{"id":"0a861e8b-1105-4a0e-aca4-926cd5141735","arxiv_id":"2507.19783","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A sharper discrepancy bound for multi-frequency shift sequences is derived and applied to quantum dynamics, but the supporting lower-bound proof counts points outside the allowed window.","lead":"This math paper proves a sharper upper bound on how often a multi-frequency rotation lands in a curved set, then applies it to bound quantum wave-packet spreading in long-range quasi-periodic Schrödinger operators. The upper-bound argument is clean, but the proof of the matching lower bound appears to count points outside the allowed time window, so the sharpness claim is unsupported.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1.3's lower-bound proof counts tuples k_i without enforcing the resulting index n=∑k_i n_i≤N, so the claimed sharpness of Theorem 1.2 is not established.","rationale":"The reader's weakest-assumption analysis identifies exactly the load-bearing flaw. The proof of Theorem 1.3 constructs a hyperplane S from vectors {n_iα} and then counts all integer combinations ∑k_i n_i whose summands satisfy certain box constraints. The transition from the counting set to the product of k_i-counts silently drops the condition that the combined index n lies in [1,N]. Since the lower bound is the basis for the paper's claim that Theorem 1.2 is 'asymptotically sharp', this is a central defect, not a cosmetic one. The upper-bound discrepancy estimate in Theorem 1.2 appears sound: the ε-ball covering argument and the WDC repulsion at scale γ/N^τ correctly show at most one orbit point per ball. The quantum-dynamics application is conditional on Theorem 1.2 and the imported LDT framework, so it may survive if separated from the sharpness claim. However, the paper as a whole promises sharp semi-algebraic discrepancy estimates; without a valid lower bound, that central claim is unsupported. No verdict adjustment is needed: the existing rejection is consistent with this assessment. I do not see a separate concern that outweighs this one, and I would not manufacture an additional objection.","tokens_in":10918,"tokens_out":7507,"duration_ms":87371,"concrete_test":"Re-derive the counting in §3.2 with the missing restriction n=∑k_i n_i≤N imposed. Specifically, for b=2 take n_1≈N^{2/3} and k_1 up to cN^{1/3} as the proof allows, and compute the proportion of such tuples with k_1 n_1>N. More generally, replace the tuple count by A(N)=#{k∈Z^{b−1}_{≥1}: 1≤k_i n_i≤N, k_i∥{n_iα}∥∞<1/b, ∑k_i n_i≤N} and compare A(N) with the product bound ∏_{i=1}^{b−1} min{b^{−1}N^{1−r(1+2ε)}, b^{−1}N^{r/b}} used in the proof. If A(N)<N^{(b−1)/(b+1)−ε}, the claimed lower bound fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central sharpness claim rests on Theorem 3.3/Theorem 1.3. In the proof, after constructing n_i≤N^{r(1+2ε)} with r=b/(b+1), the argument counts all tuples (k_1,...,k_{b-1}) satisfying 1≤k_i n_i≤N^b (likely intended N) and k_i∥{n_iα}∥∞<1/b, and asserts these give distinct points in {1≤n≤N: nα∈S}. But the index produced is n=∑k_i n_i, and the constraints only give n≤(b−1)N (or (b−1)N^b with the displayed bound), not n≤N. Thus the cardinality inequality immediately after (19) is invalid: the tuple count is not a lower bound for the left-hand side because many counted tuples may correspond to sequence indices outside the theorem's window. The product factorization then cannot support the claimed N^{(b−1)/(b+1)−ε} lower bound. Consequently the asymptotic-sharpness assertion of Theorem 1.2 is unproved. This concern does not invalidate the upper-bound Theorem 1.2 or the quantum-dynamics corollary, but it invalidates the paper's advertised sharpness result as stated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies semi-algebraic discrepancy for multi-frequency Kronecker sequences. Theorem 1.2 gives an upper bound for the number of visits of the orbit to a semi-algebraic set of small measure, under a weak Diophantine condition on the frequency. Theorem 1.3 claims a matching lower bound for almost every frequency, using a hyperplane constructed from b-1 small multiples of alpha. Theorem 1.5 applies the upper bound to long-range quasi-periodic Schr\\\"odinger operators, combining a large deviation theorem from [Liu22] with a dynamical reduction from [Liu23] to obtain logarithmic