the Schauder estimates
/ SHOW-der /
The Schauder estimates are the regularity theory of elliptic equations written in the Holder scale — the natural language of classical, pointwise smoothness. A function is Holder continuous with exponent alpha if its variation over a small distance r is controlled by r^alpha; the Holder space C^(2,alpha) collects functions whose second derivatives are Holder continuous. Schauder theory answers a precise question: if the data and coefficients of an elliptic equation are Holder continuous, exactly how smooth is the solution?
The answer, again, is the gain of two derivatives, now measured in Holder norms. The interior Schauder estimate says: if L u = f with L uniformly elliptic and its coefficients in C^(0,alpha) (Holder continuous), and f is in C^(0,alpha), then u is in C^(2,alpha) on any interior subregion, with the bound — the C^(2,alpha) norm of u on the smaller ball is not greater than a constant times (the C^(0,alpha) norm of f plus the sup norm of u). There is a matching boundary version: if in addition the boundary and the boundary data are C^(2,alpha), the estimate holds all the way up to the boundary. The proof rests on a beautiful idea — freeze the coefficients at a point so the operator becomes constant-coefficient, use the explicit estimates for that frozen operator, and treat the variation of the coefficients as a small Holder-controlled perturbation.
Why C^(0,alpha) rather than merely continuous? Because the bare gain fails for continuous-but-not-Holder data — there are continuous f for which the solution of - Laplacian u = f is NOT twice continuously differentiable. The Holder exponent is the precise amount of extra modulus that makes the two-derivative gain go through. Schauder estimates are the standard route to upgrading a weak solution to a classical C^2 solution, the workhorse behind existence proofs for nonlinear problems (via the method of continuity and fixed-point theorems), and the Holder-scale partner of the Sobolev H^k and the L^p Calderon-Zygmund estimates.
Take - Laplacian u = f on the unit ball with f in C^(0, 1/2) (Holder exponent one-half). The Schauder estimate guarantees u is in C^(2, 1/2): not just twice differentiable, but with second derivatives that are themselves Holder continuous with the same exponent. If instead f were merely continuous, this could fail — a classic counterexample produces a continuous f whose solution has unbounded second derivatives.
Holder data buys C^(2,alpha) solutions; merely-continuous data does not — the exponent is essential.
The Schauder constant blows up as alpha approaches 0 or 1, which is exactly why the endpoints fail. At alpha = 0 (mere continuity) and alpha = 1 (Lipschitz) the clean two-derivative gain is false; Schauder theory lives strictly in the open interval 0 < alpha < 1, and that restriction is not a technicality but the truth of the matter.