Calculus of variations and optimal control with generalized derivative

From MaRDI portal

DOI10.1216/RMJ.2023.53.1337MaRDI QIDQ6184982zbMATH OpenOpenAlexFDO

Authors Maria N. F. Barreto, Gastão S. F. Frederico, J. Vanterler da Costa Sousa, Juan E. Nápoles Valdes

Publication date 5 January 2024

Published in Rocky Mountain Journal of Mathematics (Search for Journal in Brave)

Full work available at URL https://projecteuclid.org/journals/rocky-mountain-journal-of-mathematics/volume-53/issue-5/CALCULUS-OF-VARIATIONS-AND-OPTIMAL-CONTROL-WITH-GENERALIZED-DERIVATIVE/10.1216/rmj.2023.53.1337.full


zbMATH Keywords

calculus of variationsd'Alembert principledu Bois-Reymond optimality conditionEuler-Lagrange equationLagrangianNoether's theoremoptimal controltime-fractional Schrödinger equation

Mathematics Subject Classification ID

Fréchet and Gateaux differentiability in optimization (49J50) Optimality conditions for free problems in one independent variable (49K05) Optimality conditions for problems involving partial differential equations (49K20) Variational principles of physics (49S05) Closed and approximate solutions to the Schrödinger, Dirac, Klein-Gordon and other equations of quantum mechanics (81Q05)




