Pith. sign in

REVIEW 4 major objections 5 minor 35 references

Machine learning in phase transition analysis of lattice quantum gravity

T0 review · 4 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Automated machine learning reliably detects known phase transitions in causal dynamical triangulations, using only geometric observables and often producing sharper signals than conventional order parameters.

desk verdict Useful first ML benchmark on CDT transitions, but copy-leakage in the validation protocol undermines the 'outperforming' claim. read the letter →

arxiv 2510.02159 v2 pith:QDVQHPDN submitted 2025-10-02 hep-lat hep-th

Jan Ambjorn , Zbigniew Drogosz , Jakub Gizbert-Studnicki , Andrzej Görlich , Dániel Németh , Marcus Reitz This is my paper · ORCID
classification hep-lathep-th MSC 81T2583C2782B26 PACS 04.60.Nc05.70.Fh07.05.Mh
keywords CausalDynamicalTriangulationslatticequantumgravityphasetransitionsmachinelearningsupervisedunsupervisedMonteCarlosimulationorderparameters
topics Quantum Gravity
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tests whether off-the-shelf supervised and unsupervised machine learning models can recognize the phase transitions of four-dimensional Causal Dynamical Triangulations. Using 30 purely geometric observables from Monte Carlo simulations, most supervised classifiers identify the phases with high accuracy and locate the A-B, A-C, and B-Cb transition points in agreement with standard order-parameter methods. For several transitions, the ML probability signals are sharper than the traditional order parameters, allowing precise transition-point identification. The authors view this as a first step toward automated exploration of the CDT parameter space and possible discovery of new phases.

What carries the argument

The pipeline uses 30 geometric observables (simplex counts, vertex counts, coordination numbers), augmented fourfold by time-shift symmetry, as input features. Seven supervised and seven unsupervised models are trained on the two parameter points deepest inside each pair of phases, validated on separate data from the same runs, then applied to intermediate parameter values. The phase transition is identified as the location where the mean classification or clustering probability changes from about 0 to 1, typically with a peak in the probability susceptibility.

What would settle it

Train the same classifiers on data from only one half of a long Monte Carlo run and test on data from the other half; if the estimated transition point shifts or the probability jump becomes diffuse, the signal is dominated by within-run correlations rather than phase physics.

Watch

Extended reading notes

Core claim

Generic machine learning classifiers, trained on geometric features extracted from Monte Carlo configurations, reproduce the positions of three known CDT phase transitions. The classifiers achieve over 99.9% validation accuracy on data drawn from deep within each phase, and the mean class probability jumps sharply at the expected transition points. In several cases this probability signal outperforms standard order parameters in sharpness, indicating that the geometric feature distributions already encode the phase structure without needing hand-designed order parameters.

Load-bearing premise

The assumption that better-than-99.9% validation accuracy on random splits of the same Monte Carlo data reflects genuine separation of phases rather than leakage from correlated Markov-chain samples.

Editorial extensions

If this is right

  • Supervised machine learning can serve as a phase-transition detector in lattice quantum gravity without requiring hand-crafted order parameters.
  • Sharper probability signals may enable more precise localization of first-order transition lines, which is relevant for locating the continuum limit.
  • The success of unsupervised methods with a fixed number of clusters suggests a route toward label-free phase identification.
  • The approach can be extended to multi-phase classification and to different spatial topologies, as the authors propose.
  • The method's success at known transitions motivates its use as an automated scan tool over the full CDT parameter space.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The >99.9% validation accuracy likely overstates generalization because validation sets are random splits of the same Monte Carlo runs used for training; Markov-chain autocorrelations could cause leakage and inflate the reported sharpness.
  • A stronger test of the method as a discovery tool would be to train at one lattice volume or parameter region and predict transitions at another, checking transferability rather than interpolation on the same runs.
  • The failure of Decision Tree and Random Forest on the A-B transition suggests that high training accuracy alone is insufficient; the choice of features and model bias can distort the transition location.
  • The same geometric-observable-plus-classification template could be exported to other lattice quantum field theories, where local observables would play the role of the 30 CDT features.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents a supervised and unsupervised machine-learning analysis of phase transitions in four-dimensional Causal Dynamical Triangulations. Using 30 geometric observables from Monte Carlo simulations of toroidal CDT, the authors train seven classifiers (and test seven clustering methods) on data points deep inside pairs of phases (A-B, A-C, B-C_b) and then apply the trained models to parameter scans. The models produce probability curves whose jump/susceptibility peak is used to locate the transition, and the paper claims that most supervised models and some unsupervised models not only reproduce standard order-parameter results but yield sharper signals, 'outperforming standard methods.'

