Nonsmooth problems in calculus of variations

MaRDI QIDQ4026949zbMATH OpenFDO

Authors V. F. Demyanov

Publication date 21 February 1993

zbMATH Keywords

nondifferentiability nonsmooth calculus

Mathematics Subject Classification ID

Nonsmooth analysis (49J52)

Review text

The author defines \(N\) variational functionals \[ J_ i(y) = \int_ a^ b F_ i(x,y(x),y'(x)) dx, \quad i=1,\ldots,N, \] and considers the problem of minimizing the composite functional \(J(y) = \Phi(J_ 1(y),\ldots,J_ N(y))\) over all piecewise smooth functions \(y\) defined on the given interval \([a,b]\) and satisfying prescribed endpoint conditions \(y(a)=y_ a\), \(y(b)=y_ b\). Although the problem is posed for a general real-valued function \(\Phi\) on \({\mathbf R}^ n\), which may be nondifferentiable, the analysis is confined to the case of \(\Phi(z_ 1,\ldots,z_ N) = \max\left\{z_ 1,\ldots,z_ N\right\}\). The function \(\Phi\) is the only source of nondifferentiability in the problem: each integrand \(F_ i\) is required to be smooth. The author proves that if the function \(y\) provides a strong local minimum for the composite function \(J\), then \(y\) must satisfy certain generalized forms of the Euler-Lagrange equation (EL) and the Weierstrass necessary condition (W). Equation (EL) is the usual Euler-Lagrange equation for the integrand \(\sum_{i\in R(y)} \lambda_ i F_ i\), where the Lagrange multipliers \(\lambda_ i\) are nonnegative, and \(R(y)=\left\{ i : J_ i(y)=\Phi(y)\right\}\) denotes the set of active indices. Regrettably, the precise statement of (W) -- part of Theorem 3.1 -- seems to have been lost during the printing of the article. However, the proof survives, so interested readers should be able to reconstruct the statement. The proof depends on constructing a family of admissible trajectories based on the minimizer \(y\) by applying \(n\) ``needle variations to the derivative. The first \(n-1\) of these are arbitrary; the last one is chosen carefully to satisfy the endpoint constraints. The result is a mapping \(\Gamma\) from \({\mathbf R}^{n-1}\) to the space of admissible arcs; necessary conditions are obtained by noticing that the zero vector minimizes \(J\circ\Gamma\) over \({\mathbf R}^{n- 1}\), and then decoding the resulting optimality condition with the help of suitable nonsmooth calculus for the function \(\Phi\). For the entire collection see [Zbl 0746.00073].

Recommendations

Cited in

(10)

This page was built for publication: Nonsmooth problems in calculus of variations

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4026949)