Quantum simulation of exact electron dynamics can be more efficient than classical mean-field methods

The paper revisits the premise that quantum computers offer advantages for simulating dynamics by enabling exact, systematically improvable simulations without exponential scaling. It advances a complementary idea: certain exact quantum algorithms for real-time electron dynamics can be more efficient than classical methods that make uncontrolled approximations, such as real-time time-dependent Hartree-Fock (RT-TDHF) and time-dependent density functional theory (RT-TDDFT). The authors define exactness as producing a time-evolved state within 1-norm error ε of full configuration interaction dynamics in a chosen basis, with refinement cost O(log(1/ε)). They focus on interacting fermions relevant to chemistry, physics, and materials science, arguing that even when mean-field approximations describe systems well, quantum algorithms can still provide speedups. The work targets comparisons against mean-field approaches due to their popularity, clear formulations, and typical use at modest system sizes, while noting that assumptions like nearsightedness may fail in dynamics of highly excited states and at high temperatures.

The study situates itself within a line of research starting from Feynman’s proposal of quantum computers for simulating dynamics and Lloyd’s universality results. It reviews classical mean-field methods (RT-TDHF, RT-TDDFT), their popularity and limitations (e.g., adiabatic approximations, self-interaction error), and emerging classical ideas like quantized tensor train compression whose efficacy for dynamics and finite temperature is unclear. On the quantum side, it discusses first-quantized algorithms for dynamics that exploit compressed representations (O(n log N) space) and early Trotterized real/reciprocal-space switching methods via quantum Fourier transforms. It highlights recent bounds on Trotter errors and interaction-picture simulation enabling sublinear-in-basis-size gate complexities. It contrasts first- and second-quantized approaches, noting regimes where second quantization can have lower gate complexity but higher qubit counts. It also reviews measurement strategies including classical shadows and multi-observable estimation, and classical algorithms such as fast multipole/Barnes–Hut/PME, noting challenges translating these to the quantum circuit model without assumptions like QRAM.

The paper develops and analyzes resource estimates for both classical mean-field dynamics and exact quantum dynamics in first quantization.

• Classical mean-field cost model: The dynamics are integrated with a step size Δt chosen for target precision ε over total simulation time t. Let T = ||F|| t where F is the effective mean-field operator; they upper bound ||F|| in a local basis with open boundary conditions by ||F|| = O(n^{7/3} δ^{-4} + n^{11/3} δ^{-5}), where δ = (n/N)^{1/3} is the minimum grid spacing given cell volume ∝ n. The first term arises from Coulomb interactions and the second from kinetic energy. For a k-th order integrator, per-step error scales as O((||F||Δt)^{k+1}), giving total error O(t ||F||^{k+1} Δt^{k}) and thus steps t/Δt = O(T^{1+1/k} ε^{-1/k}). Multiplying by the per-step update cost yields the classical operation count scaling (N^{7/3} ε^{-1} t + N^{11/3} ε^{-5/4} t) (ε/t)^{1/k}.
• Exact first-quantized quantum dynamics: The wavefunction is encoded across n registers (one per electron/occupied orbital), each of size log N, giving space O(n log N). The authors tighten bounds for Trotter-based first-quantized simulations using higher-order product formulas, achieving gate complexity scaling as (N^{1/3} η^{7/3} + N^{2/3} η^{4/3}) up to subpolynomial factors, improving prior Trotter analyses. They also review interaction-picture algorithms in plane-wave and grid bases that realize O(N^{1/3} η^{8/3}) gate complexity by implementing efficient block encodings whose preparation only scales polylogarithmically in N; the N^{1/3} dependence arises from the potential operator norm. They discuss that softening the Coulomb potential while preserving target precision could reduce the potential norm to polylog(N), enabling exponential advantages in N.
• Comparison with second-quantized approaches: Trotter-based second-quantized methods can achieve gate complexity scaling as N^{5/3} (when η = O(N)), outperforming first-quantized Trotter in some regimes (N < Θ(η^{3})) but requiring at least O(N) qubits (and O(N log N) for fast multipole-based implementations), which can be prohibitive.
• Measurement (quantum sampling) protocols: They introduce a first-quantized classical shadows scheme: apply independent random Clifford channels to each of the η registers (each on log N qubits), costing O(η log N) gates per repetition. Repeating O(η/ε^{2}) times suffices to estimate all 1-RDM elements to additive error ε, and O(k η/ε^{2}) for all elements of k-RDMs. The argument exploits antisymmetry-induced anticorrelations across registers, allowing parallelization across all n registers. For observables with norms growing with system size (e.g., energy), they propose direct estimation using block encodings and amplitude amplification, yielding O(Λ/ε) repetitions, where Λ is the 1-norm of the block encoding (for energy: Λ = O(N^{1/3} η^{5/3} + N^{2/3} η^{1/3})). They also discuss costs of sampling at multiple time points L, typically scaling as O(L) or O(L t) with time-resolved algorithms, with an added O(L) space overhead for certain improved schemes.
• State preparation: They present a new first-quantized Slater determinant preparation algorithm with O(N n) gates. It generates a superposition over ordered configurations of occupied orbitals and then applies an antisymmetrization procedure requiring O(n log n log N) gates. The construction maps a second-quantized Givens-rotation-based preparation (which uses O(N n) rotations) into first quantization on-the-fly, storing only O(n) nonzero qubits at any time and converting stabilized qubits to first-quantized registers. The Toffoli count can be reduced to O(η N) with further optimizations.
• Finite-temperature simulations: They suggest initializing circuits with Slater determinants sampled from a thermal Hartree-Fock distribution, which for quantum processors does not increase repetitions beyond those needed for sampling observables, whereas classical trajectory sampling would incur an additional O(1/ε^{2}) multiplicative cost.

