REVIEW 2 major objections 2 minor 5 references
On local non-tangential growth of the resolvent of a banded Toeplitz operator
T0 review · 2 major / 2 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read A banded Toeplitz operator's resolvent grows at most like the inverse distance to the spectrum at every non-branch boundary point, within non-tangential approach sectors.
desk verdict A solid local refinement of global resolvent growth for banded Toeplitz operators; the main theorem is correct, the remaining issues are expositional. 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 runs on the algebraic equation $P(z,w)=z^m(b(z)-w)=0$ and its divisor $Z(w)$, split into roots inside, outside, and on the unit circle. The central object is the finite critical set $K=\{b(\lambda): b'(\lambda)=0\}$. Outside $K$, the roots $z_j(w)$ are analytic in $w$, with the expansion $z_j(w)=z_j(w_0)+\frac{w-w_0}{b'(z_j(w_0))}+O((w-w_0)^2)$, the linear separation estimate $C_{11}|w-w_0|\le |z_j(w)-z_j(w_0)|\le C_{12}|w-w_0|$, and the Stolz-angle condition $|1-|z_j(w)|| \ge C_{14}|z_j(w)-z_j(w_0)|$. The non-tangential domain $\Omega'(w_0,\varepsilon)$ is the intersection of the sectors $G_j(w_0)$ over all roots near the unit circle; the Wiener–Hopf resolvent formula and a Hardy-space evaluation bound convert sector control into the inverse-linear estimate.
What would settle it
Take a non-exceptional boundary point such as $w_0=0$ for $b(z)=z^{-1}+z^2$, and compute the ratio $\|(T_b-w)^{-1}\|\operatorname{dist}(w,\sigma(T_b))$ along rays $w=\varepsilon e^{i\varphi}$ with $\varphi$ inside one of the sectors $\Gamma^\delta$. If the ratio is unbounded as $\varepsilon\to0$, Theorem 2.1 would be false; if it stays bounded, the sector restriction is doing real work and the bound is sharp in that direction.
Extended reading notes
Core claim
The central claim is Theorem 2.1: for a Hardy–Toeplitz operator $T_b$ whose symbol is a Laurent polynomial $b(z)=b_{-m}z^{-m}+\cdots+b_k z^k$, fix a boundary point $w_0\in\partial\sigma(T_b)$ that is not in the finite set $K=\{b(\lambda): b'(\lambda)=0\}$. Then for all small $\varepsilon>0$ there is a non-tangential domain $\Omega'(w_0,\varepsilon)$ at $w_0$ in which $\|(T_b-w)^{-1}\|\le C(b,w_0,\varepsilon)/\operatorname{dist}(w,\sigma(T_b))$ for every $w$ in that domain. The proof splits the roots of $z^m(b(z)-w)$ into those near the unit circle and those away from it, uses analyticity of the near-circle roots when $w_0\notin K$ to control their motion by a Stolz-angle condition, and feeds that control into the Wiener–Hopf resolvent formula. The paper also proves a complementary local result (Theorem 1.1): under weaker geometric conditions—one boundary root, or all roots inside, or all roots outside—the inverse-linear bound holds in a full disk neighborhood, not just in sectors.
Load-bearing premise
The proof requires that every solution of $b(z)=w_0$ be a simple root, so the roots move analytically and with a linear rate as $w$ varies near $w_0$; at a multiple-root point the sector geometry collapses and the theorem is silent.
Editorial extensions
If this is right
- At every boundary point outside $K$, the resolvent of a banded Toeplitz operator satisfies $\|(T_b-w)^{-1}\|\le C/\operatorname{dist}(w,\sigma(T_b))$ inside the non-tangential domain $\Omega'(w_0,\varepsilon)$, so the spectrum cannot be approached by resolvent blow-up faster than inverse-linearly in those sectors.
- Under the geometric conditions of Theorem 1.1—one boundary root, or all roots inside, or all roots outside—the inverse-linear bound holds in a full disk neighborhood of $w_0$, not only in sectors.
- Because $K$ is finite, all but finitely many boundary points admit the weak linear resolvent growth; the exceptional set consists exactly of the values $b(\lambda)$ where $b'(\lambda)=0$.
- In the three-petal example $b(z)=z^{-1}+z^2$, Theorem 2.1 yields the inverse-linear bound in each strictly interior angular sector $\Gamma^\delta$ near the origin, even though the origin is a self-intersection point of the spectral curve.
