22  Download (0)

Full text


Memoirs on Dierential Equations and Mathematical Physics

Volume 11, 1997, 67{88

Guram Kharatishvili and Tamaz Tadumadze





initial functions are proved in the form of an integral maximum principle and conditions of transversality for nonlinear systems with a variable structure, delays and a piecewise continuous initial function in the case where values of the initial function (prehistory) and of the trajectory at a non-xed initial moment and at a moment of variation of the structure do not, generally speaking, coincide.

1991 Mathematics Subject Classication. 49K25.

Key words and Phrases. Maximum principle, transversality conditions, delays, variable structure systems.

reziume. araCrPivi dagvianebul argumentiani cvladstruqturi- ani sistemisaTvis uban-uban uCKveti saCKisi PunqciiT damtkicebulia marTvis da saCKisi Punqciis optimalurobis aucilebeli pirobebi in- tegraluri maqsimumis principisa da transversalobis pirobebis saxiT im SemTxvevaSi, roca saCKisi Punqciis (Cinaistoriis) da traeqtoris mniS- vnelobebi araPiqsirebul saCKis momentSi da struqturis cvlilebis momentSi, sazogadod, ar emTxvevian erTmaneTs.


1. Introduction

Necessary conditions of optimality for the problem given below (see 2) are derived from a necessary condition of criticality [1,2,3] the basis of which forms the notion of quasi-convex set introduced by R.V. Gamkrelidze [4].

Variation of the structure of a system means that the system at some beforehand unknown moment may go over from one law of movement to another. After variation of the structure, the values of the initial function of the system depend on its previous state. This joins them into a single system with variable structure.

In conclusion, it should be noted that some particular cases of the prob- lem under consideration have been studied in [5].

2. Statement of the Problem. Necessary Conditions of Optimality

Let Oi Rni, Gi Rri, i = 1;:::;m, be open sets, Rn be an n- dimensional Euclidean space, J = [a;b] be a nite interval; let the func- tionsfi :J O2i G2i !Rni,i= 1;:::;m, be continuously dierentiable with respect to (xi;zi) 2 O2i, i = 1;:::;m, respectively, let i : R1 ! R1, i : R1 ! R1, i = 1;:::;m, be continuously dierentiable functions satisfying the conditions i(t) t, _i(t) > 0, i(t) < t, _i(t) > 0; let qk(t1;:::;tm+1;x10;x1;:::;xm0;xm),k= 0;:::;l, be scalar functions con- tinuously dierentiable in all argumentsti 2J,i= 1;:::;m+ 1, (xi0;xi)2 Oi2, i = 1;:::;m; let the functions gi : J Oi 1 ! Oi, i = 2;:::;m be continuous and continuously dierentiable with respect to xi 12Oi 1, i= 1;:::;m; respectively, let i= (Ji1;Ni) be the set of piecewise contin- uous functions'i:Ji1!Niwith a nite number of points of discontinuity, Ji1[i(a);b], Ni Oi be a convex bounded set,k'ik= supf'i(t)jt 2Ji1g; let i be the set of measurable functions ui :Ji2!Ui satisfying the con- dition: clui(Ji2)-is a compactum lying inGi,Ji2= [i(a);b], UiGi is an arbitrary set.

Consider the sets Ai=J1+iYi



p=1p; i= 1;:::;m;

with the elements i = (t1;:::;ti+1;x10;:::;xi0;'1;:::;'i;u1;:::;ui),i= 1;:::;m;Qip=1Op=O1Oi, respectively.

To every element m 2 Am, ti < ti+1, i = 1;:::;m, we assign the dierential equation of variable structure

x_i(t) =fi(t;xi(t);xi(i(t));ui(t);ui(i(t))); t2[ti;ti+1]; (1i) xi(t) ='i(t) +gi(t;xi 1(t));t2[i(ti);ti);xi(ti) =xi0; (2i)

i= 1;:::;m:


Here and in the sequel we assume that g1 = 0, that is, x1(t) = '1(t);t 2 [1(t1);t1).

