Two-scale convergence and compactness

A sequence vε two-scale converges weakly to v(x,y) if its integral against φ(x)ψ(x/ε) tends to the double integral against φ(x)ψ(y), for smooth compactly supported φ and smooth periodic ψ. Such tests are admissible without pointwise-evaluation difficulties. General tests require their own measurability and approximation conditions, so the definition should not quantify over arbitrary L² functions composed with x/ε. Bounded L² sequences have two-scale convergent subsequences in this setting, providing a compactness theorem for microscopic oscillations.

Taking ψ=1 shows that the ordinary weak limit is the cell average of v(x,y). Different microscopic profiles can therefore have the same ordinary weak limit. For instance, sin(2πx/ε) on a one-dimensional interval converges weakly to zero but has two-scale limit sin(2πy). Its squared norm tends to half the interval length rather than zero. The profile survives in two-scale space even though every fixed smooth macroscopic test averages it away.

If uε is bounded in H¹(Ω), compactness gives an ordinary weak H¹ limit u and strong L² convergence along a subsequence. Its gradient's two-scale limit has the form ∇u(x)+∇ᵧu₁(x,y), with u₁ square-integrable in x and periodic H¹ in y modulo constants. A mean-zero normalization can again select a representative. The microscopic component is a gradient, not an arbitrary vector field; compatibility with the original gradient structure constrains it.

Applying this result to periodic elliptic solutions and testing their weak equation with oscillatory functions yields the cell problem. The periodic coefficient is then combined with the retained microscopic gradient before averaging its flux. This order matters because weak convergence does not preserve products. The cell equation is what identifies the correlation for solutions; two-scale compactness alone merely supplies candidate microscopic fields. It does not make an arbitrary bounded sequence solve a homogenized PDE.

Weak two-scale convergence controls oscillatory test integrals but may still lose additional microscopic energy. A common Hilbert-space definition of strong two-scale convergence adds convergence of the L² norm to the L²(Ω×Y) norm of the two-scale limit. This differs from ordinary strong convergence because the limiting microscopic profile may depend on y. Suitable strong/weak two-scale product results can then justify correlations that ordinary weak limits cannot capture, under the integrability and admissibility hypotheses of those results.

The simplest diagnostic is to compare averages with squares. A zero-mean oscillation has a vanishing weak limit while its square has a positive mean. In a diffusion problem, the analogous lost correlation occurs between the coefficient and gradient. Confusing the averaged profile with the full two-scale limit can therefore produce an incorrect effective flux or energy. A rigorous calculation states which convergence applies to each factor and identifies whether a compactness theorem, a strong convergence result, or the PDE structure justifies the product passage.