{"id":"cffb772e-f5be-4b5f-b667-f7a0083a49d9","arxiv_id":"2505.24328","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"On any algebraic variety, dimension plus one generic measurements drawn from any irreducible, non-degenerate measurement family uniquely identify the point; dimension many measurements generally do not suffice.","lead":"This paper proves that points on an algebraic variety can always be uniquely identified from one more than the dimension many generic linear measurements, even when those measurements must come from a restricted family. The result provides a general identifiability guarantee for problems like low-rank matrix recovery, polynomial reconstruction, and tensor network learning.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the proof of Theorem 1.1 is sound under its stated hypotheses.","rationale":"The reader's verdict of ACCEPT with high confidence is justified. Theorem 1.1's proof is standard and correct: the irreducibility and linear non-degeneracy of L are exactly the hypotheses needed for Lemma 2.1, and the induction in Proposition 2.2 is valid. The final separation argument using a finite union of proper subvarieties is sound because L is irreducible. The sharpness example is convincing and shows the bound is tight in general. The minor issues noted in the paper, such as the missing v1 ≠ v2 condition in Lemma 2.1 and the coordinate-count typo in the proof of Theorem 2.3, are local and do not undermine the central claim. No load-bearing concern survives scrutiny.","tokens_in":11602,"tokens_out":28332,"duration_ms":367469,"concrete_test":"Formalize the recursive open set construction in Section 2.2 with the final step included: define Ω ⊆ L^{×(n+1)} by the conditions that each successive hyperplane section drops dimension on every irreducible component and that ℓ0 separates the resulting finite set X0. Verify that the complement of Ω is a finite union of proper closed subvarieties, so Ω is Zariski open and nonempty. If this verification fails, the genericity step in Theorem 1.1 would need repair.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central theorem is correct as stated. The proof of Theorem 1.1 is a clean induction: Lemma 2.1 provides that for any distinct pair v1, v2, the set of measurements identifying them is a proper subvariety of L; Proposition 2.2 then cuts the variety by generic hyperplane sections until a finite set remains; a final generic measurement separates the candidate set. The load-bearing assumption that L is irreducible and not contained in any hyperplane is explicitly stated and is exactly what makes Lemma 2.1 true. The only caveats are minor and do not affect the argument: Lemma 2.1 should state v1 ≠ v2 (otherwise H is not a hyperplane), and the recursive genericity construction after Proposition 2.2 should explicitly include the final separating measurement ℓ0. These are exposition issues, not correctness gaps. The sharpness example with the cubic curve correctly shows that dim X measurements can fail in general, so the theorem's bound is tight in the intended sense.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the minimum number of generic linear measurements, chosen from a prescribed irreducible algebraic family L ⊆ V*, needed to uniquely identify a point x on an algebraic variety X ⊆ V. The main result (Theorem 1.1) states that if L is not contained in any hyperplane of V*, then dim X + 1 generic measurements suffice. The proof is an induction on dimension: Lemma 2.1 shows that a generic element of L separates any fixed pair of distinct points, and Proposition 2.2 shows that n generic hyperplane sections cut an n-dimensional variety down to a finite set; a final generic measurement then separates the finite candidates. A projective analogue (Theorem 2.3) and examples illustrating sharpness (a cubic curve requiring dim X + 1 measurements) are also given.","tokens_in":11798,"tokens_out":29362,"duration_ms":305333,"significance":"The result is a clean, broadly applicable identifiability statement that generalizes the Noether Normalization Lemma to structured measurement families. It applies to bilinear measurements, point evaluations of polynomials, and tensor-network feature maps. The proof is self-contained, short, and uses only standard facts from algebraic geometry. The sharpness example is explicit and convincing. If the minor issues below are fixed, the paper will be a useful reference for algebraic compressed sensing and related areas.","major_comments":[],"minor_comments":[{"comment":"The lemma should explicitly assume v1 ≠ v2; otherwise H = V* and the conclusion that C is strictly contained in L is false.","section":"Section 2.2, Lemma 2.1"},{"comment":"The step \"By Lemma 2.1, a generic element ℓ ∈ L is non-constant on every X^(i)\" is not a direct consequence of Lemma 2.1, since the latter applies to a fixed pair (v1,v2). The intended argument is that the set of linear forms constant on a positive-dimensional irreducible component is a proper linear subspace W of V*, and L ∩ W is a proper closed subset because L is not contained in any hyperplane; this should be stated explicitly.","section":"Section 2.2, Proposition 2.2"},{"comment":"There is a typo: \"ξ1\" should be \"y1\" in the definition of X1.","section":"Section 2.2, proof of Proposition 2.2"},{"comment":"The coordinate description of the map should be [v] ↦ [ℓ0(v):...:ℓ_{n+1}(v)] into P^{n+1} (or the appropriate projective space of the quotient), and the well-definedness argument should conclude ⟨ℓ0,...,ℓ_{n+1}⟩^⊥ ∩ X = ∅ rather than ⟨ℓ0,...,ℓ_n⟩^⊥ ∩ X = ∅.","section":"Section 2.3, Theorem 2.3 and its proof"},{"comment":"After dehomogenizing by ℓ_{n+1}=1, the proof should note that the restricted measurement variety remains irreducible