REVIEW 1 major objections 3 minor 1 cited by
Some lemmata on the perturbation of the spectrum
T0 review · 1 major / 3 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read A Hilbert-space metric perturbation whose matrix elements decay fast enough leaves the two-term power asymptotics of a compact operator's eigenvalues unchanged.
desk verdict Theorem 1 has a real gap: the matrix-element decay condition on B does not pass to its positive and negative parts, so the main theorem is not proven as stated. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The argument is carried by the Rayleigh quotient $J(x) = (Kx,x)_H / ((x,x)_H + (Bx,x)_H)$ associated with the generalized eigenproblem. Restricting $J$ to the span of the first $n$ eigenfunctions of $K$ (or, for the negative part, to the tail subspace) reduces the problem to finite-dimensional comparisons. The pivotal identity (6) expresses the eigenfunction coefficients of the minimizer in terms of $(B\hat x, h_k)_H$; the decay condition on the matrix elements then makes the sums $\hat A$ and $\check A$ in (7)\textendash(8) small. This forces the Rayleigh minimizer value within $O(n^{-(1+\delta)})$ of $\lambda_n$, and the standard spectral-measure argument in Lemma 1 locks each generalized eigenvalue into a small interval free of other eigenvalues.
What would settle it
Take a diagonal $K$ with $\lambda_n = n^{-2}$, choose a compact self-adjoint $B$ whose entries in the eigenbasis satisfy $|(B h_n,h_m)_H| = (mn)^{-(1+\delta)/2}$ with alternating signs, and compute the generalized eigenvalues for large $n$. If the two-term expansion loses its $O(n^{-\delta})$ remainder for any $\delta>0$, the theorem's condition would not be sufficient.
Extended reading notes
Core claim
The central discovery is Theorem 1: the preservation of the two-term asymptotics follows from the individual matrix-element bounds $|(B h_n, h_m)_H| \le c (mn)^{-(1+\delta)/2}$ together with the assumed $\lambda_n = (a n + b + O(n^{-\delta}))^{-B}$. Lemma 1 first settles the simpler case where $\|B h_n\|_H \le c n^{-(1+\delta)}$; Theorem 1 replaces that operator-level decay by a globally weaker, more pointwise decay condition. In both cases the conclusion is the same: the generalized eigenvalues $\tilde\lambda_n$ admit the expansion $(a n + b + O(n^{-\delta}))^{-B}$. Remark 2 extends the same conclusion to spectra consisting of two interlacing sequences, such as even and odd indices with distinct constants.
Load-bearing premise
Everything rests on the requirement that the perturbation's off-diagonal coefficients in the unperturbed eigenbasis decay like the product of the two indices raised to a power strictly above one-half; if that decay fails, the coefficient estimates (7) and (8) no longer close.
Editorial extensions
If this is right
- For every operator satisfying the hypothesis of Theorem 1, the generalized eigenproblem has two-term asymptotics with the same constants $a$, $b$, $B$, and the same $O(n^{-\delta})$ remainder.
- The positive and negative cases are handled separately, so for a general self-adjoint $B$ the conclusion is obtained by squeezing between the min-max bounds for $B_+$ and $B_-$.
- By Remark 2, the result covers spectra with two interlacing eigenvalue sequences, so it applies to parity-class asymptotics such as $\lambda_n^{(1)} = ((2n-1)a + b_1 + O(n^{-\delta}))^{-B}$ and $\lambda_n^{(2)} = (2 n a + b_2 + O(n^{-\delta}))^{-B}$.
- The sufficient condition is aimed at integro-differential operators arising in the theory of fractional Gaussian processes, where the metric perturbation corresponds to a concrete covariance term.
Reading between the lines
- The proof only uses the spacing and decay of the unperturbed eigenvalues, so the same trapping argument should survive for asymptotics with logarithmic corrections, provided consecutive $\lambda_n$ stay separated by a power-like gap.
- The threshold $(mn)^{-(1+\delta)/2}$ suggests a natural boundary: perturbations with matrix elements of size $(mn)^{-1/2}$ might be able to shift the $b$-term, a question the paper does not address.
- For concrete covariance kernels, the matrix-element condition can be checked from kernel smoothness, making the theorem a practical stability criterion for two-term Weyl asymptotics in Gaussian-process settings.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies whether the two-term power-type spectral asymptotics lambda_n = (a n + b + O(n^{-delta}))^{-B} of a compact self-adjoint positive operator K are preserved when the Hilbert space scalar product is replaced by (h,g) -> ((I+B)h,g)_H, giving the generalized eigenproblem K h = lambda (I+B) h. Lemma 1 proves preservation under the norm decay condition ||B h_n|| <= c n^{-(1+delta)}. Theorem 1 claims the same conclusion under the weaker off-diagonal matrix-element decay |(B h_n, h_m)_H| <= c (mn)^{-(1+delta)/2}. Remark 2 extends the statement to two interlacing eigenvalue sequences. The proofs use spectral measure arguments in Lemma 1 and min-max/variational arguments in Theorem 1.
