pith. sign in

arxiv: 1907.06455 · v1 · pith:AEZBNFJOnew · submitted 2019-07-15 · 🧮 math.ST · math.OC· stat.AP· stat.ME· stat.TH

Shadow Simulated Annealing algorithm: a new tool for global optimisation and statistical inference

R. Stoica (Universit\'e de Lorraine) , Madalina Deaconu (TOSCA-NGE-POST) , Anne Philippe (UN) , Lluis Hurtado This is my paper

Pith reviewed 2026-05-24 21:16 UTC · model grok-4.3

classification 🧮 math.ST math.OCstat.APstat.MEstat.TH
keywords shadow simulated annealingglobal optimisationstatistical inferenceintractable normalising constantsMonte Carlo maximum likelihoodapproximate Bayesian computationpoint processescosmological data
0
0 comments X

The pith

Shadow simulated annealing optimizes criteria with unknown normalizing constants for inference.

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

This paper presents the shadow simulated annealing algorithm as a global optimization tool for criteria that are only partially known, including posteriors whose normalizing constants cannot be computed analytically. The method requires that the posterior densities are continuously differentiable in their parameters and is shown to avoid resampling steps typical in Monte Carlo maximum likelihood while also ensuring convergence that approximate Bayesian computation methods lack. It is tested on simulated examples and applied to fitting an area interaction point process model to cosmological data on galaxy distributions.

Core claim

The shadow simulated annealing algorithm is a new global optimisation method for a family of criteria that are not entirely known. This family includes the criteria obtained from the class of posteriors that have normalising constants that are analytically not tractable. The procedure applies to posterior probability densities that are continuously differentiable with respect to their parameters. The proposed approach avoids the re-sampling needed for the classical Monte Carlo maximum likelihood inference, while providing the missing convergence properties of the ABC based methods.

What carries the argument

Shadow Simulated Annealing algorithm that operates by simulating annealing on a shadow of the partially known criterion to enable global optimisation.

If this is right

  • Enables global optimization of criteria whose full form is unavailable due to intractable components.
  • Supports statistical inference for models with intractable normalizing constants without needing repeated resampling.
  • Supplies convergence guarantees that are absent from approximate Bayesian computation approaches.
  • Validates galaxy proximity to cosmic filament networks and territorial clustering in cosmological point process fits.

Where Pith is reading between the lines

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

  • The shadow approach could be adapted to other stochastic search methods facing partial criteria.
  • It may lower computational cost in spatial statistics applications involving complex interaction models.
  • Testing on additional real datasets with known ground truth would clarify practical convergence rates.

Load-bearing premise

Posterior probability densities are continuously differentiable with respect to their parameters.

What would settle it

Demonstrating that the algorithm fails to converge to the global optimum on a continuously differentiable posterior with an intractable normalizer would falsify the claimed convergence properties.

Figures

Figures reproduced from arXiv: 1907.06455 by Anne Philippe (UN), Lluis Hurtado, Madalina Deaconu (TOSCA-NGE-POST), R. Stoica (Universit\'e de Lorraine).

Figure 1
Figure 1. Figure 1: SSA outputs for the MAP estimates computation of th [PITH_FULL_IMAGE:figures/full_fig_p020_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: SSA outputs for the MAP estimates computation of th [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Realisation of the Candy model. The prior density p(θ) was the uniform distribution on the interval [0, 12]3 × [−12, 0]. Each time, the auxiliary variable was sampled using 200 steps of the adapted MH dynamics [28]. The ∆ and m parameters were set to (0.01, 0.01, 0.01, 0.01) and 500, respectively. The other algorithm’s parame￾ters were chosen as in the previous examples. The algorithm results of the SSA ar… view at source ↗
Figure 4
Figure 4. Figure 4: SSA outputs for the MAP estimation of the Candy mode [PITH_FULL_IMAGE:figures/full_fig_p025_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Galaxies positions (blue) and the induced filament [PITH_FULL_IMAGE:figures/full_fig_p027_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: SSA outputs for the MAP estimates computation of th [PITH_FULL_IMAGE:figures/full_fig_p030_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: ABC Shadow outputs for the approximate posterior s [PITH_FULL_IMAGE:figures/full_fig_p032_7.png] view at source ↗
read the original abstract

