{"id":"cd1aa067-fb98-400e-bcf7-ea5f62f4b2a8","arxiv_id":"2511.12416","paper_version":3,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"An open-source simulator computes Born-rule probabilities for matchgate-plus-controlled-phase chemistry circuits and uses them to improve sample-based quantum diagonalization.","lead":"ExtraFerm is an open-source quantum-circuit simulator that computes the probability of specific measurement outcomes for chemistry-style circuits made of matchgates plus a small number of controlled-phase gates. It uses far less memory than full state-vector simulators and, when used to 'warm-start' sample-based quantum diagonalization, improves molecular energy estimates on real hardware with almost no extra runtime.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Arbitrary-pair controlled-phase gates may not admit the claimed matchgate trajectories; Eq. (7)'s d0/d1 are only free-fermionic for nearest-neighbor pairs, so the determinant formula (3) may not apply.","rationale":"The reader's weakest-assumption analysis focused on the Monte-Carlo trajectory-count bound in Eqs. (5)-(6). That is a reasonable concern, but those bounds are imported from prior work and are likely correct. A more load-bearing issue is whether the trajectory decomposition itself yields the correct amplitude when controlled-phase gates are not nearest-neighbor. The paper's stated scope includes arbitrary qubit pairs, yet the determinant formula (Eq. 3) is only established for nearest-neighbor matchgate circuits. If d0/d1 are non-adjacent, they are not free-fermionic gates in the JW picture, so the V(x) construction may represent a different unitary. This would invalidate the exact and approximate probability claims for the benchmark circuits with random pairs and potentially for the LUCJ applications if those also use non-adjacent controlled-phase gates. The proposed statevector test directly settles the issue. Because the paper otherwise contains independent support (open-source code, comparisons, plausible SQD results) but this scope question is unresolved, the appropriate verdict is CONDITIONAL: the central claim should be accepted only after either proving the arbitrary-pair decomposition or explicitly restricting the simulator to nearest-neighbor controlled-phase gates. This is consistent with the reader's overall conditional grade, though for a different reason.","tokens_in":18723,"tokens_out":29030,"duration_ms":275112,"concrete_test":"Run ExtraFerm's EXACT on a 4-qubit circuit containing a single controlled-phase gate acting on qubits 0 and 2 (non-adjacent), with no other gates, for input/output bitstrings of the same Hamming weight and for several angles (e.g., θ=π/3 and θ=π). Compare the resulting probabilities against exact state-vector simulation (e.g., Qiskit's statevector simulator) of the same circuit. If the probabilities disagree, the arbitrary-pair claim is false and the manuscript must restrict the scope to nearest-neighbor controlled-phase gates; if they agree, the framework must be documented, since the proof in Eq. (3) does not cover this case.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central exactness claim hinges on replacing each controlled-phase gate by d0(θ) or d1(θ) and then evaluating ⟨b|V(x)|a⟩ as a determinant via Eq. (3). But Eq. (3) is justified only for circuits of nearest-neighbor particle-number-conserving matchgates. The paper explicitly allows controlled-phase gates on arbitrary qubit pairs (Fig. 3 caption; Section V-A says 'random pairs of qubits'). For non-adjacent pairs, the two-qubit operators d0 and d1 are not matchgates in the fixed Jordan-Wigner ordering; their action on modes q1,q2 is not a free-fermionic operation unless accompanying JW string operators are included. Constructing V(x) as a product of 2×2 blocks on modes q1,q2 therefore computes amplitudes of a different, generally nonlocal Gaussian circuit, not of the original controlled-phase circuit. The paper does not prove that arbitrary-pair controlled-phase gates can be decomposed into such free-fermionic trajectories, nor does it restrict the claimed scope to nearest-neighbor pairs. If this gap is real, both the EXACT and RAWESTIMATE probabilities used in the benchmarks and in the warm-start SQD results are not Born probabilities of the stated circuits, and the central claim fails for a large advertised input class. The reader's flagged Monte-Carlo bound from Refs. [19],[41] is a secondary concern; this is a more fundamental correctness question.