arxiv: 2605.08471 · v2 · submitted 2026-05-08 · 🧮 math.ST · stat.TH

Recognition: 2 theorem links

· Lean Theorem

Asymptotics for likelihood ratio tests of boundary points with singular information and unidentifiable nuisance parameters

J. M. Patrik Albin, Karl Oskar Ekvall, Matteo Bottai, Ola H\"ossjer

Pith reviewed 2026-05-13 07:42 UTC · model grok-4.3

classification 🧮 math.ST stat.TH

keywords asymptoticslikelihood ratio testboundary inferenceunidentifiable parameterssingular information matrixmixture modelsbar-chi-squared process

0 comments

The pith

The likelihood ratio test statistic for boundary parameters with unidentifiable nuisance parameters converges in distribution to the supremum of a bar-chi-squared process.

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

This paper derives the limiting behavior of likelihood ratio tests when testing boundary points in the presence of unidentifiable nuisance parameters and possibly singular information. The central result is that the test statistic's null distribution is the supremum of a stochastic process whose marginals are chi-squared mixtures depending on the nuisance. Readers should care because this provides a general framework for inference in models like finite mixtures and linkage analysis, where standard approximations do not apply, and it unifies many previous special-case results. The derivation holds under local alternatives as well, with a noncentral version of the process.

Core claim

Under suitable regularity conditions the asymptotic distribution of the LRT statistic under the null hypothesis is the supremum of a bar-chi-squared process, that is, a stochastic process whose marginal distributions are mixtures of chi-squared distributions with weights depending on the nuisance parameter. Under local alternatives the asymptotic distribution is the supremum of a noncentral bar-chi-squared process. Singularity in the information matrix stemming from the nuisance parameter does not alter the form of the limit, unlike when singularity arises from the parameter of interest.

What carries the argument

The supremum of a bar-chi-squared process, a stochastic process with marginal distributions that are mixtures of chi-squared distributions weighted according to the value of the nuisance parameter.

If this is right

The results recover all existing cases with nonsingular information or absent nuisance parameters as special cases.
Several application-specific results for mixture models and genetic linkage analysis are recovered and extended.
The same limiting form applies even when the information matrix is singular due to the nuisance parameters.
Analogous results hold for change-point detection problems that share these features.

Where Pith is reading between the lines

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

If the regularity conditions hold in a given model, one can simulate the bar-chi-squared process to obtain critical values for the test.
This framework suggests similar asymptotic analyses could apply to other models with unidentifiable components, such as certain clustering or regime-switching models.
The local alternative results enable power calculations that depend on the nuisance in a structured way.

Load-bearing premise

The statistical model satisfies suitable regularity conditions ensuring the information matrix behaves appropriately and the supremum of the process is well-defined.

What would settle it

If in a concrete model such as a finite Gaussian mixture the finite-sample distribution of the LRT statistic under the null differs substantially from what the supremum of the bar-chi-squared process predicts, even as sample size grows, the asymptotic claim would be falsified.

Figures

Figures reproduced from arXiv: 2605.08471 by J. M. Patrik Albin, Karl Oskar Ekvall, Matteo Bottai, Ola H\"ossjer.

**Figure 1.** Figure 1: Pedigree types (Pj , Φj ) used in Tables 1-2. For j = 1, 2, 3, 4, Pj consists of two parents with unknown phenotypes and k + 1 affected offspring. (In particular, (P1, Φ1) is an affected sib pair.) P5 (upper), P6 (middle) and P7 (lower) are shown above with individual numbers. Males and females correspond to squares and circles, affected individuals have black and unaffected ones have white symbols. Indivi… view at source ↗

