{"id":"c12c495d-e740-4aab-abd9-aaa9b7f95a5d","arxiv_id":"1908.03221","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Period integrals of a hypersurface determine an ideal of polynomials, and for algebraic cycles this ideal can be lifted exactly to recover the subvariety's equations, demonstrated by proving Picard numbers 8 and 14 for two quartics.","lead":"The authors present a computer-assisted method that recovers the exact equations of curves and other subvarieties of a projective hypersurface from numerical period computations. The method yields rigorously verified Picard numbers for two explicit quartic surfaces, a case where previous methods could not certify non-extremal Picard groups.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Numerical lattice identification in §3.1 is load-bearing: exact reconstruction of 56 conics does not by itself certify that no conic class was missed.","rationale":"The reader's weakest assumption identifies the same load-bearing step: the numerical lattice Λ is found 'with high confidence' and no rigorous error bound for the 5000-digit period computation is supplied. My reading confirms that this is the single point on which the exactness of the Picard number claims turns. The algebraic framework in Section 2 is coherent, and the symbolic verification of the reconstructed planes is a genuine piece of evidence: it converts each approximated Hodge class into an actual algebraic cycle. However, the proof needs the converse direction—that no conic class was missed—and that is supplied only by the asserted containment Pic(X)|_B⊂Λ, which is not established in the paper. This does not warrant rejection: the cited [LS19] may contain such a theorem, and the missing ingredient is reproducible, explicit certification rather than a conceptual flaw. Hence the verdict should remain CONDITIONAL: the method is promising and the examples are probably correct, but the proof as written is not fully self-contained until the numerical lattice identification is made rigorous or the cited bound is stated with its hypotheses and implementation.","tokens_in":15851,"tokens_out":3482,"duration_ms":42151,"concrete_test":"Recompute the period matrix for f=x^4+x^3z-xy^3+y^4+z^4+w^4 using a certified ball-arithmetic implementation of the period algorithm in [Ser19] at 5000 digits, obtain a rigorous interval enclosure of the period isomorphism, and recompute Λ by LLL with validated error bounds. Then verify that every class δ∈H2(X,Z) with δ²=-2 and h·δ=2 (or at least all classes within the stated height bound B) is contained in the certified Λ, and that the 56 conic vectors used in Proposition 3.1 satisfy the Hodge condition within the intervals. If even one vector changes, or an additional conic class appears, the proof of Proposition 3.1 fails as written.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proofs of Propositions 3.1 and 3.2 depend on the numerically computed lattice Λ⊂H2(X,Z)≃Z22. In the proof of Proposition 3.1 the text first says Λ 'is the Picard group of X with high confidence [LS19]' and then uses a stronger, rigorous-sounding assertion: 'The lattice Λ ... contains Pic(X)|_B for some explicit B≫10^1000 [LS19].' The exact symbolic verification that 56 planes cut out bitangent conic pairs proves only that those 56 selected classes are algebraic. The conclusion that these are all conic classes, and hence that Λ is the Picard group, rests on the containment Pic(X)|_B⊂Λ, which is not proved or reproduced in this paper. If the numerical period computation or the LLL identification of Λ erred at the relevant precision, a genuine conic class could lie outside Λ and the lower-bound argument would be invalid. This is load-bearing for the central claim that the Picard numbers are exactly 8 and 14: without a certified version of the lattice step, the lower bounds are conditional, even though the upper bounds from reduction modulo 101 appear independent. The concern is not internal inconsistency in the algebraic framework, but an unstated gap between 'high confidence' numerical output and the rigorous containment statement needed for the proof.