed manner.
721
+ 4.1 Contract-Based Robust Distributed MPC
722
+ By robust control, we refer to computing control inputs that ensure all con-
723
+ straints to a system with uncertain influences being met in all possible cases.
724
+ In [5], Lucia et al. introduce a multi-stage scheme for robust nonlinear MPC
725
+ (NMPC), which considers discrete sets of scenarios and represents the possible
726
+ evolution of the system state in a scenario tree like the one shown in Figure 2.
727
+ In a distributed dynamic system, the neighbors’ couplings zNI behave in an
728
+ uncertain way to the eyes of subsystem I, and, therefore, robust MPC can
729
+ also be used to design distributed MPC methods as long as each subsystem
730
+ is provided with information about the range of possible neighboring coupling
731
+ values. In [3], this idea is implemented by Lucia et al. introducing so-called
732
+ contracts ZI, which are corridors containing predicted reachable values of the
733
+ coupling variables zI and are exchanged among neighbors. At time k, the
734
+ reachable state set X l+1,[k]
735
+ I
736
+ of all values that the local state xl+1
737
+ I
738
+ may attain
739
+
740
+ Springer Nature 2021 LATEX template
741
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
742
+ 11
743
+
744
+ X 1,[0]
745
+ I
746
+
747
+ X 2,[0]
748
+ I
749
+
750
+ X 3,[0]
751
+ I
752
+
753
+ X 4,[0]
754
+ I
755
+ x0
756
+ I
757
+ x1,s7
758
+ I
759
+ x2,s9
760
+ I
761
+ x3,s9
762
+ I
763
+ x4,s9
764
+ I
765
+ x2,s8
766
+ I
767
+ x3,s8
768
+ I
769
+ x4,s8
770
+ I
771
+ x2,s7
772
+ I
773
+ x3,s7
774
+ I
775
+ x4,s7
776
+ I
777
+ x1,s4
778
+ I
779
+ x2,s6
780
+ I
781
+ x3,s6
782
+ I
783
+ x4,s6
784
+ I
785
+ x2,s5
786
+ I
787
+ x3,s5
788
+ I
789
+ x4,s5
790
+ I
791
+ x2,s4
792
+ I
793
+ x3,s4
794
+ I
795
+ x4,s4
796
+ I
797
+ x1,s1
798
+ I
799
+ x2,s3
800
+ I
801
+ x3,s3
802
+ I
803
+ x4,s3
804
+ I
805
+ x2,s2
806
+ I
807
+ x3,s2
808
+ I
809
+ x4,s2
810
+ I
811
+ x2,s1
812
+ I
813
+ x3,s1
814
+ I
815
+ x4,s1
816
+ I
817
+ Fig. 2: A scenario tree as in the multi-stage approach to robust MPC [5], here
818
+ shown for time k = 0 and Np = 4, provides a natural and computationally
819
+ efficient way to approximate the reachable sets X l,[k]
820
+ I
821
+ (indicated in gray) by
822
+ discrete node sets �
823
+ X l,[k]
824
+ I
825
+ (blue) explored by the tree.
826
+ at time l + 1 under all possible uncertainty realizations, is computed as
827
+ X l+1,[k]
828
+ I
829
+ :=
830
+
831
+ fI
832
+
833
+ xl
834
+ I, ul
835
+ I + al
836
+ I, �zl
837
+ NI, wl
838
+ I
839
+
840
+ :
841
+ xl
842
+ I ∈ X l,[k]
843
+ I
844
+ , al
845
+ I ∈ Al,[k−1]
846
+ I
847
+ , �zl
848
+ NI ∈ �
849
+ Zl,[k−1]
850
+ NI
851
+ , wl
852
+ I ∈ Wl,[k−1]
853
+ I
854
+
855
+ with X k,[k]
856
+ I
857
+ := {xk
858
+ I}. From this, the contract Zl,[k]
859
+ I
860
+ for zl
861
+ I at time k is derived as
862
+ Zl,[k]
863
+ I
864
+ :=
865
+
866
+ hI
867
+
868
+ xl
869
+ I
870
+
871
+ : xl
872
+ I ∈ X l,[k]
873
+ I
874
+
875
+ .
876
+ Similarly, contracts �
877
+ Zl,[k]
878
+ I
879
+ for the coupling coefficients �zl
880
+ I are obtained using
881
+ the dense coupling function �ζ. These sets are computed locally at time k,
882
+ provided that each subsystem knows attack and parameter uncertainty sets
883
+ Al,[k−1]
884
+ I
885
+ and Wl,[k−1]
886
+ I
887
+ and additionally receives its neighbors’ contracts �
888
+ Zl,[k−1]
889
+ I
890
+ .
891
+ If all these uncertainty sets are discrete or subsystem I chooses finite subsets
892
+ as sample scenarios, it can locally build a scenario tree as in Figure 2. The
893
+ tree contains one node xl,s
894
+ I
895
+ for each time l ∈ {k, . . . , k + Np} with prediction
896
+ horizon Np and each scenario s ∈ Σ[k−1]
897
+ I
898
+ , where Σ[k−1]
899
+ I
900
+ is the finite local index
901
+ set of scenario indices s. The local scenario trees allow to efficiently compute
902
+ finite approximations �
903
+ X l,[k]
904
+ I
905
+ of the reachable sets X l,[k]
906
+ I
907
+ as the set of tree nodes
908
+ xl,s
909
+ I
910
+ that are reached by subsystem I at stage l in any scenario s ∈ Σ[k−1]
911
+ I
912
+ .
913
+ This is indicated by blue shapes in Figure 2 and explained in detail in [8].
914
+
915
+ Springer Nature 2021 LATEX template
916
+ 12
917
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
918
+ Corresponding approximated contracts �
919
+ Zl,[k]
920
+ I
921
+ are obtained as
922
+
923
+ Zl,[k]
924
+ I
925
+ :=
926
+
927
+ �ζI
928
+
929
+ xl,s
930
+ I , ul,s
931
+ I + al,s
932
+ I , �zl,s
933
+ NI, wl,s
934
+ I
935
+
936
+ : s ∈ Σ[k−1]
937
+ I
938
+
939
+ ⊆ �
940
+ Zl,[k]
941
+ I
942
+ and have been proven to work well in practice [8, 17]. Considering every pos-
943
+ sible evolution of the uncertain system for the future time steps k, . . . , k + Np
944
+ according to the finite scenario set Σ[k−1]
945
+ I
946
+ , contract-based DMPC using multi-
947
+ stage NMPC computes robust control inputs uk
948
+ I, . . . , uk+Np−1
949
+ I
950
+ according to the
951
+ following optimal control problem based on the work of Lucia et al. in [3, 5]
952
+ min
953
+ xl,s
954
+ I ,ul,s
955
+ I
956
+
957
+ s∈Σ[k−1]
958
+ I
959
+ αs
960
+ I
961
+ k+Np−1
962
+
963
+ l=k
964
+ ℓI
965
+
966
+ xl,s
967
+ I , ul,s
968
+ I + al,s
969
+ I , �zl,s
970
+ NI, wl,s
971
+ I
972
+
973
+ s.t.
974
+ xk,s
975
+ I
976
+ = xk
977
+ I,
978
+ xl+1,s
979
+ I
980
+ = fI
981
+
982
+ xl,s
983
+ I , ul,s
984
+ I + al,s
985
+ I , �zl,s
986
+ NI, wl,s
987
+ I
988
+
989
+ ,
990
+ gI
991
+
992
+ xl,s
993
+ I , ul,s
994
+ I + al,s
995
+ I , �zl,s
996
+ NI, wl,s
997
+ I
998
+
999
+ ≤ 0,
1000
+ (9)
1001
+ xl+1,s
1002
+ I
1003
+ ∈ XI, ul,s
1004
+ I
1005
+ ∈ UI,
1006
+ xl,s
1007
+ I
1008
+ = xl,s′
1009
+ I
1010
+ ⇒ ul,s
1011
+ I
1012
+ = ul,s′
1013
+ I
1014
+ ,
1015
+ min
1016
+
1017
+
1018
+ Zl,[k−1]
1019
+ I
1020
+
1021
+ ≤ �ζI
1022
+
1023
+ xl,s
1024
+ I , ul,s
1025
+ I + al,s
1026
+ I , �zl,s
1027
+ NI, wl,s
1028
+ I
1029
+
1030
+ ≤ max
1031
+
1032
+
1033
+ Zl,[k−1]
1034
+ I
1035
+
1036
+ ,
1037
+ for all
1038
+ s ∈ Σ[k−1]
1039
+ I
1040
+ , s′ ∈ Σ[k−1]
1041
+ I
1042
+ , l ∈ {k, . . . , k + Np − 1} .
1043
+ An optimal solution of problem (9) provides a set of state trajectories starting
1044
+ at xk
1045
+ I for all scenarios, behaving according to the local discrete-time dynamics
1046
+ as in (2), and taking only feasible states xl+1,s
1047
+ I
1048
+ ∈ XI. The optimal inputs are
1049
+ chosen to be feasible, to satisfy the constraints in (4) in all scenarios s ∈ Σ[k−1]
1050
+ I
1051
+ and at all times l, and to minimize the local costs ℓI weighted over all scenarios
1052
+ with weights αs
1053
+ I ∈ R≥0. The problem formulation takes into account that
1054
+ future control inputs can be adapted when new measurements are available,
1055
+ while input values ul,s
1056
+ I , ul,s′
1057
+ I
1058
+ that are applied to the same tree node have to
1059
+ coincide because a real-time controller cannot anticipate the future. Finally,
1060
+ for consistency, we require each element �zl,s
1061
+ I
1062
+ of the updated contract �
1063
+ Zl,[k]
1064
+ I
1065
+ to be within the bounds of the previous contract �
1066
+ Zl,[k−1]
1067
+ I
1068
+ . For details on the
1069
+ purpose and the theoretical consequences of the last two groups of constraints
1070
+ we refer to the original works [3, 5] and our own work [8].
1071
+ 4.2 Adaptively Robust Distributed MPC
1072
+ While we have explained in Section 4.1 how updated contracts �
1073
+ Zl,[k]
1074
+ I
1075
+ are
1076
+ calculated at each time k from a solution of problem (9), we have not yet com-
1077
+ mented on how to obtain similar scenario sets �
1078
+ Al,[k]
1079
+ I
1080
+ and �
1081
+ Wl,[k]
1082
+ I
1083
+ for unknown
1084
+
1085
+ Springer Nature 2021 LATEX template
1086
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
1087
+ 13
1088
+ attacks al
1089
+ I and uncertain parameters wl
1090
+ I. For the latter, suitable samples are
1091
+ usually provided by forecasts, historical data, or technical properties of the
1092
+ system components. For unknown attacks, however, it would be very restric-
1093
+ tive to assume that appropriate scenario sets �
1094
+ Al,[k]
1095
+ I
1096
+ are provided. Choosing
1097
+ few random attacks as samples as in [8] cannot be expected to achieve sat-
1098
+ isfied constraints in all cases, while choosing a very large number of samples
1099
+ may cover the set AI of possible attacks sufficiently well, but leads to com-
1100
+ putationally intractable problems since the size of the scenario tree grows
1101
+ exponentially in the number of scenarios. To address this issue, we proposed a
1102
+ more general, adaptively robust MPC approach in [17] that utilizes available
1103
+ knowledge about the attackers gained from attack identification to design the
1104
+ sets �
1105
+ Al,[k]
1106
+ I
1107
+ and is repeated in this section. Unlike in [17], here the distributed
1108
+ ADI approaches from Section 3 are embedded in a DMPC setup, resulting in a
1109
+ fully distributed control framework that does not require any central instance.
1110
+ The approach has already been described in [18] and is presented here in
1111
+ further depth.
1112
+ The method is designed for local attacks aI that follow a probability distri-
1113
+ bution with unknown, time-invariant expected value µI ∈ RnuI and standard
1114
+ deviation σI ∈ R
1115
+ nuI
1116
+ ≥0 . The basic idea is to repeatedly estimate these parame-
1117
+ ters at each time k based on the solutions a∗,l
1118
+ I
1119
+ of the local attack identification
1120
+ problem at previous times l ≤ k, and to adapt the uncertainty sets �
1121
+ Al,[k] for
1122
+ possible attacks al accordingly. More precisely, at time k the mean µ[k]
1123
+ I
1124
+ and
1125
+ sample standard deviation σ[k]
1126
+ I
1127
+ of all previously identified values a∗,l
1128
+ I
1129
+ given as
1130
+ µ[k]
1131
+ I
1132
+ :=
1133
+ 1
1134
+ k + 1
1135
+ k
1136
+
1137
+ l=0
1138
+ a∗,l
1139
+ I
1140
+ and
1141
+ σ[k]
1142
+ I
1143
+ :=
1144
+
1145
+ 1
1146
+ k
1147
+ k
1148
+
1149
+ l=0
1150
+
1151
+ a∗,l
1152
+ I
1153
+ − µ[k]
1154
+ I
1155
+ �2
1156
+ � 1
1157
+ 2
1158
+ (10)
1159
+ serve as estimates for µI and σI. According to the local identification results
1160
+ until time k, the uncertainty of possible attacks al
1161
+ I for future time steps l is
1162
+ represented by three scenarios for each component (ak
1163
+ I)i for i ∈ {1, . . . , nuI}
1164
+
1165
+ Al,[k]
1166
+ I
1167
+ =
1168
+
1169
+ i∈I
1170
+
1171
+ µ[k]
1172
+ i , µ[k]
1173
+ i
1174
+ + σ[k]
1175
+ i , µ[k]
1176
+ i
1177
+ − σ[k]
1178
+ i
1179
+
1180
+ .
1181
+ (11)
1182
+ The combination of contract-based robust DMPC from Section 4.1 and the dis-
1183
+ tributed ADI method from Section 3 results in an adaptively robust distributed
1184
+ MPC method that is summarized in Algorithm 2.
1185
+ We formulate Algorithm 2 involving the local identification problem (5)
1186
+ and thus the first version of Algorithm 1 since this is what we apply in the
1187
+ numerical experiments presented in Section 6. Clearly, Algorithm 2 can also
1188
+ be defined based on the second version of Algorithm 1 solving problem (8).
1189
+ In this case, subsystem I can additionally modify the transmitted contracts
1190
+
1191
+ ZNI in such a way that the locally identified suspicions a∗
1192
+ NI, ∆�z∗
1193
+ NI about
1194
+ neighboring attacks and coupling deviations are taken into account. While this
1195
+
1196
+ Springer Nature 2021 LATEX template
1197
+ 14
1198
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
1199
+ Algorithm 2 Adaptively robust distributed MPC
1200
+ Input: local dynamic model for each subsystem I ∈ D,
1201
+ initial contracts �
1202
+ Zl,[0]
1203
+ I
1204
+ for all I, l, e.g., �
1205
+ Zl,[0]
1206
+ I
1207
+ = {hI(x0
1208
+ I)},
1209
+ finite parameter scenario sets �
1210
+ Wl,[k]
1211
+ I
1212
+ for all l, k
1213
+ 1: set �
1214
+ Al,[0]
1215
+ I
1216
+ := {} for all I, l
1217
+ 2: for time step k do
1218
+ 3:
1219
+ for I ∈ D do
1220
+ 4:
1221
+ build scenario tree by branching on �
1222
+ Al,[k−1]
1223
+ I
1224
+ , �
1225
+ Zl,[k−1]
1226
+ NI
1227
+ , and �
1228
+ Wl,[k−1]
1229
+ I
1230
+ 5:
1231
+ solve problem (9) to compute inputs ul
1232
+ I
1233
+ 6:
1234
+ derive new contracts �
1235
+ Zl,[k]
1236
+ I
1237
+ ▷ update contracts
1238
+ 7:
1239
+ transmit �
1240
+ Zl,[k]
1241
+ I
1242
+ to neighbors
1243
+ 8:
1244
+ end for
1245
+ 9:
1246
+ apply first control input uk = (uk
1247
+ I)I∈D
1248
+ 10:
1249
+ for I ∈ D do
1250
+ 11:
1251
+ solve problem (5) to obtain a suspicion a∗,k
1252
+ I
1253
+ ▷ local ADI
1254
+ 12:
1255
+ update estimates µ[k]
1256
+ I , σ[k]
1257
+ I
1258
+ as in (10)
1259
+ 13:
1260
+ adapt uncertainty set �
1261
+ Al,[k]
1262
+ I
1263
+ as in (11)
1264
+ ▷ update attack scenarios
1265
+ 14:
1266
+ end for
1267
+ 15: end for
1268
+ is not reasonable if the neighbors and thus their transmitted sensitivities �Sa
1269
+ NI
1270
+ and �Sz
1271
+ NI are generally deemed untrustworthy, it is useful if the communication
1272
+ channel to the neighbors is considered secure, but the neighbors themselves do
1273
+ not apply ADI and therefore do not adapt their contracts to attacks.
1274
+ By enhancing distributed MPC with local attack identification in each
1275
+ subsystem, we obtain a distributed adaptively robust control framework, in
1276
+ which only locally available model knowledge and some information exchange
1277
+ among neighbors is involved. Unlike the related method introduced in [17],
1278
+ Algorithm 2 requires no central authority and, in particular, no confidential
1279
+ model knowledge is published globally. Such a procedure has the advantages
1280
+ that all local identification problems can be solved in parallel, that it can be
1281
+ employed even if the subsystems fail to agree on a central authority, and that
1282
+ no private model knowledge has to be shared with the entire network. Further-
1283
+ more, all distributed ADI approaches have in common that it is challenging to
1284
+ agree on system-wide countermeasures based on multiple, possibly contradic-
1285
+ tory local identification results. Our approach provides an answer to this issue
1286
+ as it transfers the insights from distributed ADI into local countermeasures by
1287
+ adjusting the local control inputs in a suitable robust way.
1288
+ 5 Dynamic Model for Microgrids Under Attack
1289
+ Distributed microgrids that include local generation, demands, and often stor-
1290
+ age units, increase the security of supply within the microgrid area but create
1291
+
1292
+ Springer Nature 2021 LATEX template
1293
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
1294
+ 15
1295
+ new challenges: Several optimal control tasks have to be addressed under the
1296
+ uncertainty of renewables and possibly even adversarial attacks, e.g., economic
1297
+ generator dispatch, efficient battery use, or optimal power import and export
1298
+ strategies to benefit from fluctuating energy prices [24, 25]. Therefore, we aim
1299
+ to apply the resilient control framework proposed in Section 4 to the task of
1300
+ microgrid control and derive a suitable dynamic model in this section.
1301
+ The main characteristics of the model are nonlinear battery dynam-
1302
+ ics, physical coupling of neighboring microgrids through dispatchable power
1303
+ exchange, and the threat of possible attacks. Each microgrid contains an aggre-
1304
+ gated load pl
1305
+ I ≤ 0 and a set of dispatchable generation units that generate a
1306
+ total power output pg
1307
+ I ≥ 0. How uncertain load and nondispatchable generation
1308
+ from renewable energy sources are modeled is discussed below. As illustrated
1309
+ in Figure 3, each microgrid is connected to the main grid, to or from which it
1310
+ can export or import power pm
1311
+ I ∈ R. While power import is modeled by posi-
1312
+ tive values pm
1313
+ I > 0, negative values pm
1314
+ I < 0 indicate power export to the main
1315
+ grid. In addition, power transfers are possible between two neighboring micro-
1316
+ grids I, L with L ∈ NI. The power that microgrid I provides to L is denoted
1317
+ as ptr
1318
+ IL and the resulting directed power flow from I to L is given as
1319
+ pflow
1320
+ IL := ptr
1321
+ IL − ptr
1322
+ LI.
1323
+ Finally, each microgrid has a storage unit that provides or consumes storage
1324
+ power pst
1325
+ I ∈ R and the state variable sI ∈ [0.0, 1.0] indicates its state of charge
1326
+ (SoC). Power values pst
1327
+ I > 0 indicate discharging and pst
1328
+ I < 0 charging. Unlike
1329
+ other works investigating economic dispatch problems in microgrid settings,
1330
+ for example Ananduta et al. in [15], we take into account that power cannot
1331
+ change instantaneously. Instead, the dynamic evolution of pg
1332
+ I, pm
1333
+ I , and ptr
1334
+ IL is
1335
+ controlled by inputs ug
1336
+ I, um
1337
+ I , and utr
1338
+ IL and behaves according to
1339
+ ˙pg
1340
+ I = 1
1341
+ T g
1342
+ I
1343
+ (ug
1344
+ I + ag
1345
+ I − pg
1346
+ I) ,
1347
+ (12)
1348
+ ˙pm
1349
+ I =
1350
+ 1
1351
+ T m
1352
+ I
1353
+ (um
1354
+ I + am
1355
+ I − pm
1356
+ I ) ,
1357
+ (13)
1358
+ ˙ptr
1359
+ IL =
1360
+ 1
1361
+ T tr
1362
+ IL
1363
+
1364
+ utr
1365
+ IL + atr
1366
+ IL − ptr
1367
+ IL
1368
+
1369
+ .
1370
+ (14)
1371
+ The various delay parameters T g
1372
+ I , T m
1373
+ I , T tr
1374
+ IL ∈ R>0 depending on technical char-
1375
+ acteristics capture how quickly a change in the respective input affects the
1376
+ corresponding state. Compared to the generation delay T g
1377
+ I , typically smaller
1378
+ delay times T m
1379
+ I and T tr
1380
+ IL apply for power transfers with the main grid or neigh-
1381
+ boring microgrids. In line with the generic description of distributed systems
1382
+ under attack introduced in Section 2, we model attacks as additional, unknown
1383
+ inputs that impair the dynamic behavior of the microgrid systems as in (12)
1384
+ to (14). In each microgrid I ∈ D, we consider generator attacks ag
1385
+ I ∈ R, grid
1386
+ attacks am
1387
+ I ∈ R affecting the power exchange with the main grid, and transfer
1388
+
1389
+ Springer Nature 2021 LATEX template
1390
+ 16
1391
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
1392
+ pst
1393
+ I = -ΣpI
1394
+ pg
1395
+ I
1396
+ pl
1397
+ I
1398
+ pm
1399
+ I
1400
+ ptr
1401
+ IK
1402
+ ptr
1403
+ IL
1404
+ I
1405
+ L
1406
+ K
1407
+ zLI
1408
+ zKI
1409
+ Main grid
1410
+ Fig. 3: Schematic overview of the model for interconnected microgrids taken
1411
+ from [18, Fig. 1], showing the local model components for microgrid I. Apart
1412
+ from internal states, each microgrid only requires knowledge of its neighboring
1413
+ couplings (zLI)J∈NI. For power balance, storage units are used as a buffer.
1414
+ attacks atr
1415
+ IL ∈ R on power transfers to or from any neighbor L ∈ NI. While the
1416
+ inputs are computed by the local controller in I, the attack values are unknown
1417
+ to the control system. Thus, we deliberately make no difference in modeling
1418
+ attacks and renewable generation but consider both as uncertain influences
1419
+ resolved by the resilient control framework presented in Section 4.2. Similarly,
1420
+ uncertain load can be considered an attack al
1421
+ I modifying the load pl
1422
+ I = ul
1423
+ I that
1424
+ is modeled as a noncontrollable input with equal upper and lower bounds.
1425
+ The storage is used as a buffer providing the required power reserves at
1426
+ all times and thus assuring that the power balance in microgrid I is always
1427
+ satisfied, even when an attack occurs. Therefore, the storage power pst
1428
+ I is a
1429
+ dependent variable according to
1430
+ pst
1431
+ I = −pg
1432
+ I − pm
1433
+ I − pl
1434
+ I −
1435
+
1436
+ L∈NI
1437
+
1438
+ ptr
1439
+ LI − ptr
1440
+ IL
1441
+
1442
+ .
1443
+ It is important to distinguish that for microgrid I, the local state ptr
1444
+ IL can
1445
+ be controlled via utr
1446
+ IL as in (14), whereas the neighboring state ptr
1447
+ LI is neither
1448
+ controllable nor is its dynamic behavior known by microgrid I. The physical
1449
+ interconnection of neighboring microgrids is instead modeled by a coupling
1450
+ variable zLI = ptr
1451
+ LI and is treated locally as an uncertain parameter as we
1452
+ discussed in detail in Section 4.1. Figure 3 illustrates that the local knowledge
1453
+ is limited to local power variables and neighboring couplings.
1454
+ According to the storage power pst
1455
+ I , the storage is charged or discharged
1456
+ and the resulting change in the SoC sI is modeled as
1457
+ ˙sI = bI
1458
+
1459
+ sI, pst
1460
+ I
1461
+
1462
+ with some function bI : [0.0, 1.0] × R → R modeling the battery dynamics.
1463
+ While a linear approximation of this charging behavior is usually sufficient in
1464
+ the middle range of [0.0, 1.0], it is not accurate for marginal values of the SoC
1465
+
1466
+ Springer Nature 2021 LATEX template
1467
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
1468
+ 17
1469
+ which become extremely relevant in case of an attack. Following the line of
1470
+ [26, 27], the dynamics of the SoC are given as
1471
+ ˙sI = − Ist
1472
+ I
1473
+ Qst
1474
+ I
1475
+ ,
1476
+ (15)
1477
+ with Qst
1478
+ I denoting the maximum capacity of the battery and Ist
1479
+ I
1480
+ being the
1481
+ battery current. Denoting the battery voltage by U st
1482
+ I , the storage power pst
1483
+ I
1484
+ and the voltage U st
1485
+ I are given as
1486
+ pst
1487
+ I = U st
1488
+ I Ist
1489
+ I
1490
+ and U st
1491
+ I = U OCV
1492
+ I
1493
+ (sI) + Rst
1494
+ I Ist
1495
+ I .
1496
+ (16)
1497
+ in line with [26]. The term U OCV
1498
+ I
1499
+ denotes the open circuit voltage (OCV), that
1500
+ depends on the SoC sI, and the second summand determining U st
1501
+ I
1502
+ models
1503
+ the ohmic effect with resistance Rst
1504
+ I . Rewriting (16) results in the following
1505
+ relation for the storage power pst
1506
+ I :
1507
+ pst
1508
+ I = U OCV
1509
+ I
1510
+ (sI)Ist
1511
+ I + Rst
1512
+ I
1513
+
1514
+ Ist
1515
+ I
1516
+ �2 .
1517
+ Solving this equation for Ist
1518
+ I , the battery current Ist
1519
+ I = nI (sI, pst
1520
+ I ) is obtained
1521
+ from sI and pst
1522
+ I for some nonlinear function nI : [0.0, 1.0] × R → R. Together
1523
+ with (15), this results in a nonlinear function
1524
+ bI(sI, pst
1525
+ I ) := −nI(sI, pst
1526
+ I )
1527
+ Qst
1528
+ I
1529
+ that describes the dynamic behavior of the battery.
1530
+ It remains open to specify the open circuit voltage U OCV
1531
+ I
1532
+ (sI) using the
1533
+ model in [27], that is accurate also for low and high SOCs: With parameters
1534
+ αI, βI, γI, δI, µI, and νI depending on the type of battery, the OCV is given by
1535
+ U OCV
1536
+ I
1537
+ (sI) := αI + βI(−ln(sI))µI + γIsI + δIeνI(sI−1).
1538
+ (17)
1539
+ Bringing all of the above together, we have characterized a distributed
1540
+ dynamic system of interconnected microgrids, which results in a model of the
1541
+ form as in (2) when discretizing. Each microgrid is described by a local state
1542
+ xI =
1543
+
1544
+ sI, pg
1545
+ I, pm
1546
+ I , ptr
1547
+ I
1548
+ �⊤ ∈ R3+|NI|
1549
+ (18)
1550
+ with ptr
1551
+ I := (ptr
1552
+ IL)L∈NI and controlled by a local input
1553
+ uI =
1554
+
1555
+ ug
1556
+ I, um
1557
+ I , utr
1558
+ I
1559
+ �⊤ ∈ R2+|NI|,
1560
+ (19)
1561
+ that may be disturbed by an attack input
1562
+ aI =
1563
+
1564
+ ag
1565
+ I, am
1566
+ I , atr
1567
+ I
1568
+ �⊤ ∈ R2+|NI|
1569
+ (20)
1570
+
1571
+ Springer Nature 2021 LATEX template
1572
+ 18
1573
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
1574
+ with utr
1575
+ I := (utr
1576
+ IL)L∈NI and atr
1577
+ I := (atr
1578
+ IL)L∈NI. Power transfers to other micro-
1579
+ grids physically couple neighboring microgrids to each other, which is modeled
1580
+ by local coupling variables
1581
+ zI =
1582
+
1583
+ ptr
1584
+ IL
1585
+ �⊤
1586
+ L∈NI
1587
+ with zNI =
1588
+
1589
+ ptr
1590
+ LI
1591
+ �⊤
1592
+ L∈NI .
1593
+ (21)
1594
+ Each microgrid I ∈ D is operated locally to meet the respective load pl
1595
+ I at
1596
+ the lowest possible cost according to some objective function JI : R>0 → R,
1597
+ which specifies the costs incurred during some time window [0, T] of length
1598
+ T ∈ R>0 and is defined as
1599
+ JI(T) :=
1600
+ � T
1601
+ 0
1602
+ qI
1603
+
1604
+ pg
1605
+ I, ptr
1606
+ I , pst
1607
+ I
1608
+
1609
+ + ℓI
1610
+
1611
+ pflow
1612
+ I
1613
+ , pm
1614
+ I
1615
+
1616
+ dt + mI (sI (T)) .
1617
+ (22)
1618
+ It consists of quadratic stage costs qI, piecewise linear stage costs ℓI, and
1619
+ terminal costs mI. The quadratic costs qI : R≥0 × RnzI × R → R≥0 with cost
1620
+ parameters Cg
1621
+ I , Ctr
1622
+ I , Cst
1623
+ I ∈ R≥0 are given as
1624
+ qI
1625
+
1626
+ pg
1627
+ I, ptr
1628
+ I , pst
1629
+ I
1630
+ � := Cg
1631
+ I (pg
1632
+ I)2 +
1633
+
1634
+ L∈NI
1635
+ Ctr
1636
+ I
1637
+
1638
+ ptr
1639
+ IL
1640
+ �2 + Cst
1641
+ I
1642
+
1643
+ pst
1644
+ I
1645
+ �2 .
1646
+ They capture the per-unit costs of using the units for power generation, power
1647
+ transfers to neighbors, and the respective storage operations. In contrast, the
1648
+ piecewise linear costs ℓI model the economic profit or loss from selling or
1649
+ buying energy in trade with neighbors or the main grid. Defining the positive
1650
+ and negative part functions
1651
+ (v)+ :=
1652
+
1653
+ 0 if v < 0,
1654
+ v if v ≥ 0,
1655
+ and (v)− :=
1656
+
1657
+ v if v < 0,
1658
+ 0 if v ≥ 0,
1659
+ the piecewise linear cost function ℓI : RnzI × R → R is given as
1660
+ ℓI
1661
+
1662
+ pflow
1663
+ I
1664
+ , pm
1665
+ I
1666
+ � :=
1667
+
1668
+ L∈NI
1669
+ Cflow,ex
1670
+ LI
1671
+
1672
+ pflow
1673
+ LI
1674
+
1675
+ − +
1676
+
1677
+ L∈NI
1678
+ Cflow,im
1679
+ LI
1680
+
1681
+ pflow
1682
+ LI
1683
+
1684
+ +
1685
+ + Cm,ex
1686
+ I
1687
+ (pm
1688
+ I )− + Cm,im
1689
+ I
1690
+ (pm
1691
+ I )+
1692
+ for each microgrid I ∈ D, with local export and import per-unit prices Cflow,ex
1693
+ LI
1694
+ ,
1695
+ Cflow,im
1696
+ LI
1697
+ , Cm,ex
1698
+ I
1699
+ , Cm,im
1700
+ I
1701
+ ∈ R≥0, which may fluctuate throughout the day. In
1702
+ the numerical example in Section 6, we will consider import prices that are
1703
+ considerably higher than the export prices and thus focus on small producers,
1704
+ for which in practice it is often more profitable to generate power for their
1705
+ own demand than to buy electricity from the main grid, and for which power
1706
+ exports to the grid are only worthwhile at times of high demand. The terminal
1707
+
1708
+ Springer Nature 2021 LATEX template
1709
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
1710
+ 19
1711
+ costs mI : [0.0, 1.0] → R≥0 account for degradation costs of the battery as
1712
+ mI(sI) := Cdis
1713
+ I
1714
+ (sI(0) − sI(T))+ Qst
1715
+ I .
1716
+ If the state of charge sI(T) at the end of the considered horizon is smaller than
1717
+ sI(0) at the beginning, each unit of power discharge is penalized by some cost
1718
+ Cdis
1719
+ I
1720
+ ∈ R≥0.
1721
+ 6 Numerical Experiments with Microgrids
1722
+ Under Attack
1723
+ In this section, we present a numerical case study to analyze the performance
1724
+ of adaptively robust DMPC from Section 4 in the context of interconnected
1725
+ microgrids under attack using the model from Section 5. In contrast to our
1726
+ earlier work [17], we apply distributed ADI based on the local identification
1727
+ problem (5). In the experiments, we address the question of how to achieve
1728
+ an economic operation of microgrids at minimum costs despite uncertainties.
1729
+ Whether these emerge in form of disturbances with rather small impact, fluc-
1730
+ tuating generation from renewables, or malicious attacks; all represent critical
1731
+ yet all the more relevant threats to energy supply.
1732
+ To this end, we consider three microgrids I, II, and III with renewable gen-
1733
+ eration that are each connected to the main grid and the other two microgrids
1734
+ as in Figure 3. The initial values and bounds for all variables of the microgrid
1735
+ model are given in Table 1 and the parameters are chosen as in Table 2, using
1736
+ those for lithium-titanate (Li4Ti5O12) batteries from [27] in (17).
1737
+ For a timespan of two days, robust NMPC is applied locally with step size
1738
+ ∆t = 0.25 h by each microgrid. At time t ∈ [0.0, 48.0] h, the local cost func-
1739
+ tion JI in (22) takes into account the upcoming time window [t, t + Np] with
1740
+ prediction horizon Np = 6.0 h and uses the cost parameters from Table 2. The
1741
+ values Cm,im
1742
+ I
1743
+ and Cm,ex
1744
+ I
1745
+ , that describe the cost or revenue of power imports
1746
+ from or exports to the main grid, vary in the course of the day. In our example,
1747
+ we focus on microgrids that represent small local prosumers and use the fol-
1748
+ lowing fictitious values for all microgrids, which are based on real prices on the
1749
+ Table 1: This table lists lower and upper bounds as well as initial values at
1750
+ time t = 0 for all state and input variables of the microgrid model. For the state
1751
+ of charge, three distinct initial values sI(0) for the three microgrids I, II, and
1752
+ III are given. In all other cases, the indicated values apply for all subsystems.
1753
+ Variable
1754
+ Lower Bound
1755
+ Upper Bound
1756
+ Initial Value
1757
+ Unit
1758
+ sI
1759
+ 0.0
1760
+ 0.1
1761
+ 0.9, 0.5, 0.6
1762
+ -
1763
+ pg
1764
+ I
1765
+ 0.0
1766
+ 1000.0
1767
+ 0.0
1768
+ kW
1769
+ pm
1770
+ I
1771
+ -1000.0
1772
+ 2000.0
1773
+ 0.0
1774
+ kW
1775
+ ptr
1776
+ IL
1777
+ -100.0
1778
+ 100.0
1779
+ 0.0
1780
+ kW
1781
+ ug
1782
+ I
1783
+ 0.0
1784
+ 1000.0
1785
+ -
1786
+ kW
1787
+ um
1788
+ I
1789
+ -1000.0
1790
+ 2000.0
1791
+ -
1792
+ kW
1793
+ utr
1794
+ IL
1795
+ -100.0
1796
+ 100.0
1797
+ -
1798
+ kW
1799
+
1800
+ Springer Nature 2021 LATEX template
1801
+ 20
1802
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
1803
+ Table 2: This table lists all model and cost parameters that are used in
1804
+ the numerical experiments presented in this section. All values apply to all
1805
+ subsystems I ∈ {I, II, III}, except for Qst
1806
+ I , Rst
1807
+ I , and Cg
1808
+ I , where individual values
1809
+ for the respective subsystems are specified.
1810
+ (a) Model Parameters
1811
+ Param.
1812
+ Value
1813
+ Unit
1814
+ pl
1815
+ I
1816
+ -2.0
1817
+ kW
1818
+ T g
1819
+ I
1820
+ 0.1
1821
+ h
1822
+ T m
1823
+ I
1824
+ 0.001
1825
+ h
1826
+ T tr
1827
+ IL
1828
+ 0.001
1829
+ h
1830
+ Qst
1831
+ I
1832
+ 100, 200, 100
1833
+ kAh
1834
+ Rst
1835
+ I
1836
+ 1.5, 2.0, 3.0
1837
+ mΩ
1838
+ (b) OCV Parameters
1839
+ Param.
1840
+ Value
1841
+ Unit
1842
+ αI
1843
+ 2.23
1844
+ V
1845
+ βI
1846
+ -0.001
1847
+ V
1848
+ γI
1849
+ -0.35
1850
+ V
1851
+ δI
1852
+ 0.6851
1853
+ V
1854
+ µI
1855
+ 3.0
1856
+ -
1857
+ νI
1858
+ 1.6
1859
+ -
1860
+ (c) Cost Parameters
1861
+ Param.
1862
+ Value
1863
+ Cg
1864
+ I
1865
+ 0.2, 3.0, 2.0
1866
+ Ctr
1867
+ I
1868
+ 4.0
1869
+ Cst
1870
+ I
1871
+ 1.0
1872
+ Cdis
1873
+ I
1874
+ 2000
1875
+ Cflow,im
1876
+ IL
1877
+ 4.0
1878
+ Cflow,ex
1879
+ IL
1880
+ 0.04
1881
+ German electricity market in 2021 [28] and reflect typical market fluctuations
1882
+ with rising prices in the morning and evening hours:
1883
+ Cm,im
1884
+ I
1885
+ (t) =
1886
+
1887
+
1888
+
1889
+
1890
+
1891
+
1892
+
1893
+
1894
+
1895
+ 275 if (t mod 24 h) ∈ [15, 20) h,
1896
+ 200 if (t mod 24 h) ∈ [6, 9) ∪ [20, 22) h,
1897
+ 150 if (t mod 24 h) ∈ [9, 15) ∪ [22, 24) h,
1898
+ 100 otherwise,
1899
+ Cm,ex
1900
+ I
1901
+ (t) =
1902
+
1903
+
1904
+
1905
+
1906
+
1907
+ 15 if (t mod 24 h) ∈ [15, 20) h,
1908
+ 10 if (t mod 24 h) ∈ [6, 9) ∪ [20, 22) h,
1909
+ 0 otherwise.
1910
+ Here, mod is the modulo operator and (t mod 24 h) denotes the time of day.
1911
+ To achieve a resilient operation, the system is controlled using the adap-
1912
+ tively robust distributed NMPC scheme from Section 4.2. Based on the local
1913
+ control problem (9), at each sampling time k every microgrid computes con-
1914
+ tracts
1915
+
1916
+ X l,[k]
1917
+ I
1918
+ to confine the behavior of its future coupling values zl
1919
+ I for
1920
+ l ∈ {k, . . . , k + Np − 1} and shares them with its neighbors. In contrast to the
1921
+ experiments in [17], which involve a centralized ADI method, each microgrid
1922
+ consults locally identified solutions a∗,k
1923
+ I
1924
+ of problem (5) to update its estimates
1925
+ µ[k]
1926
+ I
1927
+ and σ[k]
1928
+ I
1929
+ of the expected value and standard deviation of the unknown
1930
+ random attack aI as in (10). In our numerical experiments, the nonlinear
1931
+ identification problem (5) is solved to an accuracy of εI = 10−3 using the
1932
+ interior-point solver Ipopt [29]. The states xI are assumed to be only partially
1933
+ observable with linear output function cI : XI → YI that is defined as
1934
+ cI (xI) := diag(1, 1, 1, 0, 0)xI.
1935
+ This means that for each microgrid I, the outputs yI = (sI, pg
1936
+ I, pm
1937
+ I )⊤ are con-
1938
+ sidered by the local identification process, but not the transfer variables ptr
1939
+ IL
1940
+ for all L ∈ NI. Based on the suspected attacks a∗,k
1941
+ I
1942
+ and the derived estimates
1943
+
1944
+ Springer Nature 2021 LATEX template
1945
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
1946
+ 21
1947
+
1948
+
1949
+
1950
+
1951
+
1952
+
1953
+
1954
+ sI robust
1955
+ sI non-robust
1956
+ SoC in %
1957
+ 100
1958
+ 96
1959
+ 92
1960
+ 88
1961
+ (a) State of Charge
1962
+
1963
+
1964
+
1965
+
1966
+
1967
+
1968
+
1969
+ Generation in kW
1970
+ 25
1971
+ 20
1972
+ 15
1973
+ 10
1974
+ 5
1975
+ 0
1976
+ ug
1977
+ I
1978
+ pg
1979
+ I
1980
+ (b) Power Generation
1981
+
1982
+
1983
+
1984
+
1985
+
1986
+
1987
+
1988
+ Time in hours
1989
+ 0.0
1990
+ 12.0
1991
+ 24.0
1992
+ 36.0
1993
+ 48.0
1994
+ Imports / exports in kW
1995
+ 0
1996
+ -5
1997
+ -10
1998
+ -15
1999
+ -20
2000
+ -25
2001
+ -30
2002
+ pm
2003
+ I
2004
+ (c) Power exchange with main grid
2005
+
2006
+
2007
+
2008
+
2009
+
2010
+
2011
+
2012
+ Time in hours
2013
+ 0.0
2014
+ 12.0
2015
+ 24.0
2016
+ 36.0
2017
+ 48.0
2018
+ Transfers in kW
2019
+ 2.0
2020
+ 1.5
2021
+ 1.0
2022
+ 0.5
2023
+ 0.0
2024
+ ptr
2025
+ I,II
2026
+ ptr
2027
+ I,III
2028
+ (d) Power exchange with neighbors
2029
+ Fig. 4: Selected state and input trajectories for microgrid I, showing all powers
2030
+ in kW. The microgrid is exposed to a generator attack, causing the generation
2031
+ pg
2032
+ I to be considerably larger than planned by ug
2033
+ I . The different SoC trajectories,
2034
+ computed by adaptively robust versus nonrobust NMPC, show the benefit of
2035
+ the proposed resilient control framework.
2036
+ µ[k]
2037
+ I
2038
+ and σ[k]
2039
+ I , the uncertainty sets �
2040
+ Al,[k]
2041
+ I
2042
+ are approximated as in (11). The local
2043
+ control problem (9) is repeatedly adapted to new contracts and identification
2044
+ results that become available. As a consequence, the inputs ul
2045
+ I computed at
2046
+ time k+1 for l ∈ {k+1, . . . , k+Np} are robust toward deviations in neighboring
2047
+ couplings within �
2048
+ Zl,[k]
2049
+ NI
2050
+ and identified attacks in �
2051
+ Al,[k]
2052
+ I
2053
+ .
2054
+ We examine the behavior of the system, controlled with Algorithm 2, in
2055
+ two attack scenarios. For comparison, we repeat each experiment with nonro-
2056
+ bust DMPC, where neither contracts are exchanged nor attack identification
2057
+ is considered. First, we assume that all generation units are dispatchable and
2058
+ a constant attack ag
2059
+ I = 10.0 kW disrupts the generator dynamics in micro-
2060
+ grid I according to (12). The attacker is active over the entire time window
2061
+ [0.0, 48.0] h and causes a severe deviation of the generated power pg
2062
+ I in micro-
2063
+ grid I from the control input ug
2064
+ I as Figure 4 reveals. The distributed ADI
2065
+ method based on the local identification problem (5) successfully identifies
2066
+ the unknown attack input with very high precision in every time step as
2067
+ pointed out by Figure 5, which shows the mean of the suspected attack values
2068
+ ag,∗
2069
+ I
2070
+ ≈ 9.9989 kW at all times. This allows the local robust NMPC scheme to
2071
+ adapts its prediction very accurately and adjust the control inputs accordingly.
2072
+ As a result, the microgrid takes advantage of the additional power generation
2073
+
2074
+ Springer Nature 2021 LATEX template
2075
+ 22
2076
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
2077
+
2078
+
2079
+ Actual attack ag
2080
+ I
2081
+ Identified mean µ[k]
2082
+ I
2083
+ Time in hours
2084
+ 0.0
2085
+ 12.0
2086
+ 24.0
2087
+ 36.0
2088
+ 48.0
2089
+ Attack ag
2090
+ I in kW
2091
+ 10.015
2092
+ 10.010
2093
+ 10.005
2094
+ 10.000
2095
+ 9.995
2096
+ Fig. 5: Actual attack value ag
2097
+ I and average identified value µ[k]
2098
+ I
2099
+ in the first
2100
+ attack scenario examined, in which only dispatchable generation units are in
2101
+ use and microgrid I is exposed to a generator attack.
2102
+ by charging the battery and exporting the power to the main grid during times
2103
+ with high profit. In the solution computed with nonrobust NMPC, on the con-
2104
+ trary, the battery reaches and violates its maximum state of charge of 1.0 after
2105
+ about 5.0 h as the red SoC trajectory in Figure 4a reveals. Due to bound vio-
2106
+ lations, the nonrobust scheme fails in 171 of 192 time steps when more power
2107
+ than planned is generated and the storage is charged to maintain power bal-
2108
+ ance. Since SoC values larger than 1.0 are physically invalid, the next MPC
2109
+ step in our study continues at sI = 1.0.
2110
+ It should be noted that power balance can be ensured in other ways than
2111
+ using the storage as a buffer. For instance, if power imports from and exports
2112
+ to the main grid are allowed at all times, using the grid as a buffer would not
2113
+ lead to bound violations as above. However, this can cause very high costs, for
2114
+ example, if electricity has to be imported in the evening at expensive prices. In
2115
+ contrast, the battery allows power to be stored until exports to the main grid
2116
+ become profitable. Indeed, over the entire period of two days, the adaptively
2117
+ robust NMPC scheme achieves total costs of −5.2·103 in microgrid I and thus
2118
+ makes profit despite the attack. On the contrary, nonrobust NMPC causes
2119
+ total local costs of 2.3 · 104, which is orders of magnitudes larger. Considering
2120
+ that we aim for a strategy to increase the resilience of the system, which takes
2121
+ into account not only robustness but also performance in terms of induced
2122
+ costs, the battery as a buffer is therefore a reasonable choice that enables and
2123
+ favors high resilience.
2124
+ In the second experiment, we consider a modified generator attack
2125
+ ag
2126
+ I = 10.0 kW + rg
2127
+ I , where rg
2128
+ I ∼ N(0.0, 8.0) kW represents the uncertainty in
2129
+ renewable generation and is randomly drawn from a normal distribution with
2130
+ mean 0.0 kW and standard deviation 8.0 kW, independently at each time step.
2131
+ Together, the malicious attack of 10.0 kW and the renewable fluctuations rg
2132
+ I
2133
+
2134
+ Springer Nature 2021 LATEX template
2135
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
2136
+ 23
2137
+
2138
+
2139
+
2140
+ Actual attack ag
2141
+ I
2142
+ Identified mean µ[k]
2143
+ I
2144
+ Sample std. dev. σ[k]
2145
+ I
2146
+ Time in hours
2147
+ 0.0
2148
+ 12.0
2149
+ 24.0
2150
+ 36.0
2151
+ 48.0
2152
+ Disturbance ag
2153
+ I in kW
2154
+ 40.0
2155
+ 30.0
2156
+ 20.0
2157
+ 10.0
2158
+ 0.0
2159
+ -10.0
2160
+ Fig. 6: Course of the mean µ[k]
2161
+ I
2162
+ of identified values a∗,k
2163
+ I
2164
+ over time, with sample
2165
+ standard deviation σ[k]
2166
+ I . The actual disturbance ag,k
2167
+ I
2168
+ at each time k is shown
2169
+ in orange. The figure is taken from [18, Fig. 3].
2170
+ may cause more power than planned to be generated (i. e., ag
2171
+ I > 0) or less (i. e.,
2172
+ ag
2173
+ I < 0), but are chosen such that the total generator input ug
2174
+ I + ag
2175
+ I is nonneg-
2176
+ ative. Due to the fluctuating generation, the actual value ag
2177
+ I of the unknown
2178
+ disturbance in the generator dynamics ranges from −11.3 kW to 42.9 kW as
2179
+ can be seen in Figure 6. For the examined generator with parameters as in
2180
+ Table 2, this is a very broad range, which also becomes clear in comparison
2181
+ with Figure 4b. As an apparent consequence of the continually changing val-
2182
+ ues, the local identification problem (5) yields a different suspicion ag,∗
2183
+ I
2184
+ in each
2185
+ time step. Nevertheless, Figure 6 shows that the mean µ[k]
2186
+ I
2187
+ of identified values
2188
+ quickly settles at about 10.0 kW, which underlines that the distributed ADI
2189
+ method is able to cope also with highly fluctuating and widely dispersed dis-
2190
+ turbances, since a new optimization problem is solved at each time step. This
2191
+ proves once again the great potential of the proposed class of optimization-
2192
+ based ADI methods and emphasizes that they are not tailored to a specific
2193
+ type of attack, but are also very well suited for challenging scenarios where
2194
+ attacks and other sources of significant uncertainty congregate.
2195
+ The sample standard deviation σ[k]
2196
+ I
2197
+ is considerably larger than before and
2198
+ the three scenarios µ[k]
2199
+ I , µ[k]
2200
+ I
2201
+ + σ[k]
2202
+ I , and µ[k]
2203
+ I
2204
+ − σ[k]
2205
+ I
2206
+ are further apart than in
2207
+ the first experiment. Figure 7 shows the obtained solution for the attacked
2208
+ microgrid I. While adaptively robust DMPC achieves total local costs of 3.1·103
2209
+ in microgrid I, the nonrobust approach causes more than ten times higher total
2210
+ costs of 3.2·104. Once again, classical nonrobust MPC proves to be unsuitable
2211
+
2212
+ Springer Nature 2021 LATEX template
2213
+ 24
2214
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
2215
+
2216
+
2217
+
2218
+
2219
+
2220
+
2221
+
2222
+ sI robust
2223
+ sI non-robust
2224
+ SoC in %
2225
+ 100
2226
+ 95
2227
+ 90
2228
+ 85
2229
+ 80
2230
+ (a) State of Charge
2231
+
2232
+
2233
+
2234
+
2235
+
2236
+
2237
+
2238
+ Generation in kW
2239
+ 60
2240
+ 40
2241
+ 20
2242
+ 0
2243
+ ug
2244
+ I
2245
+ pg
2246
+ I
2247
+ (b) Power Generation
2248
+
2249
+
2250
+
2251
+
2252
+
2253
+
2254
+
2255
+ Time in hours
2256
+ 0.0
2257
+ 12.0
2258
+ 24.0
2259
+ 36.0
2260
+ 48.0
2261
+ Imports / exports in kW
2262
+ 0
2263
+ -10
2264
+ -20
2265
+ -30
2266
+ -40
2267
+ pm
2268
+ I
2269
+ (c) Power exchange with main grid
2270
+
2271
+
2272
+
2273
+
2274
+
2275
+
2276
+
2277
+ Time in hours
2278
+ 0.0
2279
+ 12.0
2280
+ 24.0
2281
+ 36.0
2282
+ 48.0
2283
+ Transfers in kW
2284
+ 1.5
2285
+ 1.0
2286
+ 0.5
2287
+ 0.0
2288
+ -0.5
2289
+ -1.0
2290
+ ptr
2291
+ I,II
2292
+ ptr
2293
+ I,III
2294
+ (d) Power exchange with neighbors
2295
+ Fig. 7: States and inputs in microgrid I, which now contains renewable
2296
+ generation as another source of uncertainty in addition to the generator attack.
2297
+ to control the disturbed system as it computes a solution that violates the
2298
+ upper bound of the state of charge in 113 of 192 time steps.
2299
+ At this point, we would like to point out that the adaptively robust DMPC
2300
+ scheme is not guaranteed to yield admissible trajectories in all cases. In fact,
2301
+ proving rigorous guarantees of this kind is challenging for nonlinear dynam-
2302
+ ics. Moreover, in contrast to the multi-stage approach [30], adaptively robust
2303
+ NMPC lacks the recursive feasibility property when the attack uncertainty sets
2304
+ Al,[k]
2305
+ I
2306
+ are adjusted to sudden attacks. Furthermore, Figure 6 illustrates that
2307
+ in our second attack scenario involving uncertain renewable generation, even
2308
+ disturbances ag
2309
+ I occur that are not within the interval [µ[k]
2310
+ I
2311
+ − σ[k]
2312
+ I , µ[k]
2313
+ I
2314
+ + σ[k]
2315
+ I ].
