Randomized product formulas beyond optimal deterministic scaling

Leeseok Kim; Luis Pedro Garc\'ia-Pintos

Randomized product formulas beyond optimal deterministic scaling

T0 review · 0 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Randomized product formulas surpass deterministic quantum-simulation barriers.

desk verdict Rigorous, significant: randomized product formulas beat deterministic no-go bounds for weak perturbations; main caveats are an unpublished self-citation and the infinite-variance gate-count tail. read the letter →

arxiv 2608.07720 v1 pith:ECHGASHZ submitted 2026-08-07 quant-ph

Leeseok Kim , Luis Pedro García-Pintos This is my paper · ORCID

classification quant-ph MSC 81P68 PACS 03.67.Ac

keywords quantumsimulationHamiltonianproductformulasTrotterrandomizedalgorithmsperturbativeregimeTHRIFTerrorscaling

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether classical randomness can lift product formulas (Trotter formulas) for Hamiltonians of the form $H=A+\alpha B$ past error barriers that no deterministic product formula can cross, and answers yes in two access models. The main theorems show a randomized $2k$-th-order formula whose error is $O(\alpha^2 t^{2k+1}/r^{2k})$ at twice the gate depth of the deterministic Suzuki formula, and a randomized $2k$-th-order THRIFT formula whose error is $O(\alpha^3 t^{2k+1}/r^{2k})$ with expected $O(L)$ gates per step. The companion propositions prove that any finite deterministic formula in the same access models must carry an $\Omega(\alpha)$ or $\Omega(\alpha^2)$ leading error, so the improvement genuinely comes from averaging over random circuit realizations. Numerical benchmarks on spin models show the randomized formulas need fewer exponentials in the weak-coupling, high-precision regime.

What carries the argument

The carrying objects are the shifted first-order step $R_1(\tau,u)=e^{-i(1-u)\tau A}e^{-i\alpha\tau B}e^{-iu\tau A}$ with $u$ uniform on $[0,1]$, and, for THRIFT, the $W$-block $W_\ell([x,y])=e^{iyA}e^{-i(y-x)(A+\alpha B_\ell)}e^{-ixA}$ together with local correction unitaries $C_{a,b,n,q}^{[m]}=W_a(Q^R_{n,q})^m W_b(Q^L_{n,q})^m W_a(Q^R_{n,q})^{-m}W_b(Q^L_{n,q})^{-m}$. The shift makes the averaged one-step error agree with the ideal interaction-picture evolution through first order in $\alpha$, because $\delta t\,\mathbb{E}_u[B_I(u\delta t)]=\int_0^{\delta t}B_I(s)\,ds$; the correction unitaries use the group-commutator identity $C=I-\alpha^2 m^2\Gamma_{a,b,n,q}+O(\alpha^3)$ to inject the negative of each dyadic piece of the $\alpha^2$ THRIFT error with probability $1/m^2$. Suzuki recursion, reusing the same random sample inside each high-order step, converts these one-step cancellations into the Theorem 1 and Theorem 2 bounds, and the sharpened mixing lemma converts the averaged-unitary estimate into a diamond-norm estimate.

What would settle it

Fix a one-qubit example $H=\omega Z+\alpha X$ and a fixed number of steps $r$; compute the diamond distance between the averaged randomized first-order channel and the ideal unitary over a range of $\alpha$. Theorem 1 with $k=1$ predicts an error $O(\alpha^2 t^3/r^2)$, so the log-log slope in $\alpha$ should be $2$; a slope of $1$ would falsify the central cancellation. For randomized THRIFT, repeat with two bond operators $B_0,B_1$ and check that the one-step averaged operator error has log-log slope $3$ in $\alpha$ rather than $2$.

Watch

Extended reading notes

Core claim

The central discovery is that a uniform random shift in the placement of the weak-$B$ evolution averages the interaction-picture perturbation over the whole Trotter step, reproducing the exact Dyson integral that deterministic formulas can only sample at finitely many fixed times; this cancels the leading linear-in-$\alpha$ error and, after Suzuki recursion, yields $\alpha^2$ scaling with $2k$-th-order time scaling. In the stronger THRIFT access model, the paper shows how to cancel the $\alpha^2$ commutator term by randomly inserting a group-commutator correction built from $W$-blocks, sampled over dyadic rectangles that partition the time-ordering triangle, leaving an $\alpha^3$ leading error with only $O(L)$ expected gates per step. The accompanying no-go propositions demonstrate that finite deterministic formulas cannot universally achieve either improved scaling, since finite sums of point samples cannot reproduce continuum averages for all operators.

