REVIEW
The exact LCU sampling overhead of collective diagonal unitaries: resonances and a continued-fraction dichotomy
Quantum Physics · Submitted 2 Oct 2026
Pith's verdict: First exact optimality theory for the LCU sampling overhead of a genuinely central gate family, with an honest and mostly sound division of proved versus measured results—worth a serious referee, provided the authors ship their verification artifacts.
T0 · AI review
Why Pith says this
Highlight any text to comment on it.
Referee report
The paper
The paper claims
The minimal sampling overhead of the quadratic collective phase $e^{-i\gamma K^2}$ is set by the continued fraction of $\gamma/\pi$: rational angles cost exactly $q$ shots, badly approximable angles cost $\Theta(n)$, and a master formula…
The reading
What carries the argument
The load-bearing mechanism is the duality between the LCU overhead and a trigonometric-polynomial certificate: $\sqrt{\Gamma_{\min}}(U)$ is the atomic norm of the data $u_k$ with respect to the continuous dictionary of complex exponentials, and Hahn\textendash Banach duality turns every lower bound into one degree-$n$ polynomial of sup-norm at most one. At rational angles the Fej\'er-windowed conjugate chirp $P(\theta) = q^{-1/2}\sum_k (1 - |k-(q-1)|/q) u_k e^{ik\theta}$ is an exact dual certificate for the $q$-atom resonant measure at the coset $\theta^*_j = (2\pi j + \pi\sigma)/q$, establishing the threshold $n \ge 2q-2$ and uniqueness of the optimizer. The continued fraction of $\gamma/\pi$ then drives the whole classification: convergents index the rational rungs, and the partial quotients decide whether the cost is $\Theta(n)$ or dips to zero. For the register separation, a new certificate-lifting argument glues four twisted copies of a smaller certificate through two glue bits with two-bit Fourier projectors as readout polynomials, exceeding the cut rate exactly at the gap between the $\ell^2$ and $\ell^1$ norms.
Extended reading notes
Core claim
On the paper's own terms: the minimal overhead of the quadratic collective phase is a norm — $\sqrt{\Gamma_{\min}}$ is the total variation of the cheapest complex measure on the circle whose Fourier coefficients reproduce the target's spectral data, and exact duality makes every lower bound one trigonometric polynomial of degree $n$ with sup-norm at most one, both optima attained. The arithmetic law follows: at a rational angle $\gamma = \pi p/q$ in lowest terms, the cost is exactly $q$ for every $n \ge 2q-2$, certified by a Fej\'er-windowed conjugate chirp and attained uniquely by a uniform sampler over $q$ equally spaced angles — the fractional-revival decomposition of Kerr optics. The cost is $\Theta(n)$ if and only if $\gamma/\pi$ is badly approximable, with explicit constants; it is bounded if and only if the angle is rational; and in an untruncated bosonic mode rationality decides whether a finite-cost representation of the Kerr phase exists at all. A master formula, $\Gamma_{\min} \asymp \min_{p/q} q(1 + |\gamma - \pi p/q| n^2)$, is proved two-sided near low-denominator rationals. Dropping symmetry, the binary-register squaring phase of Fourier arithmetic costs at least $2^{0.557m-O(1)}$ over single-qubit gates, strictly beyond the $2^{m/2}$ ceiling of every single-cut operator-entanglement bound, by a new certificate-lifting argument.
Load-bearing premise
The proved structural theorems — rational angles cost exactly $q$, $\Theta(n)$ iff badly approximable, bounded iff rational, with explicit constants — do not rest on the certified-computation apparatus; but the quantified landscape does: the median cost near $0.6(n+1)$, the share bounds $0.29/f^2$ and $6.13/f$, the measured golden and silver constants $0.2864$ and $0.1974$, and the sharp constant $C(\gamma) \in [1.53, 2.07]$ all rely on rescaling a finite-grid dual certificate by Bernstein's inequality with no shipped code or data, so if a rescaled certificate exceeds sup-norm one at some off-grid angle those constants shift even though the explicit-constant theorems survive.
Editorial extensions
If this is right
- The Fourier-based LCU construction that guarantees $\Gamma \le n+1$ is optimal up to a constant factor at typical angles, so its worst-case bound is in fact the typical price; at rational angles the $n$-independent fractional-revival sampler achieves exactly $q$, beating that bound by a factor growing with the problem size.