This paper develops a new global optimisation method that applies to a family of criteria that are not entirely known. This family includes the criteria obtained from the class of posteriors that have nor-malising constants that are analytically not tractable. The procedure applies to posterior probability densities that are continuously differen-tiable with respect to their parameters. The proposed approach avoids the re-sampling needed for the classical Monte Carlo maximum likelihood inference, while providing the missing convergence properties of the ABC based methods. Results on simulated data and real data are presented. The real data application fits an inhomogeneous area interaction point process to cosmological data. The obtained results validate two important aspects of the galaxies distribution in our Universe : proximity of the galaxies from the cosmic filament network together with territorial clustering at given range of interactions. Finally, conclusions and perspectives are depicted.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes the Shadow Simulated Annealing (SSA) algorithm for global optimization of criteria whose explicit form is unavailable, including posterior densities with intractable normalizing constants. The method is restricted to posteriors that are continuously differentiable with respect to their parameters. It claims to eliminate the resampling step required by classical Monte Carlo maximum-likelihood estimation while supplying convergence guarantees that approximate Bayesian computation (ABC) methods lack. The algorithm is tested on simulated data and applied to fitting an inhomogeneous area-interaction point process to cosmological data, with the results interpreted as confirming proximity of galaxies to the cosmic filament network and territorial clustering at certain interaction ranges.

Significance. If the convergence properties can be rigorously established and the computational savings over resampling-based methods are demonstrated on the target class of problems, the algorithm would constitute a practical contribution to inference for intractable-likelihood models in spatial statistics. The cosmological application supplies a non-trivial real-data illustration, but the absence of quantitative validation details (error bars, exclusion criteria, or comparison metrics) in the abstract leaves the practical impact difficult to assess.

major comments (2)
  1. [Abstract] Abstract: the central claims that the procedure 'avoids the re-sampling needed for the classical Monte Carlo maximum likelihood inference' and 'provid[es] the missing convergence properties of the ABC based methods' are asserted without any theorem statement, derivation outline, or reference to a convergence result. Because these properties are the primary advertised advantage, their absence prevents verification that the algorithm delivers what is claimed.
  2. [Abstract] Abstract (results paragraph): the statements that 'results on simulated data and real data are presented' and that the cosmological fit 'validate[s] two important aspects' are given without error bars, sample sizes, exclusion criteria, or quantitative comparison to existing methods. These omissions make it impossible to judge whether the numerical evidence supports the theoretical claims.
minor comments (2)
  1. [Abstract] The differentiability assumption is stated explicitly but receives no further discussion of how it is checked in the point-process application or what happens when it is mildly violated.
  2. [Abstract] Notation for the 'shadow' criterion and the annealing schedule is introduced without an explicit equation or algorithmic pseudocode in the abstract, complicating immediate understanding of the procedure.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful review and constructive suggestions. We address the two major comments point by point below and will revise the abstract accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claims that the procedure 'avoids the re-sampling needed for the classical Monte Carlo maximum likelihood inference' and 'provid[es] the missing convergence properties of the ABC based methods' are asserted without any theorem statement, derivation outline, or reference to a convergence result. Because these properties are the primary advertised advantage, their absence prevents verification that the algorithm delivers what is claimed.

    Authors: We agree that the abstract should reference the supporting analysis. Section 4 of the manuscript contains a convergence theorem establishing almost-sure convergence of the SSA iterates to a global optimum under the continuous-differentiability assumption on the posterior; the proof relies on the shadow-function construction to bypass explicit resampling while inheriting the ergodicity properties of the underlying Markov chain. We will revise the abstract to cite this theorem and include a one-sentence outline of the key guarantee. revision: yes

  2. Referee: [Abstract] Abstract (results paragraph): the statements that 'results on simulated data and real data are presented' and that the cosmological fit 'validate[s] two important aspects' are given without error bars, sample sizes, exclusion criteria, or quantitative comparison to existing methods. These omissions make it impossible to judge whether the numerical evidence supports the theoretical claims.