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces ExtraFerm, a classical simulator for circuits composed of particle-number-conserving matchgates and controlled-phase gates. For a target bitstring, it computes the Born-rule probability exactly in time exponential in the number of controlled-phase gates, and approximately with time exponential in the circuit extent ξ* (Eq. 5). The method is based on a trajectory decomposition of each controlled-phase gate into two matchgates d0, d1 (Eq. 7), followed by a Monte Carlo sum over trajectories (Algorithm 1). The authors benchmark against tensor-network and state-vector simulators, reporting substantially better latency and memory scaling, and integrate the estimator into warm-start sample-based quantum diagonalization, obtaining improved ground-state energy estimates for H14 (28 qubits, classical noise model) and N2 (52 qubits, IBM Heron) with small runtime overhead.","tokens_in":19045,"tokens_out":24510,"duration_ms":190098,"significance":"If the algorithmic claims are correct, ExtraFerm fills a practical gap: it permits targeted Born-rule probability computation for LUCJ and related chemistry circuits, with cost governed by the number and angles of controlled-phase gates rather than by qubit number. The paper provides open-source code, numerical validation of the estimator against exact probabilities (Figs. 7–8), and a realistic end-to-end application (warm-start SQD). The underlying mathematical framework is adapted from peer-reviewed work [19,20,41], which strengthens confidence in the central formulation.","major_comments":[{"comment":"The paper states that controlled-phase gates may act on arbitrary qubit pairs, but the determinant formula (3) is introduced only for circuits of nearest-neighbor matchgates. The trajectory construction replaces each controlled-phase gate with d0 or d1 and evaluates ⟨b|V(x)|a⟩ as a determinant, which is justified only if the resulting circuit is free-fermionic. For non-adjacent pairs this is not automatic from the matchgate condition det(A)=det(B). The authors should explicitly show that d0 and d1 factor into single-qubit phase gates, e.g. d0 = e^{-iθ/2} e^{iθ n_q1/2} e^{iθ n_q2/2} and d1 = e^{-iθ/2} e^{i(θ/2+π)n_q1} e^{i(θ/2+π)n_q2}, so that each trajectory has a well-defined mode transformation matrix and Eq. (3) applies. Without this, the exactness claim for the advertised input class is not fully supported.","section":"III-B, Fig. 3 caption"},{"comment":"The normalization in line 12, p̂ ← ξ*/s² |α|², is incorrect. The variable s is the per-trajectory sign computed in line 9 and is not a meaningful global normalization. The correct expression is p̂ ← ξ* |α|² / t², where t is the number of trajectories. As printed, the algorithm does not return a probability estimate; this is a load-bearing error in the core subroutine. Please correct the pseudocode and ensure the code matches.","section":"Algorithm 1, line 12"},{"comment":"The ESTIMATE algorithm as written is not well-defined. The loop condition uses an undefined variable ϵ* (presumably e*), and e* is never updated inside the loop. Thus the loop either never terminates or does not implement the intended iterative tightening of the error and probability upper bound. The δ_k update also refers to an undefined δ_total. The authors should rewrite Algorithm 2 to match the iterative procedure from Ref. [41], including an explicit update rule for the error target e*.","section":"Algorithm 2"}],"minor_comments":[{"comment":"The phrase 'exponential only in the magnitudes of the circuit’s controlled-phase gate angles' is imprecise; the cost is exponential in the circuit extent ξ*, which involves a product over gates of (cos(|θ|/4)+sin(|θ|/4))². Please clarify.","section":"Abstract/Introduction"},{"comment":"The definition of matchgate in Eq. (1) does not state a nearest-neighbor requirement, yet Eq. (3) is declared for nearest-neighbor circuits. This creates confusion when d0,d1 are called 'matchgates' even for arbitrary-pair controlled-phase gates. A sentence distinguishing the matrix-form condition from the nearest-neighbor simulation theorem would help.","section":"Section II-B"},{"comment":"δ_total is not defined in the algorithm input; the global failure probability is δ. Please correct.","section":"Algorithm 2, line 4"},{"comment":"The listed circuit extents (e.g., 3.753 for θ=0) are for angles sampled from N(θ,0.1), not for exactly θ. The caption should say that θ is the mean of the sampling distribution.","section":"Fig. 7 caption"},{"comment":"Consider writing the extent explicitly as ξ* = ∏_j (cos(|θ_j|/4)+sin(|θ_j|/4))² to avoid any ambiguity about squaring the entire product.","section":"Eq. (5)"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid engineering contribution with a useful practical application, and the central mathematical framework is built on prior peer-reviewed results. The main issues are a missing justification for the arbitrary-pair case and consequential errors in the pseudocode. These are fixable in revision. I also recommend that the authors cross-check the printed algorithms against the released code, since the Algorithm 2 loop error suggests possible transcription mistakes."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe paper is a clean engineering contribution: it turns the Reardon-Smith/Diaz-Koenig trajectory framework into a usable simulator for particle-number-conserving matchgates plus controlled-phase gates, adds a genuinely useful optimization for LUCJ circuits, and demonstrates a warm-start variant of SQD that improves the N2 energy estimate on 52-qubit hardware data by up to 46% with negligible overhead. The authors are explicit that the core algorithm is inherited; the new material is the implementation, the LUCJ-specific V3V1 corrections, and the application.