and is not contained in a hyperplane of the affine space, so that Theorem 1.1 applies.","section":"Section 2.3, proof of Theorem 2.3"},{"comment":"The statement \"Theorem 1.1 holds over every infinite field\" should be qualified, since for a general infinite field a nonempty Zariski-open subset of L may contain no k-rational points; the standard interpretation over R and C is clear.","section":"Section 1.1, remark after Theorem 1.1"}],"recommendation":"minor_revision","confidential_remarks":"The paper is a well-written note with a correct central result. The main theorem is likely to be useful and the proof is elementary. The projective section has a few typos but the intended argument is sound. I would accept after minor revisions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis note proves the clean result that a point on an algebraic variety X can be uniquely identified by dim X + 1 generic linear measurements drawn from any irreducible variety L ⊆ V* that is not contained in a hyperplane. The proof is a short induction: one generic cut at a time, using the fact that a measurement family with those two properties separates any two distinct vectors outside a proper subvariety. I checked the proof and it is correct; the sharpness examples, especially the cubic curve, honestly show that dim X measurements do not suffice in general.\n\nWhat is genuinely new here is the move from generic measurements in the full dual space (classical Noether Normalization, as used in BGMV23) to generic measurements inside a structured family L. That matters for rank-one measurements, point evaluations, and tensor network recovery, where L is itself a nontrivial variety. The paper gives a self-contained proof and spells out the genericity condition explicitly, which is useful.\n\nThe soft spots are mostly cosmetic. Lemma 2.1 should state v1 ≠ v2; as written the hyperplane claim fails when the vectors coincide. The projective statement and its proof have notation typos (P^n vs P^{n+1}, ℓ_n vs ℓ_{n+1}) that don't affect the argument. The claim that the result holds over every infinite field is stated without proof; the argument is written in the language of algebraically closed fields, and over a general field one has to be a little careful about whether the Zariski open set of \"generic\" measurements actually contains rational points. That's a fixable expository issue rather than a substantive gap, and the intended applications over R and C are fine. The section on higher-degree curves is explicitly a sketch, which is acceptable for a short note.\n\nThis paper is for anyone working on identifiability in compressed sensing or tensor recovery. It deserves a serious referee and, after the minor fixes, publication. I would cite it.","headline":"A correct and genuinely useful generalization of Noether Normalization to structured linear measurements; the main proof holds up, with only minor exposition issues.","tokens_in":12291,"tokens_out":5446,"would_cite":true,"duration_ms":62324,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14Q15","15A29","90C30"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that dim X + 1 generic measurements from any irreducible, non-degenerate family L of linear functionals uniquely identify any point of an algebraic variety X, and this bound cannot be improved in general.","keywords":["linear measurements","algebraic varieties","identifiability","Noether normalization","low-rank matrix recovery","algebraic compressed sensing","point evaluations","projective geometry"],"falsifier":"A concrete way to test the claim is to search for an irreducible algebraic variety $L\\subseteq V^*$ not contained in a hyperplane and a variety $X$ with a point $x$ such that every tuple of $\\dim X+1$ elements of $L$ in some Zariski-open subset leaves at least two points of $X$ with identical measurement values; the existence of such a pair would falsify Theorem 1.1. A tractable first check would be small cases such as conics and cubic plane curves with $L$ the Veronese surface of point evaluations.","tokens_in":11440,"feed_emoji":"📐","tokens_out":7751,"duration_ms":72350,"temperature":0.7,"pith_summary":"This paper establishes a general identifiability threshold for recovering a point on an algebraic variety from linear measurements that are themselves constrained to a variety. The result is: if X is any algebraic variety in a finite-dimensional vector space and L is any irreducible algebraic variety of measurement functionals not contained in one hyperplane, then $\\dim X + 1$ generic measurements chosen from L determine every point $x\\in X$ uniquely. The theorem holds over every infinite field and covers settings where both the model and the measurements are nonlinear, such as low-rank matrices with rank-one measurements or polynomials from point evaluations. The bound is sharp: in typical cases, $\\dim X$ generic measurements leave several candidates, as shown by a parabola in the plane and by generic plane curves of degree at least four.","feed_headline":"One extra generic measurement is enough to identify a variety point","feed_subtitle":"Sharp identifiability for low-rank matrices, tensors, and polynomials from any irreducible measurement family","key_machinery":"The load-bearing object is the pair of statements Lemma 2.1 and Proposition 2.2. Lemma 2.1 says that because $L$ is irreducible and not contained in any hyperplane, for any two distinct vectors $v_1,v_2$ the set of $\\ell\\in L$ with $\\ell(v_1)=\\ell(v_2)$ is a proper subvariety of $L$; hence a generic measurement separates any fixed pair. Proposition 2.2 iterates this: a generic first measurement strictly lowers the dimension of every positive-dimensional irreducible component of $X$ on the affine hyperplane where