Review text

For given \(0\leq t_{0}<t_{1}\in \mathbb{R}\), \(\mathbb{U}\) a connected and open subset of \(\mathbb{R}^{n}\), and two closed subsets \(\mathbb{U}_{0}, \mathbb{U}_{1}\subset \overline{\mathbb{U}}\), \(AC([t_{0},t_{1}],\mathbb{R} ^{n})\) is the class of all absolutely continuous arcs \(q:[t_{0},t_{1}] \rightarrow \mathbb{R}^{n}\), and \(\Upsilon \) the set of admissible arcs: \( \Upsilon =\{q\in AC([t_{0},t_{1}],\mathbb{R}^{n})\mid \forall t\in \lbrack t_{0},t_{1}]\), \(q(t)\in \overline{\mathbb{U}}\), \(q(t_{0})\in U_{0}\), \( q(t_{1})\in U_{1}\}\). The authors consider the generalized problem of calculus of variations: find \(q^{\triangleright }\in \Upsilon \) such that: \( I(q^{\triangleright })=min\{I(q)\mid q\in \Upsilon \}\), where \(I(q)=\phi _{0}(q(t_{0}))+\phi _{1}(q(t_{1}))+J_{F,t_{0}}^{\alpha }[L(t,q(t),\frac{ d_{F}^{\alpha }q}{dt^{\alpha }}(t))](t_{1})\), the Lagrangian \( L:[t_{0},t_{1}]\times \overline{\mathbb{U}}\times \mathbb{R}^{n}\rightarrow \mathbb{R}\) belonging to \(L_{\alpha }^{1}([t_{0},t_{1}])\), \(\phi _{0},\phi _{1}:\overline{\mathbb{U}}\rightarrow \mathbb{R}\) being given continuous functions, \(J_{F,a}^{\alpha }f\) being the generalized integral of order \( \alpha \in (0,1]\) of a continuous function \(f:[a,t]\rightarrow \mathbb{R}\) defined as: \(J_{F,a}^{\alpha }f(t)=\int_{a}^{t}f(x)F(x,\alpha )dx\), and \( \frac{d_{F}^{\alpha }}{dt^{\alpha }}\) being the generalized derivative of order \(\alpha \in (0,1]\) of a function \(f:[a,b]\rightarrow \mathbb{R}\) defined as: \(\frac{d_{F}^{\alpha }f}{dt^{\alpha }}=lim_{\varepsilon \rightarrow 0}\frac{f(t+\varepsilon F(\alpha ,t))-f(t)}{\varepsilon }\). \(F\) is a positive and absolutely continuous function which will be chosen as the one-parameter Mittag-Leffler function \(E_{\alpha }\), defined by \(E_{\alpha }(z)=\sum_{k=0}^{\infty }\frac{z^{k}}{\Gamma (1+\alpha k)}\), \(\alpha ,z\in \mathbb{C}\), \(Re(\alpha )>0\), where \(\Gamma \) is the Gamma function. The authors recall the main properties of these generalized derivative and integral. Defining the augmented Lagrangian as \(\widetilde{L}(t,q(t),\frac{ d_{F}^{\alpha }q}{dt^{\alpha }}(t);F(t,\alpha ))=\frac{L(t,q(t),\frac{ d_{F}^{\alpha }q}{dt^{\alpha }}(t))}{F(t,\alpha )}\), the authors assume that: \(\widetilde{L}\) is a convex function with respect to the third variable; There exists \(k\geq 0\) and a Nagumo function \(\rho :\mathbb{R} _{+}\rightarrow \mathbb{R}_{+}\) for the functional \(I\) such that \(\widetilde{ L}(t,x,v;F(t,\alpha ))\geq \rho (\left\vert v\right\vert )-k\), \(\phi _{0}(q),\phi _{1}(q)\geq -k\); For all \(r>0\) there exists \(K(r)>0\) such that \( \left\vert \widetilde{L}(t,x,v;F(t,\alpha ))-\widetilde{L}(t,y,v;F(t,\alpha ))\right\vert <K(r)\theta (\left\vert x-y\right\vert )\rho (\left\vert v\right\vert )\), for all \(t\in \lbrack t_{0},t_{1}]\), \(x,y\in \mathbb{U}\cap B_{r}\), \(v\in \mathbb{R}^{n}\), \(\theta :\mathbb{R}_{+}\rightarrow \mathbb{R} _{+}\) being a nondecreasing function such that \(lim_{r\rightarrow 0}\theta (r)=0\); Either \(\mathbb{U}_{0}\) or \(\mathbb{U}_{1}\) is a compact set. The main result proves that the problem has at least one solution. If \( \widetilde{L}\) is strictly convex for any \(t\in \lbrack t_{0},t_{1}]\), this solution is unique. For the proof, the authors consider a minimizing sequence \((q_{n})_{n\in \mathbb{N}}\) and, assuming that \(\mathbb{U}_{0}\) is compact, they prove that the sequence of generalized fractional conformable derivatives \((\frac{d_{F}^{\alpha }q_{n}}{dt^{\alpha }})_{n\in \mathbb{N}}\) is equi-absolutely fractional conformable integrable, from which they deduce that \((q_{n})_{n\in \mathbb{N}}\) is equicontinuous and uniformly bounded. They conclude applying Arzela-Ascoli theorem and the generalized du Bois-Reymond lemma. They derive the Euler-Lagrange equation, the d'Alembert principle and the du Bois-Reymond optimality condition, first computing the first-order Gâteaux-differential of \(I\). Within the present context, the authors prove a generalized Noether theorem, assuming that the variational integral of \(I\) is invariant in an appropriate sense and proving that a quantity \(C(t,q(t),\frac{d_{F}^{\alpha }q}{dt^{\alpha }}(t))\) is preserved along any solution \(q\) of the Euler-Lagrange equations. The authors then consider the optimal control problem in Lagrange form: \(J_{F,t_{0}}^{\alpha }[L(t,q(t),u(t))](t_{1})\rightarrow min \), subject to the Cauchy problem \( \frac{d_{F}^{\alpha }q}{dt^{\alpha }}(t)=\varphi (t,q(t),u(t))\), \( q(t_{0})=q_{0}\in \mathbb{R}^{d}\), \(q\in C^{1}([t_{0},t_{1}],\mathbb{R}^{d})\) , \(u\in C^{0}([t_{0},t_{1}],\mathbb{R}^{l})\).\ The Lagrangian \( L:[t_{0},t_{1}]\times \mathbb{R}^{d}\times \mathbb{R}^{l}\rightarrow \mathbb{ R}\) and the state equation \(\varphi :[t_{0},t_{1}]\times \mathbb{R} ^{d}\times \mathbb{R}^{l}\rightarrow \mathbb{R}^{d}\) are assumed to be \(C^{1} \)-functions with respect to all their arguments. The authors first prove the existence of a solution to the equation \(\frac{d_{F}^{\alpha }y}{dt^{\alpha } }(t)=f(x,y(x))\), assuming regularity properties on \(f\), applying a fixed-point theorem for contractive mappings in complete metric spaces. They prove a Grönwall-Bellman inequality in integral form, a weak Pontryagin maximum principle, a Noether principle for optimal control. In the last part of the paper, the authors consider the time-fractional linear Schrödinger equation: \(i\hbar \frac{d_{F}^{\alpha }\psi }{dt^{\alpha }} (t,x)+\frac{\hbar ^{2}}{2m}\sum_{j=1}^{d}\frac{\partial ^{2}\psi }{\partial x_{j}^{2}}(t,x)=U(x)\psi (t,x)\), where \(\hbar =\frac{h}{2\pi }\), \(h\) being the Planck constant, \(m>0\) the mass of particle, the potential energy \(U: \mathbb{R}^{d}\rightarrow \mathbb{R}\) and \(\psi :\mathbb{R}^{d}\times \mathbb{R}\rightarrow \mathbb{C}\) is the wave function associated to the particle on \(C^{2}(\mathbb{R}\times \mathbb{R}^{d},\mathbb{C})\). The authors here prove that the solutions to the Schrödinger equation coincide with the extremals of some quantum Lagrangian.


Recommendations


Cites work


Cited in
(1)








This page was built for publication: Calculus of variations and optimal control with generalized derivative

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6184982)

Retrieved from "https://portal.mardi4nfdi.de/w/index.php?title=Calculus_of_variations_and_optimal_control_with_generalized_derivative&oldid=102578432"
Powered by MediaWiki