{"id":"b0dfa493-1768-452f-ab47-bb56a663ccc5","arxiv_id":"2501.16779","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"By systematically codifying and optimizing known exponent relations in the ANTEDB database, the paper derives four new exponent pairs, new zero density bounds, and new additive energy bounds for the Riemann zeta-function.","lead":"The authors introduce a database that stores known bounds and logical relations between many analytic number theory exponents, then use it to automatically combine those known results into new numerical bounds. They report four new exponent pairs, improved zero density estimates for the Riemann zeta-function, and new additive energy estimates for its zeroes.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The headline results are delegated to an uncertified ANTEDB computation: Theorem 20, the bold rows of Table 2, and Theorem 64(ii)-(ix) have no complete human-readable proof in the paper, so a code bug or mis-encoded hypothesis would invalidate the central claims.","rationale":"The reader's weakest assumption -- correctness and fidelity of the ANTEDB code -- is exactly the load-bearing point. The four new exponent pairs, the bold zero density estimates, and most of the additive energy bounds are generated by the database, and the paper explicitly states that the code is not formally certified. This is not an ad hominem or a disagreement with the mathematical strategy; the paper is coherent and the derivations are plausible. The issue is verifiability: a single error in the code or in the transcription of a cited theorem into a Hypothesis object would break the central claims, and the printed text does not contain enough detail to rule that out. The concrete test I propose is deliberately narrow: re-running the relevant ANTEDB routines at a fixed commit, plus an exact rational-arithmetic check of the Theorem 20 pairs against the piecewise beta-envelope, would settle whether the code actually proves what the paper claims. Because the authors themselves flag the lack of formal certification, a CONDITIONAL verdict is appropriate; my concern does not change that verdict, so I recommend UNCHANGED.","tokens_in":39527,"tokens_out":20828,"duration_ms":181339,"concrete_test":"Pin the ANTEDB repository at the commit used for the v1 posting and re-run, in exact rational arithmetic, the routines that emit the four Theorem 20 pairs, the bold rows of Table 2, and the certificates for Theorem 64(ii)-(ix). Then independently verify the Theorem 20 certificates by checking, for each listed pair (k,l), that k+(l-k)alpha is at least every beta-bound in Table 1 at all breakpoints, and audit the Lemma 14 encoding to confirm that D(k,l) is promoted to an exponent pair only when the D-line dominates the auxiliary line on the whole interval [0,1]. Any mismatch between the claimed outputs and this independent check means the conditional accept should be withdrawn.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claims are not self-contained. Theorem 20 is proved only by \"further computer calculation\" via ANTEDB; the bold zero density rows in Table 2 rely on Theorem 51 and Corollary 53, whose vertex choices were found with computer assistance; and Theorem 64 gives a proof only for part (i), with parts (ii)-(ix) deferred to the ANTEDB. Section 1.2 explicitly concedes that the ANTEDB routines \"are not formally certified to be error-free.\" Since the optimization combines many input theorems with different hypotheses, a single mis-encoded hypothesis, an incorrect inference from a piecewise beta-bound to a global exponent pair, or an arithmetic bug in the polytope intersection would invalidate the headline results even if every cited input theorem is true. One concrete place this matters is Lemma 14: as printed it gives only a max-bound on beta, and the following sentence claims this implies D(k,l) is an exponent pair, which is not immediate and needs the code (or a further argument) to verify. The human-readable proofs for Theorem 51 and Theorem 64(i) also contain numerous \"one can check\" numerical inequalities, so they do not by themselves certify the full set of new bounds.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces the Analytic Number Theory Exponent Database (ANTEDB), a Python-based system for recording known bounds and relations among exponents in analytic number theory and for optimizing them via polytope operations. Using this framework, the authors claim four new exponent pairs (Theorem 20), several new zero-density estimates for the Riemann zeta-function (Table 2, via Theorems 50, 51, and 53), and new additive-energy estimates for the zeroes (Theorem 64). The paper also abstracts a number of standard relationships, such as Huxley subdivision, the duality between exponent pairs and the β function, large-value-to-zero-density conversions, and Heath-Brown's additive-energy relation, presenting many of these as human-readable lemmas.","tokens_in":39737,"tokens_out":4496,"duration_ms":41553,"significance":"If the new numerical claims are correct, they constitute genuine quantitative advances in three related areas: exponent pairs, zero-density estimates, and additive energy of zeta zeroes. A further strength of the paper is its systematic abstraction of relationships that often appear only implicitly in the literature, and the