{"id":"6e48b8b9-dc98-4e67-b659-1508f3f1ff00","arxiv_id":"1908.01037","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"Products of quasimodes on compact manifolds can be approximated in H^{-1} and L2 norms by low-degree spectral subspaces with explicit dimension bounds, assuming the stated higher-dimensional bilinear estimate is repaired.","lead":"This mathematics paper presents approximation bounds for products of high-frequency quasimodes on compact manifolds, proposing that such products can be captured by a low-dimensional basis with explicit error estimates. It extends prior eigenfunction product approximations and contributes bilinear quasimode estimates, but a key step in the higher-dimensional proof is not justified.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The dyadic tail estimate (4.12) is false for d≥5; admissible quasimodes with high-frequency spectral tails violate it, so the proof of Theorems 4 and 1 does not go through in d≥4.","rationale":"The reader's weakest assumption, Eq. (4.12), is indeed the load-bearing step, and the stress-test confirms it for essentially the stated reason: the dyadic weights are not l²-summable, and the quasimode condition does not control the weighted l¹ sum used in (4.12). An explicit quasimode made of high-frequency spectral components on S^d makes the left side of (4.12) exceed the claimed bound, so the estimate is not merely unproved but false as a consequence of (1.1). Because Theorem 1's d≥4 range is derived through Theorem 4 and depends on (4.13), the central claim is not established by the paper. The d=2,3 arguments use different controls and appear internally consistent. A revised argument might repair the mixed-term estimate by another route, so this is a proof gap rather than a disproof of the theorem statement; but for the manuscript as written, the reader's REJECT verdict is appropriate and no adjustment is needed.","tokens_in":13275,"tokens_out":16929,"duration_ms":173150,"concrete_test":"On S^d, fix μ and choose N distinct dyadic frequencies Λ_r ∈ [2^r μ, 2^{r+1} μ] for r=1,...,N. Take zonal harmonics ψ_{Λ_r} with L^∞ norm comparable to Λ_r^{(d−1)/2} and with signs chosen so that their peaks add, and set v = a ψ_μ + Σ_{r=1}^N [μ/(√N (Λ_r²−μ²))] ψ_{Λ_r}, choosing a to make ||v||_2 = 1. Then ||(−Δ−μ²)v||_2 = μ, so v satisfies (1.1). Compute both sides of (4.12): the left is ≳ (μ/√N) Σ Λ_r^{(d−5)/2}, while the stated right side is C μ^{(d−2)/2}. For d=5 choose N > μ (e.g., N = μ²), and for d=6 choose the top frequency sufficiently large, so the left exceeds the right. This directly disproves (4.12) as written and confirms that the proof of (4.13) has no valid substitute in the text.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The main gap is the reduction of Theorem 1 to Theorem 4 for d≥4, specifically Eq. (4.12) in Section 4. The proof of Theorem 4 controls the mixed term ||L_λ u H_μ v||_2 in (4.13) solely through ||H_μ v||_∞ ≤ C μ^{(d−2)/2} obtained in (4.12). But (4.12) does not follow from the quasimode condition. Writing f = (−Δ−μ²)v with ||f||_2 ≤ Cμ, the spectral pieces satisfy 2^{2j}||S_jv||_2 ≈ ||S_jf||_2, so the dyadic sum in (4.12) is a weighted l¹ sum of pieces of f, while the quasimode condition gives only the l² norm. The weights (μ2^{−j})^{(4−d)/2} are ≥1 for d≥4 and grow like (2^j/μ)^{(d−4)/2} for d≥5, so they cannot be discarded and replaced by Cμ^{(d−4)/2}||f||_2. The failure is not a missing logarithmic factor: on S^d, take a μ-eigenfunction plus N high-frequency eigenfunctions ψ_{Λ_r} in dyadic bands with coefficients c_r = μ/(√N(Λ_r²−μ²)). Then ||(−Δ−μ²)v||_2 = μ and ||v||_2 ≈ 1, so v is an admissible quasimode, but ||H_μ v||_∞ ≳ (μ/√N) ∑ Λ_r^{(d−5)/2}. For d=5 this is ≈ μ√N, which exceeds the claimed Cμ^{3/2} once N>μ; for d≥6 it is even worse. Hence (4.13) and Theorem 4 are not established, and Theorem 1's d≥4 claim rests on an invalid estimate. The d=2,3 arguments use different controls and appear unaffected.