Significance. If the results are correct, they provide a concrete and checkable sufficient condition for the stability of two-term spectral asymptotics under metric perturbations, with potential applications to integro-differential operators arising in the theory of Gaussian processes. The paper is concise and fully self-contained, and Lemma 1 is cleanly proved. However, the main new result, Theorem 1, has a load-bearing gap in the reduction to sign-definite perturbations; the paper in its current form therefore does not establish the theorem for arbitrary (indefinite) B.
major comments (1)
- [Theorem 1, proof, first paragraph] The reduction to the cases B >= 0 and B <= 0 is not justified. The proof applies the estimates in equations (6)-(8) to the sign-definite operators B_+ and B_- individually, and these estimates use the matrix-element bound |(B_+ h_n, h_m)| <= c (mn)^{-(1+delta)/2} (respectively for B_-). The hypothesis supplies this bound only for the original operator B. Since B_+ and B_- are obtained from B by the spectral calculus, their matrix elements in the fixed basis {h_n} are not determined by the matrix elements of B; cancellations that make (B h_n, h_m) small need not persist in B_+ or B_-. Thus the sandwich lambda_n^+ <= lambda_n <= lambda_n^- does not imply the conclusion from the two sign-definite cases without an additional inheritance lemma. This is a load-bearing gap: as written, Theorem 1 is not proven for an arbitrary (indefinite) B satisfying the stated hypothesis.
minor comments (3)
- [Equations (1) and (3)] Throughout Lemma 1 and Theorem 1 the letter B denotes both the perturbation operator in (2) and the positive exponent in the asymptotics (1), (3); this makes statements such as lambda_n = (a n + b + O(n^{-delta}))^{-B} ambiguous. The exponent should be given a different symbol, for example beta.
- [Theorem 1, parts 1 and 2] The perturbed eigenvalues are not distinguished typographically from the unperturbed ones; for example, in part 1, "evidently lambda_n <= lambda_n" uses the same symbol on both sides, which is confusing and, taken literally, a tautology. Please introduce a distinct notation for the eigenvalues of the perturbed problem.
- [Theorem 1, derivative computation] In the displayed formula for the derivative of the Rayleigh quotient there is an extra closing parenthesis after (B x, h_k)_H, making the quotient ambiguous as printed. The subsequent equation (6) shows the intended formula is correct, so this is a typographical issue rather than a mathematical one.
Circularity Check
No circularity: the results are established by self-contained spectral and variational arguments, with the conclusions derived from explicit hypotheses rather than assumed or fitted.
full rationale
The paper's derivations are self-contained and do not reduce to their own inputs. Lemma 1 proves the two-term asymptotics for the generalized eigenproblem (2) by introducing the equivalent scalar product, using the spectral measure estimate (4), and bounding the residual B~h_n - lambda_n h_n via the assumption ||B h_n|| <= c n^{-(1+delta)}. No parameter is fitted and no target quantity is used as an input. Theorem 1 then replaces this assumption by the matrix-element decay condition |(B h_n, h_m)_H| <= c (mn)^{-(1+delta)/2} and proceeds by variational/min-max estimates on finite- or infinite-dimensional subspaces. The conclusion (3) is obtained by comparing Rayleigh quotients and bounding B~A or A~B; the theorem is a genuine sufficient condition, not a restatement of the hypothesis. The only potentially problematic step, the reduction to sign-definite perturbations via B = B_+ + B_-, is a mathematical correctness concern about whether the matrix-element bound is inherited by B_+ and B_-, but it is not a circularity: the target result is not assumed, and the argument does not depend on a self-citation or on a fitted value disguised as a prediction. Reference [3], the author's own preprint, is mentioned only as an application of the results and plays no role in the proofs. There is no self-definitional step, no fitted input renamed as a prediction, no load-bearing self-citation, no imported uniqueness theorem, and no known result merely renamed. The derivation chain is therefore independent of the conclusion, and the appropriate circularity score is 0.
Assumptions & free parameters
assumptions (3)
- standard math Spectral theorem for compact self-adjoint operators provides the spectral measure and eigenfunction basis.
- standard math Min-max principle for self-adjoint operators.
- standard math Eigenvalues of a self-adjoint operator respond continuously to a continuous change in the operator.
Cite this review
Pith. "Pith review of Some lemmata on the perturbation of the spectrum." pith.science (2026). https://pith.science/paper/M7S7RDTD
@misc{pith2026190809365,
author = {Pith},
title = {Pith review of: Some lemmata on the perturbation of the spectrum},
year = {2026},
howpublished = {\url{https://pith.science/paper/M7S7RDTD}},
note = {Machine review of arXiv:1908.09365}
}
read the original abstract
We give some sufficient conditions for preserving of the second term in the spectral asymptotics of a compact operator under the perturbation of the metrics in the Hilbert space.
Forward citations
Cited by 1 Pith paper
-
Spectral asymptotics for a class of integro-differential equations arising in the theory of fractional Gaussian processes
For fractional Gaussian covariance operators satisfying (K_alpha psi)(x) = -lambda psi''(x), the eigenvalues obey lambda_n = C (pi n - c_alpha - kappa pi/(3-alpha) + O(n^{-1}))^{alpha-3}, and the same formula persists...
Reference graph
Works this paper leans on
-
[1]
Quantitative analysis in Sobolev imbedding theorems and applications to spectral theory
Birman, M.S., and Solomyak, M.Z. Quantitative analysis in Sobolev imbedding theorems and applications to spectral theory. In: Proc eed. of X Summer Mathematical School. Yu.A. Mitropol’skiy and A.F. Shestopal (Eds), 1974, 5–189 (Russian); English transl. in: AMS T rans- lations, Series 2, 114. AMS, Providence, R.I. 1980
work page 1974
-
[2]
Spectral theory of self-adjoint operators in Hilbert space, 2nd ed., revised and extended
Birman, M.S., and Solomyak, M.Z. Spectral theory of self-adjoint operators in Hilbert space, 2nd ed., revised and extended. Lan’, St.Petersburg, 2010 [in Russian]; English transl. of the 1st ed.: Math - ematics and Its Applications. Soviet Series. 5, Kluwer, Dordrecht etc. 1987
work page 2010
-
[3]
Nazarov, A.I. Spectral asymptotics for a class of integro-diffe rential equations arising in the theory of fractional Gaussian processes. Preprint (2019), 31 pp. 6
work page 2019
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.