On the stability of the solution in an optimal control problem
for a Schrödinger equation
Murat Subasßı
a,⇑, Hülya Durur
b aAtaturk University, Science Faculty, Department of Mathematics, 25240 Erzurum, Turkey b
Ardahan University, Vocational School of Social Sciences, Ardahan, Turkey
a r t i c l e
i n f o
Keywords: Optimal control Frechet differentiability
Time dependent Schrödinger equation
a b s t r a c t
This research paper deals with the Frechet differentiability and necessary condition for optimality in a problem governed by the coefficient of a Schrödinger equation. In order to obtain a stabile solution by means of strong convergence of any minimizer, we have used a more regular space and considered second adjoint problem for obtaining the gradi-ent of the cost functional. Proving the Lipschitz continuity, we have stated the necessary condition for optimal solution.
Ó 2014 Elsevier Inc. All rights reserved.
1. Introduction and statement of the problem
In this paper, we consider the minimization of the functional
Jað
v
Þ ¼ Z l 0 jwðx; T;v
Þ yðxÞj2dx þa
kv
wk2W1 2ð0;TÞ ð1Þ on the set V ¼v
ðtÞ :v
2 H1ð0; TÞ; kv
ðtÞkH1ð0;TÞ6l
n o ð2Þsubject to the following Schrödinger initial-boundary value problem given on the domainX¼ ð0; lÞ ð0; TÞ;
i@w @tþ a0
@2w
@x2
v
ðtÞw ¼ 0; ðx; tÞ 2X
ð3Þwðx; 0Þ ¼
u
ðxÞ; x 2 ð0; lÞ; wð0; tÞ ¼ wðl; tÞ ¼ 0; t 2 ð0; TÞ ð4ÞEquation(3)describes the wave function of a particle having mass of unity and acting in a magnetic field for h ¼ 1. If it is known that particle moves only inside of (0, l) and it has the initial status
u
(x) then the conditions given in(4)are valid. The control functionv
(t) is the potential of outer electric field. The function w(t) is in H1(0, T) and plays a vital role in order toobtain a unique solution.
The parameter
a
is a positive number used for both obtaining the unique solution and ensuring a balance between the norms of kwðx; TÞ yðxÞk2 L2ð0;lÞand kv
wk 2 W1 2ð0;TÞin numerical investigations. http://dx.doi.org/10.1016/j.amc.2014.10.069 0096-3003/Ó 2014 Elsevier Inc. All rights reserved.⇑Corresponding author.
E-mail addresses:msubasi@atauni.edu.tr(M. Subasßı),hulyadurur@ardahan.edu.tr(H. Durur).
Contents lists available atScienceDirect
Applied Mathematics and Computation
Similar optimal control problems have been considered in the studies[1–6]with the functional Iað
v
Þ ¼ Z l 0 jwðx; T;v
Þ yðxÞj2dx þa
kv
wk2L2ð0;TÞ ð5Þ on the set ^ V ¼v
ðtÞ :v
2 L2ð0; TÞ; kv
ðtÞkL2ð0;TÞ6g
n o ð6Þsubject to the Schrödinger equations like(3)–(4). With these researches the existence and uniqueness for the optimal control problems have been proved. But the stability of the solution could have not been guaranteed by minimizing the functional
(5)on the set(6). Namely, instability often occurs in the manner that there are some minimizing sequences f
v
mg ^V for thefunctional Ia(
v
) such as lim m!1Iaðv
m Þ ¼ Ia¼ inf v2VIaðv
Þbut these sequences may not converge on the norm of L2(0, T) to the minimum element
v
⁄; limm!1k
v
mðtÞ
v
ðtÞkL2ð0;TÞ9 0
In this study, we propose the minimization of the functional Ja(
v
) on the set V given by(1)–(2)instead of minimization Ia(v
) on the set ^V. Thanks to the compact embedding of the space H1(0, T) to the space L2(0, T), the weakly minimizingsequences in H1(0, T), converge strongly in L
2(0, T)[7]. So the desired convergence of limm!1k
v
mðtÞv
ðtÞkL2ð0;TÞ¼ 0 willbe hold with the construction of any minimizing sequence
lim m!1Jað
v
mÞ ¼ J a¼ inf v2VJaðv
Þ for problem(1)–(2).But, at the same time, some problems may arise when we choose the problem(1)–(2). For example how the gradient in the space H1(0, T) can be computed? How the Lipschitz continuity of this gradient can be proved and how the necessary
con-dition for obtaining optimal solution can be gotten? Finding stabile solution by means of strong convergence of any mini-mizer requires the usage of a second adjoint problem in the stage of obtaining gradient. In the literature the answers to these types of questions are missing in the space H1(0, T). This paper deals with these questions.
