REVIEW 2 major objections 1 minor 12 references
Positive-definiteness in separable priors: effects on prior interpretability and inference
T0 review · 2 major / 1 minor · reviewed 2026-06-30 · grok-4.3
Pith's one-line read Truncation to enforce positive definiteness in separable priors can systematically favor sparser matrices unless off-diagonal variances are adjusted with dimension.
desk verdict The truncation bias claim rests on comparing to an untruncated independent-entry distribution that isn't supported on positive definite matrices, so the sparsity effect may be an artifact rather than a real distortion. 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 truncation operation applied to an independent-entry distribution to restrict support to the set of positive-definite matrices.
What would settle it
Compute or estimate the prior probability of matrices with a given number of zero off-diagonal entries under both the truncated and untruncated versions for fixed variance parameters as dimension increases; persistent deviation from equality for sparser cases would support the claim.
Extended reading notes
Core claim
The paper claims that for priors on symmetric positive-definite matrices that start with independent entries and then truncate to the positive-definite cone, the resulting distribution differs from the untruncated one in ways that affect interpretability and shrinkage. Specifically, unless the variance of off-diagonal entries is set to decrease appropriately with matrix dimension, the truncated prior assigns higher probability to sparser matrices, and this bias carries over to the posterior.
Load-bearing premise
The untruncated independent-entry distribution is the desired target that truncation should preserve as closely as possible for interpretability and shrinkage characterization.
Editorial extensions
If this is right
- Setting the variance of off-diagonal entries to scale with dimension mitigates the truncation effect for both dense and sparse matrices.
- In sparse inference, careful parameter choice prevents the truncated prior and posterior from assigning systematically higher mass to sparser structures.
- Posterior inference can be affected in unanticipated ways if truncation effects on mass assignment are ignored.
- The shrinkage properties of the prior become harder to characterise without matching the untruncated margins.
Reading between the lines
- Users of these priors in high-dimensional settings may need explicit scaling rules in software defaults to avoid unintended sparsity bias.
- Similar truncation adjustments could be required for other constrained matrix distributions such as correlation matrices.
- Direct Monte Carlo comparison of truncated and untruncated samples in moderate dimensions would quantify the mass shift on sparsity levels.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript examines the effects of adding a truncation to independent-entry priors on symmetric matrices to enforce positive-definiteness. It claims that, unless prior parameters (especially off-diagonal variances) are chosen carefully, the truncated prior and resulting posterior assign systematically higher mass to sparser structures than the untruncated counterpart, for both dense and sparse settings; the paper provides guidance on parameter settings to mitigate this discrepancy as matrix dimension grows.
Significance. If the claimed truncation effects and mitigation rules hold under rigorous derivation, the work would be significant for Bayesian covariance modeling and sparse precision-matrix inference, as it directly addresses interpretability and unintended shrinkage in a widely used prior class.
major comments (2)
- [Abstract] Abstract and introduction: the central claim rests on comparing the truncated prior to the untruncated independent-entry distribution as the reference whose sparsity properties should be preserved. However, the untruncated distribution is supported on all symmetric matrices and places positive mass outside the positive-definite cone; any observed difference in mass on sparse structures could therefore be an artifact of the projection onto the cone rather than an intrinsic effect of truncation. This comparison requires explicit justification or re-framing as a diagnostic rather than a normative target.
- [Abstract] The mitigation strategy of setting off-diagonal variances to control the effect as dimension grows inherits the same reference-distribution issue; without a clear statement of what properties of the untruncated margins are desirable on the PD cone, it is unclear whether the recommended parameter scaling achieves the intended preservation of interpretability.
minor comments (1)
- [Abstract] The abstract states that the analysis covers both dense and sparse matrices, but does not indicate whether the mitigation rules differ between the two regimes or whether the same variance scaling applies.
Simulated Author's Rebuttal
We thank the referee for their constructive comments. We address the major comments point by point below.
read point-by-point responses
-
Referee: [Abstract] Abstract and introduction: the central claim rests on comparing the truncated prior to the untruncated independent-entry distribution as the reference whose sparsity properties should be preserved. However, the untruncated distribution is supported on all symmetric matrices and places positive mass outside the positive-definite cone; any observed difference in mass on sparse structures could therefore be an artifact of the projection onto the cone rather than an intrinsic effect of truncation. This comparison requires explicit justification or re-framing as a diagnostic rather than a normative target.
