{"id":"a9071326-4a1e-44fa-9087-ec7661fdec86","arxiv_id":"2608.12204","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For a rational quadratic disk-slab on a shifted integer grid with the Babai point outside the constructed inner ellipse, exactly four integer levels of the reduction-selected functional can be occupied.","lead":"This paper proves that lattice points inside a planar ellipse cut by a strip can occupy at most four layers of a particular integer direction, in the case where the Babai nearest-plane point falls outside a chosen inner ellipse. A rational example reaches four layers, so the bound is exact.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Hard-mode exclusion rests on an unprinted 17,640-coefficient Bernstein certificate; without a checkable certificate, Theorem 7.1 is conditional on an unverified computation.","rationale":"I reviewed the chain: strict diameter estimate, five-level trichotomy, central-block exclusion via the nearest-integer inequality, side normalization to eight modes, and the shared-cap contradiction. Each step is internally consistent; the central-block proof and the side-mode cell table check out. The delta intervals and the bounds in (30) follow from the stated parameter ranges. The displayed '725/4' in Section 5 is a typo: the bracket equals 145/4, and the final 85/8 is correct once this is fixed; this is minor. The genuinely load-bearing assumption is the Bernstein certificate. Proposition 5.2 is the only step that cannot be checked by a reader from the printed text. The appendix gives per-mode minima and a hash digest but no coefficients and no code. If the certificate is wrong, the hard modes may survive, and the upper bound L*_Z <= 4 could fail. Because this is a verification gap rather than an identified logical contradiction, conditional acceptance remains the appropriate verdict; the reader's weakest-assumption analysis already identified exactly this point, so no verdict change is warranted.","tokens_in":12266,"tokens_out":19834,"duration_ms":152153,"concrete_test":"Provide the full coefficient table or the ancillary program and independently recompute the Bernstein coefficients of B for each of the four hard modes using exact rational arithmetic (e.g., a fresh implementation of the conversion formula in Appendix A.2 in Sage, sympy, or Mathematica). Verify that all 2205 coefficients per mode are strictly positive and that the global minimum is 625/2048; also confirm that the corrected lower-bound bracket in Section 5 is 145/4, not 725/4. If any coefficient is nonpositive or the minimum differs, Proposition 5.2 is not established and the proof of Theorem 7.1 must be re-examined.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim L*_Z = 4 depends on Proposition 5.2, which excludes the four hard modes by proving the polynomial B = r*delta^2*(7p-30g)^2 - 400*t^2*g^2 is positive on the closed box in variables (R,E,H,Q). Positivity is certified by converting B to the Bernstein basis and asserting that all 17,640 coefficients are positive with minimum 625/2048. Appendix A.2 lists only per-mode minima and a hash digest; it does not print the coefficients, the explicit polynomial B, or the program. Section 8 states that an ancillary program 'exact_bernstein_certificate.py' exists, but the manuscript text does not include it. If any Bernstein coefficient is nonpositive or the conversion has a sign error, the inequality sqrt(r)*|delta|*(7p-30g) > 20*t*g may fail for some admissible parameters, and the contradiction in Proposition 6.1 would not close. The proof has no fallback for the four hard modes; the easy modes use only elementary bounds. Confidence is further eroded by a displayed arithmetic slip in Section 5: the lower bound for 7*(3-vartheta)^2 - 30*vartheta^2 is written as 725/4, but the correct value is 145/4 (the final 85/8 is still correct if 145/4 is used). This is a typo, not a logical gap, but it shows the exact arithmetic needs independent checking.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies lattice points in a rational positive-definite quadratic sublevel set intersected with a slab, after a deterministic delta=3/4 reduction selects a primitive-dual integer functional w and an inner ellipse. In the strict-failure branch where the Babai point lies outside that ellipse, the paper claims that the occupied integer levels of w number at most four, and that an explicit rational instance attains four, so L*_Z=4. The proof proceeds by a strict diameter estimate bounding the real body to five levels, a five-level trichotomy reducing to three blocks, a nearest-integer inequality excluding the central block, and layer reversal and cell localization reducing the side blocks to eight modes; four easy modes are handled by a shared-cap inequality and the four hard modes by an exact Bernstein certificate with 17,640 positive rational coefficients. An appendix provides the rational attainment data and the claimed certificate minima.","tokens_in":12592,"tokens_out":13079,"duration_ms":111728,"significance":"If the certificate