{"id":"f3393415-5099-460b-920f-69deca18ad52","arxiv_id":"2506.19971","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A framework for multiple operator integrals on non-Hermitian matrices via Jordan decomposition, with perturbation and derivative formulas that are not rigorously established.","lead":"This paper defines Generalized Multiple Operator Integrals for arbitrary matrices via Jordan decomposition, aiming to extend operator integral calculus beyond Hermitian matrices. It states norm bounds, perturbation formulas, and derivative formulas, but the proofs rely on unproven smoothness of Jordan forms and some estimates appear invalid.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 7's derivative formula is not defined for generic non-normal X: it requires differentiating the Jordan data of X+tY, which diverges for X=E12, Y=E21.","rationale":"I agree with the reader that the smooth-variation assumption on Jordan decompositions is a load-bearing gap, but I locate the decisive failure in Theorem 7, the paper's stated application. The theorem claims an n-th derivative representation for 'any matrices X and Y.' Its proof passes through correction terms X^(i) that require derivatives with respect to t of the Jordan projectors and nilpotents of X+tY at t=0. For non-normal X these derivatives generically do not exist; the explicit 2x2 example X = E12, Y = E21 has X+tY with eigenvalues ±√t, so the Jordan data is not differentiable at t=0. This is not a minor technicality: it makes the RHS of Eq. (110) undefined for an elementary case where the LHS is perfectly well defined. The same failure also undermines Lemma 3 and Theorem 5, but the derivative theorem is the central deliverable promised in the abstract and introduction. I additionally note that the lower norm bound in Theorem 2 appears invalid for related reasons, since the reduction from ||A0|| to min|beta| times ||Y1...Yzeta|| is not justified for zeta >= 2; however, fixing that bound would not rescue Theorem 7. The concrete test with a 2x2 nilpotent perturbation would settle the matter computationally and directly.","tokens_in":49584,"tokens_out":16142,"duration_ms":170308,"concrete_test":"Compute exactly, for X = E12, Y = E21, f(z) = z^3, the second derivative at t=0: (X+tY)^3 = [[0,t],[t^2,0]], so d²/dt²|0 = Y. Then evaluate the right-hand side of Theorem 7's Eq. (113): the three GMOI terms are finite, but the X^(1) terms are defined in Section 5.1 through derivatives of P(t) and N(t) for X+tY, which for this choice have entries whose t-derivatives diverge as t→0. If the formula cannot be evaluated or yields a value different from Y, the theorem is disproved.","verdict_should_be":"REJECT","load_bearing_attack":"The central advertised result, Theorem 7 (Eq. 110), is not established for the stated generality 'any matrices X and Y.' The proof of Lemma 5 and the subsequent X^(i) correction terms require differentiating, at t=0, the Jordan projectors P_{kd,id}(t) and nilpotents N_{kd,id}(t) of X+tY. These derivatives are not guaranteed to exist for non-normal X, and generically do not. Take X = [[0,1],[0,0]] and Y = [[0,0],[1,0]]. Then X+tY = [[0,1],[t,0]] has eigenvalues ±√t for t>0; as t→0 the two Jordan blocks coalesce and the spectral projectors contain factors involving √t, so their t-derivatives diverge. Hence the correction terms X^(1) in Eq. (113) are undefined, even though d²/dt² f(X+tY)|_{t=0} exists and is finite for analytic f. Theorem 7 therefore fails as a statement about arbitrary matrices; it would need a hypothesis such as stable geometric multiplicities along the ray, which is neither stated nor satisfied in this basic non-normal example. The manuscript itself acknowledges after Lemma 5 that the exact computation of X^(1) 'is tedious and there is no simple formula,' confirming that this is a claim-without-derivation at the load-bearing point of the paper.