Significance. The potential value is real: if the validation protocol is sound, the paper would demonstrate that off-the-shelf ML classifiers can separate CDT phases from geometric observables alone and locate known transitions, a useful step toward automated phase exploration in quantum gravity. The paper's breadth—three transitions, several volumes, 14 algorithms, comparison with standard order parameters—is a strength. However, the central quantitative claims rest on a validation gate that is vulnerable to data leakage through the time-shift augmentation and to autocorrelation effects, and the 'outperforming standard methods' comparison uses order parameters that are algebraic functions of the ML input features. These issues must be resolved before the claims can be taken as established.

major comments (4)
  1. [Section III steps 3–5 and Appendix] The Appendix states that the dataset size was quadrupled by cyclically shifting the time coordinate of all local parameters while global parameters are kept unchanged. Hence every MC configuration appears four times in the input data. The paper does not specify whether the train/validation split of Section III (step 3) is made before or after this augmentation. If the split is after augmentation, a random split can place different shifted copies of the same configuration in both training and validation. The reported >99.9% validation accuracy can then be achieved by recognizing exact duplicates rather than by generalizing to new geometries, even for statistically independent configurations. Because step 6 enters the transition-location pipeline only after this accuracy gate, the central results in Figs. 2 and 4 are not securely supported as reported. Please either split before augmentati
  2. [Section IV/V and Appendix] The claim of 'outperforming standard methods based on order parameters' is not a level comparison. As the Appendix shows, OP1=N0/N41 and OP2=N32/N41=N4/N41-1 are algebraic combinations of N0, N41, N32, N4, all of which are included in the 30 ML input features. The ML classifiers therefore have direct access to the information contained in the order parameters (plus 26 additional features). A sharper probability curve in Fig. 2 may reflect the classifier's calibration or nonlinear combination of features rather than an intrinsically more powerful observable. The authors should either benchmark against order parameters constructed from features not given to the ML model, or explicitly qualify the statement as an in-feature comparison.
  3. [Section III steps 3–5] No thinning, binning, or autocorrelation analysis is reported. CDT Monte Carlo data are generally correlated along the Markov chain, and a random split of individual configurations from the same runs can inflate validation accuracy because neighboring configurations are near-duplicates. The paper should report the effective number of independent samples (integrated autocorrelation time) or use block-based splitting/independent runs for training and validation. Without this, the >99.9% accuracy and the quantitative transition locations are difficult to assess.
  4. [Section III step 9 and Fig. 2] The transition point is defined only qualitatively: 'where the probability jumps from approximately 0 to approximately 1.' No objective estimator (e.g., crossing of 0.5 or peak of susceptibility) or statistical error bar is given. Given that each scan includes only a handful of parameter values (e.g., 11 values of Δ for A-B at N41=100k), the claimed 'very precise identification of the phase transition points' and agreement with standard methods need a quantitative, reproducible criterion and uncertainty estimate.
minor comments (5)
  1. [Fig. 4] 'DecissionTree' is a typo; use 'Decision Tree' consistently with the text and Fig. 3.
  2. [Footnote 1] 'build-in' should be 'built-in'.
  3. [Fig. 2 caption] The rescaling/shifting of order parameters and susceptibilities is not described. State the exact transformation so the visual comparison is meaningful.
  4. [General] No data or code availability statement is included. For reproducibility, consider releasing the 30-feature datasets and the Wolfram scripts with the actual hyperparameters used.
  5. [Section IV] The Wolfram 'Automatic' hyperparameter selection is a black box; the authors should report the actual hyperparameters or the exact version of Mathematica functions used, since the footnote acknowledges this is unclear.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the ML transition signals are empirical benchmarks against externally known CDT phase structure, not derivations from their inputs.