Denition 1. The set of functionsfxi(t) =xi(t;i);t 2[i(ti);ti+1];i= 1;:::;mg, is said to be a solution of the equation with variable structure which correspond to an elementm2Am, if the functionxi(t)2Oi on the interval [i(ti);ti] satises the condition (2i), while on the interval [ti;ti+1] it is absolutely continuous and satises the equation (1i) almost everywhere.

Denition 2. The elementm2Amis said to be admissible if the corre- sponding solution fxi(t);t 2[i(ti);ti+1];i= 1;:::;mgsatises the condi- tions

qk(t1;:::;tm+1;x10;x1(t2);:::;sm0;xm(tm+1)) = 0; k= 1;:::;l:

The set of admissible elements will be denoted byA0m.

Denition 3. The element em = (et1;:::;etm+1;xe10;:::;xem0;'e1;:::;'em; ue1:::;uem)2A0m is said to be locally optimal if there exist a number >0 and a compact setKi 0i;i= 1;:::;m, such that for arbitrary elements m2A0m satisfying

mX+1 i=1


eti tij+Xm


fjexi0 xi0j+ke' 'ik+kfei fikKig the inequality


q0(t1;:::;tm+1;x10;x1(t2);:::;xm0;xm(tm+1)) is fullled.


fei fiKi =


J H t;fi;Kidt;H t;fi;Ki=

= supfei t;xi;yi fi t;xi;yi+@fei()





@yi @fi()


xi;yi2Ki2 ; fei t;xi;yi=fi t;xi;yi;eui(t);eui(i(t)));fi(t;xi;yi) =

=fi(t;xi;yi;ui(t);ui(i(t)); xei=xi(t;ei); xi(t) =xi(t;i): The problem of optimal control consists in nding a locally optimal ele- ment.


Theorem 1. (Necessary conditions of optimality). Let em 2 A0m be a locally optimal element; a < et1 < < etm+1 < b, eti < i = i(eti) <

i+1(eti+1) i = 1;:::;m 1, etm < m = m(etm) 6=etm+1; let the functions (eui(t);uei(i(t)))i= 1;:::;m, be continuous at the pointseti;i;eti+1, respec- tively,i= 1;:::;m, and the functions('ei(t);'ei(i(t))),i= 1;:::;mbe con- tinuous at the pointseti, i= 1;:::;m, respectively. Then there exist a non- zero vectorn= (n0;:::;nl), n0 0, and a solution i(t), t2[eti;i(eti+1)]

of the conjugate equation

_i(t) = i(t)@fei(t)

@xi i((t))@fei(i(t))

@yi _i(t) i(t) i+1(i+1(t))@fei+1(i+1(t))

@yi+1 @egi+1(t)

@xi _i+1(t) (3i) t2[eti;eti+1]; i(t) = 0; t2(eti+1;i(eti+1)];

i=m;:::;1; such that the following conditions are fullled:

1) the integral maximum principle





@yi _i(t)'ei(t)dt





@yi _i(t)'i(t)dt; 8'i2i; i= 1;:::;m; (4)







8'i2i; i= 1;:::;m: (5) 2) transversality conditions

@Q@ti = ai i 1(eti)fei 1(tei) +bif i(eti)fei(eti) + i(i) [fei(i;xei(i);exi0) fei(i;exi(i);'ei(eti) +egi(eti))]_i(eti)g;

i= 1;:::;m+ 1; (6)

@Q@xi0 = i(eti); @Q@xi = i(eti+1); i= 1;:::;m: (7)



fei(t) =fei(t;xei(t);exi(i(t))); @fei(t)

@xi = @


egi(t) =gi(t;exi 1(t)); @egi+1(t)

@xi = @


i(t) is the characteristic function of the segment [i(eti+1);eti+1)]; gm+1= 0, i.e., the last summand in the right-hand side of the equation (3m) equals zero; i(t) is the function inverse to i(t); a1 = 0, a2 == am+1 = 1, b1 = = bm = 1, bm+1 = 0; tilde over Q denotes that the gradient is calculated at the point (et1;:::;etm+1;xe10;xe1(et2);:::, exm0;exm(etm+1),