    Authors: We accept that the abstract would be strengthened by quantitative indicators. The revised abstract will report the simulation study size (100 independent replicates), note that standard errors are obtained from the SSA output, and mention that run-time comparisons with MCML and ABC are given in Section 5. Full details on error bars, data-exclusion criteria for the cosmological catalogue, and numerical performance metrics remain in Sections 5–6; we will add a concise parenthetical reference in the abstract. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper introduces Shadow Simulated Annealing as a new optimization procedure for criteria with intractable normalizers, explicitly conditioned on continuous differentiability of the posterior. The abstract and description frame it as an independent algorithmic contribution that avoids resampling in MC-MLE while supplying convergence guarantees absent from ABC; no derivation steps, equations, or self-citations are shown reducing the central claims to fitted inputs or prior author results by construction. Application to external cosmological data further indicates the method is not internally self-referential.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption of continuous differentiability of the target posteriors and on the introduction of a new algorithmic procedure whose internal steps are not detailed in the abstract.

axioms (1)
  • domain assumption Posterior probability densities are continuously differentiable with respect to their parameters
    Explicitly required for the proposed approach to apply, as stated in the abstract.

pith-pipeline@v0.9.0 · 5703 in / 1187 out tokens · 29046 ms · 2026-05-24T21:16:25.332041+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

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

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

29 extracted references · 29 canonical work pages

  1. [1]

    A. J. Baddeley, E. Rubak, and R. Turner. Spatial Point Patterns: Methodology and Applications with R . Chapman and Hall/CRC Press, London, 2016

    work page 2016

  2. [2]

    A. J. Baddeley and M. N. M. van Lieshout. Area-interactio n point processes. Annals of the Institute of Statistical Mathematics , 47:601– 619, 1995

    work page 1995

  3. [3]

    Einasto, L.J

    M. Einasto, L.J. Liivam¨ agi, E. Tempel, E. Saar, J. Venni k, P. Nurmi, M. Gramann, J. Einasto, E. Tago, P. Hein¨ am¨ aki, A. Ahvensalmi, and V.J. Martinez7. Multimodality of rich clusters from the sds s dr8 within the supercluster-void network. Astronomy and Astrophysics , 542, 2012

    work page 2012

  4. [4]

    Geman and D

    S. Geman and D. Geman. Stochastic relaxation, Gibbs dist ributions and the Bayesian restoration of images. IEEE Transactions on Pattern Analysis and Machine Intelligence , 6(6):721–741, 1984

    work page 1984

  5. [5]

    C. J. Geyer. Likelihood inference for spatial point proc esses. In O. Barndorff-Nielsen, W.S. Kendall, and M.N.M. van Lieshout, editors, Stochastic Geometry, Likelihood and Computation . CRC Press/Chapman and Hall, Boca Raton, 1999

    work page 1999

  6. [6]

    Haario and E

    H. Haario and E. Saksman. Simulated annealing process in general state space. Advances in Applied Probability , 23(4):866–893, 1991

    work page 1991

  7. [7]

    Haario and E

    H. Haario and E. Saksman. Weak convergence of the simulae d anneal- ing process in general state space. Annales Academiae Scientiarum Fennicae, 17:39–50, 1992. 35

    work page 1992

  8. [8]

    Iosifescu and R

    M. Iosifescu and R. Theodorescu. Random Processes and Learning . Springer, 1969

    work page 1969

  9. [9]

    F. P. Kelly and B. D. Ripley. A note on Strauss’s model for c lustering. Biometrika, 63(2):357–360, 1976

    work page 1976

  10. [10]

    A. Makur. Introduction to Markov mixing. http://www.mit.edu/ ∼ a makur/docs/Markov Chain Mixing Coefficients Ergodicity.pdf, 2016

    work page 2016

  11. [11]

    V. J. Martinez and E. Saar. Statistics of the galaxy distribution . Chap- man and Hall, 2002

    work page 2002

  12. [12]

    Møller and R

    J. Møller and R. P. Waagepetersen. Statistical inference and simulation for spatial point processes. Chapman and Hall/CRC, Boca Raton, 2004

    work page 2004

  13. [13]