2316
+ Despite these unforeseen disruptions and the lack of theoretical guarantees,
2317
+ however, all state bounds are satisfied and the solution in Figure 7 is not overly
2318
+ conservative judging from the fact that considerably lower costs are obtained
2319
+ than with nonrobust DMPC. This underlines that adaptively robust NMPC,
2320
+ using ADI results as estimates for an unknown attack, is a very powerful tool
2321
+ even under challenging circumstances with broadly dispersed disturbances.
2322
+ 7 Conclusion and Future Directions
2323
+ We introduced a comprehensive distributed MPC framework for nonlinear
2324
+ control systems under attack, which is based on local multi-stage control and
2325
+ novel distributed attack identification methods in each subsystem. To enable
2326
+
2327
+ Springer Nature 2021 LATEX template
2328
+ Resilient MPC of Distributed Systems Under Attack Using Local ADI
2329
+ 25
2330
+ the system to respond autonomously and robustly to identified perturbations,
2331
+ each control scheme represents the uncertain influence of neighboring cou-
2332
+ plings and attack inputs by scenario sets that are continuously updated based
2333
+ on newly gained knowledge. For this purpose, each subsystem applies local
2334
+ attack identification and repeatedly transmits new contract information to its
2335
+ neighbors. Using the example of microgrids interconnected by power transfers,
2336
+ the methodology was demonstrated to robustly control a distributed system
2337
+ and achieve constraint satisfaction at all times despite unknown attacks and
2338
+ uncertain renewable generation.
2339
+ We have identified two promising directions with great potential for future
2340
+ research. The first would be to derive theoretical conditions under which Algo-
2341
+ rithm 1 can be rigorously proven to successfully identify the correct inputs,
2342
+ similar to the guarantees for our centralized ADI method [16]. While some
2343
+ ideas from [16] can be transferred with few changes, further required theoreti-
2344
+ cal arguments could be based on the research results on nonlinear compressed
2345
+ sensing. For example, in [22] the restricted isometry property from [20], a cen-
2346
+ tral component of linear compressed sensing, is generalized and the iterative
2347
+ hard thresholding algorithm involving a form of gradient projection is extended
2348
+ to nonlinear systems. Furthermore, in [23] two coordinate descent methods
2349
+ are introduced that build upon the simplex algorithm for linear programming
2350
+ and are of a greedy type in the sense that they add nonzero variables one by
2351
+ one. When suitable success guarantees for the new distributed ADI approaches
2352
+ provably hold, a combination with the robustness and stability analysis of
2353
+ multi-stage NMPC and contract-based DMPC described in [3, 30, 31] could be
2354
+ the next step to strengthen the excellent numerical performance of adaptively
2355
+ robust DMPC by theoretical arguments.
2356
+ The second research direction consists in investigating a hierarchical com-
2357
+ bination of several ADI approaches that complement each other and provide
2358
+ system operators with different options suiting their needs. There is, on the
2359
+ one hand, the centralized ADI method from [16], which is based on an approx-
2360
+ imation of the dynamics and provides quick insights into the network-wide
2361
+ attack situation, but requires all subsystems to make specific sensitivity infor-
2362
+ mation publicly available and agree on a central instance to solve the global
2363
+ identification problem. On the other hand, there are distributed ADI methods
2364
+ like Algorithm 1 involving problems (5) and (8), which use local models to
2365
+ analyze possible attacks on one subsystems or its neighborhood locally. Sev-
2366
+ eral gradations or variants of these approaches may be applied, depending on
2367
+ the available model knowledge and the willingness of individual subsystems to
2368
+ cooperate or agree on a common decision instance.
2369
+ 8 Statement on Conflict of Interests
2370
+ On behalf of all authors, the corresponding author states that there is no
2371
+ conflict of interest.
2372
+
2373
+ Springer Nature 2021 LATEX template
2374
+ 26
2375
+ REFERENCES
2376
+ References
2377
+ [1] Christofides, P., Scattolini, R., de la Pena, D., Liu, J.: Distributed
2378
+ model predictive control: A tutorial review and future research directions.
2379
+ Computers & Chemical Engineering 51, 21–41 (2013)
2380
+ [2] Arauz, T., Chanfreut, P., Maestre, J.: Cyber-security in networked and
2381
+ distributed model predictive control. Annual Reviews in Control 53, 338–
2382
+ 355 (2022)
2383
+ [3] Lucia, S., K¨ogel, M., Findeisen, R.: Contract-based predictive control of
2384
+ distributed systems with plug and play capabilities. IFAC-PapersOnLine
2385
+ 48, 205–211 (2015)
2386
+ [4] Mayne, D., Seron, M., Rakovi´c, S.: Robust model predictive control of
2387
+ constrained linear systems with bounded disturbances. Automatica 41,
2388
+ 219–224 (2005)
2389
+ [5] Lucia, S., Finkler, T., Engell, S.: Multi-stage nonlinear model predictive
2390
+ control applied to a semi-batch polymerization reactor under uncertainty.
2391
+ Journal of Process Control 23, 1306–1319 (2013)
2392
+ [6] Wang, Y., Ishii, H.: A distributed model predictive scheme for resilient
2393
+ consensus with input constraints. In: IEEE Conference on Control Tech-
2394
+ nology and Applications, pp. 349–354 (2019)
2395
+ [7] Braun, S., Albrecht, S., Lucia, S.: Identifying attacks on nonlinear cyber-
2396
+ physical systems in a robust model predictive control setup. In: European
2397
+ Control Conference, pp. 513–520 (2020). IEEE
2398
+ [8] Braun, S., Albrecht, S., Lucia, S.: Hierarchical attack identification for
2399
+ distributed robust nonlinear control. In: 21st IFAC World Congress, pp.
2400
+ 6191–6198 (2020)
2401
+ [9] Pasqualetti, F., D¨orfler, F., Bullo, F.: Attack detection and identification
2402
+ in cyber-physical systems. IEEE Transactions on Automatic Control 58,
2403
+ 2715–2729 (2013)
2404
+ [10] Giraldo, J., Urbina, D., Cardenas, A., Valente, J., Faisal, M., Ruths, J.,
2405
+ Tippenhauer, N., Sandberg, H., Candell, R.: A survey of physics-based
2406
+ attack detection in cyber-physical systems. ACM Computing Surveys 51,
2407
+ 1–36 (2018)
2408
+ [11] Boem, F., Riverso, S., Ferrari-Trecate, G., Parisini, T.: Plug-and-play
2409
+ fault detection and isolation for large-scale nonlinear systems with
2410
+ stochastic uncertainties. IEEE Transactions on Automatic Control 64,
2411
+ 4–19 (2018)
2412
+ [12] Gallo, A., Turan, M., Boem, F., Parisini, T., Ferrari-Trecate, G.: A
2413
+ distributed cyber-attack detection scheme with application to DC micro-
2414
+ grids. IEEE Transactions on Automatic Control 65, 3800–3815 (2020)
2415
+ [13] Boem, F., Ferrari, R., Parisini, T.: Distributed fault detection and isola-
2416
+ tion of continuous-time non-linear systems. European Journal of Control
2417
+ 17, 603–620 (2011)
2418
+ [14] Pan, W., Yuan, Y., Sandberg, H., Gon¸calves, J., Stan, G.: Online fault
2419
+ diagnosis for nonlinear power systems. Automatica 55, 27–36 (2015)
2420
+ [15] Ananduta, W., Maestre, J., Ocampo-Martinez, C., Ishii, H.: Resilient
2421
+
2422
+ Springer Nature 2021 LATEX template
2423
+ REFERENCES
2424
+ 27
2425
+ distributed model predictive control for energy management of inter-
2426
+ connected microgrids. Optimal Control Applications and Methods 41,
2427
+ 146–169 (2020)
2428
+ [16] Braun, S., Albrecht, S., Lucia, S.: Attack identification for nonlinear sys-
2429
+ tems based on sparse optimization. IEEE Transactions on Automatic
2430
+ Control, early access (2021)
2431
+ [17] Braun, S., Albrecht, S., Lucia, S.: Adaptively robust nonlinear model pre-
2432
+ dictive control based on attack identification. at-Automatisierungstechnik
2433
+ 70, 367–377 (2022)
2434
+ [18] Braun, S., Albrecht, S., Lucia, S.: Resilient Control of Interconnected
2435
+ Microgrids Under Attack by Robust Nonlinear MPC. In: Conference
2436
+ on Informatics in Control, Automation and Robotics, pp. 58–66 (2022).
2437
+ INSTICC
2438
+ [19] Kozma, A., Savorgnan, C., Diehl, M.: Distributed multiple shooting for
2439
+ large scale nonlinear systems. In: Distributed Model Predictive Control
2440
+ Made Easy, pp. 327–340. Springer
2441
+ [20] Cand`es, E., Tao, T.: Decoding by linear programming. IEEE Transactions
2442
+ on Information Theory 51, 4203–4215 (2005)
2443
+ [21] Forster, O.: Analysis 2 - Differentialrechnung im Rn, Gew¨ohnliche Differ-
2444
+ entialgleichungen. 11 edn. Springer (2010)
2445
+ [22] Blumensath, T.: Compressed sensing with nonlinear observations and
2446
+ related nonlinear optimization problems. IEEE Transactions on Informa-
2447
+ tion Theory 59, 3466–3474 (2013)
2448
+ [23] Beck, A., Eldar, Y.: Sparsity constrained nonlinear optimization: Opti-
2449
+ mality conditions and algorithms. SIAM Journal on Optimization 23,
2450
+ 1480–1509 (2013)
2451
+ [24] Olivares, D., Mehrizi-Sani, A., Etemadi, A., Ca˜nizares, C., Iravani, R.,
2452
+ et al.: Trends in microgrid control. IEEE Transactions on Smart Grid 5,
2453
+ 1905–1919 (2014)
2454
+ [25] Mohammed, A., Refaat, S., Bayhan, S., Abu-Rub, H.: AC microgrid
2455
+ control and management strategies: evaluation and review. IEEE Power
2456
+ Electronics Magazine 6, 18–31 (2019)
2457
+ [26] Mathieu, J., Taylor, J.: Controlling nonlinear batteries for power sys-
2458
+ tems: Trading off performance and battery life. In: IEEE Power Systems
2459
+ Computation Conference, pp. 1–7 (2016)
2460
+ [27] Zhang, C., Jiang, J., Zhang, L., Liu, S., Wang, L., Loh, P.C.: A generalized
2461
+ SOC-OCV model for lithium-ion batteries and the SOC estimation for
2462
+ LNMCO battery. Energies 9, 1–16 (2016)
2463
+ [28] Bundesnetzagentur Deutschland: SMARD Strommarktdaten for Ger-
2464
+ many in November 2021. https://www.smard.de/home/downloadcenter/
2465
+ download-marktdaten. Online, last accessed: November 15th, 2022
2466
+ [29] W¨achter, A., Biegler, L.: On the Implementation of an Interior-Point
2467
+ Filter Line-Search Algorithm for Large-Scale Nonlinear Programming.
2468
+ Mathematical Programming 106, 25–57 (2006)
2469
+ [30] Lucia, S., Subramanian, S., Limon, D., Engell, S.: Stability properties of
2470
+
2471
+ Springer Nature 2021 LATEX template
2472
+ 28
2473
+ REFERENCES
2474
+ multi-stage nonlinear model predictive control. Systems & Control Letters
2475
+ 143, 104743 (2020)
2476
+ [31] Lucia, S., Paulen, R., Engell, S.: Multi-stage nonlinear model predic-
2477
+ tive control with verified robust constraint satisfaction. In: Conference on
2478
+ Decision and Control, pp. 2816–2821 (2014). IEEE
2479
+
EtE5T4oBgHgl3EQfUw_x/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
HNE1T4oBgHgl3EQf_Qa5/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3e26b0a71249592f3e91b80313418e8f5fb941f6c17a028a91479d594938dd2b
3
+ size 3145773
HNE1T4oBgHgl3EQf_Qa5/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe0560bc07d074c12cfd5487033be6a64175826106d3b582eb1c4a31a46878ac
3
+ size 137971
ItFAT4oBgHgl3EQfux4k/content/2301.08671v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7d651f70a5e2825466ab41a8736c160deaaf44679af2ced1d6928a14f013bf44
3
+ size 3107432
J9E3T4oBgHgl3EQfvQsy/content/tmp_files/2301.04691v1.pdf.txt ADDED
@@ -0,0 +1,2333 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ QoS Based Contract Design for Profit Maximization
2
+ in IoT-Enabled Data Markets
3
+ Juntao Chen, Member, IEEE, Junaid Farooq, Member, IEEE and Quanyan Zhu, Senior Member, IEEE
4
+ Abstract—The massive deployment of Internet of Things (IoT)
5
+ devices, including sensors and actuators, is ushering in smart and
6
+ connected communities of the future. The massive deployment of
7
+ Internet of Things (IoT) devices, including sensors and actuators,
8
+ is ushering in smart and connected communities of the future.
9
+ The availability of real-time and high-quality sensor data is
10
+ crucial for various IoT applications, particularly in healthcare,
11
+ energy, transportation, etc. However, data collection may have
12
+ to be outsourced to external service providers (SPs) due to
13
+ cost considerations or lack of specialized equipment. Hence,
14
+ the data market plays a critical role in such scenarios where
15
+ SPs have different quality levels of available data, and IoT
16
+ users have different application-specific data needs. The pairing
17
+ between data available to the SP and users in the data market
18
+ requires an effective mechanism design that considers the SPs’
19
+ profitability and the quality-of-service (QoS) needs of the users.
20
+ We develop a generic framework to analyze and enable such
21
+ interactions efficiently, leveraging tools from contract theory and
22
+ mechanism design theory. It can enable and empower emerging
23
+ data sharing paradigms such as Sensing-as-a-Service (SaaS). The
24
+ contract design creates a pricing structure for on-demand sensing
25
+ data for IoT users. By considering a continuum of user types,
26
+ we capture a diverse range of application requirements and
27
+ propose optimal pricing and allocation rules that ensure QoS
28
+ provisioning and maximum profitability for the SP. Furthermore,
29
+ we provide analytical solutions for fixed distributions of user
30
+ types to analyze the developed approach. For comparison, we
31
+ consider the benchmark case assuming complete information of
32
+ the user types and obtain optimal contract solutions. Finally, a
33
+ case study based on the example of virtual reality application
34
+ delivered using unmanned aerial vehicles (UAVs) is presented
35
+ to demonstrate the efficacy of the proposed contract design
36
+ framework.
37
+ Index Terms—Contract design, data pricing, Internet of things,
38
+ Maximum principle, quality-of-service, sensing-as-a-service.
39
+ I. INTRODUCTION
40
+ The Internet of things (IoT) applications rely heavily on
41
+ sensed data from a multitude of sources resulting in power-
42
+ ful and intelligent applications based on sensor fusion and
43
+ machine learning. For instance, smart and connected commu-
44
+ nities, industrial automation, smart grid all rely on reliable
45
+ and high quality data for automated decision-making [1]. To
46
+ This work was supported in part by the National Science Foundation (NSF)
47
+ under Grants ECCS-1847056, CNS-2027884, and BCS-2122060.
48
+ Juntao Chen is with the Department of Computer and Information
49
+ Sciences,
50
+ Fordham
51
+ University,
52
+ New
53
+ York,
54
+ NY
55
+ 10023
56
+ USA.
57
+ E-mail:
58
59
+ Junaid Farooq is with the Department of Electrical & Computer Engineer-
60
+ ing, College of Engineering and Computer Science, University of Michigan-
61
+ Dearborn, Dearborn, MI 48128 USA. E-mail: [email protected].
62
+ Quanyan Zhu is with the Department of Electrical and Computer Engi-
63
+ neering, Tandon School of Engineering, New York University, Brooklyn, NY,
64
+ 11201 USA. E-mail: [email protected].
65
+ UAVs
66
+ VR users
67
+ VR SP
68
+ VR
69
+ services
70
+ Service
71
+ fee
72
+ Sensing
73
+ data
74
+ Fig. 1.
75
+ In the UAV-enabled VR applications, the UAVs capture views
76
+ of the areas of interest. The collected data are aggregated in the cloud,
77
+ which is managed by the VR SP, and then sent to the remote users. The
78
+ real-time 3D information delivery is useful in applications such as remote
79
+ monitoring, navigation, and entertainment. Based on the application, VR users
80
+ have different QoS requirements and pay different service fees.
81
+ fulfil the data needs of intelligence-based IoT applications,
82
+ the sensing and data acquisition tasks can be outsourced to
83
+ professional service providers (SPs) in the data market [2].
84
+ It results in cost effective data collection for IoT applica-
85
+ tions, wider choice of sensing data, and on-demand service
86
+ delivery to users. For example, in an intelligent transportation
87
+ network, vehicles can choose the services to communicate
88
+ with roadside infrastructures that belong to sensing SP for
89
+ exchanging various types of data related to applications such
90
+ as GPS navigation, parking, and highway tolls inquiries. etc.
91
+ Another potential scenario is UAV-enabled virtual reality (VR)
92
+ experiences [3]. As shown in Fig. 1, the UAVs managed by
93
+ the SP capture 3D images of areas that users are interested
94
+ in, and send them to the remote users via cloud servers and
95
+ communication networks. These images can be of varying
96
+ quality and resolution suited for a range of different user
97
+ types. Therefore, the service interactions between the users
98
+ and the sensing SP requires a formal contract design, in
99
+ which IoT users make subscription contracts with the SP to
100
+ obtain (real-time) sensor data according to specific mission
101
+ requirements [4].
102
+ Depending on the particular application, IoT users have
103
+ different requirements on the quality of data provided by the
104
+ sensing SP. Note that provisioning of high-quality sensing
105
+ data demands high-level of investment in terms of equipment
106
+ deployment, maintenance, technical support, and data process-
107
+ ing from the SP. In the UAV-enabled VR, users may require
108
+ different levels of quality-of-service (QoS) in terms of the
109
+ transmission delay and resolution of the images. Therefore,
110
+ users with different QoS needs can be classified into different
111
+ arXiv:2301.04691v1 [eess.SY] 11 Jan 2023
112
+
113
+ HAKIOABtypes1. The sensing SP aims to maximize its revenue and
114
+ minimize the service costs jointly by delivering on-demand
115
+ sensing services. In contrast, the user’s goal is to choose
116
+ a service that maximizes its utility. Therefore, there is a
117
+ need to design efficient contracting strategies between the SP
118
+ and the users so that sensing technologies can be effectively
119
+ monetized. In the proposed contract design framework, the SP
120
+ needs to design a menu of contracts that specify the sensing
121
+ price and the QoS offered to each type of user. The optimal
122
+ contracts yield a matching between the available sensing data
123
+ and users in the IoT ecosystem that is suitable for both SP
124
+ and the users.
125
+ Due to the large-population feature of users in the massive
126
+ IoT [5], the SP may not be aware of the exact type of
127
+ each user and may only have high level information on
128
+ the distribution of user’s types (e.g., inferred from historical
129
+ demand data)2. Thus, the challenge of contract design lies
130
+ in the development of an incentive compatible and optimal
131
+ mechanism for the sensing SP to maximize its payoff by
132
+ serving IoT users inspite of the incomplete information. To
133
+ overcome this obstacle, we propose a market-based pricing
134
+ contract mechanism for the SaaS model that takes into account
135
+ incentive compatibility and individual rationality of the users.
136
+ Specifically, we consider a continuum of user types with a
137
+ generic probability distribution and design optimal contracts
138
+ leveraging the Pontryagin maximum principle [6].
139
+ Under a wide class of probability distributions of user’s
140
+ type, we obtain analytical expression of optimal contracts in
141
+ which the pricing scheme and the QoS mapping are mono-
142
+ tonically increasing with user’s types, creating a complete
143
+ sensing service market with all possible QoS levels. When the
144
+ probability density function of user’s type distribution has a
145
+ large or sudden decrease around some points, then nondis-
146
+ criminative pricing phenomenon occurs, which reduces the
147
+ diversity of service provisions to the IoT users. Specifically,
148
+ some users choose the same service contract in spite of their
149
+ heterogeneous types. In addition, nondiscriminative pricing for
150
+ all customers can occur when the user’s types are nested in
151
+ the lower regime. Hence, in this scenario, the SP should target
152
+ at the majority in the market to optimize the revenue. For
153
+ comparison, we study the optimal contracts under complete
154
+ information and characterize the solution differences.
155
+ We illustrate the optimal SaaS mechanism design principles
156
+ with an application to the UAV-enabled VR. Simulation results
157
+ show that the SP earns more profit by serving users with rela-
158
+ tively stringent service requirements (higher types). However,
159
+ since the users of lower types constitute most of the market,
160
+ the SP gains a large proportion of revenue from serving low
161
+ type users even though their unit benefit is smaller.
162
+ The main contributions of this paper are summarized as
163
+ follows:
164
+ 1) We propose a two-sided market-based SaaS contract
165
+ design for QoS driven data trading between the service
166
+ 1The user types can also be interpreted as the importance of tasks to the
167
+ users respectively, ranging from non mission-critical to mission-critical ones.
168
+ 2This asymmetric information assumption also aligns with the fact that the
169
+ users aim to preserve privacy of their true types.
170
+ provider and users in the IoT ecosystem under asym-
171
+ metric information.
172
+ 2) We characterize the solutions of optimal contracts for
173
+ arbitrary distributions of user types, that yield the best
174
+ matching between the sensing services and the users
175
+ leveraging the Pontryagin maximum principle.
176
+ 3) We show that under the efficient data pricing mecha-
177
+ nism, the optimal contracts either capture the diversity
178
+ of user types (discriminative pricing) or focus on the
179
+ majority of user types (nondiscriminative pricing) de-
180
+ pending on the users’ preferences.
181
+ 4) We provide an illustrative example of UAV-enabled VR
182
+ application to validate and test our proposed contract
183
+ design. We further provide a comparison between the
184
+ hidden and full information scenarios in terms of the
185
+ payoff of the SP.
186
+ A. Related Work
187
+ Contract design [7] has typically been used in operations
188
+ research with applications to retail, financial markets [8],
189
+ insurances [9], supply chains [10], etc. With the emergence
190
+ of IoT and the data markets [11], new service models such
191
+ as the SaaS are being developed enabling new possibilities
192
+ such as resource trading [12], [13], opportunistic IoT [14],
193
+ task offloading and outsourcing [15], and performance oriented
194
+ resource provisioning [16], [17]. Therefore, there is a need for
195
+ developing effective contracts [18] and pricing schemes [19],
196
+ [20] that incentivize the interactions between users and service
197
+ providers of data in the IoT ecosystem. The data markets
198
+ and contract solutions can be implemented using blockchain
199
+ infrastructure over IoT networks [21]–[23].
200
+ A variety of literature is available on using contract theory
201
+ for incentive mechanism design in wireless communication
202
+ systems [24], tailored for scenarios such as traffic offloading
203
+ [25], [26], relay selection [27], spectrum trading [28], [29],
204
+ etc. In [30], the authors have studied the resource trading pro-
205
+ cess between a mobile virtual wireless network operator and
206
+ infrastructure providers using a contract. Similar approaches
207
+ have also been used to facilitate Wi-Fi sharing in crowdsourced
208
+ wireless community networks [31]. Incentive mechanism de-
209
+ sign has also been received a lot of attention in the next-
210
+ generation crowdsensing applications. For example, a two-
211
+ stage Stackelberg game approach has been proposed in [32]
212
+ to design incentive mechanism for the crowdsensing service
213
+ provider by capturing the participation level of the mobile
214
+ users. In [33], the authors have investigated the sequential dy-
215
+ namic pricing scheme of a monopoly mobile network operator
216
+ in the social data market by considering the congestion effects
217
+ in wireless networks. In [34], a distributed computing approach
218
+ is used in crowdsourcing using contracts by focusing on
219
+ designing a reward-based collaboration mechanism. Contract
220
+ theory is also leveraged to price the sponsored content in mo-
221
+ bile service [35], where the authors developed a hierarchical
222
+ game framework to capture the service relationships between
223
+ the network operator acting as the leader and the content
224
+ provider and the end users acting as followers.
225
+ Our work focuses on establishing a sensing data trading
226
+ platform enabled by the IoT by considering the user’s ratio-
227
+
228
+ nality and market reputation in a holistic manner. Different
229
+ from [36] where the authors have focused on designing a
230
+ pricing mechanism for data delivery in massive IoT from
231
+ a routing perspective, we address the data pricing problem
232
+ based on a contract-theoretic approach. Regarding SaaS in
233
+ the IoT, [37] has established a public sensing framework
234
+ for service-based applications in smart cities where the data
235
+ is provided by a cloud platform. The authors in [38] have
236
+ investigated smart phone-based crowdsensing to enhance the
237
+ public safety via the collected sensing data. In this paper, we
238
+ use an analytical approach to create an implementable policy
239
+ framework, focusing on a large-population regime through
240
+ contract design, which facilitates the realisation of the SaaS
241
+ paradigm.
242
+ We highlight several differences of this work with the
243
+ literature that uses contracts in various service provisioning
244
+ applications related to IoT and (wireless) communications.
245
+ Different from the majority of works (e.g., [25], [26], [28]–
246
+ [35]) that have considered finite number of user ‘types’ in
247
+ the contract formulation, our framework focuses on a large-
248
+ population regime of IoT users and uses a density function
249
+ to describe the heterogeneous types of users. The second
250
+ difference is that our framework considers the reputation of
251
+ service provisioning through an average QoS constraint. This
252
+ constraint implicitly improves the inclusion of distinct types
253
+ of users in the service market. The third difference is on the
254
+ solution approach used. Instead of solving the problem from an
255
+ classical optimization angle, this work addresses the problem
256
+ from an optimal control perspective.
257
+ B. Organization of the Paper
258
+ The rest of the paper is organized as follows. Section II
259
+ introduces the SaaS framework and formulates the contracting
260
+ problem. Contract analysis under a class of user’s type distri-
261
+ butions is presented in Section III. We provide the detailed
262
+ optimal contract solutions for two special cases in Section
263
+ IV. Section V investigates the contract design under complete
264
+ information. Extensions of the contract design to general user’s
265
+ type distributions are presented in Section VI. Section VII
266
+ illustrates the obtained results with an application to UAV-
267
+ based VR, and Section VIII concludes the paper.
268
+ II. SYSTEM MODEL AND PROBLEM FORMULATION
269
+ We consider a pool of IoT users with varying QoS require-
270
+ ments, that are connected to an SP for obtaining sensing data.
271
+ We assume that the SP has similar sensing data available in
272
+ a variety of different quality levels. For instance, the same
273
+ video data can be available in many different pixel resolutions.
274
+ Each user obtains a particular quality of data from the SP for
275
+ its specific mission needs. Depending on the application and
276
+ quality of data required, the IoT users can be characterized by
277
+ their ‘type’, denoted by δ. In the following subsections, we
278
+ provide a description of the different model parameters and
279
+ an analytical formulation of the optimal contract between an
280
+ SP and IoT users.
281
+
282
+ 55%
283
+ 28%
284
+ 12%
285
+ 3%
286
+ 1%
287
+ 0%
288
+ 10%
289
+ 20%
290
+ 30%
291
+ 40%
292
+ 50%
293
+ 60%
294
+ Less than $250
295
+ $250 to $400
296
+ $400 to $600
297
+ $600 to $1000
298
+ More than $1000
299
+ Percentage of customers
300
+ Spending preferences
301
+ Types of Customers in VR
302
+ Data Points
303
+ Exponential Fitting
304
+ Fig. 2. Customers’ spending preferences in the VR headset. Note that a higher
305
+ price of VR equipment can be interpreted as the customer preferring higher
306
+ quality of VR experiences. Then, the data yields an empirically exponential
307
+ distribution of the customers’ types in our contract design for VR services.
308
+ A. User Type and Data Quality
309
+ Considering a large number of users in a massive IoT
310
+ setting, each user is characterized by its type δ ∈ ∆ := [δ, ¯δ],
311
+ which is hidden to the SP, where δ ≥ 0 and ¯δ ≥ 0 denote the
312
+ lower and upper bounds of the parameter, respectively. Here, δ
313
+ signifies the importance level of user’s task depending on the
314
+ application needs. Furthermore, considering a large number of
315
+ possible user types, we assume a continuum of δ admitting a
316
+ value from the set ∆. The incomplete information of the IoT
317
+ users to SP implies that the SP does not know the individual
318
+ attributes of the users. However, the SP may have a broad
319
+ understanding of the probability distribution of the users. This
320
+ preserves user’s privacy to a certain degree. Hence, instead
321
+ of knowing the explicit information of δ, we assume that the
322
+ sensing SP has knowledge only about the probability density
323
+ function of the users’ type, denoted by f(δ).
324
+ Example 1. Empirical Estimation of User Type Distribution
325
+ To design practical contracts for VR services in the IoT, we
326
+ plot the data of customers’ spending preferences on the VR
327
+ equipment in Fig. 2. The data is adapted from [39]. Since a
328
+ higher price of VR equipment generally yields a better quality
329
+ of VR experience, the data can be used to approximate the
330
+ distribution of customers’ types in our VR contract design.
331
+ Fig. 2 depicts five levels of customers’ types. Without loss of
332
+ generality, we can consider their types as type 0, type 1, type
333
+ 2, type 3, and type 4, respectively, from left to right. In the
334
+ proposed SaaS framework, we consider an on-demand sensing
335
+ service provision in a large-population regime. Thus, the cus-
336
+ tomer’s type parameter is continuous over a bounded support.
337
+ Motivated by Fig. 2, we consider the type parameter δ taking
338
+ a value from the interval ∆ := [δ, ¯δ], where δ = 0, ¯δ = 4,
339
+ and a larger δ indicates a higher requirement of VR data
340
+ quality. This modeling is consistent with the statistics shown in
341
+ Fig. 2. Furthermore, based on Fig. 2, δ empirically admits an
342
+ exponential distribution. Using statistical inference techniques,
343
+ we can obtain f(δ) = 0.952e−0.952δ, and F(δ) = 1−e−0.952δ.
344
+ Note that these probability density and distribution functions
345
+ are aligned with the market data in Fig. 2.
346
+
347
+ The sensed data available to the SP is characterized by its
348
+ QoS level, denoted by q ∈ R, and the corresponding price
349
+ (payment by the user), denoted by p ∈ R. The QoS level can
350
+ be related to a number of specific metrics, such as the pixel
351
+ density, latency, and jitter in the transmission of sensing data,
352
+ etc. Note that we consider a continuum of quality levels since
353
+ a large number of different versions of data are assumed to be
354
+ available to the SP. In general, we can consider a vectorized q,
355
+ where each element denotes the quality of the corresponding
356
+ metric. The set Q denotes the available QoS levels provided
357
+ by the SP.
358
+ B. QoS Provisioning and Profit of SP
359
+ The service relationships between users and SP described
360
+ above can be naturally captured by a contract-theoretic frame-
361
+ work. Specifically, due to the asymmetric information induced
362
+ by users’ hidden type, the SP needs to design a menu of
363
+ contracts, i.e., {q(δ), p(δ)} and present it to the users. Each
364
+ user will then choose one contract that maximizes its payoff.
365
+ The payoff of the user with type δ, which claims to be of type
366
+ δ′ (thus receiving contact {q(δ′), p(δ′)}) can be computed as
367
+ V (δ, δ′) = Φ (δ, q(δ′)) − p(δ′),
368
+ (1)
369
+ where V : ∆ × ∆ → R, and Φ : ∆ × Q → R. Note that the
370
+ function Φ is a measure of the utility of the user. A natural
371
+ assumption of Φ is described as follows:
372
+ Assumption 1. The function Φ is continuously differentiable
373
+ and increasing in variables δ and q, i.e., ∂Φ(δ,q(δ))
374
+ ∂δ
375
+ > 0 and
376
+ ∂Φ(δ,q(δ))
377
+ ∂q(δ)
378
+ > 0. Also, it satisfies ∂Φ2(δ,q(δ))
379
+ ∂q(δ)∂δ
380
+ > 0.
381
+ Assumption 1 indicates that with a better QoS level, the
382
+ payoff of user increases. Also, for a given QoS level, the
383
+ users with a larger type parameter δ have a higher payoff
384
+ since their tasks are more mission-critical. Furthermore, for a
385
+ same amount enhancement of QoS level, the resulting payoff
386
+ increases for higher type users exceeds the one associated with
387
+ lower types users.
388
+ The function describing the SP’s profit obtained by provid-
389
+ ing a QoS level q to a user of type δ, is defined as
390
+ U(δ) = p(δ) − C(q(δ)),
391
+ (2)
392
+ where C : Q → R+ is the cost of the SP for providing the
393
+ sensor data. Then, the expected total payoff of the SP can be
394
+ expressed as
395
+ � ¯δ
396
+ δ (p(δ) − C(q(δ)))f(δ)dδ, where f(δ) denotes
397
+ the density of type δ users. We consider that f(δ) is strictly
398
+ greater than 0, i.e., f(δ) > 0, ∀δ ∈ [δ, ¯δ], which holds in the
399
+ case of massive IoT.
400
+ C. Profit Maximizing Contract Problem
401
+ Based on the direct revelation principle [40], it is sufficient
402
+ for the SP to design/consider contracts in which the users can
403
+ truthfully select the one that is consistent with their true types;
404
+ in other words, the users will reveal their types in the selection
405
+ and do not have incentives to misrepresent their true types.
406
+ Hence, we characterize the incentive compatibility (IC) and
407
+ individual rationality (IR) constraints of the users defined as
408
+ follows.
409
+ Definition 1 (Incentive Compatibility). A menu of contracts
410
+ {q(δ), p(δ)}, ∀δ, designed by the sensing SP is incentive
411
+ compatible if the user of type δ selects the contract (q(δ), p(δ))
412
+ that maximizes its payoff, i.e.,
413
+ Φ(δ, q(δ)) − p(δ) ≥ Φ(δ, q(δ′)) − p(δ′), ∀(δ, δ′) ∈ ∆2. (3)
414
+ Definition 2 (Individual Rationality). The individual rational-
415
+ ity constraint of each user is captured by
416
+ Φ(δ, q(δ)) − p(δ) ≥ 0, ∀δ ∈ ∆.
417
+ (4)
418
+ To investigate the impact of average sensing QoS level on
419
+ the optimal contracts, the SP has an additional constraint on
420
+ the provided QoS to the IoT users as follows:
421
+ � ¯δ
422
+ δ
423
+ q(δ)f(δ)dδ ≥ q,
424
+ (5)
425
+ where the positive constant q is the mean/average QoS. The
426
+ constraint (5) can be interpreted as the reputation that the
427
+ SP aims to build in the sensing service market. Note that
428
+ the average QoS has been leveraged to guide the optimal
429
+ decision-making in various applications in literature, such as
430
+ bandwidth allocation in broadband service provisioning [41],
431
+ admission control of services in edge computing [42], and
432
+ transmit powers minimization in small cell base stations [43].
433
+ The goal of the SP is to jointly determine the pricing scheme
434
+ p(δ) and the corresponding service quality q(δ) that yields
435
+ the best return. To this end, the SP is required to solve the
436
+ following optimization problem:
437
+ (OP) :
438
+ max
439
+ {q(δ),p(δ)}
440
+ � ¯δ
441
+ δ
442
+
443
+ p(δ) − C(q(δ))
444
+
445
+ f(δ)dδ
446
+ s.t. Φ(δ, q(δ)) − p(δ) ≥ Φ(δ, q(δ′)) − p(δ′),
447
+ ∀(δ, δ′) ∈ ∆2, (IC)
448
+ Φ(δ, q(δ)) − p(δ) ≥ 0, ∀δ ∈ ∆, (IR)
449
+ � ¯δ
450
+ δ
451
+ q(δ)f(δ)dδ ≥ q. (Reputation)
452
+ III. ANALYSIS AND DESIGN OF OPTIMAL CONTRACTS
453
+ In this section, we first analyze the formulated problem (OP)
454
+ in Section II. Then we design the optimal contracts for SaaS
455
+ by using Pontryagin maximum principle [44].
456
+ A. Problem Analysis
457
+ To solve problem (OP), one challenge lies in the infinite
458
+ number of IC and IR constraints in (3) and (4), respectively.
459
+ To simplify (OP), we first present the following lemma.
460
+ Lemma 1. Under the condition that p(δ) and q(δ) are
461
+ differentiable, the set of IC constraints (3) is equivalent to
462
+ the local incentive constraints
463
+ dp(δ)
464
+
465
+ = ∂Φ(δ, q(δ))
466
+ ∂q(δ)
467
+ dq(δ)
468
+ dδ , ∀δ ∈ ∆,
469
+ (6)
470
+
471
+ and a monotonicity constraint
472
+ dq(δ)
473
+
474
+ ≥ 0.
475
+ (7)
476
+ Proof. See Appendix A.
477
+
478
+ To facilitate the optimal contract design in Section III-B, we
479
+ specify some structures of the payoff, Φ and cost, C. The user
480
+ with mission-critical tasks (higher δ) gains more by receiving
481
+ better quality of sensor data (higher q). Therefore, a reasonable
482
+ payoff function for type δ user can be chosen as follows.
483
+ Assumption 2. The payoff function for type δ user is consid-
484
+ ered as
485
+ Φ(δ, q(δ)) = δq(δ).
486
+ (8)
487
+ Then, (6) can be simplified as
488
+ dp(δ)
489
+
490
+ = δ dq(δ)
491
+ dδ . Note that
492
+ the payoff function (8) is not necessary linear. The only
493
+ requirement of Φ is to satisfy Assumption 1. The analysis
494
+ of optimal contract design in the following still holds for a
495
+ general Φ. Similarly, to obtain analytical results of optimal
496
+ contracts, one of the cost functions of sensing SP is chosen
497
+ as follows.
498
+ Assumption 3. The cost function of sensing SP is
499
+ C(q(δ)) = σ (exp(aq(δ)) − 1) ,
500
+ (9)
501
+ where σ > 0 is a normalizing constant trading off between
502
+ the sensing service costs and the revenue, and a > 0 is a
503
+ sensitivity constant, indicating that the marginal cost of the
504
+ sensor data is increasing with its quality.
505
+ Corollary 1. Based on Lemma 1 and (8), the IC constraints
506
+ in (OP) can be represented as
507
+ dV
508
+ dδ = q(δ),
509
+ (10)
510
+ together with the monotonicity constraint (7).
511
+ Proof. Based on (1), we have dV
512
+ dδ = dΦ
513
+ dδ − dp(δ)
514
+ dδ . Then, using
515
+
516
+ dδ = δ dq(δ)
517
+
518
+ + q(δ) and dp(δ)
519
+
520
+ = δ dq(δ)
521
+
522
+ yield the result.
523
+
524
+ Note that Corollary 1 indicates that the payoff of the user
525
+ is monotonically increasing with the type δ. Therefore, the IR
526
+ constraint can be simpli��ed as Φ(δ, q(δ)) − p(δ) ≥ 0. Indeed,
527
+ the IR constraint is binding under the optimal contracts for
528
+ type δ users, i.e.,
529
+ Φ(δ, q(δ)) − p(δ) = 0.
530
+ (11)
531
+ Otherwise, the sensing SP can earn more profits by increasing
532
+ the price p(δ) for serving the type δ users.
533
+ The reputation constraint (5) essentially divides the problem
534
+ analysis in two regimes: whether the constraint is binding
535
+ at the optimal solution or not. Denote by {p∗(δ), q∗(δ)} the
536
+ optimal solution to (OP). When q is relatively large, then it is
537
+ possible that
538
+ � ¯δ
539
+ δ q∗(δ)f(δ)dδ = q, since the SP has no incen-
540
+ tive to provide a QoS q(δ) with q(δ) > q∗(δ) which decreases
541
+ the objective value. The inequality
542
+ � ¯δ
543
+ δ q∗(δ)f(δ)dδ > q could
544
+ happen when q is relatively small. Thus, there exists a thresh-
545
+ old of q above which (5) is binding and below which is non-
546
+ binding at the optimum. In the case of
547
+ � ¯δ
548
+ δ q∗(δ)f(δ)dδ > q
549
+ which indicates that (5) is inactive, the optimal solution
550
+ {p∗(δ), q∗(δ)} to (OP) will be the same as the one to (OP)
551
+ without considering the reputation constraint. To this end, we
552
+ have the following approach to address (OP) for given q. First,
553
+ we solve (OP) without considering the reputation constraint.
554
+ If the obtained solution satisfies the reputation constraint, then
555
+ it is optimal to (OP). Otherwise, we replace the constraint (5)
556
+ in (OP) by
557
+ � ¯δ
558
+ δ
559
+ q(δ)f(δ)dδ = q,
560
+ (12)
561
+ as the reputation constraint holds as an equality at the op-
562
+ timal contract design in this regime. Solving (OP) without
563
+ incorporating the reputation constraint is a classical optimal
564
+ contract design problem. In this work, we focus on developing
565
+ a systematic approach to address the second case where (12)
566
+ is considered in the constraints.
567
+ Remark: The reputation constraint implicitly penalizes the
568
+ SP for not serving IoT users with low valuation (the users of
569
+ lower types). The reason is that not serving users is equivalent
570
+ to providing zero quality service with zero cost which indeed
571
+ decreases the average QoS. This is different from the setup
572
+ in the classical contract design in which the SP only serves
573
+ the consumers with positive valuations (e.g., based on the
574
+ metric called virtual valuation δ − 1−F (δ)
575
+ f(δ) ). Our model aims
576
+ to serve all users including those of low types that might not
577
+ contribute to the SP’s profit. Hence, the proposed framework
578
+ with the reputation constraint has the capability to enhance the
579
+ accessibility and affordability of the service to all users.
580
+ B. Optimal Contract Solution
581
+ Based on (1), we obtain p(δ) = Φ(δ, q(δ)) − V (δ), where
582
+ we suppress the notations q and p in V . This is reasonable
583
+ since the payoff of an IoT user depends on its type when the
584
+ contract is designed. Then, by regarding V (δ) as a decision
585
+ variable instead of p(δ), the problem can be rewritten as
586
+ (OP′) :
587
+ max
588
+ {q(δ),V (δ)}
589
+ � ¯δ
590
+ δ
591
+
592
+ Φ(δ, q(δ)) − V (δ) − C(q(δ))
593
+
594
+ f(δ)dδ
595
+ s.t. dV
596
+ dδ = q(δ), dq(δ)
597
+
598
+ ≥ 0, V (δ) = 0,
599
+ � ¯δ
600
+ δ
601
+ q(δ)f(δ)dδ = q.
602
+ (OP′) can be regarded as an optimal control problem.
603
+ Specifically, by following the notations in control theory, we
604
+ denote u(δ) = q(δ) by the control variable and x1(δ) = V (δ)
605
+ by the state variable. Then, we obtain ˙x1 = u(δ) with the
606
+ initial value x1(δ) = 0. The control input admits an increasing
607
+ property with the type parameter δ, i.e., ˙u(δ) ≥ 0.
608
+ The remaining difficulty in solving (OP′) lies in the reputa-
609
+ tion constraint. To facilitate the design of the optimal control
610
+ strategy, we introduce a new state variable x2(δ) satisfying
611
+ ˙x2(δ) = u(δ)f(δ). Therefore, the reputation constraint can be
612
+ replaced by
613
+ ˙x2(δ) = u(δ)f(δ),
614
+ (13)
615
+
616
+ with boundary values: x2(¯δ) = q and x2(δ) = 0.
617
+ For clarity, we present the problem (OP′) with new nota-
618
+ tions as follows:
619
+ (OP′′) :
620
+ max
621
+ {u(δ),x(δ)}
622
+ � ¯δ
623
+ δ
624
+
625
+ Φ(δ, u(δ)) − x1(δ) − C(u(δ))
626
+
627
+ f(δ)dδ
628
+ s.t.
629
+ ˙x1(δ) = u(δ), x1(δ) = 0,
630
+ ˙x2(δ) = u(δ)f(δ), x2(¯δ) = q, x2(δ) = 0,
631
+ ˙u(δ) ≥ 0.
632
+ where x = [x1, x2]T . Next, by defining λ = [λ1, λ2]T , the
633
+ Hamiltonian of (OP′′) can be expressed as
634
+ H(x(δ), u(δ), λ(δ), δ) =
635
+
636
+ Φ(δ, u(δ)) − x1(δ)
637
+ − C(u(δ))
638
+
639
+ f(δ) + λ1(δ)u(δ) + λ2(δ)u(δ)f(δ),
640
+ (14)
641
+ where λ1 and λ2 are costate variables corresponding to (10)
642
+ and (13), respectively.
643
+ By using the Pontryagin maximum principle [44], we can
644
+ obtain the optimal solution (x∗(δ), u∗(δ)) by solving the
645
+ following Hamilton system:
646
+ H(x∗(δ), u∗(δ), λ∗(δ), δ) ≥ H(x∗(δ), u(δ), λ∗(δ), δ), (15)
647
+ ˙x∗
648
+ 1 = ∂H(x∗(δ), u∗(δ), λ∗(δ), δ)
649
+ ∂λ1(δ)
650
+ = u∗(δ),
651
+ (16)
652
+ ˙x∗
653
+ 2 = ∂H(x∗(δ), u∗(δ), λ∗(δ), δ)
654
+ ∂λ2(δ)
655
+ = u∗(δ)f(δ),
656
+ (17)
657
+ ˙λ∗
658
+ 1 = −∂H(x∗(δ), u∗(δ), λ∗(δ), δ)
659
+ ∂x1(δ)
660
+ = f(δ),
661
+ (18)
662
+ ˙λ∗
663
+ 2 = −∂H(x∗(δ), u∗(δ), λ∗(δ), δ)
664
+ ∂x2(δ)
665
+ = 0,
666
+ (19)
667
+ λ1(¯δ) = 0,
668
+ (20)
669
+ λ2(¯δ) is a constant.
670
+ (21)
671
+ Note that (20) and (21) are boundary conditions. Specifically,
672
+ the initial state of x1 is fixed, and we only have freedom in
673
+ specifying boundary condition at the terminal time. Then, the
674
+ corresponding costate variable λ1 at the time ¯δ should equal to
675
+ the derivative of the terminal payoff with respect to the state
676
+ x1 at ¯δ based on the maximum principle. Since the objective
677
+ function in (OP′′) does not include the terminal payoff, then
678
+ we obtain λ1(¯δ) = 0. Similarly, the initial and terminal states
679
+ of x2 are fixed, and we can specify the boundary condition
680
+ (21) from (19) in which λ2 admits a constant value.
681
+ Furthermore, (15) ensures the optimality of control u∗(δ).
682
+ Thus, using the first-order condition, (15) can be simplified as
683
+ ∂H(x∗(δ), u(δ), λ∗(δ), δ)
684
+ ∂u(δ)
685
+ =
686
+ �∂Φ(δ, u(δ))
687
+ ∂u(δ)
688
+ − dC(u(δ))
689
+ du(δ)
690
+
691
+ ·f(δ) + λ∗
692
+ 1(δ) + λ∗
693
+ 2(δ)f(δ) = 0.
694
+ (22)
695
+ In addition, (18) and (20) indicate that
696
+ λ∗
697
+ 1(δ) = F(δ) − 1.
698
+ (23)
699
+ Note that the end-point of x2 is fixed, and hence λ2(¯δ) needs
700
+ to be determined rather than simply being 0. Based on (19),
701
+ we obtain
702
+ λ∗
703
+ 2(δ) = β, ∀δ ∈ ∆,
704
+ (24)
705
+ where β is a constant to be determined.
706
+ We have obtained the optimal solutions for λ∗
707
+ 1(δ) and λ∗
708
+ 2(δ).
709
+ To design the optimal u∗(δ), we next focus on the optimality
710
+ condition (22). The distribution of user’s type can be general,
711
+ e.g., normal, exponential, or learnt from the historical data.
712
+ We first solve (OP′′) without considering the monotonicity
713
+ constraint ˙u(δ) ≥ 0. Then, the obtained control uco(δ) from
714
+ (22) is a candidate optimal solution. By plugging (23) and
715
+ (24) into (22) and using the defined functions (8) and (9), we
716
+ obtain dC(uco(δ))
717
+ du(δ)
718
+ − ∂Φ(δ,u(δ))
719
+ ∂u(δ)
720
+ = F (δ)−1
721
+ f(δ)
722
+ + β which leads to
723
+ uco(δ) = 1
724
+ a ln
725
+ � 1
726
+
727
+ �F(δ) − 1
728
+ f(δ)
729
+ + δ + β
730
+ ��
731
+ .
732
+ (25)
733
+ The second-order condition gives
734
+ ∂2H(x∗(δ),u(δ),λ∗(δ),δ)
735
+ ∂u(δ)2
736
+ =
737
+ −σa2eau(δ)f(δ) < 0, and hence uco(δ) is a maximizer
738
+ of the Hamiltonian. The maximum principle is a necessary
739
+ condition for the optimal solution of (OP′′). Then, we further
740
+ check the sufficient condition for optimality on the maximized
741
+ Hamiltonian. Specifically, by verifying that the Hamiltonian
742
+ H(x(δ), u(δ), λ(δ), δ) is concave in both x and u, the so-
743
+ lution uco(δ) is optimal to (OP′′) without considering the
744
+ monotonicity constraint. Indeed, based on the Mangasarian
745
+ sufficiency theorem [45], a stronger conclusion is that the
746
+ obtained control uco(δ) is the unique optimal solution as the
747
+ Hamiltonian is strictly concave in u. Based on the dynamics
748
+ in (OP′′), the optimal state trajectory is also unique.
749
+ We next verify whether uco(δ) satisfying the monotonicity
750
+ constraint ˙u(δ) ≥ 0. In (25), the CDF F(δ) is increasing with
751
+ δ, but the presence of f(δ) makes the monotonicity of u(δ)
752
+ unclear. We present the following lemma which can be proved
753
+ using optimality condition to (25).
754
+ Lemma 2. If 2f 2(δ)+(1−F(δ))f ′(δ) > 0, then the obtained
755
+ solution uco(δ) is optimal. In addition, a decreasing 1−F (δ)
756
+ f(δ)
757
+ leads to an optimal uco(δ).
758
+ Proof. We
759
+ need
760
+ to
761
+ ensure
762
+ that
763
+ (25)
764
+ is
765
+ increasing
766
+ with
767
+ δ.
768
+ The
769
+ first-order
770
+ condition
771
+ of
772
+ (25)
773
+ gives
774
+
775
+ F (δ)−1
776
+ f(δ)
777
+ + δ + β
778
+ �−1 �
779
+ f 2(δ)−(F (δ)−1)f ′(δ)
780
+ f 2(δ)
781
+ + 1
782
+
783
+ >
784
+ 0.
785
+ In the first part, β is a constant determined based on
786
+ � ¯δ
787
+ δ uco(δ)f(δ)dδ =
788
+ � ¯δ
789
+ δ
790
+ 1
791
+ a ln( 1
792
+ aσ[ F (δ)−1
793
+ f(δ)
794
+ + δ + β])f(δ)dδ = q.
795
+ The integrand should be well-defined to make the equation
796
+ satisfied.
797
+ The
798
+ existence
799
+ of
800
+ such
801
+ β
802
+ is
803
+ guaranteed
804
+ as
805
+ � ¯δ
806
+ δ ln( 1
807
+ aσ[ F (δ)−1
808
+ f(δ)
809
+ + δ + β])f(δ)dδ is monotonically increasing
810
+ in β. Thus, F (δ)−1
811
+ f(δ) +δ+β > 0 which is ensured by the choice
812
+ of β. Then, we need to have
813
+ f 2(δ)−(F (δ)−1)f ′(δ)
814
+ f 2(δ)
815
+ + 1 > 0
816
+ which gives the result. We can also verify that if 1−F (δ)
817
+ f(δ)
818
+ is
819
+ decreasing in δ, then 2f 2(δ) + (1 − F(δ))f ′(δ) > 0 holds
820
+ which yields the result.
821
+
822
+ Remark: The distributions of IoT user’s type satisfying the
823
+ condition in Lemma 2 are quite general, including the uni-
824
+ form, normal and exponential ones. Note that the distributions
825
+ without a large and sudden decrease in the probability density
826
+ function (PDF) f(δ) generally satisfy the condition in Lemma
827
+ 2, and hence (25) gives the optimal solution.
828
+
829
+ Back to (24), the constant β can be obtained by solving
830
+ the reputation constraint (12), i.e.,
831
+ � ¯δ
832
+ δ
833
+ 1
834
+ a ln( 1
835
+ aσ[ F (δ)−1
836
+ f(δ)
837
+ + δ +
838
+ β])f(δ)dδ = q. The expression u∗(δ) characterizes the pro-
839
+ vided sensing QoS in terms of the user’s type. We next focus
840
+ on obtaining the pricing scheme of the sensing services. To
841
+ this end, the expression of x∗
842
+ 1(δ) becomes critical. Based on
843
+ (16), we obtain
844
+ ˙x∗
845
+ 1(δ) = 1
846
+ a ln
847
+ � 1
848
+
849
+ �F(δ) − 1
850
+ f(δ)
851
+ + δ + β
852
+ ��
853
+ .
854
+ (26)
855
+ Then, x∗
856
+ 1(δ) can be determined by (26) and x∗
857
+ 1(δ) = 0.
