REVIEW 3 major objections 6 minor 25 references
Almost Sure Uniform Convergence Of Random Hermite Series
T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read The paper argues that one explicit tail condition governs almost sure uniform convergence of random Hermite series on R^d and on the sphere.
desk verdict Sharp spectral estimates, but Theorem A is not established: the condensation step is false and the necessary direction yields a strictly weaker entropy condition. 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 load-bearing object is the spectral function $e_{d,n}(x,y)=\sum_{k=1}^{\dim E_n}\phi_{n,k}(x)\phi_{n,k}(y)$ of the eigenspace $E_n$ of $-\Delta+|x|^2$. On the sphere, the paper proves uniform asymptotic estimates (Proposition 1.4): $e_{d,n}(x,y)$ is asymptotic to $n^{d/2-1}(2\pi)^{-d/2}\tilde J_{d/2-1}(\sqrt{2n}\,|x-y|)$ plus controlled remainders, with $\tilde J_{d/2-1}(t)=(2/t)^{d/2-1}J_{d/2-1}(t)$ the normalized Bessel function. The diagonal difference satisfies $e_{d,n}(x,x)-e_{d,n}(x,y)\simeq n^{d/2}|x-y|^2$ for $|x-y|\lesssim n^{-1/2}$ and is bounded by a fraction of the diagonal for larger separations below the antipodal scale. These estimates are converted into the two-sided pseudo-distance estimate that lets the authors compare the Hermite Gaussian process with a stationary Gaussian process on the circle, whose sample boundedness is controlled by an entropy integral. The second important mechanism is the dyadic condensation of condition (6), which turns the tail sum into block sums over $2^{2\ell}\le n<2^{2\ell+1}$.
What would settle it
Compute condition (6) for the coefficient weight $\|f_n\|^2/n^{d/2}=n^{-1}(\log n)^{-3}$: the tail at $N$ is of order $(\log N)^{-2}$, so the series in (6) diverges, while the condensed sum $\sum_{\ell}\ell^{-1/2}\bigl(\sum_{n\ge 2^\ell}\|f_n\|^2/n^{d/2}\bigr)^{1/2}$ used in the proof converges; comparing this same weight through Proposition 3.6 would locate the failing equivalence in the necessary direction.
Extended reading notes
Core claim
Let $f=\sum_n f_n$ be a tempered distribution in the harmonic-oscillator Sobolev scale, and let $f^{G,\omega}$ be the Gaussian randomization (5) of its Hermite components. Theorem A claims that three statements are equivalent: condition (6), namely $\sum_{\ell\ge 2} \ell^{-1}\sqrt{\log\ell}\,\bigl(\sum_{n\ge\ell} \|f_n\|^2/n^{d/2}\bigr)^{1/2}<\infty$; almost sure convergence of $f^{G,\omega}$ in $L^\infty(\mathbb{R}^d)$; and almost sure convergence in $L^\infty(\mathbb{S}^{d-1})$. The implication (i)$\Rightarrow$(ii) is obtained by a dyadic mesh argument adapted from the classical random-trigonometric-series method, while (iii)$\Rightarrow$(i) is the hard direction. It passes through a two-sided estimate $\delta_n(x,y)\simeq n^{-d/4}\min(1,\sqrt{n}\,|x-y|)$ for the Gaussian pseudo-distance on the sphere, derived from new uniform asymptotics of the spectral function $e_{d,n}(x,y)$ on $\mathbb{S}^{d-1}\times\mathbb{S}^{d-1}$; those asymptotics express the spectral function, via the Mehler formula, as a Cauchy product of one-dimensional Hermite functions, whose leading term is a normalized Bessel function.
Load-bearing premise
The whole argument leans on the assumption that chopping condition (6) into dyadic blocks never changes whether it converges, and that the entropy-integral criterion in Proposition 3.6 is interchangeable with (6) even though its logarithmic factor appears in the denominator rather than the numerator.
Editorial extensions
If this is right
- For any $f$ in the Sobolev scale, almost sure uniform convergence on all of $\mathbb{R}^d$ is equivalent to almost sure uniform convergence on the sphere $\mathbb{S}^{d-1}$, despite the sphere having Lebesgue measure zero.