• Exact first-quantized algorithms can simulate electron dynamics with exponentially less space than classical mean-field methods (O(n log N) qubits vs O(N^{n} log(1/ε)) classical space) and polynomially fewer operations in N.
• Tightened Trotter bounds for first-quantized simulations yield gate complexity scaling as (N^{1/3} η^{7/3} + N^{2/3} η^{4/3}) up to subpolynomial factors; interaction-picture algorithms achieve O(N^{1/3} η^{8/3}). When N > O(η^{3}), interaction-picture methods are superior and can provide quintic speedups in N over classical RT mean-field scaling while incurring a quadratic slowdown in η.
• Second-quantized Trotter algorithms can achieve N^{5/3} gate complexity (for η = O(N)), outperforming first-quantized Trotter in some regimes but at the cost of O(N)–O(N log N) qubits, limiting practical applicability.
• Classical mean-field time-evolution complexity, under high-order integrators and bounded ||F||, scales as (N^{7/3} ε^{-1} t + N^{11/3} ε^{-5/4} t) (ε/t)^{1/k} operations, where k is the integrator order and T = ||F|| t with ||F|| = O(n^{7/3} δ^{-4} + n^{11/3} δ^{-5}), δ = (n/N)^{1/3}.
• New classical shadows measurement protocol: all 1-RDM elements can be estimated to additive error ε with O(η/ε^{2}) circuit repetitions; all k-RDM elements require O(k η/ε^{2}) repetitions. Each repetition uses O(η log N) gates for random Clifford channels across registers. This reduces sample complexity from naive O(N^{2}/ε) scaling and maintains speedups in N.
• New first-quantized Slater determinant preparation algorithm uses O(N n) gates and O(n) ancilla qubits, with antisymmetrization costing O(n log n log N). Toffoli complexity can be O(η N).
• Overall speedups: For zero-temperature dynamics, quantum algorithms deliver up to a seventh-power speedup in particle number when N < O(η^{2}), quartic in N when Θ(η^{2}) < N < Θ(η^{4}), super-quadratic in N when Θ(η) < N < Θ(η^{4}), and quintic in N with a quadratic slowdown in η when N > O(η^{4}). Speedups can remain super-quadratic in extremal regimes (N < O(η), N > Θ(η^{4})). Speedup is more pronounced for finite-temperature simulations where classical costs grow with the number of thermally occupied orbitals.

The work demonstrates that exact quantum simulation of interacting electrons can outperform widely used classical mean-field approximations in both space and time complexities across relevant regimes of system size N and particle number η. By tightening Trotter bounds and leveraging interaction-picture techniques, the authors close the gap between formal exactness and practical efficiency, showing polynomial improvements over classical RT-TDHF/RT-TDDFT. New measurement and state-preparation procedures address two major practical bottlenecks: efficient extraction of reduced density matrices via classical shadows at O(η/ε^{2}) sample cost, and O(N n) gate complexity for preparing arbitrary Slater determinants in first quantization. These developments preserve or expand quantum speedups even when accounting for the need to sample observables and measure at multiple time points. Nevertheless, measurement costs can erode speedups for extensive observables (e.g., total energy) or when many time samples are required, making the realizable advantage sensitive to the choice of observables and precision. The authors highlight applications where mean-field approximations are qualitatively valid yet strained—particularly finite-temperature electron dynamics in warm/hot dense matter and strongly excited non-equilibrium regimes—where quantum algorithms’ advantages are likely to be most impactful.

This paper tightens complexity bounds for first-quantized, Trotter-based exact quantum simulations of electron dynamics, refines interaction-picture resource estimates, and introduces improved protocols for first-quantized Slater determinant preparation and classical-shadows-based RDM measurement. Collectively, these advances show that exact quantum dynamics can be asymptotically more efficient than classical mean-field methods, with exponential space savings and polynomial gate-count improvements in N, and with pronounced advantages at finite temperature. Future work could refine constant factors in Trotter step counts and block encodings, extend efficient measurement to additional observables with growing norms, explore softened Coulomb potentials enabling polylog(N) dependence, integrate nuclear dynamics, and tailor analyses to restricted input classes to obtain tighter practical bounds. Application-focused studies in warm/hot dense matter and ultrafast, strongly excited systems may provide early demonstrations of quantum advantage.

• Measurement overhead: Sampling costs (O(η/ε^{2}) for RDMs; O(Λ/ε) for direct observable estimation) reduce speedups, especially for observables with norms growing with N, η (e.g., total energy) and when sampling L time points (additional O(L)–O(L t) scaling and O(L) space in some methods).
• Regime dependence: Second-quantized methods can have better gate complexity (N^{5/3}) for N < Θ(η^{3}) but require O(N)–O(N log N) qubits, potentially impractical on near-term hardware.
• Assumptions and bounds: Complexity analyses rely on worst-case upper bounds for ||F|| and Trotter errors; constant factors for required Trotter numbers in the presented representations are not fully characterized. Reported speedups may vary with basis choice, pseudization, and implementation details.
• Excluded classical methods: Linear-scaling methods and quantized tensor train compressions are not benchmarked here due to unclear efficacy for dynamics and finite temperatures; comparisons focus on conventional RT-TDHF/RT-TDDFT.
• Algorithmic prerequisites: Certain enhanced speedups (e.g., via softened Coulomb potential) assume modifications maintaining target precision; feasibility and physical fidelity require further validation. Classical fast-multipole-type accelerations do not straightforwardly translate to the quantum circuit model without strong assumptions (e.g., QRAM).
• Resource trade-offs: While first-quantized space is efficient, implementing grid-like bases and block encodings may impose architectural constraints; multiple-register Clifford randomization adds classical postprocessing effort for channel inversion.