is independently confirmed, the result is a sharp, nontrivial constant in a fixed-direction arithmetic-width setting, and it usefully distinguishes the integer-occupied layer count from the real layer-intersection count. The analytic parts of the proof are self-contained and carefully argued, and the sharpness construction is explicit and checkable. The principal deficit is reproducibility: the certificate that excludes the four hard modes is not actually included in the manuscript, only per-mode minima and a hash digest, so the central theorem is currently conditional on an unverified finite computation. Once the certificate is supplied and the exact arithmetic is checked, the result merits publication.","major_comments":[{"comment":"The proof of the hard-mode inequality (32) depends on the assertion that all 17,640 Bernstein coefficients of the polynomial B are positive, with minimum 625/2048, but the manuscript prints neither the explicit polynomial B for each mode, nor the coefficient list, nor the program that produced the certificate. The per-mode minima and the hash digest are not independently checkable by a reader. Since Proposition 6.1 uses (32) in the final contradiction and the four hard modes have no alternative elementary bound, Theorem 7.1 cannot be considered established unless the exact program or a machine-readable certificate, including the eight polynomials and their Bernstein expansions, is provided and the digest can be reproduced.","section":"Appendix A.2 and Proposition 5.2"},{"comment":"The displayed lower bound is incorrect as written: the bracket '725/4' should be '145/4', because with vartheta in [23/50,1/2] the minimum of 7(3-vartheta)^2 - 30 vartheta^2 is 145/4, not 725/4. The final value 85/8 follows only after this correction, and the derivation displayed in the paper does not justify the claimed inequality as printed. The same exact arithmetic should be re-checked throughout, since this line is used to establish the positivity of 7p-30g in the hard-mode argument.","section":"Section 5, proof of 7p-30g > 0"},{"comment":"The reproducibility statement says that the proof relies on the rational coefficient inequalities and not on a digest or an unpublished software result, but the appendix does not actually exhibit the coefficients or the program. The statement overstates the state of reproducibility. The authors should either include the ancillary program exact_bernstein_certificate.py, or provide a complete machine-readable table of all 17,640 Bernstein coefficients (or the eight explicit polynomials together with the conversion code), so that a reader can verify the positivity claim independently.","section":"Section 8 and Appendix A.2"}],"minor_comments":[{"comment":"The symbol g is used both for the auxiliary function in Lemma 3.1 and for the quantity e^2 + r vartheta^2 in (27); renaming one of them would avoid confusion.","section":"Section 5, notation"},{"comment":"The text says that boundary cases with e2=1/2 or mu=1/2 are retained only in a larger closed relaxation, but it would help to state explicitly which strict inequalities remain valid on that closed box and why the certificate's positivity on the larger box is sufficient for the strict inequality (32) on the original domain.","section":"Section 5, boundary conventions"},{"comment":"The title could state more prominently that the four-layer bound is established for the strict-failure branch of the specified deterministic reduction; the current wording is broader than the theorem.","section":"Title and abstract"}],"recommendation":"major_revision","confidential_remarks":"The non-certificate parts of the proof appear coherent and I found no circularity. The central obstacle is the unverifiable Bernstein certificate for the four hard modes; if the exact program or full coefficient data is supplied, and the arithmetic slip in Section 5 is corrected, I would support publication. The current version is not verifiable as it stands."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a real result, not a stunt. The sharp four-layer constant L*_Z = 4 for the strict Babai-failure branch is new, and the proof is mostly elementary. The one load-bearing step is a large exact Bernstein certificate that the paper doesn't actually give you, so as printed the theorem is conditional on an unverified computation.\n\nWhat is actually new: the exact extremal constant, the five-level trichotomy, the eight-mode reduction, and the rational attainment example. The reduction to three block types via the strict diameter estimate is clean, and the nearest-integer inequality for the central block is a nice piece of work. The shared-cap contradiction for the easy modes is also solid and self-contained. The attainment example is fully explicit and checks out arithmetically.