arxiv: 2605.12881 · v1 · submitted 2026-05-13 · 📊 stat.ME

Recognition: no theorem link

Change-point detection in variance-covariance matrix

Ying Lin , Benjamin Poignard

Authors on Pith no claims yet

Pith reviewed 2026-05-14 19:03 UTC · model grok-4.3

classification 📊 stat.ME

keywords changecovariancedetectionestimationlassomatrixmethodpoint

0 comments

The pith

A Group Fused LASSO plus LASSO approach with adaptive weights detects change points in piecewise-constant sparse covariance matrices and yields consistent estimators under stated conditions.

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

Many datasets track how multiple variables relate to each other over time, and those relationships can suddenly change at unknown moments. The authors treat the covariance matrix as constant between change points and use two penalties: one that encourages few changes across time and another that sets many entries to zero for sparsity. They solve the resulting optimization problem with an ADMM algorithm that breaks it into simpler steps and prove that, under certain conditions, the detected change points and the estimated sparse matrices inside each segment are consistent. Experiments on simulated data and real examples compare the method to existing procedures.

Core claim

We establish the conditions under which the estimated change points and the sparse estimators within each segment are consistent.

Load-bearing premise

The variance-covariance matrix evolves in a piecewise constant manner.

Figures

Figures reproduced from arXiv: 2605.12881 by Benjamin Poignard, Ying Lin.

**Figure 2.** Figure 2: presents the computation times for the non-adaptive and second-stage adaptive estimators as heatmaps over the pλ1, λ2q grid. For the non-adaptive estimator (Figure 2a), computation time peaks at small λ2 (approximately 1.10 seconds), reaches its minimum near λ2 “ 0.4–0.5 (approximately 0.54 seconds), and rises again as λ2 approaches 1.0; variation along the λ1 direction is mild. The second-stage adaptive e… view at source ↗

**Figure 3.** Figure 3: Computation time (in seconds) as a function of [PITH_FULL_IMAGE:figures/full_fig_p027_3.png] view at source ↗

**Figure 4.** Figure 4: Computation time (in seconds) as a function of [PITH_FULL_IMAGE:figures/full_fig_p027_4.png] view at source ↗

**Figure 5.** Figure 5: Each subfigure displays the Frobenius norm of consecutive covariance differences [PITH_FULL_IMAGE:figures/full_fig_p029_5.png] view at source ↗

read the original abstract

We consider the joint estimation of change point locations and the sparsity pattern of the variance covariance matrix, which is assumed to evolve in a piecewise constant manner. By applying Group Fused LASSO and LASSO penalties to the squared Frobenius norm, we estimate both the covariance structure and the change points. Adaptive weights are incorporated into the penalty terms to enhance change point detection and covariance estimation accuracy. We establish the conditions under which the estimated change points and the sparse estimators within each segment are consistent. To solve the resulting optimization problem efficiently, we develop an alternating direction method of multipliers (ADMM) whose updates reduce to computationally tractable subproblems. The performance of the proposed method is illustrated through synthetic and real data experiments, including comparisons with several competing procedures.

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 gives a joint group fused lasso estimator for change points and sparse covariances under a piecewise constant model, solved via ADMM with claimed consistency.

read the letter

The core contribution is a single optimization problem that detects change points in the covariance matrix of a high-dimensional time series while estimating sparse covariance within each segment. They penalize the squared Frobenius norm with group fused lasso for the changes and lasso for sparsity, add adaptive weights, derive consistency for both the locations and the estimators, and implement it with an ADMM algorithm whose steps stay simple. The experiments compare it to existing methods on synthetic cases and a couple of real datasets. That joint formulation and the practical solver are the parts that stand out as new relative to separate treatments of change detection or sparse covariance estimation. The piecewise constant assumption is stated up front and the consistency results follow from standard arguments once that holds. The ADMM approach looks workable for moderate dimensions. The main soft spot is that the piecewise constant model is restrictive; many real series have smoother or more complex shifts, so performance could degrade outside the assumed setting. The consistency conditions probably impose rates on the number of changes, the minimal jump size, and the dimension, which are not always easy to check in practice. Parameter tuning is mentioned but not given much guidance, which is common but still leaves users to handle cross-validation or similar steps. This is aimed at researchers in high-dimensional time series who already work with change-point or graphical model tools and want a combined procedure. A reader focused on applications in finance or signal processing could try the method on data that fits the assumption. It deserves peer review because the formulation is coherent, the algorithm is reproducible, and the claims are testable even if the theory needs tightening in revision.