upper bounds on position moments. The upper-bound proof in Section 2 is a clean covering argument using Bourgain's semi-algebraic covering lemma; the quantum application is a direct combination of imported theorems. The lower-bound proof in Section 3.2 has a serious gap: it counts tuples of integers without enforcing that the resulting index n is at most N.","tokens_in":11096,"tokens_out":19656,"duration_ms":235648,"significance":"If the upper-bound theorem and the quantum application are correct, they improve known semi-algebraic discrepancy exponents from roughly N^{1-1/b^2+\\varepsilon} to N^{1-1/b+\\varepsilon} and give corresponding logarithmic quantum-dynamical bounds. The upper-bound proof is transparent and uses standard tools, and the quantum application is mechanical given the imported LDT and dynamical reduction; these parts are worth publishing if the issue with the lower bound is resolved. However, the advertised 'asymptotically sharp' aspect rests entirely on Theorem 1.3, whose proof is invalid as written, so the sharpness claim is not currently supported. The paper does not ship machine-checked proofs, but the main upper-bound derivation is simple enough to be checked by hand.","major_comments":[{"comment":"The step immediately after (19) counts all tuples (k_1,...,k_{b-1}) satisfying 1 \\le k_i n_i \\le N^b and k_i\\|\\{n_i\\alpha\\}\\|_\\infty < 1/b, and treats this count as a lower bound for #{1 \\le n \\le N : n\\alpha \\bmod \\mathbb{Z}^b \\in S}. This is only legitimate if the associated integer index n = \\sum_i k_i n_i satisfies 1 \\le n \\le N. No such constraint is imposed: the written conditions give only n \\le (b-1)N^b, and even reading the displayed bound as the intended N, one gets only n \\le (b-1)N. Thus many counted tuples correspond to indices outside the window of the theorem, and the product lower bound is not a lower bound for the left-hand side. The proof needs a counting argument for the tuples that additionally satisfy \\sum_i k_i n_i \\le N; under the R-linear independence condition (18) this is a simplex count of order N^{b-1}/((b-1)!\\prod_i n_i), which would yield the claimed exponent, but that argument is absent.","section":"§3.2, proof of Theorem 3.3"},{"comment":"The application of Theorem 4.2 requires an exponent \\delta > 0 such that #B_{N,N^\\varepsilon} \\le N^{1-\\delta}. From the displayed estimate (25) the proof obtains 1-\\delta = \\tau(b-1)+\\varepsilon, so the proof requires \\tau(b-1) < 1. Theorem 1.5 is stated for arbitrary \\alpha \\in WDC(\\tau) \\cap DC(\\tau') with only \\tau \\ge 1/b; for \\tau(b-1) \\ge 1 the argument does not apply and the displayed exponent p/(1-\\tau(b-1)) in the theorem is nonpositive. The statement should either add the hypothesis \\tau < 1/(b-1) or be restricted to that range, which is sufficient for the full-measure Corollary 1.6.","section":"§4, proof of Theorem 1.5"}],"minor_comments":[{"comment":"The notation 'T^d' in Corollary 1.6 should be 'T^b' to match the dimension used in the rest of the paper.","section":"Abstract and Corollary 1.6"},{"comment":"The constraint is written as 1 \\le k_i n_i \\le N^b, while the subsequent lower bound uses N^{1-r(1+2\\varepsilon)}; these two expressions are inconsistent and need to be reconciled.","section":"§3.2, display after (19)"},{"comment":"The expression '|\\langle w, \\{n_k\\alpha\\}\\rangle|^p' appears to contain a spurious power p; the surrounding inequalities suggest the absolute value should not be raised to a variable power.","section":"§3.1, Eq. (16)"}],"recommendation":"major_revision","confidential_remarks":"My recommendation is major revision rather than reject because the upper-bound theorem and the quantum-dynamics application are independent of the flawed lower-bound proof. The missing sum constraint in Theorem 3.3 is a genuine load-bearing gap, but the surrounding construction strongly suggests a simplex-counting repair is available; I would ask the authors to supply that argument or explicitly withdraw the sharpness claim if it cannot be supplied. The reliance on [Liu22, Liu23] is not itself problematic; those results are imported as named theorems, and the application follows formally from them."