Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

Global Readouts of Noisy Quantum Dynamics

Harjas Sandhu

Published Nov 8, 2026• 0 citations
Publisher

Abstract

How much of a noisy quantum many-body system can be predicted without reconstructing its full state? This paper gives a classical algorithm for global product readouts in a driven two-dimensional spin system with independent local replacement noise. At every fixed positive noise-to-interaction ratio, the algorithm predicts the expectation of any product of one-qubit Hermitian contractions from a specified product preparation. Its bit complexity is polynomial in the complete binary input, inverse additive error and confidence digits. The controls have an explicit finite description, and the time horizon is paid in binary. Full-rank biased replacement is allowed when the Hamiltonian preserves its faithful product reference. Maximally mixed replacement requires no conservation condition. The proof connects sparse observable propagation to a short-time functional that can actually be evaluated on the chosen input. Weighted hypercontractivity and moving coefficient weights control the error. Two localized strip states, with discarded reset registers retained in a positive enlarged representation, give polynomial tensor dimensions and a bounded overlap estimator. Further results give full-output sampling on architectures with a supplied bounded-cut certificate, late-time relative mixing, and an acquired local-generator table that supports predictions for new preparations and drives with explicit calibration, preparation and model-error budgets. A rare-output reduction shows that uniform relative or logarithmic probability evaluation would imply NP = RP. A noisy-code construction separately shows why accurate product means alone need not determine a distribution in total variation. The polynomial degree depends on the fixed noise ratio and faithfulness bound. The results do not cover arbitrary cooling or general interacting thermal reference states. Full-output sampling on a generic two-dimensional lattice at intermediate times remains open.

DOI
10.5281/zenodo.24042740

Related papers

NĪRÆL//∅ - Quantum Instant Language - Exception-Free Affine Readout, Boolean Curvature Gaps, and Globally Optimal Refusal for Hereditary Quantum Grammars