Referee Report

0 major / 2 minor

Summary. The paper proposes a penalized estimator for the joint detection of change points and the sparsity pattern in a piecewise-constant variance-covariance matrix. It applies Group Fused LASSO and LASSO penalties to the squared Frobenius norm, incorporates adaptive weights, derives consistency conditions for both the change-point locations and the segment-wise sparse covariance estimators, develops an ADMM algorithm whose subproblems are tractable, and evaluates the method on synthetic and real data against competing procedures.

Significance. If the consistency theorems hold under the stated piecewise-constant assumption, the work supplies a theoretically grounded and computationally practical tool for high-dimensional covariance change-point analysis. The combination of group-fused and element-wise penalties with adaptive weighting, together with the explicit ADMM implementation and simulation comparisons, strengthens the contribution relative to separate change-point or covariance-estimation pipelines.

minor comments (2)

[Abstract] Abstract: the precise form of the Group Fused LASSO penalty (including the grouping structure and the role of the squared Frobenius norm) is not written out; adding the explicit objective function would improve immediate readability.
[Section 4 (algorithm)] The manuscript should include a short table or paragraph comparing the computational complexity of the ADMM updates with the competing procedures mentioned in the experiments.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive evaluation of our manuscript and the recommendation for minor revision. We appreciate the recognition of the theoretical consistency results, the ADMM implementation, and the empirical comparisons. Since no specific major comments were raised, we will focus on addressing any minor points in the revised version to further strengthen the presentation.

Circularity Check

0 steps flagged

No significant circularity; derivation relies on standard penalized estimation and consistency arguments

full rationale

The paper introduces a Group Fused LASSO plus LASSO penalized estimator on the squared Frobenius norm with adaptive weights, solved by ADMM, under the explicit modeling assumption that the covariance matrix is piecewise constant. Consistency of the change-point locations and segment-wise sparse estimators is derived from standard optimization and statistical arguments applied to this formulation. No load-bearing step reduces by construction to a fitted input, self-definition, or self-citation chain; the central claims remain independent of the target results and rest on external mathematical tools plus the stated piecewise-constant assumption.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the modeling assumption that the covariance matrix is piecewise constant and on standard properties of lasso-type penalties; no new entities are introduced.

free parameters (1)

regularization parameters
The weights in the Group Fused LASSO and LASSO penalties are tuning parameters whose specific values are not derived from first principles and must be selected in practice.

axioms (1)

domain assumption The covariance matrix evolves in a piecewise constant manner.
Explicitly stated in the abstract as the structural assumption enabling the change-point model.

pith-pipeline@v0.9.0 · 5416 in / 1180 out tokens · 36570 ms · 2026-05-14T19:03:46.257215+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

82 extracted references · 82 canonical work pages

[1]

Abadir, K.M. and J.R. Magnus. Matrix algebra. 2005

work page 2005
[2]

Journal of Business & Economic Statistics , volume=

Testing and valuing dynamic correlations for asset allocation , author=. Journal of Business & Economic Statistics , volume=. 2006 , publisher=

work page 2006
[3]

Merlev\`ede, Florence and Peligrad, Magda and Emmanuel Rio , journal=. A. 2011 , publisher=

work page 2011
[4]

The Econometrics Journal , volume=

Estimation of high-dimensional vector autoregression via sparse precision matrix , author=. The Econometrics Journal , volume=. 2023 , publisher=

work page 2023
[5]

Journal of Business & Economic Statistics , volume=

Large dynamic covariance matrices , author=. Journal of Business & Economic Statistics , volume=. 2019 , publisher=

work page 2019
[6]

Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

Multiple-change-point detection for high dimensional time series via sparsified binary segmentation , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2015 , publisher=

work page 2015
[7]

Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

High dimensional change point estimation via sparse projection , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2018 , publisher=

work page 2018
[8]

Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

Change point estimation in high dimensional markov random-field models , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2017 , publisher=

work page 2017
[9]

Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , year=

Network inference via the time-varying graphical lasso , author=. Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , year=

work page
[10]

The Annals of Statistics , volume=

Wild binary segmentation for multiple change-point detection , author=. The Annals of Statistics , volume=. 2014 , publisher=

work page 2014
[11]

The Annals of Statistics , volume=

Two sample tests for high-dimensional covariance matrices , author=. The Annals of Statistics , volume=. 2012 , publisher=

work page 2012
[12]

The Annals of Statistics , volume=

Sparsistency and rates of convergence in large covariance matrix estimation , author=. The Annals of Statistics , volume=. 2009 , publisher=

work page 2009
[13]