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know: the paper's main positive result, Theorem 1.2, is a genuine improvement. Replacing Diophantine with weak Diophantine (WDC) frequencies allows a short covering argument that yields exponent 1−1/b for semi-algebraic discrepancy, beating the previous HJ19, JP22, and Liu22 bounds. I checked the proof: the measure condition matches the log N hypothesis, WDC keeps at most one orbit point per eps-ball, and the count comes out as claimed. The quantum dynamics corollary (Theorem 1.5) follows mechanically if you accept the imported LDT from Liu22 and the reduction from Liu23, and Bourgain's semi-algebraic approximation is standard. Those are self-citations, but for established tools, not new claims, so I don't hold that against the paper.\n\nThe soft spot is the sharpness claim. Theorem 1.3 constructs vectors n_1,...,n_{b-1} and then counts tuples (k_i) with 1 ≤ k_i n_i ≤ N^b and k_i ||{n_i α}||∞ < 1/b, asserting each tuple gives a distinct index n = Σ k_i n_i with 1 ≤ n ≤ N. But the constraints only force each k_i n_i ≤ N^b, so n can be as large as (b−1)N^b. Even if the intended bound were k_i n_i ≤ N, the sum can exceed N. The inequality after (19) is therefore not a lower bound for #{1 ≤ n ≤ N : nα ∈ S}; it counts points outside the theorem's window. The advertised lower bound N^{(b−1)/(b+1)−ε} is not established. This is a load-bearing flaw for the ``asymptotically sharp'' headline, but it does not touch Theorem 1.2 or Theorem 1.5.\n\nWho is this for? Spectral theorists and number theorists working on discrepancy of Kronecker sequences. The upper-bound half is publishable; the sharpness half needs either a real fix or removal. I would send this to a serious referee, but the current version should be rejected or returned for major revision. The counting error is elementary and the paper's own claim of sharpness is unsupported.","headline":"The upper-bound discrepancy estimate and quantum-dynamics corollary are clean and correct; Theorem 1.3's lower-bound sharpness proof miscounts and fails as written.","tokens_in":11690,"tokens_out":3807,"would_cite":true,"duration_ms":39829,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11K38","11K55","47B80","81Q10"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves near-optimal bounds on how often multi-frequency shift sequences hit curved sets, and uses them to control wave-packet spreading in long-range quasi-periodic Schrödinger operators.","keywords":["semi-algebraic discrepancy","multi-frequency shift sequences","weak Diophantine condition","quasi-periodic Schrödinger operators","quantum dynamics","large deviation estimates","Green's function","moment bounds"],"falsifier":"Verify the Section 3.2 counting identity in dimension $b=2$: Proposition 3.2 produces $n_1\\in[1,N^{2(1+2\\varepsilon)/3}]$ with $\\|\\{n_1\\alpha\\}\\|_\\infty\\le N^{-1/3}$. The proof counts every $k_1$ satisfying $1\\le k_1 n_1\\le N^2$ and $k_1\\|\\{n_1\\alpha\\}\\|_\\infty<1/2$, and then treats $k_1 n_1$ as lying in $[1,N]$. Computing, for an explicit Diophantine $\\alpha$ and large $N$, whether the number of such $k_1$ with $1\\le k_1 n_1\\le N$ is at least $N^{1/3-\\varepsilon}$ would settle Theorem 1.3; a shortfall means the product count overcounts points outside the window.","tokens_in":10662,"feed_emoji":"📐","tokens_out":21182,"duration_ms":203619,"temperature":0.7,"pith_summary":"This paper establishes sharp counting bounds for how often a multi-frequency shift sequence $\\theta + n\\alpha \\bmod \\mathbb{Z}^b$ lands in a semi-algebraic set $S$ of small volume: if $\\alpha$ is weakly Diophantine, the number of hits among $1\\le n\\le N$ is at most $B C(b) N^{\\tau(b-1)}$, and for almost every frequency this becomes $N^{(b-1)/b+\\varepsilon}$. It also proves a matching lower construction: for almost every $\\alpha$, a hyperplane in $[0,1]^b$ is hit at least $N^{(b-1)/(b+1)-\\varepsilon}$ times, so the upper exponent is essentially optimal as the dimension grows. Because exceptional phase sets for Schrödinger Green's functions are semi-algebraic, the counting bound is converted into a uniform-in-phase statement: for long-range quasi-periodic Schrödinger operators with nonconstant analytic potential and large coupling, the $p$-th moment of the position operator grows at most like $(\\log T)^{p/(1-\\tau(b-1))+\\varepsilon}$, and like $(\\log T)^{pb+\\varepsilon}$ for almost every frequency. The results improve all previously known discrepancy and