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a practical method for reconstructing subvarieties of a projective hypersurface from its periods. For an even-dimensional hypersurface X, the authors attach to each Hodge cycle δ a homogeneous ideal ~I_δ built from the kernels of period pairings; if δ=[Y], then I(Y)⊂~I_δ, and for complete intersections the ideals agree up to a known degree. Using numerical period computations (PeriodSuite), they compute approximate period matrices, extract a Hodge lattice Λ, and then, for δ a conic class, compute the kernel ~I_{δ,1}, reconstruct the plane containing the conic as an algebraic linear form, and symbolically verify the resulting plane equation. This is used to prove that two explicit quartic surfaces have Picard numbers 8 and 14, exhibiting 56 conics (respectively 102 conics and 4 lines) whose classes generate the Picard group. A second contribution is a general section on fields generated by periods, giving bounds on the degree of the field of definition of algebraic cycles.","tokens_in":16015,"tokens_out":9878,"duration_ms":105951,"significance":"If the numerical certification is made fully rigorous, the paper's main contribution is significant: it provides a genuinely practical and partially certified route from period integrals to equations of subvarieties, and it gives nontrivial, non-extremal Picard number computations with both lower and upper bounds. The theoretical part (Section 2) is carefully written and contains useful results such as Proposition 2.15 and the discussion of perfect Hodge classes. I explicitly credit the authors for making code available and for combining symbolic verification (Magma) with modular-characteristic upper bounds. The main blockage is the load-bearing numerical lattice identification, which is currently cited rather than proved in this manuscript.","major_comments":[{"comment":"The exact lower bound rk Pic(X) ≥ 8 depends on the assertion \"The lattice Λ ... contains Pic(X)|_B for some explicit B≫10^1000 [LS19].\" In the same paragraph the lattice is described as found \"with high confidence [LS19].\" The symbolic verification of the 56 reconstructed planes proves that the 56 selected classes are algebraic, but it does not prove that no conic class was missed. If the numerical period computation or the lattice identification were incorrect at the working precision, a genuine conic class could lie outside Λ and the conclusion Λ = Pic(X) would fail. Please state the precise theorem from [LS19] that gives the containment and verify its numerical hypotheses, in particular why 5000-digit periods suffice; alternatively, explicitly label the lower bound as conditional. This is load-bearing for the central claim that the Picard numbers are exactly 8 and 14.","section":"§3.1, proof of Proposition 3.1 (and parallel proof of Proposition 3.2)"},{"comment":"The upper bound rk Pic(X) ≤ 8 (respectively 14) is asserted from \"computing the Zeta function of the reduction of X over F101\" with citations [AKR10; CHK19; Cos15], but no data from this computation is reported: the zeta function, the resulting bound, or the exact code and log are absent. A reader cannot check this half of the equality without rerunning the computation from scratch. Please include the relevant output or provide scripts and logs as supplementary material.","section":"§3.1, last paragraph and Proposition 3.2"}],"minor_comments":[{"comment":"The spelling \"ˇCeck\" should be \"Čech\", and in §1.5 \"de Rahm\" should be \"de Rham\".","section":"§2.1, around Eq. (2.4)"},{"comment":"The phrase \"twisted curve\" is likely intended to be \"twisted cubic\"; please clarify the statement and proof.","section":"§2.3.2, Proposition 2.15"},{"comment":"The minimal polynomial for a1 has degree 28, and the expression for a2 is omitted because it is roughly 4000 characters; even if too long for the main text, please provide it in supplementary data so the symbolic verification is reproducible.","section":"§3.1, proof of Proposition 3.1"},{"comment":"The notation Pic(X)|_B is defined, but the value of B is not given; the phrase \"B≫10^1000\" is not an explicit bound. Please state an actual bound or give the formula for B from [LS19].","section":"§3.1, last paragraph"},{"comment":"The statements about \"randomly sample quartics\" and \"observed\" in tens of thousands of examples are empirical; please add an explicit sentence labelling these as experiments rather than proofs.","section":"§3.2.2"}],"recommendation":"major_revision","confidential_remarks":"The main risk in this manuscript is the reliance on [LS19] for the certified containment Pic(X)|_B ⊂ Λ. If that paper indeed contains a precise theorem with verifiable hypotheses, the revision could be straightforward; if not, the headline claim should be stated as conditional. I would also ask the editor to require archiving of the Magma verification scripts and the zeta function outputs used for the upper bounds."