full rationale

This paper is an empirical benchmark rather than a derivation of new physics: it trains supervised and unsupervised ML models on Monte Carlo data from known CDT phases, then tests whether they reproduce the known transition locations. The claimed transition points are produced by applying trained classifiers to unseen MC configurations and are compared with, not constructed from, the standard order-parameter results. The only overlap between the two sides is that OP1 = N0/N41 and OP2 = N32/N41 are functions of three of the 30 ML input features; this makes the baseline a subset of the model's input space, but the ML output is a learned probability, not an algebraic renaming of the OP, so the 'outperforming standard methods' statement is an empirical comparison rather than a circular identity. The cited CDT phase structure comes from earlier numerical CDT studies, several co-authored by the present authors, but those studies are external benchmarks with published evidence, not uniqueness theorems or fitted parameters in this paper. The Appendix's time-shift augmentation could create duplicate configurations across train/validation splits (a generalization/leakage risk), and the authors' footnote acknowledges uncertainty about Wolfram's automatic preprocessing; both are validity and transparency concerns, not circularity. No load-bearing step in the paper reduces by construction to its own input, so the circularity score is low.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

No new physical parameters, particles, or entities are introduced. The free choices are ML hyperparameters and cluster counts; the load-bearing assumptions are the correctness of the known CDT phase labels, the effective independence of MC samples, and the adequacy of the 30-feature set, which includes the standard order parameters that later serve as the comparison baseline.

free parameters (3)
  • Number of clusters for Agglomerate/K-Means/Spectral = 2
    Hand-set to two for these unsupervised models; the paper says they 'allowed the maximum number of clusters to be explicitly set to two' and this choice is essential to their success.
  • Manual hyperparameters for unsupervised models = not reported per transition
    MeanShift and other models required manual tuning; the tuning depends on the transition type and volume, and the exact settings are not given.
  • Wolfram Automatic ML hyperparameters = not reported
    The built-in Classify and ClusterClassify functions perform unspecified preprocessing and hyperparameter optimization; the authors admit they could not find this information in the documentation (footnote 1).
assumptions (5)
  • domain assumption CDT path integral with Regge action and fixed toroidal topology is a valid regularisation of quantum gravity
    Taken from the preceding CDT literature (refs 8-14); the ML analysis is performed on top of that model without testing this assumption.
  • domain assumption The known CDT phase diagram (A/B/C/Cb) and the manual phase labels are correct
    Training labels are assigned by 'manual classification' guided by the established phase diagram (Section III; refs 16-22). If these labels or the deepest-in-phase points are wrong, the supervised results are meaningless.
  • domain assumption Monte Carlo configurations are effectively independent for random train/validation splits
    The paper does not discuss autocorrelation, thinning, or binning; the >99.9% validation accuracy criterion implicitly assumes that random splits of the same runs do not leak correlated samples (Section III steps 3-5).
  • domain assumption The 30 selected geometric features, including N0, N32, N41, are sufficient to separate the phases
    The whole method depends on this feature choice; the feature list is in the Appendix and includes quantities from which the standard order parameters are built.
  • domain assumption Wolfram Mathematica 12 built-in ML implementations are reliable and their automatic preprocessing is adequate
    The authors note they could not determine what data preprocessing and hyperparameter optimization Classify/ClusterClassify perform automatically (footnote 1), so the reproducibility and interpretation rely on a proprietary tool.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Machine learning in phase transition analysis of lattice quantum gravity." pith.science (2026). https://pith.science/paper/QDVQHPDN

@misc{pith2026251002159,
  author       = {Pith},
  title        = {Pith review of: Machine learning in phase transition analysis of lattice quantum gravity},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QDVQHPDN}},
  note         = {Machine review of arXiv:2510.02159}
}
read the original abstract

Using numerical data coming from Monte Carlo simulations of four-dimensional Causal Dynamical Triangulations, we study how automated machine learning algorithms can be used to recognize transitions between different phases of quantum geometries observed in lattice quantum gravity. We tested seven supervised and seven unsupervised machine learning models and found that most of them were very successful in that task, even outperforming standard methods based on order parameters.

