Quantum simulation of slow analytic time-dependent Hamiltonians

Abstract: We develop a quantum algorithm for slow analytic Hamiltonians $\widetilde H(t)=H(t/T)$ with $|H(s)|\leqα$ that achieves nearly additive query complexity and low gate overhead. Our main technical contribution is a periodic Gevrey extension of $H(s)$, together with Fourier component decay and truncation bounds that enable an efficient finite-dimensional simulation. Combined with Floquet embedding and optimal time-independent Hamiltonian simulation technique, this gives query complexity $$\widetilde{\mathcal O}!\left(αT+\log(1/\varepsilon)\right)$$ and additional gate complexity $$\widetilde{\mathcal O}!\left((αT+\log(1/\varepsilon))<sup>2\log(1/\varepsilon)\right)$$, assuming coherent access to $H&#39;(s)$ and endpoint derivatives. For slow analytic control Hamiltonians, only block encodings of the time-independent control operators are required, with the same query complexity and lower gate overhead. Our method also extends to Gevrey Hamiltonians and improves the precision dependence for simulating slow analytic semi-dissipative linear differential equations.

• The paper introduces a Floquet–QSVT algorithm that simulates slow analytic Hamiltonians with nearly additive query complexity, \(\widetilde{\mathcal O}(\alpha T+\log(1/\varepsilon))\), and polylogarithmic dependence on precision in the gate overhead.
• The method constructs a periodic Gevrey extension, exploits subexponential Fourier decay, and uses endpoint derivative access to remove periodicity assumptions while preserving efficient truncation and amplitude amplification.
• For structured analytic controls, the gate overhead improves to \(\widetilde{\mathcal O}((\alpha T+\log(1/\varepsilon))^2)\), while Gevrey-class and linear-differential-equation extensions provide broader applicability with predictable scaling trade-offs.

Problem and motivation

The paper addresses time-dependent Hamiltonian simulation: given $\widetilde H(t)$, prepare a state within $2$-norm error $\varepsilon$ of the solution of the Schrödinger equation at time $T$. For time-independent Hamiltonians, QSVT achieves the optimal query complexity $\mathcal O(\alpha T + \log(1/\varepsilon))$, where $\alpha$ is the block-encoding normalization factor. For general time-dependent Hamiltonians, the truncated Dyson series method achieves only a multiplicative dependence $$\widetilde{\mathcal O}(\alpha T \log(1/\varepsilon))$$. Floquet-based constructions achieve additive scaling for periodic and multiperiodic inputs with exponentially decaying Fourier modes, and a concurrent transducer-based work achieves optimal query complexity for general Lipschitz-continuous Hamiltonians in the HAM-T model, but its direct circuit implementation incurs additional gate complexity polynomial in $1/\varepsilon$. The central question posed by the paper is whether nearly additive query complexity can be combined with additional gate complexity polylogarithmic in $1/\varepsilon$.

The paper answers this affirmatively for slow analytic Hamiltonians of the form $\widetilde H(t) = H(t/T)$, i.e., a fixed analytic control path traversed over physical time $2$0, with $2$1 on $2$2. Analyticity is quantified via the Gevrey bound $2$3; more generally the method covers Gevrey classes $2$4 with $2$5 for $2$6.

Main results

For general slow analytic Hamiltonians, assuming coherent access to block encodings of $2$7 (a HAM'-T oracle at Clenshaw–Curtis nodes) and of high-order endpoint derivatives $2$8, $2$9 up to order $\varepsilon$0, the algorithm achieves query complexity $\varepsilon$1 and additional gate complexity $\varepsilon$2. This is the first construction of its kind to combine nearly additive query scaling with polylogarithmic precision dependence in the gate count, although the authors are explicit that the derivative-access model differs from standard HAM-T access to $\varepsilon$3 itself, so the result should not be read as an improvement under an identical input model.

For structured control Hamiltonians $\varepsilon$4 with analytic scalar coefficients and time-independent Hermitian operators $\varepsilon$5, the Fourier coefficients can be computed classically, so only controlled block encodings of the $\varepsilon$6 are queried. In that case the query complexity is unchanged and the additional gate overhead drops to $\varepsilon$7.

Two extensions broaden the scope of these bounds:

• Gevrey regularity: for $\varepsilon$8 with $\varepsilon$9, the query complexity degrades to $T$0, reflecting subexponential rather than exponential Fourier decay.
• Slow linear differential equations: combining the simulation subroutine with the linear combination of Hamiltonian simulation (LCHS) representation, the normalized solution of $T$1 with slow analytic $T$2, $T$3, can be prepared using $T$4 queries. This quadratically improves the precision dependence over the previous best $T$5 scaling for generic time-dependent coefficients, matching the time-independent LCHS scaling up to polylogarithmic factors.

Technical approach

The algorithm proceeds in three stages.