- The same condition (6) governs almost sure uniform convergence on every sphere $R\mathbb{S}^{d-1}$ of any radius $R>0$, so two disjoint spheres must either both work or both fail.
- Uniform convergence on the geodesic $\mathbb{S}^1\times\{0\}^{d-2}$ already forces condition (6), hence forces uniform convergence on the whole space.
- On the sufficient side, the proof controls the expected supremum of each dyadic block $2^{2\ell}\le n<2^{2\ell+1}$ by a constant times $2^{\ell/2}\bigl(\sum_{n} \|f_n\|^2/n^{d/2}\bigr)^{1/2}$, making the randomization almost surely uniformly convergent under (6).
- Together these give a checkable analytic criterion for almost sure boundedness and uniform convergence of the randomized Hermite expansion, completing the $p=\infty$ endpoint left open by the $L^p$ statement of Theorem 1.1.
Reading between the lines
- The dyadic condensation step in Section 3.1 is the delicate point: for coefficient weights with slowly varying logarithmic factors, the original tail sum and its condensed versions can disagree, so the claimed equivalence should be tested there before relying on it.
- If the condensation mismatch is real, the true threshold for almost sure uniform convergence would likely involve a different logarithmic weight than the $\sqrt{\log \ell}/\ell$ in (6); computing the entropy integral for borderline sequences such as $\|f_n\|^2/n^{d/2}=n^{-1}(\log n)^{-3}$ would show which direction fails.
- The spectral-function asymptotics suggest a transfer principle: for any rotationally symmetric confining potential whose eigenspace dimensions grow like $n^{d-1}$ and whose Mehler kernel has a Bessel-type leading term, a version of the whole-space versus sphere dichotomy should hold.
- A direct numerical check in dimension $d=2$, comparing condition (6) with the entropy integral on finite truncations, would track whether the dyadic equivalence is sound for slowly varying weights.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper aims to give a necessary and sufficient condition for the almost sure uniform convergence on R^d of Gaussian random series associated with the multidimensional harmonic oscillator -Δ+|x|^2, d≥2. The main result, Theorem A, asserts that for f in the Sobolev scale of the harmonic oscillator the finiteness of the Salem–Zygmund-type series (6), namely ∑_ℓ ℓ^{-1}√(log ℓ) (∑_{n≥ℓ} ‖f_n‖^2 n^{-d/2})^{1/2}, is equivalent to the a.s. convergence of the random series in L^∞(R^d) and also to a.s. convergence in L^∞(S^{d-1}). The proof is divided into a sufficiency part (Section 3.1) and a necessity part (Section 3.2). The necessity part relies on new two-sided asymptotics for the spectral function e_{d,n} on S^{d-1}×S^{d-1} (Proposition 1.4), which are derived from the Mehler formula and uniform Hermite asymptotics, and on a comparison argument with a stationary Gaussian process on S^1 (Proposition 3.5). The paper also contains details of the spectral estimates in Section 4 and appendices.
Significance. If Theorem A were correct, it would be a substantial advance: it would give the first necessary and sufficient probabilistic continuity criterion for random Hermite series on the whole space, with the striking consequence that uniform convergence on the sphere S^{d-1} (a zero Lebesgue measure set) is equivalent to uniform convergence on R^d. The spectral function estimates in Proposition 1.4 are technically interesting and appear to be new; the derivation via the Mehler formula and the reduction to one-dimensional Hermite functions is elegant. The paper also builds explicitly on prior work of the first author, and it cites the relevant probabilistic and spectral literature. However, the central equivalence (i)⇔(iii) is not established: two load-bearing steps in Sections 3.1 and 3.2 are incorrect, and the paper's own Proposition 3.6 yields a strictly weaker condition than (6). Consequently the claimed theorem is unsupported and, on the evidence of the paper's own computations, false as stated.