858
+ The following Theorem 1 explicitly characterizes the optimal
859
+ contracts in the considered scenario.
860
+ Theorem 1. Under the condition 2f 2(δ)+(1−F(δ))f ′(δ) > 0
861
+ in Lemma 2, the optimal contracts {q∗(δ), p∗(δ)} designed by
862
+ the SP are as follows:
863
+ q∗(δ) = 1
864
+ a ln
865
+ � 1
866
+
867
+ �F(δ) − 1
868
+ f(δ)
869
+ + δ + β
870
+ ��
871
+ ,
872
+ p∗(δ) = Φ(δ, q∗(δ)) − φ(δ) = δq∗(δ) − φ(δ),
873
+ (27)
874
+ where β is determined from
875
+ � ¯δ
876
+ δ q∗(δ)f(δ)dδ = q, and ˙φ(δ) :=
877
+ 1
878
+ a ln( 1
879
+ aσ[ F (δ)−1
880
+ f(δ)
881
+ + δ + β]) with φ(δ) = 0.
882
+ Structure of the optimal contracts: The q∗(δ) in (27) can be
883
+ naturally decomposed into three parts, and each one includes
884
+ a term F (δ)−1
885
+ f(δ) , δ, and β, corresponding to the incentives of
886
+ IoT users, the utility of SP, and the reputation of service
887
+ provision, respectively. Recall that λ∗
888
+ 1(δ) = F(δ) − 1. Thus,
889
+ the first term quantifying the impact of IC constraint on
890
+ q∗(δ) captures the statistics of the IoT user types. The second
891
+ term including δ arises from the maximization of objective
892
+ function of SP which yields him the largest revenue. The third
893
+ constant term β indicates that the sensitivity of reputation
894
+ constraint is the same for every type of users. This finding
895
+ is consistent with the fact that the reputation constraint takes
896
+ the aggregated service provision over all users into account,
897
+ i.e., the mean QoS. The service pricing function p∗(δ) is
898
+ characterized based on q∗(δ) through relation (1) and hence
899
+ has a similar decomposition interpretation as q∗(δ). In sum,
900
+ the structure of optimal contracts in Theorem 1 incorporates
901
+ a service payoff maximization term and two adjusting terms
902
+ for user incentives.
903
+ IV. ANALYTICAL RESULTS OF SPECIAL CASES
904
+ In this section, we present analytical results of optimal
905
+ contracts for two typical distributions of the user’s type.
906
+ A.
907
+ Uniform User Type Distribution
908
+ When δ is uniformly distributed, its PDF and cumulative
909
+ density function (CDF) admit the forms: f(δ) =
910
+ 1
911
+ ¯δ−δ and
912
+ F(δ) =
913
+ δ−δ
914
+ ¯δ−δ, δ ∈ ∆. Based on Theorem 1, the sensing
915
+ QoS function is q∗(δ) =
916
+ 1
917
+ a ln( 1
918
+ aσ[2δ − ¯δ + β]). Due to the
919
+ logarithm function, q∗(δ) is nonlinear with δ. In addition, the
920
+ marginal sensing QoS is decreasing with the IoT user’s type.
921
+ One reason is that increasing sensing QoS is harder in large q
922
+ regime than its counterpart for the SP. Further, the unknown
923
+ constant β in (24) can be solved from
924
+
925
+ β−¯δ
926
+ 2
927
+ + ¯δ
928
+
929
+ ln(β + ¯δ)−
930
+
931
+ β−¯δ
932
+ 2
933
+ + δ
934
+
935
+ ln(β − ¯δ + 2δ) = (aq + ln(aσ) + 1)(¯δ − δ).
936
+ The optimal sensing pricing scheme in the contract is
937
+ characterized in the following corollary.
938
+ Corollary 2. Under the uniform distribution of the IoT user’s
939
+ type δ, the price of sensing service is equal to p∗(δ) =
940
+ δq∗(δ) +
941
+ δ−δ
942
+ a(¯δ−δ)(ln(aσ) + 1) −
943
+ 1
944
+ a(¯δ−δ)
945
+ � �
946
+ β−¯δ
947
+ 2
948
+ + δ
949
+
950
+ ln(β − ¯δ +
951
+ 2δ) −
952
+
953
+ β−¯δ
954
+ 2
955
+ + δ
956
+
957
+ ln(β − ¯δ + 2δ)
958
+
959
+ .
960
+ B. Exponential User Type Distribution
961
+ When the user’s type δ admits the exponential distribution,
962
+ then the number of IoT users with mission-critical tasks is less
963
+ than the ones with nonmission-critical tasks. Specifically, the
964
+ PDF and CDF of δ with rate ρ are equal to f(δ) = ρe−ρδ and
965
+ F(δ) = 1−e−ρδ, respectively. Then, the optimal sensing QoS
966
+ function has the form q∗(δ) = 1
967
+ a ln( 1
968
+ aσ[δ − 1
969
+ ρ + β]), where β
970
+ can be computed from
971
+ � ¯δ
972
+ δ ln( 1
973
+ aσ[δ − 1
974
+ ρ + β])ρe−ρδdδ = aq.
975
+ Similar to the uniform distribution scenario, we can char-
976
+ acterize the optimal pricing as follows.
977
+ Corollary 3. Under the exponential distribution of the user’s
978
+ type δ, the optimal pricing of the sensing service in the
979
+ contract is p∗(δ) = δq∗(δ) − 1
980
+ a(δ + β − 1
981
+ ρ) ln(δ + β −
982
+ 1
983
+ ρ) + δ
984
+ a (1 + ln(aσ)) − γ, where the constant γ is equal to
985
+ γ = δ
986
+ a(1 + ln(aσ)) − 1
987
+ a(δ + β − 1
988
+ ρ) ln(δ + β − 1
989
+ ρ).
990
+ We elaborate more on exponential distribution scenario in
991
+ case studies in Section VII. In other cases with more general
992
+ distributions of the IoT user’s type, we can directly apply
993
+ Theorem 1 to obtain the optimal SaaS contracts. However,
994
+ note that the support of f(δ) needs to be consistent with the
995
+ range of δ. Hence, if a normal distribution is used, it needs to
996
+ be truncated in order to be compatible with the framework.
997
+ V. COMPARISON TO THE BENCHMARK SCENARIO
998
+ Under the full information scenario, the sensing SP knows
999
+ the type of each IoT user. Thus, the IC constraint (3) becomes
1000
+ no longer necessary. Then, the optimal contract design prob-
1001
+ lem for SaaS becomes:
1002
+ (OP − B) :
1003
+ max
1004
+ {q(δ),V (δ)}
1005
+ � ¯δ
1006
+ δ
1007
+
1008
+ p(δ) − C(q(δ))
1009
+
1010
+ f(δ)dδ
1011
+ s.t. V (δ) ≥ 0, ∀δ,
1012
+ � ¯δ
1013
+ δ
1014
+ q(δ)f(δ)dδ = q.
1015
+ Next, we solve (OP − B) from an optimal control perspec-
1016
+ tive again, and the results are summarized in Theorem 2. For
1017
+ clarity, we denote by qb(δ), V b(δ) the optimal solutions to
1018
+ (OP − B). Further analysis indicates that V b(δ) = 0, ∀δ, and
1019
+ the pricing scheme is charaterized by pb(δ) = Φ(δ, qb(δ)).
1020
+ By regarding q(δ) as a control variable, i.e., q(δ) = u(δ),
1021
+ and introducing a state ˙x(δ) = u1(δ)f(δ) with boundary
1022
+
1023
+ constraints x(¯δ) = q and x(δ) = 0, we can reformulate
1024
+ (OP − B) as
1025
+ (OP − B′) : max
1026
+ {u(δ)}
1027
+ � ¯δ
1028
+ δ
1029
+
1030
+ Φ(δ, u(δ)) − C(u(δ))
1031
+
1032
+ f(δ)dδ
1033
+ s.t. ˙x(δ) = u(δ)f(δ), x(¯δ) = q, x(δ) = 0.
1034
+ Note that (OP − B′) is an optimal control problem with
1035
+ fixed initial and terminal state constraints. The Hamiltonian
1036
+ of (OP − B′) is
1037
+ H(x(δ), u(δ), λ(δ), δ) =
1038
+
1039
+ Φ(δ, u(δ)) − C(u(δ))
1040
+
1041
+ ·f(δ) + λ(δ)u(δ)f(δ),
1042
+ (28)
1043
+ where
1044
+ λ
1045
+ is
1046
+ the
1047
+ costate
1048
+ variable
1049
+ associated
1050
+ with
1051
+ the
1052
+ state
1053
+ dynamics.
1054
+ The
1055
+ maximum
1056
+ principle
1057
+ yields
1058
+ the
1059
+ following Hamilton system: H(xb(δ), ub(δ), λb(δ), δ)
1060
+
1061
+ H(xb(δ), u(δ), λb(δ), δ), ˙xb = ub(δ)f(δ), ˙λb = 0, λ(¯δ) = β,
1062
+ where β is a real constant.
1063
+ The
1064
+ first-order
1065
+ condition
1066
+ of
1067
+ (28)
1068
+ with
1069
+ respect
1070
+ to
1071
+ u
1072
+ is
1073
+ ∂H(xb(δ),u(δ),λb(δ),δ)
1074
+ ∂u
1075
+ =
1076
+ ( ∂Φ(δ,u(δ))
1077
+ ∂u
1078
+
1079
+ dC(u(δ))
1080
+ du
1081
+ )f(δ) +
1082
+ λb(δ)f(δ) = 0. Further, the second-order conditions of Hamil-
1083
+ tonian (28) with respective to x and u are nonpositive, and
1084
+ hence the obtained ub(δ) is optimal. Then, the optimal control
1085
+ ub(δ) satisfies (δ − aσeaub(δ) + β)f(δ) = 0, which further
1086
+ yields ub(δ) = 1
1087
+ a ln δ+β
1088
+ aσ . The constant β can be solved from
1089
+ � ¯δ
1090
+ δ ub(δ)f(δ)dδ = q.
1091
+ We summarize the optimal contract for SaaS under the
1092
+ complete information in the following theorem.
1093
+ Theorem 2. When the SP has the complete incentive infor-
1094
+ mation of the IoT users, the optimal contracts {qb(δ), pb(δ)}
1095
+ are designed as follows:
1096
+ qb(δ) = 1
1097
+ a ln
1098
+ �δ + β
1099
+
1100
+
1101
+ ,
1102
+ pb(δ) = Φ(δ, qb(δ)) = δqb(δ),
1103
+ (29)
1104
+ where β is determined from
1105
+ � ¯δ
1106
+ δ ln δ+β
1107
+ aσ f(δ)dδ = aq.
1108
+ Remark: Theorem 2 helps to identify the fundamental dif-
1109
+ ferences of optimal contracts designed under complete and in-
1110
+ complete information structures. Comparing with the designed
1111
+ optimal contracts {q∗(δ), p∗(δ)} in Theorem 1, the sensing
1112
+ QoS mapping qb(δ) and pricing function pb(δ) in Theorem 2
1113
+ do not contain terms F (δ)−1
1114
+ f(δ)
1115
+ ≤ 0 and φ(δ) ≥ 0, respectively.
1116
+ The different values of β in q∗(δ) and qb(δ) prohibit the
1117
+ conclusion that q∗(δ) ≤ qb(δ) and p∗(δ) ≤ pb(δ). Note that
1118
+ the constraint
1119
+ � ¯δ
1120
+ δ q(δ)f(δ)dδ = q indicates the same mean
1121
+ QoS in two scenarios without/with asymmetric information
1122
+ between SP and users. Therefore, when q∗(δ) ̸= qb(δ),
1123
+ ∀δ ∈ [δ, ¯δ], we can conclude that there exists at least one ˜δ
1124
+ where q∗(˜δ) = qb(˜δ), and the IoT users in the benchmark case
1125
+ pay more for the service due to φ(δ) ≥ 0, i.e., p∗(˜δ) < pb(˜δ).
1126
+ Another remark is that the total profit of sensing SP by
1127
+ providing the optimal contracts resulting from (OP − B) is
1128
+ no less than the one from (OP) due to the removal of IC
1129
+ constraint which enlarges the feasible decision space. The
1130
+ profit difference can be interpreted as the private user’s type
1131
+ information cost which we will quantify in Section VII.
1132
+ VI. OPTIMAL CONTRACTS FOR GENERAL USER’S TYPE
1133
+ DISTRIBUTIONS
1134
+ In this section, we investigate the scenarios when the density
1135
+ condition in Lemma 2 does not hold. We provide an alternative
1136
+ maximum principle and a full characterization of optimal
1137
+ contracts for SaaS in this general case.
1138
+ A. Maximum Principle and Optimality Analysis
1139
+ Following the notations in (OP′′) except replacing u with
1140
+ x3 and introducing a new control variable µ, we formulate the
1141
+ following problem:
1142
+ (OP − E) :
1143
+ max
1144
+ {µ(δ),x1(δ),
1145
+ x2(δ),x3(δ)}
1146
+ � ¯δ
1147
+ δ
1148
+
1149
+ Φ(δ, x3(δ)) − x1(δ) − C(x3(δ))
1150
+
1151
+ f(δ)dδ
1152
+ s.t.
1153
+ ˙x1(δ) = x3(δ), x1(δ) = 0,
1154
+ ˙x2(δ) = x3(δ)f(δ), x2(¯δ) = q, x2(δ) = 0,
1155
+ ˙x3(δ) = µ(δ), µ(δ) ≥ 0.
1156
+ Note that (OP − E) is an optimal control problem with
1157
+ three state variables x1, x2, x3 and a control variable µ, where
1158
+ the initial points of x1 and x2, and the boundary points of x2
1159
+ are fixed.
1160
+ The
1161
+ Hamiltonian
1162
+ of
1163
+ (OP − E)
1164
+ can
1165
+ be
1166
+ written
1167
+ as
1168
+ H(x(δ), µ(δ), λ(δ), δ) = [Φ(δ, x3(δ)) − x1(δ) − C(x3(δ))] ·
1169
+ f(δ) + λ1(δ)x3(δ) + λ2(δ)x3(δ)f(δ) + λ3(δ)µ(δ), where
1170
+ x = [x1, x2, x3]T and λ = [λ1, λ2, λ3]T . To differentiate
1171
+ with the optimal solution (x∗(δ), u∗(δ)) in Theorem 1, we
1172
+ denote by (xo(δ), µo(δ)) the optimal solution to the cases
1173
+ with general user’s type distribution. Using the Pontryagin
1174
+ maximum principle, we obtain (xo(δ), µo(δ)) by solving the
1175
+ Hamilton system:
1176
+ H(xo(δ), µo(δ), λo(δ), δ) ≥ H(xo(δ), µ(δ), λo(δ), δ), (30)
1177
+ ˙xo
1178
+ 1 = ∂H(xo(δ), µo(δ), λo(δ), δ)
1179
+ ∂λ1(δ)
1180
+ = xo
1181
+ 3(δ),
1182
+ (31)
1183
+ ˙xo
1184
+ 2 = ∂H(xo(δ), µo(δ), λo(δ), δ)
1185
+ ∂λ2(δ)
1186
+ = xo
1187
+ 3(δ)f(δ),
1188
+ (32)
1189
+ ˙xo
1190
+ 3 = ∂H(xo(δ), µo(δ), λo(δ), δ)
1191
+ ∂λ3(δ)
1192
+ = µo(δ),
1193
+ (33)
1194
+ ˙λo
1195
+ 1 = −∂H(xo(δ), µo(δ), λo(δ), δ)
1196
+ ∂x1(δ)
1197
+ = f(δ),
1198
+ (34)
1199
+ ˙λo
1200
+ 2 = −∂H(xo(δ), µo(δ), λo(δ), δ)
1201
+ ∂x2(δ)
1202
+ = 0,
1203
+ (35)
1204
+ ˙λo
1205
+ 3 = −���H(xo(δ), µo(δ), λo(δ), δ)
1206
+ ∂x3(δ)
1207
+ = −
1208
+ �∂Φ(δ, x3(δ))
1209
+ ∂x3(δ)
1210
+ − dC(x3(δ))
1211
+ dx3(δ)
1212
+
1213
+ f(δ)
1214
+ − λo
1215
+ 1(δ) − λo
1216
+ 2(δ)f(δ),
1217
+ (36)
1218
+ λ1(¯δ) = 0,
1219
+ (37)
1220
+ λ2(¯δ) is a constant,
1221
+ (38)
1222
+ λ3(δ) = λ3(¯δ) = 0.
1223
+ (39)
1224
+ Note that (37) and (38) are boundary conditions which are
1225
+ similar to the ones in (20) and (21). In (OP − E), we include
1226
+
1227
+ another state variable x3 which does not have initial and
1228
+ terminal constraints. Then, based on the maximum principle
1229
+ [44], the corresponding costate variable λ3 at time δ and
1230
+ ¯δ should equal to the derivative of the initial and terminal
1231
+ payoff with respect to the state x3, respectively. In (OP − E),
1232
+ the objective function does not contain individual initial and
1233
+ terminal utilities, and thus we obtain condition (39).
1234
+ First, similar to (23) and (24), we observe that
1235
+ λo
1236
+ 1(δ) = F(δ) − 1,
1237
+ (40)
1238
+ λo
1239
+ 2(δ) = β,
1240
+ (41)
1241
+ where the constant β can be determined using(12) after the
1242
+ QoS mapping qo(δ) is characterized.
1243
+ In addition, by integrating (36), we obtain
1244
+ λo
1245
+ 3(δ) = −
1246
+ � δ
1247
+ δ
1248
+ �∂Φ(δ, x3(δ))
1249
+ ∂x3(δ)
1250
+ − dC(x3(δ))
1251
+ dx3(δ)
1252
+
1253
+ f(δ)
1254
+ +λo
1255
+ 1(δ) + λo
1256
+ 2(δ)f(δ)dδ.
1257
+ (42)
1258
+ Using the transversality conditions λ3(δ) = λ3(¯δ) = 0
1259
+ yields λ3(¯δ) = −
1260
+ � ¯δ
1261
+ δ ( ∂Φ(δ,x3(δ))
1262
+ ∂x3(δ)
1263
+ − dC(x3(δ))
1264
+ dx3(δ) )f(δ) + λo
1265
+ 1(δ) +
1266
+ λo
1267
+ 2(δ)f(δ)dδ = 0. Furthermore, (30) indicates that µo(δ)
1268
+ maximizes H with µo(δ) ≥ 0. Note that in the Hamil-
1269
+ tonian H, the last term λ3(δ)µ(δ) imposes a non-positive
1270
+ value constraint on λ3(δ). Otherwise, H is unbounded from
1271
+ above due to µ(δ) ≥ 0. Then, to ensure the feasibility of
1272
+ maximization, we have λ3(δ) ≤ 0 which is equivalent to
1273
+ � δ
1274
+ δ ( ∂Φ(δ,x3(δ))
1275
+ ∂x3(δ)
1276
+ −C′(x3(δ)))f(δ) +λo
1277
+ 1(δ)+λo
1278
+ 2(δ)f(δ)dδ ≥ 0.
1279
+ Thus, when λ3(δ) < 0, ˙xo
1280
+ 3(δ) = µo(δ) = 0. Therefore, the
1281
+ complementary slackness condition can be written as follows,
1282
+ ∀δ ∈ [δ, ¯δ],
1283
+ ˙xo
1284
+ 3(δ)
1285
+ � δ
1286
+ δ
1287
+ �∂Φ(δ, xo
1288
+ 3(δ))
1289
+ ∂xo
1290
+ 3(δ)
1291
+ − dC(x3(δ))
1292
+ dx3(δ)
1293
+
1294
+ f(δ)
1295
+ +λo
1296
+ 1(δ) + λo
1297
+ 2(δ)f(δ)dδ = 0.
1298
+ (43)
1299
+ We can verify that the maximum principle (30)–(39) is also
1300
+ sufficient for optimality as the associated Hamiltonian equation
1301
+ is concave in both x and µ. Furthermore, the Hamiltonian is
1302
+ strictly concave in x3 and other states are uniquely determined
1303
+ by x3. Thus, the optimal control and optimal state trajectory
1304
+ are unique [45]. We next explicitly characterize this optimal
1305
+ solution.
1306
+ B. Characterization of Optimal Contracts
1307
+ We next analyze the optimal contracts in two regimes
1308
+ regarding ˙xo
1309
+ 3(δ), i.e., ˙xo
1310
+ 3(δ) > 0 and ˙xo
1311
+ 3(δ) = 0. Based on
1312
+ (43), in the interval of δ that ˙xo
1313
+ 3(δ) > 0, then λo
1314
+ 3(δ) = 0
1315
+ for all δ in this interval, which further indicates ˙λo
1316
+ 3 = 0.
1317
+ Hence, from (36), the following equation holds: ( ∂Φ(δ,x3(δ))
1318
+ ∂x3(δ)
1319
+
1320
+ dC(x3(δ))
1321
+ dx3(δ) )f(δ) + λo
1322
+ 1(δ) + λo
1323
+ 2(δ)f(δ) = 0, which is exactly
1324
+ the same maximality condition presented in (22), where x3(δ)
1325
+ plays the role as u(δ). Following the same analysis in Section
1326
+ III-B, the optimal solutions to xo
1327
+ 1, xo
1328
+ 2, xo
1329
+ 3, λo
1330
+ 1 and λo
1331
+ 2 in
1332
+ Hamilton system (30)–(39) coincide with x∗
1333
+ 1, x∗
1334
+ 2, u∗, λ∗
1335
+ 1 and
1336
+ λ∗
1337
+ 2 in Hamilton system (15)–(21). Thus, we can conclude that
1338
+ if xo
1339
+ 3(δ) is strictly increasing over some interval and recall
1340
+ the notation x3(δ) = q(δ), the solution qo(δ) in this section
1341
+ should be the same as the one q∗(δ) in Theorem 1.
1342
+ In the other regime of ˙xo
1343
+ 3(δ) = 0, xo
1344
+ 3(δ) is unchanged.
1345
+ Then, the remaining task is to determine the intervals of δ in
1346
+ which qo(δ) admits a constant, and hence the service price is
1347
+ nondiscriminative. Note that these intervals definitely include
1348
+ the ones when q∗(δ) is decreasing, i.e., the monotonicity con-
1349
+ straint of sensing QoS is violated. For notational convenience,
1350
+ let [δ1, δ2] be the interval when qo(δ) is a constant, δ ∈ [δ1, δ2].
1351
+ We know that for δ < δ1 and δ > δ2, qo(δ) is increasing, and
1352
+ thus ˙xo
1353
+ 3(δ) > 0. Based on (43), we obtain condition λo
1354
+ 3(δ) = 0.
1355
+ Since the costate variable λo
1356
+ 3 is continuous, then at the critical
1357
+ points δ1 and δ2, λo
1358
+ 3(δ1) = λo
1359
+ 3(δ2) = 0, and using (42) yields
1360
+ � δ2
1361
+ δ1
1362
+ �∂Φ(δ, q(δ))
1363
+ ∂q(δ)
1364
+ − dC(q(δ))
1365
+ dq(δ)
1366
+
1367
+ f(δ)
1368
+ +λo
1369
+ 1(δ) + λo
1370
+ 2(δ)f(δ)dδ = 0.
1371
+ (44)
1372
+ To this end, we discuss three possible cases that qo(δ)
1373
+ is nondiscriminative over δ ∈ [δ1, δ2] subsequently. When
1374
+ analyzing qo(δ), we constantly refer to the optimal solution
1375
+ q∗(δ) in Theorem 1. Besides, we assume that both λo
1376
+ 1 and λo
1377
+ 2
1378
+ are known through (40) and (41) with an exception of β to be
1379
+ specified later.
1380
+ Case I: (δ1 = δ). In this case, (44) is reduced to
1381
+ � δ2
1382
+ δ
1383
+ �∂Φ(δ, q1)
1384
+ ∂q(δ)
1385
+ − dC(q1)
1386
+ dq(δ)
1387
+
1388
+ f(δ)
1389
+ +λo
1390
+ 1(δ) + λo
1391
+ 2(δ)f(δ)dδ = 0,
1392
+ q1 = q∗(δ2).
1393
+ (45)
1394
+ One illustrative example for this scenario is shown in Fig.
1395
+ 3(a), where for δ ∈ [δ2, ¯δ], qo(δ) = q∗(δ). In addition, the
1396
+ constant value q1 is no greater than q∗(δ), i.e., q1 ≤ q∗(δ).
1397
+ We prove this result by contradiction. If q1 > q∗(δ), then
1398
+ q1 > q∗(˜δ) for any ˜δ close enough to δ. Along with the entire
1399
+ trajectory q∗(δ), we introduce a virtual variable λ∗
1400
+ 3(δ) which
1401
+ is a counterpart of λo
1402
+ 3(δ), and thus we have λ∗
1403
+ 3(δ) = 0. Recall
1404
+ the notation x3 = q, and then the partial integrand ∂Φ(δ,q(δ))
1405
+ ∂q(δ)
1406
+
1407
+ dC(q(δ))
1408
+ dq(δ)
1409
+ in (42) decreases when the value of q increases due
1410
+ to the convexity of cost function C. Thus, the entire λo
1411
+ 3(δ)
1412
+ increases if q becomes larger. Therefore, for ˜δ close enough to
1413
+ δ and based on the assumption q1 > q∗(δ), we obtain λo
1414
+ 3(˜δ) >
1415
+ λ∗
1416
+ 3(˜δ) = 0, contradicting the condition λo
1417
+ 3(δ) ≤ 0, ∀δ ∈ [δ, ¯δ].
1418
+ Therefore, we can obtain δ2 and the corresponding value q1
1419
+ by solving two equations in (45).
1420
+ Case II: (δ < δ1 < δ2 < ¯δ). When the interval [δ1, δ2] lies
1421
+ in the interior of the entire regime δ, (44) becomes
1422
+ � δ2
1423
+ δ1
1424
+ �∂Φ(δ, q2)
1425
+ ∂q(δ)
1426
+ − dC(q2)
1427
+ dq(δ)
1428
+
1429
+ f(δ)
1430
+ +λo
1431
+ 1(δ) + λo
1432
+ 2(δ)f(δ)dδ = 0,
1433
+ q2 = q∗(δ1) = q∗(δ2).
1434
+ (46)
1435
+ We can solve for two unknowns δ1 and δ2 based on (46), and
1436
+ subsequently we obtain q2. Case II is depicted in Fig. 3(b).
1437
+
1438
+ 𝛿
1439
+ 𝑞∗(𝛿)
1440
+ 𝑞𝑜(𝛿)
1441
+ 𝛿
1442
+ 𝛿=𝛿1
1443
+ 𝛿2
1444
+ 𝑞1
1445
+ 𝑞𝑜 𝛿 , 𝑞∗(𝛿)
1446
+ 𝑞𝑜(𝛿)
1447
+ Case I: 𝛿=𝛿1
1448
+ (a) Case I: δ1 = δ
1449
+ 𝛿
1450
+ 𝑞∗(𝛿)
1451
+ 𝑞𝑜(𝛿)
1452
+ 𝛿
1453
+ 𝛿
1454
+ 𝛿2
1455
+ 𝑞2
1456
+ 𝑞𝑜 𝛿 , 𝑞∗(𝛿)
1457
+ 𝑞𝑜(𝛿)
1458
+ 𝛿1
1459
+ 𝑞𝑜 𝛿 , 𝑞∗(𝛿)
1460
+ Case II: 𝛿<𝛿1<𝛿2<𝛿
1461
+ (b) Case II: δ < δ1 < δ2 < ¯δ
1462
+ 𝛿
1463
+ 𝑞∗(𝛿)
1464
+ 𝑞𝑜(𝛿)
1465
+ 𝛿=𝛿2
1466
+ 𝛿
1467
+ 𝛿1
1468
+ 𝑞3
1469
+ 𝑞𝑜 𝛿 , 𝑞∗(𝛿)
1470
+ 𝑞𝑜(𝛿)
1471
+ Case III: 𝛿=𝛿2
1472
+ (c) Case III: δ2 = ¯δ
1473
+ Fig. 3. In all three figures, qo(δ) and q∗(δ) represent the QoS of SaaS with and without considering the monotonicity constraint, respectively. In addition,
1474
+ the optimal solution qo(δ) coincides with q∗(δ) over some interval except δ ∈ [δ1, δ2] in three cases. For δ ∈ [δ1, δ2], qo(δ) is nondiscriminative and admits
1475
+ constant values q1 q2 and q3 in (a), (b) and (c), respectively.
1476
+ Case III: (δ2 = ¯δ). When δ2 coincides with the end-point
1477
+ ¯δ, (44) can be written as
1478
+ � ¯δ
1479
+ δ1
1480
+ �∂Φ(δ, q3)
1481
+ ∂q(δ)
1482
+ − dC(q3)
1483
+ dq(δ)
1484
+
1485
+ f(δ)
1486
+ +λo
1487
+ 1(δ) + λo
1488
+ 2(δ)f(δ)dδ = 0,
1489
+ q3 = q∗(δ1).
1490
+ (47)
1491
+ Fig. 3(c) presents an example of case III. Similar to the
1492
+ analysis in Case I, the value of q3 satisfies q3 ≥ q∗(¯δ).
1493
+ Furthermore, δ1 and q3 can be obtained by solving (47).
1494
+ Note that in the optimal contracts, the intervals over which
1495
+ qo(δ) admitting a constant value can be a combination of the
1496
+ three cases, and there could exist multiple interior intervals
1497
+ as the one shown in Fig. 3(b). Another essential point is to
1498
+ determine λo
1499
+ 2 = β in (45)–(47). As the analysis in Section
1500
+ III-B, the unknown constant β can be derived using the
1501
+ constraint (12). However, (12) needs a full expression of
1502
+ optimal qo beforehand. Therefore, two procedures including
1503
+ the derivation of optimal solution qo from (45)–(47) and the
1504
+ obtaining λ2(δ) = β by (12) are intertwined. To design the
1505
+ optimal qo(δ), we thus should solve the equations (45)–(47)
1506
+ together with (12) in a holistic manner. With derived qo(δ),
1507
+ the service pricing function po(δ) then can be characterized
1508
+ with similar steps in Section III-B.
1509
+ We summarize the optimal contracts for SaaS under general
1510
+ user’s type distribution in the following theorem.
1511
+ Theorem 3. For a general user’s type distribution f(δ) where
1512
+ 2f 2(δ) + (1 − F(δ))f ′(δ) > 0 does not hold, the optimal
1513
+ contracts {qo(δ), po(δ)} designed by the SP are detailed as
1514
+ follows. The QoS mapping qo(δ) is piecewise continuous and
1515
+ weakly increasing over δ ∈ [δ, ¯δ].
1516
+ 1) qo(δ) and po(δ) coincide with q∗(δ) and p∗(δ) in
1517
+ Theorem 1 except on a finite number N of disjoint
1518
+ intervals In = (δn
1519
+ 1 , δn
1520
+ 2 ), for n = 1, ..., N, and δn
1521
+ 1 and
1522
+ δn
1523
+ 2 increase with n. Furthermore,, qo(δ) = qn, ∀δ ∈ In.
1524
+ 2) For the interior interval In where δn
1525
+ 1 ̸= δ and δn
1526
+ 2 ̸= ¯δ,
1527
+ the optimal qo(δ) satisfies
1528
+ � δn
1529
+ 2
1530
+ δn
1531
+ 1
1532
+ �∂Φ(δ, qn)
1533
+ ∂q
1534
+ − dC(qn)
1535
+ dq
1536
+
1537
+ f(δ)
1538
+ +λo
1539
+ 1(δ) + λo
1540
+ 2(δ)f(δ)dδ = 0,
1541
+ qn = q∗(δn
1542
+ 1 ) = q∗(δn
1543
+ 2 ).
1544
+ (48)
1545
+ 3) If δ1
1546
+ 1 = δ, i.e., the interval I1 starts with δ, then the
1547
+ optimal qo(δ) satisfies
1548
+ � δ1
1549
+ 2
1550
+ δ
1551
+ �∂Φ(δ, q1)
1552
+ ∂q
1553
+ − dC(q1)
1554
+ dq
1555
+
1556
+ f(δ)
1557
+ +λo
1558
+ 1(δ) + λo
1559
+ 2(δ)f(δ)dδ = 0,
1560
+ q1 = q∗(δ1
1561
+ 2) ≤ q∗(δ).
1562
+ (49)
1563
+ 4) If δN
1564
+ 2
1565
+ = ¯δ, i.e., the interval IN ends with ¯δ, then the
1566
+ optimal qo(δ) satisfies
1567
+ � ¯δ
1568
+ δN
1569
+ 1
1570
+ �∂Φ(δ, qN)
1571
+ ∂q
1572
+ − dC(qN)
1573
+ dq
1574
+
1575
+ f(δ)
1576
+ +λo
1577
+ 1(δ) + λo
1578
+ 2(δ)f(δ)dδ = 0,
1579
+ qN = q∗(δN
1580
+ 1 ) ≥ q∗(¯δ).
1581
+ (50)
1582
+ 5) Based on (48)–(50) and together with (12), (40), (41),
1583
+ qn, δn
1584
+ 1 and δn
1585
+ 2 , n = 1, ..., N, can be computed. After
1586
+ obtaining the sensing QoS function qo(δ), the optimal
1587
+ pricing po(δ) can be derived via the relation
1588
+ po(δ) = Φ(δ, qo(δ)) − φ(δ),
1589
+ (51)
1590
+ where ˙φ(δ) = qo(δ) with φ(δ) = 0.
1591
+ Remark: For the intervals where qo(δ) = q∗(δ), po(δ) is
1592
+ monotonically increasing. For δ ∈ In, n = 1..., N, qo(δ) is a
1593
+ constant and then ˙qo(δ) = 0. Based on (51) and Φ(δ, qo(δ)) =
1594
+ δqo(δ), we obtain ˙po(δ) = δ ˙qo(δ) + qo(δ) − ˙φ(δ) = 0.
1595
+ Therefore, IoT users with a type lying in the same interval
1596
+ In, n = 1, ..., N, are provided with a menu of contracts with
1597
+ the same quality of sensing data as well as the service price.
1598
+ C. Some Analytical Results
1599
+ We end up this section by presenting analytical results on
1600
+ the pricing of sensing services. These results give insights on
1601
+
1602
+ the obtained solutions, and they also contribute to the design
1603
+ of practical market-based contracts.
1604
+ (1) Structure of the optimal contracts: Comparing with the
1605
+ optimal contracts in Theorem 1, the ones in Theorem 3 have
1606
+ an additional feature of nondiscriminative service intervals.
1607
+ Specifically, in addition to the profit maximization and service
1608
+ reputation construction of SP, the IC constraints of users are
1609
+ completely considered in the contracts, where the additional
1610
+ monotonicity part is reflected by (48)–(50). Note that the
1611
+ nondiscriminative pricing reduces the diversity of service
1612
+ provisions to the IoT users which has an interpretation that
1613
+ the SP treats heterogeneous users equally. Different with the
1614
+ contracts in Theorem 1 of full separation, the pooling behavior
1615
+ (users of different types are offered with the same contract) in
1616
+ Theorem 3 due to irregular type distribution is to ensure the
1617
+ incentive compatibility of designed optimal contracts.
1618
+ (2) Number of intervals with nondiscriminative pricing: Fig.
1619
+ 3 shows that the intervals with a decreasing q∗(δ) are included
1620
+ in In, n = 1, ..., N. Then, N is equal to the number of peaks
1621
+ (local maximum) of q∗(δ). Based on Theorem 1, we analyze
1622
+ the monotonicity of F (δ)−1
1623
+ f(δ)
1624
+ + δ, indicating that the number
1625
+ of nondiscriminative pricing regimes N coincides with the
1626
+ number of intervals where 2f 2(δ) + (1 − F(δ))f ′(δ) takes
1627
+ a negative value.
1628
+ (3) Nondiscriminative pricing for all users: When q∗(δ) is
1629
+ decreasing over δ ∈ [δ, ¯δ], then based on Theorem 3, the opti-
1630
+ mal service pricing qo(δ) is nondiscriminative for all types of
1631
+ users. In this scenario, we obtain 2f 2(δ)+(1−F(δ))f ′(δ) < 0
1632
+ for all δ. From Lemma 2, an equivalent condition is that
1633
+ 1−F (δ)
1634
+ f(δ)
1635
+ increases over δ. We summarize the results in the
1636
+ following lemma.
1637
+ Lemma 3. The optimal contracts {qo(δ), po(δ)} are nondis-
1638
+ criminative for all δ if 1−F (δ)
1639
+ f(δ)
1640
+ increases over δ ∈ [δ, ¯δ]. An
1641
+ alternative equivalent condition leading to the results is that
1642
+ function log[1 − F(δ)] is strictly convex.
1643
+ Some typical distributions satisfying Lemma 3 are worth
1644
+ highlighting. One example is when f(δ) is a gamma distri-
1645
+ bution for parameter α < 1, i.e., f(δ) =
1646
+ ψαδα−1 exp(−ψδ)
1647
+ Γ(α)
1648
+ ,
1649
+ where δ ≥ 0 and Γ(δ) is a complete Gamma function.
1650
+ Another example is when f(δ) admits a Weibull distribution
1651
+ under α < 1, i.e., f(δ) = ψαδα−1 exp(−ψδα), δ ≥ 0.
1652
+ In both types of distributions, most of the IoT users are
1653
+ with type δ = 0 or close to δ, and its number decreases
1654
+ exponentially as the parameter δ increases. Therefore, the SP
1655
+ designs nondiscriminative contracts for all users, extracting
1656
+ the profits from the majority of customers in the market.
1657
+ Moreover, this nondiscriminative service provision mechanism
1658
+ aligns with the phenomenon of focusing on the majority, where
1659
+ the small group of users with larger types are treated in a
1660
+ homogeneous manner as the major population nested in lower
1661
+ types.
1662
+ (4) Invariant nondiscriminative service pricing: One natural
1663
+ question is the impact of convexity of log[1 − F(δ)] on the
1664
+ service price. For various type distributions f(δ) satisfying the
1665
+ condition in Lemma 3, we show that the convexity of F(δ)
1666
+ has no influence on the neutral service pricing. Specifically,
1667
+ based on the constraint
1668
+ � ¯δ
1669
+ δ qo(δ)f(δ)dδ = q, where qo(δ) =
1670
+ qc, ∀δ, we obtain qc � ¯δ
1671
+ δ f(δ)dδ = q. Therefore, under the the
1672
+ nondiscriminative pricing of sensing services, the QoS is qc =
1673
+ q for all users. Furthermore, the IR constraint V (δ) = 0 leads
1674
+ to the optimal constant pricing pc = δq. Hence, whenever the
1675
+ SP offers a nondiscriminative price scheme to all IoT users,
1676
+ the price must be invariant equaling to δq in spite of the user’s
1677
+ type distributions.
1678
+ VII. CASE STUDIES: UAV-ENABLED VIRTUAL REALITY
1679
+ In this section, we apply the SaaS paradigm to UAV-enabled
1680
+ virtual reality as depicted in Fig. 1 to illustrate the optimal
1681
+ contract design principles. We envision a large VR service
1682
+ market in the future, and thus a huge number of users will
1683
+ purchase the VR services. This SaaS paradigm can be also
1684
+ applied to other personalized data related service provision
1685
+ scenarios, such as virtual tourism. This virtual service modality
1686
+ becomes popular under the current disruptions caused by
1687
+ COVID-19 pandemic worldwide.
1688
+ A. UAV-Enabled VR Setting
1689
+ The VR quality can be quantified by user experience related
1690
+ metrics, including the resolution of the captured scene of
1691
+ UAV (˜q1), the delay in sensing data transmission (˜q2), and
1692
+ the reliability of UAV communicating with the tower (˜q3).
1693
+ Specifically, for the resolution quality ˜q1, it can be in the
1694
+ general classes of 240p, 360p, 480p, 720p, 1080p (commonly
1695
+ available options such as in the streaming services), and the
1696
+ qualities between these classes. The delay ˜q2 is composed of
1697
+ factors including processing delay, queuing delay, transmission
1698
+ delay, and propagation delay of sensing data. The delay can
1699
+ be reduced by using a dedicated network that streamlines the
1700
+ network path, which is more costly for the sensing service
1701
+ provider. The tolerable end-to-end delay of modern VR ap-
1702
+ plications is of an order of milliseconds, and a desired QoS
1703
+ has it less than 1 or 2 milliseconds [46]. The communication
1704
+ reliability ˜q3 between UAV and tower can be measured by
1705
+ the success rate that data packets are transmitted. According
1706
+ to a video QoS tutorial by Cisco [47], the reliability should
1707
+ be above 99% for a high QoS, and it is between 99.5% and
1708
+ 95% depending on the specific type of services. The reliability
1709
+ above is quantified by the packet loss rate.
1710
+ We can aggregate these major metrics into a single measure
1711
+ q taking values in the real space. More specifically, the QoS
1712
+ q can be determined by a linear combination in a form of
1713
+ κ1˜q1 + κ2˜q2 + κ3˜q3, where κi, i = 1, 2, 3, are positive
1714
+ weighting factors. Equal weighting refers to the scenario with
1715
+ κ1 = κ2 = κ3 = 1/3. To differentiate the delivered services
1716
+ and pricing in terms of metrics considered, we consider that,
1717
+ comparing with a small q, a larger q has all higher values in
1718
+ ˜q1, ˜q2, and ˜q3. This modeling also fits the real-world scenario
1719
+ well, as the customers choose a higher QoS should receive
1720
+ better service in every factor considered (resolution, delay,
1721
+ reliability) by paying more service fee. We anticipate a large
1722
+ VR service market in the future, and thus a huge number of
1723
+ users will purchase the VR services. We further specify the
1724
+ mean QoS q = 5. As the sensing QoS is a mapping considering
1725
+
1726
+ various metrics, we set the mean QoS q = 5 corresponding
1727
+ to the service with 720p resolution, 0.15sec delay, and 97%
1728
+ UAV transmission reliability. After obtaining the QoS in the
1729
+ optimal contract later on, we can reversely map q to the three
1730
+ specific metrics considered. Based on the current technologies
1731
+ in communication and VR, we consider the resolution, delay,
1732
+ and reliability admit a value from 240p to 1080p, 0.5 ms to 5
1733
+ ms, and 0.95% to 0.99%, respectively. Note that in the optimal
1734
+ mechanism design, higher types of users receive better quality
1735
+ of VR service from the SP.
1736
+ As depicted in Fig. 2, the user’s type distribution admits
1737
+ f(δ) = 0.952e−0.952δ, and thus F(δ) = 1 − e−0.952δ. These
1738
+ distribution functions are aligned with the market data as
1739
+ discussed in Example 1 in Section II-A.
1740
+ B. Optimal Contracts under Hidden Information
1741
+ Based on Corollary 3, we depict the optimal contracts
1742
+ of VR services in Fig. 4 with various values of a. The
1743
+ weighting factor σ admits a value of 0.16, which gives a
1744
+ reasonable comparison between the service charging fee and
1745
+ the cost of providing the service. In the cases with parameter
1746
+ a = 0.47, 0.49, 0.51, and using the results in Section IV-B,
1747
+ we obtain β = 1.14, 1.215, 1.315, respectively. With these
1748
+ selected parameters, the obtained service pricing also matches
1749
+ with the data market. One observation is that both the VR pric-
1750
+ ing and the QoS mappings are monotonically increasing with
1751
+ the user’s type, leading to an incentive compatible contract.
1752
+ Another phenomenon is that as a increases, the VR QoS is
1753
+ decreasing for a given user’s type under the regime δ > 0.47
1754
+ as shown in Fig. 4(b). The reason is that a larger a indicates
1755
+ a higher service cost of the SP which leads to a degraded VR
1756
+ QoS. Thus, the VR pricing decreases as well for a given δ as
1757
+ illustrated in Fig. 4(a). Different with the findings in regime
1758
+ δ > 0.47, the VR QoS increases with the parameter a when
1759
+ δ < 0.47, showing that a larger cost of the SP provides a
1760
+ better VR service for the customers of type δ < 0.47 while
1761
+ the customers paying less. Note that the mean VR QoS q
1762
+ stays the same for all investigated cases. Then, to maintain a
1763
+ constant reputation that the VR SP builds in the market, the
1764
+ received QoS for customers of type δ < 0.47 should increase
1765
+ with a comparing with those of δ > 0.47. This phenomenon
1766
+ also aligns with the fact that at the early stage of VR services
1767
+ promotion (a is large), the SP focuses more on the types of
1768
+ customers with a large population in the market (small δ in
1769
+ the exponential distribution), by providing a relatively better
1770
+ VR service. Based on the VR application modeling in Section
1771
+ VII-A, Fig. 4(c) presents the specific sensing QoS in terms of
1772
+ the considered resolution, delay, and reliability metrics. Under
1773
+ the the designed optimal contracts {p∗(δ), q∗(δ)}, Fig. 5 shows
1774
+ the corresponding utility of SP. As a increases which yields
1775
+ a larger service cost, the SP’s aggregate revenue decreases
1776
+ accordingly. In addition, for some small types δ close to δ,
1777
+ U(δ) can be negative. This phenomenon indicates that the SP
1778
+ makes most of the profits from the users who demand a high
1779
+ VR QoS.
1780
+ 0
1781
+ 0.5
1782
+ 1
1783
+ 1.5
1784
+ 2
1785
+ 2.5
1786
+ 3
1787
+ 3.5
1788
+ 4
1789
+ VR user's type
1790
+ 0
1791
+ 2
1792
+ 4
1793
+ 6
1794
+ 8
1795
+ 10
1796
+ 12
1797
+ 14
1798
+ VR pricing scheme p*( ) ($)
1799
+ a=0.47
1800
+ a=0.49
1801
+ a=0.51
1802
+ (a) VR pricing p∗(δ)
1803
+ 0
1804
+ 0.5
1805
+ 1
1806
+ 1.5
1807
+ 2
1808
+ 2.5
1809
+ 3
1810
+ 3.5
1811
+ 4
1812
+ VR user's type
1813
+ 0
1814
+ 1
1815
+ 2
1816
+ 3
1817
+ 4
1818
+ 5
1819
+ 6
1820
+ 7
1821
+ 8
1822
+ 9
1823
+ VR QoS q*( )
1824
+ a=0.47
1825
+ a=0.49
1826
+ a=0.51
1827
+ (b) VR QoS q∗(δ)
1828
+ 0
1829
+ 0.5
1830
+ 1
1831
+ 1.5
1832
+ 2
1833
+ 2.5
1834
+ 3
1835
+ 3.5
1836
+ 4
1837
+ VR user's type
1838
+ 400
1839
+ 600
1840
+ 800
1841
+ 1000
1842
+ Resolution (p)
1843
+ a=0.47
1844
+ a=0.49
1845
+ a=0.51
1846
+ 0
1847
+ 0.5
1848
+ 1
1849
+ 1.5
1850
+ 2
1851
+ 2.5
1852
+ 3
1853
+ 3.5
1854
+ 4
1855
+ VR user's type
1856
+ 0
1857
+ 0.1
1858
+ 0.2
1859
+ 0.3
1860
+ Delay (sec)
1861
+ a=0.47
1862
+ a=0.49
1863
+ a=0.51
1864
+ 0
1865
+ 0.5
1866
+ 1
1867
+ 1.5
1868
+ 2
1869
+ 2.5
1870
+ 3
1871
+ 3.5
1872
+ 4
1873
+ VR user's type
1874
+ 0.94
1875
+ 0.96
1876
+ 0.98
1877
+ 1
1878
+ Reliability (%)
1879
+ a=0.47
1880
+ a=0.49
1881
+ a=0.51
1882
+ (c) VR QoS in terms of resolution, delay, and reliability
1883
+ Fig. 4. (a) and (b) illustrate the optimal pricing scheme and the corresponding
1884
+ QoS of VR, respectively. (c) depicts the specific sensing QoS in terms of
1885
+ resolution, delay, and reliability metrics.
1886
+ C. Optimal Contracts under Full Information
1887
+ For comparison, we present the optimal contracts under
1888
+ the full information based on Theorem 2 and quantify the
1889
+ information cost associated with the user’s private types. Fig.
1890
+ 6 shows the optimal pricing pb(δ) and the QoS mapping qb(δ).
1891
+ Specifically, pb(δ) is larger than the counterpart p∗(δ) under
1892
+ asymmetric information. Due to the reputation constraint,
1893
+ the VR QoS qb(δ) has a similar trajectory as q∗(δ). The
1894
+ corresponding SP’s revenue is shown in Fig. 7. Similarly, a
1895
+ larger a reduces the payoff of the VR SP. Furthermore, we
1896
+ can conclude that the SP earns more by knowing the private
1897
+ user’s type information. For example, when a = 0.47, the
1898
+ average utility of serving a user is 4.4$ which is more than 4
1899
+ times larger than the one under hidden information depicted
1900
+ in Fig. 5.
1901
+ VIII. CONCLUSION
1902
+ In this paper, we have established a Sensing-as-Service
1903
+ (SaaS) framework for QoS-based data trading in the IoT
1904
+ markets using contract theory. The proposed framework is de-
1905
+ signed for massive IoT scenarios where users are characterized
1906
+ by their service requirements and sensing data available to the
1907
+ service provider (SP) is characterized by quality. Depending
1908
+ on the probability distribution of user’s QoS needs, the profit
1909
+
1910
+ 0
1911
+ 1
1912
+ 2
1913
+ 3
1914
+ 4
1915
+ VR user's type
1916
+ -1
1917
+ 0
1918
+ 1
1919
+ 2
1920
+ 3
1921
+ 4
1922
+ 5
1923
+ 6
1924
+ Utility U( ) ($)
1925
+ a=0.47
1926
+ a=0.49
1927
+ a=0.51
1928
+ a=0.47
1929
+ a=0.49
1930
+ a=0.51
1931
+ 0
1932
+ 0.1
1933
+ 0.2
1934
+ 0.3
1935
+ 0.4
1936
+ 0.5
1937
+ 0.6
1938
+ 0.7
1939
+ 0.8
1940
+ 0.9
1941
+ 1
1942
+ Average Utility of SP ($)
1943
+ Fig. 5.
1944
+ Utility of the SP under hidden information. The SP earns profits
1945
+ from the users who demand a better VR service.
1946
+ 0
1947
+ 1
1948
+ 2
1949
+ 3
1950
+ 4
1951
+ VR user's type
1952
+ 0
1953
+ 1
1954
+ 2
1955
+ 3
1956
+ 4
1957
+ 5
1958
+ 6
1959
+ 7
1960
+ 8
1961
+ 9
1962
+ VR QoS qb( )
1963
+ a=0.47
1964
+ a=0.49
1965
+ a=0.51
1966
+ 0
1967
+ 1
1968
+ 2
1969
+ 3
1970
+ 4
1971
+ VR user's type
1972
+ 0
1973
+ 5
1974
+ 10
1975
+ 15
1976
+ 20
1977
+ 25
1978
+ 30
1979
+ 35
1980
+ VR pricing scheme pb( ) ($)
1981
+ a=0.47
1982
+ a=0.49
1983
+ a=0.51
1984
+ Benchmark Scenario
1985
+ Fig. 6.
1986
+ Optimal contracts in the benchmark scenario. The VR service pricing
1987
+ pb(δ) is larger than the counterpart p∗(δ).
1988
+ maximizing contract solutions are proposed between the SP
1989
+ and users, which admit different structures. Specifically, under
1990
+ a wide class of user’s type distributions without a large or
1991
+ sudden decrease, the data pricing scheme and QoS mapping
1992
+ are monotonically increasing with the user types. Otherwise,
1993
+ nondiscriminative pricing phenomenon is observed which re-
1994
+ duces the diversity of service provisions to the IoT users.
1995
+ Moreover, invariant pricing phenomenon can occur when the
1996
+ user’s type distribution decreases exponentially, and thus the
1997
+ service provider targets the majority of users in the market to
1998
+ maximize the profits. We have also validated our results using
1999
+ a case study based on the application of the SaaS framework to
2000
+ UAV-enabled virtual reality, where the SP makes more profit
2001
+ by providing data services to higher type users. Future work
2002
+ can expand the SaaS contract design to cases when bounded
2003
+ rationality is considered in the user’s behavior, i.e., users have
2004
+ uncertainty on their type parameters, and subsequently design
2005
+ robust contract mechanisms. Another direction is to develop
2006
+ 0
2007
+ 1
2008
+ 2
2009
+ 3
2010
+ 4
2011
+ VR user's type
2012
+ -5
2013
+ 0
2014
+ 5
2015
+ 10
2016
+ 15
2017
+ 20
2018
+ 25
2019
+ 30
2020
+ Utility U( ) ($)
2021
+ a=0.47
2022
+ a=0.49
2023
+ a=0.51
2024
+ a=0.47
2025
+ a=0.49
2026
+ a=0.51
2027
+ 0
2028
+ 0.5
2029
+ 1
2030
+ 1.5
2031
+ 2
2032
+ 2.5
2033
+ 3
2034
+ 3.5
2035
+ 4
2036
+ Average Utility of SP ($)
2037
+ Benchmark Scenario
2038
+ Fig. 7.