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript introduces Generalized Multiple Operator Integrals (GMOIs) for arbitrary square matrices via Jordan decompositions, claiming to unify classical multiple operator integrals and to provide norm estimates, perturbation formulas, Lipschitz bounds, continuity theorems, and a representation of n-th derivatives of matrix functions in terms of GMOIs. The central advertised result is Theorem 7 (Eq. (110)), which purports to express d^n f(X+tY)/dt^n at t=0 for any matrices X and Y.","tokens_in":49891,"tokens_out":13226,"duration_ms":137782,"significance":"If correct, the framework would offer a useful extension of MOI calculus beyond Hermitian matrices, potentially impacting non-Hermitian perturbation theory and matrix functional calculus. The paper has a systematic algebraic setup, and the upper-norm bound in Theorem 2 (Eq. (38)) is a routine triangle inequality that appears sound. However, the main derivative and continuity claims rely on smoothness properties of Jordan decompositions that are neither stated nor generally true, and the principal theorem fails on a simple 2x2 non-normal example. The significance of the paper is therefore not established in its current form.","major_comments":[{"comment":"Theorem 7 is stated for 'any matrices X and Y', but its proof requires the derivatives at t=0 of the Jordan projectors P_{kd,id}(t) and nilpotents N_{kd,id}(t) of X+tY, as made explicit in Example 5 (Eq. (132)). For X = [[0,1],[0,0]] and Y = [[0,0],[1,0]], the matrix X+tY has eigenvalues ±√t for t>0, and the associated spectral projectors contain factors of 1/√t, so P(t) and N(t) are not differentiable at t=0 even though d²/dt² f(X+tY)|_{t=0} exists for analytic f. Consequently, the correction terms X^(i) appearing in Eq. (110) are undefined in this basic non-normal case. The theorem requires an additional hypothesis such as stable geometric multiplicities along the ray, which is neither stated nor satisfied in the example.","section":"Section 7, Theorem 7 (Eq. (110)) and Lemma 5 (Eqs. (106)-(107))"},{"comment":"The proof of Lemma 3 assumes 'under smooth variation of the Jordan decomposition (i.e., assuming stable geometric multiplicities)' and then invokes Fréchet derivatives of the Jordan projectors and nilpotents. This hypothesis is not stated in Lemma 3 or in Theorem 5, and it fails for non-normal matrices where eigenvalues can coalesce. The cancellation argument for the difference T3−T4 also relies on unproved convergence of the index pairs (kc,ic) to (kd,id). As a result, the continuity theorem (Theorem 5, Eq. (96)) is not established for the claimed generality.","section":"Section 6, Lemma 3 (proof, Eqs. (91)-(93))"},{"comment":"The perturbation formula in Theorem 3 is load-bearing for both Theorem 5 and Theorem 7, but its proof only verifies identities (65) and (66), with the remaining identities (67)-(72) asserted to follow similarly without proof. Moreover, the derivation of Eq. (73) contains the assertion 'P_{kc,ic}P_{kc,ic}=P_{kd,id}P_{kd,id}=I', which is false for idempotent projectors unless a summation over all indices is intended; the displayed algebra does not rigorously justify the claimed correction terms. This gap undermines the foundation for the subsequent continuity and differentiation results.","section":"Section 5.1, Theorem 3"},{"comment":"The proof of Lemma 4 applies Theorem 3 with C=(X+tY)_N and D=X_N, but the two GMOI terms displayed in Eq. (104) are identical (both have argument tY) and cancel trivially. No argument is given to show that the difference of nilpotent parts equals tY, nor that the correction term X(...) vanishes. Since Lemma 4 is used in the proof of Theorem 7 to discard certain derivative terms, this is a substantive gap in the main theorem.","section":"Section 7, Lemma 4 (Eq. (104))"},{"comment":"The lower-bound estimate uses the step ∥Σ β(...) P...P ∥ ≥ min|β| ∥Σ P...P ∥, which does not follow from the triangle inequality and is not generally valid for non-orthogonal projectors. The equality ∥Σ P Y P ∥ = ∥Y_1...Y_ζ∥ requires the projectors to form a resolution of the identity, which is not stated or proved in the non-Hermitian setting. A rigorous justification of Eq. (45) is needed before the lower bound in Eq. (41) can be accepted.","section":"Section 4, Theorem 2 (Eq. (41) and Eq. (45))"}],"minor_comments":[{"comment":"There are numerous typos and unclear notations, including 'Fr ´echet', 'Propotion', 'geometry multiplicities' for 'geometric multiplicities', and malformed expressions such as 'B4,3))' in Eq. (34). A careful editorial pass is needed.","section":"Throughout"},{"comment":"The foundational Theorem 1 (from the author's preprint [10]) and Theorem 6 (from the author's preprint [9]) are cited rather than proved. Since these results are load-bearing for the GMOI definition and the derivative application, the paper is not self-contained, and the