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies products of two Laplace-Beltrami quasimodes on a compact Riemannian manifold without boundary. For quasimodes u, v with frequencies λ, μ, it claims bilinear L^2 estimates in which the larger frequency λ does not appear, with a growth factor Λ(d, min(λ,μ)) equal to μ^{1/4} in d=2, μ^{1/2} log^{1/2} μ in d=3, and μ^{(d-2)/2} in d≥4 (with additional spectral-tail terms for d≥6). These bilinear estimates are then used to prove Theorem 1, which asserts that the pointwise product u_i u_j is well approximated in H^{-1} by its projection onto the first ν eigenspaces provided ν ≳ Ω(d, min(λ_m,λ_n)) ε^{-d}, with Ω as in (2.6). The paper also gives an L^2 approximation result (Theorem 5) and claims an improvement of Sogge-Zelditch L^4 quasimode bounds for d≥8. The proof splits quasimodes into low- and high-frequency parts, uses known spectral-cluster bilinear estimates, and applies Littlewood-Paley decompositions to control high-frequency contributions.","tokens_in":13767,"tokens_out":9557,"duration_ms":90068,"significance":"If the main theorems were correct, the paper would extend the Lu-Steinerberger approximation results for eigenfunction products to the substantially larger class of quasimodes, in all dimensions, and it would improve known L^4 quasimode bounds in high dimensions. The d=2 and d=3 arguments use standard ingredients (Sogge-Zelditch L^p quasimode bounds, Burq-Gerard-Tzvetkov bilinear spectral cluster estimates) and appear internally consistent. The higher-dimensional part, however, rests on a dyadic estimate that is false for admissible quasimodes. This is a load-bearing error: it invalidates the proofs of Theorem 4 for d≥4 and hence Theorem 1 for d≥4, as well as the claimed L^4 improvement for d≥8. The paper does not contain machine-checked proofs or reproducible code; its value depends entirely on the analytic arguments, and the central all-dimensions claim is not established.","major_comments":[{"comment":"The dyadic bound on ||H_μ v||_∞ is not valid for d≥4. The proof asserts, after passing to Littlewood-Paley pieces, that ∑_{2^j ≥ 2μ} (μ 2^{-j})^{(4-d)/2} || -Δ S_j v||_2 ≤ C ||(-Δ - μ^2)v||_2. For d=4 the weights are identically 1, so the left side is an infinite sum of dyadic pieces of -Δv; for d≥5 the weights grow like (2^j/μ)^{(d-4)/2}. The quasimode condition ||(-Δ - μ^2)v||_2 ≤ C μ controls the ℓ^2 sum of the spectral pieces of (-Δ-μ^2)v, which does not imply the corresponding weighted ℓ^1 sum. Concretely, on S^d, take v = ψ_μ + (μ/√N) Σ_{r=1}^N (Λ_r^2 - μ^2)^{-1} ψ_{Λ_r}, where ψ_Λ are normalized eigenfunctions and the Λ_r lie in separated dyadic bands above μ. Then ||(-Δ - μ^2)v||_2 = μ and ||v||_2 ≍ 1, so v is an admissible quasimode, but the high-frequency contribution to ||H_μ v||_∞ is ≳ (μ/√N) Σ Λ_r^{(d-5)/2}. For d=5 this is ≳ μ√N, which exceeds the claimed C μ^{3/2} once N ≫ μ; for d≥6 the divergence is even faster. Thus Eq. (4.12) fails for admissible quasimodes in all dimensions d≥4.","section":"Section 4, Eq. (4.12)"},{"comment":"The failure of Eq. (4.12) is load-bearing. The mixed term ||L_λ u H_μ v||_2 in (4.13) is controlled solely through the bound ||H_μ v||_∞ ≲ Λ(d,μ)(μ^{-1}||(-Δ-μ^2)v||_2 + ||v||_2) obtained in (4.12). Since that bound is false, the proof of Theorem 4 for d=4,5 and d≥6 does not go through, and consequently the proof of Theorem 1 for d≥4 in Section 5 collapses. The claimed improvement of the Sogge-Zelditch L^4 quasimode estimate for d≥8, which is derived from the λ=μ case of the d≥6 bilinear estimate, is also unsupported. The d=2 and d=3 arguments do not rely on this particular step and appear sound.","section":"Section 4, Eq. (4.13), and Section 5"}],"minor_comments":[{"comment":"The condition 'ν = O(Ω(d,min(λ_m,λ_n)) ε^{-d})' should be phrased as a lower bound, e.g. 