it matches $x$, so after $n=\\dim X$ cuts the candidate set is finite, and the extra measurement supplied by Lemma 2.1 eliminates all candidates but $x$. This is a restricted version of a Noether-normalization slicing argument, with all slices drawn from $L$ rather than from all of $V^*$.","core_discovery":"The central claim, Theorem 1.1, is that for a finite-dimensional vector space $V$, an algebraic variety $X\\subseteq V$, a point $x\\in X$, and an irreducible algebraic variety $L\\subseteq V^*$ not contained in any hyperplane, generic elements $\\ell_0,\\ldots,\\ell_n\\in L$ with $n=\\dim X$ uniquely identify $x$ from the measurements $\\ell_i(x)$. The proof works by cutting: $n$ generic measurements restrict $X$ to a finite set (Proposition 2.2), and one further measurement separates the finitely many remaining candidates (Lemma 2.1). The projective formulation (Theorem 2.3) says that the rational projection from a generic tuple of $n+2$ elements of $L$ is injective on $X$. The paper also shows that $n+1$ is optimal in general: for a parabola a generic single measurement leaves two points, and for a generic plane curve of degree at least four no point is determined by a single measurement.","pith_inferences":["The paper leaves open quantitative versions: how large are the Zariski-open sets of good measurements, and can failure probabilities be bounded for specific pairs $(X,L)$; the recursive construction suggests such bounds could be derived for structured families.","Irreducibility of $L$ is used only to make the separating set proper, so a finite union of irreducible non-degenerate components might behave similarly if the components are jointly non-degenerate, but the paper does not claim this.","The theorem fails for finite measurement sets such as single-entry matrix completion, so a discrete analogue would need additional combinatorial or incoherence assumptions; the gap between continuous irreducible families and finite sampling sets deserves separate study.","For generic plane curves the exceptional points recoverable with $\\dim X$ measurements are finite, and the paper sketches that for higher-dimensional varieties one might expect exceptional loci of positive dimension; this is an extrapolation beyond the curve case."],"forward_implications":["For rank-one (bilinear) measurements on the variety of rank-at-most-$k$ matrices, Theorem 1.1 guarantees that $\\dim X + 1 = (d_1+d_2-k)k + 1$ generic measurements identify a rank-$k$ matrix uniquely.","For polynomial recovery from point evaluations, any polynomial lying in a variety model of dimension $n$—sparse polynomials, low Waring rank, structured circuits—is identified by $n+1$ generic evaluations.","For tensor-network and feature-map models used in learning, the theorem gives a sharp sample complexity for noiseless exact identification with generic rank-one samples.","In the projective setting, a generic projection from $L^{n+2}$ is injective on $X$, so the result can be read as a Noether-normalization-type statement for restricted projections.","The bound cannot be improved in general: the examples show that $\\dim X$ generic measurements are not enough even for simple curves, so the extra measurement is not an artifact of the proof."],"supporting_citations":[{"why":"Supplies the dimension fact that every proper subvariety of an irreducible variety has strictly smaller dimension, the engine of the cutting step.","marker":"[Sha13]"},{"why":"Supplies the projective-geometry background, degree and B\\'ezout counts, and the rational normal curve and Veronese examples used throughout.","marker":"[Har92]"},{"why":"Provides the algebraic compressed sensing context and the Noether-normalization count for rank-variety identifiability that this theorem generalizes.","marker":"[BGMV23]"},{"why":"Supplies the dual-curve singularity facts used to show generic plane curves cannot be recovered from a single generic measurement.","marker":"[GKZ94]"}],"fun_headline_variants":["Dim+1 measurements: the exact threshold for unique identification","One extra generic measurement separates all points","Sharp result: dim+1 measurements guarantee uniqueness","Exact identifiability threshold: dim+1 linear probes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The measurement family $L$ must be an irreducible algebraic variety and must not lie inside any single hyperplane of $V^*$, so that any two distinct candidate points are separated by a Zariski-open set of measurements.","fun_headline_variants_meta":{"raw":{"variants":["Dim+1 measurements: the exact threshold for unique identification","One extra generic measurement separates all points","Sharp result: dim+1 measurements guarantee uniqueness","Exact identifiability threshold: dim+1 linear probes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00077,"raw_usage":{"total_tokens":3330,"prompt_tokens":784,"completion_tokens":2546,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":400,"completion_tokens_details":{"reasoning_tokens":2484}},"tokens_in":400,"tokens_out":2546,"duration_ms":17794,"temperature":1.0,"reasoning_tokens":2484,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:27:37.558462+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete way to test the claim is to search for an irreducible algebraic variety $L\\subseteq V^*$ not contained in a hyperplane and a variety $X$ with a point $x$ such that every tuple of $\\dim X+1$ elements of $L$ in some Zariski-open subset leaves at least two points of $X$ with identical measurement values; the existence of such a pair would falsify Theorem 1.1. A tractable first check would be small cases such as conics and cubic plane curves with $L$ the Veronese surface of point evaluations.","supporting_citations":[],"review_version":1}