{"id":"edd202d0-1590-48ed-af77-173e0d1c548e","arxiv_id":"2608.07696","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Approximately preparing an N-fold rotationally symmetric cat state with the phase space instruction set requires circuit depth Omega(phi(N)) = Omega(N / log log N), and for prime N a depth-4N protocol saturates this bound with optimal runtime Theta(alpha).","lead":"This paper proves that building a rotationally symmetric N-component cat state with a specific continuous-variable gate set (displacements plus qubit rotations) requires at least the Euler totient of N gates, roughly N divided by log log N, and gives a matching protocol for prime N. It matters because it shows that switching between universal bosonic gate sets can carry a large, state-dependent efficiency penalty.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorems 3.1 and 4.1 appear sound; the load-bearing weakness is Theorem 5.1's compilation-inefficiency claim, which is stated without proof and the abstract's 'further imply' overreaches.","rationale":"The reader's weakest assumption is the nonconstructive Diophantine phase alignment. I checked that part carefully. The factor-two phase in Lemma A.1 is correct because D(b)|a> = e^{i Im(\\bar a b)}|a+b> ≈ e^{2i Im(\\bar a b)}|a>; the N=3 explicit protocol reproduces the ideal map under this phase. Kronecker density provides a finite lambda for each step, and the thresholds alpha*_{up} are finite because T_N(epsilon_ph) is finite for fixed N. So Theorems 3.1 and 4.1 survive the reader's stated concern. The real gap is Theorem 5.1. It is stated as a theorem but its proof is a heuristic Trotter estimate and the authors concede a better cross-compilation may exist. The abstract's claim that the results 'further imply' extremely inefficient conversion is therefore not supported. Since the main lower/upper bounds are sound, the appropriate disposition is CONDITIONAL: keep the main theorems, but either prove or explicitly downgrade Theorem 5.1 and adjust the abstract. This partially agrees with the reader, who already flagged Section 5 but identified a different weakest assumption.","tokens_in":27433,"tokens_out":55148,"duration_ms":501661,"concrete_test":"Test Theorem 5.1 by solving the optimal-control compilation problem numerically: for N=3,5 and alpha up to 30, compile the QSP circuit of [17] (or a single Urot gate) into phase-space instructions using direct search/gradient optimization rather than Trotterization. Compute the minimal runtime of the compiled circuit as a function of target error epsilon. If a compilation with runtime o(alpha^3) (or near Theta(alpha)) is found, the theorem's 'no efficient compilation' assertion fails; if the minimal runtime indeed grows like 1/epsilon >> alpha^3 or worse, the proof still needs to be written down rigorously, including a lower bound that rules out better compilers.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section 5's Theorem 5.1 asserts that there exist state-preparation protocols for which there is no efficient circuit compilation between the phase-space instruction set and a gate set with Urot(θ). The proof does not establish this. It first notes that the QSP circuit of [17] prepares a cat state with O(N) gates; it then cites Theorems 3.1/4.1 to say the same state needs Theta(N) depth and Theta(alpha) runtime in S. That comparison concerns state-preparation complexity, not the cost of compiling a Urot circuit into S-gates. The subsequent Trotter estimate (5.3)-(5.5) only shows that a first-order Trotter synthesis has trun ~ 1/epsilon >> alpha^3; it does not rule out a clever compiler. The authors themselves hedge: 'there may be a more efficient cross-compilation.' No lower bound on compilation cost is supplied. Because the abstract advertises the compilation inefficiency as a consequence of the main results, this unsupported theorem is a genuine load-bearing gap, even though Theorems 3.1 and 4.1 are internally consistent.