{"id":"0f0f595d-b4e7-4b8a-a8fe-9a688fb0e1e4","arxiv_id":"2512.10810","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Quantum-to-quantum Bernoulli factories implement exactly rational functions; the minimum qubit count equals the function degree, and optimal circuits and a closed-form success probability are given.","lead":"This paper shows exactly how many qubits a quantum-to-quantum Bernoulli factory needs for a given function and gives the optimal circuit to build it. It also defines two new variants that handle several input biases or several functions with one protocol.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The all-zero-acceptance canonical form in Sec. III.A is asserted without proof; multiple successful outcomes cannot generally be unitarily merged, so the lower bound n ≥ deg(f) is not rigorously discharged.","rationale":"The optimality claim has two halves: existence of a deg(f)-copy circuit (construction, Sec. III.B) and impossibility with fewer copies (lower bound, Sec. III.A). The lower bound is the more fragile half. It depends entirely on the canonical all-zero form, which the paper does not prove. The deferred-measurement citation covers postponement but not the collapse of a multi-outcome acceptance set; as noted, such a collapse is not generally unitarily possible. I believe the theorem survives because the branch-wise linearity argument fills the gap, but that argument is absent from the manuscript. The c=0 case in Eq. (22) is a second, smaller gap: c/|c| is undefined at c=0, yet the construction needs c=0 cases (e.g., product) and only treats them by example; setting y=0 is an obvious fix. No code or formal verification accompanies the paper. These are proof-completeness issues, not demonstrated counterexamples, so the reader's CONDITIONAL verdict remains appropriate; I would not strengthen to REJECT or weaken to ACCEPT without the missing lemma.","tokens_in":19929,"tokens_out":24554,"duration_ms":248530,"concrete_test":"Construct the simplest multi-outcome QQBF: one input |z⟩, one ancilla |0⟩, a 2-qubit U such that both ancilla outcomes 0 and 1 yield the same output |f(z)⟩, with total success probability p(z). Check whether there exists a fixed 2-qubit unitary U′ such that postselecting the ancilla on 0 alone gives the same output state and the same p(z) for all z. If no U′ exists, the all-zero canonical-form reduction in Sec. III.A is false as stated, and the lower bound must instead be proved branch-wise (expected result). This directly tests the WLOG claim underlying the optimality theorem.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III.A reduces every QQBF to Fig. 1 — one fixed unitary U, computational-basis measurements, and acceptance iff all non-output qubits are 0 — by citing the deferred-measurement principle. The step 'without loss of generality ... the result is accepted when all provide outcome 0' is the load-bearing point: the lower bound n ≥ deg(f) is obtained only for this all-zero accepted branch. Deferred measurement does preserve the number n of |z⟩ copies, but it does not imply that multiple accepted outcome strings can be merged into the single all-zero string. If two orthogonal ancilla outcomes both yield the desired |f(z)⟩, no fixed unitary can map both post-measurement ancilla states to |0...0⟩ while keeping the output factorized; a unitary is injective. Thus, as written, the proof has not excluded adaptive or multi-outcome protocols that use fewer than deg(f) copies. The gap is real but likely repairable: for a fixed measurement record the success branch is a fixed linear map, so its unnormalized output amplitudes are degree-≤n polynomials in z; a branch with nonzero success probability succeeds on an open dense set, so at least one branch gives f = P/Q with deg ≤ n and the bound follows.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies the Quantum-to-Quantum Bernoulli Factory (QQBF). It claims that any QQBF can be represented by a canonical circuit with n copies of the input state |z⟩, a fixed unitary, computational-basis measurements, and acceptance only when all non-output qubits read 0; from this it derives a lower bound n ≥ deg(f) on the number of required |z⟩ copies. The paper then constructs, for every rational function f = P/Q, a circuit using n = deg(f) copies and no ancillas for n ≥ 2 (with at most one ancilla for degree 1), and gives a closed-form expression for the maximal success probability (Eq. 26). It also formalizes two extensions: multivariate QQBFs and multifunctional QQBFs, with applications to sum and product operations.","tokens_in":20214,"tokens_out":14592,"duration_ms":143510,"significance":"If the main claims hold, the paper resolves the qubit-resource complexity of QQBFs, provides an explicit circuit construction that saturates the degree lower bound, and gives a closed-form success probability. The constructive proof is self-contained and does not merely import the rational-function characterization from prior work; it also provides a unitary-completion algorithm. The multivariate and multifunctional extensions are natural and potentially