2039
+ Utility of the SP in the benchmark scenario. The SP’s revenue under
2040
+ full information is more than 4 times larger than the corresponding one under
2041
+ asymmetric information.
2042
+ an online learning approach to designing optimal contract
2043
+ solutions when the user’s type distribution is unknown to the
2044
+ SP.
2045
+ APPENDIX A
2046
+ PROOF OF LEMMA 1
2047
+ The first-order optimality condition (FOC) on (1) with
2048
+ respect to δ′ can be expressed as ∂Φ(δ,q(δ′))
2049
+ ∂q(δ′)
2050
+ dq(δ′)
2051
+ dδ′ − dp(δ′)
2052
+ dδ′
2053
+ = 0.
2054
+ The IC constraint in (3) indicates that the user of type δ
2055
+ achieves the largest payoff when claiming its true type δ. Thus,
2056
+ under δ′ = δ, the FOC becomes ∂Φ(δ,q(δ))
2057
+ ∂q(δ)
2058
+ dq(δ)
2059
+
2060
+ − dp(δ)
2061
+
2062
+ = 0,
2063
+ which yields the local incentive constraint (6). Similarly,
2064
+ the second-order optimality condition (SOC) can be written
2065
+ as:
2066
+ ∂2Φ(δ,q(δ′))
2067
+ ∂q(δ′)2
2068
+ ( dq(δ′)
2069
+ dδ′ )2 + ∂Φ(δ,q(δ′))
2070
+ ∂q(δ′)
2071
+ d2q(δ′)
2072
+ dδ′2
2073
+ − d2p(δ′)
2074
+ dδ′2
2075
+ ≤ 0.
2076
+ Differentiating (6) with respect to δ further gives
2077
+ d2p(δ)
2078
+ dδ2
2079
+ =
2080
+ ∂Φ2(δ,q(δ))
2081
+ ∂q(δ)2
2082
+ ( dq(δ)
2083
+ dδ )2 + ∂Φ2(δ,q(δ))
2084
+ ∂q(δ)∂δ
2085
+ dq(δ)
2086
+
2087
+ + ∂Φ(δ,q(δ))
2088
+ ∂q(δ)
2089
+ d2q(δ)
2090
+ dδ2 , and
2091
+ comparing it with the SOC, we obtain ∂Φ2(δ,q(δ))
2092
+ ∂q(δ)∂δ
2093
+ dq(δ)
2094
+
2095
+ ≥ 0.
2096
+ Together with Assumption 1, we obtain the monotonicity
2097
+ constraint (7). The next step is to show that (6) and (7) together
2098
+ imply the IC constraint (3). Assume that the IC constraint does
2099
+ not hold for at least one type of users, e.g., δ. Then, there
2100
+ exists a ˜δ ̸= δ such that Φ(δ, q(δ))−p(δ) < Φ(δ, q(˜δ))−p(˜δ),
2101
+ and hence
2102
+ � ˜δ
2103
+ δ ( ∂Φ(δ,q(τ))
2104
+ ∂q(τ)
2105
+ dq(τ)
2106
+
2107
+ − dp(τ)
2108
+ dτ )dτ > 0, where we can
2109
+ check that the derivative of Φ(δ, q(τ)) − p(τ) with respect
2110
+ to τ is exactly the integrand. Then when ˜δ > δ which gives
2111
+ τ > δ, we obtain ∂Φ(δ,q(τ))
2112
+ ∂q(τ)
2113
+ < ∂Φ(τ,q(τ))
2114
+ ∂q(τ)
2115
+ by Assumption 1. In
2116
+ addition, (6) indicates that
2117
+ � ˜δ
2118
+ δ ( ∂Φ(τ,q(τ))
2119
+ ∂q(τ)
2120
+ dq(τ)
2121
+
2122
+ − dp(τ)
2123
+ dτ )dτ = 0.
2124
+ Replacing
2125
+ ∂Φ(τ,q(τ))
2126
+ ∂q(τ)
2127
+ in the integrand by
2128
+ ∂Φ(δ,q(τ))
2129
+ ∂q(τ)
2130
+ yields
2131
+ � ˜δ
2132
+ δ ( ∂Φ(δ,q(τ))
2133
+ ∂q(τ)
2134
+ dq(τ)
2135
+
2136
+
2137
+ dp(τ)
2138
+ dτ )dτ
2139
+ <
2140
+ 0 since
2141
+ ∂Φ(δ,q(τ))
2142
+ ∂q(τ)
2143
+ <
2144
+ ∂Φ(τ,q(τ))
2145
+ ∂q(τ)
2146
+ and dq(τ)
2147
+
2148
+ > 0, and this inequality contradicts with
2149
+ the previous integral inequality. Similar analysis follows for
2150
+ the case when ˜δ < δ, and we can conclude that (6) and (7)
2151
+ imply the IC constraint (3).
2152
+
2153
+ REFERENCES
2154
+ [1] A. Zanella, N. Bui, A. Castellani, L. Vangelista, and M. Zorzi, “Internet
2155
+ of things for smart cities,” IEEE Internet of Things Journal, vol. 1, no. 1,
2156
+ pp. 22–32, 2014.
2157
+ [2] S. Xiong, Q. Ni, X. Wang, and Y. Su, “A connectivity enhancement
2158
+ scheme based on link transformation in IoT sensing networks,” IEEE
2159
+ Internet of Things Journal, vol. 4, no. 6, pp. 2297–2308, 2017.
2160
+ [3] Y. Zhou, C. Pan, P. L. Yeoh, K. Wang, M. Elkashlan, B. Vucetic,
2161
+ and Y. Li, “Communication-and-computing latency minimization for
2162
+ UAV-enabled virtual reality delivery systems,” IEEE Transactions on
2163
+ Communications, vol. 69, no. 3, pp. 1723–1735, 2021.
2164
+ [4] C. Perera, A. Zaslavsky, P. Christen, and D. Georgakopoulos, “Sensing
2165
+ as a service model for smart cities supported by Internet of things,”
2166
+ Transactions on Emerging Telecommunications Technologies, vol. 25,
2167
+ no. 1, pp. 81–93, 2014.
2168
+ [5] P. Semasinghe, S. Maghsudi, and E. Hossain, “Game theoretic mecha-
2169
+ nisms for resource management in massive wireless IoT systems,” IEEE
2170
+ Communications Magazine, vol. 55, no. 2, pp. 121–127, 2017.
2171
+ [6] R.
2172
+ E.
2173
+ Kopp,
2174
+ “Pontryagin
2175
+ maximum
2176
+ principle,”
2177
+ in
2178
+ Optimization
2179
+ Techniques, ser. Mathematics in Science and Engineering, G. Leitmann,
2180
+ Ed.
2181
+ Elsevier,
2182
+ 1962,
2183
+ vol.
2184
+ 5,
2185
+ pp.
2186
+ 255–279.
2187
+ [Online].
2188
+ Available:
2189
+ https://www.sciencedirect.com/science/article/pii/S0076539208620950
2190
+ [7] J.-J. Laffont and D. Martimort, The Theory of Incentives: The Principal-
2191
+ Agent Model.
2192
+ Princeton university press, 2009.
2193
+ [8] E. Fehr, A. Klein, and K. M. Schmidt, “Fairness and contract design,”
2194
+ Econometrica, vol. 75, no. 1, pp. 121–154, 2007.
2195
+ [9] N. A. Doherty and G. Dionne, “Insurance with undiversifiable risk:
2196
+ Contract structure and organizational form of insurance firms,” Journal
2197
+ of Risk and Uncertainty, vol. 6, no. 2, pp. 187–203, 1993.
2198
+ [10] C. J. Corbett and X. De Groote, “A supplier’s optimal quantity discount
2199
+ policy under asymmetric information,” Management Science, vol. 46,
2200
+ no. 3, pp. 444–450, 2000.
2201
+ [11] S. W. Driessen, G. Monsieur, and W.-J. Van Den Heuvel, “Data market
2202
+ design: A systematic literature review,” IEEE Access, vol. 10, pp.
2203
+ 33 123–33 153, 2022.
2204
+ [12] S. Sheng, R. Chen, P. Chen, X. Wang, and L. Wu, “Futures-based
2205
+ resource trading and fair pricing in real-time IoT networks,” IEEE
2206
+ Wireless Communications Letters, vol. 9, no. 1, pp. 125–128, 2020.
2207
+ [13] C. Su, F. Ye, Y. Zha, T. Liu, Y. Zhang, and Z. Han, “Matching with
2208
+ contracts-based resource trading and price negotiation in multi-access
2209
+ edge computing,” IEEE Wireless Communications Letters, vol. 10, no. 4,
2210
+ pp. 892–896, 2021.
2211
+ [14] N. Gupta, J. Singh, S. K. Dhurandher, and Z. Han, “Contract theory
2212
+ based incentive design mechanism for opportunistic IoT networks,”
2213
+ IEEE Internet of Things Journal, pp. 1–1, 2021.
2214
+ [15] C. Yi, J. Cai, and Z. Su, “A multi-user mobile computation offloading
2215
+ and transmission scheduling mechanism for delay-sensitive applica-
2216
+ tions,” IEEE Transactions on Mobile Computing, vol. 19, no. 1, pp.
2217
+ 29–43, 2020.
2218
+ [16] M. Diamanti, P. Charatsaris, E. E. Tsiropoulou, and S. Papavassiliou,
2219
+ “Incentive mechanism and resource allocation for edge-fog networks
2220
+ driven by multi-dimensional contract and game theories,” IEEE Open
2221
+ Journal of the Communications Society, vol. 3, pp. 435–452, 2022.
2222
+ [17] X. Zhou, W. Wang, N. U. Hassan, C. Yuen, and D. Niyato, “Towards
2223
+ small aoi and low latency via operator content platform: A contract
2224
+ theory-based pricing,” IEEE Transactions on Communications, vol. 70,
2225
+ no. 1, pp. 366–378, 2022.
2226
+ [18] T. N. Dang, K. Kim, L. U. Khan, S. M. A. Kazmi, Z. Han, and C. S.
2227
+ Hong, “On-device computational caching-enabled augmented reality for
2228
+ 5g and beyond: A contract-theory-based incentive mechanism,” IEEE
2229
+ Internet of Things Journal, vol. 8, no. 24, pp. 17 382–17 394, 2021.
2230
+ [19] K. Liu, X. Qiu, W. Chen, X. Chen, and Z. Zheng, “Optimal pricing
2231
+ mechanism for data market in blockchain-enhanced Internet of things,”
2232
+ IEEE Internet of Things Journal, vol. 6, no. 6, pp. 9748–9761, 2019.
2233
+ [20] Y. Lu, Y. Qi, S. Qi, Y. Li, H. Song, and Y. Liu, “Say no to price dis-
2234
+ crimination: Decentralized and automated incentives for price auditing in
2235
+ ride-hailing services,” IEEE Transactions on Mobile Computing, vol. 21,
2236
+ no. 2, pp. 663–680, 2022.
2237
+ [21] N. Weerasinghe, T. Hewa, M. Liyanage, S. S. Kanhere, and M. Ylianttila,
2238
+ “A novel Blockchain-as-a-service (BaaS) platform for local 5g opera-
2239
+ tors,” IEEE Open Journal of the Communications Society, vol. 2, pp.
2240
+ 575–601, 2021.
2241
+ [22] L. D. Nguyen, I. Leyva-Mayorga, A. N. Lewis, and P. Popovski,
2242
+ “Modeling and analysis of data trading on blockchain-based market in
2243
+ IoT networks,” IEEE Internet of Things Journal, vol. 8, no. 8, pp. 6487–
2244
+ 6497, 2021.
2245
+ [23] J. Li, T. Liu, D. Niyato, P. Wang, J. Li, and Z. Han, “Contract-theoretic
2246
+ pricing for security deposits in sharded blockchain with Internet of things
2247
+ (IoT),” IEEE Internet of Things Journal, vol. 8, no. 12, pp. 10 052–
2248
+ 10 070, 2021.
2249
+ [24] Y. Zhang, M. Pan, L. Song, Z. Dawy, and Z. Han, “A survey of contract
2250
+ theory-based incentive mechanism design in wireless networks,” IEEE
2251
+ Wireless Communications, vol. 24, no. 3, pp. 80–85, 2017.
2252
+ [25] Y. Zhang, L. Song, W. Saad, Z. Dawy, and Z. Han, “Contract-based
2253
+ incentive mechanisms for device-to-device communications in cellular
2254
+ networks,” IEEE Journal on Selected Areas in Communications, vol. 33,
2255
+ no. 10, pp. 2144–2155, 2015.
2256
+ [26] Y. Chen, S. He, F. Hou, Z. Shi, and J. Chen, “Promoting device-to-
2257
+ device communication in cellular networks by contract-based incentive
2258
+ mechanisms,” IEEE Network, vol. 31, no. 3, pp. 14–20, 2017.
2259
+ [27] Z. Hasan and V. K. Bhargava, “Relay selection for OFDM wireless sys-
2260
+ tems under asymmetric information: A contract-theory based approach,”
2261
+ IEEE Transactions on Wireless Communications, vol. 12, no. 8, pp.
2262
+ 3824–3837, 2013.
2263
+ [28] L. Duan, L. Gao, and J. Huang, “Cooperative spectrum sharing: A
2264
+ contract-based approach,” IEEE Transactions on Mobile Computing,
2265
+ vol. 13, no. 1, pp. 174–187, 2014.
2266
+ [29] L. Gao, X. Wang, Y. Xu, and Q. Zhang, “Spectrum trading in cognitive
2267
+ radio networks: A contract-theoretic modeling approach,” IEEE Journal
2268
+ on Selected Areas in Communications, vol. 29, no. 4, pp. 843–855, 2011.
2269
+ [30] Z. Chang, D. Zhang, T. H¨am¨al¨ainen, Z. Han, and T. Ristaniemi,
2270
+ “Incentive mechanism for resource allocation in wireless virtualized
2271
+ networks with multiple infrastructure providers,” IEEE Transactions on
2272
+ Mobile Computing, vol. 19, no. 1, pp. 103–115, 2018.
2273
+ [31] Q. Ma, L. Gao, Y.-F. Liu, and J. Huang, “A contract-based incen-
2274
+ tive mechanism for crowdsourced wireless community networks,” in
2275
+ Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks
2276
+ (WiOpt), 2016, pp. 1–8.
2277
+ [32] J. Nie, J. Luo, Z. Xiong, D. Niyato, and P. Wang, “A Stackelberg
2278
+ game approach toward socially-aware incentive mechanisms for mobile
2279
+ crowdsensing,” IEEE Transactions on Wireless Communications, vol. 18,
2280
+ no. 1, pp. 724–738, 2018.
2281
+ [33] Z. Xiong, D. Niyato, P. Wang, Z. Han, and Y. Zhang, “Dynamic pricing
2282
+ for revenue maximization in mobile social data market with network
2283
+ effects,” IEEE Transactions on Wireless Communications, vol. 19, no. 3,
2284
+ pp. 1722–1737, 2020.
2285
+ [34] L. Duan, T. Kubo, K. Sugiyama, J. Huang, T. Hasegawa, and J. Walrand,
2286
+ “Motivating smartphone collaboration in data acquisition and distributed
2287
+ computing,” IEEE Transactions on Mobile Computing, vol. 13, no. 10,
2288
+ pp. 2320–2333, 2014.
2289
+ [35] Z. Xiong, J. Zhao, Y. Zhang, D. Niyato, and J. Zhang, “Contract
2290
+ design in hierarchical game for sponsored content service market,” IEEE
2291
+ Transactions on Mobile Computing, vol. 20, no. 9, pp. 2763–2778, 2020.
2292
+ [36] F. Al-Turjman, “Price-based data delivery framework for dynamic and
2293
+ pervasive IoT,” Pervasive and Mobile Computing, vol. 42, pp. 299–316,
2294
+ 2017.
2295
+ [37] A. E. Al-Fagih, F. M. Al-Turjman, W. M. Alsalih, and H. S. Hassanein,
2296
+ “A priced public sensing framework for heterogeneous IoT architec-
2297
+ tures,” IEEE Transactions on Emerging Topics in Computing, vol. 1,
2298
+ no. 1, pp. 133–147, 2013.
2299
+ [38] B. Kantarci and H. T. Mouftah, “Trustworthy sensing for public safety
2300
+ in cloud-centric Internet of things,” IEEE Internet of Things Journal,
2301
+ vol. 1, no. 4, pp. 360–368, 2014.
2302
+ [39] “How much would you spend on a virtual reality headset?” Statista,
2303
+ 2015,
2304
+ [Online]
2305
+ Available:https://www.statista.com/statistics/457117/
2306
+ virtual-reality-headset-amount-willing-to-pay-in-the-united-states/.
2307
+ [40] R. B. Myerson, “Incentive compatibility and the bargaining problem,”
2308
+ Econometrica, pp. 61–73, 1979.
2309
+ [41] D. Roy, A. S. Rao, T. Alpcan, G. Das, and M. Palaniswami, “Achieving
2310
+ qos for bursty urllc applications over passive optical networks,” Journal
2311
+ of Optical Communications and Networking, vol. 14, no. 5, pp. 411–425,
2312
+ 2022.
2313
+ [42] S. Chen, L. Wang, and F. Liu, “Optimal admission control mechanism
2314
+ design for time-sensitive services in edge computing,” in IEEE Confer-
2315
+ ence on Computer Communications (INFOCOM), 2022, pp. 1169–1178.
2316
+ [43] S. Lakshminarayana, M. Assaad, and M. Debbah, “Transmit power
2317
+ minimization in small cell networks under time average qos constraints,”
2318
+ IEEE Journal on Selected Areas in Communications, vol. 33, no. 10, pp.
2319
+ 2087–2103, 2015.
2320
+ [44] D. E. Kirk, Optimal Control Theory: An Introduction.
2321
+ Courier
2322
+ Corporation, 2012.
2323
+
2324
+ [45] O. L. Mangasarian, “Sufficient conditions for the optimal control of
2325
+ nonlinear systems,” SIAM Journal on Control, vol. 4, no. 1, pp. 139–
2326
+ 152, 1966.
2327
+ [46] G. Fettweis and et al, “The Tactile Internet: ITU-T Technology
2328
+ Watch Report,” 2014, [Online] Available:https://www.itu.int/dms pub/
2329
+ itu-t/opb/gen/T-GEN-TWATCH-2014-1-PDF-E.pdf.
2330
+ [47] Cisco, “Video Quality of Service (QOS) Tutorial,” 2017, [Online] Avail-
2331
+ able:https://www.cisco.com/c/en/us/support/docs/quality-of-service-qos/
2332
+ qos-video/212134-Video-Quality-of-Service-QOS-Tutorial.html.
2333
+
J9E3T4oBgHgl3EQfvQsy/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
JtE1T4oBgHgl3EQfGQNW/content/tmp_files/2301.02911v1.pdf.txt ADDED
@@ -0,0 +1,821 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ TOWARDS EARLY PREDICTION OF NEURODEVELOPMENTAL
2
+ DISORDERS: COMPUTATIONAL MODEL FOR FACE TOUCH AND
3
+ SELF-ADAPTORS IN INFANTS
4
+ Bruno Tafur
5
+ University of Cambridge
6
+ United Kingdom
7
8
+ Marwa Mahmoud
9
+ University of Glasgow
10
+ United Kingdom
11
12
+ Staci Weiss
13
+ University of Cambridge
14
+ United Kingdom
15
16
+ ABSTRACT
17
+ Infants’ neurological development is heavily influenced by their motor skills. Evaluating a baby’s
18
+ movements is key to understanding possible risks of developmental disorders in their growth. Previous
19
+ research in psychology has shown that measuring specific movements or gestures such as face touches
20
+ in babies is essential to analyse how babies understand themselves and their context. This research
21
+ proposes the first automatic approach that detects face touches from video recordings by tracking
22
+ infants’ movements and gestures. The study uses a multimodal feature fusion approach mixing spatial
23
+ and temporal features and exploits skeleton tracking information to generate more than 170 aggregated
24
+ features of hand, face and body. This research proposes data-driven machine learning models for
25
+ the detection and classification of face touch in infants. We used cross dataset testing to evaluate
26
+ our proposed models. The models achieved 87.0% accuracy in detecting face touches and 71.4%
27
+ macro-average accuracy in detecting specific face touch locations with significant improvements over
28
+ Zero Rule and uniform random chance baselines. Moreover, we show that when we run our model to
29
+ extract face touch frequencies of a larger dataset, we can predict the development of fine motor skills
30
+ during the first 5 months after birth.
31
+ Keywords Computer Vision · Autoencoders · Neurodevelopment factors
32
+ 1
33
+ Introduction
34
+ Figure 1: An overview of our proposed framework. Spatial, temporal and appearance features are extracted, then they
35
+ are concatenated with a feature integration layer and a classification approach is used to detect and classify the infant’s
36
+ face touch, which is subsequently used to predict the neurodevelopmental scores.
37
+ Analysing body movements in early childhood gives insights into the infant’s neurological development, and it can play
38
+ an essential role in determining if a baby is suffering from injuries in the nervous system or a hereditary disease [1].
39
+ arXiv:2301.02911v1 [cs.CV] 7 Jan 2023
40
+
41
+ 777
42
+ Video
43
+ Features
44
+ Fusion Model
45
+ Output
46
+ Raw frames
47
+ Skeleton coordinates
48
+ Face touch
49
+ Geome-
50
+ Appea-
51
+ Wrists / Neck / Hips
52
+ trical
53
+ rance
54
+ Mouth / Cheek / Ears / Eyes /
55
+ ral
56
+ Features
57
+ features
58
+ Nose
59
+ No face touch
60
+ Face coordinates
61
+ Feature Integration
62
+ Ears / Eyes / Mouth / Nose
63
+ Classification
64
+ Temporal
65
+ Velocity Wrist
66
+ Neurodevelopmental
67
+ Velocity / Displacement
68
+ scores
69
+ - Feature interpolationTowards early prediction of neurodevelopmental disorders: Computational model for Face Touch and Self-adaptors in
70
+ Infants
71
+ Also, specific movements such as face touches have been found to be crucial in the development of babies since their
72
+ fetal age [2]. For example, face touches to sensitive areas of the face such as the mouth are frequent in gestational age
73
+ as babies get prepared to feed. Different cultures have also shown differences in self-touch in babies, and age has also
74
+ been established as a determinant factor [2].
75
+ Various methods are utilised to track and measure movements in babies, including 3D motion capture, sensors and
76
+ video cameras [3]. 3D motion capture and sensors are mostly used in laboratory settings instead of a more natural
77
+ setting for the infant as it requires specific equipment and tools [4]. There have been few studies that utilised computer
78
+ vision with video cameras. This method has the advantages of being highly flexible to different environments, giving
79
+ high contextual information and being easier to interpret [3]. However, it requires more complex computations, depends
80
+ on the camera quality, angle and movement, and is difficult to generalise to untrained cases.
81
+ Various studies have explored the fidgety movements of babies by analysing pixel displacement in video frames [5, 6, 7].
82
+ Recently, some studies have begun to examine more robust tracking algorithms for body parts based on methods such
83
+ as OpenPose [4, 8]. Despite being an important measure for neurodevelopment, no previous research has looked at
84
+ specific gesture detection in much detail, such as detecting specific hand movements. Most research is centred on
85
+ general fidgety movements or general statistics descriptors. Also, these approaches have primarily focused on a general
86
+ classification for high-risk infants or cerebral palsy [4, 9] or classification of movement types [5, 8, 10]. In the case of
87
+ face touches, research is limited and has centred on hand-over face gestures in adults [11] or touches from the mother to
88
+ a child [12, 13].
89
+ This research proposes a machine learning model for automatic detection and classification of face touches in newborn
90
+ infants and their location around key areas of the face using features extracted from raw videos. It proposes using
91
+ feature selection and fusion models based on temporal and spatial features, using geometric and appearance features of
92
+ the infant’s face and body. The proposed models are validated and evaluated on a couple of datasets and then applied to
93
+ a large video dataset to extract gesture descriptors of video of one-month-old infants automatically. Using regression,
94
+ we demonstrate the effectiveness of extracting these gestures in predicting Mullen neurodevelopmental scores [14] for
95
+ the same infants. To the best of our knowledge, this research represents the first study to analyse specific gestures in
96
+ infants at this level of granularity and the first to analyse self-touch. The main contributions can be described as follows:
97
+ • Proposing a data-driven machine learning model for detection and classification of face touch in infants
98
+ exploiting spatial and temporal features.
99
+ • Evaluating and validating the proposed method in a cross-dataset manner using challenging naturally collected
100
+ datasets on infants.
101
+ • Presenting preliminary results on using our proposed computational model to detect face touch features in
102
+ infants on a larger labelled dataset and demonstrating the ability to predict neurodevelopmental scores of the
103
+ infants using automatic face touch dynamics.
104
+ • Our proposed trained validated model is available on Github as an open source tool for the community. We
105
+ believe this work will enable future research in infant behaviour modelling and provide a tool for future
106
+ neurodevelopmental studies.
107
+ 2
108
+ Related Work
109
+ The most relevant studies that use computer vision in the context of neurodevelopment analysis in infants have centered
110
+ on tracking general movement indicators; such as aggregated data from pose coordinates [4, 8] or displacement
111
+ information from overall images [5, 6, 7]. In the analysis of touch, a couple of studies have centred on an analysis of a
112
+ controlled environment where a mother touches her child [12, 13].
113
+ Infants have different proportions in their limbs than adults, making them more complex for general tracking mechanisms
114
+ to work with the same accuracy. Therefore, the study carried out by Chambers et al. adapted tracking methods based on
115
+ computer vision on babies [4]. Their study focused on developing a tracking method for infants’ skeleton coordinates,
116
+ and they used these statistics to compare the risk of developing neuromotor impairment in healthy and at-risk infants.
117
+ Their study expanded on OpenPose [15] implementation for humans by tuning the model to be used on infant videos.
118
+ Regarding more specific movement patterns related to neurological disorders, a study by Das et al. [10] focused on
119
+ analysing the kicking patterns of at-risk infants. Their method tracked their movements using OpenPose and extracted
120
+ additional KAZE features [16] as image descriptors. In their experiments, the authors used an SVM classifier to
121
+ differentiate the kicking pattern types as simultaneous movement (SM), non-simultaneous movement (NSM) and no
122
+ movement (NM).
123
+ 2
124
+
125
+ Towards early prediction of neurodevelopmental disorders: Computational model for Face Touch and Self-adaptors in
126
+ Infants
127
+ In the case of touch, Chen et al. performed a couple of studies focused on the detection of interactions between a
128
+ caregiver and a child in a controlled environment [12, 13]. Specifically, their research focused on detecting touch from
129
+ the caregiver to the child in particular locations: head, arms, legs, hand, torso and feet. Their latest study applied two
130
+ main methods; firstly, they extracted tracking information by detecting the skeleton locations. Secondly, they extracted
131
+ the infant’s location in the image by applying image segmentation using the GrabCut algorithm.
132
+ Therefore, although some studies have attempted to tackle some of these issues in infants, most of them have focused
133
+ on the analysis of general movements instead of specific gestures. This study analyses hand to face gestures in infants
134
+ in larger detail and granularity and proposes novel machine learning models for automatic detection.
135
+ The relationship between face and body touch in infants and how they correlate with cognitive development has not
136
+ been studied quantitatively and systematically before in previous literature. The Mullen Scales of Early Learning is
137
+ used to measure the cognitive development of infants in five different categories: gross motor (GM), fine motor (FM),
138
+ visual reception (VR), receptive language (RL) and expressive language (EL) [14, 17]. They are a key measure of the
139
+ development of the child during the first years after birth. Previous studies have not tackled the relationship of detected
140
+ features with MSEL scores. We aim to analyse this relationship based on gesture and movement data extracted from the
141
+ infant.
142
+ 3
143
+ Datasets
144
+ For our data-driven models, we used two main datasets: BRIGHT [18] and Chambers [4]. A subset of the two datasets
145
+ was labelled and validated by a psychology expert to be later used for our models. Then, the videos from the BRIGHT
146
+ dataset were used to evaluate the correlations between face touch dynamics and neurodevelopmental scores.
147
+ 3.1
148
+ BRIGHT dataset
149
+ This dataset was provided by the ’removed for anonymous submission’ and is part of the studies carried out in the
150
+ Brain Imaging for Global Health (BRIGHT) Project [18] in which they study infants from Gambia and UK during their
151
+ first 24 months of life. The initial sample provided included 29 videos of UK infants. From the 29 videos, 23 videos
152
+ were selected as some of the babies were occluded during most of the video runtime. Each video shows the behaviour
153
+ of one infant of fewer than 2 months of age, actively responding to the input given by their mother. The videos were
154
+ recorded in different rooms, with the infant lying down with a mirror positioned on the wall behind the head of the baby.
155
+ The camera is static, and the infants generally cover a small portion of the frame but can be located in different parts
156
+ of the frame. Another complex factor that characterises this dataset includes the mother’s presence during the video,
157
+ sometimes occupying a significant part of the frame with a bigger skeleton and limbs. Also, the fact that the infants are
158
+ lying down while the camera is facing the front means that the camera generally captures the babies’ faces from a side
159
+ or the bottom, making them difficult to detect for traditional algorithms. The babies are shown rotated in the frames at
160
+ different angles between 90° and -90°.
161
+ 3.2
162
+ Chambers dataset
163
+ This is an open dataset compiled and generated by Chambers et al [4]. 25 videos were selected based on the age of the
164
+ infants in the video by filtering and selecting only the videos with babies less than 2 months to ensure better consistency
165
+ with the BRIGHT dataset. The videos show babies lying down on their own and interacting in a natural environment.
166
+ They could be dancing, playing or rolling over in their crib. The camera is sometimes moving while filming the baby,
167
+ and the babies generally cover most of the frame. The videos do not feature other people in the frame, but the babies
168
+ sometimes can move at different angles. Also, the resolutions are very varied between videos, with some of them being
169
+ more blurry and with smaller frames. The babies are shown rotated in the frames at different angles between 90° and
170
+ -90°.
171
+ 4
172
+ Labelling
173
+ The labelling process was carried out using a tagging system developed for this research which allowed efficient tagging
174
+ of the image frames. Also, the tagging was carried out with the support of a psychology expert, who helped by labelling
175
+ part of the dataset and providing her judgement about the different labelling categories.
176
+ As this study aims to detect hand over face gestures in infants automatically, the main labelling category to tag needed
177
+ to differentiate between face touch or no touch in each frame. Therefore, it was defined as follows:
178
+ 3
179
+
180
+ Towards early prediction of neurodevelopmental disorders: Computational model for Face Touch and Self-adaptors in
181
+ Infants
182
+ Table 1: Database sizes and labels
183
+ Dataset
184
+ Sizes
185
+ # Videos
186
+ 25
187
+ Total Frames
188
+ 1769
189
+ Chambers
190
+ Mean Frames per video
191
+ 70.76
192
+ % On Head
193
+ 29.2%
194
+ % Outside Head
195
+ 70.8%
196
+ # Videos
197
+ 23
198
+ Total Frames
199
+ 2039
200
+ BRIGHT
201
+ Mean Frames per video
202
+ 88.6
203
+ % On Head
204
+ 29.7%
205
+ % Outside Head
206
+ 70.3%
207
+ Total Number of Frames
208
+ 3808
209
+ • On Head: From a human perspective, it can be seen that the hand could be touching the head area. In this
210
+ study, the head area considers any of the following locations or any area enclosed by the those locations: eyes,
211
+ ears, nose, mouth, cheeks, forehead and neck.
212
+ • Outside Head: From a human perspective, it can be seen that the hand is not on the head area as defined.
213
+ Additionally, we labelled our dataset with the following non-exclusive categories: eyes, ears, nose, mouth and cheeks,
214
+ as they are the main differentiable parts of the face. The categories were also discussed and agreed upon with the
215
+ psychology expert to validate their significance and usefulness from the neurodevelopment perspective.
216
+ The final labelled datasets sizes and distributions can be seen in Table I. The final proportion of “on head” versus
217
+ “outside head” was of 29.5% to 70.5%, which is expected for this kind of natural dataset.
218
+ 5
219
+ METHOD
220
+ Because of the small size of the labelled dataset, we could not use an end-to-end deep learning model. In this section
221
+ we present the feature extraction and selection steps and the proposed feature fusion machine learning model.
222
+ 5.1
223
+ Feature extraction of face and body
224
+ Our proposed models required spatial and temporal features related to the infants’ face touch gestures. The features
225
+ extracted were selected considering the relationship between the hands of the baby and the face.
226
+ 5.1.1
227
+ Extraction of face and body landmarks
228
+ We first extracted basic face and body landmarks.
229
+ - Pose coordinates: Positions of the skeleton parts were extracted for every baby and every frame by using the fine-tuned
230
+ OpenPose [15] model trained by Chambers et al. [4]. Following the implementation of Chambers et al., the raw pose
231
+ locations were normalised, smoothed and interpolated per video.
232
+ - Face Region: Based on the extracted pose features and estimated orientation, an accurate estimate of the baby’s face
233
+ location was carried out and the image was cropped in the face region. If no possible face was found in a given frame,
234
+ the locations of the face of the nearest frames were used as guidance. Where possible, the face region was further
235
+ aligned based on the locations of the eyes and nose.
236
+ - Face coordinates: OpenPose provides general locations of the eyes, nose and ears, but its purpose is centred on getting
237
+ the whole skeleton and not on specific facial landmarks. Therefore, information about the location of facial features
238
+ based on 3D-FAN [19] was also used. The faces were extracted from the aligned cropped face regions.
239
+ 5.1.2
240
+ Extraction of geometric, appearance and temporal features
241
+ After basic landmarks features were extracted, we extracted a set of geometric and temporal feature descriptors.
242
+ Based on the initial features, the following features were calculated:
243
+ 4
244
+
245
+ Towards early prediction of neurodevelopmental disorders: Computational model for Face Touch and Self-adaptors in
246
+ Infants
247
+ Figure 2: Example body skeleton extracted using OpenPose and face keypoints extracted using 3D-FAN
248
+ - Face and body geometrical features (Distance and Angular): Based on the coordinates of the skeleton of the baby, the
249
+ normalised distances between the wrists and the ears, eyes, neck and nose were extracted. For each case, the distances
250
+ considered included differences in the X direction, differences in the Y direction and euclidean distances. Additionally,
251
+ based on the coordinates of the skeleton of the baby, the angles of the elbows and shoulders were extracted.
252
+ - Hands geometrical features (Distance): As the adapted OpenPose model by Chambers et al. [4] only generated the
253
+ skeleton up to the wrists, additional information was obtained by extending the skeleton to the hands. The MediaPipe
254
+ detection algorithm [20] was used in the area surrounding the wrists to obtain the hand coordinates. Based on the
255
+ coordinates, the normalised distances between the fingers and the eyes and nose were calculated. The distances included
256
+ differences in the X direction, differences in the Y direction and euclidean distances. Also, confidence scores were
257
+ considered as additional features based on the confidence of the MediaPipe algorithm detecting each hand.
258
+ - Temporal features: The temporal features were centred on aggregated information over various frames. We calculated
259
+ features including displacement, speed and acceleration obtained based on the coordinates of the skeleton of the baby
260
+ for the wrists and elbows.
261
+ - Appearance Features: Histogram of Oriented Gradients (HOG) [21] is a method for feature extraction based on
262
+ the directionality of the gradients in different locations in an image. This method has shown significant success
263
+ rate in different image detection tasks including detecting faces and expressions [22, 23, 24] and detecting gestures
264
+ [25, 26, 27]. These features were extracted only for the main region of interest, which is the face area. Consequently,
265
+ these features were extracted from the cropped images of the face. Additionally, we wanted to extract more localised
266
+ spatial information inside the face. Therefore, more granular HOG features were extracted in two specific face areas:
267
+ one related to the upper region of the face based on the eyes location and another related to the lower region based on
268
+ the mouth location. Also, confidence scores were considered as additional features based on the average confidence of
269
+ the landmarks in each region as calculated by 3D-FAN.
270
+ Figure 3: Examples of HOG features obtained for the face region, the upper head and the lower head. Note the
271
+ challenging nature of the dataset with extreme head poses and viewpoints.
272
+ 5.1.3
273
+ Features smoothing and data augmentation
274
+ As a final step in the feature extraction process, we smoothed and augmented the calculated features to be able to train
275
+ the classification models on the data. The outliers of the geometrical and temporal features per video were replaced by
276
+ blank values, and the data was interpolated per video to cover any deleted or missing values. If data was still missing, it
277
+ was replaced by mean values from the training data during the training stage.
278
+ Finally, to compensate for the small size of the dataset, the training data was augmented by flipping the images
279
+ horizontally, flipping all the features accordingly and considering the directionality of these features.
280
+ 5.2
281
+ Face touch detection and classification
282
+ After feature extraction and smoothing, we handled face touch detection and classification as two different classification
283
+ problems. The first is to detect when the hand touches the face as a binary classification problem; then, we classify
284
+ different touch location areas as a multi-label classification problem. The architectures proposed for both problems are
285
+ 5
286
+
287
+ 100
288
+ 200
289
+ 300
290
+ 400
291
+ 100
292
+ 200
293
+ 300
294
+ 400Input image
295
+ Input image
296
+ Input image
297
+ Histogram of Oriented Gradients
298
+ 10
299
+ 20
300
+ 20
301
+ 40
302
+ 60
303
+ 20
304
+ 100
305
+ 120
306
+ 20
307
+ 80
308
+ 100
309
+ 120
310
+ 40
311
+ 80
312
+ 60
313
+ Histogram of Oriented Gradients
314
+ Histogram of Oriented Gradients
315
+ 00
316
+ 20
317
+ 60
318
+ 20
319
+ 80
320
+ 100
321
+ 120
322
+ 20
323
+ 40
324
+ 60
325
+ 80
326
+ 100
327
+ 120
328
+ 20
329
+ 40
330
+ 60
331
+ 80
332
+ 100
333
+ 120
334
+ 20
335
+ 40
336
+ 60
337
+ 80
338
+ 100
339
+ 120Towards early prediction of neurodevelopmental disorders: Computational model for Face Touch and Self-adaptors in
340
+ Infants
341
+ very similar. The main difference lies in the method used for the classification component in the final layer. The models
342
+ can be divided into the following:
343
+ 5.2.1
344
+ Feature selection and dimensionality reduction
345
+ Our first proposed method used feature selection and dimensionality reduction and a Support Vector Machine (SVM)
346
+ classifier for solving the face touch detection problem.
347
+ There were four main categories of geometrical and temporal features: body distance features, hand distance features,
348
+ angular features and temporal features. Many of the features in the same categories correlated with each other as they
349
+ measured similar characteristics. Therefore, to ensure proper representation of the features, this method proposed
350
+ reducing these features before training a classifier.
351
+ Firstly, the features were filtered based on an automatic feature selection process. Random Forest was used to select
352
+ the most representative features and prevent skewing the classifier with features that were not that significant. The
353
+ feature selection was carried out by cross-validating with 5-folds in the training set to ensure independence. Then,
354
+ Principal Component Analysis (PCA) was used for dimensionality reduction. PCA has shown effective results in
355
+ detection problems when facing a large number of features [28, 29, 30]. PCA was used to filter a percentage of the
356
+ explained variance. The threshold of this explained variance was established as a hyperparameter that was also learned
357
+ by cross-validating with 5-folds in the training set.
358
+ After applying PCA, the classification algorithm used SVM using an RBF kernel. The model was cross-validated
359
+ with 5-folds in the training set to choose the best hyperparameters for SVM. The search for the best hyperparameters
360
+ for SVM was done in combination with the search for the threshold for PCA, as the hyperparameters were possibly
361
+ dependent on each other using a grid search method [31].
362
+ In the case of multi-class classification for the face areas, the Label Powerset model was used with underlying SVMs
363
+ to predict the multiple overlapping labels. Label Powerset transforms the labels by creating a class for each possible
364
+ combination of labels and creates a classifier for each combination [32]. Consequently, it has the advantage of
365
+ considering the possible relationships between the labels. This model was configured by tuning the hyperparameters in
366
+ the same way as in the SVM binary classifier.
367
+ 5.2.2
368
+ Feature optimisation using deep learned features
369
+ Our second proposed method used autoencoders as a feature optimisation and dimensionality reduction technique.
370
+ This method has been used in various studies as an effective way of reducing dimensionality while maintaining the
371
+ representation of the data, and it has been successfully used before with repetitive and correlated features [33]. In this
372
+ case, autoencoders were used to generate a latent representation of the input features.
373
+ Firstly, the dimensions of the features were reduced based on the autoencoder model. The model uses a neural network
374
+ architecture that learns how to represent the data in lower dimensions and reconstruct it [33]. It then minimises the error
375
+ between the reconstruction and the original input. The aim of the autoencoder is to exploit the correlations in the input
376
+ features to reduce the final dimensions without losing relevant information.
377
+ This method was used with two alternatives of input features. The first one used only geometrical and temporal features.
378
+ The second alternative also used the HOG features. The main hyperparameters that were learned for this model included
379
+ the latent dimensions and the number of epochs. These hyperparameters were selected based on the results of a 5-fold
380
+ cross-validation in the training set.
381
+ After encoding the data, the classification process was done using SVM with an RBF kernel. The input features for the
382
+ SVM classification process were the output of encoding the features with the trained encoder. The classification layer
383
+ was also cross-validated with 5-folds in the training set to choose the best hyperparameters for SVM. Finally, in the
384
+ case of the multi-label classification problem, the Label Powerset model was used with underlying SVMs to be able to
385
+ predict the multiple face touch locations.
386
+ 6
387
+ EVALUATION
388
+ We evaluated the accuracy of the detection of face touches by using a mixture of spatial and temporal features
389
+ and analysed models based on dimensionality reduction and optimisation techniques. The models were evaluated
390
+ cross-dataset to validate their effectiveness and generalisation. The approaches were evaluated with three different
391
+ configurations of the datasets to ensure the consistency of the models. Also, all segmentations of the data were grouped
392
+ by video to ensure having different videos in each set. The three configurations used were the following:
393
+ 6
394
+
395
+ Towards early prediction of neurodevelopmental disorders: Computational model for Face Touch and Self-adaptors in
396
+ Infants
397
+ • Train and Cross Validate on BRIGHT dataset - Test on Chambers dataset
398
+ • Train and Cross Validate on Chambers dataset - Test on BRIGHT dataset
399
+ • Train and Cross Validate on Chambers and on 50% of BRIGHT dataset - Test on the other 50% of BRIGHT
400
+ dataset
401
+ As there was no existing baseline for these models, the models were evaluated against Zero Rule (ZeroR) baseline and
402
+ random uniform chance. In the case of the ZeroR baseline, it is calculated by assigning the value of the majority class
403
+ to every data point [34, 35, 36] while random chance assigns a class based on random uniform probabilities. Statistical
404
+ McNemar’s tests were carried out to ensure the results were significantly different. The McNemar test was used as the
405
+ compared distributions were binary targets instead of continuous variables. All the best performing models were found
406
+ significantly different with p < 0.01 in comparison to Random Chance and Zero Rule.
407
+ 6.1
408
+ Detection of face touches
409
+ The main target was to determine if there was a face touch. This problem was treated as a binary classification task
410
+ based on the classes: “on head” and “outside head”.
411
+ The models that were analysed were the following:
412
+ • Feature selection and dimensionality reduction based on geometrical and temporal features (RF-PCA-SVM):
413
+ This model followed the components described in Section 5.2.1. It performed feature selection with Random
414
+ Forest (RF) and dimensionality reduction with PCA. Finally, it performed the classification of the labels using
415
+ SVM in the case of this binary problem. It used the geometrical distance and angular features and aggregated
416
+ temporal features.
417
+ • Feature optimisation using deep learned features based on geometrical and temporal features (AUTO ENC-
418
+ SVM-I): This model was structured as described in Section 5.2.2. It used an autoencoder neural architecture to
419
+ reduce the dimensions of the input features. Finally, it performed the classification of the labels using SVM. It
420
+ used the geometrical features (distance and angular) and the temporal features.
421
+ • Feature optimisation using deep learned features based on geometrical, temporal and HOG features
422
+ (AUTOENC-SVM-II): The model was structured as described in Section 5.2.2. Similar to the previous
423
+ model, it used an autoencoder neural architecture to reduce the dimensions of the input features and SVM for
424
+ the binary classification problem. It used the geometrical features, temporal features and HOG features.
425
+ The results for predicting between “on head” and “outside head” can be seen in Table 2, 3 and 4. All the models had
426
+ significantly higher accuracy than uniform random chance and ZeroR baselines. The best performing model reached
427
+ 87% accuracy when trained in a mixture of both datasets.
428
+ Overall the results of the three models were promising with high accuracy in comparison to the baselines. Also,
429
+ the results were relatively similar between the three models. Some performed better on different datasets, but the
430
+ performance was very competitive between them. All three models obtained better results than ZeroR or Random
431
+ Chance in accuracy, precision and recall. Therefore, the results demonstrated that these models can perform well in the
432
+ detection of face touch.
433
+ Even though the autoencoder models (AUTOENC-SVM) outperformed the random forest and PCA model (RF-PCA-
434
+ SVM) in two of the three dataset configurations, the difference in accuracy performance was limited. These results
435
+ demonstrate that the RF-PCA-SVM configuration was also very effective. Possibly in larger datasets, the autoencoder
436
+ based models could extract more representative features that could better outperform the RF-PCA-SVM model.
437
+ Similarly, the inclusion of the HOG features in the AUTOENC-SVM-II model did not show a noticeable increase in
438
+ performance. In the case of the BRIGHT dataset, it did show an improvement over the other models and a higher
439
+ improvement over AUTOENC-SVM-I. However, the improvement could have been greater. This could be caused
440
+ by the limited amount of data with very varied head poses and rotations. Therefore, the AUTOENC-SVM-II model
441
+ might perform better if trained in larger datasets where the HOG features can be learned with more generalisable
442
+ representations.
443
+ Finally, even though there were various challenges in the datasets that could have a negative impact on the models’
444
+ ability to generalise between datasets, the results demonstrated that the proposed methods had high performance in the
445
+ detection of face touches.
446
+ 7
447
+
448
+ Towards early prediction of neurodevelopmental disorders: Computational model for Face Touch and Self-adaptors in
449
+ Infants
450
+ Table 2: Results -binary classification “on head” vs “outside head”.
451
+ Training and CV dataset: Chambers. Testing dataset: Bright.
452
+ Model
453
+ Accuracy
454
+ Precision
455
+ Recall
456
+ Test
457
+ On Head
458
+ On Head
459
+ Random Chance
460
+ 50%
461
+ 29.7%
462
+ 50%
463
+ Zero Rule
464
+ 70.3%
465
+ 0%
466
+ 0%
467
+ RF-PCA-SVM
468
+ 80.3%
469
+ 68.7%
470
+ 62.2%
471
+ AUTOENC-SVM-I
472
+ 80.7%
473
+ 70.8%
474
+ 59.6%
475
+ AUTOENC-SVM-II
476
+ 80.6%
477
+ 74.4%
478
+ 53.1%
479
+ Table 3: Results -binary classification “on head” vs “outside head”.
480
+ Training and CV dataset: Bright. Testing dataset: Chambers.
481
+ Model
482
+ Accuracy
483
+ Precision
484
+ Recall
485
+ Test
486
+ On Head
487
+ On Head
488
+ Random Chance
489
+ 50%
490
+ 29.2%
491
+ 50%
492
+ Zero Rule
493
+ 70.8%
494
+ 0%
495
+ 0%
496
+ RF-PCA-SVM
497
+ 77.8%
498
+ 58.8%
499
+ 80.2%
500
+ AUTOENC-SVM-I
501
+ 75.2%
502
+ 57.9%
503
+ 54.8%
504
+ AUTOENC-SVM-II
505
+ 79.6%
506
+ 65.4%
507
+ 63.8%
508
+ 6.2
509
+ Classification of face touch descriptors
510
+ These experiments evaluate the face touch on specific locations of the face. These locations were evaluated based on the
511
+ universe of images where there is a face touch. The key locations to predict included the following: ears, nose, cheeks,
512
+ mouth, and eyes. The problem mas evaluated as a multi-label problem because the different classes could overlap and
513
+ the infant could touch more than one location at the same time.
514
+ The proposed models for this problem are the same as the ones described in Section 6.1, so we will use the same naming
515
+ abbreviations. The main difference was the change in the classification method from SVM to Label Powerset with SVM
516
+ [32] to tackle the problem as a multi-label classification problem. Therefore, the models that were analysed were the
517
+ following:
518
+ • Feature selection and dimensionality reduction based on geometrical and temporal features (RF-PCA-SVM)
519
+ • Feature optimisation using deep learned features based on geometrical and temporal features (AUTOENC-
520
+ SVM-I)
521
+ • Feature optimisation using deep learned features based on geometrical, temporal and HOG features
522
+ (AUTOENC-SVM-II)
523
+ The experiments were carried out only on the portion of images labelled as “on head” so that it could be sufficiently
524
+ balanced; therefore the dataset was even more limited in size than the original.
525
+ The obtained results can be seen in Table 5, 6 and 7. The results show the macro-average accuracy, precision and recall
526
+ of the multiple key locations per model. The highest performing model reached 71.4% average accuracy when testing
527
+ on the Chamber’s dataset.
528
+ Table 4: Results -binary classification “on head” vs “outside head”.
529
+ Training and CV dataset: Chambers + 50% Bright. Testing dataset: 50% Bright.
530
+ Model
531
+ Accuracy
532
+ Precision
533
+ Recall
534
+ Test
535
+ On Head
536
+ On Head
537
+ Random Chance
538
+ 50%
539
+ 28.3%
540
+ 50%
541
+ Zero Rule
542
+ 71.7%
543
+ 0%
544
+ 0%
545
+ RF-PCA-SVM
546
+ 87.0%
547
+ 77.2%
548
+ 76.8%
549
+ AUTOENC-SVM-I
550
+ 86.9%
551
+ 76.9%
552
+ 77.1%
553
+ AUTOENC-SVM-II
554
+ 85.7%
555
+ 71.6%
556
+ 82.2%
557
+ 8
558
+
559
+ Towards early prediction of neurodevelopmental disorders: Computational model for Face Touch and Self-adaptors in
560
+ Infants
561
+ Table 5: Results of predicting key areasTraining and CV dataset: Chambers. Testing dataset: Bright.
562
+ Model
563
+ Accuracy
564
+ Precision
565
+ Recall
566
+ Test
567
+ Key Area
568
+ Key Area
569
+ Random Chance
570
+ 50.0%
571
+ 31.1%
572
+ 50.0%
573
+ Zero Rule
574
+ 24.5%
575
+ 13%
576
+ 0%
577
+ RF-PCA-SVM
578
+ 66.6%
579
+ 33.5%
580
+ 43.8%
581
+ AUTOENC-SVM-I
582
+ 63.2%
583
+ 49.6%
584
+ 16.7%
585
+ AUTOENC-SVM-II
586
+ 63.0%
587
+ 60.9%
588
+ 13.2%
589
+ Table 6: Results of predicting key areasTraining and CV dataset: Bright. Testing dataset: Chambers.
590
+ Model
591
+ Accuracy
592
+ Precision
593
+ Recall
594
+ Test
595
+ Key Area
596
+ Key Area
597
+ Random Chance
598
+ 50.0%
599
+ 20.8%
600
+ 50%
601
+ Zero Rule
602
+ 37.8%
603
+ 15%
604
+ 0%
605
+ RF-PCA-SVM
606
+ 62.5%
607
+ 36.3%
608
+ 18.8%
609
+ AUTOENC-SVM-I
610
+ 63.8%
611
+ 29.7%
612
+ 34.3%
613
+ AUTOENC-SVM-II
614
+ 71.4%
615
+ 35.7%
616
+ 24.9%
617
+ The main metric established to select the best models during cross-validation was the average macro-accuracy of the
618
+ locations; so this was used as the main indicator of the models performance. The results demonstrated that the models
619
+ performed effectively better than the baselines.
620
+ As expected, the accuracies were lower than for the face touch problem as this was a complex multi-label problem
621
+ where multiple locations can overlap on the face, and it can be difficult even for a human to determine the exact location.
622
+ Likewise to the previous task, the results were similar between models, but these results show some indication that
623
+ HOG features might be useful in some instances. The AUTOENCODER-SVM-II model outperformed the accuracies
624
+ of the other models in two cases and demonstrated a considerable difference in accuracy when it was trained in the
625
+ BRIGHT dataset. Possibly training with HOG features in more extensive and more varied datasets could make their
626
+ representations more stable and significant in the end results.
627
+ 6.3
628
+ Predicting neurodevelopment scores
629
+ The next step was to evaluate our proposed model results - detected face touch dynamics of infants less than 2 months
630
+ old - on predicting their neurodevelopmental rates collected at ages 3 and 5 months. We chose the best-performing
631
+ model for the binary classification task (RF-PCA-SVM) and ran it on a larger dataset. Since we had the Mullen scores
632
+ only for the BRIGHT dataset, we ran our model on an average of 490 frames per video (19 videos), a total of 9298
633
+ frames. We then extracted the face touch frequency for each infant and evaluated it versus the Mullen Scales of Early
634
+ Learning (MSEL) related to gross motor (GM) skills and fine motor (FM) skills. In this case, the data was limited
635
+ because the provided metrics were evaluated per infant, and only 19 infants of the BRIGHT dataset had their information
636
+ available.