Remark 1. If rank


@t1;::: @Q@tm+1; @Q@x10;::: @Q@xm0; @Q@x1;:::; @Q@xm

= 1 +l;

then m


i=1maxfj i(t)jjt2[eti;eti+1]jg6= 0:

Remark 2. From the integral maximum principle 1) one can obtain in a standard way the pointwise maximum principle with respect to the functions 'ei(t),i= 1;:::;m:


@yi _i(t)'ei(t) i(i(t))@fei(i(t))

@yi _i(t)'i;

8'i2Ni; a.e. on [i(eti);eti]; i= 1;:::;m; with respect to the controlsuei(t),i= 1;:::;m,



p=1_pi 1(t) i(pi 1(t))fei(ip 1(t)) +Xi pi

p=1 :q_pi(t) i(qpi(t))fei(qp(t))



p=1_pi 1(t) i(pi 1(t))fi(p 1(t);exi(ip 1(t));xei(i(pi 1(t))); ui;pi p+1;ui;pi p) +Xi pi

p=1 _pi(t) i(pi(t))fi(pi(t);exi(pi(t));xei(i(pi(t))); ui;pi+p;ui;pi+p 1); 8(u10;:::;u1;i)2u1+i i; a.e. on

[i;pi;i;pi+1]; pi= 1;:::;i; i= 1;:::;m:


i;pi =i(i;pi 1); pi= 1;:::;i; i;0=i(eti); i;pi+1=eti+1;


0i(t) =t, pi(t) =i(pi 1(t));i(t) is the function inverse toi(t),1i(t) = i(t); we assume thatP0p=1p= 0.

Theorem 2. Let em 2A0m be a locally optimal element; a <et1 <<

etm+1 < b, eti < i < i+1eti+1, i= 1;:::;m 1, etm < m6= etm+1; let the functions (uei(t);uei(i(t))), i = 1;:::;m, be left- (right-) continuous at the points eti;i;eti+1, i = 1;:::;m, respectively. Then there exist a non-zero vector(+) = (0(+);:::;q(+)) and a solution i(+)(t), t2[eti;i(eti+1)]

of the conjugate equation (3i), i = 1;:::;m, such that the conditions in which we have to substitute and i(t) instead of (+) and (+)i , respec- tively, i = 1;:::;m, are fullled. Moreover, the equality (6) is replaced by the inequality


@ti ai i 1(eti)fei 1(eti ) +bif i (eti)fei(eti ) + i (i)[fei(i ;exi(i);

ex10) fei(i ;xei(i);'i(eti ) +gei(eti))]_i(eti)g; i= 1;:::;m+ 1


@ti ai +i 1(t)fei 1(eti+) +bif i+(eti)fei(eti+) i+(i)[fei(i+;xei(i); xe10) fei(i ;exi(i);'ei(eti+) +gi(eti))]_i(eti)g;


fei(et(+)i ) =fe(et(+)i ;xei(eti);xei(ei(et(+)i )):

Consider now for (1i) and (2i), i= 1;:::;m, where g2 ==gm = 0, the problem with the boundary conditions

q1(t1;x10) = 0;qi(ti;xi 1(ti)) xi0= 0; i=r;:::;m;

qm+1(tm+1;xm(tm+1)) = 0; (8) and with the functional

q0(tm+1;xm(tm+1))!min: (9) The functions

q1:J01!Rl1;qi:J0i!Rni; i= 2;:::;m;

qm+1:J0m!Rl2; q0:J0m!R1 are assumed to be continuously dierentiable in all arguments.

