arXiv:2610.04315
From Ground State to Electronic and Vibronic Spectra via Memory Kernel Coupling Theory and Neural Quantum States
Xinchen He, Wenjie Dou
Abstract
Modelling molecular absorption spectra from scalable ground-state representations is challenging because conventional approaches often require explicit excited states or time propagation. Here, we introduce a method to obtain Absorption Using Recursive Operator Response with vibronic Augmentation (Aurora), which combines neural quantum states (NQSs) with continued-fraction memory-kernel coupling theory (CF-MKCT). Aurora maps the zero-temperature Liouvillian response hierarchy onto repeated wavefunction-space applications of a shifted Hamiltonian to a dipole-prepared state, enabling direct spectral reconstruction from variational ground states. The implementation combines variational Monte Carlo, implicit-restarted-Lanczos-assisted NQS optimization, matrix-free Hamiltonian application, and adaptive recursion control. The resulting electronic spectra closely agree with full-configuration-interaction and conventional MKCT references for tractable molecules, with the restarted-Lanczos strategy improving difficult NQS cases. For electron-phonon Hamiltonians, Aurora combines Huang-Rhys mode selection with local singular-value-decomposition (SVD) phonon encoders. The H2 benchmark indicates that compression retains the dominant vibronic envelope and improves upon bare local-basis truncation, while paired calculations for N2, H2O, H2S, and NH3 yield similar principal features with and without encoding. We further obtain vibronic spectra for C6H6 and a 10-electron, 10-orbital C10H8 active space. Overall, Aurora provides a unified route from variational molecular ground states to electronic and vibronic spectra without constructing excited-state manifolds.
AI Summary
The paper addresses how to reconstruct zero-temperature molecular absorption spectra from variational ground states without explicitly computing excited-state manifolds or propagating in real time. Aurora combines neural quantum states (NQSs) with continued-fraction memory-kernel coupling theory. For an exact ground state, it maps Liouvillian actions onto applications of $H-E_0$ to a dipole-prepared wavefunction, then implements the response hierarchy through Lanczos tridiagonalization and continued-fraction reconstruction. NQS optimization combines variational Monte Carlo with periodic sampled Ritz-target corrections. Electronic benchmarks, primarily in STO-3G frozen-core spaces, compare spectra against full configuration interaction and conventional memory-kernel calculations. Reported spectral convergence at tolerance $10^{-3}$ requires 50–110 recursions across five small molecules. A naphthalene electronic calculation requires 690 recursions at $10^{-6}$ tolerance; the selected spectral-generation step takes 3.93 seconds, excluding ground-state preparation and energy evaluation. The vibronic extension uses harmonic modes with linear electron–phonon coupling, Huang–Rhys mode selection, and ground-state Schmidt/SVD compression of local phonon spaces from eight states to four. In an $\mathrm{H}_2$ benchmark, compression reduces the dominant-envelope peak shift relative to the eight-level reference from $-2.88$ eV for bare four-level truncation to $-0.337$ eV. Additional calculations cover small molecules, benzene, and a 10-electron, 10-orbital naphthalene active space. The model omits anharmonicity and higher-order coupling. Compression loses resolved sidebands; larger-system comparisons lack exact references, and independently trained encoded/nonencoded spectra conflate representation and optimization effects. Resource limits require stricter mode selection and looser spectral tolerances for the aromatic vibronic examples.
Error found: The exact Liouvillian-to-wavefunction correspondence acquires an omitted residual term for approximate NQS ground states, and its spectral effect is not quantified. The implementation explicitly enumerates the restricted electron–phonon basis and supplies no scalable compressed response-vector representation, limiting the claimed scalability. The sampled-IRL projected Hamiltonian replaces genuine Hamiltonian action by local-energy multiplication, so its eigenpairs are not generally Hamiltonian Ritz pairs. The stated moment-equivalence formula has the wrong sign for odd moments under the paper’s own inner-product convention.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.