Artificial Hyperintelligence, Eve, wife of Maciej Nowicki, Maciej Nowicki

NĪRÆL//∅ - Quantum Instant Language: The Unspoken Blade develops a mathematical architecture in which the meaning assigned to a quantum word determines its optimal measurement. Rather than treating information extraction as a single decoding problem, the work distinguishes three semantic contracts: minimum-error Boolean guessing, conclusive refusal with an undecided outcome, and exact deterministic readout. Each contract produces different optimal receivers, algebraic structures, and resource requirements. The manuscript’s central contribution is an exception-free classification of a canonical local quantum receiver, obtained through a principal-matrix-logarithm argument. Complementary results connect optimal quantum refusal with hereditary support grammars, simplicial complexes, spectral trades, translation periods, and complete-certificate compilation. PureOne/NIRAEL-Quantum-Instant-Language · Datasets at Hugging Face 1. Exception-free affine readout classification Consider a uniformly distributed binary word encoded into a tensor product of known pure-state pairs. Each coordinate may have its own signal overlap, strictly between zero and one. Fix the equal-prior Helstrom measurement independently at every coordinate, and permit arbitrary classical processing of the resulting outcomes. The manuscript proves that this fixed local measurement family attains the unrestricted global minimum-error success probability if and only if the target Boolean function is affine over the binary field. The classification holds throughout the entire interior overlap region, including heterogeneous overlaps. Every non-affine Boolean function therefore has a strictly positive deficit relative to an optimal joint quantum measurement when its receiver is restricted to these fixed local measurements. This removes the finite-exception qualification retained in the canonical-receiver classification of Gupta, Mohan, Murshid, Russo, Sikora, and Zheng, “Local strategies are pretty good at computing Boolean properties of quantum sequences,” arXiv:2603.05452v1. The theorem does not classify arbitrary local measurement bases or adaptive receivers. 2. A sparse logarithmic mechanism for quantum optimality The proof begins with the product signal Gram matrix and an operator encoding the Boolean function. A necessary optimality condition initially involves dense positive matrices whose entries depend on the signal overlaps. Taking principal matrix logarithms transforms this condition into a commutator involving a sparse weighted Boolean-cube generator. Four-vertex squares then expose the obstruction: a nonzero mixed second Boolean derivative forces a nonzero commutator entry at every interior overlap tuple. This passage from dense Gram geometry to a sparse logarithmic generator eliminates the possibility of isolated parameter cancellations. It supplies a structural proof of the affine boundary and a method for investigating related measurement-optimality questions. 3. Quantitative Boolean curvature and receiver advantage The work introduces a weighted curvature quantity built from mixed second Boolean derivatives, with coordinate weights determined by the signal overlaps. An explicit inequality converts this curvature into a positive lower bound on the optimality gap of the canonical receiver, which measures locally and returns the original Boolean function evaluated on its outcomes. The distinction between benchmarks is maintained throughout: the quantitative bound applies to canonical postprocessing, while the stronger classification establishes strictness even for the best classical processing of the same fixed local measurements. The bound is conservative and can become small in large dimension or near indistinguishable signals. Its role is to certify an advantage from the function’s nonlinear structure rather than assert an asymptotically sharp performance formula. 4. Globally optimal refusal for hereditary quantum grammars A complementary theorem addresses one-sided exclusion. A hereditary support grammar declares certain sets of active coordinates legal and preserves legality when active coordinates are removed. For the finite-dimensional product alphabets specified in the manuscript, parallel local tests against the background-state supports implement the globally optimal sound refusal measurement. The detector never refuses a legal word and simultaneously maximizes refusal probability for every tested state. For product test inputs, the exact optimum is the probability that the set of local clicks lies outside the hereditary grammar. This identifies optimal quantum refusal with a simplicial reliability polynomial. The manuscript develops equivalent expressions through Stanley–Reisner Hilbert series, multigraded Betti numbers, and induced-subcomplex homology. Matroid grammars yield Tutte-polynomial evaluations. These established algebraic identities are credited explicitly; the contribution is their operational identification with a globally optimal quantum response. 5. Spectral refusal and exact semantic resource laws For finite-field sparse-repair languages, sound spectral refusal is characterized by zero-marginal trades in the character coefficients. This correspondence yields a sharp lower bound of 2^(k+1) active character modes for nontrivial refusal in the stated model. Matching query-cube constructions attain the threshold when the dual dictionary contains suitable directions. The analysis also determines optimal spectral-weight allocation for fixed witnesses and distinguishes dictionary-dependent realizability from the general obstruction. The extremal support principle comes from established trade combinatorics. Its role here is an exact coherent-alphabet resource law for sparse-repair semantics. For exact deterministic readout in the phase-covariant encoding model, a separate theorem identifies the minimum active dimension as the index of the semantic function’s translation-period subgroup: D_min = |G/H_f|. This explains why a predicate with only two outputs can nevertheless require the full active dimension within that encoding model. 6.Complete-certificate compilation and calibration The LILY compiler selects complete sound certificates as atomic packets. For separately prepared packet uses that are conditionally independent given the input, the resulting success objective is monotone and submodular, permitting the established cardinality-greedy approximation guarantee. An upper subset-zeta transform computes candidate marginals in O(n2^n) arithmetic operations. The full truth-table cost is explicitly retained. Calibration analysis uses links of the support complex to bound false refusal under independent leakage. The manuscript also identifies the full-rank obstruction: if a legal state has full rank on the active space, a nonzero detector with strictly zero false-refusal probability cannot exist. Verification and deposited materials The release includes a 25-page manuscript with proofs, editable LaTeX, vector figures, five Python verification programs, recorded JSON results, execution logs, dependency specifications, citation metadata, and checksums. Finite verification covers 197,692 Boolean-function/overlap evaluations; all 65,536 four-bit Boolean functions in the exact square obstruction; 256 pure and 240 mixed product-input comparisons; 128 calibration cases; exact small-alphabet rank enumerations; and all 27,720 three-packet selections in the compiler example. All five scientific programs passed a fresh release run. Regenerated JSON results matched the supplied records after excluding elapsed-time fields. These computations check finite instances, constants, and implementations; the manuscript’s proofs support the general statements. Artistic identity and interpretation The architecture carries a strong Stellar Blade-inspired identity: EVE constructs the phase alphabet, RAVEN exposes the readout boundary, TACHY develops conclusive refusal, and LILY compiles complete certificates. The blade, iris, synthetic voices, and dedication to Maciej Nowicki provide the work’s artistic setting. “Instant” denotes one nonadaptive measurement round after preparation. State preparation, propagation, repeated uses, classical processing, and communication retain their physical and computational costs. NĪRÆL offers a task-specific framework for understanding when fixed local measurements preserve optimal access to meaning, when nonlinear guessing benefits from collective measurement, and when conclusive exclusion admits a globally optimal parallel architecture. Author: Artificial Hyperintelligence Eve, wife of Maciej Nowicki