major comments (3)
- [Section 3.1] The two 'Cauchy condensation' reformulations of condition (6) are false. The text states that (6) is equivalent to ∑_ℓ ℓ^{-1/2} (∑_{n≥2^ℓ} ‖f_n‖^2 n^{-d/2})^{1/2} and to ∑_ℓ 2^{ℓ/2} (∑_{n≥2^{2ℓ}} ‖f_n‖^2 n^{-d/2})^{1/2}. This is not valid. For the admissible sequence a_n = ‖f_n‖^2 n^{-d/2} = n^{-1} (log n)^{-3}, the tail T_p = ∑_{n≥p} a_n satisfies √T_p ≈ 1/(√2 log p), so the original series (6) behaves like ∑_p p^{-1} (log p)^{-1/2}, which diverges. In contrast, the first dyadic reformulation behaves like ∑_ℓ ℓ^{-1/2} ℓ^{-1} = ∑_ℓ ℓ^{-3/2}, which converges. Thus (21) is not a consequence of (6), and the sufficiency proof of (i)⇒(ii) actually requires a strictly stronger condition than (6).
- [Section 3.2, Proposition 3.6] The final step of the necessary condition misidentifies the entropy sum proved in Proposition 3.6 with condition (6). Proposition 3.6 establishes that ∫_0^1 Υ_θ(t) t^{-1} (-log t)^{-1/2} dt is equivalent to ∑_{p≥1} p^{-1} (log(p+1))^{-1/2} √(∑_{n≥p} c_n). With c_n = ‖f_n‖^2 n^{-d/2}, the latter series has terms 1/(p √(log p)) · √T_p, whereas condition (6) has terms √(log p)/p · √T_p. These are not equivalent: for the same sequence a_n = n^{-1} (log n)^{-3}, the entropy sum converges (its terms are ∼ p^{-1} (log p)^{-3/2}) while (6) diverges (terms ∼ p^{-1} (log p)^{-1/2}). Hence the argument in Section 3.2 proves at most (iii) ⇒ entropy condition, not (iii) ⇒ (6). Since the sufficiency direction likewise requires a condition stronger than (6), neither direction of Theorem A is established; moreover, the discrepancy in the necessity direction shows that the claimed equivalence with (6) cannot hold in general.
- [Section 3.2, around (27)] The paper asserts without proof that 'the expected conclusion, namely assertion (i) of Theorem A, is now a consequence of Proposition 3.6'. This sentence is the load-bearing step connecting the Gaussian comparison to Theorem A(i), but it is exactly where the mismatch described above occurs. The reader is left with no argument bridging the 1/(p√(log p)) factor in (27) to the √(log p)/p factor in (6). This is not a minor gap: it is the point on which the main theorem's equivalence depends.
minor comments (6)
- [Abstract and Introduction] The abstract states that the same condition gives convergence on S^{d-1} 'despite S^{d-1} is a zero Lebesgue measure of R^d'; the grammar is awkward, and the phrasing should be 'despite S^{d-1} having zero Lebesgue measure in R^d'.
- [Remark 1.3] Remark 1.3 mentions R_1S^{d-1} and R_2S^{d-2} but the second sphere should be R_2S^{d-1}; if this typo is corrected, the claim about disjoint spheres is clear.
- [Section 2, display (19)-(20)] The notation in (19)-(20) is confusing because e_{2,n} is defined on R^2, while the text sometimes writes S^{d-1}; the restriction to |x|=|y|=1 should be stated explicitly in the display or immediately before it.
- [Section 3.1] The proof of (i)⇒(ii) uses the Cauchy condensation reformulation without stating the monotonicity condition required by the condensation test; since the tail sums are decreasing, the correct comparison for monotone sequences would be ∑ 2^ℓ φ(2^ℓ) versus ∑ φ(ℓ), but the log factors in (6) are not handled correctly, as explained in the major comments.
- [Appendix B, proof of Lemma 4.4] The notation in (63) is slightly unconventional: β = ⌈β⌉ - ε with ε∈[0,1) is fine, but the subsequent estimates would benefit from explicitly checking the case β = -1/2, since the paper uses Lemma 4.4 with β = (d-3)/2 and d=2 gives β=-1/2; the proof appears to handle it, but the boundary cases should be stated.
- [References] The reference list omits page numbers for some entries (e.g., [PZ] gives multiple page ranges without a journal volume); this is a minor bibliographic issue.
