Semiclassical trace formulas and periodic orbits

Take a self-adjoint elliptic confining Schrödinger operator with smooth potential, for example one equal to a positive quadratic outside a compact set. For a fixed smooth compactly supported spectral test f, the relevant classical energy region is compact and f(P_h) is trace class. Its leading trace is the phase-space integral below. The h-uniform functional calculus and symbol hypotheses justify the remainder; the expression is not an unrestricted formula for the trace of the identity, which is infinite on this Hilbert space.

The normalization (2πh)^(−d) is consistent with the Weyl kernel and represents one quantum state per phase-space cell at leading order. Smooth spectral testing averages many levels and can hide finer oscillations. Recovering periodic-orbit information uses an oscillatory time representation, often with a spectral window of width comparable to h. A general formal density of delta masses is a distribution, so its expansion must be interpreted against suitable test functions rather than evaluated as an ordinary function at every energy.

At a regular energy with compact energy surface, localize the time test near a nonzero period and away from unrelated periods and zero-time contributions. For an isolated periodic orbit, nondegeneracy means its Poincaré return map on a transverse section has no eigenvalue one. The unavoidable flow direction is removed before taking the determinant. Stationary phase then produces a leading geometric factor involving the primitive period divided by sqrt of the absolute return-map determinant, together with an action and Maslov phase and the chosen test weights.

The displayed factor is one part of an orbit contribution, not a complete normalization of an unsmoothed density formula. Overall powers of h, Fourier-test factors and subprincipal transport must be supplied by a fully specified trace theorem. We use it only under an isolated nondegenerate orbit hypothesis, with a finite time localization where the contributing orbits can be controlled. Repetitions require their own return maps and indices. A zero determinant cannot be repaired by pretending its reciprocal is a large but finite coefficient.

Continuous orbit families require a clean or otherwise adapted trace analysis with different powers and coefficients. Symmetries and the isotropic oscillator in dimension at least two provide familiar sources of such families. Area preservation of a two-dimensional return map only fixes its determinant, not whether one is an eigenvalue. Even a symplectic shear can therefore fail the isolated-orbit nondegeneracy test. The lab checks that distinction without claiming its matrices arise from a constructed closed trajectory.

An exact trace-class oscillator sum gives an independent normalization check. Its exponentially decaying spectral test is not compactly supported, so it is computed directly rather than inserted into the compact-test theorem without justification. Comparing the exact sum with its convergent Gaussian phase-space integral verifies the leading cell factor and reveals a correction. This exact calculation does not imply that an untested infinite sum of orbit terms converges, or that a general chaotic spectrum can be recovered one eigenvalue at a time from the leading trace contribution.