quantum-dynamical exponents.","feed_headline":"Curved-set hits by multi-frequency shifts now have sharp bounds","feed_subtitle":"Tighter discrepancy estimates give logarithmic quantum-dynamical bounds for long-range quasi-periodic Schrödinger operators.","key_machinery":"The mechanism that carries the upper bound is the one-dimensional weak Diophantine condition $\\|n\\alpha\\|_{\\mathbb{T}^b}\\ge\\gamma/|n|^\\tau$ for all $n\\in\\mathbb{Z}\\setminus\\{0\\}$. It makes the covering argument local: any two distinct points $\\theta+n\\alpha$ and $\\theta+n'\\alpha$ lying in the same small ball would force $\\|(n-n')\\alpha\\|\\le2\\epsilon$, contradicting the condition once $\\epsilon=\\gamma/(2N^\\tau)$. Combined with the covering lemma for semi-algebraic sets (a set of degree $B$ and measure at most $\\epsilon^b$ is covered by $B C(b)\\epsilon^{1-b}$ balls), this gives at most one counted index per ball and hence the bound $B C(b)N^{\\tau(b-1)}$. The lower bound instead constructs, via a metric discrepancy lemma for Kronecker sequences, $b-1$ integer multiples $n_i\\alpha$ with $\\|\\{n_i\\alpha\\}\\|_\\infty\\le N^{-r/b}$ that are linearly independent and span a hyperplane $S$; integer combinations of these vectors with $1\\le k_i n_i\\le N^b$ are then counted as points of $S$. For quantum dynamics, the load-bearing transfer is the large deviation theorem for Green's functions of long-range quasi-periodic operators (which supplies a semi-algebraic exceptional set $\\Theta_{N_1}$ of measure at most $e^{-N^c}$) together with a criterion that bounds the $p$-th moment by $(\\log T)^{p/\\delta}$ whenever the number of bad sites in an interval of length $N$ is at most $N^{1-\\delta}$; the discrepancy estimate supplies $\\delta=1-\\tau(b-1)-\\varepsilon$.","core_discovery":"The paper's central claim is that the almost-everywhere semi-algebraic discrepancy exponent for multi-frequency shift sequences is $1-1/b$, with a lower-bound exponent $1-1/(b+1)$ that approaches it as $b\\to\\infty$. Theorem 1.2 says that for $\\alpha\\in WDC(\\gamma,\\tau)$, every semi-algebraic $S\\subseteq[0,1]^b$ of degree $B$ and Lebesgue measure at most $\\eta$, and every $N$ with $\\log N < (2\\tau b)^{-1}\\log(1/\\eta)$, satisfies $\\#\\{1\\le n\\le N:\\theta+n\\alpha\\bmod\\mathbb{Z}^b\\in S\\}\\lesssim B C(b) N^{\\tau(b-1)}$; taking $\\tau=1/b+\\varepsilon$ gives $\\le N^{(b-1)/b+\\varepsilon}$ for almost every $\\alpha$. Theorem 1.3 complements this with a hyperplane $S\\subseteq[0,1]^b$ for which $\\#\\{1\\le n\\le N:n\\alpha\\bmod\\mathbb{Z}^b\\in S\\}\\ge N^{(b-1)/(b+1)-\\varepsilon}$ for almost every $\\alpha$ and all sufficiently large $N$. The transfer to quantum dynamics is Theorem 1.5: for $\\alpha\\in WDC(\\tau)\\cap DC(\\tau')$ and nonconstant real-analytic $V$, there is a threshold $\\lambda_0$ such that for $\\lambda>\\lambda_0$, $\\sup_{\\theta}\\langle |X_{H_{\\theta,\\alpha}}|^p\\rangle_\\psi(T)\\le(\\log T)^{p/(1-\\tau(b-1))+\\varepsilon}$, and Corollary 1.6 gives the almost-everywhere version $(\\log T)^{pb+\\varepsilon}$.","pith_inferences":["The gap between the upper exponent $(b-1)/b$ and the constructed lower exponent $(b-1)/(b+1)$ suggests the almost-everywhere discrepancy exponent may be strictly between these values; replacing hyperplanes by higher-degree semi-algebraic witnesses is a concrete way to test whether the upper bound is attainable.","Because the covering argument uses only the one-dimensional small-multiples condition $\\|n\\alpha\\|\\ge\\gamma/|n|^\\tau$, the same upper bound should hold for Kronecker sequences under any nonlinear reparametrization that preserves this condition, provided the target sets remain semi-algebraic after the reparametrization.","The logarithmic moment bound in Theorem 1.5 is an upper bound; if quasi-periodic long-range models in this regime instead exhibit power-law spreading, the discrepancy method alone cannot reveal it, since its input is already asymptotically sharp in dimension.","A $C^k$ (finite-smoothness) version of the potential would remove the semi-algebraic approximation of the exceptional set on which the proof relies; whether the exponent persists for smooth but non-analytic potentials is an open question the paper does not address."],"forward_implications":["For almost every $\\alpha\\in\\mathbb{T}^b$, every semi-algebraic set of small