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the circuit depth and runtime needed to prepare N-fold rotationally invariant Schrödinger cat states using the 'phase space instruction set' (unconditional and qubit-dependent displacements plus single-qubit rotations on a single boson). The main results are Theorem 3.1, a lower bound of Ω(φ(N)) on circuit depth for sufficiently large coherent-state amplitude α, and Theorem 4.1, a matching depth-4N construction for prime N whose runtime is Θ(α) for fixed N and target error. The construction relies on a Diophantine phase-alignment condition whose solvability is proved by a Kronecker-density argument. The paper also presents numerical demonstrations for N=3 and N=5 at moderate α and, in Section 5, claims that the results imply a strong inefficiency in compiling between the phase-space instruction set and a gate set containing Urot(θ).","tokens_in":27604,"tokens_out":9610,"duration_ms":96722,"significance":"If the two main theorems stand, the paper gives a rare asymptotic separation in continuous-variable state preparation: a simple family of states that is surprisingly expensive to prepare with a natural universal gate set, with the lower bound coming from cyclotomic polynomial structure rather than from quantum speed limits. The lower-bound proof is parameter-free and the upper-bound protocol is structurally explicit, with closed-form protocols supplied for N=3 and N=5 and numerical evidence at finite α. The order-of-limits caveat in Remark 4.2 is honestly stated. The compilation-inefficiency claim in Section 5, however, is not established with the same rigor and needs to be either proved properly or removed from the abstract.","major_comments":[{"comment":"The proof as written does not establish the theorem, and the abstract's final claim overreaches. The comparison between the O(N)-runtime QSP protocol of [17] and the Θ(N)-depth, Θ(α)-runtime bounds for the phase-space instruction set is a comparison of state-preparation costs in two different gate sets; by itself it does not bound the cost of compiling one circuit into the other. The Trotter estimates in Eqs. (5.3)-(5.5) only describe a particular first-order synthesis and cannot rule out a smarter compiler; the text even concedes 'there may be a more efficient cross-compilation'. The theorem can be repaired, but only by supplying the missing lower-bound argument: if a compiler mapped the QSP circuit of (5.2) to an S-circuit with runtime o(α), that compiled circuit would prepare the same N-fold cat state with fidelity above 2/3 and runtime o(α), contradicting Proposition 3.4. The proof should be rewritten around that contradiction, with a precise definition of 'efficient compilation' and the allowed error scaling, and the abstract should be conditional on that formalized statement.","section":"Section 5, Theorem 5.1"},{"comment":"The text calls the proof of Theorem 4.1 'constructive', but the crucial parameter λ in the phase-alignment condition is shown to exist only through Kronecker density (Proposition A.4), which is nonconstructive and gives no algorithm to find λ for general prime N. This does not invalidate the existential content of Theorem 4.1, but it does mean the paper provides an existence proof plus explicit instances for N=3 and N=5, not a fully constructive recipe for every prime N. The wording in Section 4.1 should be softened accordingly, or an explicit Diophantine search procedure should be supplied.","section":"Section 4.1 and Appendix A.4"}],"minor_comments":[{"comment":"There is a typo: 'asume' should be 'assume'.","section":"Section 4.2"},{"comment":"The theorem says a protocol prepares |αN⟩ 'with squared state-vector 1−ε, i.e. achieves (2.9)', but Eq. (2.9) is written in terms of the infidelity h, and the relation between h and the squared state-vector error ε is nonlinear. The notation should be aligned.","section":"Theorem 4.1 and Eq. (2.9)"},{"comment":"The text describes the gate-level optimization as 'cutoff-free' because it uses a finite coherent-branch representation; this is accurate for the branch representation, but the representation is still finite (512 branches), so the phrase 'cutoff-free' should be qualified to avoid implying an exact infinite-dimensional optimization.","section":"Section 4.2, gate-level refinement"},{"comment":"There is a stray sentence fragment 'Calculating it explicitly, we get Finally' before Eq. (A.127); the exposition there should be cleaned up.","section":"Appendix A.9"}],"recommendation":"major_revision","confidential_remarks":"The core lower and upper bound theorems (3.1 and 4.1) appear coherent and are the paper's main contribution; the numerical section is a useful extra but not load-bearing. The Section 5 compilation claim is the main weakness: it is advertised in the abstract but not proved as stated. I would recommend major revision focused on rewriting Theorem 5.1 around Proposition 3.4, or removing the compilation claim from the abstract. The nonconstructive nature of the Diophantine step in Theorem 4.1 should also be disclosed accurately."