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Kaveh,\n\nThis paper has a genuinely new practical algorithm: recover the ideal of a subvariety from the kernel of the period pairing, then reconstruct exact algebraic equations from floating-point data. The method is clearly explained, and the twisted cubic analysis in Section 2 is careful and interesting. The two headline Picard number computations, however, are not as airtight as the abstract suggests. The lower bound that the Picard number is at least 8 depends on a numerically computed lattice Λ taken from [LS19], asserted to contain Pic(X)|_B for huge B. The symbolic verification of the 56 conics shows those conics exist, but it does not by itself prove their classes span a rank-8 sublattice of H^2(X,Z). Without a certified Λ or an explicit intersection matrix of the conic classes, the rank lower bound is conditional, even though the upper bounds from reduction mod 101 are solid.\n\nWhat deserves credit: Algorithm 3.1 is new and practical. The notion of perfect Hodge classes, the reconstruction of twisted cubics, and the combination of numerical periods with finite-characteristic upper bounds are all good ideas. Section 2's proofs are careful, and the paper is honest about the “high confidence” status of the lattice identification. My main complaint is that this load-bearing numerical step is not accompanied by a reproducible certificate, and the Magma scripts and period data are not shipped. These are fixable.\n\nSection 4 is mostly exposition of known results with modest bounds; it doesn’t affect the verdict.\n\nThe central claim—that you can find equations of subvarieties from periods—is believable and supported by the twisted cubic examples and by the fact that the reconstructed planes do come from period data. I would send this to a serious referee. The referee should be asked to check whether the lattice step can be made rigorous, or at least whether the authors can provide enough data to verify the rank of the conic classes independently.\n\nI’d bring it to reading group and cite the method if I worked in computational Hodge theory. But I would not quote the Picard numbers as theorems without checking the lattice step.","headline":"Genuinely useful reconstruction algorithm, but the two Picard number proofs rest on an uncertified numerical lattice step.","tokens_in":16617,"tokens_out":4394,"would_cite":true,"duration_ms":47251,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14C30","14C25","14J28"],"pacs":[],"model":"deepseek-v4-flash","headline":"Periods of a hypersurface can be used to reconstruct the equations of the subvarieties realizing its Hodge cycles, and the paper carries this out for two quartics, proving Picard numbers 8 and 14.","keywords":["periods","Hodge cycles","Picard number","quartic surfaces","algebraic cycles","subvariety reconstruction","Griffiths residues","period ideal"],"falsifier":"Recompute the period isomorphism for the surface $X=Z(x^4+x^3z-xy^3+y^4+z^4+w^4)$ with interval-arithmetic error bounds; for each of the 56 lattice classes $\\delta$ with $\\delta^2=-2$ and $h\\cdot\\delta=2$, the degree-one kernel $\\tilde I_{\\delta,1}$ must be generated by a single plane $Z(h)$ over the degree-28 field, and the symbolic intersection $X\\cap Z(h)$ must be two bitangent conics. If any one of those 56 verifications fails, or if the zeta function of the reduction modulo 101 does not bound the Picard number above by 8, the paper's central claim collapses.","tokens_in":15580,"feed_emoji":"📐","tokens_out":16462,"duration_ms":160558,"temperature":0.7,"pith_summary":"This paper claims that the periods of a projective hypersurface can be used to reconstruct, not just compute invariants of, the algebraic subvarieties lying on it. The construction attaches to each Hodge cycle an ideal of homogeneous polynomials defined by period integrals; any subvariety