Load-bearing premise

The load-bearing premise is that an expected-gate guarantee with an unbounded tail is acceptable: a single randomized THRIFT run can occasionally need far more gates than average, and truncating the correction levels puts a small systematic error back into the estimate.

Editorial extensions

If this is right

For $H=A+\alpha B$, the randomized first-order formula already achieves an $\alpha^2$ error with essentially the same gate count as a deterministic first-order formula, so it can outperform deterministic higher-order formulas when $\alpha$ is small.
For a fixed target error, the randomized $2k$-th-order gate count scales with $\alpha$ as $\alpha^{1/k}$ rather than the deterministic $\alpha^{1/(2k)}$, which makes the advantage grow as the perturbation weakens.
Randomized $2k$-th-order THRIFT reaches $\alpha^3$ scaling while keeping the expected number of gates per step $O(L)$, so the stronger access model buys an extra power of $\alpha$ without asymptotic gate overhead.
The $\Omega(\alpha)$ and $\Omega(\alpha^2)$ lower bounds imply the $\alpha^2$ and $\alpha^3$ scalings are unobtainable by any finite deterministic formula in those access models, making randomness essential rather than merely helpful.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

An extension the paper leaves implicit is that the same continuous-shift averaging should transfer to time-dependent Hamiltonians or to randomized-compiling settings where a perturbation is dressed by a large drive, since the mechanism only needs the interaction-picture average identity; the paper does not prove this.
A practical corollary is that randomized THRIFT gate counts have an infinite-variance tail, so an implementation must choose a truncation level and absorb the tail bias; the optimal trade-off between tail bias and expected runtime is not determined in the paper.
If the $\alpha^3$ randomized THRIFT scaling is optimal, a matching $\Omega(\alpha^3)$ deterministic lower bound would be a natural next target; the paper states only that this optimality question remains open.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Worth a serious read. The paper settles, in the affirmative, a question that has been floating around: can randomization beat the deterministic scaling limits of product formulas for weakly perturbed Hamiltonians? The answer here is yes, and the proof is solid. The continuous-shift randomization (uniform insertion point for the B-step) is a natural upgrade of the forward/reverse trick of Childs–Ostrander–Su, and the high-order Suzuki recursion works cleanly. Theorem 1 gives O(alpha^2 t^(2k+1)/r^(2k)) in the standard access model at only twice the gate depth, and Proposition 1 shows a deterministic formula cannot do better than Omega(alpha). The THRIFT variant is more surprising: dyadic sampling of group commutators cancels the O(alpha^2) term in expectation, giving O(alpha^3) with expected O(L) gates, against an Omega(alpha^2) deterministic barrier. I checked the SM; the joint-entirety argument, the dyadic partition, and the beta in (1,2) tradeoff all hold together. No load-bearing flaw.

Soft spots, in rough order of importance. First, the 'beyond optimal deterministic scaling' claim for the standard access model leans on an unpublished self-citation, Ref [44]. The theorems stand on their own, but the optimality framing depends on a manuscript I cannot verify. Second, the expected O(L) gate count for randomized THRIFT has an infinite-variance tail. The paper is upfront about this and bounds the truncation bias, so it is a practical caveat, not a mathematical one. Third, no code or data released; the numerics are reproducible in principle but a repo would help. Fourth, the plots certify error via upper bounds, not actual diamond distances—standard practice, but the crossover locations in Figs. 3–4 are only as tight as those bounds.

Who this is for: anyone working on product formulas, interaction-picture simulation, or randomized compiling. The main results will be cited. I would send it to a strong journal if the SM survives review and the authors post the companion paper. Deserves a serious referee.

Referee Report

0 major / 5 minor

Summary. This paper considers Hamiltonian simulation of H = A + alpha B with small alpha in two oracle models. In the standard access model (exponentials of A and alpha B), the authors introduce a randomized shifted Trotter formula and prove that the averaged r-step channel has diamond-norm error O(alpha^2 t^{2k+1} / r^{2k}) for the 2k-th order construction, at roughly twice the gate count of the deterministic Suzuki formula. They also prove that no finite deterministic formula can achieve better than Omega(alpha) in this model. In a stronger access model (exponentials of A and A + alpha B_ell), they develop randomized THRIFT, prove O(alpha^3 t^{2k+1} / r^{2k}) error with expected O(L) gates per step, and prove a deterministic lower bound Omega(alpha^2). Numerical experiments on a 9-qubit transverse-field Ising model and an 8-qubit random-field Heisenberg chain show gate-count advantages in the weak-coupling, high-precision regime. The proofs are contained in a detailed Supplemental Material.