637
+ In the case of the MSEL metrics, the data consisted of raw scores per visit of the infant related to the different MSEL
638
+ categories. A rate of development was calculated per infant per category based on the rate of increase during their first
639
+ five months. The data used for this case were the gross motor (GM) skills and the fine motor (FM) skills, as they are
640
+ related to the infant’s motor development and could be related to face touch behaviour. After calculating the rate of
641
+ Table 7: Results of predicting key areas
642
+ Training and cross-validation dataset: Chambers + 50% Bright. Testing dataset: 50% Bright.
643
+ Model
644
+ Accuracy
645
+ Precision
646
+ Recall
647
+ Test
648
+ Key Area
649
+ Key Area
650
+ Random Chance
651
+ 50.0%
652
+ 29.1%
653
+ 50.0%
654
+ Zero Rule
655
+ 21.8%
656
+ 17.0%
657
+ 0%
658
+ RF-PCA-SVM
659
+ 60.3%
660
+ 56.0%
661
+ 34.3%
662
+ AUTOENC-SVM-I
663
+ 58.1%
664
+ 48.2%
665
+ 19.2%
666
+ AUTOENC-SVM-II
667
+ 60.7%
668
+ 44.3%
669
+ 16.7%
670
+ 9
671
+
672
+ Towards early prediction of neurodevelopmental disorders: Computational model for Face Touch and Self-adaptors in
673
+ Infants
674
+ development of the GM and FM skills, a correlation was calculated between the ratio of face touches per frame and
675
+ these rates of development of each child.
676
+ The results showed a low to moderate positive correlation between the ratio of face touches and the rate of development
677
+ during the first months. The correlation coefficient obtained for FM was 0.599 with a significant p-value of 0.0067. The
678
+ correlation coefficient for GM was 0.186, but the p-value was not found to be significant. It is possible that face touches
679
+ are more related to fine motor skills as they are more specific and localised movements.
680
+ The results indicate that measuring infants’ face touch frequencies and dynamics in their first month or two can
681
+ be a predictive measure of their neurodevelopmental scores. It also demonstrates the effectiveness of our proposed
682
+ computational model as a tool for the early prediction of neurodevelopmental factors. We make the trained model
683
+ available to the research community at ’removed for anonymous submission’ as a baseline for infant’s face touch
684
+ detection and to facilitate future research in this area on more extensive datasets.
685
+ A dataset of 19 infants is limited, so it was not possible to test more complex prediction algorithms. However, these
686
+ results show that, in more extensive datasets, the face touch frequency could be used as one independent variable to help
687
+ predict infant neurodevelopment scores such as MSEL. Also, the models proposed during this research could support
688
+ the automation of the extraction of these face touches.
689
+ 7
690
+ Conclusion
691
+ Our research proposed a machine learning model for automatic detection of face touches in infants using features
692
+ extracted from videos. This is the first study to provide a computational model for detection and classification of these
693
+ types of gestures in infants. Our proposed models using a mix of spatial and temporal features with deep learning
694
+ features demonstrated significantly high accuracies in predicting face touch and their locations around keypoints in the
695
+ face establishing a promising step for future research in this area. We also showed the effectiveness of the proposed
696
+ model in predicting MSEL scores related to fine motor (FM) skills , demonstrating that our proposed model can be used
697
+ as en early prediction tool for neurodevelopmental disorders in infants and it is considered a baseline for future work in
698
+ this domain. We believe this research will open the door for future research in this area both on the technical as well as
699
+ neurodevelopmetanl psychology fronts.
700
+ Despite the promising results, there are several limitations to our model. The datasets used were recorded in almost
701
+ uncontrolled environments, with varied camera angles and the mum’s presence in most videos. These characteristics
702
+ made the labelling as well as the classification tasks very challenging. We are also aware of the small size of the datasets
703
+ used in this work. Obtaining datasets, especially for infants, is a challenging task due to the privacy and ethical factors
704
+ that need to be considered. However, we believe this research serves as a baseline for infant face touch detection and
705
+ classification and will open the door for further research on more extensive datasets in this area.
706
+ References
707
+ [1] Nihan Hande Akcakaya, Turgay Altunalan, Tuba Derya Do˘gan, Arzu Yılmaz, and Zuhal Yapıcı. Correlation of
708
+ Prechtl Qualitative Assessment of General Movement Analysis with Neurological Evaluation: The Importance of
709
+ Inspection in Infants. Turkish Journal Of Neurology, 25(2):63–70, June 2019.
710
+ [2] Nadja Reissland and Joe Austen. Goal directed behaviours : the development of pre-natal touch behaviours. In
711
+ Daniela Corbetta and Marco Santello, editors, Reach-to-grasp behavior: brain, behavior, and modelling across
712
+ the life span., pages 3–17. Routledge, Abingdon, Oxon, August 2018.
713
+ [3] Claire Marcroft, Aftab Khan, Nicholas D. Embleton, Michael Trenell, and Thomas Plötz. Movement Recognition
714
+ Technology as a Method of Assessing Spontaneous General Movements in High Risk Infants. Frontiers in
715
+ Neurology, 5:284, January 2015.
716
+ [4] Claire Chambers, Nidhi Seethapathi, Rachit Saluja, Helen Loeb, Samuel R. Pierce, Daniel K. Bogen, Laura
717
+ Prosser, Michelle J. Johnson, and Konrad P. Kording. Computer Vision to Automatically Assess Infant Neuromotor
718
+ Risk. IEEE transactions on neural systems and rehabilitation engineering: a publication of the IEEE Engineering
719
+ in Medicine and Biology Society, 28(11):2431–2442, November 2020.
720
+ [5] Lars Adde, Jorunn L. Helbostad, Alexander Refsum Jensenius, Gunnar Taraldsen, and Ragnhild Støen. Using
721
+ computer-based video analysis in the study of fidgety movements. Early Human Development, 85(9):541–547,
722
+ September 2009.
723
+ 10
724
+
725
+ Towards early prediction of neurodevelopmental disorders: Computational model for Face Touch and Self-adaptors in
726
+ Infants
727
+ [6] Lars Adde, Jorunn Helbostad, Alexander R. Jensenius, Mette Langaas, and Ragnhild Støen. Identification of
728
+ fidgety movements and prediction of CP by the use of computer-based video analysis is more accurate when based
729
+ on two video recordings. Physiotherapy Theory and Practice, 29(6):469–475, August 2013.
730
+ [7] Lars Adde, Jorunn L Helbostad, Alexander R Jensenius, Gunnar Taraldsen, Kristine H Grunewaldt, and Ragnhild
731
+ Støen. Early prediction of cerebral palsy by computer-based video analysis of general movements: a feasibility
732
+ study. Developmental Medicine & Child Neurology, 52(8):773–778, 2010.
733
+ [8] Simon Reich, Dajie Zhang, Tomas Kulvicius, Sven Bölte, Karin Nielsen-Saines, Florian B. Pokorny, Robert
734
+ Peharz, Luise Poustka, Florentin Wörgötter, Christa Einspieler, and Peter B. Marschik. Novel AI driven approach
735
+ to classify infant motor functions. Scientific Reports, 11(1):9888, May 2021.
736
+ [9] Ragnhild Støen, Nils Thomas Songstad, Inger Elisabeth Silberg, Toril Fjørtoft, Alexander Refsum Jensenius,
737
+ and Lars Adde. Computer-based video analysis identifies infants with absence of fidgety movements. Pediatric
738
+ Research, 82(4):665–670, October 2017.
739
+ [10] Devleena Das, Katelyn Fry, and Ayanna M Howard. Vision-Based Detection of Simultaneous Kicking for
740
+ Identifying Movement Characteristics of Infants At-Risk for Neuro-Disorders. In 2018 17th IEEE International
741
+ Conference on Machine Learning and Applications (ICMLA), pages 1413–1418, December 2018.
742
+ [11] Marwa Mahmoud and Peter Robinson. Interpreting Hand-Over-Face Gestures. In Sidney D’Mello, Arthur
743
+ Graesser, Björn Schuller, and Jean-Claude Martin, editors, Affective Computing and Intelligent Interaction,
744
+ Lecture Notes in Computer Science, pages 248–255, Berlin, Heidelberg, 2011. Springer.
745
+ [12] Qingshuang Chen, Rana Abu-Zhaya, Amanda Seidl, and Fengqing Zhu. CNN Based Touch Interaction Detection
746
+ for Infant Speech Development. In 2019 IEEE Conference on Multimedia Information Processing and Retrieval
747
+ (MIPR), pages 20–25, March 2019.
748
+ [13] Qingshuang Chen, He Li, Rana Abu-Zhaya, Amanda Seidl, Fengqing Zhu, and Edward J Delp. Touch Event
749
+ Recognition For Human Interaction. Electronic Imaging, 2016(11):1–6, February 2016.
750
+ [14] Eileen M Mullen and American Guidance Service. Mullen Scales of Early Learning. AGS, Circle Pines, Minnesota,
751
+ 1995. OCLC: 860820407.
752
+ [15] Z. Cao, G. Hidalgo Martinez, T. Simon, S. Wei, and Y. A. Sheikh. OpenPose: Realtime Multi-Person 2D Pose
753
+ Estimation using Part Affinity Fields. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019.
754
+ [16] Pablo Fernández Alcantarilla, Adrien Bartoli, and Andrew J. Davison. KAZE Features. In Andrew Fitzgibbon,
755
+ Svetlana Lazebnik, Pietro Perona, Yoichi Sato, and Cordelia Schmid, editors, Computer Vision – ECCV 2012,
756
+ Lecture Notes in Computer Science, pages 214–227, Berlin, Heidelberg, 2012. Springer.
757
+ [17] Bosiljka Milosavljevic, Perijne Vellekoop, Helen Maris, Drew Halliday, Saikou Drammeh, Lamin Sanyang,
758
+ Momodou K. Darboe, Clare Elwell, Sophie E. Moore, and Sarah Lloyd-Fox. Adaptation of the Mullen Scales of
759
+ Early Learning for use among infants aged 5- to 24-months in rural Gambia. Developmental Science, 22(5):e12808,
760
+ 2019. _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/desc.12808.
761
+ [18] Global fNIRS. The bright project. https://www.globalfnirs.org/the-bright-project/.
762
+ [19] Adrian Bulat and Georgios Tzimiropoulos. How far are we from solving the 2d & 3d face alignment problem?
763
+ (and a dataset of 230,000 3d facial landmarks). In International Conference on Computer Vision, 2017.
764
+ [20] Fan Zhang, Valentin Bazarevsky, Andrey Vakunov, Andrei Tkachenka, George Sung, Chuo-Ling Chang,
765
+ and Matthias Grundmann.
766
+ MediaPipe Hands: On-device Real-time Hand Tracking.
767
+ Technical Report
768
+ arXiv:2006.10214, arXiv, June 2020. arXiv:2006.10214 [cs] type: article.
769
+ [21] N. Dalal and B. Triggs. Histograms of oriented gradients for human detection. In 2005 IEEE Computer Society
770
+ Conference on Computer Vision and Pattern Recognition (CVPR’05), volume 1, pages 886–893 vol. 1, June 2005.
771
+ ISSN: 1063-6919.
772
+ [22] Derry Alamsyah and Muhammad Fachrurrozi. Happy and Sad Classification using HOG Feature Descriptor in
773
+ SVM Model Selection. In 2021 International Conference on Informatics, Multimedia, Cyber and Information
774
+ System (ICIMCIS, pages 74–79, October 2021.
775
+ [23] Mustafa Keman Binli, Burak Can Demiryilmaz, Pinar O˘guz Ekim, and Faezeh Yeganli. Face Detection via HOG
776
+ and GA Feature Selection with Support Vector Machines. In 2019 11th International Conference on Electrical
777
+ and Electronics Engineering (ELECO), pages 610–613, November 2019.
778
+ [24] J.A. Mahajan and A. N. Paithane. Face detection on distorted images by using quality HOG features. In 2017
779
+ International Conference on Inventive Communication and Computational Technologies (ICICCT), pages 439–444,
780
+ March 2017.
781
+ 11
782
+
783
+ Towards early prediction of neurodevelopmental disorders: Computational model for Face Touch and Self-adaptors in
784
+ Infants
785
+ [25] Lukas Prasuhn, Yuji Oyamada, Yoshihiko Mochizuki, and Hiroshi Ishikawa. A HOG-based hand gesture
786
+ recognition system on a mobile device. In 2014 IEEE International Conference on Image Processing (ICIP),
787
+ pages 3973–3977, October 2014. ISSN: 2381-8549.
788
+ [26] Kevin Nathanael Krisandria, Bima Sena Bayu Dewantara, and Dadet Pramadihanto. HOG-based Hand Gesture
789
+ Recognition Using Kinect. In 2019 International Electronics Symposium (IES), pages 254–259, September 2019.
790
+ [27] Kai-ping Feng and Fang Yuan. Static hand gesture recognition based on HOG characters and support vector
791
+ machines. In 2013 2nd International Symposium on Instrumentation and Measurement, Sensor Network and
792
+ Automation (IMSNA), pages 936–938, December 2013.
793
+ [28] Nasser H. Dardas and Emil M. Petriu. Hand gesture detection and recognition using principal component analysis.
794
+ In 2011 IEEE International Conference on Computational Intelligence for Measurement Systems and Applications
795
+ (CIMSA) Proceedings, pages 1–6, September 2011. ISSN: 2159-1555.
796
+ [29] Johanna Degen, Jan Modersitzki, and Mattias P. Heinrich. Dimensionality reduction of medical image descriptors
797
+ for multimodal image registration. Current Directions in Biomedical Engineering, 1(1):201–205, September 2015.
798
+ Publisher: De Gruyter.
799
+ [30] Jinxian Qi, Guozhang Jiang, Gongfa Li, Ying Sun, and Bo Tao. Surface EMG hand gesture recognition system
800
+ based on PCA and GRNN. Neural Computing and Applications, 32(10):6343–6351, May 2020.
801
+ [31] Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel,
802
+ Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, Jake Vanderplas, Alexandre Passos, David
803
+ Cournapeau, Matthieu Brucher, Matthieu Perrot, and Édouard Duchesnay. Scikit-learn: Machine Learning in
804
+ Python. The Journal of Machine Learning Research, 12(null):2825–2830, November 2011.
805
+ [32] Jesse Read, Antti Puurula, and Albert Bifet.
806
+ Multi-label Classification with Meta-Labels.
807
+ In 2014 IEEE
808
+ International Conference on Data Mining, pages 941–946, December 2014. ISSN: 2374-8486.
809
+ [33] Yasi Wang, Hongxun Yao, and Sicheng Zhao. Auto-encoder based dimensionality reduction. Neurocomputing,
810
+ 184:232–242, April 2016.
811
+ [34] Alex G. C. de Sá, Cristiano G. Pimenta, Gisele L. Pappa, and Alex A. Freitas. A robust experimental evaluation of
812
+ automated multi-label classification methods. In Proceedings of the 2020 Genetic and Evolutionary Computation
813
+ Conference, GECCO ’20, pages 175–183, New York, NY, USA, June 2020. Association for Computing Machinery.
814
+ [35] Marwa Mahmoud, Louis-Philippe Morency, and Peter Robinson. Automatic multimodal descriptors of rhythmic
815
+ body movement. In Proceedings of the 15th ACM on International conference on multimodal interaction, ICMI
816
+ ’13, pages 429–436, New York, NY, USA, December 2013. Association for Computing Machinery.
817
+ [36] Mert Bal, M. Fatih Amasyali, Hayri Sever, Guven Kose, and Ayse Demirhan. Performance Evaluation of the
818
+ Machine Learning Algorithms Used in Inference Mechanism of a Medical Decision Support System. The Scientific
819
+ World Journal, 2014:137896, 2014.
820
+ 12
821
+
JtE1T4oBgHgl3EQfGQNW/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
K9FRT4oBgHgl3EQf1Th-/content/tmp_files/2301.13656v1.pdf.txt ADDED
The diff for this file is too large to render. See raw diff
 
K9FRT4oBgHgl3EQf1Th-/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
KtFRT4oBgHgl3EQf1Dhl/content/tmp_files/2301.13655v1.pdf.txt ADDED
@@ -0,0 +1,4881 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.13655v1 [math.CA] 31 Jan 2023
2
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR
3
+ ESTIMATES
4
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
5
+ ABSTRACT. We develop both bilinear theory and commutator estimates in the context of
6
+ entangled dilations, specifically Zygmund dilations (x1, x2, x3) �→ (δ1x1, δ2x2, δ1δ2x3) in
7
+ R3. We construct bilinear versions of recent dyadic multiresolution methods for Zygmund
8
+ dilations and apply them to prove a paraproduct free T 1 theorem for bilinear singular in-
9
+ tegrals invariant under Zygmund dilations. Independently, we prove linear commutator
10
+ estimates even when the underlying singular integrals do not satisfy weighted estimates
11
+ with Zygmund weights. This requires new paraproduct estimates.
12
+ 1. INTRODUCTION
13
+ “Entangled” systems of dilations, see Nagel-Wainger [22], in the m-parameter product
14
+ space Rd = �m
15
+ i=1 Rdi have the general form
16
+ (x1, . . . , xm) �→ (δλ11
17
+ 1
18
+ · · · δλ1k
19
+ k
20
+ x1, . . . , δλm1
21
+ 1
22
+ · · · δλmk
23
+ k
24
+ xm),
25
+ δ1, . . . , δk > 0,
26
+ and appear naturally throughout analysis. For instance, in R3 the Zygmund dilations
27
+ (x1, x2, x3) �→ (δ1x1, δ2x2, δ1δ2x3) are compatible with the group law of the Heisenberg
28
+ group, see e.g. Müller–Ricci–Stein [21]. Even these simplest entangled dilations are not
29
+ completely understood, especially when it comes to the associated Calderón–Zygmund
30
+ type singular integral operators (SIOs).
31
+ Until recently, multiresolution methods were still missing in the Zygmund dilations
32
+ setting, as pointed out in [5]. This was a big restriction on how to go about developing
33
+ singular integral theory. However, the last two authors together with T. Hytönen and
34
+ E. Vuorinen recently developed this missing Zygmund multiresolution analysis in [14].
35
+ Such dyadic representation theorems and related multiresolution techniques had been
36
+ highly influential in recent advances on SIOs and their applications (see e.g. [12, 13, 20,
37
+ 23]), but developing them in the entangled situation required new ideas. These tools
38
+ then yielded very delicate weighted norm inequalities Lp(w) → Lp(w) for general non-
39
+ convolution form Zygmund singular integrals in the optimal generality of Zygmund
40
+ weights (introduced by Fefferman–Pipher [6])
41
+ [w]Ap,Z := sup
42
+ I∈RZ
43
+ � 1
44
+ |I|
45
+ ˆ
46
+ I
47
+ w(x) dx
48
+ �� 1
49
+ |I|
50
+ ˆ
51
+ I
52
+ w−1/(p−1)(x) dx
53
+ �p−1
54
+ < ∞,
55
+ 1 < p < ∞,
56
+ 2010 Mathematics Subject Classification. 42B20.
57
+ Key words and phrases. singular integrals, multi-parameter analysis, Zygmund dilations, multiresolution
58
+ analysis, weighted estimates.
59
+ E.A. was supported by Academy of Finland through Grant No. 321896 “Incidences on Fractals” (PI =
60
+ Orponen) and No. 314829 “Frontiers of singular integrals” (PI = Hytönen).
61
+ K. L. was supported by the National Natural Science Foundation of China through project number
62
+ 12222114 and 12001400.
63
+ 1
64
+
65
+ 2
66
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
67
+ where the supremum is over Zygmund rectangles I = I1 × I2 × I3, ℓ(I3) = ℓ(I1)ℓ(I2).
68
+ In fact, there is a precise threshold: if the kernel decay in terms of the deviation of
69
+ z ∈ R3 from the “Zygmund manifold” |z1z2| = |z3| is not fast enough, singular integrals
70
+ invariant under Zygmund dilations fail to be bounded with Zygmund weights. We con-
71
+ structed counterexamples and showed the delicate positive result in the optimal range
72
+ using the new multiresolution analysis. Previous results include [5,6,11,24].
73
+ This rather striking threshold for weighted estimates means that it is, in particular,
74
+ unclear in what generality natural estimates for commutators [b, T] = bT − T(b · ) hold.
75
+ Of course, ever since the classical one-parameter result of Coifman–Rochberg–Weiss [2],
76
+ stating that ∥[b, T]∥Lp→Lp ∼ ∥b∥BMO, commutator estimates have been a large and fun-
77
+ damental part of the theory of SIOs and their applications. Commutator estimates in the
78
+ Zygmund dilation setting were previously considered in [5] using the so-called Cauchy
79
+ integral trick. That method requires weighted bounds with Zygmund weights – this is
80
+ because it uses the fact that natural Zygmund adapted BMO functions generate Zyg-
81
+ mund weights. But we now know [14] that such weighted bounds are quite delicate –
82
+ and it turns out that the commutator bounds are true even in the regime where weighted
83
+ estimates fail. We prove the following.
84
+ 1.1. Theorem. Let b ∈ L1
85
+ loc and T be a linear paraproduct free Calderón-Zygmund operator
86
+ adapted to Zygmund dilations as in [14]. Let θ ∈ (0, 1] be the kernel exponent measuring the
87
+ decay in terms of the Zygmund ratio
88
+ Dθ(x) :=
89
+
90
+ |x1x2|
91
+ |x3|
92
+ +
93
+ |x3|
94
+ |x1x2|
95
+ �−θ
96
+ .
97
+ Then for all such θ we have
98
+ ∥[b, T]∥Lp→Lp ≲ ∥b∥bmoZ,
99
+ 1 < p < ∞.
100
+ As weighted estimates only hold with θ = 1, this requires a proof based on the mul-
101
+ tiresolution decomposition [14] and a new family of “Zygmund paraproducts”. Study-
102
+ ing paraproducts is also interesting from the technical viewpoint that, generally, proofs
103
+ of T1 theorems display a structural decomposition of SIOs into their cancellative parts
104
+ and paraproducts. The new Zygmund theory in [14] is designed for the fully cancellative
105
+ case leaving out paraproducts and BMO considerations, so this is the first paper, as far as
106
+ we know, where paraproducts are considered in the Zygmund situation. They are tricky
107
+ objects in the entangled situation. However, while this is also a step forward towards a
108
+ full T1 theorem in the Zygmund setting, the commutator theory that we develop does
109
+ not require so-called partial paraproducts, and so the paraproduct tools developed here
110
+ are not yet sufficient to prove a T1 theorem in the non-cancellative case. We also men-
111
+ tion that during our proof we include some results of independent interest, mainly, a
112
+ new, extremely short proof of the A∞ extrapolation theorem [3].
113
+ Moving to a different direction, we push the Zygmund multiresolution methods [14]
114
+ to the multilinear setting and study bilinear SIOs invariant under Zygmund dilations. A
115
+ classical model of an n-linear SIO T in Rd is obtained by setting
116
+ T(f1, . . . , fn)(x) = U(f1 ⊗ · · · ⊗ fn)(x, . . . , x),
117
+ x ∈ Rd, fi : Rd → C,
118
+ where U is a linear SIO in Rnd. See e.g. Grafakos–Torres [9] for the basic theory. Estimates
119
+ for classical multilinear SIOs play a fundamental role in pure and applied analysis – for
120
+
121
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
122
+ 3
123
+ example, Lp estimates for the homogeneous fractional derivative Dαf = F−1(|ξ|α �f(ξ))
124
+ of a product of two or more functions, the fractional Leibniz rules, are used in the area of
125
+ dispersive equations, see e.g. Kato–Ponce [15] and Grafakos–Oh [8]. We do not otherwise
126
+ attempt to summarize the massive body of literature here and simply mention that the
127
+ closest existing result is perhaps [18], which develops multiresolution methods in the
128
+ non-entangled multilinear bi-parameter case.
129
+ In this paper we prove the following “paraproduct free” T1 theorem for bilinear Zyg-
130
+ mund SIOs.
131
+ 1.2. Theorem. Let T be a bilinear paraproduct free Calderón-Zygmund operator adapted to Zyg-
132
+ mund dilations as in Definition 3.5. Let 1 < p1, p2 < ∞ and 1
133
+ 2 < p < ∞ with 1
134
+ p :=
135
+ 1
136
+ p1 + 1
137
+ p2.
138
+ Then we have
139
+ ∥T(f1, f2)∥Lp ≲ ∥f1∥Lp1∥f2∥Lp2.
140
+ Notice that we can conclude the full bilinear range, including the quasi-Banach range,
141
+ just from the paraproduct free T1 type assumptions. Also relevant is the fact that e.g. the
142
+ appearing weak boundedness condition only involves Zygmund rectangles – that is, the
143
+ T1 assumptions of Definition 3.5 are Zygmund adapted and in this respect weaker than
144
+ the corresponding tri-parameter assumptions.
145
+ It would also be very interesting to develop weighted theory with suitable kernel as-
146
+ sumptions like in the linear case [14]. That is, to generalize our recent paper [19] from
147
+ the standard multi-parameter setting to this entangled Zygmund setting. Recall that it
148
+ would be key to deal with “genuine” multilinear weights, i.e., only impose a joint Ap
149
+ condition on the associated tuple of weights ⃗w = (w1, . . . , wn). While such multilinear
150
+ weighted estimates had been known for one-parameter SIOs for over 10 years by the
151
+ influential paper [16], the multi-parameter version was only recently solved in [19]. The
152
+ entangled situation is very difficult, though, and we do not achieve such estimates in
153
+ this paper. Indeed, we are splitting our operators in a way that is sufficient for the un-
154
+ bounded estimates, but not for the weighted estimates. In fact, already the unweighted
155
+ estimates are surprisingly delicate and the only way we found to achieve them was with
156
+ using this additional decomposition and even some sparse domination tools.
157
+ Here is an outline of the paper. In Section 2 we develop the fundamental Zygmund
158
+ adapted multiresolution methods in the bilinear setting. Section 3 introduces the sin-
159
+ gular integrals and the corresponding testing conditions, and Section 4 uses the kernel
160
+ estimates to bound the various coefficients arising in the multiresolution analysis. Sec-
161
+ tion 5 contains a further decomposition of our dyadic model operators – this is then
162
+ required in Section 6, where the Lp estimates of these model operators are proved. Sec-
163
+ tion 6 concludes with the proof of Theorem 1.2. Section 7 contains the proof of the linear
164
+ commutator estimates, Theorem 1.1, and the corresponding theory of product and lit-
165
+ tle BMO commutators in the Zygmund setting. Appendix A considers bilinear variants
166
+ of the multipliers studied by Fefferman-Pipher [6] – this is motivation for the abstract
167
+ definitions of Section 3.
168
+ 2. BILINEAR ZYGMUND MULTIRESOLUTION ANALYSIS
169
+ 2.A. Dyadic intervals, Zygmund rectangles and basic randomization. Given a dyadic
170
+ grid D, I ∈ D and k ∈ Z, k ≥ 0, we use the following notation:
171
+ (1) ℓ(I) is the side length of I.
172
+
173
+ 4
174
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
175
+ (2) I(k) ∈ D is the kth parent of I, i.e., I ⊂ I(k) and ℓ(I(k)) = 2kℓ(I).
176
+ (3) ch(I) is the collection of the children of I, i.e., ch(I) = {J ∈ D: J(1) = I}.
177
+ (4) EIf = ⟨f⟩I1I is the averaging operator, where ⟨f⟩I =
178
+
179
+ I f =
180
+ 1
181
+ |I|
182
+ ´
183
+ I f.
184
+ (5) ∆If is the martingale difference ∆If = �
185
+ J∈ch(I) EJf − EIf.
186
+ (6) ∆I,kf or ∆k
187
+ If is the martingale difference block
188
+ ∆I,kf = ∆k
189
+ If =
190
+
191
+ J∈D
192
+ J(k)=I
193
+ ∆Jf.
194
+ We will have use for randomization soon. While often the grids are fixed and we sup-
195
+ press the dependence on the random parameters, it will be important to understand the
196
+ definitions underneath. So we go ahead and introduce the related notation and standard
197
+ results now. Let D0 be the standard dyadic grid in R. For ω ∈ {0, 1}Z, ω = (ωi)i∈Z, we
198
+ define the shifted lattice
199
+ D(ω) :=
200
+
201
+ L + ω := L +
202
+
203
+ i: 2−i<ℓ(L)
204
+ 2−iωi: L ∈ D0
205
+
206
+ .
207
+ Let Pω be the product probability measure on {0, 1}Z. We recall the following notion of a
208
+ good interval from [10]. We say that G ∈ D(ω, k), k ≥ 2, if G ∈ D(ω) and
209
+ (2.1)
210
+ d(G, ∂G(k)) ≥ ℓ(G(k))
211
+ 4
212
+ = 2k−2ℓ(G).
213
+ Notice that for all L ∈ D0 and k ≥ 2 we have
214
+ (2.2)
215
+ Pω({ω: L + ω ∈ D(ω, k)}) = 1
216
+ 2.
217
+ The key implication (of practical use later) of G ∈ D(ω, k) is that for n ∈ Z with |n| ≤ 2k−2
218
+ we have
219
+ (2.3)
220
+ (G ∔ n)(k) = G(k),
221
+ G ∔ n := G + nℓ(G).
222
+ In fact, we will not need much more of randomization – it only remains to move the
223
+ notation to our actual setting of R3 = R × R2. We define for
224
+ σ = (σ1, σ2, σ3) ∈ {0, 1}Z × {0, 1}Z × {0, 1}Z
225
+ that
226
+ D(σ) := D(σ1) × D(σ2) × D(σ3).
227
+ Let
228
+ Pσ := Pσ1 × Pσ2 × Pσ3.
229
+ For k = (k1, k2, k3), k1, k2, k3 ≥ 2, we define
230
+ D(σ, k) = D(σ1, k1) × D(σ2, k2) × D(σ3, k3).
231
+ We also e.g. write
232
+ D(σ, (k1, 0, k3)) = D(σ1, k1) × D(σ2) × D(σ3, k3),
233
+ that is, a 0 will designate that we do not have goodness in that parameter.
234
+
235
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
236
+ 5
237
+ As for most of the argument σ is fixed, it makes sense to mainly suppress it from the
238
+ notation and abbreviate, whenever possible, that
239
+ Dm = D(σm),
240
+ D(σm, km) = Dm(km),
241
+ m = 1, 2, 3.
242
+ Then also
243
+ D = D(σ) =
244
+ 3
245
+
246
+ m=1
247
+ Dm,
248
+ D(k) =
249
+ 3
250
+
251
+ m=1
252
+ Dm(km).
253
+ We define the Zygmund rectangles DZ ⊂ D by setting
254
+ (2.4)
255
+ DZ =
256
+
257
+ I =
258
+ 3
259
+
260
+ m=1
261
+ Im ∈ D: ℓ(I1)ℓ(I2) = ℓ(I3)
262
+
263
+ .
264
+ Obviously, DZ(k) is defined similarly as above but also requires �3
265
+ m=1 Im ∈ D(k).
266
+ 2.B. Zygmund martingale differences. Given I = �3
267
+ m=1 Im we define the Zygmund
268
+ martingale difference operator
269
+ ∆I,Zf := ∆I1∆I2×I3f.
270
+ 2.5. Remark. We highlight that the martingale difference ∆I2×I3 is the one-parameter
271
+ (and not the bi-parameter) martingale difference on the rectangle I2 × I3:
272
+ ∆I2×I3 = ∆I2∆I3 + EI2∆I3 + ∆I2EI3 ̸= ∆I2∆I3.
273
+ Moreover, the above operators really act on the full product space but only on the given
274
+ parameters – for instance, ∆I1f(x1, x2, x3) = ∆1
275
+ I1f(x1, x2, x3) = (∆I1f(·, x2, x3))(x1).
276
+ We recall the following facts from [14]. For a dyadic λ > 0 define the dilated lattices
277
+ D2,3
278
+ λ
279
+ = {I2,3 ∈ D2,3 := D2 × D3 : ℓ(I3) = λℓ(I2)}.
280
+ The basic Zygmund expansion goes as follows:
281
+ f =
282
+
283
+ I1∈D1
284
+ ∆I1f =
285
+
286
+ I1∈D1
287
+
288
+ I2,3∈D2,3
289
+ ℓ(I1)
290
+ ∆I1∆I2,3f =
291
+
292
+ I∈DZ
293
+ ∆I,Zf.
294
+ (2.6)
295
+ However, the way we split our operators will not be this simple.
296
+ The following basic results hold for the martingale differences. For I, J ∈ DZ we have
297
+ ∆I,Z∆J,Zf =
298
+
299
+ ∆I,Z
300
+ if I = J,
301
+ 0
302
+ if I ̸= J.
303
+ Notice also that the Zygmund martingale differences satisfy
304
+ ˆ
305
+ R
306
+ ∆I,Zf dx1 = 0
307
+ and
308
+ ˆ
309
+ R2 ∆I,Zf dx2 dx3 = 0.
310
+ Moreover, we have
311
+ ˆ
312
+ (∆I,Zf)g =
313
+ ˆ
314
+ f∆I,Zg.
315
+
316
+ 6
317
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
318
+ 2.C. Haar functions. For an interval J ⊂ R we denote by Jl and Jr the left and right
319
+ halves of J, respectively. We define
320
+ h0
321
+ J = |J|−1/21J
322
+ and
323
+ h1
324
+ J = hJ = |J|−1/2(1Jl − 1Jr).
325
+ The reader should carefully notice that h0
326
+ I is the non-cancellative Haar function for us
327
+ and that in some other papers a different convention is used.
328
+ As we mostly work on R3 = R × R2 we require some Haar functions on R2 as well.
329
+ For I2 × I3 ⊂ R2 and η = (η2, η3) ∈ {0, 1}2 define
330
+
331
+ I2×I3 = hη2
332
+ I2 ⊗ hη3
333
+ I3.
334
+ Similarly, as hI1 denotes a cancellative Haar function on R, we let hI2×I3 denote a can-
335
+ cellative one-parameter Haar function on I2 × I3. This means that
336
+ hI2×I3 = hη
337
+ I2×I3
338
+ for some η = (η2, η3) ∈ {0, 1}2 \ {(0, 0)}. We only use a 0 to denote a non-cancellative
339
+ Haar function: h0
340
+ I2×I3 = h(0,0)
341
+ I2×I3.
342
+ We suppress this η dependence in all that follows in the sense that a finite η summation
343
+ is not written. For example, given I = I1 × I2 × I3 ∈ DZ ⊂ �3
344
+ m=1 Dm decompose
345
+ ∆I,Zf = ∆I1∆I2×I3f = ⟨f, hI1 ⊗ hI2×I3⟩hI1 ⊗ hI2×I3 =: ⟨f, hI,Z⟩hI,Z.
346
+ 2.D. Bilinear Zygmund shifts. In preparation for defining the shifts, we define the fol-
347
+ lowing notation. Let I1, I2, I3 be rectangles, Ij = I1
348
+ j × I2
349
+ j × I3
350
+ j = I1
351
+ j × I2,3
352
+ j , and f1, f2, f3 be
353
+ functions defined on R3. For j1, j2 ∈ {1, 2, 3} define
354
+ Aj1,j2
355
+ I1,I2,I3 = Aj1,j2
356
+ I1,I2,I3(f1, f2, f3) :=
357
+ 3
358
+
359
+ j=1
360
+ ⟨fj, vIj⟩,
361
+ where
362
+ vIj = �hI1
363
+ j ⊗ �hI2,3
364
+ j ;
365
+ �hI1
366
+ j1 = hI1
367
+ j1
368
+ and
369
+ �hI1
370
+ j = h0
371
+ I1
372
+ j , j ̸= j1;
373
+ �hI2,3
374
+ j2 = hI2,3
375
+ j2
376
+ and
377
+ �hI2,3
378
+ j
379
+ = h0
380
+ I2,3
381
+ j , j ̸= j2.
382
+ For a dyadic λ > 0 define
383
+ Dλ = {K = K1 × K2 × K3 ∈ D: λℓ(K1)ℓ(K2) = ℓ(K3)}.
384
+ Moreover, for a rectangle I = I1 × I2 × I3 and k = (k1, k2, k3) define
385
+ I(k) = I(k1)
386
+ 1
387
+ × I(k2)
388
+ 2
389
+ × I(k3)
390
+ 3
391
+ .
392
+ 2.7. Definition. Let k = (k1, k2, k3), ki ∈ {0, 1, 2, . . .}, be fixed. A bilinear Zygmund shift
393
+ Q = Qk of complexity k has the form
394
+ ⟨Qk(f1, f2), f3⟩
395
+ =
396
+
397
+ K∈D2−k1−k2+k3
398
+
399
+ I1,I2,I3∈DZ
400
+ I(k)
401
+ j
402
+ =K
403
+ aK,(Ij)
404
+
405
+ Aj1,j2
406
+ I1,I2,I3 − Aj1,j2
407
+ I1
408
+ j1×I2,3
409
+ 1
410
+ ,I1
411
+ j1×I2,3
412
+ 2
413
+ ,I1
414
+ j1×I2,3
415
+ 3
416
+
417
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
418
+ 7
419
+ − Aj1,j2
420
+ I1
421
+ 1×I2,3
422
+ j2 ,I1
423
+ 2×I2,3
424
+ j2 ,I1
425
+ 3×I2,3
426
+ j2
427
+ + Aj1,j2
428
+ I1
429
+ j1×I2,3
430
+ j2 ,I1
431
+ j1×I2,3
432
+ j2 ,I1
433
+ j1×I2,3
434
+ j2
435
+
436
+ for some j1, j2 ∈ {1, 2, 3}. The coefficients aK,(Ij) satisfy
437
+ |aK,(Ij)| ≤ |I1|1/2|I2|1/2|I3|1/2
438
+ |K|2
439
+ = |I1|3/2
440
+ |K|2 .
441
+ Now, the game is to represent bilinear singular integrals using the operators Qk and
442
+ also – independently – bound the operators Qk suitably. We start with the representa-
443
+ tion part and deal with bounding the operators later. We have not defined our singular
444
+ integrals carefully yet, however, a lot of the required decomposition can be formally car-
445
+ ried out for an arbitrary operator T. The singular integral part is later required to get
446
+ sufficient decay for the appearing scalar coefficients and to handle the paraproducts.
447
+ 2.E. Zygmund decomposition of ⟨T(f1, f2), f3⟩. For now, we focus on the multireso-
448
+ lution part and start formally decomposing a general bilinear operator. We begin by
449
+ writing ⟨T(f1, f2), f3⟩ as
450
+
451
+ I1
452
+ 1,I1
453
+ 2,I1
454
+ 3∈D1
455
+ ⟨T(∆I1
456
+ 1f1, ∆I1
457
+ 2f2), ∆I1
458
+ 3f3⟩
459
+ =
460
+
461
+ I1
462
+ 1,I1
463
+ 2,I1
464
+ 3∈D1
465
+ ℓ(I1
466
+ 1),ℓ(I1
467
+ 2)>ℓ(I1
468
+ 3)
469
+ ⟨T(∆I1
470
+ 1f1, ∆I1
471
+ 2f2), ∆I1
472
+ 3f3⟩
473
+ +
474
+
475
+ I1
476
+ 1,I1
477
+ 2,I1
478
+ 3∈D1
479
+ ℓ(I1
480
+ 1),ℓ(I1
481
+ 3)>ℓ(I1
482
+ 2)
483
+ ⟨T(∆I1
484
+ 1f1, ∆I1
485
+ 2f2), ∆I1
486
+ 3f3⟩
487
+ +
488
+
489
+ I1
490
+ 1,I1
491
+ 2,I1
492
+ 3∈D1
493
+ ℓ(I1
494
+ 2),ℓ(I1
495
+ 3)>ℓ(I1
496
+ 1)
497
+ ⟨T(∆I1
498
+ 1f1, ∆I1
499
+ 2f2), ∆I1
500
+ 3f3⟩
501
+ +
502
+
503
+ I1
504
+ 1,I1
505
+ 2,I1
506
+ 3∈D1
507
+ ℓ(I1
508
+ 1)>ℓ(I1
509
+ 2)=ℓ(I1
510
+ 3)
511
+ ⟨T(∆I1
512
+ 1f1, ∆I1
513
+ 2f2), ∆I1
514
+ 3f3⟩
515
+ +
516
+
517
+ I1
518
+ 1,I1
519
+ 2,I1
520
+ 3∈D1
521
+ ℓ(I1
522
+ 2)>ℓ(I1
523
+ 1)=ℓ(I1
524
+ 3)
525
+ ⟨T(∆I1
526
+ 1f1, ∆I1
527
+ 2f2), ∆I1
528
+ 3f3⟩
529
+ +
530
+
531
+ I1
532
+ 1,I1
533
+ 2,I1
534
+ 3∈D1
535
+ ℓ(I1
536
+ 3)>ℓ(I1
537
+ 1)=ℓ(I1
538
+ 2)
539
+ ⟨T(∆I1
540
+ 1f1, ∆I1
541
+ 2f2), ∆I1
542
+ 3f3⟩
543
+ +
544
+
545
+ I1
546
+ 1,I1
547
+ 2,I1
548
+ 3∈D1
549
+ ℓ(I1
550
+ 1)=ℓ(I1
551
+ 2)=ℓ(I1
552
+ 3)
553
+ ⟨T(∆I1
554
+ 1f1, ∆I1
555
+ 2f2), ∆I1
556
+ 3f3⟩.
557
+ We collapse the first six sums, which are not already diagonal sums, into diagonal sums
558
+
559
+ I1
560
+ 1,I1
561
+ 2,I1
562
+ 3∈D1
563
+ ℓ(I1
564
+ 1)=ℓ(I1
565
+ 2)=ℓ(I1
566
+ 3)
567
+ .
568
+
569
+ 8
570
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
571
+ This has the effect that whenever we have an inequality ℓ(I1
572
+ i ) > ℓ(I1
573
+ j ), the martingale
574
+ difference operator ∆I1
575
+ i corresponding with the larger cube is changed to the averaging
576
+ operator EI1
577
+ i . Thus, in the first three sums we now have two averaging operators, and in
578
+ the next three we have one averaging operator. The more averaging operators we have,
579
+ the less cancellation we have, and thus the main challenge are the first three sums with
580
+ the least cancellation. We mainly focus on the first three sums for this reason.
581
+ In addition, the first three sums are symmetric, so we may focus on only one of them,
582
+ and choose to look at
583
+
584
+ I1
585
+ 1,I1
586
+ 2,I1
587
+ 3∈D1
588
+ ℓ(I1
589
+ 1),ℓ(I1
590
+ 2)>ℓ(I1
591
+ 3)
592
+ ⟨T(∆I1
593
+ 1f1, ∆I1
594
+ 2f2), ∆I1
595
+ 3f3⟩ =
596
+
597
+ I1
598
+ 1,I1
599
+ 2,I1
600
+ 3∈D1
601
+ ℓ(I1
602
+ 1)=ℓ(I1
603
+ 2)=ℓ(I1
604
+ 3)
605
+ ⟨T(EI1
606
+ 1f1, EI1
607
+ 2f2), ∆I1
608
+ 3f3⟩.
609
+ Now, we fix I1
610
+ 1, I1
611
+ 2, I1
612
+ 3 ∈ D1 with ℓ(I1
613
+ 1) = ℓ(I1
614
+ 2) = ℓ(I1
615
+ 3) and repeat the argument for
616
+ ⟨T(EI1
617
+ 1f1, EI1
618
+ 2f2), ∆I1
619
+ 3f3⟩ using the lattice D2,3
620
+ ℓ(I1), where recall that for a dyadic λ > 0 we
621
+ have
622
+ D2,3
623
+ λ
624
+ = {I2 × I3 ∈ D2,3 := D2 × D3 : ℓ(I3) = λℓ(I2)}.
625
+ This produces seven terms, and we again focus on
626
+
627
+ I2
628
+ 1×I3
629
+ 1,I2
630
+ 2×I3
631
+ 2,I2
632
+ 3×I3
633
+ 3∈D2,3
634
+ ℓ(I1)
635
+ ℓ(I2
636
+ 1)=ℓ(I2
637
+ 2)=ℓ(I2
638
+ 3)
639
+ ⟨T(EI1
640
+ 1EI2
641
+ 1×I3
642
+ 1f1, EI1
643
+ 2EI2
644
+ 2×I3
645
+ 2f2), ∆I1
646
+ 3∆I2
647
+ 3×I3
648
+ 3f3⟩.
649
+ Altogether, our focus, for now, is on the key term
650
+ (2.8)
651
+
652
+ I1,I2,I3∈DZ
653
+ ℓ(I1)=ℓ(I2)=ℓ(I3)
654
+ ⟨T(EI1f1, EI2f2), ∆I3,Zf3⟩,
655
+ where ℓ(I1) = ℓ(I2) = ℓ(I3) means that
656
+ ℓ(Im
657
+ 1 ) = ℓ(Im
658
+ 2 ) = ℓ(Im
659
+ 3 ),
660
+ m = 1, 2, 3.
661
+ This was completely generic – we now go a step further to the direction of Zygmund
662
+ shifts and start introducing Haar functions into the mix.
663
+ 2.F. Further decomposition of (2.8). Write
664
+ ⟨T(EI1f1, EI2f2), ∆I3,Zf3⟩ = ⟨T(h0
665
+ I1, h0
666
+ I2), hI3,Z⟩⟨f1, h0
667
+ I1⟩⟨f2, h0
668
+ I2⟩⟨f3, hI3,Z⟩.
669
+ Now, we perform a rather complicated decomposition of the product ⟨f1, h0
670
+ I1⟩⟨f2, h0
671
+ I2⟩.
672
+ To this end, start by writing
673
+ ⟨f1, h0
674
+ I1⟩⟨f2, h0
675
+ I2⟩
676
+ =
677
+
678
+ ⟨f1, h0
679
+ I1⟩⟨f2, h0
680
+ I2⟩ − ⟨f1, h0
681
+ I1
682
+ 3h0
683
+ I2,3
684
+ 1 ⟩⟨f2, h0
685
+ I1
686
+ 3h0
687
+ I2,3
688
+ 2 ⟩
689
+
690
+ + ⟨f1, h0
691
+ I1
692
+ 3h0
693
+ I2,3
694
+ 1 ⟩⟨f2, h0
695
+ I1
696
+ 3 h0
697
+ I2,3
698
+ 2 ⟩
699
+ =: A1 + A2.
700
+ We then further decompose A1 as follows
701
+ A1 =
702
+
703
+ ⟨f1, h0
704
+ I1⟩⟨f2, h0
705
+ I2⟩ − ⟨f1, h0
706
+ I1
707
+ 3h0
708
+ I2,3
709
+ 1 ⟩⟨f2, h0
710
+ I1
711
+ 3h0
712
+ I2,3
713
+ 2 ⟩
714
+ − ⟨f1, h0
715
+ I1
716
+ 1h0
717
+ I2,3
718
+ 3 ⟩⟨f2, h0
719
+ I1
720
+ 2h0
721
+ I2,3
722
+ 3 ⟩ + ⟨f1, h0
723
+ I1
724
+ 3h0
725
+ I2,3
726
+ 3 ⟩⟨f2, h0
727
+ I1
728
+ 3 h0
729
+ I2,3
730
+ 3 ⟩
731
+
732
+
733
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
734
+ 9
735
+ +
736
+
737
+ ⟨f1, h0
738
+ I1
739
+ 1 h0
740
+ I2,3
741
+ 3 ⟩⟨f2, h0
742
+ I1
743
+ 2 h0
744
+ I2,3
745
+ 3 ⟩ − ⟨f1, h0
746
+ I1
747
+ 3h0
748
+ I2,3
749
+ 3 ⟩⟨f2, h0
750
+ I1
751
+ 3h0
752
+ I2,3
753
+ 3 ⟩
754
+
755
+ .
756
+ When we later specialize to singular integrals, we will in particular make the following
757
+ assumption. We say that T is a paraproduct free operator, if for all cancellative Haar
758
+ functions hI1 and hI2,3 we have
759
+ ⟨T(1 ⊗ 1J2,3
760
+ 1 , 1 ⊗ 1J2,3
761
+ 2 ), hI1 ⊗ 1J2,3
762
+ 3 ⟩ = ⟨T ∗,j
763
+ 1 (1 ⊗ 1J2,3
764
+ 1 , 1 ⊗ 1J2,3
765
+ 2 ), hI1 ⊗ 1J2,3
766
+ 3 ⟩
767
+ = ⟨T(1I1
768
+ 1 ⊗ 1, 1I1
769
+ 2 ⊗ 1), 1I1
770
+ 3 ⊗ hI2,3⟩ = ⟨T ∗,j
771
+ 2,3 (1I1
772
+ 1 ⊗ 1, 1I1
773
+ 2 ⊗ 1), 1I1
774
+ 3 ⊗ hI2,3⟩ = 0
775
+ for all the adjoints j ∈ {1, 2}. With this assumption in the full summation (2.8) everything
776
+ else vanishes except
777
+
778
+ I1,I2,I3∈DZ
779
+ ℓ(I1)=ℓ(I2)=ℓ(I3)
780
+ ⟨T(h0
781
+ I1, h0
782
+ I2),hI3,Z⟩
783
+
784
+ ⟨f1, h0
785
+ I1⟩⟨f2, h0
786
+ I2⟩ − ⟨f1, h0
787
+ I1
788
+ 3×I2,3
789
+ 1 ⟩⟨f2, h0
790
+ I1
791
+ 3×I2,3
792
+ 2 ⟩
793
+ − ⟨f1, h0
794
+ I1
795
+ 1 ×I2,3
796
+ 3 ⟩⟨f2, h0
797
+ I1
798
+ 2×I2,3
799
+ 3 ⟩ + ⟨f1, h0
800
+ I3⟩⟨f2, h0
801
+ I3⟩
802
+
803
+ ⟨f3, hI3,Z⟩.
804
+ So we eliminated the paraproducts by assumption, and now we have to manipulate this
805
+ remaining term to a suitable form involving shifts.
806
+ In the above sum we will relabel I3 = I = I1 × I2 × I3 = I1 × I2,3. Then, for n1 =
807
+ (n1
808
+ 1, n2
809
+ 1, n3
810
+ 1) = (n1
811
+ 1, n2,3
812
+ 1 ) we write
813
+ I1 = I ∔ n1 = (I1 + n1
814
+ 1ℓ(I1)) × (I2 + n2
815
+ 1ℓ(I2)) × (I3 + n3
816
+ 1ℓ(I3)) = (I1 ∔ n1
817
+ 1) × (I2,3 ∔ n2,3
818
+ 1 ).
819
+ We write I2 similarly as I2 = I ∔ n2. Notice that if n1
820
+ 1 = n1
821
+ 2 = 0, then the term inside
822
+ the summation vanishes. Similarly, if n2,3
823
+ 1
824
+ = n2,3
825
+ 2
826
+ = (0, 0), the term inside the summation
827
+ vanishes. So we need to study
828
+
829
+ n1,n2∈Z3
830
+ max(|n1
831
+ 1|,|n1
832
+ 2|)̸=0
833
+ max(|n2
834
+ 1|,|n2
835
+ 2|)̸=0 or max(|n3
836
+ 1|,|n3
837
+ 2|)̸=0
838
+
839
+ I∈DZ
840
+ cI,n1,n2,
841
+ where
842
+ cI,n1,n2
843
+ = ⟨T(h0
844
+ I∔n1, h0
845
+ I∔n2), hI,Z⟩
846
+
847
+ ⟨f1, h0
848
+ I∔n1⟩⟨f2, h0
849
+ I∔n2⟩ − ⟨f1, h0
850
+ I1×(I2,3∔n2,3
851
+ 1 )⟩⟨f2, h0
852
+ I1×(I2,3∔n2,3
853
+ 2
854
+ )⟩
855
+ − ⟨f1, h0
856
+ (I1∔n1
857
+ 1)×I2,3⟩⟨f2, h0
858
+ (I1∔n1
859
+ 2)×I2,3⟩ + ⟨f1, h0
860
+ I⟩⟨f2, h0
861
+ I⟩
862
+
863
+ ⟨f3, hI,Z⟩.
864
+ We write
865
+
866
+ n1,n2∈Z3
867
+ max
868
+ j=1,2 |n1
869
+ j|̸=0
870
+ max
871
+ j=1,2 |n2
872
+ j|̸=0 or max
873
+ j=1,2 |n3
874
+ j|̸=0
875
+
876
+ I∈DZ
877
+ cI,n1,n2
878
+ =
879
+
880
+
881
+ k1,k2,k3=2
882
+
883
+ n1,n2∈Z3
884
+ max
885
+ j=1,2 |nm
886
+ j |∈(2km−3,2km−2]
887
+ m=1,2,3
888
+
889
+ I∈DZ
890
+ cI,n1,n2
891
+
892
+ 10
893
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
894
+ +
895
+
896
+
897
+ k1,k2=2
898
+
899
+ n1,n2∈Z3
900
+ max
901
+ j=1,2 |nm
902
+ j |∈(2km−3,2km−2]
903
+ m=1,2
904
+ n3
905
+ 1=n3
906
+ 2=0
907
+
908
+ I∈DZ
909
+ cI,n1,n2
910
+ + Σsym,
911
+ where Σsym is symmetric to the second term and has n2
912
+ 1 = n2
913
+ 2 = 0.
