{"id":"5722a876-9080-4e38-8718-71e6293210ca","arxiv_id":"2507.01443","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Discrete Dirac operators on 3D lattices converge to continuum Dirac operators strongly but not in norm as the mesh size goes to zero.","lead":"Mathematicians prove that discretized Dirac operators on 3D cubic lattices converge to the continuous Dirac operator in the strong resolvent sense as the mesh size shrinks to zero, but not in the uniform operator norm sense. The result clarifies which discretizations faithfully approximate relativistic quantum dynamics and why standard lattice Dirac operators fail to converge uniformly.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Second lower bound in Theorem 2.1(ii) uses u_h=(0,0,y_h,0)^T, but at the chosen Fourier point e_3 is not a −m eigenvector; e_4 is, so the 'analogously' step needs correction.","rationale":"The central strong-convergence claim (i) is supported by a standard Fourier argument; its reliance on Lemmas 3.3, 3.5 and 3.6 of [3] is a presentation concern, not an identified error, since the cube-lattice analogues of those lemmas are routine. The gap I find is in part (ii): the second lower bound is claimed to follow 'analogously' from u_h=(0,0,y_h,0)^T, but at the Fourier point where y_h is concentrated the third basis vector is mixed by an O(1/h) off-diagonal symbol entry, so the analogue fails literally. The paper's own computations (Step 2 and the D1,h/D2,h estimates) are otherwise consistent, and the first lower bound 1/|m−z| is sound. Because the correction is a one-character change (e3 to e4) and all estimates then go through, the theorem is very likely true, but the manuscript as printed does not prove the second half of (ii). A conditional acceptance, pending the corrected test vector, is the honest verdict.","tokens_in":9261,"tokens_out":29789,"duration_ms":300227,"concrete_test":"Compute the block structure of bD_{m,h}(π/(2h),−π/(2h),0) from (2.4) and (1.2): verify that e_4 is an eigenvector with eigenvalue −m while e_3 is not, then re-run Step 3 of the proof with u_h=(0,0,0,y_h)^T. The error terms should follow the same O(h^{1/2})+O(h^{3/2}) estimates, yielding liminf_{h→0} ∥(D_{m,h}⊕0_h−z)^{-1}−(D_m−z)^{-1}∥_B ≥ 1/|m+z|.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proof of the second lower bound in Theorem 2.1(ii) is not valid as written. With y_h(x)=h^{3/4}e^{iπ(x1−x2)/(2h)}e^{-h|x|^2}, the Fourier transform F y_h is concentrated at ξ0=(π/(2h),−π/(2h),0). At this point, using (2.4) and the matrix (1.2), the discrete symbol acts on the third standard basis vector as bD_{m,h}(ξ0)e_3=(0, 2(1−i)/h, −m, 0)^T, because i∂*_1−∂*_2 has symbol (1−i)/h−(i−1)/h=2(1−i)/h. Hence e_3 is not an eigenvector, and the O(1/h) off-diagonal entry prevents (bD_{m,h}−z)^{-1}P_h(0,0,y_h,0)^T from being close to (−m−z)^{-1}P_h(0,0,y_h,0)^T; the claimed liminf ≥1/|m+z| does not follow from the cited 'analogously' argument. The correct test vector is (0,0,0,y_h)^T: at the same ξ0, i∂*_1+∂*_2 has symbol (1−i)/h+(i−1)/h=0, and bD_{m,h}(ξ0)e_4=−m e_4. With that replacement, Step 3's estimates carry over and give the missing bound. This is a concrete gap in the printed proof, although it is repairable.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the continuum limit of the free Dirac operator discretized on a three-dimensional cubic lattice with forward and backward difference operators. The discrete operator is embedded into L2(R^3)^4 by extending lattice functions to step functions, and the main theorem (Theorem 2.1) states that the resolvent of the embedded discrete Dirac operator, (D_{m,h} ⊕ 0_h - z)^{-1}, converges strongly to (D_m - z)^{-1} as h → 0, but not in the norm resolvent sense, with an explicit lower bound liminf_{h→0} ||(D_{m,h} ⊕ 0_h - z)^{-1} - (D_m - z)^{-1}||_B ≥ max(1/|m-z|, 1/|m+z|). The proof uses the discrete Fourier transform on the torus T^3_{1/h}, compares the discrete and continuum Fourier symbols, and constructs oscillatory Gaussian test functions to establish both the strong convergence and the lower bound on the norm discrepancy.","tokens_in":9639,"tokens_out":11765,"duration_ms":102844,"significance":"If the result is correct, it provides a clean three-dimensional analogue of the authors' two-dimensional result and sharpens the known picture by giving an explicit quantitative obstruction to norm resolvent convergence for the natural step-function embedding. The paper contains detailed, checkable estimates, including explicit constants in the