'ν ≥ C Ω(d,min(λ_m,λ_n)) ε^{-d}', since the proof shows that the H^{-1} error is small when ν is sufficiently large, not merely when ν is of a given order.","section":"Section 2, Eq. (2.5)"},{"comment":"The displayed formula for σ(p) is garbled: 'd(d − 1/2(1/2 − 1/p))' should presumably be 'max{d(1/2 − 1/p) − 1/2, (d−1)/2 (1/2 − 1/p)}', matching the standard Sogge exponent used elsewhere in the paper.","section":"Section 6, Eq. (6.3)"},{"comment":"The sentence ending 'we have are desired' is an incomplete editorial remnant and should be removed or rewritten.","section":"Section 6, after Theorem 5"},{"comment":"The proof of Theorem 5 would benefit from explicitly defining σ_d; currently the statement says 'there is a σ = σ_d' but the value σ = (2/d)σ(4) appears only in the proof.","section":"Section 7.1, Eq. (7.2)"}],"recommendation":"reject","confidential_remarks":"The d=2 and d=3 parts seem coherent and could potentially form a short separate paper. As written, however, the advertised all-dimensions results are invalid because of the false dyadic estimate in (4.12), and the concrete counterexample shows this is not a mere gap in the exposition. The central claims for d≥4 therefore cannot be fixed by local revision within the current scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about arXiv:1908.01037. First, the d=2,3 bilinear quasimode estimates and the H^{-1} and L2 approximation results built on them are a credible extension of Lu–Steinerberger to quasimodes. Second, the all-dimensions claim, Theorem 1 for d≥4, rests on a false dyadic estimate, so the central advertised result is not established.\n\nWhat's new: the paper aims to approximate pointwise products of quasimodes without the microlocal support assumption used by Guo–Han–Tacy, and to remove the higher frequency in bilinear estimates. The d=2,3 arguments are assembled from known Lp quasimode bounds and Burq–Gerard–Tzvetkov cluster estimates; they appear internally consistent and give the stated mu^{1/4} and mu^{1/2} log^{1/2} mu factors.\n\nThe problem is in Section 4. Equation (4.12) claims that ||H_mu v||_∞ ≤ C mu^{(d-4)/2} ||(-Δ-mu^2)v||_2. The dyadic sum used to get there has weights (2^j/mu)^{(d-4)/2} for d≥5, which grow with j. The quasimode condition only gives l2 control on the spectral pieces of the residual, not the l1 control needed to justify the sum. The stress-test counterexample is an honest one: add high-frequency eigenfunctions with coefficients ~mu/(sqrt(N) Lambda_r^2) to a mu-eigenfunction. This satisfies (1.1) with ||v||_2 ~ 1, but the dyadic tail sum diverges, so the claimed bound fails. For d=4 the same issue appears (weights are 1, still no decay to kill the infinite l1 sum). Consequently Theorem 4 is not proved for d≥4, and Theorem 1's d≥4 part collapses with it. The claimed L4 improvement for d≥8 is also unsupported.\n\nThe paper is not sloppy in general: the low-dimensional parts are careful, the literature is engaged honestly, and the error is subtle rather than a sign of carelessness. But it is load-bearing. This should be a reject in current form; a revised version might still rescue the low-dimensional results and possibly find a different argument for some higher-dimensional cases.\n\nWho is this for? People working on bilinear eigenfunction estimates and density fitting might want the d=2,3 parts. But as it stands, the main theorem overclaims. I would send it to a serious referee—the issue deserves expert eyes, and the d=2,3 portion may be publishable—but the editor should flag the (4.12) step.\n\nRecommendation: reject with major revision; do not send to peer review expecting acceptance as-is.","headline":"The d=2,3 quasimode product results look solid, but the all-dimensions theorem rests on a false dyadic estimate in (4.12), so the paper overclaims.","tokens_in":14268,"tokens_out":4547,"would_cite":false,"duration_ms":43473,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35A15","35C99","47B06","81Q05"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that products of quasimodes can be approximated in H^{-1} by O(Ω(d,min(λ,μ)) ε^{-d}) low-frequency eigenfunctions.","keywords":["quasimodes","Laplace-Beltrami operator","bilinear estimates","approximation in H^{-1} norm","spectral clusters","Littlewood-Paley decomposition","eigenfunction products","Weyl law"],"falsifier":"Inspect the dyadic sum ∑_{2^j ≥ 2μ} $2^{{(d-4)j/2}}$ ||Δ S_j v||_2 for a family of quasimodes v on a compact d-manifold with d ≥ 5 and check whether it is bounded by a constant times $μ^{{(d-4)/2}}$ ||(−Δ−$μ^{2}$)v||_2; if a quasimode sequence makes ||H_μ v||_∞ grow faster than $μ^{{(d-4)/2}}$ times the residual, then Theorem 4 and the corresponding part of Theorem 1 are