Sufficient conditions for proper posteriors in fully-Bayesian Functional PCA

Ciprian Crainiceanu; Joseph Sartini; Scott Zeger

arxiv: 2604.06146 · v2 · submitted 2026-04-07 · 📊 stat.ME · math.ST· stat.TH

Sufficient conditions for proper posteriors in fully-Bayesian Functional PCA

Joseph Sartini , Scott Zeger , Ciprian Crainiceanu This is my paper

Pith reviewed 2026-05-10 18:21 UTC · model grok-4.3

classification 📊 stat.ME math.STstat.TH

keywords functional principal components analysisBayesian inferenceproper posteriorsspline basismixed effects modelsmoothing priorsfunctional data analysis

0 comments

The pith

Projecting functional principal components onto an orthonormal spline basis makes the smoothing prior and posterior proper without extra conditions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that projecting unknown functional principal components onto a rich orthonormal spline basis turns their orthonormality requirement into an equivalent orthonormality condition on the spline coefficients. This projection also lets the integral of the second-derivative penalty be represented directly as a separate smoothing parameter for each component, which enters the model as an inverse variance in the associated mixed-effects formulation. Because the resulting smoothing prior is proper under this construction, the full posterior is proper as well, even when the prior on the smoothing parameters is made less informative. Practitioners can therefore let the data determine the degree of smoothing rather than relying on strong prior restrictions to avoid impropriety.

Core claim

By projecting the functional principal components on a rich orthonormal spline basis, we show that orthonormality of the principal components is equivalent to orthonormality of the spline coefficients. A penalty on the integral of the second derivative of the functional principal components can be induced on the spline coefficients, where each function has its own smoothing parameter. Finally, each smoothing parameter is treated as an inverse variance component in the associated mixed effects model. We demonstrate that no additional conditions are required to ensure that the corresponding smoothing prior, and thus the posterior distribution, is proper. This allows the choice of less informed

What carries the argument

The equivalence between orthonormality of functional principal components and orthonormality of their spline coefficients under projection onto a rich orthonormal basis, together with inducement of the second-derivative penalty as per-function inverse variances in the mixed-effects model.

If this is right

The posterior distribution remains proper for any choice of less informative prior on the smoothing parameters.
Smoothing of each functional principal component is driven by the data through the mixed-effects variance components.
Implementation in infinite-dimensional settings is feasible without imposing further model restrictions.
The same construction applies to the full set of principal components while maintaining propriety.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This construction could reduce the need for careful prior tuning in related Bayesian functional data models that employ basis expansions.
Data-driven smoothing may lead to more adaptive estimates of principal components in applications with varying signal-to-noise ratios.
The approach opens the possibility of embedding the FPCA model inside larger hierarchical structures without introducing impropriety.

Load-bearing premise

The projection onto a rich orthonormal spline basis preserves exact orthonormality equivalence and represents the second-derivative penalty precisely as per-function inverse variances.

What would settle it

A concrete simulation or dataset where the posterior for the functional principal components becomes improper despite using the described spline projection and mixed-effects formulation would falsify the claim.

read the original abstract

In a fully-Bayesian Functional Principal Components Analysis (FPCA) the principal components are treated as unknown infinite-dimensional parameters. By projecting the functional principal components on a rich orthonormal spline basis, we show that orthonormality of the principal components is equivalent to orthonormality of the spline coefficients. A penalty on the integral of the second derivative of the functional principal components can be induced on the spline coefficients, where each function has its own smoothing parameter. Finally, each smoothing parameter is treated as an inverse variance component in the associated mixed effects model. In this work, we demonstrate that no additional conditions are required to ensure that the corresponding smoothing prior, and thus the posterior distribution, is proper. This allows the choice of less informative priors, such that smoothing is driven by the data.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Projecting FPCs onto a rich orthonormal spline basis and recasting the smoothing penalty as mixed-effects inverse variances makes the posterior proper with no further conditions.

