arxiv: 2603.12054 · v2 · submitted 2026-03-12 · 🪐 quant-ph

Recognition: 2 theorem links

· Lean Theorem

Noise Correlations as a Resource in Pauli-Twirled Circuits

Antoine Brillant , Rohan N Rajmohan , Peter Groszkowski , Alireza Seif , Jens Koch , Aashish Clerk

Authors on Pith no claims yet

Pith reviewed 2026-05-15 12:04 UTC · model grok-4.3

classification 🪐 quant-ph

keywords randomized compilingPauli twirlingnoise correlationscircuit fidelityClifford circuitsGaussian noisequantum error mitigation

0 comments

The pith

Correlations increase the fidelity of randomly compiled Clifford circuits under Gaussian noise.

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

The paper establishes that randomized compiling reduces both the strength and temporal range of correlations for a broad class of Gaussian noise. In Clifford circuits an exact analytical formula for the circuit fidelity shows that this quantity always rises when correlations are present. This turns memory effects from a liability into a resource that improves overall performance. Simulations indicate the fidelity gain persists in many non-Clifford circuits as well. To leading order the protocol also suppresses the quantum part of bath correlations, justifying a classical treatment of weak noise.

Core claim

For a broad class of correlated Gaussian noise, randomized compiling reduces both the strength and temporal range of correlations. For Clifford circuits the circuit fidelity is given by a simple analytical expression that increases with the presence of correlations. To leading order in system-bath coupling, randomized compiling suppresses the quantum component of bath correlations, allowing weak noise to be treated as classical. The result remains valid for many relevant non-Clifford circuits according to numerical simulations.

What carries the argument

Randomized compiling via random Pauli twirls that convert general noise into Pauli errors, together with the derived analytical fidelity expression for Clifford circuits.

If this is right

Randomized compiling shortens the temporal range of noise correlations and thereby mitigates memory effects.
To leading order the quantum component of bath correlations is suppressed, permitting a classical noise model for weak coupling.
The fidelity increase holds for many non-Clifford circuits according to numerical checks.
Correlations can be treated as a resource that enhances robustness rather than purely as an error source.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Deliberately engineered correlations could be introduced to improve the performance of compiled circuits on hardware.
The result may guide error-mitigation design for devices whose dominant errors already contain spatial or temporal correlations.
Similar benefits might appear in other twirling or randomization protocols beyond the Gaussian case examined here.

Load-bearing premise

The noise belongs to a broad class of correlated Gaussian noise.

What would settle it

A controlled experiment that measures fidelity in a randomly compiled Clifford circuit with and without added Gaussian correlations and checks whether fidelity rises with correlation strength.

Figures

Figures reproduced from arXiv: 2603.12054 by Aashish Clerk, Alireza Seif, Antoine Brillant, Jens Koch, Peter Groszkowski, Rohan N Rajmohan.

**Figure 2.** Figure 2: Noise-averaged circuit fidelity of random Clif [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗

**Figure 4.** Figure 4: Logical survival probability of the [[3, 1, 1]] repetition code initialized in the logical |0⟩ state after 250 errorcorrection cycles as a function of the noise correlation time. Each data and ancilla qubit is subject to non-Markovian dephasing noise after each two-qubit gate, with correlation function Eq.(53) and noise strength σ = 0.05. We compare the Pauli-twirled (green) and bare (orange) circuit i… view at source ↗

**Figure 5.** Figure 5: Circuit fidelity for 100 random Clifford and non [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗

read the original abstract

Randomized compiling (RC) is an established tool to tailor arbitrary quantum noise channels into Pauli errors. The effect of both spatial and temporal noise correlations in randomly compiled circuits, however, is not fully understood. Here, we show that for a broad class of correlated Gaussian noise, RC reduces both the strength and temporal range of correlations. For Clifford circuits, we derive a simple analytical expression for the circuit fidelity of randomly compiled circuits. Surprisingly, we show that this fidelity is always increased by the presence of correlations, suggesting that correlations are a resource in randomly compiled circuits. To leading order in system-bath coupling, we also show that RC suppresses the quantum component of bath correlations, implying that one can safely treat weak noise as being classical. Finally, through extensive numerical simulations, we show that our results remain valid for many relevant non-Clifford circuits. These results clarify how RC mitigates memory effects and enhances circuit robustness.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Correlations boost fidelity in RC Clifford circuits when marginal variances stay fixed, but the normalization choice needs explicit checks.

