{"id":"1000f200-e975-436f-a2f2-bbd040722a78","arxiv_id":"2506.11707","paper_version":2,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For free-fermion determinantal processes associated with Berezin-Toeplitz operators, smooth linear statistics satisfy a two-term Szegő-type expansion and a Gaussian CLT with variance equal to half the H1 seminorm in the bulk plus a flow-dependent H1/2 seminorm on the boundary.","lead":"This paper proves a central limit theorem for the particle positions in a large family of two-dimensional fermionic point processes, generalizing the Ginibre random matrix ensemble. The limiting variance has a bulk term and an edge term that depends on the classical Hamiltonian flow of the confining potential, so it is not conformally invariant.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The edge variance Σ1_D rests on an imported Bohr–Sommerfeld/spectral-simplicity input (Remark 2.10, Prop. 3.6) that is not derived from Assumption 1(c); if the O(1)-window eigenvalues are not simple with N^{-1} spacing, the Toeplitz replacement and hence Σ1_D fail.","rationale":"The reader identified the spectral-simplicity/Bohr–Sommerfeld structure in an O(1) energy window as the weakest assumption, and I agree that this is the most load-bearing point. The paper's genuinely new boundary variance is obtained by replacing the spectral projection near the edge with a Toeplitz matrix whose symbol is f evaluated along the Hamiltonian flow. That replacement is exactly Proposition 3.6, and its hypotheses are inherited from Remark 2.10 and [36, Prop. 2.11]. If those hypotheses fail, the edge computation collapses and the non-conformal variance formula Σ1_D has no basis. The concern is not an ad hominem or a dispute with consensus: it is a request to verify an imported technical input that the paper asserts but does not prove. The text contains several additional local inconsistencies—factor-of-1/2 discrepancies in Proposition 3.2 and Section 5, a bounded-vs-linear-growth mismatch, and conflicting definitions of the cutoff functions in Corollary 2.18—but these appear fixable and do not by themselves threaten the central claim. The spectral-simplicity input is different: it is the gate through which the Hamiltonian dynamics enters the variance, and it is not merely a bookkeeping error. Because the concern is substantial but the missing ingredient is plausibly true and checkable, the verdict should remain CONDITIONAL rather than being upgraded or downgraded.","tokens_in":94,"tokens_out":29398,"duration_ms":1039095,"concrete_test":"For an explicit non-radial family V_ε(z)=|z|²+ε Re(z²), 0<ε<1/2, µ=1, the level set {V_ε=1} is a connected ellipse and Assumption 1(c) holds. Derive the Bohr–Sommerfeld quantization to order N^{-1} from the action I_λ=γ({V_ε<λ}) and independently compute the O(1)-window spectrum of P_N V_ε P_N, e.g. numerically at N=200, 400, 800 in a holomorphic basis truncated to about 2N modes. Check two things: (i) all eigenvalues in [1−δ, 1+δ] are simple with spacings c/N+o(1); (ii) for a fixed smooth g supported in {|V_ε−1|<δ/2}, the matrix elements in (3.9) satisfy (3.10) with ∠a_n(I_λ) computed from the exact Hamiltonian flow (1.6). If either check fails, the Toeplitz replacement and the boundary variance formula are not justified; if both pass, the load-bearing input is confirmed for a nontrivial non-radial case.