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe thing to know: the paper's core is sound and worth engaging. Theorem 3.1 gives a depth lower bound Ω(φ(N)) for preparing N-fold cat states with the phase space instruction set, and Theorem 4.1 gives a matching depth-4N protocol for prime N with runtime Θ(α). The abstract's further claim about cross-gate-set compilation inefficiency does not follow from the theorems and should be walked back.\n\nWhat is new: the cyclotomic dimension argument (Proposition 3.2) used as a control obstruction is a nice idea, and the Diophantine phase-alignment construction in the appendix is a genuine existence proof for the matching protocol. The runtime merging trick—combining adjacent conditional displacements so the total displacement cost wraps the polygon perimeter rather than traversing radius N times—is clean and gives the Θ(α) runtime. The error analysis via telescoping sums is coherent. Credit where due: these are real results, not numerical speculation; the numerics in Section 4.2 are a demonstration and are not load-bearing.\n\nSoft spots, in order of importance. Section 5's Theorem 5.1 is not proved. The argument compares the QSP state-preparation circuit's O(N) cost with the S-gate state-preparation lower bound; that says the two gate sets prepare the same family of states at different cost, not that converting an existing Urot circuit into S-gates is inefficient. The Trotter estimate (5.3)–(5.5) is a heuristic upper bound on one compilation strategy, and the authors admit a cleverer cross-compilation might exist. The abstract's \"further imply\" overreaches. This is a load-bearing gap only for the compilation narrative; Theorems 3.1 and 4.1 stand independently. Second, the Diophantine phase-alignment step is an existence result via Kronecker density; it gives no explicit λ and the thresholds α* are unquantified and pointwise in N (Remark 4.2 acknowledges this). That is acceptable for an asymptotic theorem, but it means the protocol is not yet a hands-on recipe. Third, no code or data accompanies the GRAPE claims, so Section 4.2 is not reproducible as written; minor. There is also a small conjugation slip in the definition of θ_j in Lemma A.1, which the reader flagged; it doesn't affect the bounds.\n\nVerdict: the core lower bound and matching construction deserve serious refereeing. I would send it out. The authors should be asked to fix Theorem 5.1 (either prove a real compilation lower bound or reframe it as a conjecture/observation) and to state the nonconstructive nature of λ more prominently.\n\nFor reading group: yes, the cyclotomic obstruction is worth discussing. I'd cite Theorems 3.1 and 4.1 in future work.","headline":"Theorems 3.1 and 4.1 hold up and give a real depth separation for the phase space instruction set, but the compilation-inefficiency claim in Section 5 overreaches and should be cut or heavily qualified.","tokens_in":28196,"tokens_out":2346,"would_cite":true,"duration_ms":22139,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","11R18"],"pacs":["03.67.-a"],"model":"deepseek-v4-flash","headline":"Preparing an N-fold rotationally symmetric Schrödinger cat state from a natural continuous-variable gate set requires circuit depth at least the Euler totient φ(N); for prime N, a depth-4N protocol matches the bound, settling the…","keywords":["N-fold cat state","phase space instruction set","circuit depth","Euler totient function","cyclotomic polynomials","bosonic state preparation","qubit-dependent displacement","Diophantine phase alignment"],"falsifier":"For the lower bound, fix N = 5 (so φ(5) = 4) and optimize the best possible depth-3 circuit from the phase space instruction set against the 5-fold cat state at α = 50; Theorem 3.1 predicts no such circuit can beat squared error (2N)^{-1} = 0.1, so finding one would refute it. For the upper bound, take N = 11 and a generic fixed angle φ, then search explicitly for the finite λ that satisfies the phase-alignment conditions (A.58) at the required tolerance; a systematic failure of this search across many generic φ would show the density-based existence guarantee is not realized in practice.","tokens_in":27147,"feed_emoji":"🐈","tokens_out":10025,"duration_ms":84964,"temperature":0.7,"pith_summary":"Preparing an N-fold Schrödinger cat state — a superposition of N coherent states arranged on a circle in phase space — is a basic task for bosonic quantum computing. This paper asks how many elementary instructions it takes when the only allowed gates are single-qubit rotations and displacements of the cavity field that depend on the qubit state, a natural phase space instruction set for circuit-QED hardware. The answer is a no-go theorem: for large coherent amplitude, any circuit that achieves small error