914
+ Recall how everything implicitly depends on the random parameter σ, so that we can
915
+ average over it. By independence, we have by (2.2) that
916
+
917
+
918
+
919
+ k1,k2,k3=2
920
+
921
+ n1,n2∈Z3
922
+ max
923
+ j=1,2 |nm
924
+ j |∈(2km−3,2km−2]
925
+ m=1,2,3
926
+
927
+ I∈DZ
928
+ cI,n1,n2
929
+ = 8Eσ
930
+
931
+
932
+ k1,k2,k3=2
933
+
934
+ n1,n2∈Z3
935
+ max
936
+ j=1,2 |nm
937
+ j |∈(2km−3,2km−2]
938
+ m=1,2,3
939
+
940
+ I∈DZ(k)
941
+ cI,n1,n2,
942
+ k = (k1, k2, k3).
943
+ (2.9)
944
+ For the other two terms, where n2
945
+ j = 0 or n3
946
+ j = 0, we perform the above but do not add
947
+ goodness to the second and third parameters, respectively. For example, we have
948
+
949
+
950
+
951
+ k1,k2=2
952
+
953
+ n1,n2∈Z3
954
+ max
955
+ j=1,2 |nm
956
+ j |∈(2km−3,2km−2]
957
+ m=1,2
958
+ n3
959
+ 1=n3
960
+ 2=0
961
+
962
+ I∈DZ
963
+ cI,n1,n2
964
+ = 4Eσ
965
+
966
+
967
+ k1,k2=2
968
+
969
+ n1,n2∈Z3
970
+ max
971
+ j=1,2 |nm
972
+ j |∈(2km−3,2km−2]
973
+ m=1,2
974
+ n3
975
+ 1=n3
976
+ 2=0
977
+
978
+ I∈DZ(k1,k2,0)
979
+ cI,n1,n2.
980
+ Continuing with (2.9), we write it as
981
+ C8Eσ
982
+
983
+
984
+ k1,k2,k3=2
985
+ (|k| + 1)2ϕ(k)
986
+
987
+ K∈Dλ
988
+
989
+ I∈DZ(k)
990
+ I(k)=K
991
+
992
+ n1,n2∈Z3
993
+ maxj=1,2 |nm
994
+ j |∈(2km−3,2km−2]
995
+ m=1,2,3
996
+ cI,n1,n2
997
+ C(|k| + 1)2ϕ(k),
998
+ where
999
+ Dλ = {K = K1 × K2 × K3 ∈ D: λℓ(K1)ℓ(K2) = ℓ(K3)},
1000
+ λ = 2k3−k1−k2,
1001
+
1002
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
1003
+ 11
1004
+ and C is some suitably large constant depending on T. Recall that by (2.3) we also have
1005
+ (I ∔ n1)(k) = (I ∔ n2)(k) = I(k) = K.
1006
+ We have arrived to a point where we cannot go further without talking about singular
1007
+ integrals. Indeed, we need kernel estimates to control the coefficients. But on a structural
1008
+ level (with the paraproduct free assumption), we have obtained a reasonable representa-
1009
+ tion of the main term (2.8) in terms of sums of bilinear Zygmund shifts.
1010
+ 3. BILINEAR ZYGMUND SINGULAR INTEGRALS
1011
+ We begin by defining the required kernel estimates and cancellation conditions for
1012
+ bilinear singular integrals T invariant under Zygmund dilations. For motivation for the
1013
+ form of the kernel estimates, see Appendix A for kernel bounds of bilinear multipliers.
1014
+ This viewpoint makes the kernel estimates natural – on the other hand, they are also of
1015
+ the right form so that we will be able to bound the coef���cients from the multiresolution
1016
+ decomposition and obtain reasonable decay.
1017
+ 3.A. Full kernel representation. Our bilinear singular integral T invariant under Zyg-
1018
+ mund dilations is related to a full kernel K in the following way. The kernel K is a
1019
+ function
1020
+ K : (R3 × R3 × R3) \ ∆ → C,
1021
+ where
1022
+ ∆ = {(x, y, z) ∈ R3 × R3 × R3 : xi = yi = zi for at least one i = 1, 2, 3}.
1023
+ We look at the action of T on rectangles like I1 × I2 × I3 =: I1 × I2,3 in R3 = R × R × R =
1024
+ R × R2. So let Ii = I1
1025
+ i × I2
1026
+ i × I3
1027
+ i be rectangles, i = 1, 2, 3. Assume that there exists
1028
+ i1, i2, j1, j2 ∈ {1, 2, 3} so that I1
1029
+ i1 and I1
1030
+ i2 are disjoint and also I2,3
1031
+ j1
1032
+ and I2,3
1033
+ j2
1034
+ are disjoint.
1035
+ Then we have the full kernel representation
1036
+ ⟨T(1I1, 1I2), 1I3⟩ =
1037
+ ˚
1038
+ K(x, y, z)1I1(x)1I2(y)1I3(z) dx dy dz.
1039
+ The kernel K satisfies the following estimates.
1040
+ First, we define the decay factor
1041
+ Dθ(x, y) =
1042
+ ��2
1043
+ i=1(|xi| + |yi|)
1044
+ |x3| + |y3|
1045
+ +
1046
+ |x3| + |y3|
1047
+ �2
1048
+ i=1(|xi| + |yi|)
1049
+ �−θ
1050
+ ,
1051
+ θ ∈ (0, 2],
1052
+ and the tri-parameter bilinear size factor
1053
+ S(x, y) =
1054
+ 3
1055
+
1056
+ i=1
1057
+ 1
1058
+
1059
+ |xi| + |yi|
1060
+ �2 .
1061
+ We demand the following size estimate
1062
+ (3.1)
1063
+ |K(x, y, z)| ≲ Dθ(x − z, y − z)S(x − z, y − z).
1064
+
1065
+ 12
1066
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
1067
+ Let now c = (c1, c2, c3) be such that |ci − xi| ≤ max(|xi − zi|, |yi − zi|)/2 for i = 1, 2, 3.
1068
+ We assume that K satisfies the mixed size and Hölder estimates
1069
+ |K((c1,x2, x3), y, z) − K(x, y, z)|
1070
+
1071
+
1072
+ |c1 − x1|
1073
+ |x1 − z1| + |y1 − z1|
1074
+ �α1Dθ(x − z, y − z)S(x − z, y − z),
1075
+ (3.2)
1076
+ and
1077
+ |K((x1, c2, c3), y, z) − K(x, y, z)|
1078
+
1079
+
1080
+ |c2 − x2|
1081
+ |x2 − z2| + |y2 − z2| +
1082
+ |c3 − x3|
1083
+ |x3 − z3| + |y3 − z3|
1084
+ �α23Dθ(x − z, y − z)S(x − z, y − z),
1085
+ (3.3)
1086
+ where α1, α23 ∈ (0, 1]. Finally, we assume that K satisfies the Hölder estimate
1087
+ |K(c, y, z) − K((c1, x2, x3), y, z) − K((x1, c2, c3), y, z) + K(x, y, z)|
1088
+
1089
+
1090
+ |c1 − x1|
1091
+ |x1 − z1| + |y1 − z1|
1092
+ �α1�
1093
+ |c2 − x2|
1094
+ |x2 − z2| + |y2 − z2| +
1095
+ |c3 − x3|
1096
+ |x3 − z3| + |y3 − z3|
1097
+ �α23
1098
+ × Dθ(x − z, y − z)S(x − z, y − z).
1099
+ (3.4)
1100
+ We also demand the symmetrical mixed size and Hölder estimates and Hölder estimates.
1101
+ For j = 1, 2, define the adjoint kernels K∗,j, K∗,j
1102
+ 1
1103
+ and K∗,j
1104
+ 2,3 via the natural formulas, e.g.,
1105
+ K∗,1(x, y, z) = K(z, y, x),
1106
+ K∗,2
1107
+ 1 (x, y, z) = K(x, (z1, y2, y3), (y1, z2, z3)).
1108
+ We assume that each adjoint kernel satisfies the same estimates as the kernel K.
1109
+ 3.B. Partial kernel representations. Let �θ ∈ (0, 1]. For every interval I1 we assume that
1110
+ there exists a kernel
1111
+ KI1 : (R2 × R2 × R2) \ {(x2,3, y2,3, z2,3): xi = yi = zi for i = 2 or i = 3} → C,
1112
+ so that if I2,3
1113
+ j1 and I2,3
1114
+ j2 are disjoint for some j1, j2 ∈ {1, 2, 3}, then
1115
+ ⟨T(1I1 ⊗ 1I2,3
1116
+ 1 , 1I1 ⊗ 1I2,3
1117
+ 2 ), 1I1 ⊗ 1I2,3
1118
+ 3 ⟩
1119
+ =
1120
+ ˚
1121
+ KI1(x2,3, y2,3, z2,3)1I2,3
1122
+ 1 (x2,3)1I2,3
1123
+ 2 (y2,3)1I2,3
1124
+ 3 (z2,3) dx2,3 dy2,3 dz2,3.
1125
+ We demand the following estimates for the kernel KI1 : The size estimate
1126
+ |KI1(x2,3, y2,3, z2,3)|
1127
+
1128
+ �|I1|(|x2 − z2| + |y2 − z2|)
1129
+ |x3 − z3| + |y3 − z3|
1130
+ +
1131
+ |x3 − z3| + |y3 − z3|
1132
+ |I1|(|x2 − z2| + |y2 − z2|)
1133
+ �−�θ
1134
+ |I1|
1135
+ �3
1136
+ i=2
1137
+
1138
+ |xi − zi| + |yi − zi|
1139
+ �2
1140
+ and the continuity estimate
1141
+ |KI1(c2,3, y2,3, z2,3) − KI1(x2,3, y2,3, z2,3)|
1142
+
1143
+
1144
+ |c2 − x2|
1145
+ |x2 − z2| + |y2 − z2| +
1146
+ |c3 − x3|
1147
+ |x3 − z3| + |y3 − z3|
1148
+ �α23
1149
+ ×
1150
+ �|I1|(|x2 − z2| + |y2 − z2|)
1151
+ |x3 − z3| + |y3 − z3|
1152
+ +
1153
+ |x3 − z3| + |y3 − z3|
1154
+ |I1|(|x2 − z2| + |y2 − z2|)
1155
+ �−�θ
1156
+ |I1|
1157
+ �3
1158
+ i=2
1159
+
1160
+ |xi − zi| + |yi − zi|
1161
+ �2
1162
+
1163
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
1164
+ 13
1165
+ whenever c2,3 = (c2, c3) is such that |ci − xi| ≤ max(|xi − zi|, |yi − zi|)/2 for i = 2, 3. We
1166
+ also assume the symmetrical continuity estimates.
1167
+ We assume similar one-parameter conditions for the other partial kernel representa-
1168
+ tion. That is, for every rectangle I2,3, there exists a standard bilinear Calderón-Zygmund
1169
+ kernel KI2,3 so that if I1
1170
+ j1 and I1
1171
+ j2 are disjoint for some j1, j2 ∈ {1, 2, 3}, then
1172
+ ⟨T(1I1
1173
+ 1 ⊗ 1I2,3, 1I1
1174
+ 2 ⊗ 1I2,3), 1I1
1175
+ 3 ⊗ 1I2,3⟩
1176
+ =
1177
+ ˚
1178
+ KI2,3(x1, y1, z1)1I1
1179
+ 1 (x1)1I1
1180
+ 2 (y1)1I1
1181
+ 3(z1) dx1 dy1 dz1.
1182
+ The kernel KI2,3 satisfies the standard estimates
1183
+ |KI2,3(x1, y1, z1)| ≤ CKI2,3
1184
+ 1
1185
+ (|x1 − z1| + |y1 − z1|)2 ,
1186
+ |KI2,3(x1, y1, z1) − KI2,3(c1, y1, z1)| ≤ CKI2,3
1187
+ |x1 − c1|α1
1188
+ (|x1 − z1| + |y1 − z1|)2+α1
1189
+ whenever |x1 − c1| ≤ max(|x1 − z1|, |y1 − z1|)/2, and the symmetric continuity estimates.
1190
+ The smallest possible constant CKI2,3 in these inequalities is denoted by ∥KI2,3∥CZα1. We
1191
+ then assume that
1192
+ ∥KI2,3∥CZα1 ≲ |I2,3|.
1193
+ 3.C. Cancellation assumptions: paraproduct free operators. We say that T is a para-
1194
+ product free operator, if for all cancellative Haar functions hI1 and hI2,3 we have
1195
+ ⟨T(1 ⊗ 1J2,3
1196
+ 1 , 1 ⊗ 1J2,3
1197
+ 2 ), hI1 ⊗ 1J2,3
1198
+ 3 ⟩ = ⟨T ∗,j
1199
+ 1 (1 ⊗ 1J2,3
1200
+ 1 , 1 ⊗ 1J2,3
1201
+ 2 ), hI1 ⊗ 1J2,3
1202
+ 3 ⟩
1203
+ = ⟨T(1I1
1204
+ 1 ⊗ 1, 1I1
1205
+ 2 ⊗ 1), 1I1
1206
+ 3 ⊗ hI2,3⟩ = ⟨T ��,j
1207
+ 2,3 (1I1
1208
+ 1 ⊗ 1, 1I1
1209
+ 2 ⊗ 1), 1I1
1210
+ 3 ⊗ hI2,3⟩ = 0
1211
+ for all the adjoints j ∈ {1, 2}. We always assume that all bilinear Zygmund operators
1212
+ in this article satisfy this cancellation condition. The intention of this condition is to
1213
+ guarantee that our operator is representable using cancellative shifts only.
1214
+ 3.D. Weak boundedness property. We say that T satisfies the weak boundedness prop-
1215
+ erty if
1216
+ |⟨T(1I, 1I), 1I⟩| ≲ |I|
1217
+ for all Zygmund rectangles I = I1 × I2 × I3.
1218
+ 3.5. Definition. We say that a bilinear operator T is a paraproduct free Calderón-
1219
+ Zygmund operator adapted to Zygmund dilations (CZZ operator) if T has the full kernel
1220
+ representation, the partial kernel representations, is paraproduct free and satisfies the
1221
+ weak boundedness property.
1222
+ 4. ESTIMATES FOR THE SHIFT COEFFICIENTS
1223
+ We now move to consider the shift coefficients that appeared in the decomposition
1224
+ of T in Section 2.F. When T is a CZZ operator, we can estimate them. Without loss of
1225
+ generality, we estimate
1226
+ ⟨T(h0
1227
+ I ˙+n1, h0
1228
+ I ˙+n2), hI,Z⟩
1229
+ for I ∈ DZ and different values of n1, n2 ∈ Z3, and without loss of generality we assume
1230
+ θ = ˜θ < 1. The coefficients related to the other terms of the decomposition (other than the
1231
+ main term (2.8)) may have a different set of Haar functions, but they are treated similarly.
1232
+
1233
+ 14
1234
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
1235
+ We show that
1236
+ (4.1)
1237
+ |⟨T(h0
1238
+ I ˙+n1, h0
1239
+ I ˙+n2), hI,Z⟩| ≲ (|k| + 1)2ϕ(k) |I|
1240
+ 3
1241
+ 2
1242
+ |K|2 ,
1243
+ where
1244
+ ϕ(k) := 2−k1α1−k2 min{α23,θ}−max{k3−k1−k2,0}θ.
1245
+ For terms of this particular form, we would not actually need to analyze some of the
1246
+ diagonal cases (see Section 2.F). However, these diagonal terms would appear in some
1247
+ other forms, so it makes sense to deal with them here (even though in the real situation
1248
+ the Haar functions might be permuted differently, this does not matter, and the calcula-
1249
+ tions we present apply). It is very helpful to study the linear case [14], since the kernel
1250
+ estimates are relatively involved and we will not repeat every detail when they are simi-
1251
+ lar.
1252
+ Let mi := maxj=1,2 |ni
1253
+ j|. The analysis of the coefficients splits to combinations of
1254
+
1255
+
1256
+
1257
+
1258
+
1259
+ |m1| ∈ (2k1−3, 2k1−2],
1260
+ k1 = 3, 4, . . . ,
1261
+ (Separated)
1262
+ |m1| = 1,
1263
+ (Adjacent)
1264
+ |m1| = 0,
1265
+ (Identical)
1266
+ and
1267
+
1268
+
1269
+
1270
+
1271
+
1272
+
1273
+
1274
+
1275
+
1276
+
1277
+
1278
+
1279
+
1280
+
1281
+
1282
+
1283
+
1284
+
1285
+
1286
+ |mi| ∈ (2ki−3, 2ki−2],
1287
+ i = 2, 3, ki = 3, 4, . . . ,
1288
+ (Separated)
1289
+ |m2| < 2 and |m3| ∈ (2k3−3, 2k3−2],
1290
+ k3 = 3, 4, . . . ,
1291
+ (Separated)
1292
+ |m2| ∈ (2k2−3, 2k2−2] and |m3| < 2
1293
+ k2 = 3, 4, . . . ,
1294
+ (Separated)
1295
+ |m2| = 1 and |m3| ≤ 1
1296
+ (Adjacent)
1297
+ |m2| = 0 and |m3| = 1
1298
+ (Adjacent)
1299
+ m2 = 0 = m3.
1300
+ (Identical)
1301
+ It is enough to consider mi = ni
1302
+ 1 since the case mi = ni
1303
+ 2 is symmetrical. We will not go
1304
+ through explicitly every combination – rather, we choose some illustrative examples.
1305
+ 4..1. Separated/Separated. We begin with the case |ni
1306
+ 1| ≥ 2 for all i = 1, 2, 3. Hence,
1307
+ |xi − zi| ≥ |ni
1308
+ 1|ℓ(Ii) ≥ 2ki−3ℓ(Ii)
1309
+ and
1310
+ |xi − zi| ≤ |ni
1311
+ 1|ℓ(Ii) + 2ℓ(Ii) ≤ 2ki−1ℓ(Ii)
1312
+ for i = 1, 2, 3. Moreover, |xi − zi| ≥ |yi − zi|/2 ≥ 0 for i = 1, 2, 3. Thus, we have the
1313
+ estimate
1314
+ ��2
1315
+ i=1(|xi − zi| + |yi − zi|)
1316
+ (|x3 − z3| + |y3 − z3|)
1317
+ +
1318
+ |x3 − z3| + |y3 − z3|
1319
+ �2
1320
+ i=1(|xi − zi| + |yi − zi|)
1321
+ �−θ
1322
+
1323
+ ��2
1324
+ i=1 |xi − zi|
1325
+ |x3 − z3|
1326
+ +
1327
+ |x3 − z3|
1328
+ �2
1329
+ i=1 |xi − zi|
1330
+ �−θ
1331
+
1332
+ ��2
1333
+ i=1 2kiℓ(Ii)
1334
+ 2k3ℓ(I3)
1335
+ +
1336
+ 2k3ℓ(I3)
1337
+ �2
1338
+ i=1 2kiℓ(Ii)
1339
+ �−θ
1340
+ = (2k1+k2−k3 + 2k3−k1−k2)−θ.
1341
+
1342
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
1343
+ 15
1344
+ Using the cancellation of the Haar function we then have
1345
+ ���
1346
+ ˚
1347
+ K(x, y, z)h0
1348
+ I ˙+n1(x)h0
1349
+ I ˙+n2(y)hI,Z(z) dx dy dz
1350
+ ���
1351
+ =
1352
+ ���
1353
+ ˚ �
1354
+ K(x, y, z) − K(x, y, (cI1, z2,3)) − K(x, y, (z1, cI2,3)) + K(x, y, cI)
1355
+
1356
+ × h0
1357
+ I ˙+n1(x)h0
1358
+ I ˙+n2(y)hI,Z(z) dx dy dz
1359
+ ���
1360
+
1361
+ ˚
1362
+ 2−k1α1(2−k2 + 2−k3)α23 (2k1+k2−k3 + 2k3−k1−k2)−θ
1363
+ |K|2
1364
+ h0
1365
+ I ˙+n1(x)h0
1366
+ I ˙+n2(y)h0
1367
+ I(z) dx dy dz
1368
+ = 2−k1α1(2−k2 + 2−k3)α23(2k1+k2−k3 + 2k3−k1−k2)−θ |I|
1369
+ 3
1370
+ 2
1371
+ |K|2 ≤ ϕ(k) |I|
1372
+ 3
1373
+ 2
1374
+ |K|2 .
1375
+ Let us then consider the case, where we have separation in the parameter 3 but not in
1376
+ the parameter 2 – that is, |n2
1377
+ 1| < 2 ≤ |n3
1378
+ 1|. Then
1379
+ ��2
1380
+ i=1(|xi − zi| + |yi − zi|)
1381
+ |x3 − z3| + |y3 − z3|
1382
+ +
1383
+ |x3 − z3| + |y3 − z3|
1384
+ �2
1385
+ i=1(|xi − zi| + |yi − zi|)
1386
+ �−θ
1387
+ (4.2)
1388
+
1389
+ �|x2 − z2| + |y2 − z2|
1390
+ 2k3−k1|I2|
1391
+ +
1392
+ 2k3−k1|I2|
1393
+ |x2 − z2| + |y2 − z2|
1394
+ �−θ
1395
+
1396
+ � |x2 − z2|
1397
+ 2k3−k1|I2| + 2k3−k1|I2|
1398
+ |x2 − z2|
1399
+ �−θ
1400
+ +
1401
+ � |y2 − z2|
1402
+ 2k3−k1|I2| + 2k3−k1|I2|
1403
+ |y2 − z2|
1404
+ �−θ
1405
+ ,
1406
+ and so using the mixed estimates
1407
+ ���
1408
+ ˚
1409
+ K(x, y, z)h0
1410
+ I ˙+n1(x)h0
1411
+ I ˙+n2(y)hI,Z(z) dx dy dz
1412
+ ���
1413
+ =
1414
+ ���
1415
+ ˚ �
1416
+ K(x, y, z) − K(x, y, (cI1, z2,3))
1417
+
1418
+ h0
1419
+ I ˙+n1(x)h0
1420
+ I ˙+n2(y)hI,Z(z) dx dy dz
1421
+ ���
1422
+
1423
+ ˚
1424
+ 2−k1α1|K1|−2|K3|−2
1425
+
1426
+ |x2−z2|+|y2−z2|
1427
+ 2k3−k1|I2|
1428
+ +
1429
+ 2k3−k1|I2|
1430
+ |x2−z2|+|y2−z2|
1431
+ �−θ
1432
+
1433
+ |x2 − z2| + |y2 − z2|
1434
+ �2
1435
+ × h0
1436
+ I ˙+n1(x)h0
1437
+ I ˙+n2(y)h0
1438
+ I(z) dx dy dz
1439
+ = 2−k1α1 |I1|
1440
+ 3
1441
+ 2|I3|
1442
+ 3
1443
+ 2
1444
+ |K1|2|K3|2
1445
+ ˚
1446
+
1447
+ |x2−z2|+|y2−z2|
1448
+ 2k3−k1|I2|
1449
+ +
1450
+ 2k3−k1|I2|
1451
+ |x2−z2|+|y2−z2|
1452
+ �−θ
1453
+
1454
+ |x2 − z2| + |y2 − z2|
1455
+ �2
1456
+ × h0
1457
+ I2 ˙+n2
1458
+ 1(x2)h0
1459
+ I2 ˙+n2
1460
+ 2(y2)h0
1461
+ I2(z2) dx2 dy2 dz2
1462
+ ≲ ϕ(k) |I|
1463
+ 3
1464
+ 2
1465
+ |K|2 .
1466
+ We note that the last inequality requires a case study (see also [14, Lemma 8.5]) and we
1467
+ used the standard estimate
1468
+ ˆ
1469
+ Rd
1470
+ du
1471
+ (r + |u0 − u|)d+α ≲ r−α.
1472
+ (4.3)
1473
+ Symmetrical estimates hold if |n2
1474
+ 1| ≥ 2 > |n3
1475
+ 1|.
1476
+
1477
+ 16
1478
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
1479
+ 4..2. Adjacent/Separated. We look at the example case |n2
1480
+ 1| ≥ 2 > |n3
1481
+ 1| and |n1
1482
+ 1| = 1. By the
1483
+ size estimate we have
1484
+ |⟨T(h0
1485
+ I ˙+n1, h0
1486
+ I ˙+n2), hI,Z⟩|
1487
+
1488
+ |I2|3/2
1489
+ |I1,3|3/2|K2|2
1490
+ ¨
1491
+
1492
+ (|x1−z1|+|y1−z1|)2k2ℓ(I2)
1493
+ |x3−z3|+|y3−z3|
1494
+ +
1495
+ |x3−z3|+|y3−z3|
1496
+ (|x1−z1|+|y1−z1|)2k2ℓ(I2)
1497
+ �−θ
1498
+
1499
+ |x1 − z1| + |y1 − z1|
1500
+ �2�
1501
+ |x3 − z3| + |y3 − z3|
1502
+ �2
1503
+ × 1I1,3 ˙+n1,3
1504
+ 1 (x1,3)1I1,3 ˙+n1,3
1505
+ 2 (y1,3)1I1,3(z1,3) dx1,3 dy1,3 dz1,3.
1506
+ Similarly as (4.2), we can split the integral into two terms. Then by (4.3) we reduce the
1507
+ problem to estimating
1508
+ ¨
1509
+
1510
+ (|x1−z1|+|y1−z1|)2k2ℓ(I2)
1511
+ |x3−z3|
1512
+ +
1513
+ |x3−z3|
1514
+ (|x1−z1|+|y1−z1|)2k2ℓ(I2)
1515
+ �−θ
1516
+
1517
+ |x1 − z1| + |y1 − z1|
1518
+ �2|x3 − z3|
1519
+ × 1I1,3 ˙+n1,3
1520
+ 1 (x1,3)1I1 ˙+n1
1521
+ 2(y1)1I1,3(z1,3) dx1,3 dy1 dz1,3
1522
+ +
1523
+ ¨
1524
+
1525
+ (|x1−z1|+|y1−z1|)2k2ℓ(I2)
1526
+ |y3−z3|
1527
+ +
1528
+ |y3−z3|
1529
+
1530
+ |x1−z1|+|y1−z1|
1531
+
1532
+ 2k2ℓ(I2)
1533
+ �−θ
1534
+ (|x1 − z1| + |y1 − z1|)2|y3 − z3|
1535
+ × 1I1,3 ˙+n1,3
1536
+ 1 (x1,3)1I1,3 ˙+n1,3
1537
+ 2 (y1,3)1I1(z1) dx1,3 dy1,3 dz1.
1538
+ Since they are similar, we only bound the first one. Note that
1539
+ �(|x1 − z1| + |y1 − z1|)2k2ℓ(I2)
1540
+ |x3 − z3|
1541
+ +
1542
+ |x3 − z3|
1543
+ (|x1 − z1| + |y1 − z1|)2k2ℓ(I2)
1544
+ �−θ
1545
+ × (|x1 − z1| + |y1 − z1|)−2
1546
+
1547
+ �(|x1 − z1| + |y1 − z1|)2k2ℓ(I2)
1548
+ |x3 − z3|
1549
+ �−θ
1550
+ (|x1 − z1| + |y1 − z1|)−2χ{|x1−z1|2k2ℓ(I2)≥|x3−z3|}
1551
+ +
1552
+
1553
+ |x3 − z3|
1554
+ (|x1 − z1| + |y1 − z1|)2k2ℓ(I2)
1555
+ �−θ
1556
+ (|x1 − z1| + |y1 − z1|)−2χ{|x1−z1|2k2ℓ(I2)<|x3−z3|}.
1557
+ Then apply (4.3) to the integral over y1, then by following the linear case [14, Lemma
1558
+ 8.11] we get that the above integral is bounded by |I1,3|k22−k2θ. Thus, we get
1559
+ |⟨T(h0
1560
+ I ˙+n1, h0
1561
+ I ˙+n2), hI,Z⟩| ≲
1562
+ |I2|3/2
1563
+ |I1,3|1/2|K2|2 k22−k2θ ≲ k2ϕ(k) |I|
1564
+ 3
1565
+ 2
1566
+ |K|2 .
1567
+ 4..3. Adjacent/Adjacent. We again have no major changes to the linear case but in order
1568
+ to use the estimate
1569
+ (4.4)
1570
+ ˆ
1571
+ R
1572
+
1573
+ t
1574
+ |u| + |u|
1575
+ t
1576
+ �−θ
1577
+ t|u|
1578
+ |f(u)| du ≲ t−1Mf(0)
1579
+
1580
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
1581
+ 17
1582
+ we need to first use (4.3) repeatedly. For example, consider |n1
1583
+ 1| = 1 and |n2
1584
+ 1| = 1, |n3
1585
+ 1| ≤ 1.
1586
+ By the size estimate of the kernel, we need to control
1587
+ � �2
1588
+ i=1(|xi−zi|+|yi−zi|)
1589
+ |x3−z3|+|y3−z3|
1590
+ +
1591
+ |x3−z3|+|y3−z3|
1592
+ �2
1593
+ i=1(|xi−zi|+|yi−zi|)
1594
+ �−θ
1595
+ �3
1596
+ i=1
1597
+
1598
+ |xi − zi| + |yi − zi|
1599
+ �2
1600
+ h0
1601
+ I ˙+n1(x)h0
1602
+ I ˙+n2(y)h0
1603
+ I(z).
1604
+ As before, we split this into two terms, one of them is
1605
+ � �2
1606
+ i=1(|xi−zi|+|yi−zi|)
1607
+ |x3−z3|
1608
+ +
1609
+ |x3−z3|
1610
+ �2
1611
+ i=1(|xi−zi|+|yi−zi|)
1612
+ �−θ
1613
+ �3
1614
+ i=1
1615
+
1616
+ |xi − zi| + |yi − zi|
1617
+ �2
1618
+ h0
1619
+ I ˙+n1(x)h0
1620
+ I ˙+n2(y)h0
1621
+ I(z).
1622
+ We then apply (4.3) to the integral over y3, and then use the previous trick repeatedly.
1623
+ That is, we write
1624
+ ��2
1625
+ i=1(|xi − zi| + |yi − zi|)
1626
+ |x3 − z3|
1627
+ +
1628
+ |x3 − z3|
1629
+ �2
1630
+ i=1(|xi − zi| + |yi − zi|)
1631
+ �−θ
1632
+
1633
+ ��2
1634
+ i=1(|xi − zi| + |yi − zi|)
1635
+ |x3 − z3|
1636
+ �−θ
1637
+ χ{|x1−z1|(|x2−z2|+|y2−z2|)≥|x3−z3|}
1638
+ +
1639
+
1640
+ |x3 − z3|
1641
+ �2
1642
+ i=1(|xi − zi| + |yi − zi|)
1643
+ �−θ
1644
+ χ{|x1−z1|(|x2−z2|+|y2−z2|)<|x3−z3|}
1645
+ and apply (4.3) to the integral over y1. Then, after a similar argument on y2, we finally
1646
+ arrive at
1647
+ 1
1648
+ |I|
1649
+ 1
1650
+ 2
1651
+ ¨
1652
+ � �2
1653
+ i=1 |xi−zi|
1654
+ |x3−z3|
1655
+ +
1656
+ |x3−z3|
1657
+ �2
1658
+ i=1 |xi−zi|
1659
+ �−θ
1660
+ �3
1661
+ i=1 |xi − zi|
1662
+ h0
1663
+ I ˙+n1(x)h0
1664
+ I(z) dx dz
1665
+
1666
+ 1
1667
+ |I|
1668
+ 1
1669
+ 2
1670
+ ≲ |I|
1671
+ 3
1672
+ 2
1673
+ |K|2 .
1674
+ 4..4. Adjacent/Identical. We consider the case |n1
1675
+ 1| = 1 and n2
1676
+ j = n3
1677
+ j = 0, j = 1, 2. We write
1678
+
1679
+ Q2,3
1680
+ 1
1681
+ ,Q2,3
1682
+ 2 ,Q2,3
1683
+ 3 ∈ch(I2,3)
1684
+ ⟨T(h0
1685
+ I ˙+n11Q2,3
1686
+ 1 , h0
1687
+ I ˙+n21Q2,3
1688
+ 2 ), hI,Z1Q2,3
1689
+ 3 ⟩.
1690
+ It is enough to consider Q2,3
1691
+ 1
1692
+ = Q2,3
1693
+ 2
1694
+ = Q2,3
1695
+ 3
1696
+ since otherwise we have adjacent intervals,
1697
+ and we are back in the Adjacent/Adjacent case. Hence, the partial kernel representation
1698
+ 3.B yields that
1699
+ ��� ± |I2,3|− 3
1700
+ 2
1701
+ ˚
1702
+ KQ2,3
1703
+ 1 h0
1704
+ I1 ˙+n1
1705
+ 1h0
1706
+ I1 ˙+n1
1707
+ 2hI1
1708
+ ���
1709
+ ≲ |I2,3|
1710
+ 3
1711
+ 2
1712
+ |K2,3|2
1713
+ ˚
1714
+ 1
1715
+ (|x1 − z1| + |y1 − z1|)2 h0
1716
+ I1 ˙+n1
1717
+ 1(x1)h0
1718
+ I1 ˙+n1
1719
+ 2(y1)hI1(z1) dx1 dy1 dz1.
1720
+ Then, first using (4.3) and then standard integration methods we get the following in-
1721
+ equality
1722
+ ˚
1723
+ 1
1724
+ (|x1 − z1| + |y1 − z1|)2 h0
1725
+ I1 ˙+n1
1726
+ 1(x1)h0
1727
+ I1 ˙+n1
1728
+ 2(y1)hI1(z1) dx1 dy1 dz1
1729
+
1730
+ 18
1731
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
1732
+
1733
+ 1
1734
+ |I1|
1735
+ 1
1736
+ 2
1737
+ ¨
1738
+ 1
1739
+ |x1 − z1|h0
1740
+ I1 ˙+n1
1741
+ 1(x1)hI1(z1) dx1 dz1
1742
+ ��
1743
+ 1
1744
+ |I1|
1745
+ 1
1746
+ 2
1747
+ ∼ |I1|
1748
+ 3
1749
+ 2
1750
+ |K1|2
1751
+ as desired.
1752
+ 4..5. Identical/Identical. Just like in above we split the pairing to
1753
+
1754
+ Q1
1755
+ 1,Q1
1756
+ 2,Q1
1757
+ 3∈ch(I1)
1758
+
1759
+ Q2,3
1760
+ 1
1761
+ ,Q2,3
1762
+ 2 ,Q2,3
1763
+ 3 ∈ch(I2,3)
1764
+ ⟨T(h0
1765
+ I ˙+n1(1Q1
1766
+ 1 ⊗ 1Q2,3
1767
+ 1 ), h0
1768
+ I ˙+n2(1Q1
1769
+ 2 ⊗ 1Q2,3
1770
+ 2 )), hI,Z(1Q1
1771
+ 3 ⊗ 1Q2,3
1772
+ 3 )⟩.
1773
+ The cases when Q1
1774
+ i ̸= Q1
1775
+ j for some i, j = 1, 2, 3, i ̸= j are essentially included in the cases
1776
+ of the two previous subsections. Hence, we consider Q1
1777
+ 1 = Q1
1778
+ 2 = Q1
1779
+ 3. Then there are two
1780
+ cases left, that is, either Q2,3
1781
+ i
1782
+ ̸= Q2,3
1783
+ j
1784
+ for some i, j = 1, 2, 3, i ̸= j, or Q2,3
1785
+ 1
1786
+ = Q2,3
1787
+ 2
1788
+ = Q2,3
1789
+ 3 .
1790
+ Beginning from the latter one, we directly see that
1791
+ |⟨T(1Q1
1792
+ 1 ⊗ 1Q2,3
1793
+ 1 , 1Q1
1794
+ 1 ⊗ 1Q2,3
1795
+ 1 ), 1Q1
1796
+ 1 ⊗ 1Q2,3
1797
+ 1 ⟩| ≲ |Q1
1798
+ 1||Q2,3
1799
+ 1 |
1800
+ by the weak boundedness property 3.D. Hence, we get the desired bound
1801
+ |⟨T(h0
1802
+ I ˙+n1(1Q1
1803
+ 1 ⊗ 1Q2,3
1804
+ 1 ), h0
1805
+ I ˙+n2(1Q1
1806
+ 1 ⊗ 1Q2,3
1807
+ 1 )), hI,Z(1Q1
1808
+ 1 ⊗ 1Q2,3
1809
+ 1 )⟩| ≲ |Q1|
1810
+ |I|
1811
+ 3
1812
+ 2
1813
+ ≤ |I|
1814
+ 3
1815
+ 2
1816
+ |K|2 .
1817
+ We handle the remaining case Q2,3
1818
+ i
1819
+ ̸= Q2,3
1820
+ j
1821
+ for some i, j = 1, 2, 3, i ̸= j. By the partial
1822
+ kernel representation and its size estimate we get
1823
+ ��� ± |I|− 3
1824
+ 2
1825
+ ˚
1826
+ KQ1
1827
+ 11Q2,3
1828
+ 1 1Q2,3
1829
+ 2 1Q2,3
1830
+ 3
1831
+ ���
1832
+
1833
+ 1
1834
+ |I1|
1835
+ 1
1836
+ 2
1837
+ 1
1838
+ |I2,3|
1839
+ 3
1840
+ 2
1841
+ ˚ �|I1|(|x2 − z2| + |y2 − z2|)
1842
+ |x3 − z3| + |y3 − z3|
1843
+ +
1844
+ |x3 − z3| + |y3 − z3|
1845
+ |I1|(|x2 − z2| + |y2 − z2|)
1846
+ �−θ
1847
+ ×
1848
+ 3
1849
+
1850
+ i=2
1851
+ 1
1852
+
1853
+ |xi − zi| + |yi − zi|
1854
+ �2 1Q2,3
1855
+ 1 1Q2,3
1856
+ 2 1Q2,3
1857
+ 3 dx2,3 dy2,3 dz2,3.
1858
+ Then using similar arguments as in the Adjacent/Adjacent case and (4.4) gives us the
1859
+ desired bound.
1860
+ 5. STRUCTURAL DECOMPOSITION OF ZYGMUND SHIFTS
1861
+ In this section we decompose the bilinear Zygmund shifts (see Section 2.D) as a sum
1862
+ of operators with simpler cancellation properties. The decomposition is not optimal (in
1863
+ the sense that weighted estimates with Zygmund weights cannot be obtained via this) –
1864
+ however, it is sufficient for unweighted boundedness in the full range that we later obtain
1865
+ via tri-parameter theory. Recall that k = (k1, k2, k3) is the complexity of the bilinear
1866
+ Zygmund shift.
1867
+
1868
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
1869
+ 19
1870
+ 5.1. Definition. Bilinear operators of the form
1871
+ (5.2)
1872
+ S(l1,l2,l3)(f1, f2) =
1873
+
1874
+ L∈Dλ
1875
+
1876
+ I
1877
+ (ℓj)
1878
+ j
1879
+ =L
1880
+ aL,(Ij)⟨f1, hI1
1881
+ 1 ⊗ h0
1882
+ I2,3
1883
+ 1 ⟩⟨f2, h0
1884
+ I1
1885
+ 2 ⊗ hI2,3
1886
+ 2 ⟩hI3,
1887
+ where λ = 2n, n ∈ Z, |n| ≤ 3 max(ki) and
1888
+ |aL,(Ij)| ≤ |Ij|
1889
+ 3
1890
+ 2
1891
+ |L|2 ,
1892
+ are tri-parameter bilinear shifts of Zygmund nature if at least one rectangle I1
1893
+ i1 ×I2,3
1894
+ i2 , i1 =
1895
+ 1, 3, i2 = 2, 3 is a Zygmund rectangle and
1896
+ (1) ℓi
1897
+ j ≤ ki for all i, j = 1, 2, 3;
1898
+ (2) (ℓ3
1899
+ j − ℓ2
1900
+ j)+ ≤ (k3 − k2)+ for all j = 1, 2, 3.
1901
+ Moreover, any adjoint
1902
+ S
1903
+ j∗
1904
+ 1,j∗
1905
+ 2,3
1906
+ (l1,l2,l3),
1907
+ j1, j2,3 ∈ {0, 1, 2},
1908
+ is also considered to be a tri-parameter bilinear shift of Zygmund nature. Here, the ad-
1909
+ joint j∗
1910
+ 2,3 means that, for example, in case j2,3 = 1 functions h0
1911
+ I2,3
1912
+ 1
1913
+ and hI2,3
1914
+ 3
1915
+ switch places.
1916
+ Note that these operators share a ‘weaker’ Zygmund structure. Ideally, we would
1917
+ want to have I3 ∈ DZ and I1
1918
+ 1 × I2,3
1919
+ 2
1920
+ ∈ DZ.
1921
+ 5.3. Proposition. Let Qk, k = (k1, k2, k3), be a bilinear Zygmund shift operator as defined in
1922
+ Section 2.D. Then
1923
+ Qk = C
1924
+ c
1925
+
1926
+ u=1
1927
+ k1−1
1928
+
1929
+ l1=0
1930
+ k2,3−1
1931
+
1932
+ l2,3=0
1933
+ Su,
1934
+ where Su is a bilinear operator as in Definition 5.1 with complexity depending on l and k,and
1935
+ k2,3−1
1936
+
1937
+ l2,3=0
1938
+ :=
1939
+
1940
+
1941
+
1942
+
1943
+
1944
+
1945
+
1946
+
1947
+
1948
+
1949
+
1950
+
1951
+ 0≤l2=l3≤k2−1
1952
+ +
1953
+
1954
+ l2=k2
1955
+ k2≤l3≤k3−1
1956
+ ,
1957
+ if k3 ≥ k2
1958
+
1959
+ 0≤l2=l3≤k3−1
1960
+ +
1961
+
1962
+ k3≤l2≤k2−1
1963
+ l3=k3
1964
+ ,
1965
+ if k3 < k2.
1966
+ Proof. The argument is similar in spirit to the purely bi-parameter decomposition in [1].
1967
+ For notational convenience, we consider a shift Qk of the particular form
1968
+ ⟨Qk(f1, f2), f3⟩
1969
+ =
1970
+
1971
+ K∈D2−k1−k2+k3
1972
+
1973
+ I1,I2,I3∈DZ
1974
+ I(k)
1975
+ j
1976
+ =K
1977
+ aK,(Ij)
1978
+
1979
+ A3,3
1980
+ I1,I2,I3 − A3,3
1981
+ I1
1982
+ 3×I2,3
1983
+ 1
1984
+ ,I1
1985
+ 3×I2,3
1986
+ 2
1987
+ ,I3
1988
+ − A3,3
1989
+ I1
1990
+ 1×I2,3
1991
+ 3
1992
+ ,I1
1993
+ 2×I2,3
1994
+ 3
1995
+ ,I3 + A3,3
1996
+ I3,I3,I3
1997
+
1998
+ =
1999
+
2000
+ K∈D2−k1−k2+k3
2001
+
2002
+ I1,I2,I3∈DZ
2003
+ I(k)
2004
+ j
2005
+ =K
2006
+ aK,(Ij)⟨f3, hI3⟩
2007
+
2008
+ ⟨f1, h0
2009
+ I1⟩⟨f2, h0
2010
+ I2⟩ − ⟨f1, h0
2011
+ I1
2012
+ 3h0
2013
+ I2,3
2014
+ 1 ⟩⟨f2, h0
2015
+ I1
2016
+ 3h0
2017
+ I2,3
2018
+ 2 ⟩
2019
+
2020
+ 20
2021
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
2022
+ − ⟨f1, h0
2023
+ I1
2024
+ 1h0
2025
+ I2,3
2026
+ 3 ⟩⟨f2, h0
2027
+ I1
2028
+ 2h0
2029
+ I2,3
2030
+ 3 ⟩ + ⟨f1, h0
2031
+ I3⟩⟨f2, h0
2032
+ I3⟩
2033
+
2034
+ .
2035
+ There is no essential difference in the general case. Let us also use the usual abbreviation
2036
+ D2−k1−k2+k3 = Dλ.
2037
+ We define
2038
+ bK,(Ij) = |I1|aK,(Ij)
2039
+ and
2040
+ B3,3
2041
+ I1,I2,I3 = ⟨f1⟩I1⟨f2⟩I2⟨f3, hI3⟩.
2042
+ We can write the shift Qk using these by replacing a with b and A with B.
2043
+ Recall the notation
2044
+ ∆l1
2045
+ K1f =
2046
+
2047
+ L1∈D1
2048
+ (L1)(l1)=K1
2049
+ ∆L1f,
2050
+ P k1
2051
+ K1f =
2052
+ k1−1
2053
+
2054
+ l1=0
2055
+ ∆l1
2056
+ K1f,
2057
+ EK1f = ⟨f⟩K11K1,
2058
+ Ek1
2059
+ K1f =
2060
+
2061
+ L1∈D1
2062
+ (L1)(k1)=K1
2063
+ ⟨f⟩L11L1.
2064
+ Let us define
2065
+ (5.4)
2066
+ P k2,3
2067
+ K2,3f :=
2068
+ k2,3−1
2069
+
2070
+ l2,3=0
2071
+ ∆(l2,l3)
2072
+ K2,3 f :=
2073
+
2074
+
2075
+
2076
+
2077
+
2078
+
2079
+
2080
+
2081
+
2082
+ k2−1
2083
+
2084
+ l2=0
2085
+ ∆l2
2086
+ K2,3f +
2087
+ k3−1
2088
+
2089
+ l3=k2 Ek2
2090
+ K2∆l3
2091
+ K3f,
2092
+ if k3 ≥ k2
2093
+ k3−1
2094
+ ���
2095
+ l3=0
2096
+ ∆l3
2097
+ K2,3f +
2098
+ k2−1
2099
+
2100
+ l2=k3 ∆l2
2101
+ K2Ek3
2102
+ K3f,
2103
+ if k3 < k2,
2104
+ where we have the standard one-parameter definition
2105
+ ∆li
2106
+ K2,3f =
2107
+
2108
+ L2,3∈D2,3
2109
+ (L2)(li)×(L3)(li)=K2×K3
2110
+ ∆L2,3f.
2111
+ We also use a similar shorthand for the expanded martingale blocks
2112
+ k2,3−1
2113
+
2114
+ l2,3=0
2115
+ ∆(l2,l3)
2116
+ K2,3 f =
2117
+ k2,3−1
2118
+
2119
+ l2,3=0
2120
+
2121
+ (L2,3)(l2,3)=K2,3
2122
+ ⟨f, hL2,3⟩hL2,3,
2123
+ where we allow, for example, that hL2,3 = h0
2124
+ L2 ⊗ hL3 when k3 > k2 and l2 = k2.
2125
+ Using this notation we define the following. For a cube I and integers l, j0 ∈ {1, 2, . . . }
2126
+ we define
2127
+ (5.5)
2128
+ DI,l(j, j0) =
2129
+
2130
+
2131
+
2132
+
2133
+
2134
+ EI,
2135
+ if j ∈ {1, . . . , j0 − 1},
2136
+ P l
2137
+ I,
2138
+ if j = j0,
2139
+ id,
2140
+ if j ∈ {j0 + 1, j0 + 2, . . . },
2141
+ where id denotes the identity operator, and if we have a rectangle I2,3 and a tuple l2,3 we
2142
+ use the modified P l2,3
2143
+ I2,3.
2144
+
2145
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
2146
+ 21
2147
+ Let I1, I2, I3 be as in the summation of Qk. We use the above notation in parameter one
2148
+ DI1,l1(j, j0) and for the other two parameters we use DI2,3,l2,3(j, j0). Thus, expanding to
2149
+ the martingale blocks leads us to
2150
+ B3,3
2151
+ I1,I2,I3
2152
+ =
2153
+ 3
2154
+
2155
+ m1,m2=1
2156
+ 2
2157
+
2158
+ j=1
2159
+ ⟨D1
2160
+ K1,k1(j, m1)D2,3
2161
+ K2,3,k2,3(j, m2)fj⟩Ij⟨f3, hI3⟩.
2162
+ Hence, we may write
2163
+
2164
+ K∈Dλ
2165
+
2166
+ I1,I2,I3∈DZ
2167
+ I(k)
2168
+ j
2169
+ =K
2170
+ B3,3
2171
+ I1,I2,I3 =:
2172
+ 3
2173
+
2174
+ m1,m2=1
2175
+ Σ1
2176
+ m1,m2.
2177
+ Also, we have that
2178
+ B3,3
2179
+ I1
2180
+ 3×I2,3
2181
+ 1
2182
+ ,I1
2183
+ 3×I2,3
2184
+ 2
2185
+ ,I1
2186
+ 3×I2,3
2187
+ 3
2188
+ =
2189
+ 3
2190
+
2191
+ m2=1
2192
+ 2
2193
+
2194
+ j=1
2195
+ ⟨D2,3
2196
+ K2,3,k2,3(j, m2)fj⟩I1
2197
+ 3×I2,3
2198
+ j ⟨f3, hI3⟩
2199
+ and
2200
+ B3,3
2201
+ I1
2202
+ 1×I2,3
2203
+ 3
2204
+ ,I1
2205
+ 2×I2,3
2206
+ 3
2207
+ ,I1
2208
+ 3×I2,3
2209
+ 3
2210
+ =
2211
+ 3
2212
+
2213
+ m1=1
2214
+ 2
2215
+
2216
+ j=1
2217
+ ⟨D1
2218
+ K1,k1(j, m1)fj⟩I1
2219
+ j ×I2,3
2220
+ 3 ⟨f3, hI3⟩,
2221
+ which gives that
2222
+
2223
+ K∈Dλ
2224
+
2225
+ I1,I2,I3∈DZ
2226
+ I(k)
2227
+ j
2228
+ =K
2229
+ B3,3
2230
+ I1
2231
+ 3×I2,3
2232
+ 1
2233
+ ,I1
2234
+ 3×I2,3
2235
+ 2
2236
+ ,I1
2237
+ 3×I2,3
2238
+ 3
2239
+ =:
2240
+ 3
2241
+
2242
+ m2=1
2243
+ Σ2
2244
+ m2
2245
+ and
2246
+
2247
+ K∈Dλ
2248
+
2249
+ I1,I2,I3∈DZ
2250
+ I(k)
2251
+ j
2252
+ =K
2253
+ B3,3
2254
+ I1
2255
+ 1×I2,3
2256
+ 3
2257
+ ,I1
2258
+ 2×I2,3
2259
+ 3
2260
+ ,I1
2261
+ 3×I2,3
2262
+ 3
2263
+ =:
2264
+ 3
2265
+
2266
+ m1=1
2267
+ Σ3
2268
+ m1.
2269
+ Finally, we just set
2270
+
2271
+ K∈Dλ
2272
+
2273
+ I1,I2,I3∈DZ
2274
+ I(k)
2275
+ j
2276
+ =K
2277
+ B3,3
2278
+ I3,I3,I3 =: Σ4.
2279
+ Thus, we have the following decomposition
2280
+ ⟨Qk(f1, f2), f3⟩ =
2281
+ 2
2282
+
2283
+ m1,m2=1
2284
+ Σ1
2285
+ m1,m2 +
2286
+ 2
2287
+
2288
+ m2=1
2289
+ (Σ1
2290
+ 3,m2 − Σ2
2291
+ m2)
2292
+ +
2293
+ 2
2294
+
2295
+ m1=1
2296
+ (Σ1
2297
+ m1,3 − Σ3
2298
+ m1) + (Σ1
2299
+ 3,3 − Σ2
2300
+ 3 − Σ3
2301
+ 3 + Σ4).
2302
+
2303
+ 22
2304
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
2305
+ First, we take one Σ1
2306
+ m1,m2 with m1, m2 ∈ {1, 2}. For notational convenience, we choose
2307
+ the case m1 = m2 = 2. Recall that
2308
+ Σ1
2309
+ 2,2 =
2310
+
2311
+ K∈Dλ
2312
+
2313
+ I1,I2,I3∈DZ
2314
+ I(k)
2315
+ j
2316
+ =K
2317
+ bK,(Ij)⟨f1⟩K⟨P k1
2318
+ K1P k2,3
2319
+ K2,3f2⟩I2⟨f3, hI3⟩.