Figures

Figures reproduced from arXiv: 2510.02159 by the authors.

Figure 1
Figure 1. FIG. 1. The phase diagram of the toroidal CDT. Solid lines [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Summary of the results obtained by different ML [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Machine learning analysis of the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Machine learning analysis of the [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 11 linked inside Pith

  1. [1]

    Take all data measured for the highest value of∆ (∆max =−0.108), i.e., deepest inside phaseA

  2. [2]

    Take all data measured for the lowest value of∆ (∆min =−0.128), i.e., deepest inside phaseB

  3. [3]

    Split data from points 1 and 2 intotrainingand validationdatasets

  4. [4]

    Train the selected ML model using the training sets from points 1 and 2, this step also includes optimiz- ing hyperparameters of the ML model, if necessary

  5. [5]

    Check accuracy of the trained model using valida- tion sets from points 1 and 2

  6. [6]

    If the accuracy test was passed then use the trained ML model to classify / cluster other datasets mea- sured for∆between∆ min and∆ max, i.e., closer to the phase transition point than data from points 1 and 2

  7. [7]

    For each data pointd j(∆), j= 1, ...,dataset length, that comes from a dataset measured for a given value of∆compute the probability Pr(dj(∆)∈A)that the data belong to phaseA

  8. [8]

    Compute the mean value of the probability ⟨Pr(dj(∆)∈A)⟩measured for each dataset (∆) and its susceptibility (variance)

Show all 35 references
  1. [9]

    Find the transition point predicted by a given ML model∆ crit M L, where the probability⟨Pr(d j(∆)∈ A)⟩jumps from approximately0to approximately 1; see Fig. 2. That point is also usually associated with the jump in the measured susceptibility

  2. [10]

    manual” hyperparameter optimization; light green – the model correctly identifies phase transitions but requires “manual

    Compare∆ crit M Lwith∆ crit measured using standard CDT order parameters. IV. RESUL TS The results obtained for all tested ML models using datasets comprising 30 selected features measured in the CDT Monte Carlo simulations are summarized in Fig. 3. For each studied phase tran...

  3. [11]

    Ultraviolet Divergences in Quantum The- ories of Gravitation,

    S. Weinberg, “Ultraviolet Divergences in Quantum The- ories of Gravitation,” inGeneral Relativity: An Einstein Centenary Survey(Cambridge University Press, 1980) pp. 790–831

    1980

  4. [12]

    Codello, R

    A. Codello, R. Percacci, and C. Rahmede, Annals of Physics324, 414 (2009)

    2009

  5. [13]

    Reuter and F

    M. Reuter and F. Saueressig, New Journal of Physics14, 055022 (2012)

    2012

  6. [14]

    Reuter and F

    M. Reuter and F. Saueressig,Quantum Gravity and the Functional Renormalization Group: The Road towards Asymptotic Safety, Cambridge Monographs on Mathe- matical Physics (Cambridge University Press, 2019)

    2019

  7. [15]

    Ambjørn, A

    J. Ambjørn, A. Görlich, J. Jurkiewicz, A. Kreienbuehl, and R. Loll, Class. Quant. Grav.31, 165003 (2014), arXiv:1405.4585 [hep-th]

    2014 arXiv

  8. [16]

    Ambjorn, J

    J. Ambjorn, J. Gizbert-Studnicki, A. Görlich, J. Ju- rkiewicz, and R. Loll, Frontiers in PhysicsV olume 8 - 2020(2020), 10.3389/fphy.2020.00247

    2020

  9. [17]

    Ambjørn, J

    J. Ambjørn, J. Gizbert-Studnicki, A. Görlich, and D. Németh, Phys. Rev. D110, 126006 (2024)

    2024

  10. [18]