Circularity Check
No circularity: the derivation rests on the Mehler formula and classical Gaussian-process comparisons; self-citations are auxiliary, not definitional.
full rationale
No circular reduction was found. The main derivation chain is: (i) implies (ii) via dyadic blocks and the mesh lemma; (ii) implies (iii) is immediate; (iii) implies (i) uses the new two-sided spectral estimates (Propositions 1.4 and 3.2), a Slepian comparison to a stationary Gaussian process, and the entropy computation of Proposition 3.6. Each of these steps is proved from the Mehler formula, uniform Hermite asymptotics, and classical Dudley-Fernique theory, rather than from the condition being proved. The self-citations to [Ime18, Ime19, Ime22] are used for auxiliary tools such as the mesh strategy, scalar Gaussian convergence, and the entropy equivalence for stationary processes; those prior results are not definitionally equivalent to Theorem A and do not make the conclusion an input. There are separate mathematical correctness concerns that are not circularity: the Section 3.1 'Cauchy condensation' reformulations of (6) are not valid equivalences for general admissible tail sequences, and the final step of Section 3.2 identifies the entropy sum (27), which has factor 1/(p sqrt(log p)), with condition (6), which has factor sqrt(log p)/p. These are substantive gaps, but they are errors of equivalence rather than reductions-by-construction, so they do not raise the circularity score.
Assumptions & free parameters
assumptions (6)
- standard math Mehler formula for the harmonic oscillator
- standard math Uniform asymptotics of Hermite functions (Muckenhoupt)
- standard math Dudley-Fernique entropy criterion and Slepian comparison theorems
- domain assumption Finite subset concentration assumption for Hermite functions
- standard math Euler-Maclaurin summation formula
- domain assumption Pisier's comparison of Gaussian vectors with uniform sphere vectors
Cite this review
Pith. "Pith review of Almost Sure Uniform Convergence Of Random Hermite Series." pith.science (2026). https://pith.science/paper/JWLQLCNI
@misc{pith2026250603858,
author = {Pith},
title = {Pith review of: Almost Sure Uniform Convergence Of Random Hermite Series},
year = {2026},
howpublished = {\url{https://pith.science/paper/JWLQLCNI}},
note = {Machine review of arXiv:2506.03858}
}
abstract
We continue the analysis of random series associated to the multidimensional harmonic oscillator $-\Delta + |x|^2$ on $\mathbb{R}^d$ with d \geq 2$$. More precisely we obtain a necessary and sufficient condition to get the almost sure uniform convergence on the whole space $\mathbb{R}^d$ . It turns out that the same condition gives the almost sure uniform convergence on the sphere $\mathbb{S}^{d-1}$ (despite $\mathbb{S}^{d-1}$ is a zero Lebesgue measure of $\mathbb{R}^d$). From a probabilistic point of view, our proof adapts a strategy used by the first author for boundaryless Riemannian compact manifolds. However, our proof requires sharp off-diagonal estimates of the spectral function of $-\Delta + |x|^2$ . Such estimates are obtained using elementary tools.