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central new content of the paper is the boundary variance Σ1_D, and the proof of that term goes through Proposition 3.6, which asserts that matrix elements of A = X_N(e^f−1)X_N in the eigenbasis of H_N have the approximate Toeplitz form (3.10). The proof of Proposition 3.6 is not self-contained: it invokes Remark 2.10, which in turn appeals to Charles [27] for simplicity and Bohr–Sommerfeld spacing of eigenvalues, and then to Proposition 2.11 of [36] for the matrix-element expansion. For (3.10) to hold uniformly in an O(1) energy window around µ, one needs the eigenvalues λ_k in that window to be simple, spaced by ~N^{-1}, with the index condition I_{λ_k}=k+O(1) and with eigenfunction phases fixed along the Hamiltonian flow. These are nontrivial spectral facts on the non-compact space C, and the text does not verify that the hypotheses of [27] are satisfied, nor does it rule out accidental degeneracies within the O(1) window for non-radial potentials. Assumption 1(c) only gives one connected regular level curve at energy µ; it does not by itself control the flow uniformly for all energies in [µ−δ, µ+δ] or the uniformity in N of the quantization. The paper explicitly excludes multi-cut resonances and says generically no resonances are expected, but no proof of absence of resonances in the single-cut case is supplied. If the spectral simplification failed, the replacement principle of Proposition 3.7 would break, the Szegő-limit step in Section 3.3 would not apply, and the boundary variance would not be Σ1_D. This is a correctness-risk point rather than a contradiction, because the missing input may well be true; however, it is load-bearing precisely because the paper's main novelty depends on it.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies determinantal point processes obtained from the ground-state projection Π_N = 1{P_N V P_N ≤ μ} of Berezin–Toeplitz operators on the Fock–Bargmann space, generalizing the N-Ginibre ensemble. Under Assumptions 1 on the potential V, the main result Theorem 1.2 gives a two-term Szegő-type asymptotic expansion for the log-Laplace transform of smooth linear statistics X(f), and hence a central limit theorem with variance Σ(f) = Σ^1_D(f) + Σ^2_D(f). Here Σ^2_D is the bulk H^1-seminorm and Σ^1_D is an H^{1/2}-seminorm computed along the Hamiltonian flow of V on the level set {V = μ}. The proof combines exponential decay of the kernel, decorrelation estimates separating bulk and boundary, a replacement principle reducing the boundary contribution to a Toeplitz determinant, and strong Szegő asymptotics.","tokens_in":41012,"tokens_out":13086,"duration_ms":158728,"significance":"If the proof is completed, the result is a substantial extension of the Rider–Virág CLT for the Ginibre ensemble to a natural family of fermionic point processes. The central new content is the boundary variance Σ^1_D, which depends on the Hamiltonian flow and is generally not conformally invariant; this is an interesting and falsifiable prediction. The paper is also methodologically valuable: the variance is constructed explicitly from the flow and the gradient with no fitted parameters, and the replacement principle and bulk mesoscopic CLT are stated as tools of independent interest. The cited prior work [27, 36, 67] provides building blocks rather than the main theorem, so I do not see circularity in the central claim.","major_comments":[{"comment":"The approximate Toeplitz expansion (3.10), and hence the edge variance Σ^1_D, depends on spectral simplicity and Bohr–Sommerfeld spacing of H_N = P_N V P_N in the O(1) window [μ−δ/2, μ+δ/2]. Remark 2.10 asserts these properties by appealing to [27] and [67, Thm. 13.10], but Assumption 1(c) only gives one connected regular level {V = μ}; it does not by itself control all levels in the window, rule out accidental degeneracies for non-radial V, or guarantee the index condition I_{λ_k} = k + O(1) uniformly. The paper explicitly says resonances in multi-cut cases are expected generically absent, but no proof of absence is supplied even in the single-cut case. Since Proposition 3.7 and the strong Szegő step use these properties uniformly, the boundary variance is currently conditional on an imported spectral input that is not derived in the manuscript. Please provide a precise verification of the hypotheses of [27] (or a generic non-resonance theorem), or state Theorem 1.2 with this spectral assumption made