This paper is organized as follows: In Section2, we introduce the definition of generalized solution that we need in the remaining part of the paper. In Section3, we affirm the existence and uniqueness of the optimal solution by means of Goebel’s study[9]. The contribution of this research begins in Section4. In Section4, we give the Theorem 1 about the Frechet differentiability of the functional and we give a second adjoint problem. In Section5, estimating the solution of this adjoint problem, the Lipschitz continuity of the gradient and necessary condition for optimal control is submitted. The gra-dient obtained in the space H1(0, T) will lead to further numerical investigations about stabile solutions.
2. The generalized solution for Schrödinger initial-boundary value problem
Let us define spaces used in this article. The space L2(X) is a Hilbert space consisting of square integrable functions with
its modulus onX. The inner product and norm on this space are defined such as
hw; /iL2ðXÞ¼ Z X wðx; tÞ/ðx; tÞdxdt; kwkL2ðXÞ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi hw; wiL2ðXÞ q :
The space H2(0, l) is a Hilbert space consisting of all elements L
2(0, l) having generalized derivatives of first and second
order from L2(0, l). The inner product and norm on this space are defined such as hw; /iH2ð0;lÞ¼ Z l 0 w/þ@w @x @/ @xþ @2w @x2 @2/ @x2 ! dx; kwkH2ð0;lÞ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi hw; wiH2ð0;lÞ q :
The space Ho2ð0; lÞ is a subspace of H2
(0, l), in which all functions in [0, l] that are equal to zero at 0 and at l are dense. The space H2,1(
X) is a Hilbert space consisting of all elements L2(X) having generalized derivatives@w@x,@
2w
@x2 and
@w @t from
L2(X). The inner product and norm on this space are defined such as hw; /iH2;1ðXÞ¼ Z X w/þ@w @x @/ @xþ @2w @x2 @2/ @x2þ @w @t @/ @t ! dxdt; kwkH2;1ðXÞ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi hw; wiH2;1ðXÞ q : The space H2;1 o
ðXÞ is a subspace of H2,1(X), in which all functions inXthat are equal to zero at 0 and at l are dense. Now, we introduce the definition of the generalized solution to the Schrödinger initial-boundary value problem.
If the initial status
u
(x) is in the spaceu
2 H2 oð0; lÞ ð7Þ
then the generalized solution of the problem(3)–(4)is the function w 2 Ho2;1ðXÞ satisfying the integral equality of Z X i@w @tþ a0 @2w @x2
v
ðtÞw " #g
ðx; tÞdxdt ¼ 0 ð8Þ for8g
2 L2ðXÞ.Using Galerkin method[8]it can be proven that the solution in the sense of(8)is exist, unique and satisfies the inequality
kwk2 H2;1 o ðXÞ 6c0k
u
k2 H2 o ð0;lÞ: ð9ÞHere c0is independent from
u
.3. The existence and uniqueness of the optimal control
To prove the existence of optimal solution for problem(1)–(2)it is enough to show that the conditions of the generalized Weierstrass theorem hold.
Let us accept that we have found a minimizer f
v
mg V for the functional Ja(
v
), namely the following equality is valid;lim
m!1Jað
v
mÞ ¼ Ja¼ inf v2VJað
v
Þ:Since the set V is closed, bounded and convex subset of H1(0, T), it is possible to choose a subsequence f
v
mkg fv
mgwhich weakly converges to a control
v
0e
H1(0, T). Again since the set is weak closed, this weak limit is in the set V. Onthe other hand the solution sequence fwmkg of the problem(3)–(4)corresponding to the controls f
v
mkg is exist. Similar to
(9), this sequence holds the inequality
kwmkk
2 Ho2;1ðXÞ
6c0k
u
k2Ho2ð0;lÞ:
Besides this sequence weakly converges to an element w02 H o
2;1ðXÞ. It can be shown that
w
0is generalized solution to the
problem.