Authors: The untruncated independent-entry prior serves as the natural baseline for separable priors, with truncation applied subsequently to enforce positive-definiteness. Our analysis demonstrates the distortion introduced by this truncation. We have revised the manuscript to explicitly frame the comparison as a diagnostic for assessing truncation effects on interpretability and sparsity, rather than positioning the untruncated distribution as a normative target on the positive-definite cone. This clarification addresses the concern directly. revision: yes
-
Referee: [Abstract] The mitigation strategy of setting off-diagonal variances to control the effect as dimension grows inherits the same reference-distribution issue; without a clear statement of what properties of the untruncated margins are desirable on the PD cone, it is unclear whether the recommended parameter scaling achieves the intended preservation of interpretability.
Authors: We agree that the target properties on the PD cone merit explicit statement. The recommended scaling of off-diagonal variances is designed to ensure that the truncated prior's marginal distributions and sparsity characteristics more closely match those of the untruncated prior as dimension increases. We have added clarification in the revised manuscript specifying the desirable properties (matching marginal variances and reduced bias in sparsity) and demonstrating how the scaling achieves this approximation within the positive-definite cone. revision: yes
Circularity Check
No significant circularity; analysis is self-contained
full rationale
The paper directly analyzes the truncation effect on independent-entry priors for positive-definite matrices by comparing truncated and untruncated versions, deriving guidance on parameter settings (e.g., off-diagonal variances) to mitigate sparsity bias as dimension grows. No equations or claims reduce by construction to fitted inputs renamed as predictions, self-definitional loops, or load-bearing self-citations. The untruncated distribution is treated as an explicit reference distribution whose properties are examined post-truncation; this is a modeling choice with independent mathematical content rather than a circular reduction. The derivation chain relies on standard truncation arguments and asymptotic analysis without importing uniqueness theorems or ansatzes from prior self-work.
Assumptions & free parameters
Cite this review
Pith. "Pith review of Positive-definiteness in separable priors: effects on prior interpretability and inference." pith.science (2026). https://pith.science/paper/EUXDLDC2
@misc{pith2026260522640,
author = {Pith},
title = {Pith review of: Positive-definiteness in separable priors: effects on prior interpretability and inference},
year = {2026},
howpublished = {\url{https://pith.science/paper/EUXDLDC2}},
note = {Machine review of arXiv:2605.22640}
}
read the original abstract
A popular class of priors for symmetric positive-definite matrices assumes independent entries and adds a truncation to ensure positive-definiteness. While conceptually simple and often computationally convenient, unless done carefully this truncation can have unintended effects. If the truncated prior or its margins are significantly different from their untruncated counterpart, then its interpretability may suffer, its shrinkage properties become harder to characterise, and posterior inference may be affected in unanticipated ways. We investigate the effect of the truncation both for dense and sparse matrices, and show how to set prior parameters such as the variance of off-diagonal entries such that said effect is mitigated as the matrix dimension grows. We pay particular attention to sparse inference where, unless prior parameters are set carefully, the truncated prior and hence its corresponding posterior assign systematically higher mass to sparser structures than the untruncated prior.
