{"id":"7995b71b-30f8-45bb-8425-ae54541a2b62","arxiv_id":"2504.14349","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"An upsampling-based divide-and-conquer circuit for quantum state preparation of probability distributions is derived, but the discrete-distribution extension is not correctly normalized and the method is not novel.","lead":"This paper derives a circuit for preparing quantum states whose amplitudes encode sampled probability distributions, using an upsampling scheme and controlled rotations, and applies it to Gaussian, Laplace, and binomial examples. It claims a polylogarithmic depth version via quantum forking, but the core algorithm reduces to known state-preparation techniques and the discrete extension has a normalization error.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The discrete-probability extension is internally inconsistent: Eq. (33) is not normalized, Eq. (38) does not match Eq. (33), and the claim that Fig. 2 cannot prepare arbitrary discrete amplitudes is contradicted by the Mottonen-style decomposition.","rationale":"The reader's weakest assumption identifies the same load-bearing failure: the discrete extension, which is half of the paper's stated contribution, is not a valid preparation of an arbitrary discrete probability distribution. My independent check confirms the normalization error in Eq. (33) and the mismatch with Eq. (38), and adds the point that the alleged impossibility of preparing (28) with Fig. 2 contradicts the universality of the Mottonen decomposition, which uses the same controlled-Ry gate count. Because the continuous part is a plausible re-derivation of known results and the forking reduction is a known technique, the continuous contribution is not the central problem; however, the abstract explicitly claims the discrete extension, and that claim is unsupported. The paper would need either to use the Fig. 2 circuit directly with Mottonen-style angles for arbitrary discrete probabilities, or to substantially restrict and reword the discrete claim, before the central claim could stand. Therefore the reader's REJECT verdict is appropriate; my stress-test does not change it.","tokens_in":10655,"tokens_out":8540,"duration_ms":79026,"concrete_test":"For n=3, choose a non-symmetric target distribution q over 8 values, e.g. q = [0.05, 0.10, 0.15, 0.20, 0.10, 0.05, 0.20, 0.15] normalized to 1. Compute the state produced by Eq. (38) using P over the first 4 values, then evaluate its norm and its fidelity to the target |φ_3>. If the norm is not 1 or the fidelity is below 1, the discrete construction does not prepare the claimed distribution. As a control, use the same Fig. 2 circuit with angles chosen by the standard Mottonen recursive formula for |φ_3>; if that achieves fidelity 1, the premise that Fig. 2 cannot prepare arbitrary discrete probabilities is experimentally disproven.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's discrete claim rests on Section II.B, and that argument fails in at least three places. First, the premise that the circuit of Fig. 2 cannot prepare the arbitrary real-amplitude state (28) for n>1 is false: Fig. 2 contains 2^n−1 controlled Ry rotations, and the standard Mottonen et al. construction [9] uses exactly such gates to prepare arbitrary real amplitude states by recursive angle selection. Thus the stated motivation for the auxiliary-state construction (33) is incorrect. Second, Eq. (33) is not normalized. The branch |φ*_{n−1}> has squared norm 2^{n−1}−1, so the squared norm of (33) is 2^{n−1}/(2^n−1), equal to 1 only in the trivial n=1 case. The correct equal-weight normalization would need a factor 2^{−(n−1)/2} rather than (2^n−1)^{−1/2}. Third, Eq. (38) does not equal Eq. (33): the m>1 rotations with θ=π/2 contribute a factor 2^{−(n−1)/2} to every amplitude, so the two expressions differ by that overall factor, and Eq. (34) integrates to 2^{n−1}/(2^n−1), not 1. Even after correcting the normalization, the construction only encodes 2^{n−1} independent probabilities P(i_{n−1}) plus their complements in the other half of the register; it cannot encode an arbitrary 2^n-point discrete distribution. The Figure 6 demonstration uses only a projection onto the |0⟩ branch of the most significant qubit, which is not the full state promised in the abstract.