**Figure 2.** Figure 2: Plot of χ 2 2 -weight w2 as function of the proportion β of affected sib pairs for a mixture of affected sib and first cousin pairs. 54 [PITH_FULL_IMAGE:figures/full_fig_p054_2.png] view at source ↗

read the original abstract

We establish the asymptotic distribution of likelihood ratio tests (LRTs) in settings where some of the nuisance parameters are unidentifiable under the null hypothesis, parameters of interest lie on the boundary of the parameter space, and the information matrix of the identifiable parameters may be singular. Our work is motivated by mixture models and genetic linkage analysis, which exhibit all three features simultaneously, but it is applicable more broadly to other problems such as change-point detection. Under suitable regularity conditions, the asymptotic distribution of the LRT statistic under the null hypothesis is the supremum of a $\bar{\chi}^2$-process, that is, a stochastic process whose marginal distributions are mixtures of $\chi^2$-distributions with weights depending on the nuisance parameter. Under local alternatives, the asymptotic distribution of the LRT statistic is the supremum of a noncentral $\bar{\chi}^2$-process, whose marginal distributions are mixtures of truncated, noncentral $\chi^2$-distributions. In contrast to prior work on singular information, where singularity stems from the parameter of interest and changes the form of the limit distribution, here singularity is determined by the nuisance parameter and the limit has the same form as in the nonsingular case. Existing results for boundary inference with nonsingular information or without nuisance parameters are obtained as special cases, and several existing application-specific results for mixture models and genetic linkage analysis are recovered and extended.

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

The paper shows the LRT limit stays a sup of a bar-chi-squared process even when singularity comes from the nuisance parameters rather than the interest parameter.

read the letter

The main new piece is that the asymptotic distribution of the LRT under the null remains the supremum of a bar-chi-squared process when the information singularity is driven by unidentifiable nuisance parameters. This keeps the same functional form as the nonsingular case, which differs from earlier results where singularity tied to the parameter of interest altered the limit. The paper recovers the mixture-model and genetic-linkage examples as special cases and adds the local-alternative version with noncentral mixtures, all under one set of regularity conditions on the information matrix and process supremum. That unification is useful because those two application areas often hit boundary, singularity, and nuisance issues at the same time. The derivation appears to follow standard tightness and weak-convergence arguments for these processes, with no obvious circularity or fitting inside the limit statement. One soft spot is that the regularity conditions are invoked at a general level and will still need model-specific checks in practice, for instance to confirm the supremum is well-defined for finite mixtures where identifiability can be delicate. This is typical for the literature rather than a flaw in the argument itself. The work is aimed at statisticians who need correct critical values for LRTs in non-regular models such as mixtures, change-point detection, or linkage analysis. A reader focused on asymptotic theory for boundary problems will get direct value from the extension. I would bring it to a reading group to walk through the process construction and would cite it if working on similar non-regular inference. It deserves a serious referee because the distinction from prior singular-information cases is clear and the applications are concrete.

Referee Report

1 major / 2 minor

Summary. The paper establishes the asymptotic distribution of likelihood ratio tests (LRTs) in settings where nuisance parameters are unidentifiable under the null, parameters of interest lie on the boundary, and the information matrix may be singular. Under suitable regularity conditions, the null asymptotic distribution of the LRT statistic is the supremum of a bar-chi-squared process whose marginals are mixtures of chi-squared distributions with nuisance-dependent weights. Under local alternatives, it is the supremum of a noncentral bar-chi-squared process. The result recovers special cases from mixture models, genetic linkage, and change-point detection, and shows that nuisance-induced singularity does not change the form of the limit unlike interest-parameter singularity.

Significance. This work is significant as it provides a unified asymptotic theory for non-regular LRT problems that arise frequently in statistical applications involving mixtures and boundary constraints. By demonstrating that the limit retains the standard bar-chi-squared form when singularity comes from the nuisance parameter, it simplifies inference in these models and extends prior results on boundary testing and singular information. The recovery of application-specific results as special cases adds credibility, and the local alternative analysis supports power calculations.

major comments (1)

The proof of convergence to the supremum of the bar-chi-squared process (main theorem) relies on tightness and stochastic equicontinuity arguments that are outlined but not fully detailed for the case of singular information induced by the nuisance parameter; this is load-bearing for the central asymptotic claim.

minor comments (2)

The abstract introduces the bar-chi-squared process without a brief inline definition or pointer to its marginal distribution, which reduces accessibility for readers new to the area.
In the introduction and comparison section, the contrast with prior singular-information results (where singularity arises from the parameter of interest) could be made more explicit by citing the specific change in limit form that is avoided here.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading and positive assessment of the manuscript. We address the single major comment below and will revise accordingly.

read point-by-point responses

Referee: The proof of convergence to the supremum of the bar-chi-squared process (main theorem) relies on tightness and stochastic equicontinuity arguments that are outlined but not fully detailed for the case of singular information induced by the nuisance parameter; this is load-bearing for the central asymptotic claim.