\n\nThe soft spot is exactly where the reader put their finger: Proposition 5.2 rests on 17,640 positive Bernstein coefficients, and Appendix A only gives the per-mode minima plus a hash digest. That is not enough for a reader to verify the positivity without rerunning the computation themselves. Section 8 says an ancillary program exists, but it isn't in the manuscript. For a computer-assisted proof this is a normal and fixable gap: put the code or the full coefficient table in the ancillary files, or better, make the certificate machine-checkable (e.g., a text file with all coefficients). Until then, Theorem 7.1 is conditional on a computation nobody can see.\n\nThere is also a minor arithmetic slip in Section 5: the displayed lower bound for 7(3-vartheta)^2 - 30*vartheta^2 is written as 725/4, but the value at vartheta=1/2 is 145/4. The final conclusion 85/8 is actually correct if you compute the intended expression (-30/4 + (1/2)*(145/4) = 85/8), so this is a typo, not a hole. Still, it shows the exact arithmetic needs a careful pass.\n\nThe class is narrow: planar disk-slab, strict failure branch. The paper is honest about that, and the constant is exact for that class. I don't see circularity or hidden fitting; the reduction and Babai failure are fixed algorithmic inputs.\n\nWho's this for: people working on lattice enumeration, fixed-dimension integer programming, or arithmetic width. It gives a sharp worst-case level count for one branch where previously there was no exact constant.\n\nRecommendation: send it to peer review. The mathematical structure is coherent and the result is worth checking. Ask for the certificate/code before final acceptance. If the computation confirms, the paper is publishable as is modulo the typo.","headline":"Genuinely new sharp four-layer constant for a narrow but well-defined branch, with a clean proof structure; just make the 17,640-coefficient Bernstein certificate actually available before publication.","tokens_in":13072,"tokens_out":5530,"would_cite":false,"duration_ms":38620,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11H06","90C10"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that every admissible planar disk-slab instance in the strict-failure branch has at most four occupied integer levels of the reduction-selected functional, and a rational instance attains four.","keywords":["lattice enumeration","arithmetic width","Babai nearest plane","planar convex bodies","Bernstein certificates","integer programming","lattice reduction","sharp bound"],"falsifier":"Recompute the Bernstein coefficient table for the four hard modes in independent exact rational arithmetic and check whether every entry is strictly positive; a single nonpositive coefficient would invalidate Proposition 5.2 and open the possibility of a five-level counterexample.","tokens_in":12097,"feed_emoji":"📐","tokens_out":10151,"duration_ms":91988,"temperature":0.7,"pith_summary":"The paper asks how many integer levels of a deterministically selected direction can contain lattice points when a rational ellipse on an affine lattice coset is cut by a slab. In the nondegenerate strict-failure branch, where the Babai nearest-plane point lies outside the inner ellipse, it proves that the answer is at most four, and it exhibits a rational instance with exactly four occupied levels. The proof rules out every possible block of five consecutive occupied levels, so the bound is exact rather than an artifact of the particular example. A reader should care because the constant four is uniform: it does not grow with the size of the disk or slab once the reduction branch is fixed.","feed_headline":"Lattice points in a planar disk-slab fill at most four layers","feed_subtitle":"A rational example attains four; the exact worst-case count for integer-occupied slices is four.","key_machinery":"The carrying object is the reduction-selected primitive-dual functional $w = U^{-T}u_2$ together with the inner-ellipse gauge $H(t,j) = \\alpha(t-e_1+\\mu j)^2 + \\beta(j-e_2)^2$. A deterministic $\\delta = 3/4$ reduction fixes the parameters $\\alpha,\\beta,\\mu$ and the half-open error cells; strict Babai failure gives $E_B > 1$ and forces $\\beta > 4/3$, which limits occupation to the seven levels $\\{-3,\\dots,3\\}$. A strict diameter estimate $\\operatorname{diam}(\\tilde C) < 4\\sqrt2$ then bounds the span of occupied levels, a nearest-integer inequality kills the central block, and cell localization reduces the two side blocks to eight modes. Four of those modes are excluded by the shared-cap inequality; the remaining four require proving positivity of a four-variable polynomial, which is done by converting its power coefficients to Bernstein coefficients on a rational box and verifying that all 17,640 coefficients are positive.","core_discovery":"The central claim is $L^*_Z = 4$: for every admissible instance satisfying the strict-failure assumptions, the set $J_Z(I)$ of occupied integer levels of the reduction-induced primitive-dual functional $w = U^{-T} u_2$ has size at most four, and there is a rational admissible instance with $J_Z = \\{-2,-1,0,1\\}$. The proof first shows that at most seven candidate levels can be occupied, then uses a strict disk-slab diameter estimate to compress any hypothetical five-level occupation into one of