lower bound, and it explains why modifications such as those in [1] and [2] are needed to achieve norm convergence. These are genuine strengths. The main limitations are the reliance on unstated three-dimensional versions of technical lemmas from the authors' prior two-dimensional paper [3], and a local but load-bearing error in the second lower bound of Theorem 2.1(ii). Both are repairable without changing the main claim.","major_comments":[{"comment":"The claimed second lower bound 1/|m+z| is not proved by the stated 'analogously' argument with u_h = (0,0,y_h,0)^T. The Fourier transform of y_h is concentrated at ξ0 = (π/(2h), −π/(2h), 0); at this point, using (2.4) and the matrix (1.2), bD_{m,h}(ξ0)e_3 = (0, 2(1−i)/h, −m, 0)^T, because i∂*_1 − ∂*_2 has symbol 2(1−i)/h there. Thus e_3 is not an eigenvector of the discrete symbol, and the O(1/h) off-diagonal entry prevents (bD_{m,h}−z)^{-1}P_h u_h from being close to (−m−z)^{-1}P_h u_h. The correct vector is u_h = (0,0,0,y_h)^T: at the same ξ0, i∂*_1 + ∂*_2 has symbol 0 and bD_{m,h}(ξ0)e_4 = −m e_4. With this replacement, the Step 3 estimates carry over and give the claimed lower bound. Please make the replacement explicit and repeat the Step 1–3 estimates for the fourth component.","section":"Section 2, proof of Theorem 2.1(ii)"},{"comment":"The strong convergence argument uses Lemmas 3.3, 3.5 and 3.6 of [3], which are stated in [3] for two-dimensional square lattices. The current paper is three-dimensional, and these lemmas control the strong convergence of P_h to the identity, of F_h to F, and of the difference F_h − F on Schwartz functions. Since these properties are load-bearing for the convergence proof in (2.9)–(2.12), the authors should either state the three-dimensional versions with proofs or explicitly indicate that the proofs in [3] are dimension-independent and carry over verbatim to R^3.","section":"Section 2, equations (2.9)–(2.12)"}],"minor_comments":[{"comment":"The displayed formula for the norm of (8/π^2)(y_h)_h has a typo: the leading constant should be 8/(π^{5/4} 2^{3/4}), not 8/π^{5/4} 2^{3/4}; the printed version differs by a factor of 2^{3/2}. This does not affect the subsequent argument, which only needs the norm to tend to a nonzero constant.","section":"Section 2, after equation (2.17)"},{"comment":"In the line containing '∥(Dmh − z)^{-1}Ph uh∥', the operator 'Dmh' should be 'D_{m,h}'.","section":"Section 2, final displayed estimate of Theorem 2.1(ii)"},{"comment":"The notation (bD_{m,h} − z)^{-1} ⊕ 0_h is used after the Fourier extension of F_h is introduced, but the extension of 0_h to the Fourier side is not defined. Please define the action of this operator on L2(T^3_{1/h}) and its orthogonal complement explicitly.","section":"Section 2, equation (2.9)"}],"recommendation":"major_revision","confidential_remarks":"The main theorem is likely correct, and the error in the second lower bound is local and straightforward to repair by replacing the test vector with (0,0,0,y_h)^T. The reliance on [3] for technical lemmas is acceptable if the authors confirm the dimension independence. I see no ground for rejection; the paper is within the scope of the journal and the result is a solid extension of the authors' earlier work."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [colleague],\n\nThe paper does what it says: it proves that the embedded discrete Dirac operator on a 3D cubic lattice converges to the continuum Dirac operator in strong resolvent sense, and that norm resolvent convergence fails, with an explicit liminf bound max(1/|m−z|, 1/|m+z|). The Fourier proof is clean and shorter than the authors' 2D version, and the lower-bound construction with the oscillating Gaussian is a neat trick.\n\nI checked the symbol computations and the estimates in Steps 1–3; the remainder bounds and the D1,h/D2,h estimates are fine. The strong convergence argument uses the standard Fourier truncation plus [3, Lemmas 3.3/3.5/3.6]. Those lemmas are about the projection Ph and the discrete Fourier transform and should hold in 3D verbatim, but the paper should say so explicitly instead of sending the reader back to the 2D paper.\n\nHere is the genuinely soft spot. The second lower bound in (ii) is not proved as written. The last line claims it follows 'analogously' from u_h=(0,0,y_h,0)^T. At the Fourier point (π/(2h), −π/(2h), 0) where the Gaussian concentrates, e_3 is not an eigenvector of the discrete symbol: bD_{m,h}(ξ0)e_3 has an O(1/h) component in the second slot, so the resolvent applied to that vector is not close to 1/(−m−z) times the vector. The correct test vector is (0,0,0,y_h)^T; then e_4 is a −m eigenvector and the 'analogously' argument goes through with the same remainder estimates. This is a typo-level error, not a fatal one, but as printed that half of (ii) has a gap.