public availability of the ANTEDB code and database; this has the potential to make future incremental improvements easier to propagate. The paper is also commendably explicit in several places about the limits of the computer assistance (e.g., Section 1.2 concedes that ANTEDB routines are not formally certified), and many of the structural lemmas are given with human-readable proofs. However, because several headline results are delegated to computations whose correctness is not established within the manuscript, the central claims are not currently self-contained.","major_comments":[{"comment":"The proof of Theorem 20 consists of the sentence 'The claim follows from Lemma 15 after some further computer calculation.' Since these four exponent pairs are a headline result, and since Section 1.2 explicitly states that the ANTEDB routines 'are not formally certified to be error-free', the central claim is not verified within the paper. A single miscoded hypothesis or arithmetic bug in the polytope intersection would invalidate all four pairs. Please provide either an explicit, independently checkable transcript of the computation (for instance, the exact polytope data and the piecewise-linear β bounds used for each of the four pairs) or human-readable derivations that the reader can verify without executing external code.","section":"Theorem 20; Section 1.2"},{"comment":"Lemma 14 gives only an upper bound for β(α) in terms of an exponent pair (k, ℓ), namely β(α) ≤ max{k₁ + α(ℓ₁ − k₁), 1/12 + 2α/3}. The following sentence asserts that in practice this implies D(k, ℓ) is an exponent pair, but this implication is not immediate: it requires checking, via Lemma 15, that k₁ + (ℓ₁ − k₁)α is bounded by the displayed maximum for all α ∈ [0,1], with appropriate splitting of the range of α. That verification is omitted, and Table 1 relies on the D-process in several rows. Since Theorem 20 depends on Table 1, this gap propagates to the headline exponent-pair results.","section":"Lemma 14; Table 1"},{"comment":"The proof of Theorem 53 asserts that S(σ) is a convex polygon ('one may verify'), lists eight exponent pairs 'found with the aid of computer assistance', and then states that applying Theorem 52 and taking a minimum yields the piecewise formulas. The manuscript does not show that each listed pair lies in the required region S(σ) on each stated σ-interval, nor does it display the endpoint computations that produce the interval boundaries (such as 2841/3016, 859/908, 1625/1692, etc.). Because Theorem 53 supplies several bold rows of Table 2, which are advertised as new zero-density estimates, this leaves a load-bearing part of the paper unverifiable without the external database. Please supply the verified optimization data or a detailed enumeration of the interval checks.","section":"Theorem 53; Table 2"},{"comment":"Only part (i) of Theorem 64 is proved in the paper; parts (ii)–(ix) are deferred to the ANTEDB. Section 1.2 concedes that the ANTEDB routines are not formally certified to be error-free. Since the additive energy estimates are one of the three headline contributions, the current manuscript does not provide a complete proof of eight of the nine claimed cases. Even part (i) contains many 'one can check' inequalities, which are likely routine but are not shown. Please either provide complete proofs for all parts of Theorem 64 or make available a formally verified or independently checkable computation transcript that certifies the deferred parts.","section":"Theorem 64"}],"minor_comments":[{"comment":"In the definition of an exponent pair, the phrase 'for all T ≥ N ≤ 1' should presumably read 'for all T ≥ N ≥ 1'.","section":"Definition 11"},{"comment":"In the definition of the large value energy region, the condition 'V = N^{σ+o(1)}' appears twice; the second occurrence is redundant and should be removed or replaced.","section":"Definition 58"},{"comment":"Several rows in Table 2 are attributed to 'Corollary 53', but the text labels the relevant result as Theorem 53. Please correct the cross-reference.","section":"Table 2"},{"comment":"In the proof of Theorem 51, the second case is stated as '16/21 ≤ σ ≤ 38/49', while the theorem statement restricts to '17/22 ≤ σ ≤ 4/5'. Please clarify whether the proof is intended for the wider range or align the statement and proof.","section":"Theorem 51"},{"comment":"The phrase 'one can check' appears in several places where numerical inequalities are asserted. For a computer-assisted paper, it would considerably aid verification if a supplementary file enumerated these inequalities or if the code scripts that certify them were clearly designated and archived.","section":"Various proofs, e.g., Theorem 51 and Theorem 64"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern about uncertified computer assistance is legitimate and lands directly on the central claims. The paper would be strengthened by making the ANTEDB computations reproducible and verified in a way that a reader can check without trusting the Python implementation. This is a fixable issue, so I do not recommend rejection; however, in its current form the manuscript does not yet provide complete proofs of its headline results."