arXiv:2610.33800
Efficient Block Encodings of Vibrational Hamiltonians
Shashank G. Mehendale, Ignacio Loaiza, Shreyas Malpathak, Stepan Fomichev, Danial Motlagh, Artur F. Izmaylov
Abstract
Simulation of molecular vibrations is emerging as a promising application of quantum computing in quantum chemistry. In this work, we develop new strategies for block-encoding the second-quantized $n$-mode representation of the vibrational Hamiltonian, whose algebraic structure closely parallels that of the second-quantized electronic Hamiltonian. We derive linear-combination-of-unitaries (LCU) representations of this Hamiltonian based on Pauli decomposition and tensor hypercontraction (THC) and construct efficient circuits for the corresponding block encodings. We also extend the Block-Invariant Symmetry Shift (BLISS) technique to vibrational Hamiltonians, yielding a lower LCU 1-norm and, consequently, a lower overall circuit cost. For the molecules considered here, BLISS-THC reduces the Toffoli cost of quantum phase estimation by more than two orders of magnitude relative to existing block-encoding approaches. In comparison with recent Trotter-based methods, it shows up to a ninefold reduction in the T-gate count, with the relative savings tending to increase with system size. Asymptotically, and up to polylogarithmic factors, both the Pauli and THC LCUs trade additional ancillas for a quadratic improvement in scaling with the number of modes, while THC provides an additional quadratic improvement in scaling with the number of modals. These results highlight the importance of Hamiltonian factorizations and Toffoli-qubit tradeoffs for reducing the computational cost of quantum simulations of molecular vibrations.
AI Summary
The paper addresses the fault-tolerant cost of simulating molecular vibrations in a finite-modal, second-quantized $n$-mode representation. It constructs Pauli and tensor-hypercontraction (THC) linear combinations of unitaries, associated block-encoding circuits, and a vibrational extension of Block-Invariant Symmetry Shift (BLISS) that lowers the LCU normalization without changing physical-sector eigenvalues. Loading coefficients globally enables ancilla-assisted QROM tradeoffs rather than repeated one-mode block encodings. For $M$ modes, $N$ modals per mode, fixed coupling order $n\geq2$, and energy precision $\epsilon$, the QPE Toffoli costs are $\widetilde{\mathcal O}(\lambda_{\mathrm{Pauli}}M^{n/2}N^n/\epsilon)$ and $\widetilde{\mathcal O}(\lambda_{\mathrm{THC}}(MN)^{n/2}/\epsilon)$, respectively. These scalings require sufficient ancillas; THC additionally assumes average rank $\mathcal O(N)$. The constructions and circuit counts are supported by numerical tensor optimizations and logical-resource estimates. Seven two-mode-coupling polyacene Hamiltonians with 30–138 modes and five modals yield 111–281-fold Toffoli reductions for BLISS-THC relative to the prior Triangular LCU, with 13–33% more logical qubits. Two three-mode-coupling benchmarks give 95–178-fold reductions with approximately 27–32% more qubits. Comparisons against a recent Trotter method on small molecules show approximately 2–9-fold fewer T gates but 5–7-fold more qubits. The resource estimates assume access to an exact eigenstate. Tensor-approximation tolerances use a $1\,\mathrm{cm}^{-1}$ Frobenius-error threshold, supported by small-system ground-state tests rather than certified large-system energy errors. THC ranks are empirical, and optimization is nonlinear. Broader THC-compatible BLISS shifts proved difficult, so BLISS-THC uses only one-mode shifts. Comparisons with real-space block encodings and benchmarks on classically challenging target systems remain open.
Error found: The abstract omits that THC's claimed asymptotic improvement with modal count assumes an average THC rank proportional to the number of modals, an assumption not established by the small-modal benchmarks. The large-system resource savings are not certified comparisons at equal total energy accuracy because tensor approximation errors are controlled through a heuristic Frobenius-error proxy, separately from QPE error.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 42 pages, 10 figures, 4 tables