The Annals of Statistics , volume=

Covariance regularization by thresholding , author=. The Annals of Statistics , volume=. 2008 , publisher=

work page 2008
[14]

Soviet Mathematics Doklady , volume=

Detecting disorder in multidimensional random processes , author=. Soviet Mathematics Doklady , volume=

work page
[15]

Bernoulli , volume=

Quadratic shrinkage for large covariance matrices , author=. Bernoulli , volume=. 2022 , publisher=

work page 2022
[16]

The Annals of Statistics , volume=

High-dimensional covariance matrix estimation in approximate factor models , author=. The Annals of Statistics , volume=. 2011 , publisher=

work page 2011
[17]

The Annals of Statistics , volume=

Estimating sparse precision matrix: Optimal rates of convergence and adaptive estimation , author=. The Annals of Statistics , volume=. 2016 , publisher=

work page 2016
[18]

CRC Monographs on Statistics and Applied Probability

Statistical Inference , author=. 1975 , series="CRC Monographs on Statistics and Applied Probability", publisher=

work page 1975
[19]

Bernoulli , volume=

Optimal covariance change point localization in high dimensions , author=. Bernoulli , volume=. 2021 , publisher=

work page 2021
[20]

Proceedings of the 40th International Conference on Machine Learning , series=

Divide and Conquer Dynamic Programming: An Almost Linear Time Change Point Detection Methodology in High Dimensions , author=. Proceedings of the 40th International Conference on Machine Learning , series=. 2023 , publisher=

work page 2023
[21]

Journal of the American Statistical Association , volume=

Estimating a change point in a sequence of very high-dimensional covariance matrices , author=. Journal of the American Statistical Association , volume=. 2022 , publisher=

work page 2022
[22]

Journal of the American Statistical Association , volume=

Two-sample covariance matrix testing and support recovery in high-dimensional and sparse settings , author=. Journal of the American Statistical Association , volume=. 2013 , publisher=

work page 2013
[23]

Journal of the American Statistical Association , volume=

CARE: Large precision matrix estimation for compositional data , author=. Journal of the American Statistical Association , volume=. 2024 , publisher=

work page 2024
[24]

Biometrics , volume=

Brain connectivity alteration detection via matrix-variate differential network model , author=. Biometrics , volume=. 2021 , publisher=

work page 2021
[25]

Bioinformatics , volume=

Compositional data network analysis via lasso penalized D-trace loss , author=. Bioinformatics , volume=. 2019 , publisher=

work page 2019
[26]

Journal of the American Statistical Association , volume=

Variable selection via nonconcave penalized likelihood and its oracle properties , author=. Journal of the American Statistical Association , volume=. 2001 , publisher=

work page 2001
[27]

Journal of the American Statistical Association , volume=

Panel data models with interactive fixed effects and multiple structural breaks , author=. Journal of the American Statistical Association , volume=. 2016 , publisher=

work page 2016
[28]

Journal of the American Statistical Association , volume=

Group LASSO for structural break time series , author=. Journal of the American Statistical Association , volume=. 2014 , publisher=

work page 2014
[29]

Journal of the American Statistical Association , volume=

Multiple change-point estimation with a total variation penalty , author=. Journal of the American Statistical Association , volume=. 2010 , publisher=

work page 2010
[30]

Biometrika , volume=

Differential network analysis via lasso penalized D-trace loss , author=. Biometrika , volume=. 2017 , publisher=

work page 2017
[31]

Biometrika , volume=

Sparse precision matrix estimation via lasso penalized D-trace loss , author=. Biometrika , volume=. 2014 , publisher=

work page 2014
[32]

Journal of the American Statistical Association , volume=

The adaptive lasso and its oracle properties , author=. Journal of the American Statistical Association , volume=. 2006 , publisher=

work page 2006
[33]

Annals of the Institute of Statistical Mathematics , volume=

Asymptotic theory of the adaptive Sparse Group Lasso , author=. Annals of the Institute of Statistical Mathematics , volume=. 2020 , publisher=

work page 2020
[34]

Scandinavian Journal of Statistics , volume=

Change-point detection in a linear model by adaptive fused quantile method , author=. Scandinavian Journal of Statistics , volume=. 2020 , publisher=

work page 2020
[35]

Arxiv , volume=

Sparse factor models of high dimension , author=. Arxiv , volume=. 2025 , publisher=

work page 2025
[36]

Econometric Theory , volume=

Shrinkage estimation of regression models with multiple structural changes , author=. Econometric Theory , volume=. 2016 , publisher=

work page 2016
[37]

Journal of Multivariate Analysis , volume=

High dimensional change point inference: Recent developments and extensions , author=. Journal of Multivariate Analysis , volume=. 2022 , publisher=

work page 2022
[38]