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I'll cut to the punchline: this is a serious results paper with genuine new numerical inequalities, and it should be refereed. The four exponent pairs, the bolded rows in Table 2, and the additive energy bounds in Theorem 64 are not in the prior literature, and they appear to follow from known theorems combined through the ANTEDB optimizer rather than from curve-fitting to the outputs. The database framing is a useful contribution by itself.\n\nWhat it does well: the nonstandard asymptotic setup is handled cleanly and makes the epsilon-management much less painful; the abstraction of Huxley subdivision, large value exponents, and zero density from large values is genuinely clarifying. The paper recovers Ingham, Huxley, and Guth-Maynard as special cases of one small set of corollaries. Many of the auxiliary lemmas (Bourgain large values theorem Theorem 32, Lemma 39, Lemma 40, the reflection Lemma 34) have honest human-readable proofs, and Theorem 50 and Theorem 51 give substantial 'one can check' details.\n\nWhere the soft spots are, in proportion: the headline results are not fully self-contained. Theorem 20 is proved by 'further computer calculation' with no details. Theorem 64 gives a full proof only for (i) and sends (ii)-(ix) to the ANTEDB. The authors state in Section 1.2 that the ANTEDB routines are not formally certified to be error-free. That is a real limitation for a paper whose central claims depend on those routines. The stress-test note points at Lemma 14: as printed, the D-process is stated as a max-bound on beta, and the jump to 'therefore D(k,l) is an exponent pair' is not immediate without invoking Lemma 15 plus a range check. That is worth fixing as an expositional gap, but I don't think it's a load-bearing flaw. The self-citation concern is minor: [TY23] is prior work by two of the authors and is a legitimate input. The circularity burden is low.\n\nBottom line: this paper is for analytic number theorists who want the current best numerical bounds and for anyone building similar automated databases. It deserves serious referee time. My recommendation: send it to peer review with the condition that the authors supply a versioned snapshot of ANTEDB plus independent verification or formalisation of the computer-assisted parts, or at minimum a detailed appendix for Theorem 20 and Theorem 64(ii)-(ix). With that, accept.","headline":"A serious results paper with genuinely new numerical bounds, but the headline claims lean on an uncertified database computation and need a versioned, checkable artifact before they should be accepted.","tokens_in":40298,"tokens_out":2166,"would_cite":true,"duration_ms":20647,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11L07","11M06","11T23"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that a systematic database of exponent relations yields four new exponent pairs, new zero-density bounds, and new additive-energy bounds for the Riemann zeta-function.","keywords":["exponent pairs","zero density estimates","additive energy","Riemann zeta-function","large value theorems","computer-assisted proof","exponential sums","analytic number theory"],"falsifier":"An independent re-implementation that starts from the same stated inequalities and recomputes the polytope intersections would settle the central claim: if any of the four exponent pairs in Theorem 20 or any bold zero-density entry in Table 2 is not implied by the stated inputs, the paper's conclusion fails.","tokens_in":39287,"feed_emoji":"🧮","tokens_out":7458,"duration_ms":64840,"temperature":0.7,"pith_summary":"The paper argues that the best-known bounds on exponential-sum, zero-density, and additive-energy exponents for the Riemann zeta-function can be improved by systematically collecting every known relation among these exponents and running a computer search over their consequences. It reports four new exponent pairs, several new zero density estimates, and new upper bounds on the additive energy of zeroes of the zeta-function. The point is that these are not new analytic inputs but optimal combinations of existing theorems, made visible by a database that abstracts away the routine optimization. If the claims are right, then any future improvement to a single upstream bound can be propagated automatically to the best implied bounds across the whole network.","feed_headline":"Four new exponent pairs and sharper zero-density bounds","feed_subtitle":"New database automatically combines known large-value theorems into best-implied zeta bounds.","key_machinery":"The central object is the Analytic Number Theory Exponent Database (ANTEDB), a database in which each known theorem or conjecture about an exponent is stored both as human-readable text and as an executable object with dependencies that specify how the result is proved. The carrying mechanism is polytope optimisation: each relation among exponents (an exponent pair, a large value estimate, a zero density implication, a subdivision or power-raising rule) is converted into a convex region in the space of possible exponent tuples, and intersecting these regions yields the best bounds implied by the stored inputs. A 'cheap non-standard analysis' asymptotic formalism keeps epsilon losses uniform throughout. The output is a set of implied bounds, which the routine can convert back into machine-checkable derivations and, in many cases, human-readable proofs.","core_discovery":"In the paper's own terms, the