- Roots away from the unit circle contribute only bounded factors in the proof, so the inverse-linear rate is controlled entirely by the roots that stay close to the unit circle.
Reading between the lines
- The paper leaves open whether points in $K$, where roots branch, might also admit inverse-linear growth in some approach sectors; answering that would require a Puiseux-series treatment rather than the analytic-root expansion used here.
- The sector restriction is probably essential: along directions where a near-circle root moves tangentially to the unit circle the condition $|1-|z_j(w)||\ge C|z_j(w)-z_j(w_0)|$ fails, so testing whether the bound survives on tangent directions would illuminate the sharpness of the non-tangential geometry.
- Remark 2.2 suggests the exceptional set can be shrunk whenever the multiple roots lie outside the unit disk, so the practical obstruction to linear resolvent growth may be smaller than $K$ for many symbols.
- These local bounds feed directly into estimates for the pseudospectrum of infinite banded Toeplitz matrices: near non-branch boundary points the $\varepsilon$-pseudospectrum should extend into the resolvent set only to order $\varepsilon$, a prediction one could test numerically.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies local resolvent growth for banded Toeplitz operators T_b with Laurent polynomial symbol b. It introduces a local version of linear resolvent growth, LLRG(w0), and proves in Theorem 1.1 that local regularity of the divisor at a boundary point w0 implies LLRG(w0). The main result, Theorem 2.1, states that if w0 is not in the finite set K of branch points of the algebraic relation b(z)=w, then the resolvent admits an inverse-linear upper bound on a non-tangential domain Ω'(w0,ε) near w0. The proof uses Wiener-Hopf factorization, partial fraction expansions of the resolvent kernel, point-evaluation estimates in H^2, and an angular separation of the simple roots near the unit circle.
Significance. If the proof is repaired as indicated below, the paper gives a useful local refinement of the global resolvent growth bounds in [3]. The finite exceptional set K is explicit, and the theorem is honest about being silent at branch points. The argument is constructive and does not rely on numerical fitting; the constants depend only on the symbol and the fixed neighborhood. The main contribution is a quantitative mechanism, via simple-root expansions and Stolz-angle estimates, for obtaining inverse-linear growth at non-critical boundary points.
major comments (2)
- [§2, Eq. (2.9)–(2.11)] Equation (2.9) is incorrect as printed. From (2.7), z_j(w)=z_j(w0)+(w-w0)/b'(z_j(w0))+O((w-w0)^2), so the first-order term in |z_j(w)|^2 is 2 Re( conjugate(z_j(w0))(w-w0)/b'(z_j(w0)) ), not 2 Re( z_j(w0)(w-w0)/b'(z_j(w0)) ). Since (2.10) defines G_j through this coefficient and (2.11) is derived from (2.9), the proof as written does not establish the Stolz-angle estimate for the stated domains. The error is consequential: for b(z)=z^3 and z0=e^{iπ/4}, the radial direction toward w0=b(z0) gives zero first-order change under the printed formula, whereas the true first-order change is nonzero. Section 3 actually computes with the conjugated coefficient, so the intended fix is clear: replace z_j(w0) by its conjugate in (2.9), (2.10), and the corresponding displayed computation in Section 3, and re-check the subsequent estimates with this corrected definition.
- [§2, Eq. (2.12)] The definition of Ω'(w0,ε) is not well-posed as written. The intersection is over z_j in \hat Z(w), but \hat Z(w) depends on w, so Ω' cannot be understood as a fixed subset of B(w0,ε) before w is chosen. It should be defined using the fixed finite set \hat Z(w0)=Z_un(w0), with the natural bijection to the roots near the unit circle for w near w0. In addition, the constant C13 appearing in (2.10) is never quantified in the statement; the theorem should specify, for example, that C13 is any fixed sufficiently small positive number and that the constant C(b,w0,ε) in (2.13) may depend on this choice.
minor comments (2)
- [§1, proof of Theorem 1.1] The displayed inequality ||b_ext^{-1}||_∞ ≤ C3 / ||b-w||_∞ is false as written: the denominator should be the infimum of |b(t)-w| over t∈T, i.e., dist(w,b(T)), because the proof establishes a lower bound for |b_ext(t,w)|. The subsequent use of dist(w,σ(T_b)) shows the intended statement, and the line should be corrected accordingly.