read the letter

The one thing to know is that this paper shows a standard spline projection plus mixed-effects link is enough to guarantee proper posteriors in fully Bayesian FPCA, so you can drop extra restrictions and let the data drive the smoothing parameters via a proper hyperprior on the variances. They start from the infinite-dimensional FPCs, project them onto an orthonormal spline basis that preserves orthonormality exactly, turn the integrated squared second-derivative penalty into a quadratic form on the coefficients, and treat each per-function smoothing parameter as an inverse-variance component. This equivalence lets the hierarchical model inherit propriety from the mixed-effects representation without additional assumptions. The result is new in this specific Bayesian FPCA setting even though it rests on well-known spline and penalized-spline equivalences. It does the job cleanly and directly addresses a real barrier for people who want less informative priors in biostatistical or longitudinal applications. The finite-dimensional projected model is internally consistent, and the chain from basis choice to quadratic penalty to variance-component prior holds without circularity or self-reference. Credit for keeping the argument explicit and for showing that the construction works with standard tools rather than inventing new machinery. A soft spot is the infinite-dimensional limit: the paper needs to spell out how the propriety carries over when the basis is only approximately rich or when the number of components grows, and any uniformity or measurability details should be checked in the proofs. Edge cases with sparse observations or very flexible bases could still bite in practice. This paper is for statisticians who build or apply Bayesian FPCA models and have run into posterior propriety worries. Readers who care about theoretical backing for prior choices in functional data will get value from the targeted guarantee. It deserves a serious referee because the claim is focused, the model equivalences are reproducible in principle, and the practical payoff is clear. I would send it for peer review; the derivations look solid enough to benefit from expert eyes on the details.

Referee Report

0 major / 3 minor

Summary. The paper claims that in fully-Bayesian FPCA, projecting the unknown infinite-dimensional functional principal components onto a rich orthonormal spline basis makes their orthonormality equivalent to orthonormality of the coefficient vectors; the integrated squared second-derivative penalty then induces a quadratic form on those coefficients with per-function smoothing parameters; and treating each smoothing parameter as an inverse-variance component in the associated mixed-effects model renders the hierarchical smoothing prior (and hence the posterior) proper with no further restrictions, thereby permitting less informative, data-driven priors.

Significance. If the central derivations hold, the result supplies a clean, internally consistent route to proper posteriors in Bayesian FPCA by leveraging standard mixed-model propriety results and the exact preservation of orthonormality under the chosen basis. This is a useful contribution to functional data analysis: it removes the need for ad-hoc restrictions on priors and connects FPCA directly to well-understood hierarchical models, potentially simplifying both theory and computation in high-dimensional functional settings.

minor comments (3)

The abstract states that 'no additional conditions are required,' but the manuscript should explicitly list the standing assumptions on the spline basis (richness, orthonormality) and on the hyperprior for the smoothing parameters so that readers can verify the claim without ambiguity.
Notation for the functional principal components, their spline coefficients, and the per-function smoothing parameters should be introduced once and used consistently; a small table or glossary of symbols would improve readability.
The transition from the infinite-dimensional penalty to the finite-dimensional quadratic form on coefficients is central; a short self-contained lemma or displayed equation showing the exact induction of the penalty matrix would help readers follow the argument.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary of our manuscript and for recommending minor revision. The referee's description accurately captures our central claim that the spline projection and mixed-effects equivalence suffice to ensure propriety of the smoothing prior and posterior in fully-Bayesian FPCA, with no further restrictions required. As no specific major comments appear in the report, we have no point-by-point responses to offer.

Circularity Check

0 steps flagged

No significant circularity; derivation uses standard spline basis and mixed-model equivalences

full rationale

The paper's central demonstration—that projecting FPCs onto a rich orthonormal spline basis makes orthonormality of the functions equivalent to orthonormality of coefficients, induces the integrated squared second-derivative penalty as per-function inverse variances, and thereby renders the smoothing prior (and posterior) proper without extra conditions—rests on algebraic properties of orthonormal bases and the standard representation of penalized splines as mixed-effects models. These are external mathematical facts, not self-definitions or fitted quantities renamed as predictions. No load-bearing self-citation chain, uniqueness theorem imported from the authors' prior work, or ansatz smuggled via citation is required; the finite-dimensional hierarchical model is proper once hyperpriors are placed on the variance components. The argument is therefore self-contained and does not reduce to its inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The central claim rests on standard properties of orthonormal spline bases and the equivalence of roughness penalties to mixed-effects variance components; no free parameters or new entities are introduced.