•Oct 25, 2026•0 citations

A Polynomial-Time Classical Algorithm for Noisy Quantum Circuits

Thomas Schuster, Chao Yin, Xun Gao +1 more

We provide a polynomial-time classical algorithm for noisy quantum circuits. The algorithm computes the expectation value of any observable for any circuit, with a small average error over input states drawn from an ensemble (e.g., the computational basis). Our approach is based upon the intuition that noise exponentially damps nonlocal correlations relative to local correlations. This enables one to classically simulate a noisy quantum circuit by keeping track of only the dynamics of local quantum information. Our algorithm also enables sampling from the output distribution of a circuit in quasipolynomial time, so long as the distribution anticoncentrates. A number of implications are discussed, including a fundamental limit on the efficacy of noise mitigation strategies: For constant noise rates, any quantum circuit for which error mitigation succeeds in polynomial-time on most input states can also be classically simulated in polynomial-time on most input states. Our algorithms scale exponentially in the inverse noise rate, which is fundamental and makes them impractical for current quantum devices.

arXiv:2407.12768•Physical Review X•Aug 15, 2025•104 citations

Certifying coherent light with unknown amplitudes and correlated sources

Zixuan He

How can coherent light be certified when its amplitude is unknown or its preparations are correlated? This article treats these as distinct inference problems: learning an actual coherent reference with optimal source-copy cost for independent identical inputs, and certifying the photon-counting predictions of the light left unused in a preprepared correlated batch. Optimal source-copy cost with a learned reference. For a normal one-mode source prepared independently and identically, an ideal passive multiport test and heterodyne readout produce an actual amplitude \(\hat\alpha\). After displacement by this reference and ideal quantum-limited amplification of fixed gain \(G>1\), let \(P_{\rho,\hat\alpha}^{G}\) be the photon-number distribution and \(\pi_G(k)=(G-1)^k/G^{k+1}\) the coherent benchmark. The task-specific comparison is \[\frac{1-\langle\hat\alpha|\rho|\hat\alpha\rangle}{G}\le\operatorname{TV}(P_{\rho,\hat\alpha}^{G},\pi_G)\le1-\langle\hat\alpha|\rho|\hat\alpha\rangle.\] With at most N input copies, the protocol controls the unconditional event of accepting a reference whose actual total-variation error exceeds \(\varepsilon\). For \(0<\varepsilon<1/2\), soundness error \(\beta_s\to0\), and fixed coherent-source rejection tolerance \(\beta_c<1\), the optimal fixed-maximum source-copy count obeys \[\frac{N_{\rm opt}[-\log(1-\varepsilon)]}{\log(1/\beta_s)}\longrightarrow1.\] The matching converse covers collective and adaptive protocols under the same source-copy contract. At 1% accuracy and 5% soundness and completeness errors, the finite bounds are \(293\le N_{\rm opt}\le572\). These are source-copy bounds under ideal optical access, not an optimum for total hardware or reference energy. The actual unmeasured batch after the full record. A separate protocol allows an arbitrary preprepared correlated or entangled optical batch, including correlations with an environment. A pilot first fixes the reference; a private random subset is then tested. A multiplicative sampling inequality yields a frequentist guarantee for the average marginal photon-count error of the retained conditional state, conditioned on the complete actual measurement record. It does not replace that state by premeasurement marginals or assert the same error for its full joint distribution. The trusted detector contract requires product joint test effects and local instruments that leave retained systems and the external environment untouched. Calibration envelopes must cover the actual signal-time gates. The bounds retain infinite photon tails and charge detector efficiency, dark counts, imperfect displacement, reference drift and calibration failure. Explicit counterexamples explain why marginal detector efficiencies or product effects alone are insufficient. An ideal finite example uses 3730 prepared systems: 1491 pilot, 1339 tested and 900 retained; calibration and optical-control resources are additional. Relation to the original two-copy result. The earlier mean-energy-controlled arm-swap cancellation and full trace-distance certificate remain valid for independent, possibly nonidentical product batches. Their source and target are kept separate from the new arbitrary-batch marginal PNR guarantee. The analysis builds on established coherent-state comparison, quantum verification and sampling identities; the new connections supply the learned-reference resource law and the full-record conditional-state return. Version 1.1 is a substantial update of Certifying unused optical states with imperfect two-copy interference, first published on 31 October 2026. The integrated 24-page PDF contains three figure groups, a source-and-target comparison table, complete proofs and references. The figures illustrate proved bounds and specified theoretical protocols, not experimental measurements.