- [Abstract and Section 3] There are several small typos: the abstract's BpσpTbqq should presumably be ∂σ(Tb); Section 3 refers to 'Theorem 2.13' instead of Theorem 2.1; and the figure reference 'Figure ??' is unresolved. Please fix these before final submission.
Circularity Check
No circularity found: Theorem 2.1 is proved directly from root expansions and standard resolvent identities, with no fitted parameter or conclusion built into an input.
full rationale
The proof of Theorem 2.1 does not define any input in terms of the target resolvent bound. The main ingredients are: the Wiener-Hopf factorization (1.1)/(2.3), the resolvent formula (2.14), the first-order root expansion (2.7)-(2.9) valid because w0 is assumed outside the branch set K, the sector condition (2.10) that yields the Stolz-angle separation (2.11), the standard point-evaluation estimate (2.17), and the partial-fraction estimate for ||(b-w)^(-1)||_2. Each of these is stated independently of the conclusion (2.13). The citations to [3] provide the divisor-counting lemma |Zin(w)|=m for w in the resolvent set and the global regularity proof pattern; this lemma is parameter-free, does not assume LLRG, and is not the source of the 1/dist estimate. No parameter is fitted to a subset of data and then renamed a prediction, and no uniqueness theorem is imported from the authors. The only questionable display is a notational issue in Theorem 1.1's proof involving ||b(t)-w||_∞ versus its minimum, which affects exposition but not circularity. The exclusion w0 not in K is explicit and the result is silent at branch points by design, so there is no hidden assumption. Overall the derivation is self-contained in the relevant sense.
Assumptions & free parameters
free parameters (1)
- C13 =
unspecified positive constant
assumptions (5)
- standard math Wiener-Hopf factorization b(z)-w = b_ext(z,w) b_in(z,w) for w in the resolvent set
- domain assumption For w in rho(Tb), |Zin(w)| = m and |Zext(w)| = k (count lemma)
- standard math Krein's bound ||(Tb-w)^{-1}|| <= ||b_in^{-1}||_∞ ||b_ext^{-1}||_∞
- standard math H^2 point evaluation bound |h(λ)| <= ||h||_2 / sqrt(1-|λ|^2)
- standard math Formula (2.14) for the resolvent of a Toeplitz operator
Cite this review
Pith. "Pith review of On local non-tangential growth of the resolvent of a banded Toeplitz operator." pith.science (2026). https://pith.science/paper/OIVZBZX3
@misc{pith2026250600584,
author = {Pith},
title = {Pith review of: On local non-tangential growth of the resolvent of a banded Toeplitz operator},
year = {2026},
howpublished = {\url{https://pith.science/paper/OIVZBZX3}},
note = {Machine review of arXiv:2506.00584}
}
abstract
We study the growth of the resolvent of a Hardy--Toeplitz operator $T_b$ with a Laurent polynomial symbol (\emph{i.e., } the matrix $T_b$ is banded), at the neighborhood of a point $w_0\in\partial(\sigma(T_b))$ on the boundary of its spectrum. We show that such growth is inverse linear in some non-tangential domains at the vertex $w_0$, provided that $w_0$ does not belong to a certain finite set on the complex plane.
Figures
Reference graph
Works this paper leans on
-
[3]
L. Golinskii, S. Kupin and A. Vishnyakova, On the growth of resolvent of Toeplitz operators, MPAG, 2024
work page 2024
-
[1]
A. B¨ ottcher and S. Grudsky,Spectral Properties of Banded Toeplitz Matrices, SIAM, Philadelphia 2005
work page 2005
-
[2]
A. B¨ ottcher and B. Silbermann, Introduction to Large Truncated Toeplitz Matrices , Springer Science NY, 1999
work page 1999
-
[4]
Halmos, A Hilbert Space Problem Book , Springer–Verlag, Berlin, 1982
P. Halmos, A Hilbert Space Problem Book , Springer–Verlag, Berlin, 1982
work page 1982
-
[5]
Nikolski, Toeplitz Matrices and Operators, Cambridge Studies in Advanced Math- ematics, v.182, 2020
N. Nikolski, Toeplitz Matrices and Operators, Cambridge Studies in Advanced Math- ematics, v.182, 2020. Mathematics Division, Institute for Low Temperature Physics and Engi- neering, 47 Science’s ave., Kharkiv 61103, Ukraine Email address : golinskii@ilt.kharkov.ua, leonid.golinskii@gmail.com Institut de Math ´ematiques de Bordeaux UMR5251, CNRS, Universi...
work page 2020
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Sign in to comment.