Using the above inequality, we can say that
wmð:; TÞ!w0ð:; TÞ; weakly in L2ð0; lÞ
for m ? 1. Since the norm in L2(0, l) is weakly lower semi continuous we can pass the inequality; kw0ðx; TÞ yðxÞkL2ð0;lÞ6 lim
m!1kwmðx; TÞ yðxÞkL2ð0;lÞ
:
Taking into account that the sequence f
v
mg V weakly converges in H1(0, T), we conclude that Ja6Jaðv
0Þ 6 limm!1
Jað
v
mÞ ¼ Ja:Namely the conditions of generalized Weierstrass theorem hold. Therefore the limit
v
0is optimal solution to the problem (1)–(2);Jað
v
0Þ ¼ Ja¼ Jaðv
Þ:As to the uniqueness of the solution we refer to the following theorem given by Goebel in[9].
Theorem: Let H be a uniformly convex Banach space and the set V be a closed, bounded and convex subset of H. If
a
> 0, b P 1 are given numbers and the functional J(v) is lower semi continuous and bounded from below on the set V then there is a dense set G of H that the functionalJað
v
Þ ¼ Jðv
Þ þa
kv
wkbHtakes its minimum on the set V for8w 2 G. If b > 1 then minimum is unique. h The following statements are valid in problem(1)–(2);
(a) the set H1(0, T) is a uniformly convex Banach space[10],
(b) the set V is a closed, bounded and convex subset of H1(0, T),
(c) the functional J(
v
) is lower semi continuous and bounded from below on the set V, (d) b = 2.4. The Frechet differentiability of the functional
In this section we prove the Frechet differentiability of the functional(1)on the set(2). To do this, we first consider the following adjoint boundary value problem
i@/ @t þ a0 @2/ @x2
v
ðtÞ/ ¼ 0; ðx; tÞ 2X
/ðx; TÞ ¼ 2iðwðx; TÞ yÞ; /ð0; tÞ ¼ /ðl; tÞ ¼ 0: ð10ÞAs the solution of this problem, we mean the function / ¼ /ðx; tÞ 2 C0ð½0; T; L2ð0; lÞÞ which satisfies the integral equality Z X / i@
c
@tþ a0 @2c
@x2v
ðtÞc
! dxdt ¼ Zl 0 i/ðx; 0Þc
ðx; 0Þdx 2 Z l 0 ðwðx; TÞ yÞc
ðx; TÞdx ð11Þ for8c
2 H o2;1ðXÞ. The solution for this problem is exist, unique and satisfies the following inequality k/ð:; tÞk2L2ð0;lÞ6c1 k
u
k 2 Ho2ð0;lÞþ kyk 2 L2ð0;lÞ ;8
t 2 ½0; T: ð12ÞHere c1is independent from
u
and y.Now, we state following theorem about the Frechet differentiability of the functional using the solution of adjoint bound-ary value problem.
Theorem 1. The functional Ja(v) is Frechet differentiable on the set V and its gradient is
J0að
v
Þ ¼ n þ 2a
ðv
wÞ: ð13ÞHere, the function n(t) is the solution of the adjoint problem
n00 n ¼ Z l 0 Reðw/Þdx n0ð0Þ ¼ n0ðTÞ ¼ 0 ð14Þ
Proof. The increment of the functional Ja(
v
) according to the increment of the controlv
+ dve
V isdJað
v
Þ ¼ Jaðv
þ dv
Þ Jaðv
Þ ¼ 2 Z l 0 Re½ðwðx; TÞ yÞdwðx; TÞdx þ 2a
Z T 0 ½ðv
wÞdv
þ ðv
t wtÞdv
tdt þ kdwð:; TÞk2L2ð0;lÞþa
kdv
k 2 H1ð0;TÞ: ð15ÞHere the difference function d
w
= dw
(x, t) is the solution of the following difference problem according to the increment of the controlv
+ dve
V; i@dw @t þ a0 @2dw @x2 ðv
ðtÞ þ dv
ðtÞÞdw ¼ dv
ðtÞw; ðx; tÞ 2X
ð16Þ dwðx; 0Þ ¼ 0; dwð0; tÞ ¼ dwðl; tÞ ¼ 0:The solution of the problem(16)satisfies the inequality
kdwð:; tÞk2L2ð0;lÞ6c2kd
v
k2
L2ð0;TÞ;
8
t 2 ½0; T: ð17ÞHere c2is independent from dv.