read the letter

The main thing to know is that this paper finds correlations in Gaussian noise can increase the fidelity of randomly compiled Clifford circuits, which runs against the usual assumption that correlations are purely a liability in error mitigation. They derive a simple analytical expression for that fidelity and show it rises with correlation strength under their noise model. The work also demonstrates that randomized compiling reduces both the strength and the temporal range of the correlations, and that weak bath correlations lose their quantum character to leading order, letting the noise be treated as classical. The numerics extend the fidelity gain to several non-Clifford cases, which makes the claim more relevant to actual hardware circuits. This is new relative to the RC literature they cite; earlier papers treat correlations mainly as leftover error after twirling, not as something that can be leveraged. The derivation for the Clifford case appears direct and the numerics are presented as broad validation. The clearest soft spot is the noise parameterization. The fidelity increase assumes fixed per-qubit variances on the covariance diagonal while off-diagonal terms vary. If the comparison instead holds total noise power fixed, the benefit can disappear or reverse, as the stress-test note suggests. The abstract does not flag this distinction, so the full text should make the normalization explicit. The non-Clifford simulations are useful but would be easier to assess with more detail on error bars and circuit selection. This is aimed at people working on randomized compiling and noise tailoring for near-term devices. Anyone designing RC protocols would get concrete value from the analytical result and the suggestion that certain correlations can be turned to advantage. I would send it for peer review; the central claim is grounded enough to deserve referee scrutiny even if the normalization point needs tightening.

Referee Report

1 major / 2 minor

Summary. The manuscript analyzes the impact of spatial and temporal correlations in a broad class of Gaussian noise on randomized compiling (RC) for quantum circuits. It shows that RC reduces both the strength and temporal range of correlations. For Clifford circuits an analytical expression for the circuit fidelity is derived, with the surprising result that this fidelity is always increased by the presence of correlations. To leading order in system-bath coupling, RC is shown to suppress the quantum component of bath correlations, allowing weak noise to be treated as classical. The findings are supported by numerical simulations for many relevant non-Clifford circuits.

Significance. If the central analytical result holds, the work establishes that noise correlations can act as a resource that improves fidelity in randomly compiled circuits rather than degrading it, which has direct implications for error mitigation in near-term quantum devices. The closed-form fidelity expression for Clifford circuits and the leading-order demonstration that RC renders weak noise effectively classical are particularly valuable contributions that could inform circuit design and noise modeling.

major comments (1)

[Analytical derivation for Clifford circuits] The analytical derivation of the Clifford-circuit fidelity under correlated Gaussian noise (the section presenting the closed-form expression): the reported monotonic increase in fidelity with correlation strength is load-bearing on the choice to hold the diagonal entries of the covariance matrix fixed while varying the off-diagonal terms. The manuscript does not explicitly confirm that the uncorrelated reference case is normalized to the same per-qubit marginal variance (or equivalently the same total integrated noise power); under the alternative normalization the sign of the effect can reverse for positive correlations, undermining the claim that correlations are always a resource.

minor comments (2)

[Abstract] The abstract states that results remain valid for 'many relevant non-Clifford circuits' but provides no quantitative bounds on circuit depth, noise strength, or the specific gate sets tested; adding these details would strengthen the generality statement.
[Numerical simulations] In the numerical validation section, the handling of statistical error bars, the number of sampled circuits, and any data-selection criteria should be stated explicitly to permit independent reproduction.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading and constructive feedback. We address the single major comment below, clarifying the normalization choice in our analytical derivation while acknowledging the referee's point on alternative normalizations.

read point-by-point responses

Referee: [Analytical derivation for Clifford circuits] The analytical derivation of the Clifford-circuit fidelity under correlated Gaussian noise (the section presenting the closed-form expression): the reported monotonic increase in fidelity with correlation strength is load-bearing on the choice to hold the diagonal entries of the covariance matrix fixed while varying the off-diagonal terms. The manuscript does not explicitly confirm that the uncorrelated reference case is normalized to the same per-qubit marginal variance (or equivalently the same total integrated noise power); under the alternative normalization the sign of the effect can reverse for positive correlations, undermining the claim that correlations are always a resource.