central discovery is that four points in the exponent-pair triangle—$(89/1282,997/1282)$, $(652397/9713986,7599781/9713986)$, $(10769/351096,609317/702192)$, and $(89/3478,15327/17390)$—are genuine exponent pairs, that the zero density bounds marked in bold in Table 2 are consequences of the stated large value inputs, and that for $7/10 \\le \\sigma <1$ the additive energy exponent $A^*(\\sigma)$ obeys the piecewise bounds recorded in Theorem 64. These results are reached by encoding the known bounds and the relations between them in a database, representing each relation as a polytope of feasible exponent tuples, and intersecting these polytopes to find the best implied bounds. The paper presents human-readable proofs for the new claims where possible, and stores machine-readable derivations for the rest.","pith_inferences":["Beyond the paper: the same machinery could be pointed at other exponent collections, such as those for $L$-functions or for prime gaps, and the paper notes this as a possible expansion; one testable extension is to run the optimiser on the announced $L$-function analogues.","Beyond the paper: the heuristic in Section 6.2 suggests that the bottleneck for zero density bounds is the value of $LV(\\sigma,\\tau_0)$; checking the database's polytope against the Montgomery conjecture threshold would show how close current methods come to the heuristic limit.","Beyond the paper: because the uncertified code is the weakest link, an independent implementation or formal verification of the optimisation would be a direct way to convert these numerical claims into fully certified theorem status."],"forward_implications":["The four new exponent pairs immediately improve every bound that depends on exponent pairs via the standard A/B/C/D processes, including the $\\beta(\\alpha)$ table and the large value estimates derived from it.","The new zero density estimates sharpen the best known upper bounds on $N(\\sigma,T)$ in several ranges, bringing known bounds closer to the density hypothesis without reaching it.","The additive energy bounds in Theorem 64 improve the classical estimates for $A^*(\\sigma)$ over the stated intervals, which is relevant to the distribution of primes in short intervals.","Within the database, future improvements in any single exponent input automatically yield optimised improvements to all dependent exponents, making the routine propagation of new bounds largely mechanical."],"supporting_citations":[{"why":"Supplies the recent large value estimates for Dirichlet polynomials that serve as inputs to the zero-density and additive-energy optimisation.","marker":"[GM24]"},{"why":"Supplies the classical additive-energy relation and the zero-density method that Theorem 50 and Theorem 64 extend.","marker":"[HB79c]"},{"why":"Supplies the baseline table of beta bounds and the exponent pairs that the database combines.","marker":"[TY23]"},{"why":"Supplies the large values inequality used in the optimised density bound.","marker":"[Bou00]"},{"why":"Supplies the seed exponent pair that the A and D processes convert into new pairs.","marker":"[Bou17]"},{"why":"Supplies the converse lemma that converts zero density bounds into large value bounds, used in the heuristic.","marker":"[MT24]"},{"why":"Supplies the subdivision method and the standard table entries for the beta bounds.","marker":"[Hux96]"},{"why":"Supplies the kth-derivative exponential sum estimate encoded as one of the beta bounds.","marker":"[HB17]"},{"why":"Supplies the D-process that the optimisation uses alongside the A and B processes.","marker":"[Sar95]"}],"fun_headline_variants":["Four new exponent pairs via systematic optimization","Database yields sharper zero-density and additive-energy bounds","New exponent pairs and zeta estimates from ANTEDB","Systematic approach nets four exponent pairs and more","Optimized database delivers fresh number theory bounds"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the computer code implementing the database correctly encodes every cited theorem's hypotheses and performs the optimisation without error; a bug in that code could invalidate the new bounds even though each individual input theorem is true.","fun_headline_variants_meta":{"raw":{"variants":["Four new exponent pairs via systematic optimization","Database yields sharper zero-density and additive-energy bounds","New exponent pairs and zeta estimates from ANTEDB","Systematic approach nets four exponent pairs and more","Optimized database delivers fresh number theory bounds"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000171,"raw_usage":{"total_tokens":1214,"prompt_tokens":829,"completion_tokens":385,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":445,"completion_tokens_details":{"reasoning_tokens":315}},"tokens_in":445,"tokens_out":385,"duration_ms":4278,"temperature":1.0,"reasoning_tokens":315,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T10:45:31.399188+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"An independent re-implementation that starts from the same stated inequalities and recomputes the polytope intersections would settle the central claim: if any of the four exponent pairs in Theorem 20 or any bold zero-density entry in Table 2 is not implied by the stated inputs, the paper's conclusion fails.","supporting_citations":[],"review_version":1}