With some manipulations we have the following equation;
2 Z l 0 Re½ðwðx; TÞ yðxÞÞdwðx; TÞdx ¼ Z X d
v
ðtÞReðw/Þdxdt Z X dv
ðtÞReðdw/Þdxdt:Using the above equality we can rewrite(15)as
dJað
v
Þ ¼ Z X dv
ðtÞReðw/Þdxdt þ 2a
Z T 0 ½ðv
wÞdv
þ ðv
t wtÞdv
tdt dt þ Rðdv
Þ ð18Þ where the term R(dv) isRðd
v
Þ ¼ kdwð:; TÞk2L2ð0;lÞþa
kdv
k 2 W1 2ð0;TÞ Z X dv
ðtÞReðdw/Þdxdt ð19Þand it satisfies the inequality
jRðd
v
Þj 6 c3kdv
k2H1ð0;TÞ ð20ÞHere c3is independent from dv.
Therefore the increment dJa(
v
) in(18)is rearranged such asdJað
v
Þ ¼ ZT 0 Z l 0 Reðw/Þdx ( ) dv
ðtÞdt þ 2a
hv
w; dv
iH1 ð0;TÞþ oðkdv
kH1 ð0;TÞÞ ð21ÞIn order to obtain the inner product in the space H1(0, T), we purpose a second adjoint problem in the following; n00 n ¼ Z l 0 Reðw/Þdx n0ð0Þ ¼ n0ðTÞ ¼ 0 ð22Þ
With the solution of this problem the statement(21)is written such as
dJað
v
Þ ¼ ZT 0 nðtÞdv
ðtÞ þ n0ðtÞdv
0ðtÞ ½ dt þ 2a
hv
w; dv
iH1ð0;TÞþ oðkdv
kH1ð0;TÞÞ or dJaðv
Þ ¼ n þ 2ha
ðv
wÞ; dv
iH1 ð0;TÞþ oðkdv
kH1 ð0;TÞÞ: ð23ÞHence the functional Ja(
v
) is Frechet differentiable on the set V and its gradient is given byJ0
að
v
Þ ¼ n þ 2a
ðv
wÞ: ð24Þ5. Lipschitz continuity of the gradient and necessary condition for optimal control
In this section, we give a theorem about Lipschitz continuity of the gradient. Then, to state the necessary condition will be possible.
Theorem 2. The gradient J0að
v
Þ holds the following Lipschitz inequalitykJ0að
v
þ dv
Þ J0
að
v
Þk2
H1ð0;TÞ6c4kd
v
k2H1ð0;TÞ ð25ÞHere c4is independent from dv.