must have depth at least φ(N), Euler's totient function, which grows like N/log log N. For prime N, the paper also gives an explicit depth-4N protocol that saturates this bound, with total runtime Θ(α), so the two bounds together pin down the exact asymptotic cost. The broader implication is that a universal gate set can be surprisingly inefficient at preparing even a simple family of states, and that converting bosonic circuits between different universal gate sets can carry a large overhead.","feed_headline":"N-fold cat states need φ(N) gate depth — and primes saturate it","feed_subtitle":"A natural cavity gate set needs near-linear depth for symmetric cats — evidence that bosonic circuit compilation can be very costly.","key_machinery":"Two mechanisms carry the argument. The lower bound rests on the cyclotomic identity dim_Q span_Q{$e^{{2πij/N}}$} = φ(N): the N-th roots of unity are the roots of the N-th cyclotomic polynomial, which is irreducible of degree φ(N) over Q, so at least φ(N) independent complex directions are needed just to place coherent-state centres at the N target points, and each phase-space displacement contributes only one such direction. The upper bound rests on the phase-alignment condition: a single real scaling λ must simultaneously bring every phase difference 2(θ* − θ_j) to within tolerance of the ideal angle θ_κ modulo 2π, and Proposition A.4 guarantees such a λ exists because the sine-difference frequencies are Q-linearly independent for almost every fixed angle φ, making the multiples of λ dense on the torus.","core_discovery":"The paper's central claim is Theorem 3.1: if a circuit built from the phase space instruction set prepares the N-fold cat state |α_N⟩ with squared state-vector error below (2N)^{-1}, then for all sufficiently large |α| the circuit must contain at least φ(N) instructions. The reason is algebraic rather than kinematic: after M conditional displacements, all reachable coherent-state centres are signed sums Σ_j σ_j β_j of the M displacement parameters, so they lie in a Q-vector space of dimension at most M; but the N target centres α $e^{{2πij/N}}$ generate a Q-vector space of dimension φ(N), the degree of the cyclotomic field extension. For prime N, φ(N) = N − 1, and the companion Theorem 4.1 provides a matching depth-4N protocol whose four-gate V blocks each add one coherent component, with the required phase-alignment parameters supplied by a Diophantine density argument. Because successive large displacements merge into chords of the cat's polygon, the protocol's runtime is Θ(α), which Proposition 3.4 shows is optimal.","pith_inferences":["The cyclotomic argument is a template for other target states whose peaks occupy points of high algebraic degree over Q: grids or lattices of coherent states would presumably lower-bound circuit depth by the dimension of the field generated by their coordinates, not merely by the number of peaks.","The matching protocol is an existence result: the phase-alignment parameter λ is guaranteed by torus density, but the proof supplies no algorithm to find it, so at experimentally relevant α the protocol likely needs to be wrapped in numerical optimization, as the paper's own pipeline does.","The small-α numerics suggest a practical, testable claim: for α ≳ 7 the number-theoretic protocol is a better seed for pulse optimization than random initialization, and benchmarking it against SNAP- or QSP-based preparation at the same α would show whether this structural cost translates into wall-clock control time.","The perimeter-merging trick that keeps runtime Θ(α) — adjacent conditional displacements combine into short chords of the cat's polygon — could be reused in other multi-step bosonic protocols where successive displacements are nearly collinear."],"forward_implications":["For any fixed N, the minimum circuit depth needed to prepare an N-fold cat state with the phase space instruction set grows at least as φ(N) ≳ N/log log N, so the cost is nearly linear in the number of cat legs rather than logarithmic.","When N is prime, the depth-4N protocol saturates the lower bound, so the asymptotic cost of preparing prime-fold cats is settled: Θ(N) depth and Θ(α) runtime.","The runtime Θ(α) is optimal — no protocol can do better, because separating the peaks of a cat with finite fidelity already forces a total displacement of order α (Proposition 3.4).","There is no efficient runtime-preserving compilation between the phase space instruction set and a quantum-signal-processing gate set: the same cat state that a QSP circuit prepares in O(N) time provably needs Θ(α) runtime