Second variation

The second variation is the quadratic term in a functional's expansion along $u+\epsilon \eta$; positivity on admissible variations is a local-minimum test.

For fixed-endpoint variations, a second variation is coercively positive if the displayed estimate holds for one constant $c>0$ and every admissible variation. The endpoint Poincaré inequality makes this equivalent to a positive lower bound in the $W^{1,2}$ norm. It is stronger than strict positivity on each nonzero variation: $q(\phi )=\int {\phi }^2$ is strictly positive, but high-frequency variations make its ratio to $\int {\phi }^{\prime 2}$ tend to zero.

For $q=\int (p{\phi }^{\prime 2}+2r\phi {\phi }^{\prime }+s{\phi }^2)$ with continuous coefficients, coercive positivity implies $p\ge c$. Indeed, if $p-c<0$ on an interval, multiply a smooth function supported there by $\sin (nx)$. Its negative highest-derivative contribution has order $n^2$, while the mixed term has order at most $n$ and the remaining term is bounded. This contradicts the displayed inequality for large $n$. In particular coercive positivity implies the Strengthened Legendre condition, whereas mere strict positivity does not.

A nonnegative second variation requires its highest-derivative coefficient $P=f_{vv}$ to be nonnegative. If it were negative on an interval, a smooth compactly supported factor times $\sin (nx)$ would have negative derivative energy of order $n^2$, while its zero-order energy stays bounded. Large $n$ would contradict nonnegativity. This is a local necessary condition; it does not rule out negative directions caused by the lower-order terms.

Strict positivity of the highest-derivative coefficient on the closed parameter interval is the strengthened Legendre condition. Continuity then gives a uniform positive lower bound and makes the Jacobi equation a regular second-order equation. It permits the Jacobi positivity criterion. It is not necessary for mere strict positivity of a second variation: $\int {\phi }^2$ is strictly positive on nonzero variations even though its derivative coefficient vanishes.

The Jacobi equation is the linearization of an Euler-Lagrange equation about a stationary path. Its solutions describe infinitesimal one-parameter families of stationary paths.

If $w$ is a nonvanishing solution of $-(P{w}^{\prime }{)}^{\prime }+Qw=0$, expanding the right-hand side gives the displayed identity. Integrating expresses a scalar second variation as a square plus a boundary term. A simple zero of $w$ at a fixed endpoint causes no difficulty for Lipschitz endpoint-vanishing variations, since $\phi =O(x-a)$ makes the boundary term tend to zero. This identity is the scalar form of a ground-state transform for a weighted Dirichlet energy.

A conjugate point marks a nonzero endpoint-vanishing Jacobi variation and the loss of positive definiteness of the second variation.

For the energy functional of a Riemannian manifold, this variational condition yields conjugate points along a geodesic.

Assume the Strengthened Legendre condition $P>0$ and let $q(\phi )={\int }_a^b(P{\phi }^{\prime 2}+Q{\phi }^2)$. Then $q$ is strictly positive on all nonzero fixed-endpoint Lipschitz variations exactly when the normalized Jacobi equation solution with $w(a)=0$, $P(a)w^{\prime }(a)=1$ has no zero in $(a,b]$. A zero at $c$ gives a nonzero variation equal to $w$ up to $c$ and zero thereafter, with $q=0$. Conversely, if no zero occurs, the Jacobi identity for the second variation writes $q$ as a nonnegative square. Equality forces $\phi =Cw$, and the endpoint at $b$ forces $C=0$.

For continuous $P\ge 0,Q$, a scalar second variation $q=\int (P{\phi }^{\prime 2}+Q{\phi }^2)$ is strictly positive on nonzero fixed-endpoint Lipschitz variations if and only if every regularized form $q_{\eta }$ for $\eta >0$ has no conjugate point in $(a,b]$, and the original weak Jacobi boundary problem has no nonzero Lipschitz solution. The Jacobi positivity criterion applies to each regularized coefficient $P+\eta >0$. Letting $\eta$ decrease to zero gives $q\ge 0$. For a nonnegative quadratic form, $q(\phi )=0$ forces its mixed bilinear form with every test function to vanish, by considering $q(\phi +t\psi )$ for both signs of $t$. This is the weak Jacobi equation, so the kernel condition gives strict positivity. Necessity follows by the same argument in reverse. It distinguishes strict positivity from a uniform coercive estimate when the derivative coefficient degenerates.

Along a nonconstant affinely parametrized geodesic, the multiplicity of a conjugate point at positive time $t_0$ is the dimension of the space of Jacobi fields vanishing at both endpoints. The tangential component of such a Jacobi field is affine and vanishes at both endpoints, hence is zero. Uniqueness for the Jacobi field equation embeds this space into the $(n-1)$-dimensional normal space by $J\mapsto D_{t}J(0)$. Thus its dimension is at most $n-1$.

For a Jacobi field along a geodesic with nonpositive sectional curvature, the curvature convention with positive round-sphere curvature givesThus $|J{|}^2$ is nonnegative and convex. If it vanishes at both ends of a segment, convexity forces it to vanish everywhere. No nonzero Jacobi field can vanish at both ends, so the segment has no conjugate points. Only curvature of planes containing the geodesic tangent is needed, and completeness is unnecessary.

For $f(0)=f(T)=0$,with equality for a multiple of $\sin (\pi t/T)$.

Expand a real function with zero endpoint values in its Fourier sine series, $y={\sum }_{n\ge 1}{a}_n\sin (n\pi x/L)$. The Parseval identity gives $\int y^2=(L/2)\sum a_n^2$. For its weak derivative, the Fourier cosine series has coefficients $(n\pi /L)a_n$ and zero constant term, soThe derivative coefficient identity follows by integration by parts, rather than by assuming pointwise differentiation of an arbitrary series. This proves the Wirtinger inequality and its equality case. Under unit L2 norm, the only real minimizers are $y=\pm \sqrt{2/L}\sin (\pi x/L)$.

If a continuously differentiable $L$-periodic function $f$ has mean zero, thenEquality holds exactly for $f(s)=A\cos (2\pi s/L)+B\sin (2\pi s/L)$.