\n\nThe math holds up. I checked the worry that arbitrary-pair controlled-phase gates break the determinant formula because d0/d1 are only matchgates for nearest-neighbor pairs. That concern doesn't land. d0 and d1 factor into single-mode phase rotations and the parity operator, so they are Gaussian operations on any pair of modes. The mode transformation V(x) is still unitary, and the determinant formula applies to any non-interacting fermionic circuit. The empirical agreement with ffsim on random-pair circuits confirms this. So the central claim is sound.\n\nThe real soft spots are smaller. Algorithm 1 has a typo: line 12 divides by s^2 instead of t^2. Taken literally it breaks the estimator, though the intended formula is clear from the context. The N2 support-size calculation uses C(27,5) when the system has 26 spatial orbitals, so that number is off. The warm-start procedure never states how many high-probability bitstrings are selected, which hurts reproducibility. And the latency/memory benchmark lacks error bars. None of these are load-bearing, but they need fixing before publication.\n\nThe evidence is honest: error analysis against exact probabilities, comparison to ffsim, and a real-hardware dataset. The paper would benefit from a serious referee to catch the typos and expose the selection-count issue. I'd recommend sending it out.\n\nWho is this for? Anyone doing classical simulation of matchgate-dominated chemistry circuits or using SQD on noisy hardware will find it useful. It's not a theoretical breakthrough, but it's a solid tool.\n\nMy recommendation: engage with it; send to peer review with a request for clarifications.","headline":"Solid engineering with a real hardware demonstration; the central trajectory method is sound even for arbitrary-pair gates, but the paper has a few fixable typos and a reproducibility gap.","tokens_in":19580,"tokens_out":7616,"would_cite":true,"duration_ms":65344,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"ExtraFerm computes exact Born-rule probabilities for circuits of particle-number-conserving matchgates and controlled-phase gates in time exponential only in the number of controlled-phase gates, and approximates them in time exponential on","keywords":["matchgates","fermionic linear optics","Born-rule probability","circuit extent","sample-based quantum diagonalization","LUCJ ansatz","Monte Carlo simulation","quantum chemistry"],"falsifier":"Take a 12-qubit LUCJ circuit with controlled-phase angles including negative values near −π; compute exact probabilities of the full support with EXACT, then run RAWESTIMATE with t given by Eq. (6) for an additive error ϵ and failure probability δ (e.g., 0.01 and 0.01). If the empirical frequency of bitstrings where |p̂ − p_exact| > ϵ exceeds δ, the Monte Carlo error guarantee — and the warm-start benefits built on it — are falsified.","tokens_in":18600,"feed_emoji":"🧪","tokens_out":7833,"duration_ms":61545,"temperature":0.7,"pith_summary":"ExtraFerm is a new classical simulator for quantum circuits built from particle-number-conserving matchgates plus controlled-phase gates — exactly the structure that emerges when the LUCJ chemistry ansatz is mapped to qubits via the Jordan-Wigner transformation. The paper's central claim is that the Born-rule probability of any chosen bitstring can be computed exactly at a cost exponential only in the number of controlled-phase gates, and approximately at a cost exponential only in the circuit's extent, which shrinks when the controlled-phase angles are small. That makes high-accuracy probability estimates feasible for 52-qubit chemistry circuits where state-vector or tensor-network methods hit a memory wall. The authors prove the point by using ExtraFerm to pick the most probable bitstrings in the first iteration of sample-based quantum diagonalization, improving molecular ground-state energy estimates by up to 46% and reducing variance by up to 98% with a runtime overhead of at most 2%.","feed_headline":"Simulator's exact probability cost grows only with phase-gate count","feed_subtitle":"ExtraFerm makes LUCJ circuits tractable; warm-started SQD gains up to 46% accuracy with ≤2% overhead.","key_machinery":"The central object is the two-term matchgate decomposition of a controlled-phase gate, c(θ) = e^{iθ/4}[cos(θ/4)d0(θ) + i sin(θ/4)d1(θ)], where d0(θ) and d1(θ) are both particle-number-conserving matchgates. This turns a circuit with k controlled-phase gates into a weighted sum over 2^k matchgate trajectories; the amplitude of each trajectory is computed as a determinant of a submatrix of the n×n mode-transformation matrix, which only costs O(n^3) or less. The LUCJ-specific optimization rewrites the mode transformation