Significance. The results are significant and, based on my reading of the Supplemental Material, correct. The main theorems are derived self-containedly from the interaction picture, Dyson series, and Suzuki recursion, with no fitted parameters, and the lower bounds use elegant quadrature/Fourier-witness arguments. The honest treatment of the infinite-variance gate-count tail and the truncation bias in the numerics is a strength. If the results stand, they demonstrate a genuine separation between randomized and deterministic product formulas with only constant overhead, which is likely to influence practical Hamiltonian simulation in perturbative regimes.

minor comments (5)

[SM S1.2, Lemma S1] The stated inequality 1/2 ||U - V||_diamond <= 2 ||U - E_omega V_omega|| is a factor of two looser than the standard mixing lemma, which gives ||U - V||_diamond <= 2 ||U - E[V]||. Since the scaling claims are unaffected, this is a presentation point, but the statement and the label 'sharpened' should be checked against Ref. [48].
[Main text, stronger access model and Fig. 4] The base formula in Eq. (27) is called 'randomized first-order THRIFT' although its error scales as O(delta t^3), i.e., it is second order in the Trotter step; the label should be clarified to avoid confusion with the order of the deterministic formula.
[Conclusion] The 'known quadratic lower bounds [43,44]' include an unpublished manuscript [44] by the same authors; either remove it from the 'known' claim or explicitly mark it as unpublished, and add a sentence explaining why Ref. [43] applies to the alpha-scaling of product formulas.
[References and title page] There are minor typos: Ref. [19] spells 'Physiscal Review Letters', and the author name on the title page ('Garc \'ia-Pintos') has unusual spacing. These should be corrected in the final version.
[SM S4.1, Eq. (S145)] The expression for qbar_N is typeset ambiguously; please clarify the exponent and denominator, for example by using parentheses around 1 - 2 beta in the exponent.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the central theorems are derived self-containedly, with only minor non-load-bearing first-party citations.

full rationale

The main claims (Theorem 1 and Theorem 2) are proven from first principles in the Supplemental Material using the interaction picture, Dyson series, the sharpened mixing lemma (Ref. [48], external), Suzuki recursion, and explicit group-commutator expansions. No parameter is fitted to data and no 'prediction' is a renamed input. The randomized first-order cancellation follows from the explicit identity delta t E_u[B_I(u delta t)] = integral_0^{delta t} B_I(s) ds, and the high-order alpha-scaling is established by the joint-entire expansion of F_{2k}(tau, alpha), not by assuming the conclusion. For randomized THRIFT, the mean-zero property E_omega[D_omega(x)]=0 follows by construction from p_{a,b,n,q}=1/m_n^2 and the dyadic partition of the time-ordering triangle; the error and expected gate-count estimates are then independent calculations involving beta in (1,1+1/(2k+2)). The deterministic lower bounds in Propositions 1 and 2 are separate quadrature/Fourier-witness arguments. The only first-party references are Ref. [44], an unpublished manuscript by one of the authors cited only to 'suggest' optimality of the alpha-scaling, and Ref. [49], cited only as inspiration for a polynomial witness in the proof of Proposition 1; neither is load-bearing for any theorem. The known limitation of infinite-variance gate count is explicitly acknowledged in SM Eq. (S96), and the truncation bias is explicitly bounded in SM S4.1. Thus no circular step can be exhibited, and the paper is self-contained against external benchmarks.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

No new physical entities or forces are introduced. The free parameters are algorithm-design hyperparameters (beta, c, m_n) that are chosen by hand to ensure convergence and valid probabilities; they are not fitted to data. The axioms are standard mathematical tools and the explicit bounded-operator domain assumption.