2320
+ We expand
2321
+ ⟨P k1
2322
+ K1P k2,3
2323
+ K2,3f2⟩I2 =
2324
+ k1−1
2325
+
2326
+ l1=0
2327
+ k2,3−1
2328
+
2329
+ l2,3=0
2330
+
2331
+ (L1)(l1)=K1
2332
+ (L2,3)(l2,3)=K2,3
2333
+ ⟨f2, hL1 ⊗ hL2,3⟩⟨hL1 ⊗ hL2,3⟩I2
2334
+ and note that L is not necessarily a Zygmund rectangle. It holds that
2335
+ Σ1
2336
+ 2,2 =
2337
+ k1−1
2338
+
2339
+ l1=0
2340
+ k2,3−1
2341
+
2342
+ l2,3=0
2343
+
2344
+ K∈Dλ
2345
+
2346
+ L(l1,l2,l3)=K
2347
+
2348
+ I3∈DZ
2349
+ I(k)
2350
+ 3
2351
+ =K
2352
+ � �
2353
+ I1
2354
+ I(k)
2355
+ 1
2356
+ =K
2357
+
2358
+ I2⊂L
2359
+ I(k)
2360
+ 2
2361
+ =K
2362
+ bK,(Ij)⟨hL⟩I2
2363
+ |K|
2364
+ 1
2365
+ 2
2366
+
2367
+ ⟨f1, h0
2368
+ K⟩⟨f2, hL⟩⟨f3, hI3⟩.
2369
+ Now, since we can easily check that
2370
+ ���
2371
+
2372
+ I1
2373
+ I(k)
2374
+ 1
2375
+ =K
2376
+
2377
+ I2⊂L
2378
+ I(k)
2379
+ 2
2380
+ =K
2381
+ bK,(Ij)⟨hL⟩I2
2382
+ |K|
2383
+ 1
2384
+ 2
2385
+ ��� ≤ |K|
2386
+ 1
2387
+ 2 |L|
2388
+ 1
2389
+ 2 |I3|
2390
+ 1
2391
+ 2
2392
+ |K|2
2393
+ ,
2394
+ we get a sum of operators we wanted
2395
+ Σ1
2396
+ 2,2 =
2397
+ k1−1
2398
+
2399
+ l1=0
2400
+ k2,3−1
2401
+
2402
+ l2,3=0
2403
+ ⟨S(0,(l1,l2,l3),k)(f1, f2), f3⟩,
2404
+ where S(0,(l1,l2,l3),k) is a type of the shift (5.2). The general case Σ1
2405
+ m1,m2 is analogous.
2406
+ We turn to the terms Σ1
2407
+ 3,m2 − Σ2
2408
+ m2. Let us take, for example, the case m2 = 1. After
2409
+ expanding P k2,3
2410
+ K2,3 in the first slot, Σ1
2411
+ 3,1 − Σ2
2412
+ 1 can be written as
2413
+ k2,3−1
2414
+
2415
+ l2,3=0
2416
+
2417
+ K∈Dλ
2418
+
2419
+ (L2,3)(l2,3)=K2,3
2420
+
2421
+ I1,I2,I3
2422
+ I(k)
2423
+ j
2424
+ =K
2425
+ bK,(Ij)⟨hL2,3⟩I2,3
2426
+ 1
2427
+ ��
2428
+ f1, 1K1
2429
+ |K1| ⊗ hL2,3
2430
+
2431
+ ⟨f2⟩K1×I2,3
2432
+ 2
2433
+
2434
+
2435
+ f1,
2436
+ 1I1
2437
+ 3
2438
+ |I1
2439
+ 3| ⊗ hL2,3
2440
+
2441
+ ⟨f2⟩I1
2442
+ 3×I2,3
2443
+ 2
2444
+
2445
+ ⟨f3, hI3⟩.
2446
+ For the moment, we fix one l2,3 and write g1 = ⟨f1, hL2⟩ and g2 = ⟨f2⟩I2,3
2447
+ 2 . We write inside
2448
+ the brackets
2449
+ 2
2450
+
2451
+ j=1
2452
+ ⟨gj⟩K1 −
2453
+ 2
2454
+
2455
+ j=1
2456
+ ⟨gj⟩I1
2457
+ 3 = −
2458
+ k1−1
2459
+
2460
+ l1=0
2461
+
2462
+ 2
2463
+
2464
+ j=1
2465
+ ⟨gj⟩(I1
2466
+ 3)(l1) −
2467
+ 2
2468
+
2469
+ j=1
2470
+ ⟨gj⟩(I1
2471
+ 3)(l1+1)
2472
+
2473
+
2474
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
2475
+ 23
2476
+ and then expand �2
2477
+ j=1⟨gj⟩(I1
2478
+ 3)(l1) − �2
2479
+ j=1⟨gj⟩(I1
2480
+ 3)(l1+1) as
2481
+ ⟨∆(I1
2482
+ 3)(l1+1)g1⟩I1
2483
+ 3⟨g2⟩(I1
2484
+ 3)(l1) + ⟨g1⟩(I1
2485
+ 3)(l1+1)⟨∆(I1
2486
+ 3)(l1+1)g2⟩I1
2487
+ 3.
2488
+ We get
2489
+ 2
2490
+
2491
+ j=1
2492
+ ⟨gj⟩K1 −
2493
+ 2
2494
+
2495
+ j=1
2496
+ ⟨gj⟩I1
2497
+ 3
2498
+ = −
2499
+ k1−1
2500
+
2501
+ l1=0
2502
+
2503
+ ⟨∆(I1
2504
+ 3)(l1+1)g1⟩I1
2505
+ 3⟨g2⟩(I1
2506
+ 3)(l1) + ⟨g1⟩(I1
2507
+ 3)(l1+1)⟨∆(I1
2508
+ 3)(l1+1)g2⟩I1
2509
+ 3
2510
+
2511
+ ,
2512
+ where we can expand
2513
+ ⟨∆(I1
2514
+ 3)(l1+1)gj⟩I1
2515
+ 3 = ⟨gj, h(I1
2516
+ 3)(l1+1)⟩⟨h(I1
2517
+ 3 )(l1+1)⟩I1
2518
+ 3.
2519
+ For fixed l1 and l2,3 the expansion leads to the term
2520
+
2521
+ K∈Dλ
2522
+
2523
+ (L2,3)(l2,3)=K2,3
2524
+
2525
+ I1,I2,I3
2526
+ I(k)
2527
+ j
2528
+ =K
2529
+ bK,(Ij)⟨h(I1
2530
+ 3)(l1+1) ⊗ hL2,3⟩I1
2531
+ 3×I2,3
2532
+ 1
2533
+
2534
+ f1, h(I1
2535
+ 3 )(l1+1) ⊗ hL2,3
2536
+
2537
+ ⟨f2⟩(I1
2538
+ 3)(l1)×I2,3
2539
+ 2 ⟨f3, hI3⟩,
2540
+ and to the symmetrical one, where the cancellation h(I1
2541
+ 3)(l1+1) is paired with the second
2542
+ function and f1 is averaged over (I1
2543
+ 3)(l1+1). Again, we want to reorganize the summations
2544
+ and verify the correct normalization for the shifts of the form (5.2). In the first parameter
2545
+ we will now take (I1
2546
+ 3)(l1+1) as the new top cube, that is,
2547
+
2548
+ K1
2549
+
2550
+ (L1)(k1−l1)=K1
2551
+
2552
+ K2,3∈D2−l1−k2+k3 ℓ(L1)
2553
+
2554
+ (I1
2555
+ 3)(l1)=L1
2556
+
2557
+ (L2,3)(l2,3)=K2,3
2558
+
2559
+ I2,3
2560
+ 2
2561
+ ,I2,3
2562
+ 3
2563
+ (Ii
2564
+ j)(ki)=Ki
2565
+ cK1,L1,I1
2566
+ 3,K2,3,L2,3,I2,3
2567
+ 2
2568
+ ,I2,3
2569
+ 3
2570
+
2571
+ f1, h(L1)(1) ⊗ hL2,3
2572
+
2573
+ ⟨f2⟩L1×I2,3
2574
+ 2 ⟨f3, hI3⟩,
2575
+ (5.6)
2576
+ where
2577
+ cK1,L1,I1
2578
+ 3,K2,3,L2,3,I2,3
2579
+ 2
2580
+ ,I2,3
2581
+ 3
2582
+ =
2583
+
2584
+ I1
2585
+ 1,I1
2586
+ 2
2587
+ (I1
2588
+ j )(k1)=K1
2589
+
2590
+ I2,3
2591
+ 1
2592
+ ⊂L2,3
2593
+ (Ii
2594
+ 1)(ki)=Ki
2595
+ bK,(Ij)⟨h(L1)(1)×L2,3⟩I1
2596
+ 3×I2,3
2597
+ 1 .
2598
+ Moreover, we have
2599
+ |cK1,L1,I1
2600
+ 3,K2,3,L2,3,I2,3
2601
+ 2
2602
+ ,I2,3
2603
+ 3 | ≤ |(L1)(1)|
2604
+ 3
2605
+ 2|I1
2606
+ 3|
2607
+ 1
2608
+ 2
2609
+ |(L1)(1)|2
2610
+ × |L2,3|
2611
+ 1
2612
+ 2|I2,3
2613
+ 2 ||I2,3
2614
+ 3 |
2615
+ 1
2616
+ 2
2617
+ |K2,3|2
2618
+ .
2619
+ Notice that this is the right normalization for (5.2), since f2 is related to L1 and |(L1)(1)| =
2620
+ 2|L1|, and we can change the averages into pairings against non-cancellative Haar func-
2621
+ tions.
2622
+
2623
+ 24
2624
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
2625
+ We conclude that for some C ≥ 1 we have
2626
+ C−1(5.6) = ⟨S((0,l2,3),(1,k2,3),(l1+1,k2,3))(f1, f2), f3⟩,
2627
+ where S((0,l2,3),(1,k2,3),(l1+1,k2,3)) is an operator of the desired type and of complexity
2628
+ (0, l2,3), (1, k2,3), (l1 + 1, k2,3).
2629
+ The other term and the other case of Σ1
2630
+ 3,2 − Σ2
2631
+ 2 are analogous.
2632
+ The cases Σ1
2633
+ m1,3 − Σ3
2634
+ m1 are handled almost identically, however, we need to treat
2635
+ 2
2636
+
2637
+ j=1
2638
+ ⟨gj⟩K2,3 −
2639
+ 2
2640
+
2641
+ j=1
2642
+ ⟨gj⟩I2,3
2643
+ 3
2644
+ slightly differently. We expand the rectangles I2,3
2645
+ 3
2646
+ in the one-parameter fashion until we
2647
+ reach the smaller of the cubes K2, K3. Then we continue with one-parameter expansion
2648
+ with only one of the cubes until we reach the bigger of the cubes K2, K3. For example, if
2649
+ k3 > k2, we expand as
2650
+ 2
2651
+
2652
+ j=1
2653
+ ⟨gj⟩K2,3 −
2654
+ 2
2655
+
2656
+ j=1
2657
+ ⟨gj⟩I2,3
2658
+ 3
2659
+ = −
2660
+ k2−1
2661
+
2662
+ l2=0
2663
+
2664
+ ⟨∆(I2,3
2665
+ 3
2666
+ )(l2+1,l2+1)g1⟩(I2,3
2667
+ 3
2668
+ )(l2,l2)⟨g2⟩(I2,3
2669
+ 3
2670
+ )(l2,l2)
2671
+ + ⟨g1⟩(I2,3
2672
+ 3
2673
+ )(l2+1,l2+1)⟨∆(I2,3
2674
+ 3
2675
+ )(l2+1,l2+1)g2⟩(I2,3
2676
+ 3
2677
+ )(l2,l2)
2678
+
2679
+
2680
+ k3−1
2681
+
2682
+ l3=k2
2683
+
2684
+ ⟨EK2∆(I3
2685
+ 3)(l3+1)g1⟩K2×(I3
2686
+ 3)(l3)⟨g2⟩K2×(I3
2687
+ 3)(l3)
2688
+ + ⟨g1⟩K2×(I3
2689
+ 3)(l3+1)⟨EK2∆(I3)(l3+1)g2⟩K2×(I3
2690
+ 3)(l3)
2691
+
2692
+ ,
2693
+ The case k3 ≤ k2 can be expanded similarly. Similarly as in the previous cases, we can
2694
+ now write the terms in the particular form (5.2). For example, related to the latter term,
2695
+
2696
+ K∈Dλ
2697
+
2698
+ L2,3∈D2,3
2699
+ λl,kℓ(K1)
2700
+ L2=K2
2701
+ (L3)(k3−l3)=K3
2702
+
2703
+ (L1)(l1)=K1
2704
+
2705
+ (I1
2706
+ 3)(k1)=K1
2707
+
2708
+ (I2
2709
+ 3)(k2)=K2
2710
+ (I3
2711
+ 3)(l3)=L3
2712
+ cK,L,I3
2713
+
2714
+ f1, 1K1
2715
+ |K1| ⊗ h(L2,3)(0,1)
2716
+ ��
2717
+ f2, hL1 ⊗ 1L2,3
2718
+ |L2,3|
2719
+
2720
+ ⟨f3, hI3⟩,
2721
+ where l3 ∈ {k2, . . . , k3 − 1}, λl,k = 2−k1−k2+l3 and
2722
+ |cK,L,I3| =
2723
+ ���
2724
+
2725
+ I1,I2
2726
+ (Ij)(k)=K
2727
+ I1
2728
+ 2⊂L1
2729
+ aK,(Ij)|I1|⟨hL1 ⊗ hK2×(L3)(1)⟩I1
2730
+ 2×K2×L3
2731
+ ���
2732
+
2733
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
2734
+ 25
2735
+
2736
+
2737
+ I1,I2
2738
+ (Ij)(k)=K
2739
+ I1
2740
+ 2⊂L1
2741
+ |I3|
2742
+ 1
2743
+ 2|I1||I2|
2744
+ |K|2
2745
+ |K2|− 1
2746
+ 2|(L3)(1)|− 1
2747
+ 2 |L1|− 1
2748
+ 2
2749
+ = |L1|
2750
+ 1
2751
+ 2
2752
+ |K1|
2753
+ |I3|
2754
+ 1
2755
+ 2|K2 × (L3)(1)|
2756
+ 3
2757
+ 2
2758
+ |K2 × (L3)(1)|2
2759
+ .
2760
+ This normalization is an absolute constant away from the correct one since we consider
2761
+ that K2 × (L3)(1) is the top rectangle in parameters 2 and 3.
2762
+ Finally, we consider Σ1
2763
+ 3,3 − Σ2
2764
+ 3 − Σ3
2765
+ 3 + Σ4 that equals to
2766
+
2767
+ K∈Dλ
2768
+
2769
+ I1,I2,I3∈DZ
2770
+ I(k)
2771
+ j
2772
+ =K
2773
+ bK,(Ij)
2774
+
2775
+ 2
2776
+
2777
+ j=1
2778
+ ⟨fj⟩K −
2779
+ 2
2780
+
2781
+ j=1
2782
+ ⟨fj⟩I1
2783
+ 3×K2,3 −
2784
+ 2
2785
+
2786
+ j=1
2787
+ ⟨fj⟩K1×I2,3
2788
+ 3
2789
+ +
2790
+ 2
2791
+
2792
+ j=1
2793
+ ⟨fj⟩I3
2794
+
2795
+ ⟨f3, hI3⟩.
2796
+ (5.7)
2797
+ As we already showed, we can expand
2798
+ 2
2799
+
2800
+ j=1
2801
+ ⟨fj⟩K −
2802
+ 2
2803
+
2804
+ j=1
2805
+ ⟨fj⟩I1
2806
+ 3×K2,3
2807
+ = −
2808
+ k1−1
2809
+
2810
+ l1=0
2811
+
2812
+ ⟨∆(I1
2813
+ 3)(l1+1)g1⟩I1
2814
+ 3⟨g2⟩(I1
2815
+ 3)(l1) + ⟨g1⟩(I1
2816
+ 3)(l1+1)⟨∆(I1
2817
+ 3)(l1+1)g2⟩I1
2818
+ 3
2819
+
2820
+ ,
2821
+ where gj = ⟨fj⟩K2,3, and similarly for
2822
+ n
2823
+
2824
+ j=1
2825
+ ⟨fj⟩I3 −
2826
+ 2
2827
+
2828
+ j=1
2829
+ ⟨fj⟩K1×I2,3
2830
+ 3
2831
+ we get same expansion with the positive sign and gj = ⟨fj⟩I2,3
2832
+ 3 .
2833
+ Then we sum the two expansions together and expand in the parameters 2 and 3. That
2834
+ is, we will expand
2835
+ k1−1
2836
+
2837
+ l1=0
2838
+ ⟨h(I1
2839
+ 3 )(l1+1)⟩(I1
2840
+ 3)(l1)
2841
+
2842
+ f1, h(I1
2843
+ 3 )(l1+1) ⊗ 1K2,3
2844
+ |K2,3|
2845
+
2846
+ ⟨f2⟩(I1
2847
+ 3 )(l1)×K2,3
2848
+
2849
+
2850
+ f1, h(I1
2851
+ 3 )(l1+1) ⊗
2852
+ 1I2,3
2853
+ 3
2854
+ |I2,3
2855
+ 3 |
2856
+
2857
+ ⟨f2⟩(I1
2858
+ 3)(l1)×I2,3
2859
+ 3 .
2860
+ Thus, we get, for example when k2 < k3, that
2861
+ k1−1
2862
+
2863
+ l1=0
2864
+ k2−1
2865
+
2866
+ l2=0
2867
+
2868
+ K∈Dλ
2869
+
2870
+ L1∈D1
2871
+ (L1)(k1−l1)=K1
2872
+
2873
+ L2,3∈D2,3
2874
+ 2−l1 ℓ(L1)
2875
+ (L2,3)(k2−l2,k3−l2)=K2,3
2876
+
2877
+ I3∈DZ
2878
+ (I3)(l1,l2,l2)=L
2879
+ × cK,L,I3
2880
+
2881
+ f1, h(L1)(1) ⊗ h0
2882
+ (L2,3)(1,1)
2883
+ ��
2884
+ f2, h0
2885
+ L1 ⊗ h(L2,3)(1,1)
2886
+
2887
+ ⟨f3, hI3⟩
2888
+
2889
+ 26
2890
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
2891
+ +
2892
+ k1−1
2893
+
2894
+ l1=0
2895
+ k3−1
2896
+
2897
+ l3=k2
2898
+
2899
+ K∈Dλ
2900
+
2901
+ L1∈D1
2902
+ (L1)(k1−l1)=K1
2903
+
2904
+ L2,3∈D2−l1−k2+l3 ℓ(L1)
2905
+ L2=K2
2906
+ (L3)(k3−l3)=K3
2907
+
2908
+ I3∈DZ
2909
+ (I3)(l1,k2,l3)=L
2910
+ × cK,L,I3
2911
+
2912
+ f1, h(L1)(1) ⊗ h0
2913
+ (L2,3)(0,1)
2914
+ ��
2915
+ f2, h0
2916
+ L1 ⊗ h(L2,3)(0,1)
2917
+
2918
+ ⟨f3, hI3⟩.
2919
+ Here
2920
+ |cK,L,I3| =
2921
+ ���
2922
+
2923
+ I1,I2∈DZ
2924
+ Ik
2925
+ j =K
2926
+ aK,(Ij)|I1||L1|− 1
2927
+ 2 |(L2,3)(1)|− 1
2928
+ 2 ⟨h(L1)(1) ⊗ h(L2,3)(1)⟩L1×L2,3
2929
+ ���
2930
+ ≤ |I3|
2931
+ 1
2932
+ 2|(L1)(1)|
2933
+ 3
2934
+ 2
2935
+ |(L)(1)|2
2936
+ |L1|− 1
2937
+ 2 |(L2,3)(1)|− 1
2938
+ 2 ∼ |I3|
2939
+ 1
2940
+ 2 |(L1)(1)|
2941
+ 1
2942
+ 2 |L1|
2943
+ 1
2944
+ 2|(L2,3)(1)|
2945
+ 3
2946
+ 2
2947
+ |(L1)(1)|2|(L2,3)(1)|2
2948
+ .
2949
+ We abused notation slightly by (L2,3)(1) meaning both (L2,3)(1,1) and (L2,3)(0,1). The other
2950
+ terms are handled analogously.
2951
+
2952
+ 6. BOUNDEDNESS OF ZYGMUND SHIFTS
2953
+ In this section we prove the boundedness of Zygmund shifts. We first prove the fol-
2954
+ lowing. A collection S is called γ-sparse if there are pairwise disjoint subsets E(S) ⊂ S,
2955
+ S ∈ S , with |E(S)| ≥ γ|S|. Often the precise value of γ is not important and we just talk
2956
+ about sparse collections.
2957
+ 6.1. Proposition. Let λ = 2k for some k ∈ Z and
2958
+ Λ(f1, f2, f3) =
2959
+
2960
+ K∈D2,3
2961
+ λ
2962
+
2963
+ (Ij)(ℓj )=K
2964
+ �3
2965
+ j=1 |Ij|
2966
+ 1
2967
+ 2
2968
+ |K|2
2969
+ |⟨f1, h0
2970
+ I1⟩| · |⟨f2, hI2⟩| · |⟨f3, hI3⟩|.
2971
+ Then there exists a sparse collection S ⊂ D2,3
2972
+ λ
2973
+ such that
2974
+ Λ(f1, f2, f3) ≲ max{k2, k3}
2975
+
2976
+ S∈S
2977
+ |S|
2978
+ 3
2979
+
2980
+ j=1
2981
+ ⟨|fj|⟩S.
2982
+ Proof. The proof is an easy adaptation of the sparseness argument in [17, Section 5]. In
2983
+ fact, we only need to check the validity of
2984
+ Λ(f1, f2, f3) ≲ ∥f1∥Lp∥f2∥Lq∥f3∥Lr,
2985
+ where p, q, r ∈ (1, ∞) and 1/p + 1/q + 1/r = 1. This can be done by direct computation:
2986
+ Λ(f1, f2, f3) ≤
2987
+ ˆ
2988
+ f1
2989
+
2990
+ K∈D2,3
2991
+ λ
2992
+ ⟨|∆ℓ2
2993
+ Kf2|⟩K⟨|∆ℓ3
2994
+ Kf3|⟩K1K
2995
+ ≤ ∥f1∥Lp
2996
+ ���
2997
+
2998
+
2999
+ K∈D2,3
3000
+ λ
3001
+
3002
+ MD2,3
3003
+ λ |∆ℓ2
3004
+ Kf2|
3005
+ �2� 1
3006
+ 2 ���
3007
+ Lq
3008
+ ���
3009
+
3010
+
3011
+ K∈D2,3
3012
+ λ
3013
+
3014
+ MD2,3
3015
+ λ |∆ℓ3
3016
+ Kf3|
3017
+ �2� 1
3018
+ 2 ���
3019
+ Lr
3020
+ ≲ ∥f1∥Lp∥f2∥Lq∥f3∥Lr.
3021
+
3022
+
3023
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
3024
+ 27
3025
+ 6.2. Proposition. Let Qk, k = (k1, k2, k3), be a bilinear Zygmund shift as in Section 2.D, and
3026
+ let 1 < p1, p2 < ∞ and 1
3027
+ 2 < p < ∞ with 1
3028
+ p :=
3029
+ 1
3030
+ p1 + 1
3031
+ p2. Let
3032
+ w1, w2 ∈ Ap(R × R × R),
3033
+ and
3034
+ w := w
3035
+ p
3036
+ p1
3037
+ 1 w
3038
+ p
3039
+ p2
3040
+ 2 .
3041
+ Then, for every η ∈ (0, 1) we have
3042
+ ∥Qk(f1, f2)∥Lp(w) ≲ max
3043
+ i {ki}22k1η∥f1∥Lp1(w1)∥f2∥Lp2(w2).
3044
+ Proof. We prove the weighted boundedness L4(w1)×L4(w2) → L2(w), of the tri-parameter
3045
+ bilinear shifts of Zygmund nature (5.2). We do this with tri-parameter weights wi ∈ A4.
3046
+ We then extrapolate the result to the full bilinear range using the traditional multilinear
3047
+ extrapolation by Grafakos–Martell (and Duoandikoetxea) [4,7]. Our result then follows
3048
+ from Proposition 5.3.
3049
+ Note that if we have I3 ∈ DZ in (5.2), then the related λ in Proposition 6.1 is
3050
+ 2ℓ3
3051
+ 3−ℓ2
3052
+ 3−ℓ1
3053
+ 3|L1|.
3054
+ (For other cases, for instance if I1
3055
+ 1 × I2,3
3056
+ 2
3057
+ ∈ DZ, then λ = 2ℓ3
3058
+ 2−ℓ2
3059
+ 2−ℓ1
3060
+ 1|L1|). Assume v ∈
3061
+ A4,λ(R2); recall that Ap,λ(R2) is defined similarly as Ap(R2) except that the supremum is
3062
+ taken over rectangles R = I × J with |J| = λ|I|. Then
3063
+
3064
+ S∈S
3065
+ |S|
3066
+ 3
3067
+
3068
+ j=1
3069
+ ⟨|fj|⟩S =
3070
+
3071
+ S∈S
3072
+ ⟨|f1|⟩S⟨|f2|⟩S⟨|f3|v−1⟩v
3073
+ Sv(S).
3074
+ Since for any R ∈ S,
3075
+
3076
+ S⊂R
3077
+ S∈S
3078
+ v(S) =
3079
+
3080
+ S⊂R
3081
+ S∈S
3082
+ v(S)
3083
+ |S| |S| ≲
3084
+
3085
+ S⊂R
3086
+ S∈S
3087
+ v(S)
3088
+ |S| |ES| ≤
3089
+ ˆ
3090
+ R
3091
+ MD2,3
3092
+ λ (v1R) ≲[v]A4,λ(R2) v(R),
3093
+ by the Carleson embedding theorem we have
3094
+ (6.3)
3095
+
3096
+ S∈S
3097
+ |S|
3098
+ 3
3099
+
3100
+ j=1
3101
+ ⟨|fj|⟩S ≲[v]A4,λ(R2)
3102
+ ˆ
3103
+ R2 MD2,3
3104
+ λ |f1|MD2,3
3105
+ λ |f2|Mv
3106
+ D2,3
3107
+ λ (|f3|v−1)v.
3108
+ Now, given weights wj ∈ A4(R3), j = 1, 2, we know that w = w1/2
3109
+ 1
3110
+ w1/2
3111
+ 2
3112
+ ∈ A4(R3). We
3113
+ have
3114
+ |⟨S(f1, f2), f3⟩| =
3115
+
3116
+ L1
3117
+
3118
+ (I1
3119
+ j )
3120
+ (ℓ1
3121
+ j )=L1
3122
+ �3
3123
+ j=1 |I1
3124
+ j |
3125
+ 1
3126
+ 2
3127
+ |L1|2
3128
+ Λ(⟨f1, hI1
3129
+ 1⟩, ⟨f2, h0
3130
+ I1
3131
+ 2⟩, ⟨f3, hI1
3132
+ 3 ⟩).
3133
+ Note that ⟨w⟩L1 ∈ A4,λ(R2) with [⟨w⟩L1]A4,λ(R2) ≤ [w]A4 for any λ. Thus, applying (6.3)
3134
+ with v = ⟨w⟩L1 we have
3135
+ |⟨S(f1, f2), f3⟩|
3136
+ ≲ max
3137
+ i
3138
+ {ki}
3139
+
3140
+ L1
3141
+
3142
+ (I1
3143
+ j )
3144
+ (ℓ1
3145
+ j )=L1
3146
+ �3
3147
+ j=1 |I1
3148
+ j |
3149
+ 1
3150
+ 2
3151
+ |L1|2
3152
+ ˆ
3153
+ R2 MD2,3
3154
+ λ ⟨f1, hI1
3155
+ 1⟩MD2,3
3156
+ λ ⟨f2, h0
3157
+ I1
3158
+ 2⟩Mv
3159
+ D2,3
3160
+ λ (⟨f3, hI1
3161
+ 3⟩v−1)v
3162
+
3163
+ 28
3164
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
3165
+ = max
3166
+ i
3167
+ {ki}
3168
+
3169
+ L1
3170
+ ˆ
3171
+ R3⟨MD|∆ℓ1
3172
+ 1
3173
+ L1f1|⟩L1⟨MD|f2|⟩L1
3174
+
3175
+ (I1
3176
+ 3)(ℓ1
3177
+ 3)=L1
3178
+ |I1
3179
+ 3|
3180
+ 1
3181
+ 2 M
3182
+ ⟨w⟩L1
3183
+ D2,3
3184
+ λ
3185
+ (⟨f3, hI1
3186
+ 3⟩⟨w⟩−1
3187
+ L1 ) 1L1
3188
+ |L1|w
3189
+ ≤ max
3190
+ i
3191
+ {ki}
3192
+ ���
3193
+ � �
3194
+ L1
3195
+
3196
+ MD1MD|∆ℓ1
3197
+ 1
3198
+ L1f1|
3199
+ �2� 1
3200
+ 2 ���
3201
+ L4(w1)∥MD1MD|f2|∥L4(w2)
3202
+ ×
3203
+ ���
3204
+ � �
3205
+ L1
3206
+
3207
+
3208
+ (I1
3209
+ 3)(ℓ1
3210
+ 3)=L1
3211
+ |I1
3212
+ 3|
3213
+ 1
3214
+ 2 M
3215
+ ⟨w⟩L1
3216
+ D2,3
3217
+ λ
3218
+ (⟨f3, hI1
3219
+ 3⟩⟨w⟩−1
3220
+ L1 )|L1|−1�2
3221
+ 1L1
3222
+ � 1
3223
+ 2 ���
3224
+ L2(w).
3225
+ By the well-know square function and maximal function estimates we have
3226
+ ���
3227
+ � �
3228
+ L1
3229
+
3230
+ MD1MD|∆ℓ1
3231
+ 1
3232
+ L1f1|
3233
+ �2� 1
3234
+ 2 ���
3235
+ L4(w1) ≲ ∥f1∥L4(w1)
3236
+ and
3237
+ ∥MD1MD|f2|∥L4(w2) ≲ ∥f2∥L4(w2).
3238
+ The estimate of the last term is a bit tricky. By the (one parameter)vector-valued estimates
3239
+ of M
3240
+ ⟨w⟩L1
3241
+ D2,3
3242
+ λ
3243
+ (see e.g. [19, Proposition 4.3] for a bi-parameter version (the proof easily adapts
3244
+ to the one-parameter case)), we have
3245
+ ���
3246
+ � �
3247
+ L1
3248
+
3249
+
3250
+ (I1
3251
+ 3)(ℓ1
3252
+ 3)=L1
3253
+ |I1
3254
+ 3|
3255
+ 1
3256
+ 2M
3257
+ ⟨w⟩L1
3258
+ D2,3
3259
+ λ
3260
+ (⟨f3, hI1
3261
+ 3⟩⟨w⟩−1
3262
+ L1 )|L1|−1�2
3263
+ 1L1
3264
+ � 1
3265
+ 2 ���
3266
+ L2(w)
3267
+ ≤ 2ℓ1
3268
+ 3�����
3269
+ � �
3270
+ L1
3271
+
3272
+
3273
+ (I1
3274
+ 3)(ℓ1
3275
+ 3)=L1
3276
+ |I1
3277
+ 3|
3278
+ s
3279
+ 2M
3280
+ ⟨w⟩L1
3281
+ D2,3
3282
+ λ
3283
+ (⟨f3, hI1
3284
+ 3⟩⟨w⟩−1
3285
+ L1 )s|L1|− s
3286
+ 2
3287
+ � 2
3288
+ s � 1
3289
+ 2���
3290
+ L2(⟨w⟩L1)
3291
+ ≲ 2ℓ1
3292
+ 3��
3293
+ � �
3294
+ L1
3295
+
3296
+
3297
+ (I1
3298
+ 3)(ℓ1
3299
+ 3)=L1
3300
+ |I1
3301
+ 3|
3302
+ s
3303
+ 2��⟨f3, hI1
3304
+ 3⟩⟨w⟩−1
3305
+ L1
3306
+ ��s|L1|− s
3307
+ 2
3308
+ � 2
3309
+ s � 1
3310
+ 2 ���
3311
+ L2(⟨w⟩L1)
3312
+ ≤ 2ℓ1
3313
+ 3��
3314
+ � �
3315
+ L1
3316
+
3317
+
3318
+ (I1
3319
+ 3)(ℓ1
3320
+ 3)=L1
3321
+ |I1
3322
+ 3|
3323
+ 1
3324
+ 2|⟨f3, hI1
3325
+ 3⟩|⟨w⟩−1
3326
+ L1 |L1|− 1
3327
+ 2
3328
+ �2� 1
3329
+ 2 ���
3330
+ L2(⟨w⟩L1)
3331
+ ≲ 2ℓ1
3332
+ 3η∥f3∥L2(w−1),
3333
+ where s = (1/η)′ and in the last step we have used [19, Proposition 5.8]. Thus,
3334
+ ∥S(f1, f2)∥L2(w) ≲ max
3335
+ i
3336
+ {ki}2k1η∥f1∥L4(w1)∥f2∥L4(w2).
3337
+
3338
+ Now we are able to conclude the proof of Theorem 1.2.
3339
+ Proof of Theorem 1.2. By the representation formula discussed in Sections 2.E and 2.F, the
3340
+ coefficient estimates in Section 4 (in particular (4.1)) we get that
3341
+ ⟨T(f1, f2), f3⟩ =CEσ
3342
+
3343
+
3344
+ k1,k2,k3=2
3345
+ (|k| + 1)2ϕ(k)
3346
+
3347
+ I∈DZ(k)
3348
+ ⟨Q(k1,k2,k3)(f1, f2), f3⟩
3349
+ C(|k| + 1)2ϕ(k)
3350
+ .
3351
+
3352
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
3353
+ 29
3354
+ Thus, for p1, p2 ∈ (1, ∞) so that p ∈ (1, ∞), we conclude by Proposition 6.2 that
3355
+ ∥T(f1, f2)∥Lp(w) ≲
3356
+
3357
+
3358
+ k1,k2,k3=2
3359
+ (|k| + 1)2ϕ(k) max
3360
+ i {ki}22k1η∥f1∥Lp1(w1)∥f2∥Lp2(w2)
3361
+ ≲ ∥f1∥Lp1(w1)∥f2∥Lp2(w2),
3362
+ where we need to take η < α1. Consequently, we can now pass the result to the full
3363
+ bilinear range using the traditional multilinear extrapolation [4,7].
3364
+
3365
+ 7. LINEAR COMMUTATORS IN THE ZYGMUND DILATION SETTING
3366
+ In this section we return to the linear theory and complete the following commutator
3367
+ estimate left open by previous results. This requires new and interesting paraproduct
3368
+ estimates. For the context, see the explanation below.
3369
+ 7.1. Theorem. Let b ∈ L1
3370
+ loc and T be a linear CZZ operator as in [14]. Let θ ∈ (0, 1] be the
3371
+ kernel exponent measuring the decay in terms of the Zygmund ratio
3372
+ Dθ(x) :=
3373
+ �|x1x2|
3374
+ |x3|
3375
+ +
3376
+ |x3|
3377
+ |x1x2|
3378
+ �−θ
3379
+ .
3380
+ Then
3381
+ ∥[b, T]∥Lp→Lp ≲ ∥b∥bmoZ
3382
+ whenever p ∈ (1, ∞).
3383
+ Here the definition of the little BMO is given by
3384
+ ∥b∥bmoZ := sup
3385
+ DZ
3386
+ sup
3387
+ R∈DZ
3388
+ 1
3389
+ |R|
3390
+ ˆ
3391
+ R
3392
+ |b(x) − ⟨b⟩R| dx < ∞,
3393
+ where the supremum is over all different collections of Zygmund rectangles DZ and then
3394
+ over all R ∈ DZ.
3395
+ This theorem was previously considered in [5] using the so-called Cauchy trick. That
3396
+ method requires weighted bounds with Zygmund weights. But we now know [14] how
3397
+ delicate such weighted bounds are – weighted bounds with Zygmund weights do not
3398
+ in general hold if θ < 1. However, the commutator bounds are still true – but we need
3399
+ a different proof, presented here. It suffices to prove the boundedness of commutators
3400
+ [b, Qk] for any linear shift Qk of the Zygmund dilation type.
3401
+ For θ = 1 we could use the Cauchy trick and the weighted bounds from [14] – this
3402
+ would give weighted commutator estimates with Zygmund weights.
3403
+ We begin by recording lemmas that we need for the main proofs of this section.
3404
+ 7.2. Lemma. Let b be a locally integrable function. Then the following are equivalent
3405
+ (1) b ∈ bmoDZ,
3406
+ (2)
3407
+ max
3408
+
3409
+ sup
3410
+ I1∈D1 ∥⟨b⟩I1,1∥BMOD2,3
3411
+ ℓ(I1)
3412
+ , ess sup
3413
+ (x2,x3)∈R2 ∥b(·, x2, x3)∥BMO
3414
+
3415
+ < ∞,
3416
+ (3)
3417
+ max
3418
+
3419
+ sup
3420
+ I2∈D2 ∥⟨b⟩I2,2∥BMOD2,3
3421
+ ℓ(I2)
3422
+ , ess sup
3423
+ (x1,x3)∈R2 ∥b(x1, ·, x3)∥BMO
3424
+
3425
+ < ∞.
3426
+
3427
+ 30
3428
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
3429
+ For completeness, we give the proof.
3430
+ Proof. Let us begin showing that bmoZ =⇒ (2) (and by symmetry also (3)). Clearly, for
3431
+ all Zygmund rectangles I = I1 × I2 × I3 ∈ DZ we have
3432
+ ∥b∥bmoZ ≥ 1
3433
+ |I|
3434
+ ˆ
3435
+ I
3436
+ |b − ⟨b⟩I| ≥
3437
+ 1
3438
+ |I2,3|
3439
+ ˆ
3440
+ I2,3 |⟨b⟩I1,1 − ⟨b⟩I|.
3441
+ (7.3)
3442
+ So by uniform boundedness we immediately get
3443
+ ∥⟨b⟩I1,1∥BMOD2,3
3444
+ ℓ(I1)
3445
+ :=
3446
+ sup
3447
+ I2,3∈D2,3
3448
+ ℓ(I1)
3449
+ 1
3450
+ |I2,3|
3451
+ ˆ
3452
+ I2,3 |⟨b⟩I1,1 − ⟨⟨b⟩I1,1⟩I2,3| ≤ ∥b∥bmoZ < ∞.
3453
+ We move on to proving the second assertion inside (2). For fixed I1 ∈ D1 we define
3454
+ fI1(x2, x3) :=
3455
+ ´
3456
+ I1 |b(x1, x2, x3) − ⟨b⟩I1(x2, x3)| dx1. Then for every I2,3 ∈ D2,3
3457
+ ℓ(I1) we have
3458
+ ⟨fI1⟩I2,3 ≤
3459
+ 1
3460
+ |I2,3|
3461
+ ˆ
3462
+ I2,3
3463
+ ˆ
3464
+ I1 |b − ⟨b⟩I| +
3465
+ 1
3466
+ |I2,3|
3467
+ ˆ
3468
+ I2,3
3469
+ ˆ
3470
+ I1 |⟨b⟩I1,1 − ⟨b⟩I| ≤ 2|I1|∥b∥bmoZ,
3471
+ where last inequality holds by definition and the above estimate (7.3). Now, by the
3472
+ Lebesgue differentiation theorem we get for (x2, x3) ∈ R2 \ N(I1), where N(I1) is a
3473
+ null set depending on I1, that
3474
+ fI1(x2, x3) ≤ 2|I1|∥b∥bmoZ.
3475
+ It is then easy to conclude that
3476
+ ∥b(·, x2, x3)∥BMO ≤ 2∥b∥bmoZ
3477
+ for almost every (x2, x3) ∈ R2.
3478
+ Conversely,
3479
+ ˆ
3480
+ I
3481
+ |b − ⟨b⟩I| ≤
3482
+ ˆ
3483
+ I
3484
+ |b − ⟨b⟩I1,1| +
3485
+ ˆ
3486
+ I
3487
+ |⟨b⟩I1,1 − ⟨b⟩I|
3488
+ ≤ |I1|
3489
+ ˆ
3490
+ I2,3 ∥b(·, x2, x3)∥BMO + |I|∥⟨b⟩I1,1∥BMOℓ(I1) ≤ |I|(C1 + C2),
3491
+ where C1 := ess sup(x2,x3)∈R2 ∥b(·, x2, x3)∥BMO and C2 := supI1 ∥⟨b⟩I1,1∥BMOℓ(I1).
3492
+
3493
+ Then the usual duality results imply the following.
3494
+ 7.4. Corollary. If b ∈ bmoZ and I1 is fixed, then
3495
+
3496
+ I2,3∈D2,3
3497
+ ℓ(I1)
3498
+ ⟨⟨b⟩I1, hI2,3⟩ϕI2,3 ≲ ∥b∥bmoZ
3499
+ ���
3500
+
3501
+
3502
+ I2,3∈D2,3
3503
+ ℓ(I1)
3504
+ ϕI2,3 1I2,3
3505
+ |I2,3|
3506
+ � 1
3507
+ 2���
3508
+ L1.
3509
+ Also, for fixed (x2, x3), we have
3510
+
3511
+ I1∈D1
3512
+ ⟨b, hI1⟩1ϕI1 ≲ ∥b∥bmoZ
3513
+ ���
3514
+ � �
3515
+ I1∈D1
3516
+ ϕI1 1I1
3517
+ |I1|
3518
+ � 1
3519
+ 2���
3520
+ L1.
3521
+ Using the duality type estimates we can use the square function lower bounds to prove
3522
+ the inclusion of product type spaces.
3523
+
3524
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
3525
+ 31
3526
+ 7.5. Definition. Given a lattice of Zygmund rectangles DZ and a sequence of scalars
3527
+ B = (bI)I∈DZ we define
3528
+ ∥B∥BMOprod := sup
3529
+
3530
+
3531
+ 1
3532
+ |Ω|
3533
+
3534
+ I∈DZ
3535
+ I⊂Ω
3536
+ |bI|2
3537
+ � 1
3538
+ 2
3539
+ .
3540
+ The inclusion of the little BMO space can be easily seen from the duality estimate
3541
+ (7.6)
3542
+ ∥B∥BMOprod ∼ sup
3543
+ � �
3544
+ I∈DZ
3545
+ |aI||bI|:
3546
+ ���
3547
+ � �
3548
+ I∈DZ
3549
+ |aI|2 1I
3550
+ |I|
3551
+ � 1
3552
+ 2 ���
3553
+ L1 ≤ 1
3554
+
3555
+ .
3556
+ 7.A. Paraproduct expansions. Here the correct expansions style is the Zygmund mar-
3557
+ tingale expansion similar to [14, Equation (5.22)]. This gives
3558
+ bf =
3559
+
3560
+ I∈DZ
3561
+
3562
+ ∆I,Zb∆I,Zf + ∆I,Zb∆I1EI2,3f + ∆I1EI2,3b∆I,Zf
3563
+ (7.7)
3564
+ + ∆I,ZbEI1∆I2,3f + ∆I,ZbEI1EI2,3f + ∆I1EI2,3bEI1∆I2,3f
3565
+ + EI1∆I2,3b∆I,Zf + EI1∆I2,3b∆I1EI2,3f + EI1EI2,3b∆I,Zf
3566
+
3567
+ =:
3568
+ 3
3569
+
3570
+ i,j=1
3571
+ ai,j(b, f),
3572
+ where, for example, a1,1 = �
3573
+ I∈DZ ∆I,Zb∆I,Zf and
3574
+ a1,2 =
3575
+
3576
+ I∈DZ
3577
+ ∆I,Zb∆I1EI2,3f,
3578
+ i.e., interpret so that rows correspond to the first index i and columns correspond with
3579
+ the second index j.
3580
+ 7.8. Lemma. If b ∈ bmoZ, then the paraproducts ai,j such that (i, j) ̸= (3, 3) are bounded. That
3581
+ is,
3582
+ ∥ai,j(b, f)∥Lp ≲ ∥b∥bmoZ∥f∥Lp,
3583
+ 1 < p < ∞.
3584
+ Proof. Case 1: product type i ̸= 3 ̸= j. We begin with the paraproducts where it would
3585
+ suffice to have a product BMO type assumption (but recall that little BMO is a subset).
3586
+ The symmetry Π = a1,1 is essentially trivial. By (7.6) we have
3587
+ |⟨Πf, g⟩| ≲
3588
+ ���
3589
+ � �
3590
+ I∈DZ
3591
+ |⟨f, hI,Z⟩|2⟨|g|⟩2
3592
+ I
3593
+ 1I
3594
+ |I|
3595
+ � 1
3596
+ 2���
3597
+ L1
3598
+
3599
+ ���
3600
+ � �
3601
+ I∈DZ
3602
+ ⟨|∆I,Zf|⟩2
3603
+ I1I
3604
+ � 1
3605
+ 2 ���
3606
+ Lp∥MZg∥Lp′
3607
+
3608
+ ���
3609
+ � �
3610
+ I∈DZ
3611
+ MZ(∆I,Zf)2� 1
3612
+ 2 ���
3613
+ Lp∥g∥Lp′
3614
+ ≲ ∥SZf∥Lp∥g∥Lp′ ≲ ∥f∥Lp∥g∥Lp′ .
3615
+
3616
+ 32
3617
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
3618
+ The ‘twisted’ case Π = a1,2 (and the symmetrical a2,1) is trickier. Indeed, to decouple
3619
+ f and g we cannot blindly take maximal functions only in some parameters – this would
3620
+ break the Zygmund structure. In any case, we begin with the application of (7.6) to get
3621
+ |⟨Πf, g⟩| ≲
3622
+ ���
3623
+ � �
3624
+ I∈DZ
3625
+ ���
3626
+
3627
+ f, 1I1
3628
+ |I1| ⊗ hI2×I3
3629
+ ��
3630
+ g, hI1 ⊗
3631
+ 1I2×I3
3632
+ |I2 × I3|
3633
+ ����
3634
+ 2 1I
3635
+ |I|
3636
+ � 1
3637
+ 2 ���
3638
+ L1.
3639
+ The above is an L1 norm, while L2 would be nice. This is where A∞ extrapolation
3640
+ comes in. We fix ν ∈ A∞,Z, and move to estimate
3641
+ ���
3642
+ � �
3643
+ I∈DZ
3644
+ ���
3645
+
3646
+ f, 1I1
3647
+ |I1| ⊗ hI2×I3
3648
+ ��
3649
+ g, hI1 ⊗
3650
+ 1I2×I3
3651
+ |I2 × I3|
3652
+ ����
3653
+ 2 1I
3654
+ |I|
3655
+ � 1
3656
+ 2 ���
3657
+ L2(ν).
3658
+ We will soon show that
3659
+ ���
3660
+ � �
3661
+ I∈DZ
3662
+ ���
3663
+
3664
+ f, 1I1
3665
+ |I1| ⊗ hI2×I3
3666
+ ��
3667
+ g, hI1 ⊗
3668
+ 1I2×I3
3669
+ |I2 × I3|
3670
+ ����
3671
+ 2 1I
3672
+ |I|
3673
+ � 1
3674
+ 2 ���
3675
+ L2(ν)
3676
+
3677
+ ���MZf
3678
+ � �
3679
+ I1∈D1
3680
+ MZ(∆I1g)2�1/2���
3681
+ L2(ν).
3682
+ (7.9)
3683
+ The A∞ extrapolation, Theorem 7.10, then implies that this inequality holds also in Lp(ν),
3684
+ p ∈ (0, ∞), ν ∈ A∞,Z. We take p = 1 and ν ≡ 1 to get that
3685
+ |⟨Πf, g⟩| ≲
3686
+ ���MZf
3687
+ � �
3688
+ I1∈D1
3689
+ MZ(∆I1g)2�1/2���
3690
+ L1
3691
+ ≤ ∥MZf∥Lp
3692
+ ���
3693
+ � �
3694
+ I1∈D1
3695
+ MZ(∆I1g)2�1/2���
3696
+ Lp′ ≲ ∥f∥Lp∥g∥Lp′ .
3697
+ It remains to prove (7.9). We write
3698
+ ���
3699
+ � �
3700
+ I∈DZ
3701
+ ���
3702
+
3703
+ f, 1I1
3704
+ |I1| ⊗ hI2×I3
3705
+ ��
3706
+ g, hI1 ⊗
3707
+ 1I2×I3
3708
+ |I2 × I3|
3709
+ ����
3710
+ 2 1I
3711
+ |I|
3712
+ � 1
3713
+ 2���
3714
+ 2
3715
+ L2(ν)
3716
+ =
3717
+
3718
+ I1∈D1
3719
+
3720
+ I2×I3∈D2,3
3721
+ ℓ(I1)
3722
+ ���
3723
+
3724
+ f, 1I1
3725
+ |I1| ⊗ hI2×I3
3726
+ ����
3727
+ 2���
3728
+
3729
+ g, hI1 ⊗
3730
+ 1I2×I3
3731
+ |I2 × I3|
3732
+ ����
3733
+ 2
3734
+ ⟨ν⟩I.
3735
+ Fix some I1 ∈ D1. Let I2
3736
+ 0 × I3
3737
+ 0 ∈ D2,3
3738
+ ℓ(I1) and suppose ϕ1, ϕ2 and ϕ3 are locally inte-
3739
+ grable functions in R2. Then, there exists a sparse collection S = S(I2
3740
+ 0 × I3
3741
+ 0, ϕ1, ϕ2, ϕ3) ⊂
3742
+ D2,3
3743
+ ℓ(I1)(I2
3744
+ 0 × I3
3745
+ 0) so that
3746
+
3747
+ I2×I3∈D2,3
3748
+ ℓ(I1)
3749
+ I2×I3⊂I2
3750
+ 0×I3
3751
+ 0
3752
+ |⟨ϕ1, hI2×I3⟩|2|⟨ϕ2⟩I2×I3|2⟨ϕ3⟩I2×I3 ≲
3753
+
3754
+ Q∈S
3755
+ ⟨|ϕ1|⟩2
3756
+ Q⟨|ϕ2|⟩2
3757
+ Q⟨|ϕ3|⟩Q|Q|.
3758
+
3759
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
3760
+ 33
3761
+ We use this with the functions ϕ1 = ⟨f⟩I1, ϕ2 = ⟨g, hI1⟩ and ϕ3 = ⟨ν⟩I1 to have that for
3762
+ some sparse collection S = S(I1, I2
3763
+ 0 × I3
3764
+ 0, f, g, ν) ⊂ D2,3
3765
+ ℓ(I1) there holds that
3766
+
3767
+ I2×I3∈D2,3
3768
+ ℓ(I1)
3769
+ I2×I3⊂I2
3770
+ 0×I3
3771
+ 0
3772
+ ���
3773
+
3774
+ f, 1I1
3775
+ |I1| ⊗ hI2×I3
3776
+ ����
3777
+ 2���
3778
+
3779
+ g, hI1 ⊗
3780
+ 1I2×I3
3781
+ |I2 × I3|
3782
+ ����
3783
+ 2
3784
+ ⟨ν⟩I
3785
+
3786
+
3787
+ Q∈S
3788
+ ⟨|⟨f⟩I1|⟩2
3789
+ Q⟨|⟨g, hI1⟩|⟩2
3790
+ Q⟨ν⟩I1(Q)
3791
+
3792
+
3793
+ Q∈S
3794
+ ���
3795
+ M2,3
3796
+ ℓ(I1)⟨f⟩I1
3797
+ ��
3798
+ M2,3
3799
+ ℓ(I1)⟨g, hI1⟩
3800
+ ��⟨ν⟩I1
3801
+ Q
3802
+ �2
3803
+ ⟨ν⟩I1(Q)
3804
+
3805
+ ˆ
3806
+ R2
3807
+
3808
+ M2,3
3809
+ ℓ(I1)⟨f⟩I1
3810
+ �2�
3811
+ M2,3
3812
+ ℓ(I1)⟨g, hI1⟩
3813
+ �2⟨ν⟩I1,
3814
+ where in the last step we used the fact that ⟨ν⟩I1 ∈ A∞,ℓ(I1)(R2) and the Carleson embed-
3815
+ ding theorem.
3816
+ Since the last estimate holds uniformly for every I2
3817
+ 0 × I3
3818
+ 0 ∈ D2,3
3819
+ ℓ(I1), we get that
3820
+
3821
+ I1∈D1
3822
+
3823
+ I2×I3∈D2,3
3824
+ ℓ(I1)
3825
+ ���
3826
+
3827
+ f, 1I1
3828
+ |I1| ⊗ hI2×I3
3829
+ ����
3830
+ 2���
3831
+
3832
+ g, hI1 ⊗
3833
+ 1I2×I3
3834
+ |I2 × I3|
3835
+ ����
3836
+ 2
3837
+ ⟨ν⟩I
3838
+
3839
+
3840
+ I1∈D1
3841
+ ˆ
3842
+ R2
3843
+
3844
+ M2,3
3845
+ ℓ(I1)⟨f⟩I1
3846
+ �2�
3847
+ M2,3
3848
+ ℓ(I1)⟨g, hI1⟩
3849
+ �2⟨ν⟩I1
3850
+
3851
+
3852
+ I1∈D1
3853
+ ˆ
3854
+ R3
3855
+
3856
+ M2,3
3857
+ ℓ(I1)⟨f⟩I1
3858
+ �2�
3859
+ M2,3
3860
+ ℓ(I1)⟨|∆I1g|⟩I1
3861
+ �21I1ν
3862
+
3863
+ ˆ
3864
+ R2[MZf]2 �
3865
+ I1∈D1
3866
+ MZ(∆I1g)2ν.