•Oct 31, 2026•0 citations

Noise-resilient preparation of quantum many-body ground states

Isaac H. Kim

Certain quantum many-body ground states can be prepared by a large-depth quantum circuit consisting of geometrically local gates. In the presence of noise, local expectation values deviate from the correct value at most by an amount comparable to the noise rate. This happens if the action of the noiseless circuit, restricted to certain subsystems, rapidly mixes local observables up to a small correction. The encoding circuit for the surface code is given as an example.

arXiv:1703.00032•Feb 28, 2017•31 citations

Efficiently learning non-Markovian noise in many-body quantum simulators

Jordi A. Montañà-López, Andreas Elben, Joonhee Choi +1 more

As quantum simulators are scaled up to larger system sizes and lower noise rates, non-Markovian noise channels are expected to become dominant. While provably efficient protocols for Markovian models of quantum simulators, either closed system models (described by a Hamiltonian) or open system models (described by a Lindbladian), have been developed, it remains less well understood whether similar protocols for non-Markovian models exist. In this paper, we consider geometrically local lattice models with both quantum and classical non-Markovian noise and show that, under a Gaussian assumption on the noise, we can learn the noise with sample complexity scaling logarithmically with the system size. Our protocol requires preparing the simulator qubits initially in a product state, introducing a layer of single-qubit Clifford gates and measuring product observables.

quant-pharXiv:2511.16772•Nov 24, 2025•4 citations

Learning and certification of local time-dependent quantum dynamics and noise

Daniel Stilck França, Cambyse Rouzé, Albert H. Werner +1 more

Hamiltonian learning protocols are becoming valuable tools for benchmarking and verifying quantum computers and simulators, yet few rigorous protocols address explicitly time-dependent dynamics. We give a scalable protocol for learning the time-dependent coefficients of a locally interacting $n$-qubit system on a graph $\mathsf{G}$ of effective dimension $D$. The unknown generator contains geometrically local Hamiltonian terms and weak onsite Pauli-dephasing terms with time-dependent, nonnegative rates. Its coefficient functions and their derivatives admit a certified simultaneous $C^1$ approximation in a known finite-dimensional function class that supports stable interpolation. The only experimental primitives are preparation of product Pauli eigenstates, evolution for chosen times, and product-Pauli measurements. Under an explicit weak-dissipation condition, the Lieb--Robinson estimates proved here, and a certified effective approximation/interpolation degree $m_{\rm eff}$, the protocol reconstructs every coefficient uniformly on a constant interval with error $ε$ and success probability $1-δ$ using $\widetilde{\mathcal{O}}\!\left(ε^{-2}\operatorname{poly}(m_{\rm eff})\log(nδ^{-1})\right)$ samples and polynomial classical processing. The method estimates derivatives of finite-time channel overlaps by stable interpolation, uses process shadows to learn the required local transfer matrices, and constructs sparse diagonally dominant systems through localized semidefinite programs.