The functionQfor the problem under consideration is of the form Q= (q0;q1;q2 x20;:::;qm xm0;qm+1): (10) Taking into account (10), from Theorem 1 there follows


Theorem 3. Letmbe a locally optimal element of the problem(1i), (2i), i= 1;:::;m, (8), (9) and let the conditions of Theorem 1 be fullled. Then there exist a non-zero vector = (0;1;:::;m+1), 0 0, 1 2 Rl1, i 2Rni, i= 1;:::;m, m+1 2Rl2 and a solution i(t), t2[eti;i(eti+1)], of the conjugate equation(3i)i= 1;:::;m, where the last term in the right- hand side equals zero, such that the conditions (4), (5) are fullled, while the conditions of transversality take the form


@ti = ai i 1(eti)fei 1(eti) +bif i(eti)fei(eti) +

+ i(i)[fei(i;exi(i);xei0) fei(i;xei(i);'ei(eti))]i(eti)g ci0 @qe0i

@tm+1; i= 1;:::;m+ 1;1 @qe1

@x10 = 1(et1); i(eti) =i; i= 2;:::;m;

i @qi

@xi 1 = i 1(eti); i= 2;:::;m;


@xm +m+1@qem+1

@xm = m(etm+1): . Here

c1==cm= 0; cm+1=1; xei0=qi(eti;exi 1(eti); i= 2;:::;m:

3. Proof of Theorem 1.

The necessary conditions of optimality are proved by the scheme given in [1, 2, 3]. When applying this scheme, the principal moment is the construc- tion of a continuous and dierentiable mapping which plays an important role in deriving the necessary conditions of optimality. To this end we present below and prove (see 3.1 and 3.2) the appropriate theorems.

3.1. Continuous Dependence and Dierentiability of the Solution. LetO Rn be an open set: E(JO2;Rn) be a space of n-dimensional functions f;J02!Rn satisfying the following conditions:

3) for a xed t 2 J the function f(t;x;y) is continuously dierentiable with respect to (x;y)2O2;

4) for a xed (x;y) 2 O2 the functions f;fx;fy are measurable with respect tot; for an arbitrary compactumKO and an arbitrary function f there exists a functionmf;k(t)2L1(J;R1+),R1+= [0;1) such that

jf(t;x;y)j+jfx(_)j+jfy(_)jmf;K(t) 8(t;x;y)2JK2:

In the space E(J O2;Rn), let us introduce by means of the bases of neighborhoods of zero two locally convex separable topologies [1, 2, 3],


fVs(K;) E(J 02;Rn) a compactum K and a number > 0 are arbitraryg,s= 1;2, where

V1(K;) =ff2E(JO2;Rn)jh(f;K)g; (11) V2(K;) =ff2E(JO2;Rn)j


J H0(t;f;K)dtg; (12) h(f;K) = sup


t0 f(t;x;y)dtj(t0;t00;x;y)2J2K2

; H0(t;f;K) = supnjf(t;x;y)j+jfx()j+jfy()j(x;y)2K2o: Consider the sets




p=1E(JOi2;Rni); i= 1;:::;m;

with the elementsi = (t1;:::;ti+1;x10;:::;xi0;'1;:::;'i;f1;:::;fi),i= 1;:::;m respectively. In what follows we will assume that the topologies T1, T2 are prescribed in the spaces E(J Oi2;Rni) and i = 1;:::;m (see (11) and (12)).

To every elementm2Bmthere corresponds the dierential equation of variable structure:

x_i(t) =fi(t;xi(t);xi((t))); t2[ti;ti+1]; (13i) xi(t) ='i(t) +gi(t;xi 1(t)); t2[i;(ti);ti+1);

xi(ti) =xi0; i= 1;:::;m: (14i) Denition 4. A set of functions fxi(t) = xi(t;i);t 2 [i(ti);ti+1];i = 1;:::;mgis said to be a solution of the equation of variable structure cor- responding to the element m 2 Bm, if the function xi(t) 2 0i on the segment [i(ti);ti] satises the condition (14i) and on the interval [ti;ti+1] is absolutely continuous and satises the equation (13i) a.e.

