= Second variation
{wiki}
For a twice differentiable <functional>, the <second variation> at $u$ in an admissible direction $\eta$ is
$$
\delta^2E[u;\eta]=\left.\frac{d^2}{d\epsilon^2}E[u+\epsilon\eta]\right|_{\epsilon=0}.
$$
Its contribution to the <Taylor expansion> is $\epsilon^2\delta^2E[u;\eta]/2$. At a stationary local minimum it must be nonnegative in every direction. A negative direction rules out a minimum, but mere directional strict positivity in an infinite-dimensional space is weaker than a uniform coercive bound and does not alone supply a sufficient-minimum theorem.
For zero-endpoint functions on $(0,L)$, consider
$$
E[u]=\frac12\int_0^L(u\prime{}^2-\lambda u^2)dx.
$$
At $u=0$, the second variation is $\int_0^L(\eta\prime{}^2-\lambda\eta^2)dx$. The <Wirtinger inequality> makes this coercively positive when $0\le\lambda<\pi^2/L^2$. At equality, $\eta=\sin(\pi x/L)$ is a zero direction; above it, this same direction has negative second variation. Thus lower-order terms can destroy minimality despite a positive highest-derivative coefficient. The <Legendre necessary condition>, <Jacobi equation> and <conjugate point> organize these tests for general variational integrals.
Back to article page