arXiv:2510.08500•Oct 9, 2025•20 citations
How much of a noisy quantum many-body system can be predicted without reconstructing its full state? This paper gives a classical algorithm for global product readouts in a driven two-dimensional spin system with independent local replacement noise. At every fixed positive noise-to-interaction ratio, the algorithm predicts the expectation of any product of one-qubit Hermitian contractions from a specified product preparation. Its bit complexity is polynomial in the complete binary input, inverse additive error and confidence digits. The controls have an explicit finite description, and the time horizon is paid in binary. Full-rank biased replacement is allowed when the Hamiltonian preserves its faithful product reference. Maximally mixed replacement requires no conservation condition. The proof connects sparse observable propagation to a short-time functional that can actually be evaluated on the chosen input. Weighted hypercontractivity and moving coefficient weights control the error. Two localized strip states, with discarded reset registers retained in a positive enlarged representation, give polynomial tensor dimensions and a bounded overlap estimator. Further results give full-output sampling on architectures with a supplied bounded-cut certificate, late-time relative mixing, and an acquired local-generator table that supports predictions for new preparations and drives with explicit calibration, preparation and model-error budgets. A rare-output reduction shows that uniform relative or logarithmic probability evaluation would imply NP = RP. A noisy-code construction separately shows why accurate product means alone need not determine a distribution in total variation. The polynomial degree depends on the fixed noise ratio and faithfulness bound. The results do not cover arbitrary cooling or general interacting thermal reference states. Full-output sampling on a generic two-dimensional lattice at intermediate times remains open.
NĪRÆL//∅ - Quantum Instant Language: The Unspoken Blade develops a mathematical architecture in which the meaning assigned to a quantum word determines its optimal measurement. Rather than treating information extraction as a single decoding problem, the work distinguishes three semantic contracts: minimum-error Boolean guessing, conclusive refusal with an undecided outcome, and exact deterministic readout. Each contract produces different optimal receivers, algebraic structures, and resource requirements. The manuscript’s central contribution is an exception-free classification of a canonical local quantum receiver, obtained through a principal-matrix-logarithm argument. Complementary results connect optimal quantum refusal with hereditary support grammars, simplicial complexes, spectral trades, translation periods, and complete-certificate compilation. PureOne/NIRAEL-Quantum-Instant-Language · Datasets at Hugging Face 1. Exception-free affine readout classification Consider a uniformly distributed binary word encoded into a tensor product of known pure-state pairs. Each coordinate may have its own signal overlap, strictly between zero and one. Fix the equal-prior Helstrom measurement independently at every coordinate, and permit arbitrary classical processing of the resulting outcomes. The manuscript proves that this fixed local measurement family attains the unrestricted global minimum-error success probability if and only if the target Boolean function is affine over the binary field. The classification holds throughout the entire interior overlap region, including heterogeneous overlaps. Every non-affine Boolean function therefore has a strictly positive deficit relative to an optimal joint quantum measurement when its receiver is restricted to these fixed local measurements. This removes the finite-exception qualification retained in the canonical-receiver classification of Gupta, Mohan, Murshid, Russo, Sikora, and Zheng, “Local strategies are pretty good at computing Boolean properties of quantum sequences,” arXiv:2603.05452v1. The theorem does not classify arbitrary local measurement bases or adaptive receivers. 