Authors: We agree that the tightness and stochastic equicontinuity arguments in the proof of the main theorem (Theorem 3.1) are outlined at a high level in the main text and appendix but would benefit from fuller expansion when the information singularity arises from the nuisance parameter. In the revised manuscript we will add a dedicated subsection in the appendix that derives these arguments explicitly under the stated regularity conditions. The derivation will verify the moment and Lipschitz conditions on the score process, confirm that nuisance-induced singularity does not alter the modulus-of-continuity bounds, and show that the same chaining arguments used in the nonsingular case continue to apply, thereby establishing the required tightness without changing the form of the limiting bar-chi-squared process. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper derives the asymptotic distribution of the LRT statistic as the supremum of a bar-chi-squared process under regularity conditions on the model, information matrix, and nuisance parameters. This follows from standard likelihood theory extensions for boundary and singular cases without reducing the target limit distribution to any fitted parameter, self-defined quantity, or load-bearing self-citation chain. Existing results are recovered as special cases, but the central claim is obtained directly from the likelihood and information structure rather than by construction from inputs. No equation equates the predicted limit to a data-dependent fit or renames a known result via ansatz smuggling.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on a set of regularity conditions that guarantee the existence of the limiting process and the validity of the supremum representation; these conditions are standard in asymptotic statistics but must hold for the concrete models of interest.

axioms (1)

domain assumption suitable regularity conditions on the likelihood and information matrix
Invoked throughout the derivation to obtain the bar-chi-squared limit and to ensure the process is well-defined.

pith-pipeline@v0.9.0 · 5570 in / 1306 out tokens · 50182 ms · 2026-05-13T07:42:22.017152+00:00 · methodology

Review history (2 revisions) →

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/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
Under suitable regularity conditions, the asymptotic distribution of the LRT statistic under the null hypothesis is the supremum of a bar-chi-squared process... X(t) = sup delta in Delta(t) Q(delta,t) with Q = 2 delta^T Z(t) - ||delta||^2
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear
X(t) has a bar-chi^2 distribution... mixture of chi^2_j with weights w_j(t) determined by Delta(t)

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages

[1]

Donald W. K. Andrews. Admissibility of the likelihood ratio test when the parameter space is restricted under the alternative. Econometrica, 64 0 (3): 0 705--718, 1996

work page 1996
[2]

Donald W. K. Andrews. Testing when a parameter is on the boundary of the maintained hypothesis. Econometrica, 69 0 (3): 0 683--734, 2001

work page 2001
[3]

Donald W. K. Andrews and W. Ploberger. Admissibility of the likelihood ratio test when a nuisance parameter is present only under the alternative. Annals of Statistics, 23 0 (5): 0 1609--1629, 1995

work page 1995
[4]

P. J. Bickel and H. Chernoff. Asymptotic distribution of the likelihood ratio statistic in a prototypical non-regular problem. In J.K. Ghosh, S.K. Mitra, K.R. Parthasarathy, and Prakasa B.L.S Rao, editors, A Raghu Raj Bahadur Festschrift, pages 83--96. Wiley, New York, 1993

work page 1993
[5]

Confidence regions when the Fisher information is zero

Matteo Bottai. Confidence regions when the Fisher information is zero. Biometrika, 90 0 (1): 0 73--84, 2003. doi:10.1093/biomet/90.1.73

work page doi:10.1093/biomet/90.1.73 2003
[6]

Large sample distribution of the likelihood ratio test for normal mixtures

Hanfeng Chen and Jiahua Chen. Large sample distribution of the likelihood ratio test for normal mixtures. Statistics & Probability Letters, 52: 0 125--133, 2001. doi:10.1016/S0167-7152(00)00171-1

work page doi:10.1016/s0167-7152(00)00171-1 2001
[7]

Tests for homogeneity in normal mixtures in the presence of a structural parameter