dependence on unpublished preprints should be addressed.","section":"Sections 2 and 7"},{"comment":"The definition of the GMOI sets Y_{ζ+1}=I but the notation and index ranges for the parameter matrices and argument matrices should be checked for consistency; the current presentation is confusing.","section":"Eq. (17)"},{"comment":"The assumption that all partial derivatives of β[ζ+1] up to any order are bounded is odd because β[ζ+1] is evaluated on finite spectra; the intended smoothness hypothesis should be stated precisely.","section":"Section 6, Theorem 5"},{"comment":"The formulas (132)-(138) for high-order derivatives of the correction terms are stated without proof and assume the differentiability of spectral data. Given the counterexample above, these formulas are not valid for arbitrary non-normal matrices.","section":"Section 7, Example 5"}],"recommendation":"reject","confidential_remarks":"The manuscript depends heavily on the author's own unpublished preprints ([8], [9], [10]) for its foundational expansion theorem and first-derivative formula. The central derivative theorem fails for a basic non-normal example, and the continuity proof rests on an unstated smoothness assumption. The framework may be salvageable under restrictive hypotheses on geometric multiplicities, but as stated the main claims are not correct. The editor may also wish to consider whether the extensive reliance on self-citations to unpublished work meets the journal's standards."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Shih-Yu Chang's paper develops a Jordan-decomposition-based multiple operator integral for arbitrary finite matrices, with norm bounds, perturbation formulas, and a claimed formula for derivatives of matrix functions. The algebraic core is the genuinely new part: it rewrites the standard non-commutative divided-difference expansion of f(X1,...,Xr) in terms of projectors and nilpotents, and the composition rule in Proposition 2 shows the formalism has internal coherence. The upper norm bound is a routine triangle inequality, and the Lipschitz estimate in Theorem 4 follows from telescoping plus that bound; those parts are basically fine.\n\nThe soft spots are where the reader's report put them. The lower bound in Eq. (41) is not justified for non-orthogonal projectors: the step bounding the norm of A0 below by min|beta| times the product of the Y-norms uses a spectral-norm-type property that Frobenius sums of non-orthogonal projections do not have. That is a real error, though it is not load-bearing for everything else.\n\nThe load-bearing problem is Theorem 7. The proof, including the X^(i) correction terms, requires differentiating the Jordan projectors and nilpotents of X+tY at t=0. For non-normal X this fails in general. The stress-test example X=E12, Y=E21 is clean: X+tY has eigenvalues ±√t, so the spectral projectors have 1/√t singularities as t→0, and the t-derivatives of the Jordan data do not exist, while the second derivative of f(X+tY) is perfectly finite for analytic f. So Theorem 7 as stated, for arbitrary matrices X and Y, is false. The manuscript itself concedes after Lemma 5 that X^(1) has \"no simple formula,\" which is a claim-without-derivation at the hinge. A hypothesis of stable geometric multiplicities along the ray might repair it, but that is not what is stated, and the example shows the missing hypothesis is not cosmetic.\n\nThe continuity theorem also rests on Lemma 3, whose proof explicitly assumes stable geometric multiplicities; the theorem statement omits that assumption, and norm convergence alone does not imply it. So the continuity claim is not established in the stated generality.\n\nWho is this for? A functional analyst working on non-Hermitian perturbation theory will find the GMOI definition useful as a formal calculus, and the algebraic properties are worth knowing. But the analytic claims as stated are not proved. I would not cite the main theorems in current form, though I might cite the GMOI definition as a notation if needed. The paper deserves a serious referee—it is not incoherent algebra—but it should return only after the derivative theorem is either proved under explicit stable-multiplicity assumptions or downgraded to a formal identity. Send it to peer review with the smooth-Jordan hypothesis and the projector lower bound flagged as mandatory revisions.","headline":"The GMOI construction is a