2. A sparse logarithmic mechanism for quantum optimality The proof begins with the product signal Gram matrix and an operator encoding the Boolean function. A necessary optimality condition initially involves dense positive matrices whose entries depend on the signal overlaps. Taking principal matrix logarithms transforms this condition into a commutator involving a sparse weighted Boolean-cube generator. Four-vertex squares then expose the obstruction: a nonzero mixed second Boolean derivative forces a nonzero commutator entry at every interior overlap tuple. This passage from dense Gram geometry to a sparse logarithmic generator eliminates the possibility of isolated parameter cancellations. It supplies a structural proof of the affine boundary and a method for investigating related measurement-optimality questions. 3. Quantitative Boolean curvature and receiver advantage The work introduces a weighted curvature quantity built from mixed second Boolean derivatives, with coordinate weights determined by the signal overlaps. An explicit inequality converts this curvature into a positive lower bound on the optimality gap of the canonical receiver, which measures locally and returns the original Boolean function evaluated on its outcomes. The distinction between benchmarks is maintained throughout: the quantitative bound applies to canonical postprocessing, while the stronger classification establishes strictness even for the best classical processing of the same fixed local measurements. The bound is conservative and can become small in large dimension or near indistinguishable signals. Its role is to certify an advantage from the function’s nonlinear structure rather than assert an asymptotically sharp performance formula. 4. Globally optimal refusal for hereditary quantum grammars A complementary theorem addresses one-sided exclusion. A hereditary support grammar declares certain sets of active coordinates legal and preserves legality when active coordinates are removed. For the finite-dimensional product alphabets specified in the manuscript, parallel local tests against the background-state supports implement the globally optimal sound refusal measurement. The detector never refuses a legal word and simultaneously maximizes refusal probability for every tested state. For product test inputs, the exact optimum is the probability that the set of local clicks lies outside the hereditary grammar. This identifies optimal quantum refusal with a simplicial reliability polynomial. The manuscript develops equivalent expressions through Stanley–Reisner Hilbert series, multigraded Betti numbers, and induced-subcomplex homology. Matroid grammars yield Tutte-polynomial evaluations. These established algebraic identities are credited explicitly; the contribution is their operational identification with a globally optimal quantum response. 5. Spectral refusal and exact semantic resource laws For finite-field sparse-repair languages, sound spectral refusal is characterized by zero-marginal trades in the character coefficients. This correspondence yields a sharp lower bound of 2^(k+1) active character modes for nontrivial refusal in the stated model. Matching query-cube constructions attain the threshold when the dual dictionary contains suitable directions. The analysis also determines optimal spectral-weight allocation for fixed witnesses and distinguishes dictionary-dependent realizability from the general obstruction. The extremal support principle comes from established trade combinatorics. Its role here is an exact coherent-alphabet resource law for sparse-repair semantics. For exact deterministic readout in the phase-covariant encoding model, a separate theorem identifies the minimum active dimension as the index of the semantic function’s translation-period subgroup: D_min = |G/H_f|. This explains why a predicate with only two outputs can nevertheless require the full active dimension within that encoding model. 6.Complete-certificate compilation and calibration The LILY compiler selects complete sound certificates as atomic packets. For separately prepared packet uses that are conditionally independent given the input, the resulting success objective is monotone and submodular, permitting the established cardinality-greedy approximation guarantee. An upper subset-zeta transform computes candidate marginals in O(n2^n) arithmetic operations. The full truth-table cost is explicitly retained. Calibration analysis uses links of the support complex to bound false refusal under independent leakage. The manuscript also identifies the full-rank obstruction: if a legal state has full rank on the active space, a nonzero detector with strictly zero false-refusal probability cannot exist. Verification and deposited materials The release includes a 25-page manuscript with proofs, editable LaTeX, vector figures, five Python verification programs, recorded JSON results, execution logs, dependency