\n\nThe citation pattern is honest. The authors compare with [1] and [2] and state that those papers use modified operators to achieve norm resolvent convergence; this paper treats the unmodified operator with the natural step-function embedding. The overlap is real but the explicit lower bound in 3D is new.\n\nThis paper is for people working on lattice approximations of Dirac operators and for numerical analysts who need to know that the naive discretization of the resolvent does not converge uniformly. It deserves a serious referee; the fix I described is small and should be easy to make. My recommendation: accept after that correction is inserted.\n\nBest,","headline":"Clean 3D proof of strong-but-not-norm resolvent convergence for unmodified lattice Dirac operators with an explicit lower bound; the printed second lower bound has a typo-level error that is easy to fix.","tokens_in":10183,"tokens_out":8390,"would_cite":true,"duration_ms":79543,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["47A10","47B37","47B93"],"pacs":[],"model":"deepseek-v4-flash","headline":"Lattice Dirac resolvents converge strongly, not in norm","keywords":["discrete Dirac operator","resolvent convergence","strong convergence","norm resolvent convergence","cubic lattice","continuum limit","discrete Fourier transform","spectral theory"],"falsifier":"Numerically compute the operator norm of $(D_{m,h}\\oplus 0_h-z)^{-1}-(D_m-z)^{-1}$ on a finite but large cubic lattice for a fixed $m$ and non-real $z$, taking $h$ smaller at each step; if the norm is ever found to drop below $\\max(1/|m-z|,1/|m+z|)$ by a non-vanishing amount, the lower bound of Theorem 2.1(ii) is false.","tokens_in":9052,"feed_emoji":"📐","tokens_out":7911,"duration_ms":79980,"temperature":0.7,"pith_summary":"This paper proves that the free Dirac operator on a three-dimensional cubic lattice, embedded into the continuum by extending lattice functions to step functions, converges to the continuum Dirac operator in the strong resolvent sense as the lattice spacing $h$ goes to zero. It also proves that the same resolvents do not converge in the operator norm, and gives an explicit lower bound for the norm of their difference in the limit. Strong resolvent convergence is enough to guarantee the matching of spectra and of many spectral quantities, while norm resolvent convergence would be needed for uniform control over all states. The three-dimensional proof is shorter and more direct than the prior two-dimensional argument.","feed_headline":"Lattice Dirac resolvents converge strongly, not in norm","feed_subtitle":"Even as the mesh shrinks, the resolvent difference never vanishes uniformly; the paper pins down the exact lower bound.","key_machinery":"The proof rests on the step-function embedding $J_h$ that maps lattice functions into $L^2(\\mathbb{R}^3)$, its orthogonal projection $P_h$, and the discrete Fourier transform $F_h$ on the torus $[-\\pi/h,\\pi/h]^3$. Working in the Fourier representation turns the discrete Dirac operator into a multiplication operator whose symbol $\\widehat{D}_{m,h}(\\xi)$ approaches the continuum symbol $\\widehat{D}_m(\\xi)$ pointwise, but not uniformly. The failure of uniform convergence is detected by a family of Gaussian test functions $u_h$ concentrated near the boundary of the Brillouin zone; explicit asymptotic estimates show that the projected vector $(D_{m,h}-z)^{-1}P_h u_h$ has norm approaching $\\|u_h\\|/|m-z|$, proving the lower bound.","core_discovery":"The central result is Theorem 2.1: for any non-real $z$ in the resolvent set, the embedded discrete resolvent $(D_{m,h}\\oplus 0_h-z)^{-1}$ converges strongly to the continuum resolvent $(D_m-z)^{-1}$ in $L^2(\\mathbb{R}^3)^4$ as $h\\to 0$, yet the operator norm of their difference satisfies $\\liminf_{h\\to 0}\\|(D_{m,h}\\oplus 0_h-z)^{-1}-(D_m-z)^{-1}\\| \\ge \\max(1/|m-z|,1/|m+z|)$. The lower bound is constructed by test functions whose Fourier transform is localised near the corners of the Brillouin zone, where the discrete symbol differs from the continuum symbol by an error that does not vanish in norm.","pith_inferences":["The construction of the test functions suggests that the non-convergence is a high-frequency