realizing the cycle has its ideal contained in this period ideal, and in useful cases the two agree up to the degree needed to read off equations. The authors implement the procedure numerically, reconstruct the coefficients of the equations as algebraic numbers, and verify them symbolically. As proof of concept, they determine the Picard numbers of two specific quartic surfaces, 8 and 14, by exhibiting the 56 conics or 102 conics plus 4 lines that generate the Picard lattice, with upper bounds coming from zeta functions in finite characteristic. The method matters because it replaces expensive symbolic searches through Hilbert schemes with numerical linear algebra followed by exact verification.","feed_headline":"Periods alone rebuild the curves inside quartic surfaces","feed_subtitle":"Two quartic surfaces are proven to have Picard numbers 8 and 14 by recovering their conics from period data.","key_machinery":"The central object is the period ideal $\\tilde I_{\\delta} = \\bigoplus_{u\\ge 0}\\tilde I_{\\delta,u}$, where $\\tilde I_{\\delta,u}$ is the kernel of the map $S_u\\to S_v^{\\vee}$, $p\\mapsto (q\\mapsto\\int_{\\delta}\\omega_{pq})$, with $v=N+\\frac n2 d-u$ and $\\omega_p$ the residue of the rational differential form $p\\,\\Omega/f^{\\ell+1}$ on the ambient projective space. A Hodge cycle $\\delta$ is an integral class orthogonal to half of the Hodge filtration, the natural candidate for a class of an algebraic subvariety. The ideal $\\tilde I_{\\delta}$ contains the ideal $I(Y)$ of every subvariety $Y$ with $[Y]=\\delta$; the key structural results used here are the complete-intersection description from [Dan17], which gives agreement up to an explicit degree, and the conic case $\\tilde I_{\\delta,1}=\\langle \\text{plane equation}\\rangle$ in a quartic. The numerical part computes $\\tilde I_{\\delta,1}$ by linear algebra on the period matrix; the exactness part reconstructs its coefficients as algebraic numbers and verifies the resulting equations by substitution.","core_discovery":"The central claim, stated in the authors' own terms, is that algebraic cycles on a hypersurface can be recovered from their periods: given a Hodge cycle $\\delta$, the kernel $\\tilde I_{\\delta}$ of the pairings $(p,q)\\mapsto\\int_{\\delta}\\omega_{pq}$ is an ideal that contains the ideal $I(Y)$ of any subvariety $Y$ with $[Y]=\\delta$, and in the cases treated it can be computed and then identified with $I(Y)$ after symbolic verification. The paper proves this concretely for quartic surfaces: for $X=Z(x^4+x^3z-xy^3+y^4+z^4+w^4)$ it exhibits 56 conics whose classes generate an integral lattice of rank 8, and for $X=Z(5x^4-4x^2zw+8y^4-5z^4+4zw^3)$ it exhibits 102 conics and 4 lines generating rank 14, with matching upper bounds from reduction to characteristic $\\mathbb F_{101}$. Because the period computations are numerical, the equations are first obtained approximately and then reconstructed as algebraic numbers; the subsequent symbolic check that each plane indeed cuts $X$ into two conics is what turns the numerical output into a proof. The paper also proposes the notion of a perfect Hodge class to isolate exactly when this reconstruction works, and proves that twisted cubics on quintic and higher-degree surfaces are always reconstructible.","pith_inferences":["Editorial inference: The same period-ideal pipeline should transfer to higher-dimensional cycles, such as curves on threefolds, provided the residue pairings can be computed and the subvariety is cut out by equations of low known degree; the practical bottleneck is the cost of the numerical period map, not the reconstruction step.","Editorial inference: The gap between the enormous uniform bound $M(\\rho-1)$ and the observed degree-28 field in Proposition 3.1 suggests the Galois representation on the period lattice carries much sharper information about fields of definition; tracking that representation could replace uniform bounds with cycle-specific ones.","Editorial inference: If Question 2.12 has a positive answer, the reconstruction method is complete for algebraic classes on hypersurfaces, converting Hodge-conjecture checks in computable cases into a verification problem: recover candidate subvarieties from periods and then verify their equations symbolically."],"forward_implications":["For