- For bosonic simulation, the Kerr-gate phase-shifter decomposition is the unique optimal decomposition over phase shifters, so its $\ell^1$-norm bound is tight, and because finite-cost decompositions exist only at rational $\gamma/\pi$, approximation at irrational Kerr strengths is unavoidable.
- At rational couplings, the $q$-component cat decomposition of one-axis twisting is total-variation optimal and unique above a threshold and strictly beatable below it; at badly approximable couplings no cheap signed mixture of rotations exists at all.
- The squaring phase of Fourier arithmetic separates implementation resource sets: two-local diagonal gates implement it at overhead one, yet no LCU over arbitrary single-qubit gates does so below an exponential in the register size, with an exponent above what operator entanglement across any single cut accounts for.
Reading between the lines
- If the master formula extends off the deep-resonance range (the paper's Open Problem M.1), the overhead at any angle and register size would be computable from a single continued-fraction datum, which could become a practical recipe for choosing penalty angles in constrained quantum optimization.
- The certificate-lifting technique behind the register separation reads as a general template for lower bounds that operator entanglement across a single bipartition cannot reach, potentially applicable to other quadratic phase functions and other depth parameters.
- The 2-adic sheet map of Theorem 8.3 suggests gate libraries shape LCU overhead arithmetically: adjoining symmetric diagonal Clifford phases removes exactly the power-of-two factor from a rung's cost, a fact a library designer could exploit when a rung lies just beyond reach.
- The triangle of equivalences — $\Theta(n)$ cost, data-polynomial flatness, and bounded type for chirp data — hints that LCU sampling overhead is a faithful quantitative proxy for Diophantine structure in a broader class of collective phases than the quadratic one.
Original abstract
The minimal sampling overhead of the quadratic collective phase $e^{-i\gamma K^2}$ over one layer of single-qubit rotations is decided by the continued fraction of $\gamma/\pi$. Here $K$ is a collective observable with spectrum $\{0,\dots,n\}$, such as the permutation-symmetric Hamming weight. This gate is the cardinality-penalty layer of constrained optimization, the one-axis-twisting gate of spin squeezing and the Kerr phase of a bosonic mode. Instead of compiling it to two-qubit gates, we sample it as an LCU over such layers at overhead $\Gamma$. Our results give an optimality theory for this overhead. Ancilla-free sampling reproduces the target's outcome probabilities up to the minimal $\Gamma$, and no smaller factor works for every input; for permutation-symmetric $K$, independent per-qubit angles or one layer of arbitrary single-qubit gates give no further reduction. Rational angles $\pi p/q$ in lowest terms cost exactly $q$ once $n\ge2q-2$, attained uniquely by a uniform $q$-point sampler; the cost is $\Theta(n)$ if and only if $\gamma/\pi$ is badly approximable, and bounded in $n$ if and only if it is rational. A second result shows what the symmetry buys: without it, the squaring phase of Fourier arithmetic on $m$ qubits costs at least $2^{0.557m-O(1)}$ over single-qubit gates, beyond the $2^{m/2}$ of every cut bound, by a new certificate-lifting argument. For constructions known only through upper bounds, it gives the optimum: the Fourier-based LCU of Carrera Vazquez, Egger and Woerner is, at fixed $n$, optimal up to a constant at typical angles and beaten by $(n+1)/q$ at rational ones with an $n$-independent sampler; the Kerr decomposition of Upreti, Quesada and Chabaud is the unique optimum over phase shifters at rational parameters without photon-number cutoff; and it identifies the rational couplings where one-axis twisting admits optimal cat decompositions.
