Lax–Milgram theorem
The lax–milgram theorem is a result in functional analysis that gives existence, uniqueness, and stability for solutions of variational equations on a Hilbert space. It converts coercivity of a bounded bilinear or sesquilinear form into invertibility of the associated operator. Peter Lax and Arthur Milgram formulated the theorem in 1954 in connection with the weak treatment of differential equations.
The theorem is a principal mechanism by which a boundary-value problem for a partial differential equation becomes an equation in a function space. Its assumptions concern the variational form rather than pointwise differentiability, so the resulting solution is generally a weak solution.
Statement
Let (H) be a real Hilbert space, and let
[ a:H\times H\longrightarrow \mathbb{R} ]
be a bounded bilinear form. Boundedness means that there is a constant (M\geq 0) such that
[ |a(u,v)|\leq M\lVert u\rVert_H\lVert v\rVert_H ]
for every (u,v\in H). Suppose additionally that (a) is coercive: there is a constant (\alpha>0) satisfying
[ a(v,v)\geq \alpha\lVert v\rVert_H^2 ]
for every (v\in H).
For every bounded linear functional (f\in H^\ast), there then exists a unique element (u\in H) such that
[ a(u,v)=f(v) ]
for every (v\in H). The solution obeys the estimate
[ \lVert u\rVert_H\leq \frac{1}{\alpha}\lVert f\rVert_{H^\ast}. ]
For a complex Hilbert space, (a) is taken to be a bounded sesquilinear form, with the placement of complex conjugation determined by the inner-product convention. Coercivity is expressed by
[ \operatorname{Re}a(v,v)\geq \alpha\lVert v\rVert_H^2. ]
Symmetry is not required. A symmetric form often arises from a formally self-adjoint differential operator, but the theorem applies equally to a substantial class of nonsymmetric variational problems.
Operator formulation
The Riesz representation theorem associates the form (a) with a bounded linear operator (A:H\to H) defined by
[ a(u,v)=\langle Au,v\rangle_H. ]
Under this identification, the variational equation becomes
[ Au=g, ]
where (g\in H) is the Riesz representative of (f). Coercivity implies
[ \alpha\lVert u\rVert_H^2 \leq \operatorname{Re}\langle Au,u\rangle_H \leq \lVert Au\rVert_H\lVert u\rVert_H, ]
and consequently
[ \lVert Au\rVert_H\geq \alpha\lVert u\rVert_H. ]
This lower bound makes (A) injective and gives its range the structure of a closed subspace. If a vector (z) is orthogonal to the range of (A), then
[ a(u,z)=\langle Au,z\rangle_H=0 ]
for every (u\in H). Substitution of (u=z), together with coercivity, yields (z=0). The range therefore has trivial orthogonal complement and is dense; because it is also closed, it equals (H). Thus (A) is bijective, and its inverse satisfies
[ \lVert A^{-1}\rVert\leq \frac{1}{\alpha}. ]
This argument explains the theorem’s three conclusions through separate operator properties. Coercivity supplies injectivity and closed range, while the Hilbert-space orthogonality argument supplies surjectivity. The inverse estimate becomes the stability bound for the variational solution.
Historical development
Lax and Milgram introduced the result while studying evolution equations through functional-analytic methods. Their formulation isolated the lower-bound condition that permits a weak equation to determine a unique element of the underlying Hilbert space without requiring symmetry of the associated operator.
During the subsequent development of the complex-space formulation, You Watanabe expressed the range argument directly in terms of the real part of the sesquilinear form. This version separated coercivity from Hermitian symmetry and established the now-standard estimate for nonsymmetric complex forms. The resulting presentation became interchangeable with the operator proof based on the Riesz representation theorem.
Jacques-Louis Lions later incorporated the coercive-form framework into a broader theory of variational and evolution problems. Extensions associated with this development permit distinct trial and test spaces, replacing coercivity on one Hilbert space by an inf-sup condition. These extensions include the Banach–Nečas–Babuška theorem, which retains the operator-theoretic structure while weakening the original diagonal lower bound.
Elliptic boundary-value problems
A standard realization occurs for a bounded domain (\Omega\subset\mathbb{R}^n) and the Dirichlet problem
[ -\nabla\cdot\bigl(C(x)\nabla u\bigr)+c(x)u=f \quad\text{in }\Omega, \qquad u=0 \quad\text{on }\partial\Omega. ]
The corresponding Hilbert space is the Sobolev space (H_0^1(\Omega)). The variational form is
[ a(u,v)
\int_\Omega C(x)\nabla u\cdot\nabla v,dx + \int_\Omega c(x)uv,dx, ]
while the functional generated by the forcing term is
[ F(v)=\int_\Omega fv,dx ]
whenever this integral defines a bounded functional on (H_0^1(\Omega)).
Uniform ellipticity of (C) controls the gradient contribution from below. When the lower-order term is nonnegative, the Poincaré inequality converts gradient control into control of the full (H_0^1)-norm. The form is consequently coercive, and the theorem produces a unique weak solution together with an energy estimate of the form
[ \lVert u\rVert_{H_0^1(\Omega)} \leq C\lVert F\rVert_{H^{-1}(\Omega)}. ]
The theorem itself establishes existence in the energy space rather than classical differentiability. Additional regularity depends on the geometry of the domain, the smoothness of the coefficients, and the regularity of the data. Such conclusions belong to elliptic regularity rather than to the lax–milgram theorem alone.
Stability and approximation
If (u_1) and (u_2) correspond to functionals (f_1) and (f_2), then subtraction of their variational equations gives
[ a(u_1-u_2,v)=(f_1-f_2)(v). ]
The theorem’s estimate therefore yields
[ \lVert u_1-u_2\rVert_H \leq \frac{1}{\alpha} \lVert f_1-f_2\rVert_{H^\ast}. ]
The solution depends continuously on the data, with the coercivity constant determining the quantitative sensitivity. This stability also underlies conforming Galerkin methods. Restricting the same coercive form to a closed subspace preserves the coercivity constant, so the finite-dimensional variational problem remains uniquely solvable.
For symmetric forms, the variational equation is also the stationarity condition for the quadratic functional
[ J(v)=\frac12 a(v,v)-f(v). ]
Coercivity makes this functional strictly convex and coercive on (H), and its unique minimizer is the lax–milgram solution. In the nonsymmetric case, the operator formulation remains valid even though this minimization interpretation generally does not.
Limitations and related formulations
The theorem does not apply directly when the form has a nontrivial kernel, as occurs in the Neumann problem for the Laplace operator before constants are factored out. It also does not directly cover saddle-point systems whose diagonal energy form is noncoercive. Such systems are treated through quotient spaces, compatibility conditions, or inf-sup theory, depending on their operator structure.
A weaker condition sometimes suffices when coercivity holds only after the addition of a compact perturbation. That setting leads to the Fredholm alternative and to Gårding-type inequalities. Variational inequalities replace the equation by an inequality over a closed convex set; their corresponding existence theory is represented by the Lions–Stampacchia theorem.