arXiv:2610.02183
Single-Particle Spectral Estimation
Adrian Chapman, Charles Derby, Steven T. Flammia, Yeongwoo Hwang, Joel Klassen, Calum McCartney
Abstract
We study Hamiltonian learning in a novel setting, where we are promised a physical structure that is obfuscated by a global unitary. In particular, we consider the task of learning the weights $E_i$ of a Hamiltonian $H=U^{\dagger}(\sum_{i \in I} E_i n_i + H_O)U$ comprised of a set of non-interacting modes $I$ and an arbitrary ``spectator'' $H_O$ acting on a disjoint set of modes $O$. The non-interacting structure is hidden by an unknown global unitary $U$, and our goal is to recover the $E_i$ without reconstructing $U$. Such Hamiltonians are well-motivated in the setting of materials modeling, where they find application for classical methods such as mean-field theory, Green's functions, and density functional theory. We show that this problem is DQC1-hard, despite the apparent simplification given by the non-interacting structure. We then introduce SPICES (single-particle inference via contour estimators), which combines a Hadamard test with a novel classical postprocessing procedure to efficiently recover the spectral distribution in Wasserstein distance under certain natural assumptions. The method exploits the correspondence between single-particle energies of the individual quasiparticles and the Fourier frequencies governing zeros of the partition function at complex inverse temperature. As quasiparticle spectra feature ubiquitously in downstream materials calculations, and with SPICES being simple and practical to execute, we expect that SPICES is a promising route to quantum applications in materials simulation.
AI Summary
The paper studies single-particle spectral estimation when a free sector is hidden by an unknown global unitary: $UHU^\dagger=\sum_{i\in I}E_i n_i+H_O$, with a corresponding transformed occupation observable $N$. The objective is to recover the thermally weighted distribution of protected energies in a window $\Omega$, without reconstructing $U$. SPICES estimates normalized complex grand-canonical partition functions using Hadamard tests on Gibbs states. Protected energies correspond to inverse-fugacity roots $e^{-(\beta+it)E_i}$. Logarithmic contour integrals isolate their first root moment; Fourier sampling and linear-programming reconstruction then produce an estimate within Wasserstein-1 error $\epsilon$, using evolution times $O(\epsilon^{-1})$. Polynomial complexity requires spectator roots to remain outside the target annulus, radial separation $q<1$ from its boundaries, a normalized signal floor $b$, efficient Gibbs preparation and controlled evolution, and polynomial bounds on $b^{-1}$, $(1-q)^{-1}$ and $e^{2\beta|\Omega|}$. For free systems, $\beta\Delta\ge\log(3M)$ guarantees signal magnitude at least $e^{-1}$. For spectator-only interactions of norm $|\lambda|$, sufficient stability conditions include $|\lambda|\sqrt{\beta^2+t_{\max}^2}<\log(4/3)$. A reduction establishes DQC1-hardness even for stable instances with one protected energy in $\{16,17\}$; the construction encodes difficulty in the occupation observable and its operational access. The structural promise requires exact decoupling of protected modes, excluding general quasiparticle interactions. The sufficient temperature and conditioning bounds restrict window widths relative to endpoint gaps. Interacting-spectator guarantees can require finite time horizons. Numerical validation and comparisons with Green’s-function methods are deferred.
AI was used to assist with paper drafting, editing and the design of figures and diagrams.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 70 pages, 3 figures