Figures
Reference graph
Works this paper leans on
-
[1]
doi: 10.1016/j.jmva.2015.01.015
ISSN 10957243. doi: 10.1016/j.jmva.2015.01.015. URL http://dx.doi.org/10.1016/j.jmva.2015.01.015. S. Boucheron, G. Lugosi, and P. Massart.Concentration inequalities: A nonasymptotic theory of indepen- dence. Oxford university press, Oxford,
-
[2]
doi: 10.1093/biostatistics/kxm045
ISSN 14654644. doi: 10.1093/biostatistics/kxm045. Lingrui Gan, Naveen N Narisetty, and Feng Liang. Bayesian Regularization for Graphical Models With Unequal Shrinkage.Journal of the American Statistical Association, 114(527):1218–1231,
-
[3]
Fengshi Niu, Harsha Nori, Brian Quistorff, Rich Caruana, Donald Ngwe, and Aadharsh Kannan
ISSN 1537274X. doi: 10.1080/01621459.2018.1482755. Jack Jewson, Li Li, Laura Battaglia, Stephen Hansen, David Rossell, and Piotr Zwiernik. Graphical model inference with external network data.Biometrics, 80(4):ujae151,
-
[4]
Locally associated graphical models and mixed convex exponential families.arXiv, 2008.04688:1–34,
18 REFERENCES REFERENCES Steffen Lauritzen and Piotr Zwiernik. Locally associated graphical models and mixed convex exponential families.arXiv, 2008.04688:1–34,
-
[5]
Deborah Sulem, Jack Jewson, and David Rossell. Bayesian computation for high-dimensional gaussian graphical models with spike-and-slab priors.arXiv, 2511.01875:1–139,
-
[6]
ISSN 19360975. doi: 10.1214/12-BA729. Hao Wang. Scaling it up: Stochastic search structure learning in graphical models.Bayesian Analysis, 10 (2):351–377,
-
[7]
ISSN 19316690. doi: 10.1214/14-BA916. 19 A SECTION 1 PROOFS A Section 1 proofs A.1 Proof of Proposition 1 LetSbe the set of symmetric matrices andS + the set of PD matrices. The TV distance is given by TV(p, p+) = sup A⊆S p(A)−p +(A) . The supremum is achieved by taking anyAsuch that{Θ :p(Θ)< p +(Θ)} ⊆A⊆ {Θ :p(Θ)≤p +(Θ)}, provided thatAis measurable. If Θ...
-
[8]
B.3 Proof of Theorem 1 We decompose Θ as Θ =µI+σX k whereX k has zero diagonal and i.i.d. off-diagonals with densityπ. Standard Wigner matrix theory shows that Wk = Xk√ k = Θ−µI σ √ k has eigenvalues converging to the semicircle distribution and, in particular, has minimum eigenvalueλmin(Wk)→ −2 with probability 1 ask→ ∞(Bai and Yin, 1988). Sinceλ min(Θ) ...
work page 1988
Show all 12 references
-
[9]
It follows that we still have limk→∞ c= 1 if one sets anyσ= µ (2+δk) √ k such thatk −2/3 =o(δ k)
This follows from the Wigner matrix theory in Lee and Yin (2014), which shows that deviations of the smallest eigenvalue of eΘ from−2 are of orderk −2/3 in probability. It follows that we still have limk→∞ c= 1 if one sets anyσ= µ (2+δk) √ k such thatk −2/3 =o(δ k). B.4 Proof ...
2014
-
[10]
To ease notation, letl z(Θ) andl z′(Θ) be random variables with distribution equal to the conditional distributionλ min(Θ)|Z=zandλ min(Θ)|Z=z ′ respectively
From this, any otherz ′ such thatz ′ ≥zentry- wise follows by induction. To ease notation, letl z(Θ) andl z′(Θ) be random variables with distribution equal to the conditional distributionλ min(Θ)|Z=zandλ min(Θ)|Z=z ′ respectively. The goal is to show thatE[l z(Θ)]≥E[l z′(Θ)]. ...
2018
-
[11]
To apply Theorem 4, we need to find ν=∥E(W 2)∥= X i>j E [zijθij(Eij +E ji)]2 = X i>j zij(Eii +E jj)E(θ2 ij) . where we used thatz ijθij(Eij +E ji) are independent,z 2 ij =z ij, that simple algebra shows that (Eij +E ji)2 = (Eii +E jj), and that for any set independent and zero...
2015
-
[12]
(2012), Theorem 2.7 shows that deviations of the maximum eigenvalue of a sparse Wigner matrix from 2 are of the orderk −2/3
C.8 Proof of Corollaries 5-6 Erd˝ os et al. (2012), Theorem 2.7 shows that deviations of the maximum eigenvalue of a sparse Wigner matrix from 2 are of the orderk −2/3. In particular, the maximum eigenvalue converges to 2 ask→ ∞. Note the condition thatq > N 1/3 whereq= √kηk w...
2012
Reviewed June 30, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.