Periodic Gevrey extension. A naive periodic continuation of $T$6 from $T$7 would be nonsmooth at the seam because endpoint derivatives need not agree, destroying rapid Fourier decay. The paper constructs a 2-periodic extension $T$8 using a Gevrey-$T$9 compactly supported cutoff function and a Gevrey version of Borel's lemma: on $\mathcal O(\alpha T + \log(1/\varepsilon))$0 the extension interpolates between the full Taylor data at $\mathcal O(\alpha T + \log(1/\varepsilon))$1 and $\mathcal O(\alpha T + \log(1/\varepsilon))$2 via series of the form $\mathcal O(\alpha T + \log(1/\varepsilon))$3 with radii $\mathcal O(\alpha T + \log(1/\varepsilon))$4. All derivatives match at the boundaries, and the extension lies in $\mathcal O(\alpha T + \log(1/\varepsilon))$5 with explicit derivative bounds. Since $\mathcal O(\alpha T + \log(1/\varepsilon))$6 is chosen as $\mathcal O(\alpha T + \log(1/\varepsilon))$7, the regularity loss from the extension is asymptotically negligible — a key design choice enabling the final near-additive scaling.

Fourier decay and Floquet truncation. Repeated integration by parts combined with Stirling's inequality yields subexponential decay $\mathcal O(\alpha T + \log(1/\varepsilon))$8 with $\mathcal O(\alpha T + \log(1/\varepsilon))$9. A Lieb–Robinson-type bound on transition amplitudes in the truncated Floquet space follows via a Dyson-series argument in the interaction picture, giving truncation error $\alpha$0. Consequently $\alpha$1, which after optimizing $\alpha$2 becomes $\alpha$3.

Floquet embedding and QSVT. The effective Hamiltonian on the enlarged Floquet–Hilbert space is truncated to $\alpha$4, augmented with boundary terms to restore exact translation symmetry (the refined effective Hamiltonian), and block-encoded via LCU. Approximate translation symmetry lifts the success amplitude from $\alpha$5 to $\alpha$6, and oblivious amplitude amplification restores it to $\alpha$7. QSVT implements $\alpha$8. In the general case, Fourier coefficients are reconstructed coherently from Clenshaw–Curtis quadrature of $\alpha$9 plus truncated Gevrey series in the endpoint derivatives, with all LCU coefficient norms bounded by $$\widetilde{\mathcal O}(\alpha T \log(1/\varepsilon))$$0, which is what preserves the additive query scaling.

Relation to prior work

Relative to Mizuta et al.'s optimal algorithms for analytic periodic systems, the contribution is to remove the periodicity assumption by explicitly constructing the periodic extension and deriving the corresponding subexponential Fourier decay and Lieb–Robinson bounds, which do not follow from existing analyses since the extension is Gevrey but not real analytic (by the identity theorem). Relative to the concurrent transducer-based algorithm achieving $$\widetilde{\mathcal O}(\alpha T \log(1/\varepsilon))$$1 queries for Lipschitz Hamiltonians, the present assumptions are stronger (slow variation, analyticity, derivative oracles) and the query bound weaker by polylogarithmic factors, but the gate overhead is polylogarithmic in $$\widetilde{\mathcal O}(\alpha T \log(1/\varepsilon))$$2 rather than polynomial. Relative to product-formula analyses that also use smooth periodic extensions as an analytical device, here the extension is an explicit algorithmic ingredient and is infinitely differentiable, which is essential for the polylogarithmic precision dependence.

Limitations and open questions

Several caveats qualify the results. The general-case oracle model requires block encodings of $$\widetilde{\mathcal O}(\alpha T \log(1/\varepsilon))$$3 and of high-order endpoint derivatives rather than standard HAM-T access to $$\widetilde{\mathcal O}(\alpha T \log(1/\varepsilon))$$4; when $$\widetilde{\mathcal O}(\alpha T \log(1/\varepsilon))$$5 is specified through structured decompositions these may be built from the same primitives at comparable cost, but this must be verified case by case. The slow condition $$\widetilde{\mathcal O}(\alpha T \log(1/\varepsilon))$$6 excludes Hamiltonians whose temporal variation is not uniformly $$\widetilde{\mathcal O}(\alpha T \log(1/\varepsilon))$$7. For Gevrey indices $$\widetilde{\mathcal O}(\alpha T \log(1/\varepsilon))$$8 the query complexity loses a polynomial factor, and the gate overhead carries a nearly quadratic dependence on $$\widetilde{\mathcal O}(\alpha T \log(1/\varepsilon))$$9, worse than the Dyson series' $1/\varepsilon$0 gates. The analysis is formulated purely in the block-encoding model and does not capture structure-sensitive improvements from locality, sparsity, or small commutators. The authors identify several open problems: replacing derivative oracles with standard HAM-T access alone; relaxing the slow condition to an integral norm involving $1/\varepsilon$1; reducing the gate overhead toward $1/\varepsilon$2; and determining whether general Lipschitz-continuous Hamiltonians admit both optimal query complexity and polylogarithmic gate overhead simultaneously.

Conclusion

The paper establishes that slow analytic time-dependent Hamiltonian dynamics can be simulated with nearly additive query complexity $1/\varepsilon$3 while keeping additional gates polylogarithmic in the precision, via a periodic Gevrey extension, subexponential Fourier decay estimates, Floquet embedding, and QSVT. The same framework extends to Gevrey-class Hamiltonians with polynomially degraded scaling and improves the precision dependence of quantum algorithms for slow analytic semi-dissipative linear differential equations through LCHS. The result delineates a trade-off against the concurrent transducer approach: stronger structural assumptions buy gate efficiency, and closing the gap between the two input models remains open.