3867
+ Thus, (7.9) is proved.
3868
+ Case 2: little BMO paraproducts (i = 3, j = 1, 2 or i = 1, 2, j = 3). Actually, now we only
3869
+ have “trivial” type cases with different twist. Symmetries a1,3 and a3,1 are similar as well
3870
+ as a2,3 and a3,2. Let us choose for example Π = a1,3 first. By Corollary 7.4 we have
3871
+ |⟨Π(b, f), g⟩| ≲
3872
+ ���
3873
+ � �
3874
+ I1∈D1
3875
+
3876
+
3877
+ I2,3∈D2,3
3878
+ ℓ(I1)
3879
+ |⟨f, hI,Z⟩||⟨g, hI1hI1 ⊗ hI2,3⟩I| 1I2,3
3880
+ |I2,3|
3881
+ �2 1I1
3882
+ |I1|
3883
+ � 1
3884
+ 2 ���
3885
+ L1.
3886
+ Now we again can use similar sparse method as above and for fixed I1 prove
3887
+ ˆ
3888
+
3889
+ I2,3∈Dℓ(I1)
3890
+ |⟨f, hI,Z⟩||⟨g, hI1hI1 ⊗ hI2,3⟩I| 1I2,3
3891
+ |I2,3|⟨ν⟩I1
3892
+
3893
+ ˆ
3894
+ M2,3
3895
+ ℓ(I1)(⟨|∆I1f|⟩I1)M2,3
3896
+ ℓ(I1)⟨g⟩I11I1ν.
3897
+
3898
+ 34
3899
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
3900
+ The above estimate together with vector-valued version of Theorem 7.10 (proven in [3]
3901
+ for general Muckenhoupt basis) yields
3902
+ ���
3903
+ � �
3904
+ I1∈D1
3905
+
3906
+
3907
+ I2,3∈Dℓ(I1)
3908
+ |⟨f, hI,Z⟩||⟨g, hI1hI1 ⊗ hI2,3⟩I| 1I2,3
3909
+ |I2,3|
3910
+ �2 1I1
3911
+ |I1|
3912
+ � 1
3913
+ 2���
3914
+ L1
3915
+
3916
+ ���
3917
+ � �
3918
+ I1∈D1
3919
+ MZ(∆I1f)2 1I1
3920
+ |I1|
3921
+ � 1
3922
+ 2MZg
3923
+ ���
3924
+ L1
3925
+
3926
+ ���
3927
+ � �
3928
+ I1∈D1
3929
+ MZ(∆I1f)2�1/2���
3930
+ Lp∥MZg∥Lp′ ≲ ∥f∥Lp∥g∥Lp′ .
3931
+ Moving to the symmetry Π = a3,2 we first get
3932
+ |⟨Π(b, f), g⟩|
3933
+ =
3934
+ ���
3935
+
3936
+ I∈DZ
3937
+ ⟨⟨b⟩I1, hI2,3⟩
3938
+
3939
+ f, hI1 ⊗ 1I2,3
3940
+ |I2,3|
3941
+
3942
+ ⟨g, hI,Z⟩
3943
+ ���
3944
+ ≲ ∥b∥bmoZ
3945
+ ���
3946
+
3947
+ I1∈D1
3948
+
3949
+
3950
+ I2,3∈Dℓ(I1)
3951
+ |⟨f, hI1 ⊗ 1I2,3
3952
+ |I2,3|⟩|2|⟨g, hI,Z⟩|2 1I2,3
3953
+ |I2,3|
3954
+ � 1
3955
+ 2 1I1
3956
+ |I1|
3957
+ ���
3958
+ L1,
3959
+ where we use the other estimate in Corollary 7.4. Like above, we continue as follows
3960
+ ���
3961
+
3962
+ I1∈D1
3963
+
3964
+
3965
+ I2,3∈Dℓ(I1)
3966
+ |⟨f, hI1 ⊗ 1I2,3
3967
+ |I2,3|⟩|2|⟨g, hI,Z⟩|2 1I2,3
3968
+ |I2,3|
3969
+ � 1
3970
+ 2 1I1
3971
+ |I1|
3972
+ ���
3973
+ L1
3974
+
3975
+ ���
3976
+
3977
+ I1∈D1
3978
+ M2,3
3979
+ ℓ(I1)⟨|∆I1f|⟩I1M2,3
3980
+ ℓ(I1)⟨|∆I1g|⟩I11I1
3981
+ ���
3982
+ L1
3983
+
3984
+ ���
3985
+ � �
3986
+ I1∈D1
3987
+ MZ(∆I1f)2�1/2���
3988
+ Lp
3989
+ ���
3990
+ � �
3991
+ I1∈D1
3992
+ MZ(∆I1g)2�1/2���
3993
+ Lp′
3994
+ ≲ ∥f∥Lp∥g∥Lp′ .
3995
+
3996
+ In above proof we needed the A∞ extrapolation with Zygmund A∞ weights. In fact,
3997
+ we give a very simple proof of A∞ extrapolation [3] in general.
3998
+ 7.10. Theorem. Let (f, g) be a pair of non-negative functions. Assume that there is some 0 <
3999
+ p0 < ∞ such that for all w ∈ A∞,Z there holds
4000
+ ˆ
4001
+ f p0w ≤ C([w]A∞,Z)
4002
+ ˆ
4003
+ gp0w,
4004
+ where C is an increasing function. Then for all 0 < p < ∞ and all w ∈ A∞,Z there holds
4005
+ ˆ
4006
+ f pw ≤ C([w]A∞,Z)
4007
+ ˆ
4008
+ gpw.
4009
+ Proof. We have for all 1 < r < ∞ and all w ∈ Ar,Z that
4010
+ ˆ
4011
+ (f p0/r)rw ≤ C([w]Ar,Z)
4012
+ ˆ
4013
+ (gp0/r)rw.
4014
+
4015
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
4016
+ 35
4017
+ Thus, by the classical extrapolation with Ap,Z weights we have
4018
+ (7.11)
4019
+ ˆ
4020
+ (f p0/r)sw ≤ C([w]As,Z)
4021
+ ˆ
4022
+ (gp0/r)sw
4023
+ for all 1 < s < ∞ and w ∈ As,Z.
4024
+ Finally, let 0 < p < ∞ and w ∈ A∞,Z. Then, there exists some 1 < s0 < ∞ such that
4025
+ w ∈ As0,Z. Choose some 1 < r < ∞ and s0 ≤ s < ∞ such that
4026
+ sp0/r = p.
4027
+ For example, we can take
4028
+ s = s0p
4029
+ p0
4030
+ �p0
4031
+ p + 1
4032
+
4033
+ = s0
4034
+ � p
4035
+ p0
4036
+ + 1
4037
+
4038
+ ,
4039
+ r = s0
4040
+ �p0
4041
+ p + 1
4042
+
4043
+ .
4044
+ Since As0,Z ⊂ As,Z, we can use (7.11) with the exponents s and r to get the claim.
4045
+
4046
+ 7.B. Zygmund shift commutators. Let k = (k1, k2), ki ∈ {0, 1, 2, . . .}, be fixed. A Zyg-
4047
+ mund shift Q = Qk of complexity k, see [14], has the form
4048
+ ⟨Qkf, g⟩
4049
+ =
4050
+
4051
+ K∈D2−k1−k2+k3
4052
+
4053
+ I,J∈DZ
4054
+ I(k)=K=J(k)
4055
+ aIJK⟨f, hI1 ⊗ HI2,3,J2,3⟩⟨g, HI1,J1 ⊗ hJ2,3⟩
4056
+ or
4057
+ ⟨Qkf, g⟩
4058
+ =
4059
+
4060
+ K∈D2−k1−k2+k3
4061
+
4062
+ I,J∈DZ
4063
+ I(k)=K=J(k)
4064
+ aIJK⟨f, hI1 ⊗ hI2,3⟩⟨g, HI1,J1 ⊗ HI2,3,J2,3⟩,
4065
+ where HI,J
4066
+ (1) is supported on I ∪ J and constant on children:
4067
+ HI,J =
4068
+
4069
+ L∈ch(I)∪ch(J)
4070
+ bL1L
4071
+ (2) is L2 normalized: |HI,J| ≤ |I|− 1
4072
+ 2 , and
4073
+ (3) has zero average:
4074
+ ´
4075
+ HI,J = 0.
4076
+ We will be focusing on the mixed type form since it is the most interesting one. Usually
4077
+ the other type is much easier and the method is easily recovered from this case.
4078
+ 7.12. Proposition. Let Qk be a Zygmund shift of complexity k = (k1, k2, k3). Let 1 < p < ∞
4079
+ and b ∈ bmoZ . Then we have
4080
+ ∥[b, Qk]f∥Lp ≲ max(k1, k2, k3)(|k| + 1)2∥b∥bmoZ∥f∥Lp.
4081
+ Proof. We consider the commutator [b, Qk]f : bQkf − Qk(bf) that in the dual form equals
4082
+ to
4083
+
4084
+ K∈D2−k1−k2+k3
4085
+
4086
+ I,J∈DZ
4087
+ I(k)=K=J(k)
4088
+ aIJK
4089
+
4090
+ ⟨bf, hI1 ⊗ HI2,3,J2,3⟩⟨g, HI1,J1 ⊗ hJ2,3⟩
4091
+ −⟨f, hI1 ⊗ HI2,3,J2,3⟩⟨bg, HI1,J1 ⊗ hJ2,3⟩
4092
+
4093
+ .
4094
+
4095
+ 36
4096
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
4097
+ Now, expanding both bf and bg with the expansion (7.7) we get the terms
4098
+ ⟨Qk(ai,j(b, f)), g⟩
4099
+ and
4100
+ ⟨Qkf, ai,j(b, g)⟩
4101
+ whenever (i, j) ̸= (3, 3). These terms are directly bounded separately, in particular, we
4102
+ have Qk : Lp → Lp and ai,j : Lp → Lp. Hence, we are left with bounding
4103
+
4104
+ K∈Dλ
4105
+
4106
+ I,J∈DZ
4107
+ I(k)=K=J(k)
4108
+ aIJK
4109
+ � �
4110
+ L∈DZ
4111
+ ⟨b⟩L⟨∆L,Zf, hI1 ⊗ HI2,3,J2,3⟩⟨g, HI1,J1 ⊗ hJ2,3⟩
4112
+
4113
+
4114
+ L∈DZ
4115
+ ⟨b⟩L⟨f, hI1 ⊗ HI2,3,J2,3⟩⟨∆L,Zg, HI1,J1 ⊗ hJ2,3⟩
4116
+
4117
+ =
4118
+
4119
+ K∈Dλ
4120
+
4121
+ I,J∈DZ
4122
+ I(k)=K=J(k)
4123
+ aIJK
4124
+ ×
4125
+
4126
+
4127
+ L∈DZ
4128
+ ℓ(L1)=2−k1ℓ(K1)
4129
+ ℓ(K2)≤2k2ℓ(L2)≤2max(k2,k3)ℓ(K2)
4130
+ ⟨b⟩L⟨∆L,Zf, hI1 ⊗ HI2,3,J2,3⟩⟨g, HI1,J1 ⊗ hJ2,3⟩
4131
+
4132
+
4133
+ Q∈DZ
4134
+ Q1⊂K1, ℓ(Q1)≥ℓ(I1)
4135
+ 2−k1ℓ(K2)≤2k2ℓ(Q2)≤ℓ(K2)
4136
+ ⟨b⟩Q⟨f, hI1 ⊗ HI2,3,J2,3⟩⟨∆Q,Zg, HI1,J1 ⊗ hJ2,3⟩
4137
+
4138
+ ,
4139
+ where we have abbreviated 2−k1−k2+K3 by λ. Now, we write
4140
+ ⟨f, hI1 ⊗ HI2,3,J2,3⟩ =
4141
+
4142
+ L∈DZ
4143
+ ℓ(L1)=2−k1ℓ(K1)
4144
+ ℓ(K2)≤2k2ℓ(L2)≤2max(k2,k3)ℓ(K2)
4145
+ ⟨∆L,Zf, hI1 ⊗ HI2,3,J2,3⟩
4146
+ and
4147
+ ⟨g, HI1,J1 ⊗ hJ2,3⟩ =
4148
+
4149
+ Q∈DZ
4150
+ Q1⊂K1, ℓ(Q1)≥ℓ(I1)
4151
+ 2−k1ℓ(K2)≤2k2ℓ(Q2)≤ℓ(K2)
4152
+ ⟨∆Q,Zg, HI1,J1 ⊗ hJ2,3⟩
4153
+ for the unexpanded terms. Thus, we end up with
4154
+
4155
+ K∈Dλ
4156
+
4157
+ I,J∈DZ
4158
+ I(k)=K=J(k)
4159
+ aIJK
4160
+
4161
+ L∈DZ
4162
+ ℓ(L1)=2−k1ℓ(K1)
4163
+ ℓ(K2)≤2k2ℓ(L2)≤2max(k2,k3)ℓ(K2)
4164
+
4165
+ Q∈DZ
4166
+ Q1⊂K1, ℓ(Q1)≥ℓ(I1)
4167
+ 2−k1ℓ(K2)≤2k2ℓ(Q2)≤ℓ(K2)
4168
+ ×
4169
+
4170
+ (⟨b⟩L − ⟨b⟩Q)⟨∆L,Zf, hI1 ⊗ HI2,3,J2,3⟩⟨∆Q,Zg, HI1,J1 ⊗ hJ2,3⟩
4171
+
4172
+ .
4173
+ We write explicitly the complexity levels for Q and L. That is, in the above summations
4174
+ we have (L2)(l2) = (K2)(max(0,k3−k2)) for some l2 ∈ {0, . . . , max(k2, k3)}, (Q1)(q1) = K1,
4175
+
4176
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
4177
+ 37
4178
+ for some q1 ∈ {0, . . . , k1}, and (Q2)(q2) = K2 for some q2 ∈ {k2, . . . , k2 + k1}. We get
4179
+
4180
+ K∈Dλ
4181
+
4182
+ I,J∈DZ
4183
+ I(k)=K=J(k)
4184
+ aIJK
4185
+ max(k2,k3)
4186
+
4187
+ l2=0
4188
+
4189
+ q1∈{0,...,k1}
4190
+ q2∈{k2,...,k2+k1}
4191
+
4192
+ L∈DZ
4193
+ ℓ(L1)=2−k1ℓ(K1)
4194
+ (L2)(l2)=(K2)(max(0,k3−k2))
4195
+
4196
+ Q∈DZ
4197
+ (Q1)(q1)=K1
4198
+ (Q2)(q2)=K2
4199
+ ×
4200
+
4201
+ (⟨b⟩L − ⟨b⟩Q)⟨∆L,Zf, hI1 ⊗ HI2,3,J2,3⟩⟨∆Q,Zg, HI1,J1 ⊗ hJ2,3⟩
4202
+
4203
+ .
4204
+ Here we need to notice that R = R1 × R2 × R3 ⊃ K, L, Q, where
4205
+ R = K(k1,max(0,k3−k2),k1+max(k2−k3,0)) and R ∈ DZ.
4206
+ This is a common “Zygmund ancestor” for all of these rectangles.
4207
+ Let us expand in the difference ⟨b⟩L − ⟨b⟩Q in the following way
4208
+ ⟨b⟩L = ⟨b⟩L − ⟨b⟩L(0,1,1)
4209
+ + ⟨b⟩L(0,1,1) − ⟨b⟩L(0,2,2)
4210
+ ...
4211
+ + ⟨b⟩L(0,l2−1,l2−1) − ⟨b⟩L(0,l2,l2) + ⟨b⟩L(0,l2,l2)
4212
+ =
4213
+ l2−1
4214
+
4215
+ r2=0
4216
+
4217
+ ⟨b⟩L(0,r2,r2) − ⟨b⟩L(0,r2+1,r2+1)
4218
+
4219
+ + ⟨b⟩L(0,l2,l2).
4220
+ Notice that since ℓ(L1)ℓ(L2) = ℓ(L3), we have ℓ(L1)ℓ((L2)(r2)) = ℓ((L3)(r2)), i.e. rectan-
4221
+ gles (L2)(r2) × (L3)(r2) ∈ Dℓ(L1) which is desirable since we want to use the characteriza-
4222
+ tion (2) in Lemma 7.2. We continue with the last term
4223
+ ⟨b⟩L(0,l2,l2) = ⟨b⟩L(0,l2,l2) − ⟨b⟩L(1,l2,1+l2)
4224
+ + ⟨b⟩L(1,l2,1+l2) − ⟨b⟩L(2,l2,2+l2)
4225
+ ...
4226
+ ⟨b⟩L(k1−1,l2,k1−1+l2) − ⟨b⟩L(k1,l2,k1+l2) + ⟨b⟩G
4227
+ =
4228
+ k1−1
4229
+
4230
+ r1=0
4231
+
4232
+ ⟨b⟩L(r1,l2,r1+l2) − ⟨b⟩L(r1+1,l2,r1+1+l2)
4233
+
4234
+ + ⟨b⟩R.
4235
+ Recall that (L2)(l2) = (K2)(max(0,k3−k2)) =: R2 and observe that since ℓ((L3)(k1+l2)) =
4236
+ ℓ((L2)(l2))ℓ((L1)(k1)) = ℓ(R2)ℓ(K1) we get (L3)(k1+l2) = R3. Thus, we end up with a sum
4237
+ of terms of the forms
4238
+ ⟨b⟩L(0,r2,r2) − ⟨b⟩L(0,r2+1,r2+1)
4239
+ and
4240
+ ⟨b⟩L(r1,l2,r1+l2) − ⟨b⟩L(r1+1,l2,r1+1+l2),
4241
+ (7.13)
4242
+ and we have for fixed r1 and r2
4243
+ |(7.13)| ≲ ∥b∥bmoZ
4244
+ by Lemma 7.2.
4245
+
4246
+ 38
4247
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
4248
+ By the same argument as above we get
4249
+ ⟨b⟩Q =
4250
+ max(0,k3−k2)+q2−1
4251
+
4252
+ ρ2=0
4253
+ ⟨b⟩Q(0,ρ2,ρ2) − ⟨b⟩Q(0,ρ2+1,ρ2+1)
4254
+ +
4255
+ q1
4256
+
4257
+ ρ1=0
4258
+ ⟨b⟩Q(ρ1,�q2,ρ1+�q2) − ⟨b⟩Q(ρ1+1,�q2,ρ1+1+�q2)
4259
+ + ⟨b⟩R,
4260
+ where �q2 = max(0, k3 − k2) + q2,
4261
+ (Q2)(�q2) = (K2)(max(0,k3−k2))
4262
+ and
4263
+ (Q3)(q1+�q2) = (K3)(k1+max(k2−k3,0)).
4264
+ Notice that the last term corresponds to the last term in the previous expansion, and
4265
+ hence, their difference equals to zero. Again, here we have
4266
+ |⟨b⟩Q(0,ρ2,ρ2) − ⟨b⟩Q(0,ρ2+1,ρ2+1) + ⟨b⟩Q(ρ1,�q2,ρ1+�q2) − ⟨b⟩Q(ρ1+1,�q2,ρ1+1+�q2)| ≲ ∥b∥bmoZ
4267
+ for fixed ρ1 and ρ2.
4268
+ Now, we can split the commutator into the two terms
4269
+ Wb
4270
+ K,kf = 1K
4271
+
4272
+ L∈DZ
4273
+ ℓ(L1)=2−k1ℓ(K1)
4274
+ ℓ(K2)≤2k2ℓ(L2)≤2max(k2,k3)ℓ(K2)
4275
+ bL,K∆L,Zf,
4276
+ where
4277
+ |bL,K| ≲ max(k1, k2, k3)∥b∥bmoZ,
4278
+ and
4279
+ Vb
4280
+ K,kg =
4281
+
4282
+ Q∈DZ
4283
+ Q1⊂K1, ℓ(Q1)≥ℓ(I1)
4284
+ 2−k1ℓ(K2)≤2k2ℓ(Q2)≤ℓ(K2)
4285
+ bQ,K∆Q,Zg,
4286
+ where
4287
+ |bQ,K| ≲ max(k1, k2, k3)∥b∥bmoZ.
4288
+ Thus, the last term of the commutator is the sum of
4289
+
4290
+ K∈Dλ
4291
+
4292
+ I,J∈DZ
4293
+ I(k)=K=J(k)
4294
+ aIJK⟨Wb
4295
+ K,kf, hI1 ⊗ HI2,3,J2,3⟩⟨VK,kg, HI1,J1 ⊗ hJ2,3⟩
4296
+ and
4297
+
4298
+ K∈Dλ
4299
+
4300
+ I,J∈DZ
4301
+ I(k)=K=J(k)
4302
+ aIJK⟨WK,kf, hI1 ⊗ HI2,3,J2,3⟩⟨Vb
4303
+ K,kg, HI1,J1 ⊗ hJ2,3⟩.
4304
+ The boundedness follows via standard methods (adapt proofs of [14, Theorem 6.2 and
4305
+ Lemma 5.20].)
4306
+
4307
+
4308
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
4309
+ 39
4310
+ APPENDIX A. BILINEAR FEFFERMAN-PIPHER MULTIPLIERS
4311
+ In this section we consider bilinear variants of multipliers studied by Fefferman-Pipher [6].
4312
+ These considerations motivate the kernel estimates in Section 3. After the presented cal-
4313
+ culations, the reader can easily check how everything fits with Section 3. In fact, we will
4314
+ see that the bilinear Fefferman-Pipher multipliers produce kernels which satisfy the the
4315
+ kernel estimates in Section 3 with
4316
+ θ = 2,
4317
+ α1 = 1,
4318
+ α2,3 = 1,
4319
+ and an extra logarithm factor. In the partial kernel estimates �θ = 1 and there is also a
4320
+ harmless logarithm factor. We leave further analysis of these multipliers for future work.
4321
+ We consider the following multi-parameter dilation on R6 – define
4322
+ ρs,t(x, y) = (sx1, tx2, stx3, sy1, ty2, sty3),
4323
+ s, t > 0,
4324
+ and set
4325
+ A1 := {(ξ, η) ∈ R6 : 1
4326
+ 2 < |(ξ1, η1)| ≤ 1, 1
4327
+ 2 < |(ξ2, ξ3, η2, η3)| ≤ 1}.
4328
+ In this section we consider the parameter groups {1} and {2, 3} only. The grouping
4329
+ {{2}, {1, 3}} is similar, for example, we would set
4330
+ A2 := {(ξ, η) ∈ R6 : 1
4331
+ 2 < |(ξ2, η2)| ≤ 1, 1
4332
+ 2 < |(ξ1, ξ3, η1, η3)| ≤ 1}.
4333
+ For Schwartz functions f1, f2 we define the bilinear multiplier operator
4334
+ Tm,1(f1, f2)(x) =
4335
+ ˆ
4336
+ R3
4337
+ ˆ
4338
+ R3 m(ξ, η) �f1(ξ) �f2(η)e2πix·(ξ+η) dξ dη,
4339
+ where the symbol m ∈ CN is assumed to satisfy
4340
+ ∥m∥M1
4341
+ Z :=
4342
+ sup
4343
+ |α|∞≤N
4344
+ |β|∞≤N
4345
+ sup
4346
+ s,t>0
4347
+ sup
4348
+ (ξ,η)∈A1 |∂α
4349
+ ξ ∂β
4350
+ η (m ◦ ρs,t)(ξ, η)| < ∞.
4351
+ Thus, if (ξ, η) ∈ A1, then by definition
4352
+ |(∂α
4353
+ ξ ∂β
4354
+ η m)(sξ1, tξ2, stξ3, sη1, tη2, stη3)| ≤ ∥m∥M1
4355
+ Zs−α1−β1t−α2−β2(st)−α3−β3
4356
+ (A.1)
4357
+ = ∥m∥M1
4358
+ Zs−(α1+β1)+(α2+β2)(st)−(α2+β2)−(α3+β3).
4359
+ Now, for (ζ1, σ1) ̸= 0 and (ζ2, ζ3, σ2, σ3) ̸= 0 denote
4360
+ s = |(ζ1, σ1)|,
4361
+ st = |(sζ2, ζ3, sσ2, σ3)|,
4362
+ (ξ1, ξ2, ξ3) =
4363
+ �ζ1
4364
+ s , ζ2
4365
+ t , ζ3
4366
+ st
4367
+
4368
+ ,
4369
+ (η1, η2, η3) =
4370
+ �σ1
4371
+ s , σ2
4372
+ t , σ3
4373
+ st
4374
+
4375
+ .
4376
+ Thus, (ξ, η) ∈ A1 and
4377
+ |∂α
4378
+ ζ ∂β
4379
+ σm(ζ, σ)| ≲ ∥m∥M1
4380
+ Z(|ζ1| + |σ1|)−(α1+β1)+(α2+β2)
4381
+ (A.2)
4382
+ ×
4383
+
4384
+ |((|ζ1| + |σ1|)ζ2, ζ3)| + |((|ζ1| + |σ1|)σ2, σ3)|
4385
+ �−(α2+β2)−(α3+β3).
4386
+ We write, with two standard partition of unity φ1 on R2 \{0} and φ2,3 on R4 \{0}, that
4387
+ 1 =
4388
+
4389
+ j,k∈Z
4390
+ φ1(2−jξ1, 2−jη1)φ2,3(2−kξ2, 2−j−kξ3, 2−kη2, 2−j−kη3).
4391
+
4392
+ 40
4393
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
4394
+ Via this identity we obtain
4395
+ m =
4396
+
4397
+ j,k
4398
+ (φ1 ⊗ φ2,3 ◦ ρ2−j,2−k) · m
4399
+ =
4400
+
4401
+ j,k
4402
+ (φ1 ⊗ φ2,3 · (m ◦ ρ2j,2k)) ◦ ρ2−j,2−k =: mj,k.
4403
+ Since φ1 and φ2,3 are supported in ¯B(0, 2) \ B(0, 1
4404
+ 2) in R2 and R4, respectively, we know
4405
+ that
4406
+ spt mj,k ⊂
4407
+ ρ2j,2k
4408
+
4409
+ (ξ, η) : (ξ1, η1) ∈ ¯BR2(0, 2) \ BR2(0, 1
4410
+ 2), (ξ2,3, η2,3) ∈ ¯BR4(0, 2) \ BR4(0, 1
4411
+ 2)
4412
+
4413
+ .
4414
+ Using this we get
4415
+ ∥∂α∂βmj,k∥L∞ ≲ 2−(j,k,j+k)·(α+β)
4416
+ and
4417
+ ∥∂α∂βmj,k∥L1 ≲ 2(j,k,j+k)·(2−(α+β)),
4418
+ where 2 = (2, 2, 2).
4419
+ Let Kj,k(y, z) = ˇmj,k and K(y, z) = �
4420
+ j,k Kj,k(y, z) – then K(x − y, x − z) is the corre-
4421
+ sponding kernel. Using similar analysis as in [14] we have
4422
+ ∥yαz ˜α∂β
4423
+ y ∂γ
4424
+ z Kj,k∥L∞ ≲ ∥∂α
4425
+ ξ ∂ ˜α
4426
+ η (ξβηγmj,k)∥L1
4427
+
4428
+
4429
+ l≤α
4430
+ ˜l≤˜α
4431
+ �α
4432
+ l
4433
+ ��˜α
4434
+ ˜l
4435
+
4436
+ ∥∂l(ξβ)∂
4437
+ ˜l(ηγ) · ∂α−l∂ ˜α−˜lmj,k)∥L1
4438
+ ≲ 2(j,k,j+k)·(2+(β+γ)−(α+˜α))
4439
+ for multi-indices α, ˜α, β, γ. Hence, we get
4440
+ |yβ+1zγ+1∂β
4441
+ y ∂γ
4442
+ z Kj,k(y, z)| ≲ 2(j,k,j+k)·(2+(β+γ)−(α+˜α))|yβ+1−α| · |zγ+1−˜α|.
4443
+ Taking αi, ˜αi ∈ {0, N} we obtain
4444
+ |yβ+1zγ+1∂β
4445
+ y ∂γ
4446
+ z K(y, z)|
4447
+
4448
+
4449
+ j
4450
+ min{(2j|y1|)β1+1, (2j|y1|)β1+1−N} min{(2j|z1|)γ1+1, (2j|z1|)γ1+1−N}
4451
+ ×
4452
+
4453
+ k
4454
+ min{(2k|y2|)β2+1, (2k|y2|)β2+1−N} min{(2k|z2|)γ2+1, (2k|z2|)γ2+1−N}
4455
+ × min{(2j+k|y3|)β3+1, (2j+k|y3|)β3+1−N} min{(2j+k|z3|)γ3+1, (2j+k|z3|)γ3+1−N}.
4456
+ We can estimate the inner sum either by
4457
+
4458
+ k : 2k<1/(|y2|+|z2|)
4459
+ (2k|y2|)β2+1(2k|z2|)γ2+1(2j+k|y3|)β3+1(2j+k|z3|)γ3+1
4460
+ +
4461
+
4462
+ k : 2k≥1/(|y2|+|z2|)≥1/(2|y2|)
4463
+ (2k|y2|)β2+1−N(2k|z2|)γ2+1(2j+k|y3|)β3+1(2j+k|z3|)γ3+1
4464
+ +
4465
+
4466
+ k : 2k≥1/(|y2|+|z2|)>1/(2|z2|)
4467
+ (2k|y2|)β2+1(2k|z2|)γ2+1−N(2j+k|y3|)β3+1(2j+k|z3|)γ3+1
4468
+
4469
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
4470
+ 41
4471
+
4472
+ |y2|β2+1
4473
+ (|y2| + |z2|)β2+1 ·
4474
+ |z2|γ2+1
4475
+ (|y2| + |z2|)γ2+1 ·
4476
+ (2j|y3|)β3+1
4477
+ (|y2| + |z2|)β3+1 ·
4478
+ (2j|z3|)γ3+1
4479
+ (|y2| + |z2|)γ3+1 =: I1
4480
+ or by
4481
+
4482
+ k : 2k<2−j/(|y3|+|z3|)
4483
+ (2k|y2|)β2+1(2k|z2|)γ2+1(2j+k|y3|)β3+1(2j+k|z3|)γ3+1
4484
+ +
4485
+
4486
+ k : 2k≥2−j/(|y3|+|z3|)≥2−j/(2|y3|)
4487
+ (2k|y2|)β2+1(2k|z2|)γ2+1(2j+k|y3|)��3+1−N(2j+k|z3|)γ3+1
4488
+ +
4489
+
4490
+ k : 2k≥2−j/(|y3|+|z3|)>2−j/(2|z3|)
4491
+ (2k|y2|)β2+1(2k|z2|)γ2+1(2j+k|y3|)β3+1(2j+k|z3|)γ3+1−N
4492
+
4493
+ |y2|β2+1
4494
+ [2j(|y3| + |z3|)]β2+1 ·
4495
+ |z2|γ2+1
4496
+ [2j(|y3| + |z3|)]γ2+1 ·
4497
+ |y3|β3+1
4498
+ (|y3| + |z3|)β3+1 ·
4499
+ |z3|γ3+1
4500
+ (|y3| + |z3|)γ3+1 =: I2,
4501
+ in both cases provided that β2 + β3 + γ2 + γ3 < N − 4.
4502
+ The outer sum can then be estimated either by
4503
+
4504
+ j : 2j<1/(|y1|+|z1|)
4505
+ (2j|y1|)β1+1(2j|z1|)γ1+1I1
4506
+ +
4507
+
4508
+ j : 2j≥1/(|y1|+|z1|)≥1/(2|y1|)
4509
+ (2j|y1|)β1+1−N(2j|z1|)γ1+1I1
4510
+ +
4511
+
4512
+ j : 2j≥1/(|y1|+|z1|)>1/(2|z1|)
4513
+ (2j|y1|)β1+1(2j|z1|)γ1+1−NI1
4514
+
4515
+ |y1|β1+1|z1|γ1+1
4516
+ (|y1| + |z1|)β1+γ1+2
4517
+ |y2|β2+1|z2|γ2+1
4518
+ (|y2| + |z2|)β2+γ2+2
4519
+ |y3|β3+1|z3|γ3+1
4520
+ [(|y1| + |z1|)(|y2| + |z2|)]β3+γ3+2
4521
+ or, if (|y1| + |z1|)(|y2| + |z2|) ≤ |y3| + |z3|, by
4522
+
4523
+ j : 2j<(|y2|+|z2|)/(|y3|+|z3|)
4524
+ (2j|y1|)β1+1(2j|z1|)γ1+1I1
4525
+ +
4526
+
4527
+ j : |y2|+|z2|
4528
+ |y3|+|z3| ≤2j≤
4529
+ 1
4530
+ |y1|+|z1|
4531
+ (2j|y1|)β1+1(2j|z1|)γ1+1I2
4532
+ +
4533
+
4534
+ j : 2j>1/(|y1|+|z1|)>1/(2|z1|)
4535
+ (2j|y1|)β1+1(2j|z1|)γ1+1−NI2
4536
+ +
4537
+
4538
+ j : 2j>1/(|y1|+|z1|)>1/(2|y1|)
4539
+ (2j|y1|)β1+1−N(2j|z1|)γ1+1I2 =: I + II + III + IV.
4540
+ It is straightforward that
4541
+ I ∼
4542
+ |y1|β1+1|z1|γ1+1
4543
+ (|y1| + |z1|)β1+γ1+2
4544
+ |y2|β2+1|z2|γ2+1
4545
+ (|y2| + |z2|)β2+γ2+2
4546
+ |y3|β3+1|z3|γ3+1
4547
+ (|y3| + |z3|)β3+γ3+2
4548
+ ×
4549
+ �(|y1| + |z1|)(|y2| + |z2|)
4550
+ |y3| + |z3|
4551
+ �β1+γ1+2
4552
+ ;
4553
+ III ∼ IV ∼
4554
+ |y1|β1+1|z1|γ1+1
4555
+ (|y1| + |z1|)β1+γ1+2
4556
+ |y2|β2+1|z2|γ2+1
4557
+ (|y2| + |z2|)β2+γ2+2
4558
+ |y3|β3+1|z3|γ3+1
4559
+ (|y3| + |z3|)β3+γ3+2
4560
+
4561
+ 42
4562
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
4563
+ ×
4564
+ �(|y1| + |z1|)(|y2| + |z2|)
4565
+ |y3| + |z3|
4566
+ �β2+γ2+2
4567
+ .
4568
+ Lastly, we have
4569
+ II ∼
4570
+ |y1|β1+1|z1|γ1+1
4571
+ (|y1| + |z1|)β1+γ1+2
4572
+ |y2|β2+1|z2|γ2+1
4573
+ (|y2| + |z2|)β2+γ2+2
4574
+ |y3|β3+1|z3|γ3+1
4575
+ (|y3| + |z3|)β3+γ3+2
4576
+ ×
4577
+ �(|y1| + |z1|)(|y2| + |z2|)
4578
+ |y3| + |z3|
4579
+ �min{β1+γ1,β2+γ2}+2
4580
+ Lβ1,β2,γ1,γ2(y, z),
4581
+ where
4582
+ Lβ1,β2,γ1,γ2(y, z) := 1 + log+
4583
+
4584
+ |y3| + |z3|
4585
+ (|y1| + |z1|)(|y2| + |z2|)
4586
+
4587
+ when β1 + γ1 = β2 + γ2 and Lβ1,β2,γ1,γ2(y, z) = 1 otherwise. In conclusion, we get
4588
+ |∂β
4589
+ y ∂γ
4590
+ z K(y, z)| ≲
4591
+ 1
4592
+ [(|y1| + |z1|)(|y2| + |z2|) + |y3| + |z3|]β3+γ3+4
4593
+ ×
4594
+ 1
4595
+ (|y1| + |z1|)β1+γ1(|y2| + |z2|)β2+γ2
4596
+ × min
4597
+
4598
+ 1,
4599
+ �(|y1| + |z1|)(|y2| + |z2|)
4600
+ |y3| + |z3|
4601
+ �min{β1+γ1,β2+γ2}�
4602
+ Lβ1,β2,γ1,γ2(y, z).
4603
+ 1.A. Partial kernel estimates. Let m ∈ M1
4604
+ Z. We define truncations of m by setting
4605
+ mJ :=
4606
+
4607
+ |j|≤J1,|k|≤J2
4608
+ mj,k,
4609
+ J = (J1, J2) ∈ N2.
4610
+ A.3. Lemma. Suppose that m ∈ M1
4611
+ Z. Let mJ be defined as above and let KJ = ˇmJ. Then for
4612
+ (y2, z2) ̸= 0 ̸= (y3, z3) we have the estimate
4613
+ ���
4614
+ ˚
4615
+ I1×I1×I1 ∂β2
4616
+ y2 ∂β3
4617
+ y3 ∂γ2
4618
+ z2 ∂γ3
4619
+ z3 KJ(x1 − y1, y2, y3, x1 − z1, z2, z3) dy1 dz1 dx1
4620
+ ���
4621
+
4622
+ 1
4623
+ (|y2| + |z2|)β2+γ2 ·
4624
+ 1
4625
+ (|y3| + |z3|)β3+γ3 |I1|(|I1|(|y2| + |z2|)
4626
+ |y3| + |z3|
4627
+ +
4628
+ |y3| + |z3|
4629
+ |I1|(|y2| + |z2|))−1
4630
+ ×
4631
+ 1
4632
+ �3
4633
+ i=2(|yi| + |zi|)2 ·
4634
+
4635
+ 1 + log+
4636
+ |y3| + |z3|
4637
+ |I1|(|y2| + |z2|)
4638
+
4639
+ ,
4640
+ where I1 is an interval and β2 + β3 + γ2 + γ3 ≤ 1.
4641
+ Proof. Since mJ(0, ξ2, ξ3, 0, η2, η3) = 0, using the Fourier transform we know that
4642
+ (A.4)
4643
+ ¨
4644
+ R2 ∂β2
4645
+ y2 ∂β3
4646
+ y3 ∂γ2
4647
+ z2 ∂γ3
4648
+ z3 KJ(y1, y2, y3, z1, z2, z3) dy1 dz1 = 0.
4649
+ Suppose first that |I1|(|y2| + |z2|) ≥ |y3| + |z3| – by (A.4) we may equivalently estimate
4650
+ the integral over I1 × (R2 \ (I1 × I1)) instead of I1 × I1 × I1. By the kernel estimates we
4651
+ have ���
4652
+ ˚
4653
+ I1×(R2\(I1×I1))
4654
+ ∂β2
4655
+ y2 ∂β3
4656
+ y3 ∂γ2
4657
+ z2 ∂γ3
4658
+ z3 KJ(x1 − y1, y2, y3, x1 − z1, z2, z3) dy1 dz1 dx1
4659
+ ���
4660
+
4661
+ ˆ
4662
+ I1
4663
+ ¨
4664
+ R2\(I1×I1)
4665
+ 1
4666
+ (|y2| + |z2|)β2+γ2
4667
+
4668
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
4669
+ 43
4670
+ ×
4671
+ 1 + log+
4672
+ |y3|+|z3|
4673
+ (|x1−y1|+|x1−z1|)(|y2|+|z2|)
4674
+ [(|x1 − y1| + |x1 − z1|)(|y2| + |z2|) + |y3| + |z3|]β3+γ3+4 dy1 dz1 dx1.
4675
+ Note that we have either y1 ∈ R\I1 or z1 ∈ R\I1, and we may without loss of generality
4676
+ assume y1 ∈ R \ I1. Then the integral is dominated by
4677
+ ˆ
4678
+ I1
4679
+ ¨
4680
+ (R\I1)×R
4681
+ 1
4682
+ (|y2| + |z2|)β2+γ2
4683
+ ×
4684
+ 1 + log+
4685
+ |y3|+|z3|
4686
+ |x1−y1|(|y2|+|z2|)
4687
+ [(|x1 − y1| + |x1 − z1|)(|y2| + |z2|) + |y3| + |z3|]β3+γ3+4 dy1 dz1 dx1
4688
+
4689
+ 1
4690
+ (|y2| + |z2|)β2+γ2+β3+γ3+4
4691
+ ˆ
4692
+ I1
4693
+ ˆ
4694
+ R\I1
4695
+ 1 + log+
4696
+ |y3|+|z3|
4697
+ |x1−y1|(|y2|+|z2|)
4698
+
4699
+ |x1 − y1| + |y3|+|z3|
4700
+ |y2|+|z2|
4701
+ �β3+γ3+3 dy1 dx1.
4702
+ Let t := |y3|+|z3|
4703
+ |y2|+|z2|. By a change of variables we reduce to
4704
+ t−β3−γ3−1
4705
+ (|y2| + |z2|)β2+γ2+β3+γ3+4
4706
+ ¨
4707
+ t−1I1×(R\t−1I1)
4708
+ 1 + log+
4709
+ 1
4710
+ |x1−y1|
4711
+
4712
+ |x1 − y1| + 1
4713
+ �β3+γ3+3 dy1 dx1
4714
+
4715
+ t−β3−γ3−1
4716
+ (|y2| + |z2|)β2+γ2+β3+γ3+4
4717
+ ˆ
4718
+ t−1I1
4719
+ 1
4720
+
4721
+ d(x1, ∂(t−1I1)) + 1
4722
+ �β3+γ3+2 dx1
4723
+
4724
+ t−β3−γ3−1
4725
+ (|y2| + |z2|)β2+γ2+β3+γ3+4
4726
+ =
4727
+ 1
4728
+ (|y2| + |z2|)β2+γ2+3
4729
+ 1
4730
+ (|y3| + |z3|)β3+γ3+1
4731
+
4732
+ 1
4733
+ (|y2| + |z2|)β2+γ2 ·
4734
+ 1
4735
+ (|y3| + |z3|)β3+γ3 |I1|(|I1|(|y2| + |z2|)
4736
+ |y3| + |z3|
4737
+ +
4738
+ |y3| + |z3|
4739
+ |I1|(|y2| + |z2|))−1
4740
+ ×
4741
+ 1
4742
+ �3
4743
+ i=2(|yi| + |zi|)2 .
4744
+ Assume then that |I1|(|y2| + |z2|) < |y3| + |z3|. This time we integrate over I1 × I1 × I1.
4745
+ Proceeding as above we reduce to the integral
4746
+ ˚
4747
+ t−1I1×t−1I1×t−1I1
4748
+ t−β3−γ3−1
4749
+ (|y2| + |z2|)β2+γ2+β3+γ3+4
4750
+ 1 + log+
4751
+ 1
4752
+ (|x1−y1|+|x1−z1|)
4753
+ [(|x1 − y1| + |x1 − z1|) + 1]β3+γ3+4 dy1 dz1 dx1
4754
+
4755
+ ¨
4756
+ t−1I1×t−1I1
4757
+ t−β3−γ3−1
4758
+ (|y2| + |z2|)β2+γ2+β3+γ3+4
4759
+ 1 + log+
4760
+ 1
4761
+ |x1−y1|
4762
+ (|x1 − y1| + 1)β3+γ3+3 dy1 dx1
4763
+
4764
+ t−β3−γ3−1
4765
+ (|y2| + |z2|)β2+γ2+β3+γ3+4
4766
+ ¨
4767
+ t−1I1×t−1I1
4768
+
4769
+ 1 + log+
4770
+ 1
4771
+ |x1 − y1|
4772
+
4773
+ dy1 dx1
4774
+
4775
+ t−β3−γ3−1
4776
+ (|y2| + |z2|)β2+γ2+β3+γ3+4 (t−1|I1|)2(1 + log+(t|I1|−1))
4777
+
4778
+ 44
4779
+ EMIL AIRTA, KANGWEI LI, AND HENRI MARTIKAINEN
4780
+ =
4781
+ 1
4782
+ (|y2| + |z2|)β2+γ2 ·
4783
+ 1
4784
+ (|y3| + |z3|)β3+γ3 |I1|(|I1|(|y2| + |z2|)
4785
+ |y3| + |z3|
4786
+ +
4787
+ |y3| + |z3|
4788
+ |I1|(|y2| + |z2|))−1
4789
+ ×
4790
+ 1
4791
+ �3
4792
+ i=2(|yi| + |zi|)2 ·
4793
+
4794
+ 1 + log+
4795
+ |y3| + |z3|
4796
+ |I1|(|y2| + |z2|)
4797
+
4798
+ .
4799
+ Thus, we are done.
4800
+
4801
+ With (A.2) at hand, similarly as in the linear case we can derive the following.
4802
+ A.5. Lemma. Let m ∈ M1
4803
+ Z and denote by Tm the corresponding Fourier multiplier operator.
4804
+ Let f1, g1 ∈ L4(R), f2,3, g2,3 ∈ L4(R2) and h1 ∈ L2(R), h2,3 ∈ L2(R2). Then
4805
+ ⟨Tm(f1 ⊗ f2,3, g1 ⊗ g2,3), h1 ⊗ h2,3⟩ = ⟨Tmf2,3,g2,3,h2,3(f1, g1), h1⟩,
4806
+ where mf2,3,g2,3,h2,3 is a standard bilinear Coifman-Meyer multiplier in R satisfying the estimates
4807
+ |( d/ dξ1)α( d/ dη1)βmf2,3,g2,3,h2,3(ξ1, η1)|
4808
+ ≲ ∥m∥M1
4809
+ Z∥f2,3∥L4∥g2,3∥L4∥h2,3∥L2(|ξ1| + |η1|)−α−β.
4810
+ Thus, Tmf2,3,g2,3,h2,3 is a convolution form bilinear Calderón-Zygmund operator. In particular,
4811
+ there exists a standard bilinear Calderón-Zygmund kernel Km,f2,3,g2,3,h2,3 such that
4812
+ ∥Km,f2,3,g2,3,h2,3∥CZ1(R2) ≲ ∥f2,3∥L4∥g2,3∥L4∥h2,3∥L2.
4813
+ Moreover, if spt f1 ∩ spt g1 ∩ spt h1 = ∅, then
4814
+ ⟨Tm(f1 ⊗ f2,3, g1 ⊗ g2,3), h1 ⊗ h2,3⟩
4815
+ =
4816
+ ˚
4817
+ Km,f2,3,g2,3,h2,3(x1, y1, z1)f1(y1)g1(z1)h1(x1) dy1 dz1 dx1.
4818
+ REFERENCES
4819
+ [1] E. Airta, H. Martikainen, and E. Vuorinen, Product space singular integrals with mild kernel regularity, J.
4820
+ Geom. Anal. 32 (2022), article number 24. ↑19
4821
+ [2] R. R. Coifman, R. Rochberg, and G. Weiss, Factorization theorems for Hardy spaces in several variables, Ann.
4822
+ of Math. (2) 103 (1976), no. 3, 611–635. MR412721 ↑2
4823
+ [3] D. Cruz-Uribe, J. M. Martell, and C. Pérez, Extrapolation from A∞ weights and applications, J. Funct. Anal.
4824
+ 213 (2004), no. 2, 412–439. MR2078632 ↑2, 34
4825
+ [4] J. Duoandikoetxea, Extrapolation of weights revisited: new proofs and sharp bounds, J. Funct. Anal. 260
4826
+ (2011), no. 6, 1886–1901. ↑27, 29
4827
+ [5] X. T. Duong, J. Li, Y. Ou, J. Pipher, and B. Wick, Weighted estimates of singular integrals and commutators
4828
+ in the Zygmund dilation setting, preprint, arXiv:1905.00999 (2019). ↑1, 2, 29
4829
+ [6] R. Fefferman and J. Pipher, Multiparameter operators and sharp weighted inequalities, Amer. J. Math. 119
4830
+ (1997), no. 2, 337–369. MR1439553 ↑1, 2, 3, 39
4831
+ [7] L. Grafakos and J. M. Martell, Extrapolation of weighted norm inequalities for multivariable operators and
4832
+ applications, J. Geom. Anal. 14 (2004), no. 1, 19–46. MR2030573 ↑27, 29
4833
+ [8] L. Grafakos and S. Oh, The Kato-Ponce inequality, Comm. Partial Differential Equations 39 (2014), no. 6,
4834
+ 1128–1157. MR3200091 ↑3
4835
+ [9] L. Grafakos and R. H. Torres, Multilinear Calderón-Zygmund theory, Adv. Math. 165 (2002), no. 1, 124–
4836
+ 164. MR1880324 ↑2
4837
+ [10] A. Grau de la Herrán and T. Hytönen, Dyadic representation and boundedness of nonhomogeneous Calderón-
4838
+ Zygmund operators with mild kernel regularity, Michigan Math. J. 67 (2018), no. 4, 757–786. MR3877436
4839
+ ↑4
4840
+ [11] Y. Han, J. Li, C.-C. Lin, and C. Tan, Singular Integrals Associated with Zygmund Dilations, J. Geom. Anal.
4841
+ 29 (2019), 2410–2455. ↑2
4842
+
4843
+ ZYGMUND DILATIONS: BILINEAR ANALYSIS AND COMMUTATOR ESTIMATES
4844
+ 45
4845
+ [12] I. Holmes, S. Petermichl, and B. D. Wick, Weighted little bmo and two-weight inequalities for Journé commu-
4846
+ tators, Anal. PDE 11 (2018), no. 7, 1693–1740. MR3810470 ↑1
4847
+ [13] T. Hytönen, The sharp weighted bound for general Calderón-Zygmund operators, Ann. of Math. (2) 175 (2012),
4848
+ no. 3, 1473–1506. MR2912709 ↑1
4849
+ [14] T. Hytönen, K. Li, H. Martikainen, and E. Vuorinen, Multiresolution analysis and Zygmund dilations,
4850
+ preprint, arXiv:2203.15777 (2022). ↑1, 2, 3, 5, 14, 15, 16, 29, 31, 35, 38, 40
4851
+ [15] T. Kato and G. Ponce, Commutator estimates and the Euler and Navier-Stokes equations, Comm. Pure Appl.
4852
+ Math. 41 (1988), no. 7, 891–907. MR951744 ↑3
4853
+ [16] A. Lerner, S. Ombrosi, C. Pérez, R. Torres, and R. Trujillo-González, New maximal functions and multiple
4854
+ weights for the multilinear Calderón–Zygmund theory, Adv. Math. 220 (2009), no. 4, 1222–1264. ↑3
4855
+ [17] K. Li, H. Martikainen, Y. Ou, and E. Vuorinen, Bilinear representation theorem, Trans. Amer. Math. Soc.
4856
+ 371 (2019), no. 6, 4193–4214. MR3917220 ↑26
4857
+ [18] K. Li, H. Martikainen, and E. Vuorinen, Bilinear Calderón-Zygmund theory on product spaces, J. Math.
4858
+ Pures Appl. (9) 138 (2020), 356–412. MR4098772 ↑3
4859
+ [19]
4860
+ , Genuinely multilinear weighted estimates for singular integrals in product spaces, Adv. Math. 393
4861
+ (2021), 108099. ↑3, 28
4862
+ [20] H. Martikainen, Representation of bi-parameter singular integrals by dyadic operators, Adv. Math. 229 (2012),
4863
+ no. 3, 1734–1761. MR2871155 ↑1
4864
+ [21] D. Müller, F. Ricci, and E. M. Stein, Marcinkiewicz multipliers and multi-parameter structure on Heisenberg
4865
+ (-type) groups. I, Invent. Math. 119 (1995), no. 2, 199–233. MR1312498 ↑1
4866
+ [22] A. Nagel and S. Wainger, L2 boundedness of Hilbert transforms along surfaces and convolution operators
4867
+ homogeneous with respect to a multiple parameter group, Amer. J. Math. 99 (1977), 761–785. ↑1
4868
+ [23] F. Nazarov, S. Treil, and A. Volberg, The T b-theorem on non-homogeneous spaces, Acta Math. 190 (2003),
4869
+ no. 2, 151–239. MR1998349 ↑1
4870
+ [24] F. Ricci and E. M. Stein, Multiparameter singular integrals and maximal functions, Ann. Inst. Fourier
4871
+ (Grenoble) 42 (1992), no. 3, 637–670. MR1182643 ↑2
4872
+ (E.A.) DEPARTMENT OF MATHEMATICS AND STATISTICS, UNIVERSITY OF JYVÄSKYLÄ, P.O. BOX 35
4873
+ (MAD), FI-40014 UNIVERSITY OF JYVÄKYLÄ, FINLAND
4874
+ Email address: [email protected]
4875
+ (K.L.) CENTER FOR APPLIED MATHEMATICS, TIANJIN UNIVERSITY, WEIJIN ROAD 92, 300072 TIANJIN,
4876
+ CHINA
4877
+ Email address: [email protected]
4878
+ (H.M.) DEPARTMENT OF MATHEMATICS AND STATISTICS, WASHINGTON UNIVERSITY IN ST. LOUIS, 1
4879
+ BROOKINGS DRIVE, ST. LOUIS, MO 63130, USA
4880
+ Email address: [email protected]
4881
+