false.","tokens_in":13026,"feed_emoji":"📐","tokens_out":7803,"duration_ms":72767,"temperature":0.7,"pith_summary":"This paper claims that on any compact Riemannian manifold, the pointwise product of two quasimodes—functions that nearly solve the Laplace eigenvalue equation at frequencies λ and μ—can be nearly recovered from the span of the lowest few true eigenfunctions, with error below ε in the $H^{{-1}}$ norm. The required number of eigenfunctions grows like Ω(d,min(λ,μ)) $ε^{{-d}}$, where Ω is $μ^{{1/2}}$ in two dimensions, $μ^{{3/2}}$ $log^{{3/2}}$(μ) in three dimensions, and $μ^{{d(d-2)/2}}$ in dimensions four and higher. If correct, this gives approximation bounds for products of quasimodes in all dimensions, extending a known eigenfunction approximation result to the broader class of quasimodes. The paper also establishes bilinear quasimode estimates that remove the highest frequency from the right-hand side, which is the feature needed for nonlinear applications.","feed_headline":"Few true modes capture products of quasimodes","feed_subtitle":"Error falls below ε once the mode count passes Ω(d,min(λ,μ)) ε^{-d}, extending eigenfunction approximation to quasimodes.","key_machinery":"The central machinery is a Littlewood–Paley decomposition that splits each quasimode into a low-frequency part L_λ u = ψ(P/λ)u and a high-frequency part H_λ u = ρ(P/λ)u, where P = √(−Δ_g). The low-frequency part is controlled by bilinear spectral-cluster estimates, while the high-frequency part is handled by L^p quasimode bounds and Sobolev embedding; the two pieces are then recombined with Hölder and Cauchy–Schwarz steps. The key feature of these estimates is that the larger frequency λ drops out of the final right-hand side, leaving only the smaller frequency μ through the factor Λ(d,μ). This frequency cancellation is what makes the approximation space depend on min(λ,μ) rather than on the product of the two frequencies.","core_discovery":"The central claim is Theorem 1: for quasimodes u_i and u_j with frequencies λ_i and λ_j, projecting the product u_i u_j onto the first ν eigenfunctions leaves an $H^{{-1}}$ remainder of size less than ε whenever ν = O(Ω(d,min(λ_m,λ_n)) $ε^{{-d}}$). The proof rests on bilinear quasimode estimates in which the larger frequency disappears from the bound: for 2 ≤ d ≤ 5, ||u v||_2 is controlled by Λ(d,min(λ,μ)) times the quasimode remainders of u and v, and for d ≥ 6 a similar bound holds at the cost of an extra small high-frequency tail term. The case λ = μ of these estimates yields an $L^{4}$ quasimode bound that improves the Sogge–Zelditch bound in dimensions d ≥ 8. From the bilinear estimates, the $H^{{-1}}$ approximation follows by the Weyl law, which converts a bound on spectral coefficients into a bound on the required number of modes.","pith_inferences":["The author does not address whether the exponent d(d-2)/2 in dimensions d ≥ 4 is sharp; a natural test is whether products of two high-frequency quasimodes on the round sphere saturate the bound.","The H^{-1} norm is the natural space for Coulomb-type potentials, so the theorem may give a rigorous justification for density fitting of approximate wavefunctions on curved manifolds.","The method separates cleanly: the low-frequency part relies only on sharp spectral-cluster estimates, while the high-frequency part is where the extra quasimode assumptions enter; replacing that step with a different bound could extend the theorem to Schrödinger operators with potentials."],"forward_implications":["In dimensions 2 and 3, the approximating space size is roughly μ^{1/2} ε^{-2} and μ^{3/2} log^{3/2}(μ) ε^{-3}, respectively; in dimensions d ≥ 4 it grows like μ^{d(d-2)/2} ε^{-d}.","For λ = μ and d ≥ 8, the bilinear estimate gives an L^4 quasimode bound with a smaller remainder than the Sogge–Zelditch L^4 bound.","The same machinery yields an L^2 approximation bound: ||R_ν(u_i u_j)||_{L^2} ≤ C n^{σ} (n/ν)^{1/d}, so products of quasimodes are also well represented by low-frequency eigenfunctions in the ordinary L^2 sense.","Because the bilinear bound depends on the smaller frequency alone, products pairing a high-frequency quasimode with a low-frequency quasimode are controlled by the