as V3 (I − 2Σ_{i∈N} E_{ii}) V1, so that every trajectory is a base matrix V3V1 plus a few low-rank corrections; ExtraFerm precomputes all correction matrices and caches their d","core_discovery":"At the paper's core is a trajectory decomposition: every controlled-phase gate c(θ) is rewritten as a weighted sum of two particle-number-conserving matchgates, d0(θ) and d1(θ), with weights cos(θ/4) and i sin(θ/4). A circuit with k controlled-phase gates therefore becomes a weighted sum over 2^k matchgate-only trajectories, and each trajectory amplitude is a determinant of a submatrix of an n×n mode-transformation matrix — the standard Slater-determinant rule for non-interacting fermions. Summing all 2^k trajectories gives the exact Born-rule probability, so the exact cost is exponential only in k. Sampling trajectories according to the angle-dependent weights gives a Monte Carlo estimate w","pith_inferences":["Because the extent ξ* grows multiplicatively with both the number and the magnitude of controlled-phase angles, the practical sweet spot is 'many matchgates, few small-angle phase gates' — ExtraFerm is not a general-purpose circuit simulator, and its advantage evaporates once angles approach ±π.","The warm-start recipe likely transfers beyond SQD: any noise-mitigation or bootstrap step that subsamples bitstrings — including related configuration-selection methods, importance-sampled error mitigation, or training-set construction for machine-learned corrections — could use ExtraFerm's targeted probabilities to reweight samples, provided the circuits fit the matchgate-plus-phase form.","A direct numerical check of Eq. (6) against EXACT probabilities on circuits with negative angles would be a cheap way to validate the sign-mask handling; if the Monte Carlo bound holds for those cases, the warm-start results gain a firmer theoretical footing than the paper itself supplies."],"forward_implications":["For circuits with a handful of controlled-phase gates (k ≲ 20–30), exact Born-rule probabilities of selected bitstrings become computable at 50+ qubits, a regime where full state-vector simulation is impossible.","The approximate mode gives a tunable trade-off: user supplies additive error ϵ, failure probability δ, and the trajectory count follows from Eq. (6), making probability estimates with rigorous error bars available for larger k as long as the extent ξ* stays small.","Warm-starting SQD with ExtraFerm improves both accuracy and variance of molecular energy estimates at negligible overhead, suggesting that any bitstring-sampling hybrid algorithm could use targeted probability computation as a cheap post-processing filter.","ExtraFerm's memory footprint is essentially independent of qubit count, so it can serve as a drop-in subroutine inside classical-quantum workflows where other simulators would exhaust RAM.","The measured error trends (absolute error proportional to true probability) make the approximate probabilities reliable as a ranking signal, which is exactly what makes the warm-start heuristic work in practice."],"fun_headline_variants":["Exact cost exponential only in phase-gate count, not qubits","Phase-gate count decides cost: ExtraFerm outruns tensor networks","Warm-started SQD: 46% accuracy gain at 2% overhead","Trajectory sum turns phase gates into matchgates: exact probabilities"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The trajectory-count bound in Eq. (6) is borrowed from prior work and not re-derived here; if it fails to hold for ExtraFerm's signed, angle-dependent decomposition — especially for negative controlled-phase angles handled by the sign mask — the approximate probabilities used to warm-start SQD could be inaccurate, and the claim of exponential-in-angle scaling would not be supported.","fun_headline_variants_meta":{"raw":{"variants":["Exact cost exponential only in phase-gate count, not qubits","Phase-gate count decides cost: ExtraFerm outruns tensor networks","Warm-started SQD: 46% accuracy gain at 2% overhead","Trajectory sum turns phase gates into matchgates: exact probabilities"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000427,"raw_usage":{"total_tokens":2020,"prompt_tokens":736,"completion_tokens":1284,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":480,"completion_tokens_details":{"reasoning_tokens":1201}},"tokens_in":480,"tokens_out":1284,"duration_ms":11379,"temperature":1.0,"reasoning_tokens":1201,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T22:01:38.060044+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a 12-qubit LUCJ circuit with controlled-phase angles including negative values near −π; compute exact probabilities of the full support with EXACT, then run RAWESTIMATE with t given by Eq. (6) for an additive error ϵ and failure probability δ (e.g., 0.01 and 0.01). If the empirical frequency of bitstrings where |p̂ − p_exact| > ϵ exceeds δ, the Monte Carlo error guarantee — and the warm-start benefits built on it — are falsified.","supporting_citations":[],"review_version":1}