The compiled record
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
The abstract, sentence by sentence
We consider the cost of implementing the quadratic collective phase $e^{-i\gamma K^2}$, with $K$ a collective observable of evenly spaced spectrum, such as the permutation-symmetric Hamming weight. This gate is the cardinality-penalty layer of constrained quantum optimization, the one-axis-twisting gate of spin squeezing and the Kerr phase of a bosonic mode. Instead of two-qubit gates, it can be sampled as a linear combination of single-qubit rotation layers (an LCU) at an overhead $\Gamma$. We determine the minimal overhead. Ancilla-free sampling of the layers reproduces the target's outcome probabilities up to this factor, and no smaller factor works for every input; when $K$ is permutation symmetric, arbitrary single-qubit gates do no better. The minimal overhead's growth with the register size $n$ is set by the continued fraction of $\gamma/\pi$: rational angles $\pi p/q$ in lowest terms cost exactly $\Gamma=q$ for every $n\ge2q-2$; the cost is $\Theta(n)$ if and only if $\gamma/\pi$ is badly approximable; and a master formula, proved two-sided near low-denominator rationals, reads it off the nearest one. Without the symmetry the cost over single-qubit gates can be exponential: for the binary-weighted $K=\sum_j2^jx_j$ the squaring phase of Fourier arithmetic on $m$ qubits costs at least $2^{0.557m-O(1)}$, beyond the $2^{m/2}$ ceiling of every cut bound, by a new certificate-lifting argument. The Fourier-based LCU of Carrera Vazquez, Egger and Woerner (2026), which guarantees $\Gamma\le n+1$, is optimal up to a constant factor at typical angles at fixed $n$; at rational angles the unique optimizer is the fractional-revival decomposition of Kerr optics, a recipe independent of $n$. In bosonic simulation, the Kerr-gate phase-shifter decomposition of Upreti, Quesada and Chabaud (2026) is the unique optimum over phase shifters, and a finite-cost one exists only at rational $\gamma/\pi$.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | We consider the cost of implementing the quadratic collective phase $e^{-i\gamma K^2}$, with $K$ a collective observable of evenly spaced spectrum, such as the permutation-symmetric Hamming weight. | background 0.77 | 0.23 | 0.03 | asserted 0.72 |
| s1 | This gate is the cardinality-penalty layer of constrained quantum optimization, the one-axis-twisting gate of spin squeezing and the Kerr phase of a bosonic mode. | background 0.98 | 0.07 | 0.02 | asserted 1.00 |
| s2 | Instead of two-qubit gates, it can be sampled as a linear combination of single-qubit rotation layers (an LCU) at an overhead $\Gamma$. | method 0.57 | 0.19 | 0.05 | asserted 0.83 |
| s3 | We determine the minimal overhead. | contribution 0.91 | 0.90 | 0.15 | asserted 0.89 |
| s4 | Ancilla-free sampling of the layers reproduces the target's outcome probabilities up to this factor, and no smaller factor works for every input; when $K$ is permutation symmetric, arbitrary single-qubit gates do no better. | result 1.00 | 0.86 | 0.20 | asserted 1.00 |
| s5 | The minimal overhead's growth with the register size $n$ is set by the continued fraction of $\gamma/\pi$: rational angles $\pi p/q$ in lowest terms cost exactly $\Gamma=q$ for every $n\ge2q-2$; the cost is $\Theta(n)$ if and only if $\gamma/\pi$ is badly approximable; and a master formula, proved two-sided near low-denominator rationals, reads it off the nearest one. | result 1.00 | 0.93 | 0.74 | asserted 1.00 |
| s6 | Without the symmetry the cost over single-qubit gates can be exponential: for the binary-weighted $K=\sum_j2^jx_j$ the squaring phase of Fourier arithmetic on $m$ qubits costs at least $2^{0.557m-O(1)}$, beyond the $2^{m/2}$ ceiling of every cut bound, by a new certificate-lifting argument. | result 0.99 | 0.86 | 0.11 | asserted 1.00 |
| s7 | The Fourier-based LCU of Carrera Vazquez, Egger and Woerner (2026), which guarantees $\Gamma\le n+1$, is optimal up to a constant factor at typical angles at fixed $n$; at rational angles the unique optimizer is the fractional-revival decomposition of Kerr optics, a recipe independent of $n$. | result 0.84 | 0.60 | 0.13 | asserted 1.00 |
| s8 | In bosonic simulation, the Kerr-gate phase-shifter decomposition of Upreti, Quesada and Chabaud (2026) is the unique optimum over phase shifters, and a finite-cost one exists only at rational $\gamma/\pi$. | result 0.93 | 0.45 | 0.07 | asserted 1.00 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
Figures
Figures reproduced from arXiv:2610.03595 by the authors.
Discussion
Disagree with the review, or know something it missed? Leave a comment, highlight any passage above to comment on it, or challenge this review formally.