useful for quantum algorithms and experimental implementations. These are substantial contributions. The main caveat is that the lower-bound proof relies on an incompletely justified canonical-form reduction, and some technical details in the explicit formulas and multifunctional characterization need clarification.","major_comments":[{"comment":"The reduction to a single accepted outcome (all non-output qubits read 0) is asserted via the deferred measurement principle. Deferred measurement preserves the number n of |z⟩ copies and moves measurements to the end, but it does not imply that a protocol with several accepted outcome strings can be rewritten as one accepting only the all-zero string. Because a unitary is injective, distinct post-measurement ancilla states generally cannot be mapped to a single |0...0⟩ while keeping the output factorized. Since the lower bound n ≥ deg(f) is derived only for the all-zero branch, the proof of the lower bound is incomplete as written. A repair is available: for a fixed accepted measurement record, the unnormalized output amplitudes are degree-≤n polynomials in z; any branch with nonzero success probability succeeds on an open dense set, so its output ratio is a rational P/Q with deg ≤ n, a","section":"Sec. III.A, Fig. 1, Eq. (8)"},{"comment":"The text claims to maximize the success probability for a fixed number of qubits n and treats |w|^2 as the only free parameter. But for n > deg(f), the polynomials P and Q representing f are not unique: multiplying both by any polynomial R of degree at most n−deg(f) leaves f unchanged but changes a, b, c, and the numerator in Eq. (25). Thus Eq. (26) is not the maximum over all implementations with n qubits unless one explicitly restricts to the canonical choice deg(P), deg(Q) ≤ n with no common factor. The central tightness claim for n = deg(f) is unaffected, but the stated generality of the optimization is not proven.","section":"Sec. III.C, Eqs. (24)-(26)"},{"comment":"The closed-form solution for y contains the factor c/|c|, so it is undefined when c = 0. The paper then uses c = 0 examples (Sec. VI, product and sum operations) and claims the construction is general. A separate treatment of the c = 0 case is needed, e.g., choosing y = e^{iφ} sqrt((l − |w|^2 − a + b)/2) with φ fixed by Eq. (17). Without this, the explicit algorithm does not cover all rational functions as stated.","section":"Sec. III.B, Eqs. (21)-(23)"},{"comment":"The proof that Eqs. (29)-(30) are necessary and sufficient for compatibility is sketched rather than completed. The text says the conditions allow deriving a1 and a3, and that orthonormality allows always deriving a2, a4, H, but it does not show explicitly that the remaining orthonormality equations are satisfied, nor that setting a5 = 0 yields the maximum probability. In addition, Supplementary V's unitary-dilation construction appears to give a multifunctional circuit for any two functions, which is a different existential claim; the relationship between that construction and the compatibility conditions should be clarified.","section":"Sec. V, Eqs. (29)-(30); Supplementary V"}],"minor_comments":[{"comment":"The definition of M contains a duplicated term |10⟩⟨r| and |10⟩⟨s|; the third term should presumably be |01⟩⟨r| (with the four basis states |00⟩, |10⟩, |01⟩, |11⟩ corresponding to the two output cases and the herald q2). Please correct this typo, as it affects the subsequent dilation argument.","section":"Supplementary V, Eq. (47)"},{"comment":"The notation |s_n_j⟩ is used for symmetric states, but it is not explicitly defined whether j counts zeros or ones in the computational basis. Please define it precisely, as the binomial coefficients in Eq. (4) depend on this convention.","section":"Sec. II, Eq. (4)"},{"comment":"There are several grammatical and typographical errors: 'In this contest' should be 'In this context' (Sec. II), 'desidered' should be 'desired' (Sec. VII), and 'simulatable' and 'simulable' are used inconsistently. A proofreading pass is recommended.","section":"Throughout"},{"comment":"The derivation of the averaged success probabilities over the uniform Bloch-sphere and covariant-state ensembles is not shown. Since these formulas are used to draw conclusions about optimal n, please provide the integration details or a reference where the measures are defined.","section":"Sec. III.C, Eqs. (27)-(28)"}],"recommendation":"major_revision","confidential_remarks":"The paper's central construction and the rationality characterization are plausible and largely self-contained, but the lower-bound proof has a real gap in the canonical-form reduction that must be fixed. The remaining issues (c=0 division, representation freedom in Eq. (26), and the multifunctional proof sketch) are local but should be addressed carefully. I recommend major revision rather than rejection because the main technical gap appears repairable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: this is a useful resource-counting paper for quantum-to-quantum Bernoulli factories, and the main result — n ≥ deg(f) with a matching circuit — is almost certainly right. But two places in the proof are asserted rather than shown, and the success-probability claim is over-sold.