axioms (3)

domain assumption Projection of functional principal components onto a rich orthonormal spline basis preserves orthonormality equivalence between the functions and their coefficients.
Invoked to equate orthonormality conditions.
domain assumption A penalty on the integral of the second derivative of each principal component induces a corresponding penalty on the spline coefficients with per-function smoothing parameters.
Standard spline penalty translation.
domain assumption Each smoothing parameter can be treated as an inverse variance component in the associated mixed effects model.
Links the penalty to the Bayesian prior.

pith-pipeline@v0.9.0 · 5435 in / 1428 out tokens · 32940 ms · 2026-05-10T18:21:09.722364+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

By projecting the functional principal components on a rich orthonormal spline basis, we show that orthonormality of the principal components is equivalent to orthonormality of the spline coefficients... each smoothing parameter is treated as an inverse variance component... Theorem 1: The joint prior p(Ψ,H) is proper for all valid hyperparameter values of αψ>0 and βψ>0.

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

4 extracted references · 4 canonical work pages · 1 internal anchor

[1]

Springer, New York, 2003

ISBN 978-0-387-00160-9 978-0-387-21540-2. doi: 10.1007/978-0-387-21540-2. URL http://link.springer. com/10.1007/978-0-387-21540-2 . Edited by Bickel, P. and Diggle, P. and Fienberg, S. and Krickeberg, K. and Olkin, I. and Wermuth, N. and Zeger, S. C.M. Crainiceanu and J. Goldsmith. Bayesian Functional Data Analysis Using WinBUGS.Journal of Statistical Sof...

work page doi:10.1007/978-0-387-21540-2
[2]

doi: 10.18637/jss.v032.i11

ISSN 1548-7660. doi: 10.18637/jss.v032.i11. URL https://doi.org/10. 18637/jss.v032.i11. C.M. Crainiceanu, D. Ruppert, and M.P. Wand. Bayesian analysis for penalized spline regression using winbugs. Journal of Statistical Software, 14(14):1–24,

work page doi:10.18637/jss.v032.i11
[3]

URL https: //arxiv.org/abs/2509.03512. J. Sartini, X. Zhou, E. Selvin., S. Zeger, and C.M. Crainiceanu. Fast bayesian functional principal components analysis. Journal of Computational and Graphical Statistics, 0(0):1–12,

work page internal anchor Pith review Pith/arXiv arXiv
[4]

doi: 10.1093/biomet/90.2.289

ISSN 0006-3444, 1464-3510. doi: 10.1093/biomet/90.2.289. URL https://academic.oup.com/biomet/ article-lookup/doi/10.1093/biomet/90.2.289. G. Wahba.Spline models for observational data. CBMS-NSF regional conference series in applied mathematics

work page doi:10.1093/biomet/90.2.289

[1] [1]

Springer, New York, 2003

ISBN 978-0-387-00160-9 978-0-387-21540-2. doi: 10.1007/978-0-387-21540-2. URL http://link.springer. com/10.1007/978-0-387-21540-2 . Edited by Bickel, P. and Diggle, P. and Fienberg, S. and Krickeberg, K. and Olkin, I. and Wermuth, N. and Zeger, S. C.M. Crainiceanu and J. Goldsmith. Bayesian Functional Data Analysis Using WinBUGS.Journal of Statistical Sof...

work page doi:10.1007/978-0-387-21540-2

[2] [2]

doi: 10.18637/jss.v032.i11

ISSN 1548-7660. doi: 10.18637/jss.v032.i11. URL https://doi.org/10. 18637/jss.v032.i11. C.M. Crainiceanu, D. Ruppert, and M.P. Wand. Bayesian analysis for penalized spline regression using winbugs. Journal of Statistical Software, 14(14):1–24,

work page doi:10.18637/jss.v032.i11

[3] [3]

URL https: //arxiv.org/abs/2509.03512. J. Sartini, X. Zhou, E. Selvin., S. Zeger, and C.M. Crainiceanu. Fast bayesian functional principal components analysis. Journal of Computational and Graphical Statistics, 0(0):1–12,

work page internal anchor Pith review Pith/arXiv arXiv

[4] [4]

doi: 10.1093/biomet/90.2.289

ISSN 0006-3444, 1464-3510. doi: 10.1093/biomet/90.2.289. URL https://academic.oup.com/biomet/ article-lookup/doi/10.1093/biomet/90.2.289. G. Wahba.Spline models for observational data. CBMS-NSF regional conference series in applied mathematics

work page doi:10.1093/biomet/90.2.289