real algebraic extension, but the derivative theorem fails for generic non-normal matrices because it silently requires differentiating Jordan data that need not exist.","tokens_in":50378,"tokens_out":3261,"would_cite":false,"duration_ms":38394,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["47A60","47A55","15A60"],"pacs":[],"model":"deepseek-v4-flash","headline":"A generalized multiple operator integral extends matrix calculus to non-Hermitian matrices.","keywords":["multiple operator integrals","non-Hermitian matrices","Jordan decomposition","divided differences","perturbation formula","matrix function derivatives","Frobenius norm estimates","functional calculus"],"falsifier":"Take $D$ to be a $2\\times 2$ Jordan block and $V$ a rank-one perturbation that makes $D+tV$ diagonalizable for every $t>0$; compute the GMOI difference and the correction term $\\overline{X}(\\ldots)$ as $t\\to 0$. If the correction term is not $o(t)$, or if the second-derivative formula of Example 3 fails to match a direct finite-difference evaluation of $\\frac{d^2}{dt^2}f(D+tV)|_{t=0}$, then Lemma 3 and Theorem 7 would be contradicted.","tokens_in":49354,"feed_emoji":"🧮","tokens_out":5339,"duration_ms":50812,"temperature":0.7,"pith_summary":"The paper proposes Generalized Multiple Operator Integrals (GMOIs), a finite-dimensional replacement for Multiple Operator Integrals that works for arbitrary square matrices, not only Hermitian ones. It defines the GMOI through the Jordan decomposition of each parameter matrix, adding divided-difference terms that carry the nilpotent parts, so the integral encodes the full Jordan structure rather than just eigenvalues. The paper claims this recovers conventional MOIs as a special case, supplies Frobenius-norm upper and lower bounds, and proves a perturbation formula with explicit correction terms. From that it derives Lipschitz estimates and a continuity theorem for the GMOI under matrix perturbations, and it represents the n-th derivative of a matrix function as a combination of GMOIs and correction terms. A sympathetic reader would care because non-Hermitian matrices are the rule in numerical linear algebra and non-Hermitian physics, and this gives them a functional-calculus calculus analogous to the Hermitian one.","feed_headline":"Operator integrals extend to non-Hermitian matrices","feed_subtitle":"Jordan decomposition plus divided differences yields derivative formulas and perturbation bounds for arbitrary square matrices.","key_machinery":"The load-bearing object is the GMOI of Eq. (17): a finite sum over eigenvalue indices and geometric components in which each parameter matrix contributes either its spectral projector $P_{k,i}$ or a nilpotent power $N^q_{k,i}$, weighted by normalized partial derivatives (divided differences) of the kernel $\\beta$. The binary-expansion notation $\\Psi$ in Section 4 organizes the $2^{\\zeta+1}$ choices of projector versus nilpotent at each parameter slot; this expansion is what makes the norm estimates tractable. The perturbation formula in Theorem 3 rests on Lemma 2, a divided-difference identity that converts differences $\\beta^{[k]}(\\ldots,\\lambda,\\ldots)-\\beta^{[k]}(\\ldots,\\mu,\\ldots)$ into $(\\lambda-\\mu)\\beta^{[k+1]}(\\ldots,\\lambda,\\mu,\\ldots)$, so eigenvalue differences are traded for one more GMOI slot. The correction terms $X(\\ldots)$ collect all residual terms involving differences of nilpotents $N_C-N_D$.","core_discovery":"The central claim is that the family defined in Eq. (17), built from spectral projectors $P_{k,i}$ and nilpotents $N_{k,i}$ of each matrix $X_p$ via Jordan decomposition, is the correct generalization of multiple operator integrals. For any analytic kernel $\\beta$, the GMOI $T^{X_1,\\ldots,X_{\\zeta+1}}_{\\beta}(Y_1,\\ldots,Y_\\zeta)$ expands into one term using only projectors plus partial sums over nilpotent powers with divided-difference weights $\\beta^{(q)}(\\lambda)/q!$, so it reproduces the Taylor expansion of $f(X_1,\\ldots,X_r)$ from Theorem 1. Theorem 3 gives a perturbation formula expressing the difference between GMOIs evaluated at $C$ and $D$ as GMOIs of $C-D$ plus correction terms, and Theorem 7 uses iterated perturbation to write $d^n f(X+tY)/dt^n$ at $t=0$ as $n!