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a quantum state-preparation method based on time-series upsampling. For a continuous probability density P(x) with support on the real line, the authors define a sampling lattice over a window w, derive a normalization factor δx_n, and express the target state as a product of cosine rotations parameterized by angles θ_{n,m}; the resulting divide-and-conquer circuit has exponential gate count, which they then convert to polylogarithmic depth using quantum forking. The same formalism is extended to discrete probability distributions by preparing a two-branch auxiliary state |ψ_n> = (|0>|φ_{n-1}> + |1>|φ*_{n-1}>)/√(2^n−1). Examples include Gaussian, Laplace, and binomial distributions.","tokens_in":11054,"tokens_out":12100,"duration_ms":96676,"significance":"The continuous part is a coherent and explicit construction: Eq. (16) correctly solves the normalization condition, Eqs. (8)-(11) give correct limiting behaviors, and the angle prescription (26) is parameter-free in the sense that all angles are computed directly from the target PDF. The paper also provides explicit small circuits, which is helpful. However, the advertised extension to 'arbitrary' discrete distributions is the central novelty of the abstract, and that extension is invalid: the proposed state is not normalized, Eq. (38) is inconsistent with Eq. (33), the premise that Fig. 2 cannot prepare arbitrary real-amplitude states is false, and the construction has only half the required degrees of freedom. Because these defects are in the load-bearing discrete section, the paper's main claim is not established.","major_comments":[{"comment":"The state |ψ_n> defined in Eq. (33) is not normalized for n>1: because |φ_{n-1}> has squared norm Σ P(i_{n-1}) and |φ*_{n-1}> has squared norm 2^{n-1}−Σ P(i_{n-1}), the squared norm of (33) is 2^{n-1}/(2^n−1). Correspondingly, the PDF in Eq. (34) integrates to 2^{n-1}/(2^n−1), not to unity. The correct prefactor would be 2^{−(n−1)/2}, and the same erroneous 1/(2^n−1) factor appears in Appendix A, Eq. (A7).","section":"Section II.B, Eqs. (33)-(34) and Appendix A, Eq. (A7)"},{"comment":"The claim that 'It is not possible to prepare (28) from the set of unitary gate operations depicted in Fig. 2, except when n=1' is incorrect. The circuit of Fig. 2 contains 2^n−1 controlled Ry rotations, which is the standard parameter count for arbitrary real-amplitude state preparation; the Mottonen et al. construction cited as [9] uses exactly such uniformly controlled Ry gates. Thus the stated motivation for the auxiliary-state construction (33) is unsupported, and the apparent need for the two-branch state disappears.","section":"Section II.B, preamble to Eq. (33)"},{"comment":"Equation (38) does not follow from substituting Eq. (37) into Eq. (25). For m>1, each factor in the product in (25) evaluates to cos(π/4)=1/√2 independent of the bit, so the product contributes an overall factor 2^{−(n−1)/2}; Eq. (38) instead has prefactor 1/√(2^n−1). Consequently (38) is not equal to (33), and even after correcting the prefactor the state would have amplitudes √P(i_{n−1})/2^{(n−1)/2} in the first half and √(1−P(i_{n−1}))/2^{(n−1)/2} in the second half, which is not the arbitrary distribution |φ_n> of Eq. (28). The construction contains only 2^{n−1} free parameters P(i_{n−1}), not the 2^n−1 parameters needed for arbitrary discrete probabilities.","section":"Section II.B, Eq. (38)"},{"comment":"The binomial demonstration reports only the projection onto the |0> state of the most significant qubit. This is not the full n-qubit state promised by the abstract; it discards half of the amplitudes, and a projection of a normalized state need not match the target distribution. The example therefore does not verify preparation of the 2^n-point binomial distribution.","section":"Section III.B.1, Fig. 6"}],"minor_comments":[{"comment":"The word 'polylogarthmic' is misspelled; it should be 'polylogarithmic'.","section":"Abstract and Section I"},{"comment":"The phrase 'the compliment of P(0)' should read 'the complement of P(0)'.","section":"Section II.B, text before Eq. (29)"},{"comment":"The number of ancilla qubits is given as 