Journal of Computational and Graphical Statistics , volume=

Regularized estimation of piecewise constant Gaussian graphical models: the group-fused graphical Lasso , author=. Journal of Computational and Graphical Statistics , volume=. 2017 , publisher=

work page 2017
[39]

Journal of Machine Learning Research , volume=

Zhao, Peng and Bin Yu , title =. Journal of Machine Learning Research , volume=

work page
[40]

Annals of the Institute of Statistical Mathematics , volume=

Consistent multiple changepoint estimation with fused Gaussian graphical models , author=. Annals of the Institute of Statistical Mathematics , volume=. 2021 , publisher=

work page 2021
[41]

HAL, Technical Report, Computational Biology Center, Paris , volume=

The group fused lasso for multiple change-point detection , author=. HAL, Technical Report, Computational Biology Center, Paris , volume=. 2011 , publisher=

work page 2011
[42]

Electronic Journal of Statistics , volume=

Change-point detection in high-dimensional covariance structure , author=. Electronic Journal of Statistics , volume=. 2018 , publisher=

work page 2018
[43]

Electronic Journal of Statistics , volume=

Estimating networks with jumps , author=. Electronic Journal of Statistics , volume=. 2012 , publisher=

work page 2012
[44]

Machine Learning , volume=

Time varying undirected graphs , author=. Machine Learning , volume=. 2010 , publisher=

work page 2010
[45]

NeuroImage , volume=

Estimating time-varying brain connectivity networks from functional MRI time series , author=. NeuroImage , volume=. 2014 , publisher=

work page 2014
[46]

Journal of Econometrics , volume=

Shrinkage estimation of common breaks in panel data models via adaptive group fused Lasso , author=. Journal of Econometrics , volume=. 2016 , publisher=

work page 2016
[47]

, title =

Eckstein, Jonathan and Bertsekas, Dimitri P. , title =. Mathematical Programming , year = 1992, volume = 55, number =

work page 1992
[48]

and Sun, Defeng and Tseng, Paul , title =

Fazel, Maryam and Pong, Ting K. and Sun, Defeng and Tseng, Paul , title =. SIAM Journal on Matrix Analysis and Applications , year = 2013, volume = 34, number = 3, month = jan, pages =

work page 2013
[49]

Computers & Mathematics with Applications , year = 1976, volume = 2, number = 1, pages =

Gabay, Daniel and Mercier, Bertrand , title =. Computers & Mathematics with Applications , year = 1976, volume = 2, number = 1, pages =

work page 1976
[50]

Revue française d’automatique, informatique, recherche opérationnelle

Glowinski, Roland and Marroco, Americo , title =. Revue française d’automatique, informatique, recherche opérationnelle. Analyse numérique , year = 1975, volume = 9, number =

work page 1975
[51]

Studies in Mathematics and Its Applications , publisher =

Applications of the Method of Multipliers to Variational Inequalities , year = 1983, author =. Studies in Mathematics and Its Applications , publisher =

work page 1983
[52]

and Han, Deren and Yang, Hai , title =

He, Bingsheng and Liao, Li-Z. and Han, Deren and Yang, Hai , title =. Mathematical Programming , year = 2002, volume = 92, number = 1, month = mar, pages =

work page 2002
[53]

Optimization Methods and Software , year = 1994, volume = 4, number = 1, month = jan, pages =

Eckstein, Jonathan , title =. Optimization Methods and Software , year = 1994, volume = 4, number = 1, month = jan, pages =

work page 1994
[54]

Tyrrell , title =

Rockafellar, R. Tyrrell , title =

work page
[55]

and Combettes, Patrick L

Bauschke, Heinz H. and Combettes, Patrick L. , title =

work page
[56]

, title =

Li, Xudong and Sun, Defeng and Toh, Kim-C. , title =. Mathematical Programming , year = 2016, volume = 155, number =

work page 2016
[57]

, title =

Chen, Liang and Sun, Defeng and Toh, Kim-C. , title =. Mathematical Programming , year = 2017, volume = 161, number =

work page 2017
[58]

, title =

Li, Xudong and Sun, Defeng and Toh, Kim-C. , title =. Mathematical Programming Computation , year = 2018, volume = 10, number = 4, month = apr, pages =

work page 2018
[59]

, title =

Li, Xudong and Sun, Defeng and Toh, Kim-C. , title =. Mathematical Programming , year = 2019, volume = 175, number =

work page 2019
[60]