Proof. The increment of the functional J0
að
v
Þ according to the increment of the controlv
+ dve
V isJ0
að
v
þ dv
Þ J0
að
v
Þ ¼ ndþ 2a
ðv
þ dv
wÞ n þ 2a
ðv
wÞ ¼ dn þ 2a
dv
: ð26Þ Here the function dn(t) is the solution of the increment problemdn00ðtÞ dnðtÞ ¼ Z l
0
Reðwdd/þ dw/Þdx: ð27Þ
The problem(27)has a solution in H1(0, T) and this solution satisfies the inequality kdnðtÞkH1ð0;TÞ6 Z l 0 ðjwdkd/j þ jdwk/jÞdx L 2ð0;TÞ ð28Þ The function d/ðx; tÞ ¼ /dðx; tÞ /ðx; tÞ ¼ /ðx; t;
v
þ dv
Þ /ðx; t;v
Þthat takes place in the right hand side of(28)is the solution of the problem
i@d/ @t þ a0 @2d/ @x2 ð
v
ðtÞ þ dv
ðtÞÞd/ ¼ dv
ðtÞ/; ðx; tÞ 2X
d/ðx; TÞ ¼ 2idwðx; TÞ; d/ð0; tÞ ¼ d/ðl; tÞ ¼ 0: ð29Þand this function holds the inequality
kd/ð:; tÞk2L2ð0;lÞ6c5kd
v
k2
L2ð0;TÞ ð30Þ
The other function
w
dthat takes place in the right hand side of(28)holds the inequality kwdk 2 L2ðXÞ6ku
k 2 Ho2ð0;lÞ: ð31ÞTherefore the inequality(28)has the property
kdnðtÞk2H1ð0;TÞ62kwdk2L 2ðXÞmax06t6Tðkd/ð:; tÞk 2 L2ð0;lÞÞ þ 2k/k 2 L2ðXÞmax06t6Tðkdwð:; tÞk 2 L2ð0;lÞÞ:
If the inequalities(12), (17), (30) and (31)about the functions, /, d
w
(.,t), d/(.,t) andw
dare used then the followingassess-ment is obtained; dnðtÞ k k2W1 2ð0;TÞ 6c6kd
v
k2 L2ð0;TÞ: ð32ÞHere c6is independent from dv.
Taking the norm of(26)in the space H1(0, T), we acquire; kJ0að
v
þ dv
Þ J0
að
v
Þk2
H1ð0;TÞ62kdnðtÞk2H1ð0;TÞþ 8ð
a
Þ2kdv
ðtÞk2H1ð0;TÞand considering(32), we have the following inequality for the gradient J0
að
v
Þ; kJ0aðv
þ dv
Þ J 0 aðv
Þk 2 H1ð0;TÞ6c7kdv
k2H1ð0;TÞ:Here c7is independent from dv.
Hence, it has been demonstrated that the gradient J0
að
v
Þ is continuous on the set V and moreover it holds the Lipschitzcondition with the constant c7> 0. The fact that the functional Ja(
v
) is continuously differentiable and that the set V is convex, in that case according to theorem in[11](p. 28) the following inequality is validJ0
að
v
Þ;v
v
H1
ð0;TÞP0;
8
v
2 Vif the control
v
⁄e
V is the minimum point of the functional Ja(
v
). So the necessary condition for optimal control has been obtained.References
[1]A.D. _Iskenderov, G.Ya. Yagubov, A variational method for solving the inverse problem of determining the quantum-mechanical potential, Soviet Math. Dok1. 38 (3) (1989) 637–641.
[2]A.D. _Iskenderov, G.Ya. Yagubov, Optimal control of nonlinear quantum mechanical systems, J. Autom. Telemech. 12 (1989) 27–38. [3]B. Yıldız, G.Ya. Yagubov, On an optimal control problem, J. Comput. Appl. Math. 88 (1997) 275–287.
[4]Bünyamin Yıldız, Murat Subasßı, On the optimal control problem for linear Schrödinger equation, Appl. Math. Comput. 121 (2001) 373–381. [5]Murat Subasßı, An optimal control problem governed by the potential of a linear Schrödinger equation, Appl. Math. Comput. 131 (1) (2002) 95–106. [6]Gabil Yagubov, Fatma Toyog˘lu, Murat Subasßı, An optimal control problem for two-dimensional Schrödinger equation, Appl. Math. Comput. 218 (2012)
6177–6187.
[7]A.D. Iskenderov, R.Q. Tagıyev, Q.Y. Yagubov, Optimization Methods, Çasßıoglu Press, Baku, 2002. p. 400.
[8]O.A. Ladyzhenskaya, V.A. Solonnikov, N.N. Uraltseva, Linear and quasi-linear equations of parabolic type, in: Translations of Mathematical Monographs, American Mathematical Society, Rhode Island, 1968.
[9]M. Goebel, On existence of optimal control, Math. Nachr. 93 (1979) 67–73. [10]K. Yosida, Functional Analysis, Springer-Verlag, New York, 1980. p. 624.