Authors: We thank the referee for this observation. In the derivation, the diagonal entries of the covariance matrix (local per-qubit and per-time noise variances) are held fixed while off-diagonal terms are varied to introduce correlations. This normalization keeps the marginal noise strength per qubit constant, treating correlations as an additive resource on top of fixed local noise; the uncorrelated reference case therefore shares identical marginal variances. We will revise the manuscript to state this normalization explicitly and confirm the reference case. We acknowledge that an alternative normalization fixing total integrated noise power (by reducing diagonals as correlations increase) can reverse the sign of the fidelity change for positive correlations. However, we argue that the fixed-marginal normalization is the physically relevant one for assessing whether correlations act as a resource when local noise strengths are unchanged. We will add a short discussion of both normalizations to qualify the claim appropriately. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation; fidelity expression derived independently from Gaussian noise model

full rationale

The central result derives an analytical fidelity expression for Clifford circuits under correlated Gaussian noise directly from the noise covariance structure. No step reduces the output to a fitted input by construction, nor relies on self-citation for a uniqueness theorem or ansatz. The parameterization (fixed diagonals, variable off-diagonals) is an explicit modeling assumption stated in the derivation, not a tautological redefinition. The claim that correlations increase fidelity follows from the math rather than being presupposed. This is a standard non-circular derivation with independent content.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on the assumption of a broad class of correlated Gaussian noise and on the restriction to Clifford circuits for the analytic result; no free parameters or invented entities are mentioned in the abstract.

axioms (2)

domain assumption Noise is drawn from a broad class of correlated Gaussian channels
Invoked to derive the reduction in correlation strength and the fidelity expression
domain assumption Circuits are composed of Clifford gates for the analytic fidelity formula
Required for the closed-form fidelity result

pith-pipeline@v0.9.0 · 5470 in / 1157 out tokens · 36995 ms · 2026-05-15T12:04:31.257435+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

For Clifford circuits, we derive a simple analytical expression for the circuit fidelity... λ_Q̂ ≈ 1/√det(1+4M_Q Σ)
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We consider a broad class of correlated Gaussian noise... covariance matrix Σ

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages · 3 internal anchors

[1]

Since the inserted Pauli gates are uncorrelated between different layers, we can aver- age each layer independently

Averaging over random Paulis The first step in our analysis is thus to average over the random Pauli gates. Since the inserted Pauli gates are uncorrelated between different layers, we can aver- age each layer independently. Denoting the average over Paulis by⟨·⟩ P ≡ 1 4n P P[·], one finds ⟨P RZ(⃗θj)P⟩ P = nY α=1 h cos2(θ(α) j )I+ sin 2(θ(α) j )Z (α) i (9...

work page
[2]

We take these to be zero-mean Gaussian random variables with a covariance matrixΣ: Σ= ⃗θ ⃗θT ,(20) where overline is used to denote a noise average

Averaging over correlated noise The next step is to average over the noise variables θ(α) j . We take these to be zero-mean Gaussian random variables with a covariance matrixΣ: Σ= ⃗θ ⃗θT ,(20) where overline is used to denote a noise average. Spatial and temporal noise correlations are thus encoded in the form ofΣ. For this noise model, the noise average ...

work page
[3]

,[σ(n) l ]2),(24) so that each [σ (α) j ]2 characterizes the variance of the er- ror associated with gate layerjand qubitα

Pauli-twirling leverages noise correlations We start our analysis by writing the diagonals of the noise covariance matrix as: diag(Σ) = ([σ(1) 1 ]2,[σ (1) 2 ]2, . . . ,[σ(n) l ]2),(24) so that each [σ (α) j ]2 characterizes the variance of the er- ror associated with gate layerjand qubitα. The matrix diag(Σ) therefore contains the information to character...

work page
[4]

In the absence of correlations, the correlation matrix is simply diagonal

Limiting cases To understand the origin of the lower and upper bounds derived in the previous section, we analyze the two extremal covariance matrices, namely the ones corre- sponding to the limits of Markovian noise with no spatial correlations and of quasistatic noise with maximal spatial correlations. In the absence of correlations, the correlation mat...