explicit.","section":"§3.2, Proposition 3.6 and Remark 2.10"},{"comment":"The factors of 1/2 are inconsistent. Proposition 3.2 concludes tr[Π_N(ϑ(A)−ϑ(g))Π_N] ≃ (1/4)∫_D |∇f|² dγ = (1/4)Σ^2_D(f), while its own proof in Proposition 3.11 gives (1/4)∫_D |∇f|² dγ = (1/2)Σ^2_D(f), which is the value needed for Theorem 1.2. Moreover, the final assembly in Section 5 writes Υ(f_2,Π_N) → Σ^1_D(f_2)+Σ^2_D(f_2), omitting the factors 1/2 that appear in Propositions 3.1 and 3.11; taken literally this would give total variance 2Σ(f), contradicting Theorem 1.2. These displays must be corrected and checked consistently.","section":"Proposition 3.2 and §5"}],"minor_comments":[{"comment":"The proof invokes Theorem 4.4 'with ϵ_N = 1', but Theorem 4.4 requires dist(x_N, {V > μ}) ≥ 2η_N = 2, which is not satisfied by an arbitrary fixed cutoff f_j supported in {V ≤ μ − δ/2}. The intended statement is the fixed-scale bulk CLT from Corollary 4.2; the proof should be rephrased accordingly.","section":"§5"},{"comment":"Corollary 2.18 defines the boundary cutoff χ_2 supported in {|V−μ| ≤ 3δ}, whereas Section 3 assumes the test function is supported in {|V−μ| < δ/2}. This can be made consistent by choosing the Corollary's δ small relative to the Section 3 parameter, but the paper should say so explicitly.","section":"Corollary 2.18 and §3"},{"comment":"In the display 'N^{−1}Π_N(x,x) → 1{V(x) < x}', the argument of the indicator should be V(x) < μ.","section":"Proposition 2.13"},{"comment":"The phrase 'the eigenvalues of Π_N V_1 Π_N' should presumably read 'P_N V_1 P_N', since Π_N is the spectral projection whose eigenvalues are being described.","section":"Remark 2.10"},{"comment":"There are small presentation errors: the reference 'Proposition 4.4' should be 'Theorem 4.4', the sentence after the cutoffs has a capitalization error ('in this case f_j ∈ C_0^∞'), and the proof begins by assuming f is bounded although Theorem 1.2 assumes at most linear growth; the reduction via Lemma 2.15 should be stated.","section":"§5 and §2.4"}],"recommendation":"major_revision","confidential_remarks":"The decisive issue is the unverified Bohr–Sommerfeld and spectral-simplicity input behind Proposition 3.6. If the authors can supply a rigorous verification of the hypotheses of [27] for the whole energy window, or make the spectral assumption explicit and show it is generic, I would be inclined to accept after the factor-of-two corrections. The paper is within the scope of the journal and the main idea is coherent."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real result. The authors generalize the Ginibre fluctuation theorem to spectral projections of Berezin–Toeplitz operators and find that the edge variance is governed by the Hamiltonian flow, not by conformal geometry. That non-conformal boundary term is the genuine novelty. The proof is detailed and mostly convincing. I would send it to a serious referee.\n\nWhat is actually new: the class of processes, the two-term Szegő-type expansion for their Laplace transforms, and the CLT with variance Σ1_D + Σ2_D. Prior work (Rider–Virág, Ameur–Hedenmalm–Makarov) had conformally invariant variances; here the H^{1/2} boundary term depends on the flow (1.6). The replacement principle in Section 3.1 and the bulk CLT via the trace identity are nice, and the exponential decay/decorrelation machinery is put together cleanly.\n\nWhere I would push back: the proof of Theorem 1.2 is written for bounded test functions, while the theorem allows linear growth. There is no explicit truncation argument. That should be added. There are also several factor-of-2 typos: Proposition 3.2 states 1/4∫|∇f|^2 = 1/4Σ2_D(f) instead of 1/2Σ2_D(f), and Section 5 drops the 1/2 in several displayed formulas. These are presentation bugs, not math errors, but they must be fixed before publication.