Particle-Number-Preserving Quantum Signal Processing for Thermal Properties: Canonical-Ensemble Estimation and Grand-Canonical Reconstruction

AI summary

Overview

Research area: Quantum algorithms for quantum many-body simulation, specifically estimating thermal (finite-temperature) properties of interacting fermionic systems on quantum computers.

Technical level: Advanced. The paper assumes familiarity with second quantization, the Jordan-Wigner transformation, Trotterization, quantum signal processing, block encodings, and asymptotic complexity analysis.

Scope: The paper presents an end-to-end quantum algorithm for canonical-ensemble thermal properties within a fixed particle-number sector, plus a classical-reweighting scheme that converts that data into grand-canonical results for arbitrary chemical potentials, with complexity comparisons against classical Lanczos methods.

What This Paper Is About

Many important quantum systems, including molecules, ultracold atoms, and lattice electronic materials, conserve particle number, so their thermal properties are naturally computed inside a fixed-particle-number sector (the canonical ensemble). Classical methods such as Lanczos-based approaches have rigorous guarantees but cost time that grows with the dimension of that sector, which is exponential in system size, and the authors argue that a rigorous quantum framework for this fixed-sector problem had not yet been established. The paper builds such a framework from start to finish and then shows how the same measured data can be reweighted classically to answer grand-canonical questions at any chemical potential without rerunning the quantum circuits.

Key Contributions

1. A complete canonical-ensemble quantum algorithm (Algorithm 1). It combines three ingredients that had previously been developed separately: initialization in a "mixed Dicke state" (the maximally mixed state of a fixed-particle sector), Trotter circuits built so that every block commutes with the particle-number operator, and a generalized quantum signal processing (GQSP) circuit with one ancilla qubit that approximately block-encodes the exponential of the Hamiltonian. The probability of measuring the ancilla in |0⟩ is proportional to the partition function.

2. Particle-number-preserving Trotter construction (Example 1). The authors extend the unitary coupled cluster ansatz used in variational quantum algorithms to both symmetric and antisymmetric forms, giving explicit blocks for one-body terms, for four-distinct-index two-body terms, for the one-coincidence case, and for the two-coincidence case. Any finite product of these blocks commutes with the number operator N, so the effective Trotterized Hamiltonian H̃ satisfies [H̃, N] = 0 and Trotter error stays inside the target sector rather than leaking into others.

3. A grand-canonical reconstruction scheme (Algorithm 2). Canonical partition functions and observables are estimated sector by sector, then combined by classical reweighting using 𝒵_L(β, μ) = Σ_s e^{β μ n_s} Z_{L,s}(β) and the corresponding weighted formula for observables. The quantum circuit and circuit-sampling complexities are independent of the chemical potential μ, so a scan over many trial values of μ costs no additional circuit executions.

4. A complexity separation from classical Lanczos methods (Observation 1). For physically relevant Hamiltonians whose spectral norm scales linearly with system size, the authors prove that system-size-independent temperature thresholds exist above which both the canonical and the reconstructed grand-canonical schemes have lower asymptotic time complexity than Lanczos-based classical methods, up to polynomial factors.

Main Findings

• Resource counts (Theorem 1). Initial-state preparation needs 2L qubits for the total-particle sector s = k, or 4L qubits for the spin-resolved sector s = (k, k′). The GQSP stage uses L system qubits (s = k) or 2L system qubits (s = (k, k′)) plus a single ancilla qubit.
• Trotter step count. The number of Trotter steps per Hamiltonian simulation is r = Ω(Λ_{H(L,s)}² max[1/(λ_{L,s} δ), β/log(1+ε)]), where λ_{L,s} is a normalization factor with ‖H(L,s)/λ_{L,s}‖₂ ≤ 1 − δ.
• Gate and sampling complexity. Gate complexity is O(r L⁴ (ln(1/ε) + β λ_{L,s}) / δ). Circuit-sampling complexity is O(C d_s e^{β λ_{L,s}} / (Z_{L,s}(β) ε²) · log(1/η)), where d_s is the sector dimension, equal to binom(L,k) for s = k and binom(L,k)·binom(L,k′) for s = (k, k′), and C = 1 for the partition function or (Σ_a |c_a|)² for an observable.
• No chemical-potential dependence in the canonical bounds. Canonical-ensemble gate and circuit-sampling complexity bounds are independent of μ.
• Exponential improvement over a textbook Trotter analysis. The authors state that a standard textbook approach would require 𝒪((εP(0))^{-1/p}) Trotter steps because of the relative-error convention, and that P(0) scales exponentially with β and λ_{L,s}; their step count in (12) is independent of P(0).
• Sector-restricted normalization. The Hamiltonian simulation is normalized by a factor corresponding to the fixed-particle subspace rather than the full Hilbert space, which the authors note is consistent with the block-encoding case and generally smaller than the full-space value.
• Initial-state preparation cost (Proposition 1). Preparing the mixed Dicke state ω_{L,k} takes 2L qubits, 𝒪(kL) gates and 𝒪(L) circuit depth, by preparing a purified Dicke state and tracing out the ancilla register; ω_{L,(k,k′)} costs 4L qubits with the same gate and depth scaling. Pure Dicke state preparation requires only linear circuit depth.
• Numerical validation. The theoretical results are checked numerically for the two-dimensional Hubbard model in both weak- and intermediate-coupling regimes.
• Against the sampling-tree classical algorithm. The authors note that the recent classical sampling-tree algorithm has rigorous temperature thresholds that, for the 2D Hubbard model they study, lie far above physically relevant regimes according to their Table 2, which is why Lanczos is used as the main classical baseline.
• Comparison with prior quantum approaches. Fixed-sector initialization also appears in the LCU-based method of [25], but that construction lacks number-preserving Trotter circuits and can incur at least quadratically higher circuit-sampling complexity, a common drawback of incoherent LCU methods relative to coherent QSP implementations. Lindbladian algorithms in [15, 17] do not preserve particle-number symmetry, and the number-preserving construction in [4] is restricted to ground-state properties of Hartree-Fock models.