work page
[5]

The process fidelity can be written as a sum of terms of the form λ= 1p det(1+MΣ) ,(B12) as shown in Eq

Circuit fidelity bounds for finite-duration gates We now turn to the derivation of upper and lower bounds on the process fidelity in the presence of finite- duration gates. The process fidelity can be written as a sum of terms of the form λ= 1p det(1+MΣ) ,(B12) as shown in Eq. (21). HereΣdenotes the covariance matrix of the noise variables, andMis one of ...

work page
[6]

Lower bound The matrixMhas a block structure indexed by pairs of qubits, M=   M(1,1) · · ·M (1,n) ... ... ... M(n,1) · · ·M (n,n)   ,(B13) where each block is diagonal in the layer index. The covariance matrix can be written in an analogous form, Σ=   Σ(1,1) · · ·Σ (1,n) ... ... ... Σ(n,1) · · ·Σ (n,n)   .(B14) It is convenient to reorder the ...

work page
[7]

Upper bound We now derive an upper bound onλ. This is equivalent to finding a lower bound on the determinant det(1+MΣ).(B24) As before we write det(1+MΣ) = det 1+M 1/2ΣM1/2 ,(B25) which is well defined sinceMis positive semidefinite. For any positive semidefinite matrixA, the determi- nant satisfies the bound det(1+A)≥1 + tr(A).(B26) Applying this toA=M 1...

work page
[8]

Z. Cai, R. Babbush, S. C. Benjamin, S. Endo, W. J. Hug- gins, Y. Li, J. R. McClean, and T. E. O’Brien, Quantum error mitigation, Reviews of Modern Physics95, 045005 (2023)

work page 2023
[9]

Van Den Berg, Z

E. Van Den Berg, Z. K. Minev, A. Kandala, and K. Temme, Probabilistic error cancellation with sparse Pauli–Lindblad models on noisy quantum processors, Na- ture Physics19, 1116 (2023)

work page 2023
[10]

M. A. Nielsen, A simple formula for the average gate fi- delity of a quantum dynamical operation, Physics Letters A303, 249 (2002), arXiv:quant-ph/0205035

work page internal anchor Pith review Pith/arXiv arXiv 2002
[11]

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 (ACM Press, Dallas, Texas, United States, 1998) pp. 20– 30

work page 1998
[12]

C. A. Ryan, M. Laforest, and R. Laflamme, Random- ized benchmarking of single- and multi-qubit control in liquid-state NMR quantum information processing, New Journal of Physics11, 013034 (2009)

work page 2009
[13]

Characterizing Quantum Gates via Randomized Benchmarking

E. Magesan, J. M. Gambetta, and J. Emerson, Charac- terizing Quantum Gates via Randomized Benchmarking, Physical Review A85, 042311 (2012), arXiv:1109.6887 [quant-ph]

work page internal anchor Pith review Pith/arXiv arXiv 2012
[14]

J. M. Epstein, A. W. Cross, E. Magesan, and J. M. Gam- betta, Investigating the limits of randomized benchmark- ing protocols, Physical Review A89, 062321 (2014)

work page 2014
[15]

Erhard, J

A. Erhard, J. J. Wallman, L. Postler, M. Meth, R. Stricker, E. A. Martinez, P. Schindler, T. Monz, J. Emerson, and R. Blatt, Characterizing large-scale quantum computers via cycle benchmarking, Nature Communications10, 5347 (2019)

work page 2019
[16]

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

work page 2016
[17]

Winick, J

A. Winick, J. J. Wallman, D. Dahlen, I. Hincks, E. Ospadov, and J. Emerson, Concepts and conditions for error suppression through randomized compiling (2022), arXiv:2212.07500 [quant-ph]

work page arXiv 2022
[18]

Hashim, R

A. Hashim, R. K. Naik, A. Morvan, J.-L. Ville, B. Mitchell, J. M. Kreikebaum, M. Davis, E. Smith, C. Iancu, K. P. O’Brien, I. Hincks, J. J. Wallman, J. Emerson, and I. Siddiqi, Randomized Compiling for Scalable Quantum Computing on a Noisy Superconduct- ing Quantum Processor, Physical Review X11, 041039 (2021)

work page 2021
[19]