phenomenon at the edges of the Brillouin zone, so any discretisation that damps or filters these high frequencies (for example, using a smoother embedding or a non-uniform lattice) might achieve norm resolvent convergence.","The proof technique, relying only on the symbol comparison and projection estimates, may extend to other lattice geometries and to higher dimensions, yielding the same strong-but-not-norm dichotomy.","The lower bound depends on the specific step-function embedding; it would be informative to test whether other natural embeddings that also 'preserve the discrete operator' give different limiting behaviour."],"forward_implications":["Strong resolvent convergence implies that the spectra of the embedded discrete operators converge to the spectrum of the continuum Dirac operator in the Hausdorff distance as $h\\to 0$.","Because norm resolvent convergence fails, functions of the resolvent such as spectral projections and semigroups do not converge uniformly, so uniform approximation of time evolution on the whole space is not guaranteed.","The explicit lower bound $\\max(1/|m-z|,1/|m+z|)$ quantifies the unavoidable error near the mass-shell singularities of the continuum resolvent.","The paper's remark that central difference operators do not change the fundamental issue indicates that simply altering the finite-difference scheme cannot restore norm convergence."],"supporting_citations":[{"why":"Supplies the lemmas on discrete Fourier transform and projection convergence (Lemmas 3.3, 3.5, 3.6) that the 3D strong-convergence proof uses without restatement.","marker":"[3]"},{"why":"Discusses modified discrete Dirac operators that achieve norm resolvent convergence, providing the contrast that motivates the non-convergence result.","marker":"[1]"},{"why":"Offers additional discussion of discrete Dirac operators and their continuum limits, cited for alternative modifications that restore norm convergence.","marker":"[2]"}],"fun_headline_variants":["Lattice Dirac: strong resolvent limit, norm gap persists","Strong but not uniform: discrete Dirac resolvent limit","3D Dirac resolvents: strong limit, never uniform","Discrete Dirac limit: strong only, with explicit norm gap","Lattice Dirac: exact norm gap in resolvent convergence"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of strong convergence assumes, without re-proving, that three key lemmas on the discrete Fourier transform and the projection $P_h$ from the two-dimensional paper [3] remain valid verbatim in three dimensions; if any of these lemmas fails in $\\mathbb{R}^3$, the strong convergence argument has a gap.","fun_headline_variants_meta":{"raw":{"variants":["Lattice Dirac: strong resolvent limit, norm gap persists","Strong but not uniform: discrete Dirac resolvent limit","3D Dirac resolvents: strong limit, never uniform","Discrete Dirac limit: strong only, with explicit norm gap","Lattice Dirac: exact norm gap in resolvent convergence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001129,"raw_usage":{"total_tokens":4585,"prompt_tokens":726,"completion_tokens":3859,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":342,"completion_tokens_details":{"reasoning_tokens":3777}},"tokens_in":342,"tokens_out":3859,"duration_ms":31651,"temperature":1.0,"reasoning_tokens":3777,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T20:53:00.642322+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Numerically compute the operator norm of $(D_{m,h}\\oplus 0_h-z)^{-1}-(D_m-z)^{-1}$ on a finite but large cubic lattice for a fixed $m$ and non-real $z$, taking $h$ smaller at each step; if the norm is ever found to drop below $\\max(1/|m-z|,1/|m+z|)$ by a non-vanishing amount, the lower bound of Theorem 2.1(ii) is false.","supporting_citations":[{"cited_title":"Schmidt, T","cited_arxiv_id":null,"evidence_quote":"Supplies the lemmas on discrete Fourier transform and projection convergence (Lemmas 3.3, 3.5, 3.6) that the 3D strong-convergence proof uses without restatement."},{"cited_title":"Cornean, H","cited_arxiv_id":null,"evidence_quote":"Discusses modified discrete Dirac operators that achieve norm resolvent convergence, providing the contrast that motivates the non-convergence result."},{"cited_title":"Nakamura Remarks on discrete Dirac operators and their continuum limits.J","cited_arxiv_id":null,"evidence_quote":"Offers additional discussion of discrete Dirac operators and their continuum limits, cited for alternative modifications that restore norm convergence."}],"review_version":1}