    Ambjorn, J

    J. Ambjorn, J. Jurkiewicz, and R. Loll, Phys. Rev. D 72, 064014 (2005), arXiv:hep-th/0505154

    2005 arXiv

  11. [19]

    Ambjørn, A

    J. Ambjørn, A. Görlich, S. Jordan, J. Jurkiewicz, and R. Loll, Phys. Lett. B690, 413 (2010), arXiv:1002.3298 [hep-th]

    2010 arXiv

  12. [20]

    Ambjørn, D

    J. Ambjørn, D. Coumbe, J. Gizbert-Studnicki, and J. Jurkiewicz, JHEP08, 033 (2015), arXiv:1503.08580 [hep-th]

    2015 arXiv

  13. [21]

    Ambjørn, J

    J. Ambjørn, J. Jurkiewicz, and R. Loll, Nucl. Phys. B 610, 347 (2001), arXiv:hep-th/0105267

    2001 arXiv

  14. [22]

    Ambjørn, A

    J. Ambjørn, A. Görlich, J. Jurkiewicz, and R. Loll, Phys. Rept.519, 127 (2012), arXiv:1203.3591 [hep-th]

    2012 arXiv

  15. [23]

    Loll, Class

    R. Loll, Class. Quant. Grav.37, 013002 (2020), arXiv:1905.08669 [hep-th]

    2020 arXiv

  16. [24]

    Jurkiewicz, and D

    J.Ambjørn, Z.Drogosz, J.Gizbert-Studnicki, A.Görlich, J. Jurkiewicz, and D. Németh, Universe7, 79 (2021), arXiv:2103.15610 [gr-qc]

    2021 arXiv

  17. [25]

    Regge, Nuovo Cim.19, 558 (1961)

    T. Regge, Nuovo Cim.19, 558 (1961)

    1961

  18. [26]

    Ambjørn, S

    J. Ambjørn, S. Jordan, J. Jurkiewicz, and R. Loll, Phys. Rev. D85, 124044 (2012)

    2012

  19. [27]

    Ambjørn, J

    J. Ambjørn, J. Gizbert-Studnicki, A. Görlich, J. Ju- rkiewicz, and D. Németh, JHEP06, 111 (2018), arXiv:1802.10434 [hep-th]

    2018 arXiv

  20. [28]

    Ambjørn, D

    J. Ambjørn, D. Coumbe, J. Gizbert-Studnicki, A. Gör- lich, and J. Jurkiewicz, Classical and Quantum Gravity 36, 224001 (2019)

    2019

  21. [29]

    Ambjørn, J

    J. Ambjørn, J. Gizbert-Studnicki, A. Görlich, and D. Németh, JHEP04, 103 (2022)

    2022

  22. [30]

    Ambjørn, S

    J. Ambjørn, S. Jordan, J. Jurkiewicz, and R. Loll, Phys. Rev. Lett.107, 211303 (2011)

    2011

  23. [31]

    Coumbe, J

    D. Coumbe, J. Gizbert-Studnicki, and J. Jurkiewicz, JHEP02, 144 (2016), arXiv:1510.08672 [hep-th]

    2016 arXiv

  24. [32]

    Ambjørn, G

    J. Ambjørn, G. Czelusta, J. Gizbert-Studnicki, A. Gör- lich, J. Jurkiewicz, and D. Németh, JHEP05, 030 (2020), arXiv:2002.01051 [hep-th]

    2020 arXiv

  25. [33]

    Classify,

    W. Research, “Classify,”https://reference.wolfram. com/language/ref/Classify.html(2025)

    2025

  26. [34]

    ClusterClassify,

    W. Research, “ClusterClassify,”https://reference. wolfram.com/language/ref/ClusterClassify.html (2025)

    2025

  27. [35]

    Machine Learning Methods,

    W. Research, “Machine Learning Methods,” https://reference.wolfram.com/language/guide/ MachineLearningMethods.html(2025)

    2025

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.