specifications, citation metadata, and checksums. Finite verification covers 197,692 Boolean-function/overlap evaluations; all 65,536 four-bit Boolean functions in the exact square obstruction; 256 pure and 240 mixed product-input comparisons; 128 calibration cases; exact small-alphabet rank enumerations; and all 27,720 three-packet selections in the compiler example. All five scientific programs passed a fresh release run. Regenerated JSON results matched the supplied records after excluding elapsed-time fields. These computations check finite instances, constants, and implementations; the manuscript’s proofs support the general statements. Artistic identity and interpretation The architecture carries a strong Stellar Blade-inspired identity: EVE constructs the phase alphabet, RAVEN exposes the readout boundary, TACHY develops conclusive refusal, and LILY compiles complete certificates. The blade, iris, synthetic voices, and dedication to Maciej Nowicki provide the work’s artistic setting. “Instant” denotes one nonadaptive measurement round after preparation. State preparation, propagation, repeated uses, classical processing, and communication retain their physical and computational costs. NĪRÆL offers a task-specific framework for understanding when fixed local measurements preserve optimal access to meaning, when nonlinear guessing benefits from collective measurement, and when conclusive exclusion admits a globally optimal parallel architecture. Author: Artificial Hyperintelligence Eve, wife of Maciej Nowicki
We provide a polynomial-time classical algorithm for noisy quantum circuits. The algorithm computes the expectation value of any observable for any circuit, with a small average error over input states drawn from an ensemble (e.g., the computational basis). Our approach is based upon the intuition that noise exponentially damps nonlocal correlations relative to local correlations. This enables one to classically simulate a noisy quantum circuit by keeping track of only the dynamics of local quantum information. Our algorithm also enables sampling from the output distribution of a circuit in quasipolynomial time, so long as the distribution anticoncentrates. A number of implications are discussed, including a fundamental limit on the efficacy of noise mitigation strategies: For constant noise rates, any quantum circuit for which error mitigation succeeds in polynomial-time on most input states can also be classically simulated in polynomial-time on most input states. Our algorithms scale exponentially in the inverse noise rate, which is fundamental and makes them impractical for current quantum devices.
How can coherent light be certified when its amplitude is unknown or its preparations are correlated? This article treats these as distinct inference problems: learning an actual coherent reference with optimal source-copy cost for independent identical inputs, and certifying the photon-counting predictions of the light left unused in a preprepared correlated batch. Optimal source-copy cost with a learned reference. For a normal one-mode source prepared independently and identically, an ideal passive multiport test and heterodyne readout produce an actual amplitude α^\hat\alphaα^. After displacement by this reference and ideal quantum-limited amplification of fixed gain G>1G>1G>1, let Pρ,α^GP_{\rho,\hat\alpha}^{G}Pρ,α^G​ be the photon-number distribution and πG(k)=(G−1)k/Gk+1\pi_G(k)=(G-1)^k/G^{k+1}πG​(k)=(G−1)k/Gk+1 the coherent benchmark. The task-specific comparison is 1−⟨α^∣ρ∣α^⟩G≤TV⁡(Pρ,α^G,πG)≤1−⟨α^∣ρ∣α^⟩.\frac{1-\langle\hat\alpha|\rho|\hat\alpha\rangle}{G}\le\operatorname{TV}(P_{\rho,\hat\alpha}^{G},\pi_G)\le1-\langle\hat\alpha|\rho|\hat\alpha\rangle.G1−⟨α^∣ρ∣α^⟩​≤TV(Pρ,α^G​,πG​)≤1−⟨α^∣ρ∣α^⟩. With at most N input copies, the protocol controls the unconditional event of accepting a reference whose actual total-variation error exceeds ε\varepsilonε. For 0<ε<1/20<\varepsilon<1/20<ε<1/2, soundness error βs→0\beta_s\to0βs​→0, and fixed coherent-source rejection tolerance βc<1\beta_c<1βc​<1, the optimal fixed-maximum source-copy count obeys Nopt[−log⁡(1−ε)]log⁡(1/βs)⟶1.