free parameters (3)

dyadic sampling exponent beta_k = beta=1.7 in numerics (L=2); in general any 1<beta_k<1+1/(2k+2)
Algorithm design parameter in randomized THRIFT (SM S3.3): controls the tradeoff between correction probability and gate overhead; chosen by hand to make series converge and expected gate count O(L).
dyadic sampling constant c_k = c = sqrt(2*(C(L,2))/(1-2^(1-2 beta))) in the construction; for L=2, beta=1.7 gives c ≈ 1.579
Chosen large enough so that the total probability of applying a correction is at most 1/2, making the sampling distribution valid (SM S3.2).
minimum correction order m_n = m_n = ceil(c 2^(beta n))
The correction unitary uses 4*m_n extra W-blocks; this specific scaling with n makes the expected gate-count series converge (SM Eqs. S89, S94).

assumptions (3)

standard math Sharpened mixing lemma (Lemma S1): for a random unitary V and fixed U, (1/2)||U - E[V]||_diamond <= 2||U - E[V]||_op.
Used in SM S1.2 to convert the operator-norm distance of averaged unitaries into a diamond-norm bound for the corresponding channels; cited to Ref [48] and used throughout Theorems 1 and 2.
standard math Dyson series / time-ordered exponential expansion converges for bounded operators (SM S1.1, Eq. S7).
The proof expands e^{-i tau(A+alpha B)} and the W-blocks as Dyson series and retains leading orders in alpha; this requires bounded operators and uniform convergence.
domain assumption A and B (and B_l) are Hermitian bounded operators with comparable norms, and alpha << 1.
The error bounds are asymptotic in alpha and tau; the analysis does not address unbounded or long-range Hamiltonians beyond the finite-dimensional / bounded-operator setting used in the proofs.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Randomized product formulas beyond optimal deterministic scaling." pith.science (2026). https://pith.science/paper/ECHGASHZ

@misc{pith2026260807720,
  author       = {Pith},
  title        = {Pith review of: Randomized product formulas beyond optimal deterministic scaling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ECHGASHZ}},
  note         = {Machine review of arXiv:2608.07720}
}

abstract

Product formulas, also known as Trotter formulas, are among the most widely used and practical methods for simulating quantum systems on quantum computers. Here we introduce two new classes of randomized product formulas for simulating Hamiltonians with separated energy scales, $H=A+\alpha B$, where $\alpha$ is small. In the standard access model, where one can implement exponentials of $A$ and $B$ separately, our randomized formulas achieve $\mathcal O(\alpha^2)$ error scaling at the cost of only doubling the gate depth of the corresponding deterministic formula. We further prove an $\Omega(\alpha)$ lower bound for deterministic product formulas. In a stronger access model, allowing exponentials of $A+\alpha B_\ell$ for $B = \sum_{\ell}B_\ell$, our randomized formula, based on Trotter Heuristic Resource Improved Formulas for Time-dynamics (THRIFT)~[J. L. Bosse et al., Nat. Commun. 16, 2673 (2025)], achieves $\mathcal O(\alpha^3)$ error scaling with only constant-factor expected gate overhead. We also establish an $\Omega(\alpha^2)$ lower bound for deterministic product formulas in this access model. Numerical simulations confirm gate-count reductions for simulating physically motivated systems.

Figures

Figures reproduced from arXiv: 2608.07720 by the authors.

**Figure 2.** FIG. 2: Dyadic sampling for randomized THRIFT. Sam [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 4.** FIG. 4: Gate-count advantage of randomized first-order [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 30 canonical work pages

[44]

W. F. Braasch, Jr., L. Kim, and M. Marvian, Limits of Advantage in Randomization in Quantum Protocols (2026), manuscript in preparation

work page 2026
[1]

R. P. Feynman, Simulating physics with computers, International Journal of Theoretical Physics21, 467 (1982)

1982
[2]

A. M. Childs, On the relationship between continuous- and discrete-time quantum walk, Communications in Mathematical Physics294, 581 (2010)

2010
[3]

A. M. Childs and N. Wiebe, Hamiltonian simulation us- ing linear combinations of unitary operations, Quantum Information and Computation12, 901 (2012)

2012
[4]

D. W. Berry, A. M. Childs, and R. Kothari, Hamilto- nian simulation with nearly optimal dependence on all parameters, inProceedings of the 56th IEEE Symposium on Foundations of Computer Science(2015) pp. 792–809

work page 2015
[5]

D. W. Berry, A. M. Childs, R. Cleve, R. Kothari, and R. D. Somma, Simulating Hamiltonian dynamics with a truncated taylor series, Physical Review Letters114, 090502 (2015)

work page 2015
[6]

