text
stringlengths
0
89.3k
induction ys with
nil
simp zip Listtake Listdrop
cons y ys ih match xs with
simp zip Listtake
Listdrop Listlength
cons x xs
5unfold zip
unfold Listtake at ih
unfold Listdrop at ih
simp Listlength
rw ih
The following is a correct proof of prop_84 generated by GPT4 There are several unnecessary
definitions passed to the calls to the simplifier the simp tactic However in both the correct and
incorrect proofs GPT4 demonstrates the main ideas of the proof induct on ys pair off the first
elements of xsandys then apply the inductive hypothesis to what remains
induction ys generalizing xs with
nil simp zip Listtake Listdrop
cons y ys ih
cases xs with
nil simp zip Listtake Listdrop
cons x xs
simp Listcons_append zip Listtake
Listdrop Listlength
rw ih
342 Nonstandard Insertion Increments Length
Although the zip concatenation property is not part of Mathlib one can argue that because the
property involves a utility function defined in Mathlib with its own associated properties and proofs
it is not altogether surprising that GPT4 was able to infer a proof of a novel property of a redefinition
of the function We therefore present a property of a function ins not included in Mathlib that inserts
an element into a list before the first element it is less than Proving this property requires the ability
to reason about previously unseen code
def ins Nat List Nat List Nat
n n
n xxs if n x then n x xs else x ins n xs
The following is an incorrect proof of prop_15 by GPT4 Interestingly GPT4s main failing in this
case is syntactic the proof can be manually corrected by using the correct by_cases tactic in place
of byCases unrecognized by Lean 4 and manual removal of the following case matching the in
statements at which point the proof is completed at the rw ih step
induction xs with
nil
simp ins Listlength Natsucc
cons y ys ih
simp ins
byCases h x y
inl h_lt
simp h_lt Listlength Natsucc
inr h_geq
simp h_geq
rw ih
simp Natsucc_assoc
As one might expect from the nearly correct proof above GPT4 also produces a correct proof This
proof is remarkable for its simplicity Unlike the previous correct proof there is little wasted space a
feature often sought after by human proof experts GPT4 demonstrates understanding of how to work
with the ins function via the split_ifs tactic and the usage of the inductive hypothesis only in the
second branch where it is necessary
induction xs with
nil simp ins
cons y ys ih
simp ins
split_ifs with h
simp
simp ih
6343 Sorting Properties
The Lean source code of miniCodeProps contains implementations of 9 sorting algorithms with
associated properties to prove for each algorithm We display several of the functions and lemmas
associated with one of our two heapsort implementations each uses a different strategy for merging
heaps The toList function contains an example of a termination lemma the line beginning with
have _h references a lemma that is used by Lean to prove that the recursion in toList terminates a
property required by default for recursive functions in Lean Several of the functions and lemmas
referenced are omitted from the code for brevity
def toHeap List Nat MyHeap
xs hmerging xsmap fun x MyHeapnode MyHeapnil x MyHeapnil
lemma numElem_merge_branches_lt p q MyHeap x Nat numElem hmerge p q
numElem MyHeapnode p x q by
rw merge_elems _ _
have h numElem MyHeapnode p x q 1 numElem p numElem q rfl
rw h
linarith
def toList MyHeap List Nat
MyHeapnil
MyHeapnode p x q
have _h numElem_merge_branches_lt p q x
x toList hmerge p q
termination_by hp numElem hp
def hsort List Nat List Nat
xs toList toHeap xs
theorem prop_HSortSorts xs List Nat ordered hsort xs True by sorry
theorem prop_HSortCount x Nat xs List Nat count x hsort xs count x xs
by sorry
theorem prop_HSortPermutes xs List Nat isPermutation hsort xs xs True
by sorry
theorem prop_HSortIsSort xs List Nat hsort xs isort xs by sorry
The properties above express the following intuitive properties heapsort orders list elements pre
serves the list size returns a permutation of the original list and returns the same result as insertion
sort These properties are repeated for each sorting algorithm However given the complexity
of the sorting functions involved we expect that proving such properties will be very difficult
miniCodeProps also includes termination lemmas such as numElem_merge_branches_lt with the
humanproduced proof removed An example of an incorrect proof of one of the sorting properties
produced by GPT4 can be found in Appendix A