Abstract Generated abstract
This note addresses the construction of minimizing sequences for nonlinear variational problems, a point left open in general existence arguments based on convergence of such sequences. It proves that the Ritz method yields a minimizing sequence for a functional whose domain is linear, whose restrictions to finite-dimensional linear manifolds are continuously differentiable, and which is increasing and lower semicontinuous in a metric for which the chosen coordinate sequence is complete. The proof constructs finite-dimensional Ritz minimizers, compares them with approximations to an arbitrary minimizing sequence, and uses monotonicity of the resulting functional values to show convergence to the infimum.
Full Text
MATHEMATICS
S. G. MIKHLIN
ON THE RITZ METHOD IN NONLINEAR PROBLEMS
(Presented by Academician S. L. Sobolev on 3 X 1961)
In a number of works the existence of a solution of a variational problem for functionals of a fairly general form is established by proving the convergence of a minimizing sequence; among recent works on this topic we note the papers \((^{1,2})\). However, the important question of the actual construction of a minimizing sequence in the general case has remained open. For quadratic functionals bounded below, such a construction can be carried out by applying the Ritz method \((^3)\); I. I. Vorovich showed \((^4)\) that the Ritz method gives a minimizing sequence (whose convergence was proved by the same author) for functionals of the nonlinear theory of shells; an analogous result was obtained by A. Langenbach for some functionals of the theory of plasticity in the papers \((^2)\).
The purpose of the present article is to prove the following theorem.
Theorem. Let the functional \(\Phi(u)\) satisfy the following requirements: a) its domain of definition \(D(\Phi)\) is a linear set; b) on every finite-dimensional linear manifold in its domain of definition this functional is continuously differentiable; c) in some metric \(\rho\), defined on \(D(\Phi)\), the functional in question is increasing and lower semicontinuous.
Let, further, the coordinate sequence \(\{\varphi_n\}\) be subject to the usual conditions: 1) \(\varphi_n \in D(\Phi)\); 2) the elements \(\varphi_1, \varphi_2, \ldots, \varphi_n\) are linearly independent for every \(n\); 3) the coordinate sequence is complete in the metric \(\rho\).
Under the enumerated conditions, the Ritz method gives a minimizing sequence for the functional \(\Phi(u)\).
Proof. For any \(n\) one can construct the \(n\)-th approximate solution by Ritz’s method. Indeed, the expression \(\Phi\left(\sum_{k=1}^{n} a_k \varphi_k\right)\) is a function of the variables \(a_1, a_2, \ldots, a_n\), continuously differentiable for all values of these variables and tending to \(+\infty\) if \(a_1^2 + a_2^2 + \ldots + a_n^2 \to \infty\). This function attains its absolute minimum at least at one point, which is at a finite distance from the origin, and at this point
\[ \frac{\partial \Phi\left(\sum_{k=1}^{n} a_k \varphi_k\right)}{\partial a_j} = 0,\qquad j = 1, 2, \ldots, n. \]
We shall now prove that the approximate Ritz solutions form a minimizing sequence for the functional \(\Phi(u)\). Let \(\inf \Phi(u)=d\). Construct a minimizing sequence \(\{u^{(n)}\}\) such that
\[ \Phi\left(u^{(n)}\right) \leq d+\frac{1}{n}. \]
In view of condition 3), for each \(u^{(n)}\) one can choose such a linear combination
\[ v^{(N_n)}=\sum_{k=1}^{N_n} \alpha_k^{(n)}\varphi_k, \]
that \(\rho\bigl(u^{(n)}, v^{(N_n)}\bigr)<\delta_n\); choose the number \(\delta_n\) so small that, for any \(v\) satisfying the inequality \(\rho\bigl(u^{(n)}, v\bigr)<\delta_n\), one has \(\Phi\bigl(u^{(n)}\bigr)-\Phi(v)\geqslant -\frac{1}{n}\). Then
\[ \Phi\bigl(v^{(N_n)}\bigr)\leqslant \Phi\bigl(u^{(n)}\bigr)+\frac{1}{n} \]
and, a fortiori,
\[ \Phi\bigl(v^{(N_n)}\bigr)\leqslant d+\frac{2}{n}. \]
It follows that \(\{v^{(N_n)}\}\) is a minimizing sequence. Let
\[ u_p=\sum_{k=1}^{p} a_k\varphi_k \]
denote the \(p\)-th approximate solution by Ritz’s method. Then
\[ \Phi(u_{N_n})\leqslant \Phi\bigl(v^{(N_n)}\bigr)\leqslant d+\frac{2}{n}, \]
and \(\{\Phi(u_{N_n})\}\) is also a minimizing sequence. Finally, since \(\Phi(u_n)\) decreases monotonically as \(n\) increases, while the sequence \(\Phi(u_{N_n})\to d\), it follows that \(\Phi(u_n)\to d\), as was required to prove.
Received
28 IX 1961
REFERENCES
- I. V. Gel’man, DAN, 120, No. 3 (1958); DAN, 122, No. 4 (1958); Uch. zap. Leningr. ped. inst. im. A. I. Gertsena, 166, 255 (1958); Izv. Vyssh. uchebn. zaved., Matematika, No. 4 (17) (1960).
- A. Langenbach, DAN, 121, No. 2 (1958); Vestn. LGU, No. 1, ser. mat., mekh. i astr., issue 1, 38 (1961); A. Langenbach, Wiss. Zs. Humboldt Univ. Berlin, Math.-naturwiss. Reihe, 9 (1959–1960).
- S. G. Mikhlin, Direct Methods in Mathematical Physics, 1950.
- I. I. Vorovich, DAN, 105, No. 1 (1955); Izv. AN SSSR, ser. matem., 19, No. 4, 173 (1955).