Jiahua Chen and Hanfeng Chen. Tests for homogeneity in normal mixtures in the presence of a structural parameter. Statistica Sinica, 13: 0 351--365, 2003

work page 2003
[8]

On the distribution of the likelihood ratio

Herman Chernoff. On the distribution of the likelihood ratio. The Annals of Mathematical Statistics, 25 0 (3): 0 573--578, 1954. doi:10.1214/aoms/1177728725

work page doi:10.1214/aoms/1177728725 1954
[9]

Asymptotic distribution of the likelihood ratio test that a mixture of two binomials is a single binomial

Herman Chernoff and Eric Lander. Asymptotic distribution of the likelihood ratio test that a mixture of two binomials is a single binomial. Journal of Statistical Planning and Inference, 43: 0 19--40, 1995

work page 1995
[10]

A generalized argmax theorem with applications

Gregory Cox. A generalized argmax theorem with applications. arXiv preprint arXiv:2209.08793, 2022

work page arXiv 2022
[11]

Dacunha-Castelle and E

D. Dacunha-Castelle and E. Gassiat. Testing the order of a model using locally conic parametrization: Population mixtures and stationary arma processes. Annals of Statistics, 27 0 (4): 0 1178--1209, 1999

work page 1999
[12]

Robert B. Davies. Hypothesis testing when a nuisance parameter is present only under the alternative. Biometrika, 62 0 (2): 0 247--254, 1977

work page 1977
[13]

Robert B. Davies. Hypothesis testing when a nuisance parameter is present only under the alternative. Biometrika, 74 0 (1): 0 33--43, 1987

work page 1987
[14]

Brown, and David Siegmund

Jos \'e e Dupuis, Philip O. Brown, and David Siegmund. Statistical methods for linkage analysis of complex traits from high-resolution maps of identity by descent. Genetics, 140: 0 843--856, 1995

work page 1995
[15]

Confidence regions near singular information and boundary points with applications to mixed models

Karl Oskar Ekvall and Matteo Bottai. Confidence regions near singular information and boundary points with applications to mixed models. The Annals of Statistics, 50 0 (3): 0 1806--1832, 2022. doi:10.1214/22-AOS2177

work page doi:10.1214/22-aos2177 2022
[16]

The log likelihood ratio in segmented regression

Paul Feder. The log likelihood ratio in segmented regression. The Annals of Statistics, 3 0 (1): 0 84--97, 1975

work page 1975
[17]

Brown, and David Siegmund

Eleanor Feingold, Philip O. Brown, and David Siegmund. Gaussian models for genetic linkage analysis using complete high-resolution maps of identity by descent. American Journal of Human Genetics, 53: 0 234--251, 1993

work page 1993
[18]

Charles J. Geyer. On the asymptotics of constrained M-estimation . The Annals of Statistics, 22 0 (4): 0 1993--2010, 1994. doi:10.1214/aos/1176325768

work page doi:10.1214/aos/1176325768 1993
[19]

Bootstrap test procedure for variance components in nonlinear mixed effects models in the presence of nuisance parameters and a singular Fisher information matrix

T Gu \'e don, C Baey, and E Kuhn. Bootstrap test procedure for variance components in nonlinear mixed effects models in the presence of nuisance parameters and a singular Fisher information matrix. Biometrika, 111 0 (4): 0 1331--1348, 2024. doi:10.1093/biomet/asae025

work page doi:10.1093/biomet/asae025 2024
[20]

Asymptotic properties of affected-sib-pair linkage analysis

Peter Holmans. Asymptotic properties of affected-sib-pair linkage analysis. American Journal of Human Genetics, 52: 0 362--374, 1993

work page 1993
[21]

Determining inheritance distributions via stochastic penetrances

Ola H \"o ssjer. Determining inheritance distributions via stochastic penetrances. Journal of the American Statistical Association, 98: 0 1035--1051, 2003

work page 2003
[22]

Conditional likelihood score functions in linkage analysis

Ola H \"o ssjer. Conditional likelihood score functions in linkage analysis. Biostatistics, 6: 0 313--332, 2005 a

work page 2005
[23]

Spectral decomposition of score functions in linkage analysis

Ola H \"o ssjer. Spectral decomposition of score functions in linkage analysis. Bernoulli, 11 0 (6): 0 1093--1113, 2005 b

work page 2005
[24]