G. H. Low and I. L. Chuang, Optimal Hamiltonian sim- ulation by quantum signal processing, Physical Review Letters118, 010501 (2017)

2017
[7]

G. H. Low and I. L. Chuang, Hamiltonian simulation by qubitization, Quantum3, 163 (2019)

2019

Show all 58 references

[8]

H. F. Trotter, On the product of semi-groups of opera- tors, Proceedings of the American Mathematical Society 10, 545 (1959)

1959
[9]

M. Suzuki, Generalized trotter’s formula and systematic approximants of exponential operators and inner deriva- tions with applications to many-body problems, Commu- nications in Mathematical Physics51, 183 (1976)

1976
[10]

Suzuki, Fractal decomposition of exponential opera- tors with applications to many-body theories and monte carlo simulations, Physics Letters A146, 319 (1990)

M. Suzuki, Fractal decomposition of exponential opera- tors with applications to many-body theories and monte carlo simulations, Physics Letters A146, 319 (1990)

1990
[11]

A. M. Childs, D. Maslov, Y. Nam, N. J. Ross, and Y. Su, Toward the first quantum simulation with quan- tum speedup, Proceedings of the National Academy of Sciences115, 9456 (2018)

2018
[12]

A. M. Childs, Y. Su, M. C. Tran, N. Wiebe, and S. Zhu, Theory of trotter error with commutator scaling, Physi- cal Review X11, 011020 (2021)

2021
[13]

B. P. Lanyon, C. Hempel, D. Nigg, M. M¨ uller, R. Ger- ritsma, F. Z¨ ahringer, P. Schindler, J. T. Barreiro, M. Rambach, G. Kirchmair, M. Hennrich, P. Zoller, R. Blatt, and C. F. Roos, Universal digital quantum sim- ulation with trapped ions, Science334, 57 (2011)

2011
[14]

E. A. Martinez, C. A. Muschik, P. Schindler, D. Nigg, A. Erhard, M. Heyl, P. Hauke, M. Dalmonte, T. Monz, P. Zoller, and R. Blatt, Real-time dynamics of lattice gauge theories with a few-qubit quantum computer, Na- ture534, 516 (2016)

2016
[15]

Arute, K

F. Arute, K. Arya, R. Babbush, D. Bacon, J. C. Bardin, R. Barends, A. Bengtsson, S. Boixo, M. Broughton, B. B. Buckley,et al., Observation of separated dynamics of charge and spin in the fermi-hubbard model, arXiv preprint arXiv:2010.07965 10.48550/arXiv.2010.07965 (2020)

2020 doi arXiv
[16]

Y. Kim, A. Eddins, S. Anand,et al., Evidence for the utility of quantum computing before fault tolerance, Na- ture618, 500 (2023)

2023
[17]

T. A. Cochran, B. Jobst, E. Rosenberg, Y. D. Lensky, G. Gyawali,et al., Visualizing dynamics of charges and strings in (2 + 1)D lattice gauge theories, Nature642, 315 (2025)

2025
[18]

Viola and E

L. Viola and E. Knill, Random Decoupling Schemes for Quantum Dynamical Control and Error Suppression, Physical Review Letters94, 060502 (2005)

2005
[19]

L. F. Santos and L. Viola, Enhanced convergence and robust performance of randomized dynamical decoupling, Physiscal Review Letters97, 150501 (2006)

2006
[20]

Boixo, E

S. Boixo, E. Knill, and R. Somma, Eigenpath traver- sal by phase randomization, Quantum Info. Comput.9, 833–855 (2009)

2009
[21]

J. J. Wallman and J. Emerson, Noise tailoring for scalable quantum computation via randomized compiling, Physi- cal Review A94, 052325 (2016)

2016
[22]

K. Wan, M. Berta, and E. T. Campbell, Random- ized quantum algorithm for statistical phase estimation, Physical Review Letters129, 030503 (2022)

2022
[23]

J. M. Martyn and P. Rall, Halving the cost of quantum algorithms with randomization, npj Quantum Informa- tion11, 47 (2025)

2025
[24]

Gosset, R

D. Gosset, R. Kothari, and C. Zhang, Multi-qubit toffoli with exponentially fewer t gates (2025), arXiv:2510.07223 [quant-ph]

2025
[25]

C. Yi, L. Kim, and M. Marvian, Faster randomized dy- namical decoupling, Physical Review Letters136, 010601 (2026)

2026
[26]