Methodology in Plain English

The strategy is to turn a thermal-average question into a probability of seeing a particular bit.

First, rather than starting from a single definite electronic configuration, the computer is placed in an equal superposition of all configurations with exactly the right number of particles, with the extra register used to purify the state then discarded. This is the mixed Dicke state, and it is the maximally mixed state of the chosen sector.

Second, the exponential e^{-βH/2} is implemented not directly but through quantum signal processing, which interleaves single-qubit rotations with controlled short-time evolutions of the Hamiltonian. Those short-time evolutions are built from Trotter blocks that have been deliberately grouped so that each block conserves particle number; the authors work out explicit forms for the one- and two-body terms after the Jordan-Wigner transformation. Because the blocks conserve particle number, the simulated Hamiltonian never pushes amplitude into sectors that were not asked about.

Third, the observable consequence is simple: if the ancilla qubit is measured and found in |0⟩, the chance of that happening is proportional to the partition function, and the system-register measurement recorded on those same shots gives a sample of the thermal expectation value. Running the circuit many times and averaging yields both quantities. Comparing with a naive Trotter analysis, the authors show their step count avoids an exponential dependence on the success probability.

Fourth, for the grand-canonical ensemble, the authors exploit the fact that the grand partition function is just a chemical-potential-weighted sum of canonical partition functions over sectors. So they estimate every sector once, then do the weighted sum on a classical computer. Changing μ is a classical post-processing step, not a new experiment.

Finally, they compare the resulting cost scaling against the classical Lanczos baseline, which costs time proportional to the sector dimension regardless of temperature. Since their quantum cost grows with inverse temperature while the classical cost does not, there is necessarily a temperature threshold above which the quantum route wins, and they identify regimes where that threshold does not grow with system size.

Why This Matters

Impact on research. The paper fills a gap the authors identify explicitly: rigorous quantum algorithms existed for grand-canonical thermal properties and for ground states, but not an end-to-end construction for the canonical ensemble inside a fixed particle-number sector. It also supplies a practical separation result against the strongest general-purpose classical baseline, and it shows that the chemical potential, usually a complicating parameter, can be removed from the quantum cost entirely.

Real-world applications (as identified in the paper):

• Molecular systems, where fixed electron number is the natural setting for thermal property estimation.
• Ultracold atoms, where particle number is controlled and conserved.
• Lattice electronic systems such as twisted bilayer graphene.
• Chemical-potential scanning in practice, where practitioners need to determine an appropriate μ through an extensive sweep and would otherwise have to rerun quantum circuits for every trial value.

Industry relevance. Finite-temperature property estimation for correlated fermionic materials is directly relevant to quantum chemistry and materials discovery workflows, where phase diagrams, occupation numbers, and thermal averages are the deliverable rather than a single ground-state energy. The reweighting trick lowers the experimental cost of the parameter sweeps that such workflows require.

Future Directions

• Determining how the approach performs beyond the weak- and intermediate-coupling 2D Hubbard regimes tested, since the numerical validation reported covers those two regimes only.
• Establishing numerically where the sampling-tree classical algorithm's theoretical guarantees become practically relevant; the authors state this remains unclear and that its rigorous thresholds sit far above physically relevant temperatures for the Hubbard model they study.
• Tightening the normalization factor λ_{L,s}, which the authors obtain conservatively via Gershgorin's theorem and suggest could be improved with semidefinite programming relaxations.
• Reducing the polynomial overheads in the gate complexity, which carries an L⁴ factor and a 1/δ factor, and extending the treatment to models with different symmetry structure (for example, sectors defined by symmetries other than total particle number and spin).

Target Audience

Quantum algorithm researchers working on quantum simulation of fermionic many-body systems and finite-temperature algorithms; quantum information theorists interested in QSP, block encodings, and Trotter error analysis; and computationally oriented condensed-matter and quantum chemistry researchers who need thermal properties of particle-number-conserving models and want to understand when a quantum computer offers an asymptotic advantage over Lanczos-style classical diagonalization.

Authors’ abstract

We develop an end-to-end framework for estimating canonical thermal properties of particle-number-conserving second-quantized Hamiltonians using quantum simulations. The framework integrates initial state preparation with mixed Dicke states, particle-number-preserving Trotterized simulation, and generalized quantum signal processing to estimate canonical partition functions and observables. Building on this canonical scheme, we introduce a grand-canonical reconstruction scheme in which the same quantum measurement data can be reused for arbitrary chemical potentials through classical reweighting, thereby avoiding repeated quantum circuit executions as the chemical potential is varied. We prove the existence of system-size-independent temperature regimes in which both the canonical and reconstructed grand-canonical schemes have lower asymptotic time complexity than classical Lanczos methods, up to polynomial factors, for physically relevant Hamiltonians. We numerically validate our theoretical results for the two-dimensional Hubbard model in both weak- and intermediate-coupling regimes.