arXiv:2610.21907
Selected Stabilizer Subspaces for Compact Many-Body Wave Functions
Maria-Andreea Filip, George H. Booth
Abstract
Selected configuration interaction (SCI) obtains accurate many-body solutions by variationally diagonalising the Hamiltonian in a sparse, iteratively selected set of configurations, but can lose compactness when the target state has broad configurational support. We introduce selected stabilizer interaction (SSI), a fully classical, SCI-inspired method that instead iteratively constructs a nonorthogonal subspace spanned by a systematically enlargeable set of selected stabilizer states. A single stabilizer, generated by Clifford operations, can encode a structured superposition of (possibly exponentially) many configurations, shifting the relevant compression from configurational to stabilizer sparsity. We benchmark ground-state energies, fidelities, superposition degrees and required expansion lengths against exact diagonalization and equivalent SCI schemes for the transverse-field Ising, XXZ Heisenberg, and Hubbard models, finding particularly strong compression from the stabilizer representation in the spin models, and increasing advantage with system size for these systems. The results establish a proof of principle for heuristic selected stabilizer diagonalization, while also identifying particular challenges in symmetry handling and effective exploration of the stabilizer space.
AI Summary
Can selected-subspace diagonalization become more compact by replacing computational-basis configurations with stabilizer states? This paper introduces selected stabilizer interaction (SSI), a fully classical variational method that builds a nonorthogonal stabilizer expansion. Candidates are generated by applying sampled pairs of Hamiltonian Pauli terms to existing stabilizers, ranked by their overlap with the Ritz residual, and added one at a time before solving a generalized eigenvalue problem. Stabilizer overlaps and Pauli-Hamiltonian matrix elements are evaluated without expanding their configurational support. Benchmarks against exact diagonalization and a matched configuration-selection heuristic cover one-dimensional transverse-field Ising, XXZ, and Hubbard chains. SSI uses fewer retained states in the spin models, with increasing compression over the tested sizes. For an eight-qubit Ising chain at $|h/J|=1$, selected stabilizers approach an average support of roughly $64$ configurations. Size-dependent Ising tests use simultaneous relative-energy and overlap-error thresholds of $0.01$. XXZ calculations enforce fixed magnetization through model-specific candidate filters. Hubbard compression is more limited and sensitive to candidate generation. Greedy grouping of exact-state amplitudes gives exact stabilizer-rank upper bounds of $2$, $10$, and $104$ for four-, eight-, and twelve-qubit chains, respectively; the corresponding symmetry-sector dimensions are $4$, $36$, and $400$. Adding single-Pauli candidates improves SSI, but its expansions still exceed these bounds at high accuracy. The study measures expansion compactness rather than time-to-solution; larger-system scaling, effective symmetry handling, seed selection, and broader fermionic applications remain unresolved.
Error found: In the three-qubit TFI motivation, X_0 Z_1 X_2 is incorrectly called a generator of spatial reflection and is not a symmetry of the generic Hamiltonian; the correct reflection operator exchanges qubits 0 and 2.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 20 pages, 8 figures, 1 table