L. Kim, F. Riberi, K. Baca, and M. Marvian, Random- ized quantum optimal control (2026), arXiv:2607.10946 [quant-ph]

2026 arXiv
[27]

G¨ unther, F

J. G¨ unther, F. Witteveen, A. Schmidhuber, M. Miller, M. Christandl, and A. W. Harrow, Phase estimation with partially randomized time evolution, PRX Quantum7, 7 020332 (2026)

2026
[28]

A. W. Harrow, A. Lowe, and F. Witteveen, Randomized truncation of quantum states (2026), arXiv:2510.08518 [quant-ph]

2026
[29]

Campbell, Random compiler for fast Hamiltonian sim- ulation, Physical Review Letters123, 070503 (2019)

E. Campbell, Random compiler for fast Hamiltonian sim- ulation, Physical Review Letters123, 070503 (2019)

2019
[30]

A. M. Childs, A. Ostrander, and Y. Su, Faster quantum simulation by randomization, Quantum3, 182 (2019)

2019
[31]

Ouyang, D

Y. Ouyang, D. R. White, and E. T. Campbell, Compila- tion by stochastic Hamiltonian sparsification, Quantum 4, 235 (2020)

2020
[32]

P. K. Faehrmann, M. Steudtner, R. Kueng, M. Kieferova, and J. Eisert, Randomizing multi-product formulas for Hamiltonian simulation, Quantum6, 806 (2022)

2022
[33]

C.-H. Cho, D. W. Berry, and M.-H. Hsieh, Doubling the order of approximation via the randomized product for- mula, Physical Review A109, 062431 (2024)

2024
[34]

H. Chen, B. Li, J. Lu, and L. Ying, A Randomized Method for Simulating Lindblad Equations and Thermal State Preparation, Quantum9, 1917 (2025)

2025
[35]

G. H. Low and N. Wiebe, Hamiltonian simulation in the interaction picture, arXiv preprint arXiv:1805.00675 10.48550/arXiv.1805.00675 (2018)

2018 doi arXiv
[36]

D. W. Berry, A. M. Childs, Y. Su, X. Wang, and N. Wiebe, Time-dependent Hamiltonian simulation with L1-norm scaling, Quantum4, 254 (2020)

2020
[37]

D. An, D. Fang, and L. Lin, Time-dependent Hamilto- nian simulation of highly oscillatory dynamics and super- convergence for schr¨ odinger equation, Quantum6, 690 (2022)

2022
[38]

Sharma and M

K. Sharma and M. C. Tran, Hamiltonian simulation in the interaction picture using the magnus expansion (2024), arXiv:2404.02966 [quant-ph]

2024 arXiv
[39]

J. L. Bosse, A. M. Childs, C. Derby, F. M. Gambetta, A. Montanaro, and R. A. Santos, Efficient and practi- cal Hamiltonian simulation from time-dependent product formulas, Nature Communications16, 10.1038/s41467- 025-57580-5 (2025)

2025 doi
[40]

Bagherimehrab, L

M. Bagherimehrab, L. Mantilla Calder´ on, D. W. Berry, P. Schleich, M. Ghazi Vakili, A. Aldossary, J. A. Cam- pos Gonzalez Angulo, C. Gorgulla, and A. Aspuru-Guzik, Faster algorithmic quantum and classical simulations by corrected product formulas, PRX Quantum7, 033018 (2026)

2026
[41]

Watrous,The Theory of Quantum Information(Cam- bridge University Press, Cambridge, 2018)

J. Watrous,The Theory of Quantum Information(Cam- bridge University Press, Cambridge, 2018)

2018
[42]

A. Y. Kitaev, A. Shen, and M. N. Vyalyi,Classical and quantum computation, 47 (American Mathematical Soc., 2002)

2002
[43]

Akibue, G

S. Akibue, G. Kato, and S. Tani, Probabilistic unitary synthesis with optimal accuracy, ACM Transactions on Quantum Computing5, 10.1145/3663576 (2024)

2024 doi
[45]

J. J. Sakurai and J. Napolitano,Modern quantum me- chanics(Cambridge university press, 2020)

2020
[46]

Campbell, Shorter gate sequences for quantum com- puting by mixing unitaries, Physical Review A95, 042306 (2017)

E. Campbell, Shorter gate sequences for quantum com- puting by mixing unitaries, Physical Review A95, 042306 (2017)

2017
[47]