\n\nThe bigger soft spot is the spectral input. Proposition 3.6 needs simple eigenvalues with N^{-1} spacing and fixed phases in an O(1) energy window around µ, and that is imported from Charles [27] via Remark 2.10. Assumption 1(c) alone does not give that uniformity; the paper should state the needed result as a lemma and verify the hypotheses. I don't think this is circular—[27] is independent—and the missing facts are standard Bohr–Sommerfeld, so I expect they can be supplied. But as written it is a load-bearing citation.\n\nOverall: the central claim holds up. The paper is written for specialists in DPPs or semiclassical analysis; general probabilists can read the introduction and Theorem 1.2. I would accept for peer review and ask for the spectral lemma and the bounded-to-linear-growth step to be made explicit.","headline":"Genuinely new CLT for Berezin-Toeplitz point processes with a non-conformal edge term; proof mostly solid, with fixable typos and one imported spectral input that needs spelling out.","tokens_in":41744,"tokens_out":4635,"would_cite":true,"duration_ms":53750,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G55","60B20","47B35","35P20"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper shows that for ground-state free-fermion determinantal point processes given by spectral projections of Berezin–Toeplitz operators $P_N V P_N$, smooth linear statistics satisfy a law of large numbers and a Gaussian central…","keywords":["determinantal point processes","Berezin–Toeplitz operators","free fermions","Ginibre ensemble","Szegő limit theorem","linear statistics","central limit theorem","droplet edge fluctuations"],"falsifier":"Take two potentials with the same droplet boundary but different Hamiltonian flows on $\\partial D$, for instance a radial potential $V(|z|)$ and an anisotropic quadratic potential $V(x,y)=x^2+\\alpha y^2$ whose level set is the same ellipse, and compute the limiting variance of a boundary-supported linear statistic $f$. If the variance is identical for both potentials, the flow-dependent formula for $\\Sigma^1_D$ is wrong.","tokens_in":40445,"feed_emoji":"🎲","tokens_out":6750,"duration_ms":76822,"temperature":0.7,"pith_summary":"The paper studies the random point configuration of $N$ free fermions in the lowest Landau level with a confining potential $V$, i.e., the determinantal process with kernel the spectral projection of the Berezin–Toeplitz operator $P_N V P_N$ below energy $\\mu$. Its main result is a two-term Szegő-type expansion: the Laplace transform of any smooth linear statistic with at most linear growth is Gaussian with variance $\\Sigma(f) = \\Sigma^1_D(f)+\\Sigma^2_D(f)$, where $\\Sigma^2_D$ is the $H^1$ seminorm of $f$ over the droplet and $\\Sigma^1_D$ is a weighted $H^{1/2}$ seminorm along the level curve $\\{V=\\mu\\}$. The boundary term is computed from the Fourier coefficients of $f$ along the Hamiltonian flow of $V$, so it depends on the dynamics and is generally not conformally invariant. This generalizes the Ginibre circular-law fluctuations and explains why edge fluctuations of such fermionic droplets are not universal in the Gaussian-free-field sense.","feed_headline":"Free-fermion droplet fluctuations get a two-term variance","feed_subtitle":"Boundary fluctuations follow the Hamiltonian flow of the potential, breaking conformal invariance.","key_machinery":"The argument splits the linear statistic into bulk and edge parts using exponential decay of the correlation kernel away from the droplet and decorrelation estimates. In the bulk, the kernel is exponentially close to the Bergman projection $P_N$, so the CLT for the $\\infty$-Ginibre process (Proposition 4.3) yields the $\\Sigma^2_D$ term, even at arbitrary mesoscopic scales. At the edge, the local spectral operator $A = X_N(e^f-1)X_N$ is approximated by an infinite Toeplitz matrix $B$ with entries $\\hat b_{j-k}(I_\\mu)$ in the eigenbasis of $P_N V P_N$, using the Bohr–Sommerfeld/WKB structure of the eigenfunctions (Proposition 3.6) and a general