\n\nWhat's new: Jiang et al. showed simulable functions are exactly the rational functions; this paper adds a qubit lower bound, an explicit circuit that saturates it with no ancillas for n ≥ 2, a closed-form success probability, and two variants (multivariate and multifunctional). The construction is concrete, and the sum/product examples reproduce the known unitaries from Ref. [22], which is a good sanity check. The multivariate and multifunctional sections are plausible and come with explicit compatibility conditions.\n\nSoft spots:\n\n1. Section III.A is the load-bearing step: \"without loss of generality ... the result is accepted when all provide outcome 0.\" Deferred measurement does not merge multiple accepted measurement outcomes into a single all-zero branch. The stress-test note is right. The gap is repairable — on each accepted branch the output amplitudes are degree-≤n polynomials, and since all branches implement the same f, their ratios coincide, so one branch gives f = P/Q with deg ≤ n — but as written, the lower bound is not fully discharged.\n\n2. The c = 0 case is missing. Equations (21)–(22) contain c/|c|, so the general construction excludes c = 0. The authors handle it in examples and mention the a = b, c = 0 case, but there is no general formula or proof. This is an omission, not a fatal flaw.\n\n3. The success probability is optimized within the proposed family of unitaries, not proven maximal over all QQBF protocols. The phrase \"most efficient in terms of ... success probability\" is stronger than what is demonstrated. A referee should ask whether multi-branch acceptance could beat the single-branch construction; I suspect not, but it needs a proof.\n\nNone of these undermine the main framework. The paper is original, self-contained, and honest about the prior characterization. It deserves a serious referee and likely acceptance after the canonical-form argument is fixed, the c = 0 case is supplied, and the optimality claim is either proved or weakened.","headline":"A useful resource-counting paper for QQBFs: the qubit lower bound and matching construction are real additions, but the proof has two repairable gaps and the success-probability claim is stronger than what is shown.","tokens_in":20711,"tokens_out":5320,"would_cite":true,"duration_ms":59664,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["03.67.-a","03.67.Lx"],"model":"deepseek-v4-flash","headline":"The paper proves that quantum-to-quantum Bernoulli factories simulate exactly the rational functions, and that simulating f(z) needs deg(f) input qubits—with an explicit circuit and closed-form maximal success probability.","keywords":["quantum-to-quantum Bernoulli factory","rational function simulation","qubit complexity lower bound","success probability optimization","multivariate Bernoulli factory","multifunctional Bernoulli factory","quantum randomness manipulation","quantum circuit synthesis"],"falsifier":"A numerical or experimental search for a circuit with only one copy of |z> that outputs |z²> with nonzero probability for a dense set of z values, using any postselection or adaptive strategy, would falsify the lower bound n ≥ deg(f). On the optimality side, computing the success probability of the paper's degree-one circuit for f(z)=z and finding any input z for which a different circuit exceeds Eq. (26) would falsify the maximal-success-probability claim.","tokens_in":19821,"feed_emoji":"⚛️","tokens_out":7930,"duration_ms":76493,"temperature":0.7,"pith_summary":"Quantum-to-quantum Bernoulli factories turn quantum coins—qubits whose amplitudes encode an unknown bias z—into new quantum coins with bias f(z). This paper characterises how costly that transformation is. It proves that the class of simulable functions is exactly the complex rational functions, that any implementation of f(z) must use at least deg(f) copies of the input state, and that the bound is tight: every rational f can be implemented with exactly deg(f) copies, no ancillas (except degree-one functions, which may need at most one), via an explicit unitary circuit. It also supplies a closed-form expression for the maximum success probability and shows the construction reaches it. Knowing the exact qubit count and maximal success probability is what makes these factories viable subroutines for quantum algorithms that rely on Bayesian inference, Monte Carlo methods, and blind quantum computation.","feed_headline":"Qubit cost of quantum Bernoulli factories equals rational function degree","feed_subtitle":"A closed-form success probability and circuit construction make the resource bound optimal for all rational functions.","key_machinery":"The key object is the quantum coin state |z> = (z|0>+|1>)/sqrt(1+|z|²), a parametrisation of qubit states by a complex number. The argument is carried by the canonical-form reduction: every QQBF becomes one fixed unitary U acting on n copies of |z> plus m ancillas, followed by all-zero postselection. In that form