low frequency, a property directly suited to nonlinear arguments where the highest frequency must not appear."],"supporting_citations":[{"why":"Supplies quasimode L^p estimates, including the V ≡ 0 case used to control the high-frequency part in dimensions d ≥ 6.","marker":"[1]"},{"why":"Bilinear eigenfunction estimates on surfaces used for the low-frequency product bound in dimension 2.","marker":"[2]"},{"why":"Multilinear eigenfunction estimates on three-dimensional manifolds used for the low-frequency product bound in dimension 3.","marker":"[3]"},{"why":"Bilinear and multilinear spectral cluster estimates in higher dimensions used to bound products of low-frequency pieces for d ≥ 4.","marker":"[4, 5]"},{"why":"Semiclassical L^p estimates for spectrally localized functions used in the L^2 approximation result.","marker":"[9]"},{"why":"The eigenfunction product approximation whose argument is adapted to prove the H^{-1} bound.","marker":"[10]"},{"why":"The eigenfunction product approximation result that this paper extends to quasimodes.","marker":"[11]"},{"why":"L^p quasimode bounds of Sogge–Zelditch; the λ = μ case is compared and improved for d ≥ 8.","marker":"[15]"}],"fun_headline_variants":["Products of quasimodes need few modes","Sparse spaces capture quasimode products","Quasimode products approximate with small space","Tight bilinear bounds for quasimode products"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the dyadic tail estimate in (4.12), which assumes a uniform L^∞ bound on the high-frequency part of a quasimode; in dimensions at least five this bound does not follow from the defining quasimode equation, and the d ≥ 5 part of Theorem 1 collapses if it fails.","fun_headline_variants_meta":{"raw":{"variants":["Products of quasimodes need few modes","Sparse spaces capture quasimode products","Quasimode products approximate with small space","Tight bilinear bounds for quasimode products"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000212,"raw_usage":{"total_tokens":1442,"prompt_tokens":993,"completion_tokens":449,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":609,"completion_tokens_details":{"reasoning_tokens":392}},"tokens_in":609,"tokens_out":449,"duration_ms":4923,"temperature":1.0,"reasoning_tokens":392,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:28:13.743809+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Inspect the dyadic sum ∑_{2^j ≥ 2μ} $2^{{(d-4)j/2}}$ ||Δ S_j v||_2 for a family of quasimodes v on a compact d-manifold with d ≥ 5 and check whether it is bounded by a constant times $μ^{{(d-4)/2}}$ ||(−Δ−$μ^{2}$)v||_2; if a quasimode sequence makes ||H_μ v||_∞ grow faster than $μ^{{(d-4)/2}}$ times the residual, then Theorem 4 and the corresponding part of Theorem 1 are false.","supporting_citations":[{"cited_title":"Quasimode, eigenfunction and spectral projection bounds for Schr\\\"odinger operators on manifolds with critically singular potentials","cited_arxiv_id":"1904.09665","evidence_quote":"Supplies quasimode L^p estimates, including the V ≡ 0 case used to control the high-frequency part in dimensions d ≥ 6."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Bilinear eigenfunction estimates on surfaces used for the low-frequency product bound in dimension 2."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Multilinear eigenfunction estimates on three-dimensional manifolds used for the low-frequency product bound in dimension 3."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Semiclassical L^p estimates for spectrally localized functions used in the L^2 approximation result."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The eigenfunction product approximation whose argument is adapted to prove the H^{-1} bound."},{"cited_title":"On Pointwise Products of Elliptic Eigenfunctions","cited_arxiv_id":"1810.01024","evidence_quote":"The eigenfunction product approximation result that this paper extends to quasimodes."},{"cited_title":"Sogge, Steve Zelditch, A note on Lp-norms of quasi-modes, Adv","cited_arxiv_id":null,"evidence_quote":"L^p quasimode bounds of Sogge–Zelditch; the λ = μ case is compared and improved for d ≥ 8."}],"review_version":1}