Lander and David Botstein

Eric S. Lander and David Botstein. Mapping mendelian factors underlying quantitative traits using RFLP linkage maps. Genetics, 121: 0 185--199, 1989

work page 1989
[25]

Likelihood ratio tests for genetic linkage

Mohamed Lemdani and Odile Pons. Likelihood ratio tests for genetic linkage. Statistics and Probability Letters, 33: 0 15--22, 1997

work page 1997
[26]

Mixture Models: Theory, Geometry and Applications

B.G Lindsay. Mixture Models: Theory, Geometry and Applications. IMS, Hayward, CA, 1995

work page 1995
[27]

Optimal allele-sharing statistics for genetic mapping using affected relatives

Mary Sara McPeek. Optimal allele-sharing statistics for genetic mapping using affected relatives. Genetic Epidemiology, 16: 0 225--249, 1999

work page 1999
[28]

Skovgaard

Christian Ritz and Ib M. Skovgaard. Likelihood ratio tests in curved exponential families with nuisance parameters present only under the alternative. Biometrika, 92: 0 507--517, 2005

work page 2005
[29]

Cox, Matteo Bottai, and James Robins

Andrea Rotnitzky, David R. Cox, Matteo Bottai, and James Robins. Likelihood-based inference with singular information matrix. Bernoulli, 6: 0 243--284, 2000

work page 2000
[30]

Self and Kung-Yee Liang

Steven G. Self and Kung-Yee Liang. Asymptotic properties of maximum likelihood estimators and likelihood ratio tests under nonstandard conditions. Journal of the American Statistical Association, 82 0 (3): 0 605--610, 1987

work page 1987
[31]

P. K. Sen and M.J. Silvapulle. An appraisal of statistical inference under inequality constraints. Journal of Statistical Planning and Inference, 107: 0 3--43, 2002

work page 2002
[32]

Serfling

Robert J. Serfling. Approximation Theorems of Mathematical Statistics . Wiley Series in Probability and Mathematical Statistics. John Wiley & Sons, Hoboken, NJ, USA, 2002. ISBN 978-0-470-31648-1. doi:10.1002/9780470316481

work page doi:10.1002/9780470316481 2002
[33]

Statistics in Human Genetics

Pak Sham. Statistics in Human Genetics. Arnold Applications of Statistics Series. Arnold, London, 1998. ISBN 978-0-340-66241-0 978-0-471-19488-0

work page 1998
[34]

A. Shapiro. Towards a unified theory of inequality constrained testing in multivariate analysis. International Statistical Review / Revue Internationale de Statistique, 56 0 (1): 0 49--62, 1988. doi:10.2307/1403361

work page doi:10.2307/1403361 1988
[35]

Asymptotic distribution of test statistics in the analysis of moment structures under inequality constraints

Alexander Shapiro. Asymptotic distribution of test statistics in the analysis of moment structures under inequality constraints. Biometrika, 72 0 (1): 0 133--144, 1985. doi:10.1093/biomet/72.1.133

work page doi:10.1093/biomet/72.1.133 1985
[36]

B. K. Suarez, J. Rice, and T. Reich. The generalized sib pair IBD distribution: Its use in detection of linkage. Annals of Human Genetics, 44: 0 87--94, 1978

work page 1978
[37]

van der Vaart and Jon A

Aad W. van der Vaart and Jon A. Wellner. Weak Convergence and Empirical Processes: With Applications to Statistics . Springer Series in Statistics. Springer, Cham, Switzerland, 2 edition, 2023. ISBN 978-3-031-29038-1 978-3-031-29040-4

work page 2023
[38]

Whittemore and Jennifer Halpern

Alice S. Whittemore and Jennifer Halpern. A class of tests for linkage using affected pedigree members. Biometrics, 50: 0 118--127, 1994

work page 1994
[39]

S. S. Wilks. The large-sample distribution of the likelihood ratio for testing composite hypotheses. The Annals of Mathematical Statistics, 9 0 (1): 0 60--62, 1938. doi:10.1214/aoms/1177732360

work page doi:10.1214/aoms/1177732360 1938