'd = 2n− 1'; from context and Refs. [10,11] this should be d = 2^n−1.","section":"Section II.C"},{"comment":"The caption says the sampling window is centered about x = x̄ + wζ/2, while Eq. (3) gives x_0 = x̄ + w(ζ−1)/2; these statements are inconsistent.","section":"Fig. 1(a) caption and Eq. (3)"},{"comment":"The notation i_{n−m} is not defined explicitly; it would help to state that it denotes the integer formed by the lower n−m bits.","section":"Eq. (25)"}],"recommendation":"reject","confidential_remarks":"The continuous-probability part appears internally consistent, but the discrete extension is the headline contribution and it fails at multiple independent points: the normalization is wrong, Eq. (38) is inconsistent with Eq. (33), the impossibility premise is false, and the construction has insufficient degrees of freedom. Because the discrete construction cannot be repaired by a local normalization change, I do not see a path to acceptance within the current scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nRead the Erickson paper. The upsampling framing is the only thing that's genuinely new: sampling a PDF with a window w and a small random shift zeta, then building the state from the periodic sum. That part is worked out cleanly—the normalization factor delta_x_n, the limiting behaviors (8)-(11), and the recursive angle formulas (19)-(26) are internally consistent. But here's the rub: the circuit those angles define is exactly the Mottonen et al. divide-and-conquer construction, and the polylog depth comes entirely from Araujo's forking. So the continuous section is a competent re-derivation of known results with a new parameterization, not a new result.\n\nThe discrete section is where it falls apart. The claim in II.B that Fig. 2 cannot prepare an arbitrary real-amplitude state is simply false: the figure contains 2^n - 1 controlled Ry gates, which is the Mottonen count, and that construction does arbitrary real amplitudes. That false premise is the entire motivation for the auxiliary-state construction, so the foundation is rotten.\n\nThen the math compounds the problem. Eq. (33) as written is not normalized: the branch |phi*_{n-1}> has squared norm 2^{n-1} - 1, so the total squared norm is 2^{n-1}/(2^n - 1), equal to 1 only for n=1. The correct factor would be 2^{-(n-1)/2}, not (2^n - 1)^{-1/2}. Eq. (34) inherits the same error, integrating to 2^{n-1}/(2^n - 1). And Eq. (38), which the author says is the same state, differs from (33) by an overall factor because the m>1 rotations with theta = pi/2 each contribute 1/sqrt(2). So the two expressions are not equal.\n\nEven after fixing the normalization, the construction only encodes 2^{n-1} independent probabilities P(i_{n-1}) and their complements in the other half of the register. It cannot represent an arbitrary 2^n-point distribution, despite the abstract's promise. The Figure 6 demo shows only the |0> branch projection, which sidesteps the issue.\n\nSo: the paper is not a waste of paper—the continuous derivation is a nice, clean write-up of a known method with an interesting sampling twist. But the discrete extension, which is half the stated contribution, is unsupported. The errors are mechanical, not deep, but they are load-bearing. I wouldn't bring this to a reading group, and I wouldn't cite it. For peer review: a serious editor could desk reject on the grounds that the main novel claim fails and the rest is already known; given the fundamental premise error, I don't think it earns referee time until the author fixes the discrete section.","headline":"The continuous half is a sound but non-novel re-derivation of Mottonen-style state prep; the discrete half rests on a false premise and a normalization error, so the paper's main extension claim fails.","tokens_in":11554,"tokens_out":5368,"would_cite":false,"duration_ms":44093,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"An upsampling construction prepares quantum states encoding sampled probability distributions, with quantum forking reducing gate depth to polylogarithmic.","keywords":["quantum state preparation","upsampling algorithm","probability distributions","polylogarithmic gate depth","quantum forking","Gaussian distribution","Laplace distribution","binomial