text
stringlengths
0
89.3k
The eigenvalues of L which are also eigenvalues of A2 and
the associated eigenvectors are given by λ1095λ2093
v10201009796v20207409783
We have βtn
i1γiviλt
i with β0n
i1γivi0 where γ1
andγ2are given by
γ1
γ2v1v21β0287099
302574
andγ1v160818
296399 γ 2v259537
280858
withλ1λ20 we conclude from Theorem 5 that
βt2
i1γiviλt
i0tNAs a conclusion any solution starting in the set C1switches
not more than once We can conclude that for this example
any solution to the considered system switches not more than
twice as some can be initialized in C2 The fact that the origin
of the considered system is GES follows by Theorem 1 as A1
andA2are Schur
Remark 3 For this example it turns out that a piecewise
quadratic Lyapunov function given by 13 exists Indeed
the LMIs conditions 36 given in Appendix B are feasible
withP125662 2 6008
26008 2 6427P201744 0 1570
01570 0 1482
Y100091 Y292206106Y300017 0 0232
00232 0 0038
Y400043 0 0058
00058 0 0297
B Insulin infusion problem
We consider system 14 with n4m2and matrices A
BandKgiven by 7 and 8 Without the results presented
in this paper stability analysis of this 4thorder example is an
open question as far as we know Let A1ABK andA2A
as in Section IIC The eigenvalues of A2areλ109592
λ209512 λ309277 andλ409200 We next show that
the conditions of Theorem 4 hold with in2 To this end we
use Theorem 5 along with conditions given in 33
We start by considering 24 with i01n4andt11
We have to determine the values of t2andt3that satisfy
the conditions of Theorem 4 namely for every t2 t3N
we either have Σi11t2t30or there exists an invertible
matrix NR44 whose lines are taken from 24 such that
items i and ii of Theorem 4 are satisfied By employing
linear programming techniques we have identified 58 pairs of
values for t2andt3 as displayed in Table I
t2 1 3 4 5 6 7 8 9 10 11 12 13
t3 1 1 1 1 1 1 1 1 1 1 1 1
t214 15 16 17 18 19 20 21 22 23 24 1
t3 1 1 1 1 1 1 1 1 1 1 1 2
t2 3 4 5 6 7 8 9 10 11 12 13 14
t3 2 2 2 2 2 2 2 2 2 2 2 2
t215 16 17 1 2 3 4 5 6 7 8 9
t3 2 2 2 3 3 3 3 3 3 3 3 3
t210 1 2 3 4 1 2 3 1 2
t3 3 4 4 4 4 5 5 5 6 6
TABLE I Values of t2andt3in section IVB
For all these 58 pairs we checked that Theorem 4 holds
that is we have found for each of them an invertible matrix
NR44whose lines are taken from 24 such that items i
and ii of Theorem 4 are satisfied For item ii we use
Theorem 5 and conditions 33 Detailed computations for two
of these pairs are provided in Appendix D The conclusion is
that any solution to the considered system exhibits no more
than 4 switches As the eigenvalues of A1andA2lie inside
the unit circle the origin is GES by Theorem 112 GENERIC COLORIZED JOURNAL VOL XX NO XX XXXX 2021
VII C ONCLUSION
This paper focuses on the stability analysis of CLS whose
solutions all exhibit a finite number of switches This property
is very useful when investigating stability as we showed
and somehow move the problem to analytically establish that
all solutions indeed switch a finite number of times We
have first presented general sufficient conditions in terms of
sets intersections To illustrate how these conditions can be
exploited we have then concentrated on the case where two
cones partitions the state space for which we have developed
numerically tractable sufficient conditions Interestingly these
contributions required the development of novel results on the
nonnegativity of solutions to discretetime dynamical systems
for given initial conditions The theoretical developments have
been applied successfully to the optimizationbased control of
a model of insulin infusion for which all the existing stability
results we are aware of failed
It would be interesting in future work to develop similar
tools to prove that solutions exhibit a finite number of switches
and exploit this property for stability analysis for other classes
of hybrid dynamical systems
APPENDIX
A Sampleddata modeling for optimal insulin infusion
Letybe the BGL response uthe insulin input and fFδ
an impulse of food at time t0 In 7 impulses are used
as a mathematical abstraction of insulin applied and meals
consumed and the optimal insulin infusion model is given by
linear transfer functions as for any sCysTFsfs
TIsuswhere