three consecutive blocks. A nearest-integer inequality excludes the central block, while layer reversal and cell localization reduce the side blocks to eight explicit modes, of which four violate a shared-cap inequality and four are excluded by an exact Bernstein certificate whose 17,640 coefficients are all positive. Corollary 7.3 then sharpens the uniform bound to $L^*_Z = 4$.","pith_inferences":["Beyond the paper, the Bernstein-certificate exclusion is a finite exact-rational check that could be reused for other eight-mode reductions with the same cap structure, provided the mode-specific polynomials are recomputed.","A testable extension is to vary the deterministic parameters, such as the reduction rounding rule or the value of $\\delta$, and search for a five-level instance; finding one would show that the constant four depends on those conventions.","The paper gives no information about higher dimensions; one plausible direction is that the corresponding worst-case number of occupied levels grows with dimension, but the planar proof offers no direct route to such a statement."],"forward_implications":["No admissible instance in the strict-failure branch can occupy five or more levels; the only possible five-level blocks are the three listed in Proposition 3.2, and each is excluded.","The rational four-level fixture shows the bound cannot be improved, and in that fixture the real body still intersects five levels, so integer occupation and real layer intersection are genuinely different counts.","Because the safe candidate list is fixed to seven levels and the diameter estimate is uniform, the worst-case count four holds across all sizes of the disk and slab once the reduction branch is fixed.","The proof's exclusions are deterministic finite checks, so any search for occupied levels of the selected functional can stop once four levels have been found."],"supporting_citations":[{"why":"Defines directional arithmetic width, the counting object that $J_Z(I)$ specializes to one selected direction.","marker":"[5]"},{"why":"Introduces the vertex-based layer number whose planar worst case the paper computes for a specific functional.","marker":"[6]"},{"why":"Supplies the Babai nearest-plane construction whose strict failure defines the branch under study.","marker":"[3]"},{"why":"Provides the fixed-dimensional integer-programming setting that motivates bounding the number of occupied levels.","marker":"[10]"}],"fun_headline_variants":["Four layers max: sharp bound for integer-occupied disk-slab slices","Disk-slab lattice points occupy at most four integer levels","Sharp four-layer result: integer-occupied slices of disk-slab at most four","Lattice points in disk-slab fill at most four layers: sharp bound"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof's hard-mode exclusion rests on the claim that all 17,640 exact rational Bernstein coefficients in the certificate are positive; if even one coefficient were nonpositive, the exclusion of the four hard modes would not follow.","fun_headline_variants_meta":{"raw":{"variants":["Four layers max: sharp bound for integer-occupied disk-slab slices","Disk-slab lattice points occupy at most four integer levels","Sharp four-layer result: integer-occupied slices of disk-slab at most four","Lattice points in disk-slab fill at most four layers: sharp bound"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000562,"raw_usage":{"total_tokens":2640,"prompt_tokens":887,"completion_tokens":1753,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":503,"completion_tokens_details":{"reasoning_tokens":1674}},"tokens_in":503,"tokens_out":1753,"duration_ms":13139,"temperature":1.0,"reasoning_tokens":1674,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:15:44.817001+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the Bernstein coefficient table for the four hard modes in independent exact rational arithmetic and check whether every entry is strictly positive; a single nonpositive coefficient would invalidate Proposition 5.2 and open the possibility of a five-level counterexample.","supporting_citations":[{"cited_title":"An arithmetic measure of width for convex bodies","cited_arxiv_id":"2509.04726","evidence_quote":"Defines directional arithmetic width, the counting object that $J_Z(I)$ specializes to one selected direction."},{"cited_title":"Lattice-free polytopes and their diameter.Discrete & Compu- tational Geometry, 13(1):59–75, 1995","cited_arxiv_id":null,"evidence_quote":"Introduces the vertex-based layer number whose planar worst case the paper computes for a specific functional."},{"cited_title":"On Lov´ asz’ lattice reduction and the nearest lattice point problem.Combina- torica, 6(1):1–13, 1986","cited_arxiv_id":null,"evidence_quote":"Supplies the Babai nearest-plane construction whose strict failure defines the branch under study."},{"cited_title":"Lenstra, Jr","cited_arxiv_id":null,"evidence_quote":"Provides the fixed-dimensional integer-programming setting that motivates bounding the number of occupied levels."}],"review_version":1}