and near-linear depth in the phase space instruction set (Theorem 5.1).","The construction extends to generalized cat states with arbitrary complex coefficients and phases at the same depth 4N (Appendix A.9)."],"supporting_citations":[{"why":"Supplies the number-theoretic backbone of Theorem 3.1: cyclotomic polynomials have integer coefficients and Φ_N is irreducible over Q, so the N-th roots of unity span a Q-vector space of dimension φ(N).","marker":"[21]"},{"why":"Provides the ergodic density fact used in Proposition A.4: Q-linearly independent frequencies give dense orbits on the torus, which guarantees the existence of the phase-alignment parameter λ.","marker":"[24]"},{"why":"Gives the competing quantum-signal-processing protocol that prepares the same cat state in O(N) time, against which Theorem 5.1 shows cross-compilation is inefficient.","marker":"[17]"},{"why":"Underlies the generalized quantum signal processing used in the competing construction and in the Section 5 estimate of Trotter-based compilation cost.","marker":"[23]"},{"why":"Sets the motivational contrast: in qubit systems any state is reachable with polylog(1/ε) depth, the benchmark against which the near-linear bosonic lower bound is surprising.","marker":"[10, 11]"},{"why":"Supplies the GRAPE optimal-control algorithm used in the numerical comparison of random-seeded versus protocol-seeded pulse optimization.","marker":"[16]"},{"why":"Defines universality for continuous-variable gate sets via the dynamical Lie algebra, the notion used in Proposition 2.3 to call the phase space instruction set universal.","marker":"[19]"}],"fun_headline_variants":["N-fold cat states require φ(N) gate depth — proved","Prime N achieves optimal cat-state protocol","Universal gate set fails at symmetric cats","Cyclotomic fields dictate cat-state circuit depth"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The matching depth-4N protocol rests on the assumption that a single scalar parameter can be chosen to align all the required phase angles at once; the proof shows this parameter exists for almost every geometric configuration, but not for every one, and it does not say how to find it in practice.","fun_headline_variants_meta":{"raw":{"variants":["N-fold cat states require φ(N) gate depth — proved","Prime N achieves optimal cat-state protocol","Universal gate set fails at symmetric cats","Cyclotomic fields dictate cat-state circuit depth"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000238,"raw_usage":{"total_tokens":1501,"prompt_tokens":924,"completion_tokens":577,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":540,"completion_tokens_details":{"reasoning_tokens":518}},"tokens_in":540,"tokens_out":577,"duration_ms":6369,"temperature":1.0,"reasoning_tokens":518,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T00:26:52.637522+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For the lower bound, fix N = 5 (so φ(5) = 4) and optimize the best possible depth-3 circuit from the phase space instruction set against the 5-fold cat state at α = 50; Theorem 3.1 predicts no such circuit can beat squared error (2N)^{-1} = 0.1, so finding one would refute it. For the upper bound, take N = 11 and a generic fixed angle φ, then search explicitly for the finite λ that satisfies the phase-alignment conditions (A.58) at the required tolerance; a systematic failure of this search across many generic φ would show the density-based existence guarantee is not realized in practice.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the number-theoretic backbone of Theorem 3.1: cyclotomic polynomials have integer coefficients and Φ_N is irreducible over Q, so the N-th roots of unity span a Q-vector space of dimension φ(N)."},{"cited_title":"259 (Springer, London, 2011)","cited_arxiv_id":null,"evidence_quote":"Provides the ergodic density fact used in Proposition A.4: Q-linearly independent frequencies give dense orbits on the torus, which guarantees the existence of the phase-alignment parameter λ."},{"cited_title":"Optimal control of coupled spin dynamics: design of NMR pulse sequences by gradient ascent algorithms,","cited_arxiv_id":null,"evidence_quote":"Supplies the GRAPE optimal-control algorithm used in the numerical comparison of random-seeded versus protocol-seeded pulse optimization."},{"cited_title":"Quantum computation over continuous variables,","cited_arxiv_id":null,"evidence_quote":"Defines universality for continuous-variable gate sets via the dynamical Lie algebra, the notion used in Proposition 2.3 to call the phase space instruction set universal."}],"review_version":1}