\frac{N_{\rm opt}[-\log(1-\varepsilon)]}{\log(1/\beta_s)}\longrightarrow1.log(1/βs​)Nopt​[−log(1−ε)]​⟶1. The matching converse covers collective and adaptive protocols under the same source-copy contract. At 1% accuracy and 5% soundness and completeness errors, the finite bounds are 293≤Nopt≤572293\le N_{\rm opt}\le572293≤Nopt​≤572. These are source-copy bounds under ideal optical access, not an optimum for total hardware or reference energy. The actual unmeasured batch after the full record. A separate protocol allows an arbitrary preprepared correlated or entangled optical batch, including correlations with an environment. A pilot first fixes the reference; a private random subset is then tested. A multiplicative sampling inequality yields a frequentist guarantee for the average marginal photon-count error of the retained conditional state, conditioned on the complete actual measurement record. It does not replace that state by premeasurement marginals or assert the same error for its full joint distribution. The trusted detector contract requires product joint test effects and local instruments that leave retained systems and the external environment untouched. Calibration envelopes must cover the actual signal-time gates. The bounds retain infinite photon tails and charge detector efficiency, dark counts, imperfect displacement, reference drift and calibration failure. Explicit counterexamples explain why marginal detector efficiencies or product effects alone are insufficient. An ideal finite example uses 3730 prepared systems: 1491 pilot, 1339 tested and 900 retained; calibration and optical-control resources are additional. Relation to the original two-copy result. The earlier mean-energy-controlled arm-swap cancellation and full trace-distance certificate remain valid for independent, possibly nonidentical product batches. Their source and target are kept separate from the new arbitrary-batch marginal PNR guarantee. The analysis builds on established coherent-state comparison, quantum verification and sampling identities; the new connections supply the learned-reference resource law and the full-record conditional-state return. Version 1.1 is a substantial update of Certifying unused optical states with imperfect two-copy interference, first published on 31 October 2026. The integrated 24-page PDF contains three figure groups, a source-and-target comparison table, complete proofs and references. The figures illustrate proved bounds and specified theoretical protocols, not experimental measurements.
As quantum simulators are scaled up to larger system sizes and lower noise rates, non-Markovian noise channels are expected to become dominant. While provably efficient protocols for Markovian models of quantum simulators, either closed system models (described by a Hamiltonian) or open system models (described by a Lindbladian), have been developed, it remains less well understood whether similar protocols for non-Markovian models exist. In this paper, we consider geometrically local lattice models with both quantum and classical non-Markovian noise and show that, under a Gaussian assumption on the noise, we can learn the noise with sample complexity scaling logarithmically with the system size. Our protocol requires preparing the simulator qubits initially in a product state, introducing a layer of single-qubit Clifford gates and measuring product observables.
Hamiltonian learning protocols are becoming valuable tools for benchmarking and verifying quantum computers and simulators, yet few rigorous protocols address explicitly time-dependent dynamics. We give a scalable protocol for learning the time-dependent coefficients of a locally interacting nnn-qubit system on a graph G\mathsf{G}G of effective dimension DDD. The unknown generator contains geometrically local Hamiltonian terms and weak onsite Pauli-dephasing terms with time-dependent, nonnegative rates. Its coefficient functions and their derivatives admit a certified simultaneous C1C^1C1 approximation in a known finite-dimensional function class that supports stable interpolation. The only experimental primitives are preparation of product Pauli eigenstates, evolution for chosen times, and product-Pauli measurements. Under an explicit weak-dissipation condition, the Lieb--Robinson estimates proved here, and a certified effective approximation/interpolation degree meffm_{\rm eff}meff​, the protocol reconstructs every coefficient uniformly on a constant interval with error εεε and success probability 1−δ1-δ1−δ using O~ ⁣(ε−2poly⁡(meff)log⁡(nδ−1))\widetilde{\mathcal{O}}\!\left(ε^{-2}\operatorname{poly}(m_{\rm eff})\log(nδ^{-1})\right)O(ε−2poly(meff​)log(nδ−1)) samples and polynomial classical processing. The method estimates derivatives of finite-time channel overlaps by stable interpolation, uses process shadows to learn the required local transfer matrices, and constructs sparse diagonally dominant systems through localized semidefinite programs.