replacement principle (Proposition 3.3); the strong Szegő limit theorem then produces $\\Sigma^1_D$. A semiclassical two-term calculus for Berezin–Toeplitz operators contributes a complementary $\\tfrac12\\Sigma^2_D$ term from the edge region.","core_discovery":"In precise terms, Theorem 1.2 asserts that for $f$ smooth near the droplet $D$ with $|f|\\leq C(1+|\\cdot|)$, one has $\\mathbb{E}_{\\Pi_N}[e^{X(f)}] = \\exp(\\mathbb{E}X(f) + \\tfrac12 \\Sigma(f) + o(1))$ with $\\Sigma(f) = \\Sigma^1_D(f)+\\Sigma^2_D(f)$; consequently $(X(f)-\\mathbb{E}X(f))/\\sqrt{\\Sigma(f)}$ converges to a standard Gaussian. The bulk term $\\Sigma^2_D(f)=\\tfrac12\\int_D |\\nabla f|^2\\, d\\gamma$ comes from the $\\infty$-Ginibre process, while the boundary term $\\Sigma^1_D(f)=\\sum_{k\\geq 1} k|\\hat f_k|^2$ uses the Fourier coefficients $\\hat f_k$ of $f$ along the curve $\\{V=\\mu\\}$ parametrized by the Hamiltonian flow (1.6). The key qualitative content is that the edge fluctuations are encoded in the dynamics of $V$, not merely in the geometry of the droplet.","pith_inferences":["If the formula extends to multi-component level sets, the boundary variance should be written as a sum over components, but resonances between components may produce extra covariance terms not present in Theorem 1.2.","The replacement-principle technique suggests a route to edge CLTs for higher Landau levels by substituting the corresponding microlocal dynamics, provided the spectral projection still has Toeplitz asymptotics.","A direct numerical test is to exact-diagonalize $P_N V P_N$ for an anisotropic quadratic potential $V(x,y)=x^2+\\alpha y^2$ and a test function supported near the ellipse boundary; the predicted variance depends on the flow, so agreement with the formula would confirm the flow-dependence, whereas agreement with a purely geometric variance would refute it.","The non-conformal dependence hints that the edge noise couples to the quantum metric of the lowest Landau level, which could be probed in anisotropic quantum-Hall-droplet experiments."],"forward_implications":["Particles condense uniformly in the droplet, with exponentially small probability of lying outside any mesoscopic neighbourhood of $D$.","Bulk local statistics are universal: the microscopic point process converges to the $\\infty$-Ginibre process in the bulk.","The centred statistic $X(f)-\\mathbb{E}X(f)$ is asymptotically Gaussian with variance $\\Sigma^1_D(f)+\\Sigma^2_D(f)$; for radial $V$ this reproduces the Ginibre circular-law fluctuation formula.","Edge fluctuations are not described by the Gaussian free field: the variance functional is not conformally invariant unless the potential is harmonic.","The bulk CLT holds at arbitrary mesoscopic scales for test functions supported away from the boundary."],"supporting_citations":[{"why":"Supplies the Bohr–Sommerfeld structure of eigenfunctions near $\\{V=\\mu\\}$, which underlies the approximate Toeplitz form of edge matrix elements in Proposition 3.6.","marker":"[27]"},{"why":"Provides the one-dimensional Szegő-type edge asymptotics and the Proposition 2.11 used to derive the approximate Toeplitz structure for the two-dimensional Berezin–Toeplitz edge.","marker":"[36]"},{"why":"Establishes the Ginibre ensemble fluctuation result with bulk and boundary variance terms that the present two-term formula generalizes.","marker":"[60]"},{"why":"Gives the $\\infty$-Ginibre bulk CLT with $H^1$ variance, directly used for the $\\Sigma^2_D$ contribution in the bulk.","marker":"[59]"},{"why":"Provides the normal-matrix two-term variance with boundary term, used for comparison and for contrasting conformal invariance.","marker":"[1]"},{"why":"Supplies the pseudodifferential conjugation (Theorem 13.10) invoked in Proposition 3.6 to relate Beezin–Toeplitz operators to pseudodifferential operators.","marker":"[67]"},{"why":"Provides the exponential concentration estimates for eigenfunctions