M. B. Hastings, Turning gate synthesis errors into in- coherent errors, Quantum Info. Comput.17, 488–494 (2017)

2017
[48]

Chen, H.-Y

C.-F. Chen, H.-Y. Huang, R. Kueng, and J. A. Tropp, Concentration for random product formulas, PRX Quan- tum2, 040305 (2021)

2021
[49]

Kim and M

L. Kim and M. Marvian, High-order dynamical decoupling in the weak-coupling regime (2026), arXiv:2602.05343 [quant-ph]

2026
[50]

Aharonov, A

D. Aharonov, A. Kitaev, and N. Nisan, Quantum circuits with mixed states, inProceedings of the Thirtieth Annual ACM Symposium on Theory of Computing, STOC ’98 (Association for Computing Machinery, New York, NY, USA, 1998) p. 20–30. 1 Supplemental Material CONTENTS S1. Prelimin...

1998
[51]

Dyadic correction construction 8
[52]

Explicit sampling procedure 9
[53]

Error bound on the local correction unitary 10
[54]

(28) 11 S3.3

Derivation of Eq. (28) 11 S3.3. Proof of Theorem 2 13 S3.4. Proof of Proposition 2 16 S4. Numerical simulation details and additional remarks 19 S4.1. Error bounds and gate-cost evaluation 19 S4.2. Gate-count crossovers across target precisions 19 S4.3. Finite-sampling error 2...
[55]

The sampling distribution is chosen so that the averaged second-order 9 contribution of these corrections is the negative of the leadingα 2 error in Eq

Dyadic correction construction Randomized THRIFT appends toS(τ) a randomly sampled correction unitary built entirely fromW-blocks, and hence implementable in the same access model. The sampling distribution is chosen so that the averaged second-order 9 contribution of these co...
[56]

For an explicit implementation, it is convenient to sample this distribution hierarchically rather than enumerate all tuples (a, b, n, q)

Explicit sampling procedure The probabilities defined above specify the weightp a,b,n,q =m −2 n of each correction branch, together with the probabilityp 0 of applying no correction. For an explicit implementation, it is convenient to sample this distribution hierarchically ra...
[57]

Error bound on the local correction unitary A single randomized THRIFT step in the interaction picture is Srand(τ, ω) := ( S(τ), ω= 0, C [mn] a,b,n,qS(τ), ω= (a, b, n, q), (S73) where Pr[ω= 0] =p 0 and Pr[ω= (a, b, n, q)] =pa,b,n,q. The following lemma shows that each local co...
[58]

Derivation of Eq.(28) In the Schr¨ odinger picture, the one-step formula corresponds to Trand(τ, ω) :=e−iτ ASrand(τ, ω).(S83) Forδt=t/rand independent samplesω= (ω 1, . . . , ωr), define ther-step randomized THRIFT formula T (r) rand(t,ω) :=T rand(δt, ωr)· · ·Trand(δt, ω1).(S8...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [44]

W. F. Braasch, Jr., L. Kim, and M. Marvian, Limits of Advantage in Randomization in Quantum Protocols (2026), manuscript in preparation

work page 2026

[2] [1]

R. P. Feynman, Simulating physics with computers, International Journal of Theoretical Physics21, 467 (1982)

1982

[3] [2]

A. M. Childs, On the relationship between continuous- and discrete-time quantum walk, Communications in Mathematical Physics294, 581 (2010)

2010

[4] [3]

A. M. Childs and N. Wiebe, Hamiltonian simulation us- ing linear combinations of unitary operations, Quantum Information and Computation12, 901 (2012)

2012

[5] [4]

D. W. Berry, A. M. Childs, and R. Kothari, Hamilto- nian simulation with nearly optimal dependence on all parameters, inProceedings of the 56th IEEE Symposium on Foundations of Computer Science(2015) pp. 792–809

work page 2015

[6] [5]

D. W. Berry, A. M. Childs, R. Cleve, R. Kothari, and R. D. Somma, Simulating Hamiltonian dynamics with a truncated taylor series, Physical Review Letters114, 090502 (2015)

work page 2015

[7] [6]

G. H. Low and I. L. Chuang, Optimal Hamiltonian sim- ulation by quantum signal processing, Physical Review Letters118, 010501 (2017)

2017

[8] [7]

G. H. Low and I. L. Chuang, Hamiltonian simulation by qubitization, Quantum3, 163 (2019)

2019