complete-intersection subvarieties of smooth hypersurfaces, the ideal of the subvariety can be read off from its Hodge class up to an explicit degree, so low-degree cycles such as planes and conics are reconstructible in practice.","The same pair of arguments, reconstructed classes for the lower bound and zeta functions in finite characteristic for the upper bound, can determine Picard numbers exactly for other quartics with moderate rank, not just extremal ones.","Twisted cubics on quartic surfaces are reconstructible from periods exactly when $\\tilde I_{[T],1}=0$ and $\\dim\\tilde I_{[T],2}=3$ (Proposition 2.13), and on surfaces of degree at least five they are always reconstructible (Proposition 2.15).","For divisor classes on varieties with $H^1(X,\\mathcal O_X)=0$, Theorem 4.12 bounds the degree of a field over which a basis of the Picard group can be defined by the bound $M(\\rho-1)$, turning the search for such a basis into a finite search."],"supporting_citations":[{"why":"Supplies the residue construction that turns polynomials into cohomology classes $\\omega_p$, the foundation of the period pairings.","marker":"[Gri69]"},{"why":"Gives the Čech-cocycle representation of the middle Hodge component used to prove $I(Y)\\subset\\tilde I_{[Y]}$.","marker":"[CG80]"},{"why":"Provides the description of $I_{[Y]}$ for complete intersections (Proposition 2.5) that yields the explicit agreement degree in Corollary 2.7.","marker":"[Dan17]"},{"why":"Computes the numerical period isomorphism $Q$ used by Algorithm 3.1 and in the quartic examples.","marker":"[Ser19]"},{"why":"Extracts the integral lattice $\\Lambda\\subset H^2(X,\\mathbb Z)$ of Hodge classes that supplies the candidate conic and line classes.","marker":"[LS19]"},{"why":"Gives the p-adic-cohomology method for bounding Picard numbers that provides the matching upper bound.","marker":"[AKR10]"},{"why":"Computes the zeta function of the reductions over $\\mathbb F_{101}$ needed to apply that upper bound.","marker":"[CHK19]"},{"why":"Provides the symbolic computer algebra used to verify that each reconstructed plane cuts $X$ into the claimed pair of conics.","marker":"[BCP97]"}],"fun_headline_variants":["Periods reconstruct the conics that generate the Picard group","Period data alone reveals hidden conics in quartic surfaces","From periods to proofs: reconstructing subvarieties exactly","Perfect Hodge classes recover subvarieties from periods","Numerical periods, exact conics: a new reconstruction method"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the numerically computed period data correctly identifies the lattice of Hodge classes: the paper trusts this identification with high confidence but supplies no rigorous error bound for the 5000-digit computations, so an error there would make the 56 or 102 candidate classes untrustworthy even though symbolic checks would likely catch gross mistakes.","fun_headline_variants_meta":{"raw":{"variants":["Periods reconstruct the conics that generate the Picard group","Period data alone reveals hidden conics in quartic surfaces","From periods to proofs: reconstructing subvarieties exactly","Perfect Hodge classes recover subvarieties from periods","Numerical periods, exact conics: a new reconstruction method"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000772,"raw_usage":{"total_tokens":3439,"prompt_tokens":989,"completion_tokens":2450,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":605,"completion_tokens_details":{"reasoning_tokens":2368}},"tokens_in":605,"tokens_out":2450,"duration_ms":18627,"temperature":1.0,"reasoning_tokens":2368,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:20:43.234248+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the period isomorphism for the surface $X=Z(x^4+x^3z-xy^3+y^4+z^4+w^4)$ with interval-arithmetic error bounds; for each of the 56 lattice classes $\\delta$ with $\\delta^2=-2$ and $h\\cdot\\delta=2$, the degree-one kernel $\\tilde I_{\\delta,1}$ must be generated by a single plane $Z(h)$ over the degree-28 field, and the symbolic intersection $X\\cap Z(h)$ must be two bitangent conics. If any one of those 56 verifications fails, or if the zeta function of the reduction modulo 101 does not bound the Picard number above by 8, the paper's central claim collapses.","supporting_citations":[],"review_version":1}