in the forbidden region, used for the kernel decay and decorrelation estimates.","marker":"[35]"}],"fun_headline_variants":["Two-term variance for free-fermion droplet fluctuations","Free-fermion edge fluctuations break conformal invariance","Generalized Ginibre: bulk and boundary variance split","Boundary noise in free-fermion droplets follows Hamiltonian flow"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The spectral data of $P_N V P_N$ in an energy window of size $O(1)$ around $\\mu$ must follow the Bohr–Sommerfeld ladder governed by the Hamiltonian flow of $V$, which is derived from the assumption that the level set $\\{V=\\mu\\}$ is a single smooth curve with $\\nabla^\\perp V \\neq 0$; if the level set has several components that resonate, the edge variance formula could change.","fun_headline_variants_meta":{"raw":{"variants":["Two-term variance for free-fermion droplet fluctuations","Free-fermion edge fluctuations break conformal invariance","Generalized Ginibre: bulk and boundary variance split","Boundary noise in free-fermion droplets follows Hamiltonian flow"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000204,"raw_usage":{"total_tokens":1353,"prompt_tokens":875,"completion_tokens":478,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":491,"completion_tokens_details":{"reasoning_tokens":413}},"tokens_in":491,"tokens_out":478,"duration_ms":5402,"temperature":1.0,"reasoning_tokens":413,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T04:04:16.773188+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take two potentials with the same droplet boundary but different Hamiltonian flows on $\\partial D$, for instance a radial potential $V(|z|)$ and an anisotropic quadratic potential $V(x,y)=x^2+\\alpha y^2$ whose level set is the same ellipse, and compute the limiting variance of a boundary-supported linear statistic $f$. If the variance is identical for both potentials, the flow-dependent formula for $\\Sigma^1_D$ is wrong.","supporting_citations":[{"cited_title":"Quasimodes and Bohr-Sommerfeld Conditions for the Toeplitz Operators","cited_arxiv_id":null,"evidence_quote":"Supplies the Bohr–Sommerfeld structure of eigenfunctions near $\\{V=\\mu\\}$, which underlies the approximate Toeplitz form of edge matrix elements in Proposition 3.6."},{"cited_title":"Central limit theorem for smooth statistics of one- dimensional free fermions","cited_arxiv_id":null,"evidence_quote":"Provides the one-dimensional Szegő-type edge asymptotics and the Proposition 2.11 used to derive the approximate Toeplitz structure for the two-dimensional Berezin–Toeplitz edge."},{"cited_title":"The noise in the circular law and the Gaussian free field.International Mathematics Research Notices, (9):rnm006, 2007","cited_arxiv_id":null,"evidence_quote":"Establishes the Ginibre ensemble fluctuation result with bulk and boundary variance terms that the present two-term formula generalizes."},{"cited_title":"Complex determinantal processes and H1 noise","cited_arxiv_id":null,"evidence_quote":"Gives the $\\infty$-Ginibre bulk CLT with $H^1$ variance, directly used for the $\\Sigma^2_D$ contribution in the bulk."},{"cited_title":"Random normal matrices and Ward identities","cited_arxiv_id":null,"evidence_quote":"Provides the normal-matrix two-term variance with boundary term, used for comparison and for contrasting conformal invariance."},{"cited_title":"Semiclassical Analysis, volume 138","cited_arxiv_id":null,"evidence_quote":"Supplies the pseudodifferential conjugation (Theorem 13.10) invoked in Proposition 3.6 to relate Beezin–Toeplitz operators to pseudodifferential operators."},{"cited_title":"Fractional exponential decay in the forbidden region for Toeplitz operators","cited_arxiv_id":null,"evidence_quote":"Provides the exponential concentration estimates for eigenfunctions in the forbidden region, used for the kernel decay and decorrelation estimates."}],"review_version":1}