Wallman, C

J. Wallman, C. Granade, R. Harper, and S. T. Flam- mia, Estimating the coherence of noise, New Journal of Physics17, 113020 (2015)

work page 2015
[20]

Y. R. Sanders, J. J. Wallman, and B. C. Sanders, Bound- ing quantum gate error rate based on reported average fidelity, New Journal of Physics18, 012002 (2015), pub- lisher: IOP Publishing

work page 2015
[21]

Brillant, P

A. Brillant, P. Groszkowski, A. Seif, J. Koch, and A. A. Clerk, Randomized Benchmarking with Non-Markovian Noise and Realistic Finite-Time Gates, Physical Review Letters135, 070601 (2025)

work page 2025
[22]

Kattem¨ olle and G

J. Kattem¨ olle and G. Burkard, Ability of error correla- tions to improve the performance of variational quantum algorithms, Physical Review A107, 042426 (2023)

work page 2023
[23]

Z. Liu, Y. Xiao, and Z. Cai, Non-Markovian Noise Suppression Simplified through Channel Representation (2024), arXiv:2412.11220 [quant-ph]

work page arXiv 2024
[24]

Groszkowski, A

P. Groszkowski, A. Seif, J. Koch, and A. A. Clerk, Simple master equations for describing driven systems subject to classical non-Markovian noise, Quantum7, 972 (2023), arXiv:2207.03980 [quant-ph]

work page arXiv 2023
[25]

D. A. Lidar, Lecture Notes on the Theory of Open Quan- tum Systems (2020), arXiv:1902.00967 [quant-ph]

work page arXiv 2020
[26]

R. F. Fox, Critique of the generalized cumulant expan- sion method, Journal of Mathematical Physics17, 1148 (1976)

work page 1976
[27]

J. M. Chow, J. M. Gambetta, A. D. C´ orcoles, S. T. Merkel, J. A. Smolin, C. Rigetti, S. Poletto, G. A. Keefe, M. B. Rothwell, J. R. Rozen, M. B. Ketchen, and M. Stef- fen, Universal Quantum Gate Set Approaching Fault- Tolerant Thresholds with Superconducting Qubits, Phys- ical Review Letters109, 060501 (2012)

work page 2012
[28]

S. T. Flammia and J. J. Wallman, Efficient Estimation of Pauli Channels, ACM Transactions on Quantum Com- puting1, 3:1 (2020)

work page 2020
[29]

Viola, E

L. Viola, E. Knill, and S. Lloyd, Dynamical Decoupling of Open Quantum Systems, Physical Review Letters82, 2417 (1999)

work page 1999
[30]

This means that our starting point is an interaction pic- ture with respect to the bath only Hamiltonian, defined by ˆUB(t) =e −i ˆHB t

work page
[31]

A. A. Clerk, M. H. Devoret, S. M. Girvin, F. Mar- quardt, and R. J. Schoelkopf, Introduction to quantum noise, measurement, and amplification, Reviews of Mod- ern Physics82, 1155 (2010)

work page 2010
[32]

Merkel, T

S. Merkel, T. Proctor, S. Ferracin, J. Hines, S. Barron, L. C. G. Govia, and D. McKay, When Clifford bench- marks are sufficient; estimating application performance with scalable proxy circuits (2025), arXiv:2503.05943 [quant-ph]

work page arXiv 2025
[33]

A. W. Cross, A. Javadi-Abhari, T. Alexander, N. d. Beaudrap, L. S. Bishop, S. Heidel, C. A. Ryan, P. Sivara- jah, J. Smolin, J. M. Gambetta, and B. R. Johnson, OpenQASM 3: A broader and deeper quantum assembly language, ACM Transactions on Quantum Computing3, 1 (2022), arXiv:2104.14722 [quant-ph]

work page arXiv 2022
[34]

A. W. Cross, L. S. Bishop, J. A. Smolin, and J. M. Gambetta, Open Quantum Assembly Language (2017), arXiv:1707.03429 [quant-ph]

work page internal anchor Pith review Pith/arXiv arXiv 2017