Theorem 4. Let fexi(t);t 2 [Ji(eti);eti+1];i = 1;:::;mg, a < et1 < <

etm+1< b, be a solution corresponding to the elementem=(et1;:::;etm+1;ex10; :::;exm0;'e1;:::;'em;fe1;:::;fem) 2 Bm and let Ki1 0i be a compactum containing some neighborhood of the set


Then for every" >0 there exists a number =(")>0 such that to every element

m2V(em;K11;:::;Km1;;c0) =mY+1

i=1 V(eti;)Ym







there corresponds a solutionfxi(t;i);t2[Ji(ti);ti+1];i= 1;:::;mg. More- over, the function xi(t;i) is dened on [Ji(ti);eti+1+] (Ji(a);b), and satises almost everywhere on [ti;eti+] the equation (13i).

If sm2V(em;K11;:::;Km1;;c0); s= 1;2; then

jxi(t;1i) xi(t;2i)j"; t2[max(t1i;t2i);eti+1+];

i= 1;:::;m: (15)

Here V(eti;), V(xei0;),V('ei;), are the -neighborhoods of the points

eti,xei0, 'ei, in the spaces R1, Rn, respectively; (Ji1;Rni); V1(fei;Ki;) = fei+V1(Ki1;),V2(fei;Ki1c0) =fei+V2(Ki1;c0),V1(Ki0;)E(J02i;Rni, V2(Ki1;c0)E(J02i;Rni;c0>0 is a xed number.

Theorem 4 can be proved by the method given in [6] (see also [7]) and is used in proving the continuity of the mapping (see 3.3).

Remark 3. There exists a numbere2[0;] (see (12)) such that V2(fei;Ki1;e)V1(fei;Ki;e)\V2(fei;Ki1;c0); i= 1;:::;m:

Consequently, the inequality (15) is the more so valid for sm2mY+1

i=1 V(eti;e)Ym


i=1V2(fei;Ki1;e): This fact is used in proving the openness of the setD0 (see 3.2).

Introduce the set

Vi=fi= (t1;:::;ti+1;x10;:::;xi0;'1;:::;'i;f1;:::;fi)2

2Bi eijtijci; xi0c1;k'ikci;fp=Xs

j=1j"fjp;p= 1;:::;i;

jjjcig; i= 1;:::;m;

fjp 2 E(J 02p;Rnp), p = 1;:::;i, j = 1;:::;s, are xed points and s, ci>0 are xed numbers.

From Theorem 4 we have


Theorem 5. There exist numbers"0>0,0>0 such that for an arbitrary (";m)2[0;"0]Vm, to the elementem+"mthere corresponds solution

fxi(t;ei+"i);t2[i(ti);ti+1]; i= 1;:::;mg; ti=eti+"ti: Moreover,xi(t;ei+"i) is dened on [i(ti);eti+1+0].

Remark 4. Due to the uniqueness,xi(t;ei) on the interval [i(eti);eti+1+ 0] is a continuation of xei(t). Therefore the functionxei(t) in the sequel is assumed to be dened on the whole interval [i(eti);eti+1+0].

Using the numbers0and"0(see Theorem 5), we introduce the notation xi(t;"i) =xi(t;i+"i) xei(t);t2[max(ti;eti);eti+1+0];"2[0;"0]: Theorem 6. Let a <et1 <<etm+1 < b, eti < i =i(eti)< i+1(eti+1), i+ 1;:::;m 1,etm< m=m(etm)6=etm+1; let the functionsfei(t;xi;yi), i = 1;:::;m, be continuous respectively at the points (eti;exi0;'ei(i(eti))), (i;xei(i);exi0), (i;xei(i);'ei(eti)), i = 1;:::;m, and the functions 'ei(t), i= 1;:::;m, be continuous respectively at the pointseti, i= 1;:::;m. Then there exist numbers"12[0;"0],12[0;0] such that the following formula is valid:

xi(t;"i) ="xi(t;i) +oi(t;"i); 8(t;";i)2[i+1(eti+1);

eti+1+1][0;"1]Vi; i= 1;:::;m; (16) where

xi(t;i) =Yi(eti;t)hxi0 fei(eti)ti







h'i(s) +@egi(s)

@xi 1xi 1(s)i_i(s)ds+



Yi(s;t)feids Yi(i;t)

fei(i;xei(i);exi0) fei(i;exi(i);'ei;(eti) +egi(eti))_i(eti)ti =

=x1i(t;i) x2i(t;i); i= 1;:::;m; (17) Yi(s;t) is a matrix function satisfying the equation