distribution"],"falsifier":"Calculate the squared norm of the discrete state in Eq. (33) for $n>1$: with the prefactor $1/\\sqrt{2^n-1}$ it equals $2^{n-1}/(2^n-1)$, not $1$, so a simulation of the $n=3$ binomial circuit would reveal measured probabilities that miss the target distribution by that factor. For the continuous branch, measure bitstring frequencies from the prepared Gaussian state and compare them with $P(x_n)\\delta x_n$; any systematic deviation beyond sampling error would refute the preparation claim.","tokens_in":10442,"feed_emoji":"⚛️","tokens_out":12177,"duration_ms":94132,"temperature":0.7,"pith_summary":"The paper derives a quantum state-preparation routine that loads a sampled probability distribution into the amplitudes of an $n$-qubit state. The construction treats the distribution as a periodic upsampled function, factors the amplitudes into products of sines and cosines, and arranges those factors as a divide-and-conquer circuit of controlled $y$-rotations. Because that circuit has exponential gate count, the paper applies quantum forking to produce an equivalent circuit with polylogarithmic gate depth. The method is worked out for continuous distributions on the whole real line, demonstrated on Gaussian and Laplace examples, and extended to arbitrary discrete distributions such as the binomial. If the derivation holds, it offers a general, cheap state-loading primitive for quantum algorithms that need repeated reloading, including financial and scientific applications.","feed_headline":"Upsampling cuts quantum state preparation to polylog depth","feed_subtitle":"A divide-and-conquer amplitude circuit folds into a shallow one, loading Gauss, Laplace, and binomial into n qubits.","key_machinery":"The load-bearing object is the upsampling envelope $\\xi_n(x)$ with its period-splitting identity: splitting the period $w$ into halves turns the envelope into products of $\\cos(\\theta_{n,m}(i)/2)$ and $\\sin(\\theta_{n,m}(i)/2)$ factors. That factorization is what converts a smooth target distribution into a concrete circuit of controlled $y$-rotations. The second piece of machinery is quantum forking, which reuses the controlled rotations across a shared control qubit and ancillas to compress the circuit's time complexity from exponential to polylogarithmic gate depth. For the discrete extension, the same envelope is formed from Dirac-delta densities placed at the bitstring labels, and the reduction to $\\theta_{n,1}=2\\arccos\\sqrt{P(i)}$, $\\theta_{n,m}=\\pi/2$ for $m>1$ is what turns the general circuit into a discrete state-preparation routine.","core_discovery":"The paper's central claim is that a target probability density $P(x)$ can be captured by a periodic envelope $\\xi_n(x) = \\sqrt{\\delta x_n \\sum_{j=-\\infty}^{\\infty} P(x+jw)}$, whose period $w$ contains the $2^n$ sample points of an $n$-qubit register. Repeatedly applying the identity $\\sum_j P(x+jy) = \\sum_j P(x+2jy) + \\sum_j P(x+(2j+1)y)$ splits the envelope into a product of cosine and sine factors, so each qubit is prepared by controlled $y$-rotations with angles $\\theta_{n,m}(i)$. The same formula yields the discrete branch: for a probability distribution $P(i)$ over bitstrings, the paper introduces a complementary-probability auxiliary state and finds that the rotation angles reduce to $2\\arccos\\sqrt{P(i)}$ on the leading qubit and $\\pi/2$ on all others, leaving a circuit whose Hadamard gates and controlled rotations prepare the desired amplitudes. Applied to Gaussian, Laplace, and binomial examples, this is the construction the paper claims achieves polylogarithmic depth after quantum forking.","pith_inferences":["If the period-splitting identity is read in reverse, the same circuit gives a natural tomography dual: measured amplitudes could be inverted to reconstruct a sampled PDF, a use the paper does not pursue.","The $\\zeta$ shift in the sampling window could support an ensemble strategy — prepare several states at different $\\zeta$ values and average measurement statistics — to suppress aliasing artifacts beyond the single-window construction the paper presents.","The normalization factor in the discrete auxiliary state is one of the first things to check in an independent