text
stringlengths
0
89.3k
i1ipip1with τ0 This fact is expressed by
the condition tp11 together with τk1 τ krk0 t jk1that is to say that if jκp1 we
enforce rκ1andτκrκ0
Proof of Proposition 1 For every tN we have
Σt1tpt0
i1ipip1iΣt1tp0
i1ipip1St1tpt0
i1ipip1i
κ
k1rk
ℓ1Stj1tjk1τkℓ
ij1ijk1ijk
St1tpt0
i1ipip1i
We are ready to state conditions under which anysolution
to 14 and not only those initialized in Σt1tp0
i1ipip1as in
Proposition 1 exhibits at most pswitches
Theorem 3 Given pN if for every i1 i p1
1 mand every t2 t pN the conditions of
Proposition 1 are satisfied with t11 then all the possible
solutions of 14 admit at most pswitches
Proof Assume by contradiction that there exists x0Rn
such that the solution initialized at x0switches more that
ptimes Then there exists i1 i p ip1 ip2with ik1
ik and t1 t p tp1Nsuch that At11
i1x0belongs
toΣ1t2tp10
i1i2ip1ip2 This leads to a contradiction since by
Proposition 1 this set is empty
From a computational point of view the main difficulties
with the result of Theorem 3 are related to checking the inter
section 23 for every tNand every i1 mip1
The problem is combinatorial by nature However we provide
in the sequel tractable conditions for the case m2 In
Section IVB we propose a solution to tackle the fact that tis a
priori not bounded It consists in using Farkas lemma which is
a particular case of the Sprocedure dedicated to linear forms
to transform the problem into assessing the nonnegativity
of specific solution to an auxiliary discretetime system We
propose tractable conditions to check this nonnegativity in
Section V As to the combinatorial complexity related to
checking the intersection 23 for every i1 mip1
the situation is manageable as we will manipulate sets defined
by linear inequalities and checking if their intersection is
nonempty can be done using linear programming
Remark 2 Connection with Hellys theorem Let us first
recall that Hellys theorem 29 states that
Given snconvex sets S1 S sRn if the
intersection of every n1of these sets is nonempty
theni1sSi
A consequence of this result is that given snconvex sets
S1 S sRnfor whichs
i1Si there exist n1sets
among them whose intersection is empty Hence we can certify
whether the intersections overall the ssetsSiis empty by only
considering the intersection of n1of them
Proposition 1 is in the spirit of Hellys theorem Indeed the
sufficient condition given in Proposition 1 is expressed in term
of sets intersections and we enforce that κ
k1rkn The
result applies even if the sets St1tk1t
i1ik1ikare not convex Here
these sets are convex as the sets Ciare However Proposition 18 GENERIC COLORIZED JOURNAL VOL XX NO XX XXXX 2021
is not an application of Hellys Theorem In fact if we applied
directly Hellys Theorem we would obtain the same result as
in Proposition 1 but the set selection ie the choice of jk
rkandτkrk would depend on tandi In this case we would
have obtained a necessary and sufficient condition but this
condition might be impossible to test in practice
B When m2
To streamline the discussion and maintain clarity we focus
on the case where m2 ie the partition of Rnis made of
two cones while keeping in mind that the results and concepts
presented here can be readily extended to the general case
when mN Note that the case where m2is notoriously
difficult 4
When m2 the cones C1andC2can be written as in 10
for some matrix KR1n which is not the same as in
Section II in general
1 General result Since the indexes ikused in Sec
tion IVA are such that ik1ik the only possibilities
when m2are the sequences 121 or212 This
observation together with the expression of C1andC2given
in 10 leads to set
Si1t1tptSt1tpt
i1i1p
xRn1i1pKAt
i1p1Atp
i1p At1
i1x0
where for every iN we have defined 2i11and
2i2 Similarly we define Σi1t1tpt Given i112
if we start from an initial condition in Ci1 the solution stays
for the remaining times in Ci1or enters in Ci11 Hence the
maximal number of switches starting from Ci1is bounded by
one plus the maximal number of switches starting from Ci11