measure is visited at most $N^{(b-1)/b+\\varepsilon}$ times among the first $N$ shifts, improving the previous best exponents $1-1/(b^2(b-1)+b)$ and $1-1/(2b)$ to $1-1/b$.","For $\\alpha\\in WDC(\\tau)\\cap DC(\\tau')$ and nonconstant real-analytic $V$, the phase-uniform moment satisfies $\\sup_\\theta\\langle|X|^p\\rangle_\\psi(T)\\le(\\log T)^{p/(1-\\tau(b-1))+\\varepsilon}$; for almost every $\\alpha$ this reads $(\\log T)^{pb+\\varepsilon}$.","The lower-bound hyperplane construction shows the upper estimate cannot be improved by more than a factor of roughly $(b+1)/b$ in the exponent, so the discrepancy result is asymptotically sharp as $b\\to\\infty$.","The counting argument gives a uniform-in-phase statement for all $\\theta$, rather than only a phase-averaged localization bound.","The proof of the upper bound works under the weaker one-dimensional weak Diophantine condition, which makes the argument shorter than earlier approaches based on the full multi-frequency Diophantine class."],"supporting_citations":[{"why":"Supplies the semi-algebraic covering lemma (Corollary 9.6) used to cover $S$ by $B C(b)\\epsilon^{1-b}$ balls, and the Green's function perturbation framework used in the LDT transfer.","marker":"[Bou05]"},{"why":"Supplies the metric discrepancy bound $D_N(\\{n\\alpha\\})\\le C(\\alpha)N^{-1}(\\log N)^{b+2}$ that produces the independent small vectors in Proposition 3.2.","marker":"[Sch64]"},{"why":"Provides the large deviation theorem for Green's functions of long-range quasi-periodic operators, whose exceptional set is the semi-algebraic object counted in Theorem 1.2.","marker":"[Liu22]"},{"why":"Provides the criterion converting a bound on the number of bad Green's-function sites into the logarithmic moment bound used to state Theorem 1.5.","marker":"[Liu23]"},{"why":"Gives the previous best discrepancy/dynamical exponent $N^{1-1/(b^2(b-1)+b)+\\varepsilon}$ that the new exponent $1-1/b+\\varepsilon$ improves.","marker":"[HJ19]"},{"why":"Gives the previous bound $N^{1-1/(2b)+\\varepsilon}$ for smooth potentials, another baseline the present results improve.","marker":"[JP22]"}],"fun_headline_variants":["Sharp discrepancy for multi-frequency shifts yields logarithmic quantum bounds","Multi-frequency shift discrepancy exponent pinned down","Logarithmic quantum dynamics from sharp discrepancy estimates","Curved-set discrepancy for shifts sharpened, aiding quantum bounds"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The lower-bound proof assumes that every integer combination $n=\\sum_{i=1}^{b-1}k_i n_i$ with $1\\le k_i n_i\\le N^b$ and $k_i\\|\\{n_i\\alpha\\}\\|_\\infty<1/b$ automatically lies in the counting window $1\\le n\\le N$; the constraints only force $n_i\\le N^{1+2\\varepsilon}$ and $k_i\\le N^b/n_i$, so $n$ can be as large as $(b-1)N^b$, and the proof drops the $n\\le N$ restriction when passing from the counting set to the product of $k_i$-counts.","fun_headline_variants_meta":{"raw":{"variants":["Sharp discrepancy for multi-frequency shifts yields logarithmic quantum bounds","Multi-frequency shift discrepancy exponent pinned down","Logarithmic quantum dynamics from sharp discrepancy estimates","Curved-set discrepancy for shifts sharpened, aiding quantum bounds"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000809,"raw_usage":{"total_tokens":3548,"prompt_tokens":943,"completion_tokens":2605,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":559,"completion_tokens_details":{"reasoning_tokens":2544}},"tokens_in":559,"tokens_out":2605,"duration_ms":22688,"temperature":1.0,"reasoning_tokens":2544,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T14:02:07.190812+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Verify the Section 3.2 counting identity in dimension $b=2$: Proposition 3.2 produces $n_1\\in[1,N^{2(1+2\\varepsilon)/3}]$ with $\\|\\{n_1\\alpha\\}\\|_\\infty\\le N^{-1/3}$. The proof counts every $k_1$ satisfying $1\\le k_1 n_1\\le N^2$ and $k_1\\|\\{n_1\\alpha\\}\\|_\\infty<1/2$, and then treats $k_1 n_1$ as lying in $[1,N]$. Computing, for an explicit Diophantine $\\alpha$ and large $N$, whether the number of such $k_1$ with $1\\le k_1 n_1\\le N$ is at least $N^{1/3-\\varepsilon}$ would settle Theorem 1.3; a shortfall means the product count overcounts points outside the window.","supporting_citations":[],"review_version":1}