re-derivation: requiring unit norm fixes the prefactor at $2^{-(n-1)}$ rather than $1/(2^n-1)$, and the downstream angle formulas would remain unchanged."],"forward_implications":["For any univariate distribution supported on the whole real line, the same angle formulas prepare the corresponding state; the paper names Cauchy and Student's $t$ as immediate follow-ons to Gaussian and Laplace.","Distributions on the positive half-line, such as lognormal, are covered after a logarithmic change of variables that maps their support onto $\\mathbb{R}$.","Arbitrary discrete distributions over $2^n$ outcomes can be loaded with the discrete circuit, demonstrated on the binomial distribution.","Because quantum states cannot be cloned, a calculation that consumes the state must reload it repeatedly; cutting preparation depth to polylogarithmic makes that reloading affordable in practice.","The univariate construction is stated to extend to multivariate probability distributions."],"supporting_citations":[{"why":"defines the periodicized probability amplitude and resampling idea that the upsampling construction builds on","marker":"[12]"},{"why":"supplies the controlled y-rotation gates from which the divide-and-conquer circuit is assembled","marker":"[9]"},{"why":"provides the divide-and-conquer framework used in the derivation","marker":"[8]"},{"why":"gives the polylogarithmic-depth circuit transformation the paper applies","marker":"[10]"},{"why":"supplies the quantum-forking method that produces the polylogarithmic circuit","marker":"[11]"},{"why":"provides the reference implementation of the forking conversion used in the example calculations","marker":"[14]"}],"fun_headline_variants":["Upsampling trick flattens quantum state prep to polylog depth","Upsampling algorithm yields shallow circuits for continuous and discrete probabilities","Gauss and Laplace states built in polylog depth via upsampling","Quantum state prep: upsampling turns exponential depth into polylog"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The discrete branch rests on the premise that the divide-and-conquer circuit of Fig. 2 cannot prepare arbitrary real-amplitude states directly, which is why a complementary-probability auxiliary state with a particular normalization factor is introduced; if that premise fails, the discrete extension loses its justification.","fun_headline_variants_meta":{"raw":{"variants":["Upsampling trick flattens quantum state prep to polylog depth","Upsampling algorithm yields shallow circuits for continuous and discrete probabilities","Gauss and Laplace states built in polylog depth via upsampling","Quantum state prep: upsampling turns exponential depth into polylog"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000944,"raw_usage":{"total_tokens":4049,"prompt_tokens":979,"completion_tokens":3070,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":595,"completion_tokens_details":{"reasoning_tokens":2991}},"tokens_in":595,"tokens_out":3070,"duration_ms":20430,"temperature":1.0,"reasoning_tokens":2991,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T11:51:56.658887+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Calculate the squared norm of the discrete state in Eq. (33) for $n>1$: with the prefactor $1/\\sqrt{2^n-1}$ it equals $2^{n-1}/(2^n-1)$, not $1$, so a simulation of the $n=3$ binomial circuit would reveal measured probabilities that miss the target distribution by that factor. For the continuous branch, measure bitstring frequencies from the prepared Gaussian state and compare them with $P(x_n)\\delta x_n$; any systematic deviation beyond sampling error would refute the preparation claim.","supporting_citations":[{"cited_title":"Application of an upsampling algorithm to quantum state preparation of continuous and discrete probability distributions","cited_arxiv_id":"2504.14349","evidence_quote":"provides the divide-and-conquer framework used in the derivation"},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"gives the polylogarithmic-depth circuit transformation the paper applies"},{"cited_title":"M¨ ott¨ onen, J","cited_arxiv_id":null,"evidence_quote":"provides the reference implementation of the forking conversion used in the example calculations"}],"review_version":1}