Toh , title =

Lei Yang and Jia Li and Defeng Sun and Kim-C. Toh , title =. Journal of Machine Learning Research , year = 2021, volume = 22, pages =

work page 2021
[61]

SIAM Journal on Optimization , year = 2017, volume = 27, number = 3, month = jan, pages =

Xu, Yangyang , title =. SIAM Journal on Optimization , year = 2017, volume = 27, number = 3, month = jan, pages =

work page 2017
[62]

SIAM Journal on Imaging Sciences , year = 2015, volume = 8, number = 1, month = jan, pages =

Ouyang, Yuyuan and Chen, Yunmei and Lan, Guanghui and Pasiliao, Eduardo , title =. SIAM Journal on Imaging Sciences , year = 2015, volume = 8, number = 1, month = jan, pages =

work page 2015
[63]

Journal of Scientific Computing , year = 2019, volume = 79, number = 2, month = dec, pages =

Li, Huan and Lin, Zhouchen , title =. Journal of Scientific Computing , year = 2019, volume = 79, number = 2, month = dec, pages =

work page 2019
[64]

Mathematical Programming Computation , year = 2018, volume = 10, number = 4, month = jan, pages =

Xiao, Yunhai and Chen, Liang and Li, Donghui , title =. Mathematical Programming Computation , year = 2018, volume = 10, number = 4, month = jan, pages =

work page 2018
[65]

SIAM Journal on Optimization , year = 2022, volume = 32, number = 1, month = feb, pages =

Sabach, Shoham and Teboulle, Marc , title =. SIAM Journal on Optimization , year = 2022, volume = 32, number = 1, month = feb, pages =

work page 2022
[66]

SIAM Journal on Optimization , year = 2025, volume = 35, number = 2, pages =

Defeng Sun AND Yancheng Yuan AND Guojun Zhang AND Xinyuan Zhao , title =. SIAM Journal on Optimization , year = 2025, volume = 35, number = 2, pages =

work page 2025
[67]

Journal of Optimization Theory and Applications , year = 2025, volume = 204, pages =

Bo Yang AND Xinyuan Zhao AND Xudong Li AND Defeng Sun , title =. Journal of Optimization Theory and Applications , year = 2025, volume = 204, pages =

work page 2025
[68]

2211.14881v1 , primaryclass =

Guojun Zhang AND Yancheng Yuan AND Defeng Sun , title =. 2211.14881v1 , primaryclass =

work page arXiv
[69]

Augmented

Fortin, Michel and Glowinski, Roland , series=. Augmented. 1983 , volume=

work page 1983
[70]

2024 , eprint=

Break recovery in graphical networks with D-trace loss , author=. 2024 , eprint=

work page 2024
[71]

The Annals of Statistics , year = 2013, volume = 41, number = 5, publisher =

Wang, Lan and Kim, Yongdai and Li, Runze , title =. The Annals of Statistics , year = 2013, volume = 41, number = 5, publisher =

work page 2013
[72]

and Chen, Zehua , title =

Chen, Jiahua. and Chen, Zehua , title =. Biometrika , year = 2008, volume = 95, number = 3, pages =

work page 2008
[73]

and Siegmund, David O

Zhang, Nancy R. and Siegmund, David O. , title =. Statistica Sinica , year = 2012, volume = 22, pages =

work page 2012
[74]

Statistica Sinica , volume=

Multiscale and multilevel technique for consistent segmentation of nonstationary time series , author=. Statistica Sinica , volume=. 2012 , publisher=

work page 2012
[75]

Statistica Sinica , year = 2023, volume =

Bai, Yue and Safikhani, Abolfazl , title =. Statistica Sinica , year = 2023, volume =

work page 2023
[76]

Journal of the American Statistical Association , year =

Cai, Tony and Liu, Weidong , title =. Journal of the American Statistical Association , year =

work page
[77]

Journal of the American Statistical Association , year =

Xue, Lingzhou and Ma, Shiqian and Zou, Hui , title =. Journal of the American Statistical Association , year =

work page
[78]

Journal of Computational and Graphical Statistics , year =

Liu, Han and Wang, Lie and Zhao, Tuo , title =. Journal of Computational and Graphical Statistics , year =

work page
[79]

and Hero, Alfred O

Chen, Yilun and Wiesel, Ami and Eldar, Yonina C. and Hero, Alfred O. , title =. IEEE Transactions on Signal Processing , year =

work page
[80]

Journal of the Royal Statistical Society: Series B , year =

Fan, Jianqing and Liao, Yuan and Mincheva, Martina , title =. Journal of the Royal Statistical Society: Series B , year =

work page

Showing first 80 references.