Uniform ellipticity and principal symbols

For a divergence-form operator $Lu=-\partial_i(a_{ij}(x)\partial_j u)$, the leading derivatives are described by the coefficient matrix A. When the coefficients are differentiable, expanding the product also produces first-order terms; these do not enter the second-order principal symbol. With the convention that differentiating a Fourier mode contributes i times its frequency, the two derivatives and the leading minus sign give the real quadratic symbol xi-transpose-A-xi. Stating this convention avoids confusing a positive elliptic operator with a negative Laplacian symbol.

The nondivergence operator $-a_{ij}(x)\partial_i\partial_j u$ has the same principal symbol but is not generally the same operator when A varies in space. Divergence form has a natural weak formulation requiring only bounded measurable coefficients and first weak derivatives of u. Nondivergence equations require a different interpretation at low regularity. Agreement of the principal symbols therefore identifies the leading directional behavior without erasing differences in lower-order terms, solution spaces, or theorems.

For a real symmetric measurable matrix field, assume positive constants lambda and Lambda bound the quadratic form below and above almost everywhere, for every real vector. This is the uniform ellipticity condition used throughout the module. It bounds the eigenvalues of A between those constants and makes the energy density comparable with the squared gradient. The ratio Lambda/lambda measures the permitted anisotropy, although estimates can also depend on coefficient regularity, domain geometry, and lower-order terms.

Pointwise positive definiteness only asks each location to have some positive smallest eigenvalue. Those location-dependent values may approach zero, leaving no common lambda. For example, the matrix diag(x squared,one) is positive definite for every x between zero and one, but its first-direction energy becomes arbitrarily small near zero. On a compact set, a continuous positive-definite matrix field has a uniform positive lower bound; on an open domain without control up to its boundary, that compactness argument is unavailable.

Uniform ellipticity does not by itself select a unique solution. A harmonic function can be changed by another harmonic function unless boundary conditions or a normalization remove that freedom. It also does not turn measurable coefficients into differentiable ones, so one cannot differentiate the equation and assume all resulting terms are classical functions. Existence, uniqueness, interior regularity, and boundary regularity each require an appropriate combination of data, domain, coefficient, and function-space assumptions.

The coefficient matrix can be interpreted as a directional conductivity in a simple diffusion model. That interpretation is useful for identifying which flux is continuous across an interface, but it does not justify treating every elliptic equation as a physical material model. In computations, test the smallest and largest eigenvalues across the represented coefficient family, and identify what happens between grid points. A finite list of positive eigenvalues is evidence about samples, whereas a uniform analytical bound controls the continuum model.