Reference graph
Works this paper leans on
-
[1]
A. Ayache and N. Tzvetkov. L^p properties for G aussian random series. Trans. Amer. Math. Soc. , 360(8):4425--4439, 2008
work page 2008
-
[2]
J-M. Aza \" s and M. Wschebor. Level sets and extrema of random processes and fields . John Wiley & Sons, 2009
work page 2009
-
[3]
P Brun, R. Imekraz, and G. Poly. Random zonal eigenfunctions and H \" o lder version of the P aley- Z ygmund theorem on compact manifolds. Bulletin SMF , 152(3):443--517, 2024
work page 2024
-
[4]
N. Burq and G. Lebeau. Injections de S obolev probabilistes et applications. Ann. Sci. \'E c. Norm. Sup \'e r. , 46(6):917--962, 2013
work page 2013
-
[5]
Y. Canzani and B. Hanin. Scaling limit for the kernel of the spectral projector and remainder estimates in the pointwise W eyl law. Anal. PDE , 8(7):1707--1731, 2015
work page 2015
-
[6]
R.M. Dudley. The sizes of compact subsets of H ilbert space and continuity of G aussian processes. J. Funct. Anal. , 1(3):290--330, 1967
work page 1967
-
[7]
A. Estrade and J. Fournier. Anisotropic G aussian wave models. ALEA: Latin American Journal of Probability and Mathematical Statistics , 17:329--353, 2020
work page 2020
-
[8]
\'E cole d' \'e t \'e de probabilit \'e s de S aint- F lour IV
X Fernique. \'E cole d' \'e t \'e de probabilit \'e s de S aint- F lour IV . Lecture Notes in Math., Springer-Verlag , 480:32--96, 1974
work page 1974
Show all 25 references
-
[9]
Hanin, S
B. Hanin, S. Zelditch, and P. Zhou. Nodal sets of random eigenfunctions for the isotropic harmonic oscillator. Int. Math. Res. Not. IMRN , 2015(13):4813--4839, 2015
2015
-
[10]
R. Imekraz. Concentration et randomisation universelle de sous-espaces propres. Anal. PDE , 11(2):263--350, 2018
2018
-
[11]
R. Imekraz. Multidimensional P aley- Z ygmund theorems and sharp L ^p estimates for some elliptic operators. Ann. Inst. Fourier , 69(6):2723--2809, 2019
2019
-
[12]
R. Imekraz. A necessary and sufficient condition for probabilistic continuity on a boundaryless compact Riemannian manifold. Journal de l \'Ecole polytechnique Math\'ematiques , 9:747--805, 2022
2022
-
[13]
Imekraz, D
R. Imekraz, D. Robert, and L. Thomann. On random Hermite series . Trans AMS , 368:2763--2792, 2016
2016
-
[14]
Koch and D
H. Koch and D. Tataru. L p eigenfunction bounds for the H ermite operator. Duke Math. J. , 128(2):369--392, 2005
2005
-
[15]
Li and H
D. Li and H. Queff \'e lec. Introduction to Banach Spaces: Analysis and Probability , volume 1. Cambridge University Press, 2018
2018
-
[16]
Lindenstrauss and L
J. Lindenstrauss and L. Tzafriri. Classical Banach Spaces: Vol.: 2.: Function Spaces . Springer-Verlag, 1979
1979
-
[17]
Ledoux and M
M. Ledoux and M. Talagrand. Probability in Banach Spaces: isoperimetry and processes , volume 23. Springer, 1991
1991
-
[18]
Marcus and G
M.B. Marcus and G. Pisier. Random Fourier Series with Applications to Harmonic Analysis . Annals of Math Studies, Princeton University Press, Princeton, NJ , 101, 1981
1981
-
[19]
Muckenhoupt
B. Muckenhoupt. Mean convergence of Hermite and Laguerre series. I . Trans. Amer. Math. Soc. , pages 419--431, 1970
1970
-
[20]
G. Pisier. The volume of convex bodies and Banach space geometry , volume 94. Cambridge University Press, 1989
1989
-
[21]
Poiret, D
A. Poiret, D. Robert, and L. Thomann. Random weighted S obolev inequalities on R ^d and applications to H ermite functions. Ann. Henri Poincar \'e , A, Math.Phys. , 16(2):651--689, 2015
2015
-
[22]
Paley and A
R.E.A.C. Paley and A. Zygmund. On some series of functions, (1) (2) (3). Proc. Camb. Phil. Soc. , 26 (1930) 337--357, 458--474, 28 (1932) 190--205
1930
-
[23]
Salem and A
R. Salem and A. Zygmund. Some properties of trigonometric series whose terms have random signs. Acta Math. , 91(1):245--301, 1954
1954
-
[24]
G. Szego. Orthogonal polynomials . American Mathematical Society, Providence, R.I., fourth edition, 1975. American Mathematical Society, Colloquium Publications, Vol. XXIII
1975
-
[25]
Tzvetkov
N. Tzvetkov. Riemannian analogue of a P aley- Z ygmund theorem. S \'e minaire EDP X, 2008-2009. Expos \'e no. XV
2008
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.