$ times a GMOI with divided-difference kernel $f^{[n]}$, minus sums of GMOIs with nilpotent parameter pairs and correction terms $X^{(i)}$.","pith_inferences":["If the continuity theorem is right, the GMOI gives a route to Fréchet derivative formulas for non-analytic matrix functions defined through divided differences on non-normal matrices, not just analytic functions.","A natural test is to compute the second-derivative formula of Example 3 for a $2\\times 2$ Jordan block with $Y$ coupling the chain; direct finite differences should match the GMOI expression only if the correction terms are retained.","The paper treats finite dimensions throughout; the same projector-nilpotent expansion suggests a path toward infinite-dimensional operators with discrete non-normal spectra, but continuous-spectrum cases need a mechanism other than Jordan decomposition.","Where geometric multiplicities change (eigenvalue coalescence), the paper's smooth-Jordan assumption fails, so one would expect the perturbation formula to require additional spectral projectors or to break down entirely."],"forward_implications":["Conventional Hermitian MOIs become the projector-only summand of the GMOI, so the Hermitian theory is a special case rather than a separate construction.","The Frobenius-norm bounds give explicit control on how large a GMOI can be in terms of kernel maxima and nilpotent norms, which is what makes perturbation series manageable.","The Lipschitz estimate bounds the change in a GMOI when the argument matrices $Y_i$ change, with scalar constants computed from the upper bounds.","Continuity of the GMOI under $X_{i,\\ell}\\to X_i$ holds when the Jordan data vary continuously; under that condition perturbative expansions converge.","Every $n$-th derivative of a matrix function at a non-Hermitian matrix $X$ is expressed as a finite combination of GMOIs with divided-difference kernels plus computable correction terms."],"supporting_citations":[{"why":"Supplies Theorem 1, the Taylor expansion of analytic matrix functions via projectors and nilpotents, on which the GMOI definition is built.","marker":"[10]"},{"why":"Provides the finite-dimensional generalized double operator integral and Lemma 3 on the converse triangle inequality used in the norm estimates.","marker":"[8]"},{"why":"Gives the first-order derivative identity $\\frac{d}{dt}f(X+tY)=T^{X+tY,X+tY}_{f^{[1]}}(Y)$ that Theorem 7 iterates.","marker":"[9]"},{"why":"Defines the conventional multiple operator integral for Hermitian matrices, the object the paper recovers as a special case.","marker":"[11]"}],"fun_headline_variants":["Operator integrals branch out to non-Hermitian","Non-Hermitian matrix calculus via operator integrals","Generalized operator integrals for arbitrary matrices","Jordan decomposition unlocks operator integral theory"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes that, as one matrix is smoothly deformed into another, its Jordan blocks (the projectors and nilpotents) can also be chosen to deform smoothly as long as eigenvalue multiplicities do not change; without that assumption, the continuity and derivative conclusions do not follow.","fun_headline_variants_meta":{"raw":{"variants":["Operator integrals branch out to non-Hermitian","Non-Hermitian matrix calculus via operator integrals","Generalized operator integrals for arbitrary matrices","Jordan decomposition unlocks operator integral theory"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000321,"raw_usage":{"total_tokens":1842,"prompt_tokens":1016,"completion_tokens":826,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":632,"completion_tokens_details":{"reasoning_tokens":772}},"tokens_in":632,"tokens_out":826,"duration_ms":8668,"temperature":1.0,"reasoning_tokens":772,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T18:22:58.943777+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $D$ to be a $2\\times 2$ Jordan block and $V$ a rank-one perturbation that makes $D+tV$ diagonalizable for every $t>0$; compute the GMOI difference and the correction term $\\overline{X}(\\ldots)$ as $t\\to 0$. If the correction term is not $o(t)$, or if the second-derivative formula of Example 3 fails to match a direct finite-difference evaluation of $\\frac{d^2}{dt^2}f(D+tV)|_{t=0}$, then Lemma 3 and Theorem 7 would be contradicted.","supporting_citations":[{"cited_title":"Skripka and A","cited_arxiv_id":null,"evidence_quote":"Defines the conventional multiple operator integral for Hermitian matrices, the object the paper recovers as a special case."}],"review_version":2}