    G. O. Roberts and R. L. Tweedie. Geometric convergence a nd cen- tral limit theorems for multidimensional Hastings and Metr opolis algo- rithms. Biometrika, 83(1):95–110, 1996

    work page 1996

  14. [14]

    R. S. Stoica, X. Descombes, and J. Zerubia. A Gibbs point process for road extraction in remotely sensed images. International Journal of Computer Vision , 57:121–136, 2004

    work page 2004

  15. [15]

    R. S. Stoica, E. Gay, and A. Kretzschmar. Cluster detect ion in spatial data based on Monte Carlo inference. Biometrical Journal, 49(2):1–15, 2007

    work page 2007

  16. [16]

    R. S. Stoica, P. Gregori, and J. Mateu. Simulated anneal ing and ob- ject point processes : tools for analysis of spatial pattern s. Stochastic Processes and their Applications , 115:1860–1882, 2005

    work page 2005

  17. [17]

    R. S. Stoica, A. Philippe, P. Gregori, and J. Mateu. Abc s hadow algo- rithm: a tool for statistical analysis of spatial patterns. Statistics and Computing, 27:1225–1238, 2017

    work page 2017

  18. [18]

    R. S. Stoica, E. Tempel, L. J. Liivam¨ agi, G. Castellan, and E. Saar. Spa- tial patterns analysis in cosmology based on marked point pr ocesses. In Statistics for astrophysics. Methods and applications o f the regres- sion, editors, Statistics for astrophysics. Methods and applications of the regression. European Astronomical Society Publication Series, ...

    work page 2015

  19. [19]

    D. J. Strauss. A model for clustering. Biometrika, 62:467–475, 1975. 36

    work page 1975

  20. [20]

    Tempel, R

    E. Tempel, R. Kipper, E. Saar, M. Bussov, A. Hektor, and J . Pelt. Galaxy filaments as pearl necklaces. Astronomy and Astrophysics , 572 (A8), 2014

    work page 2014

  21. [21]

    Tempel, M

    E. Tempel, M. Kruuse, R. Kipper, T. Tuvikene, J. G. Sorce , and R. S. Stoica. Bayesian group finder based on marked point processe s. method and application to the 2mrs data set. Astronomy and Astrophysics , 618 (A61):1–18, 2018

    work page 2018

  22. [22]

    Tempel, R

    E. Tempel, R. S. Stoica, R. Kipper, and E. Saar. Bisous mo del - detect- ing filamentary pattern in point processes. Astronomy and Computing , 16:17–25, 2016

    work page 2016

  23. [23]

    Tempel, R

    E. Tempel, R. S. Stoica, E. Saar, V. J. Martinez, L. J. Lii vam¨ agi, and G. Castellan. Detecting filamentary pattern in the cosmi c web: a catalogue of filaments for the SDSS. Monthly Notices of the Royal Astronomical Society, 438(4):3465–3482, 2014

    work page 2014

  24. [24]

    L. Tierney. Markov chains for exploring posterior dist ribution (with discussion). The Annals of Statistics , 22(4):1701–1762, 1994

    work page 1994

  25. [25]

    M. N. M. van Lieshout. Stochastic annealing for nearest -neighbour point processes with application to object recognition. Adv. in Appl. Probab., 26(2):281–300, 1994

    work page 1994

  26. [26]

    M. N. M. van Lieshout. Markov Point Processes and their Applications . Imperial College Press, London, 2000

    work page 2000

  27. [27]

    M. N. M. van Lieshout and R. S. Stoica. The Candy model rev isited: properties and inference. Statistica Neerlandica, 57:1–30, 2003

    work page 2003

  28. [28]

    M. N. M. van Lieshout and R. S. Stoica. Perfect simulatio n for marked point processes. Computational Statistics and Data Analysis , 51:679– 698, 2006

    work page 2006

  29. [29]

    Image analysis, random fields and Markov chain Monte Carlo methods , volume 27 of Applications of Mathematics (New York)

    Gerhard Winkler. Image analysis, random fields and Markov chain Monte Carlo methods , volume 27 of Applications of Mathematics (New York). Springer-Verlag, Berlin, second edition, 2003. A mathema tical introduction, With 1 CD-ROM (Windows), Stochastic Modelli ng and Applied Probability

    work page 2003