@s = Yi(s;t)@fei(s)

@xi Yi(i(s);t)@fei(i(s))

@yi _i(s); s2[eti;eti+1] and also the condition

Yi(t;t) =E; Yi(s;t) = 0; s > t; next,

"lim!0oi(t;"i)="= 0; uniformly with respect to (t;i)2[i+1(eti+1);


eti+1+1]Vi;fei(t) =fi(t;xei(t);exi(i(t))); m+1(etm+1) =etm+1 1: The proof of Theorem 6 is conducted in a way described in [2, 7]. In the same way one can prove more general

Theorem 7. Let a < et1 < < etm+1 < b, eti < i < i+1(eti+1), i = 1;:::;m 1,etm< m6=etm+1, and let the conditions


limfei(t;xi;'ei(i(t))) =fei(eti(+));exi0;'e(i(eti(+)))<1; (t;xi;yi)!(i(+);xei();xei0)

limfei(t;xi;yi) =fei(ei(+);xei(i);exi0)<1;

be fullled. Then there exist numbers"12[0;"0],12[0;0], such that the formula (16) is valid for an arbitrary point (t;";i) 2[i+1(eti+1);eti+1+ 1][0;"1]Vi(+), while in the formula (17) before ti there take place respectively the expressions

Y(eti;t)fei(eti(+)) +Y(i;t)fei(i;(+);xei(i);exi0) fei(i(+);xei(i);'ei(eti(+)) +g(eti))_i(eti): Here

Vi fi2Vij tK0; K= 1;:::;i+ 1g; Vi+fi2Vij tK 0; K= 1;:::;i+ 1g:

Theorem 8 ([The Cauchy formula [8]]). Let A(t), B(t), t 2J1 = [s1;s2] be summable matrix functions of dimensionnn; letF(t), t2J1be an n- dimensional summable vector function and(t) satisfy the same conditions asi(t) do; let'(t), t2[(s1);s2] be a piecewise continuous function. Then the solution of the equation

x_(t) =A(t)x(t) +B(t)x((t)) +F(t); t2J1; x(t) ='(t); t2[(s1);s1);x(s1) =x02Rn; can be represented as

x(t) =Y(s1;t)x0+ s



(s1) Y((s);t)B((s))_(s)'(s)ds+ t


s1 Y(s;t)F(s)ds;


whereY(s;t) is a matrix function satisfying the equation


@s = Y(s;t)A(s) Y((s);t)B((s))_(s); s2[s1;t]


and also the condition

Y(t;t) =E; Y(s;t) = 0; s > t; (s) is the function inverse to(s).

On the basis of Theorem 8 we can conclude that the functionx1i(t;i) (see (17)) satises the equation

x_1i(t) =@fei(t)

@xi x1i(t) +@fei(t)

@yi x1i(i(t)) +fei(t);

t2[eti;eti+1] (18) with the initial condition

x1i(t) ='i(t) +@egi(t)

@xi 1xi 1(t); t2[i(eti);eti);

x1i(eti) =xi0 fei(et1)ti; (19) while the functionx2i(t;i) (see (17)) satises the equation

x_2i(t) =@fei(t)

@xi x2i(t) +@fei(t)

@yi x2i(i(t)); t2[i;eti+1]; (20) with the initial condition

x2i(t) = 0; t < i x2i(i) = [fei(i;exi(i);xei0)

fei(i;xei(i);'ei(eti))]_i(eti)ti: (21) For the sake of brevity we denote the function,xi(t;i),x1i(t;i) and x2i(t;i), respectively by xi(t), x1i(t) andx2i(t).

Theorem 9. The following formula is valid:





@xixi(eti+1) =Xm

i=1(Yi(eti)x1i(eti) Yi(i)x2i(i) +i); (22) where






@yi _i(t)'i(t)dt+ +



Yi(t)fei(t)dt; (23)




Related subjects :