Isospectral manifolds and inverse spectral limits

Two compact objects are isospectral when their entire ordered eigenvalue lists agree with multiplicity for the chosen self-adjoint realizations. Their heat traces then agree at every positive time, so their corresponding heat coefficients agree. The leading Weyl growth fixes dimension and volume. In the smooth closed scalar case, the next heat coefficient fixes integrated scalar curvature. The zero-mode multiplicity fixes the number of connected components, but these few invariants do not constitute a complete description of the metric.

Boundary data must be matched before comparing coefficients: a Dirichlet list is not interchangeable with a Neumann list. Smooth-boundary area terms can be audible, while polygonal corners alter later coefficients. In dimension two, additional geometric assumptions can turn integrated curvature into topological information through Gauss–Bonnet. One should not extrapolate that special relation into a claim that all topology is determined in every dimension. The spectrum constrains geometry strongly without automatically identifying every local or global feature.

Let a finite group act freely by isometries on a closed Riemannian manifold. Two subgroups are almost conjugate when they meet every conjugacy class in the same number of elements. Sunada's criterion gives isospectral quotient manifolds for such subgroups. Freeness ensures that the quotients in this statement are manifolds rather than orbifolds. The criterion alone does not guarantee they are nonisometric; that separate geometric distinction must be established in a claimed counterexample.

The reason is representation-theoretic. Each finite-dimensional Laplace eigenspace carries a group representation, and quotient eigenfunctions correspond to subgroup-invariant vectors. The invariant dimension is the average of the representation character over the subgroup. Characters are constant on conjugacy classes, so almost-conjugacy gives equal averages and hence equal multiplicities at every eigenvalue. This is an exact all-level argument, not an agreement observed by matching the first few numerically computed frequencies.

Gordon, Webb, and Wolpert constructed noncongruent simply connected planar polygonal domains with identical Dirichlet spectra, answering the general drum-shape question negatively. Their polygonal boundaries should not be described as smooth boundaries when invoking smooth heat coefficients. Closed-manifold isospectral examples also exist. Such constructions disprove universal geometric recovery, but they do not imply that every restricted family lacks rigidity; a carefully specified class may admit positive inverse results.

Finite numerical agreement has a different logical status. Two objects can share a leading asymptotic coefficient or several low modes while differing later. Conversely, numerical discrepancies can be caused by discretization or mismatched normalizations before any geometric conclusion is justified. A finite matrix model can illustrate spectral ambiguity but is not itself a pair of Riemannian counterexamples. The proof of a geometric example must connect its exact operator spectra and separately show that the underlying shapes are not related by an isometry.