the output amplitude pair (P(z), Q(z)) is a linear combination of monomials z^j with binomial weights, which directly ties the degree of the rational function to the number n of input qubits. The paper constructs the optimal U from two orthonormal vectors |v0>, |v1> whose overlaps with the symmetric states |s^n_j> store the coefficients of P and Q; a parameter w parametrises the unu","core_discovery":"Any valid QQBF can be reduced to a fixed unitary U acting on n copies of |z> plus m ancillas, followed by computational-basis measurements and acceptance only when all non-output qubits read 0. In that canonical form every output amplitude is a polynomial in z of degree at most n, so the output is |P(z)/Q(z)> and simulable functions are rational. The paper's constructive proof makes the bound tight: given P,Q of degree at most n, two orthonormal rows of U can be chosen with overlaps KP(z) and KQ(z), which fixes all but one free parameter |w|². Since |w|² only lowers the success probability, the optimal choice is w=0, giving a circuit with n=deg(f) qubits, no ancillas for degree at least 2, a","pith_inferences":["The degree-equals-qubit-count result suggests that the minimal circuit complexity is essentially the degree of a rational map on the Riemann sphere; one might therefore compare quantum factory complexity directly with classical Bernoulli factories, which are restricted to rational functions with no zeros or poles on [0,1].","The compatibility conditions define a partial ordering on rational functions: g1 is compatible with g0 when extra unitary rows can be made orthogonal to g0's rows without touching g0's success probability. A useful next step would be to characterise maximal sets of pairwise compatible functions and schedule many factories in a single circuit.","The lower bound depends on accepting only the all-zero outcome pattern. A testable extension is to search systematically for adaptive or multi-outcome-postselection factories that simulate degree-d functions with fewer than d copies of |z>; finding one would sharpen the boundary of the canonical-form argument."],"forward_implications":["Every rational function f with deg(f) ≥ 2 can be implemented with exactly deg(f) quantum coins and no auxiliary qubits; degree-one functions consume at most one auxiliary qubit.","The success probability formula (Eq. 26) gives the exact maximum for a fixed qubit budget, so no circuit using the same number of qubits can exceed it.","Using more than deg(f) copies of |z> can raise the success probability for particular inputs or state ensembles, and the optimal number of qubits depends on the function and on the input distribution.","Multivariate rational functions of k biases are simulable with per-variable qubit counts equal to their partial degrees, and one auxiliary qubit is needed only when every partial degree is 1.","Any pair of simulable functions can be combined in a single multifunctional circuit; if the pair satisfies the compatibility conditions (29)–(30), the primary function's success probability is preserved, as when the product operation is added to the sum operation."],"fun_headline_variants":["Quantum Bernoulli factory qubit cost equals rational function degree","Optimal qubit count for quantum Bernoulli factories: rational degree","Quantum Bernoulli factories: qubit bound matches rational degree","Rational degree sets qubit cost in quantum Bernoulli factories"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that every valid factory can be rearranged into a single fixed unitary followed by measuring all non-output qubits and accepting only the all-zero pattern; if adaptive mid-circuit measurements, feed-forward, or postselection on other outcome patterns could do the job with fewer than deg(f) copies of |z>, the claimed optimality would collapse.","fun_headline_variants_meta":{"raw":{"variants":["Quantum Bernoulli factory qubit cost equals rational function degree","Optimal qubit count for quantum Bernoulli factories: rational degree","Quantum Bernoulli factories: qubit bound matches rational degree","Rational degree sets qubit cost in quantum Bernoulli factories"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000224,"raw_usage":{"total_tokens":1293,"prompt_tokens":737,"completion_tokens":556,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":481,"completion_tokens_details":{"reasoning_tokens":491}},"tokens_in":481,"tokens_out":556,"duration_ms":6298,"temperature":1.0,"reasoning_tokens":491,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T17:02:11.228715+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A numerical or experimental search for a circuit with only one copy of |z> that outputs |z²> with nonzero probability for a dense set of z values, using any postselection or adaptive strategy, would falsify the lower bound n ≥ deg(f). On the optimality side, computing the success probability of the paper's degree-one circuit for f(z)=z and finding any input z for which a different circuit exceeds Eq. (26) would falsify the maximal-success-probability claim.","supporting_citations":[],"review_version":1}