Γ-convergence for oscillating energies

Γ-convergence to E₀ requires a liminf inequality for every sequence uε converging to u in the chosen topology. It prevents some hidden microscopic construction from having energy below the proposed limit. A complementary recovery sequence must exist for each u, converging to that u and attaining the limit energy from above. The recovery sequence is allowed to depend on u. Pointwise convergence Eε(u)→E₀(u) for fixed u neither replaces these two conditions nor generally follows from them.

For oscillatory quadratic energies, the recovery sequence uses cell correctors to redistribute the gradient within each period. Its potential perturbation may vanish in L² while its gradient changes by order one. This mechanism explains why the Γ-limit uses the homogenized tensor instead of the arithmetic coefficient average of a fixed smooth function. Constructing recovery sequences near the boundary requires preserving the prescribed trace, commonly through cutoffs and approximation. Ignoring that constraint can produce an inadmissible upper-bound construction.

Take Eε(u)=one half of the integral of ∇u·A(x/ε)∇u, extended by infinity outside H₀¹. Uniform ellipticity and the Poincaré inequality bound every common energy sublevel in H₀¹. Rellich compactness makes those sublevels precompact in L², giving equicoercivity in the chosen topology. A fixed affine Dirichlet class can be treated by subtracting an H¹ extension of its trace; the same compactness reasoning then applies with constants depending on that extension.

With Γ-convergence and equicoercivity, suitably chosen approximate minimizers have convergent subsequences whose limits minimize E₀, and minimum values converge when the functionals are proper with finite attainable comparison energy. If the limiting minimizer is unique, the whole minimizing sequence converges in L². This theorem concerns global minimization. It does not automatically describe local minima, gradient-flow dynamics, or convergence of microscopic gradients in a strong norm. Each additional statement requires further structure beyond the two Γ inequalities.

A fixed L² load adds the continuous strong-L² term −∫fu, so the same Γ-limit persists with that load and the minimizer solves the homogenized weak equation. A more general H⁻¹ load is continuous on weak H¹ bounded sets, but it is not a continuous functional on all of L². One must then formulate the compactness and perturbation argument in the appropriate spaces rather than invoking strong-L² continuity incorrectly. The topology is part of the theorem, not an omitted technical label.

For symmetric quadratic problems, energy convergence and G-convergence are closely linked, especially when formulated locally on all subdomains. That connection explains why variational and PDE approaches identify the same effective tensor. A single global energy problem should not be used to claim every possible localized coefficient identification without checking the theorem's scope. Γ-convergence is powerful because it records optimal microscopic relaxation under a precise topology, while leaving unrelated dynamics and quantitative rates to separate analyses.