pith. sign in

arxiv: 2606.25603 · v1 · pith:NSZX4UX3new · submitted 2026-06-24 · 📊 stat.CO

Resampling in conditional SMC algorithms

Axel Finke , Adam M. Johansen , Anthony Lee , Lawrence M. Murray This is my paper

Pith reviewed 2026-06-25 19:31 UTC · model grok-4.3

classification 📊 stat.CO
keywords conditional sequential Monte Carloresamplingparticle Markov chain Monte Carlosystematic resamplingadaptive resamplingchopthin resampling
0
0 comments X

The pith

A simple framework implements valid conditional SMC resampling for schemes with complex dependence under weak assumptions only.

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

The paper introduces a framework for resampling steps inside sequential Monte Carlo and conditional sequential Monte Carlo algorithms. The framework handles most existing schemes, including those whose particle indices depend on one another in complicated ways, such as systematic resampling, adaptive resampling, and versions of chopthin resampling. It also shows how to obtain valid conditional versions of these schemes without extra random permutation or shifting of ancestor indices. Only weak assumptions are needed; marginal unbiasedness and exchangeability are not required. This matters for particle Markov chain Monte Carlo, where CSMC is used to build valid MCMC kernels and resampling choice affects practical performance.

Core claim

The authors present a simple framework for implementing valid SMC and CSMC algorithms. The framework covers most known resampling schemes including those with a complicated dependence structure like systematic resampling, but also adaptive resampling, and even more exotic schemes like a version of chopthin resampling. It explains how to implement conditional analogues of these and other well known resampling schemes without randomly permuting or shifting the order of the ancestor indices. The framework requires only very weak assumptions which include neither marginal unbiasedness nor exchangeability.

What carries the argument

A simple framework for implementing valid SMC and CSMC algorithms that handles resampling schemes with arbitrary dependence structures.

Load-bearing premise

The proposed framework produces valid conditional SMC algorithms for the listed resampling schemes when only the stated weak assumptions hold.

What would settle it

A concrete counter-example in which the framework is applied to systematic resampling inside CSMC on a toy model and the resulting Markov chain fails to leave the target distribution invariant.

read the original abstract

Conditional sequential Monte Carlo (CSMC) algorithms arise in particle Markov chain Monte Carlo and a number of related settings. As in standard sequential Monte Carlo (SMC) algorithms, it is possible to employ a number of approaches to resampling within CSMC, but some additional care is required to arrive at a valid algorithm. We present a simple framework for implementing valid SMC and CSMC algorithms which (a) covers most known resampling schemes including those with a complicated dependence structure like systematic resampling, but also adaptive resampling, and even more `exotic' schemes like a version of chopthin resampling; (b) explains how to implement conditional analogues of these and other well known resampling schemes without randomly permuting/shifting the order of the ancestor indices; (c) requires only very weak assumptions which include neither (marginal) "unbiasedness" nor exchangeability.

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 a general framework for resampling within both standard sequential Monte Carlo (SMC) and conditional SMC (CSMC) algorithms. The framework is claimed to cover a wide range of resampling schemes—including those with complex dependence structures such as systematic resampling, adaptive resampling, and variants of chopthin resampling—while requiring only weak measurability and consistency conditions that exclude both marginal unbiasedness and exchangeability. It further provides explicit constructions for the corresponding conditional kernels that preserve the correct conditional distribution of ancestor indices without random permutation or shifting.

Significance. If the explicit kernel constructions and consistency conditions hold as stated, the framework would unify and extend existing resampling methods for CSMC, enabling valid implementations of previously difficult schemes (e.g., systematic and adaptive) under weaker assumptions. This would be a useful methodological contribution for particle MCMC and related algorithms, particularly where exchangeability or unbiasedness cannot be assumed.

major comments (2)
  1. [§3.2, Definition 3.1] §3.2, Definition 3.1: the consistency condition (3) is stated only for the marginal resampling kernel; it is not immediately clear from the text whether the same condition suffices to guarantee the joint conditional distribution required for CSMC when the resampling map depends on the full particle system (as in systematic resampling).
  2. [§4.3, Proposition 4.2] §4.3, Proposition 4.2: the proof sketch invokes a measurability argument to avoid permutation; however, the argument appears to rely on the ancestor indices being ordered by the original particle labels, which may not hold for adaptive or chopthin schemes where the selection depends on auxiliary randomness. A concrete counter-example or additional lemma would strengthen the claim.
minor comments (2)
  1. [§2] Notation for the conditional kernel K_n is introduced in §2 but used interchangeably with the unconditional kernel in later sections; a consistent subscript or superscript would improve readability.
  2. [§5.1] The abstract mentions 'a version of chopthin resampling' but the precise variant implemented is only defined in §5.1; an earlier reference or footnote would help readers.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for the constructive comments. We address each major comment below and indicate the revisions we are prepared to make.

read point-by-point responses

  1. Referee: [§3.2, Definition 3.1] §3.2, Definition 3.1: the consistency condition (3) is stated only for the marginal resampling kernel; it is not immediately clear from the text whether the same condition suffices to guarantee the joint conditional distribution required for CSMC when the resampling map depends on the full particle system (as in systematic resampling).

    Authors: The consistency condition (3) is formulated at the level of the marginal kernel, but the framework constructs the joint conditional kernels in Section 4 so that the required conditional distribution for CSMC is preserved whenever the marginal condition holds. This construction applies directly to resampling maps that depend on the full particle system (including systematic resampling) because the weak measurability assumption is taken with respect to the sigma-algebra generated by the entire collection of particles and weights. The proofs for the specific schemes in the manuscript already verify the joint property. We will add a short clarifying paragraph after Definition 3.1 to make the extension from marginal to joint explicit. revision: partial

  2. Referee: [§4.3, Proposition 4.2] §4.3, Proposition 4.2: the proof sketch invokes a measurability argument to avoid permutation; however, the argument appears to rely on the ancestor indices being ordered by the original particle labels, which may not hold for adaptive or chopthin schemes where the selection depends on auxiliary randomness. A concrete counter-example or additional lemma would strengthen the claim.

    Authors: The measurability argument in the proof of Proposition 4.2 does not assume that ancestor indices are ordered by the original particle labels. It relies only on the resampling map being measurable with respect to the filtration that includes both the particle system and any auxiliary random variables used by the scheme. For adaptive and chopthin resampling the auxiliary randomness is folded into the definition of the kernel, so the conditional distribution of the ancestor indices remains correct without permutation or shifting. We will add a short lemma (placed before Proposition 4.2) that states the measurability condition explicitly for kernels that may depend on auxiliary variables, thereby covering the schemes mentioned. revision: yes

Circularity Check

0 steps flagged

No significant circularity; framework is self-contained

full rationale

The manuscript presents an explicit algorithmic framework for constructing valid conditional kernels for a broad class of resampling schemes (systematic, adaptive, chopthin, etc.) under measurability and consistency conditions that exclude both marginal unbiasedness and exchangeability. No derivation step reduces by construction to a fitted parameter, a self-citation chain, or a renamed empirical pattern; the central contribution consists of the stated weak assumptions plus the direct construction of the conditional ancestor-index distributions, which are shown to preserve the required conditional law without additional hidden requirements. The paper therefore supplies independent methodological content rather than a tautological re-expression of its inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; the contribution is a methodological framework whose internal assumptions are not detailed here.

pith-pipeline@v0.9.1-grok · 5672 in / 1073 out tokens · 39805 ms · 2026-06-25T19:31:25.317036+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

300 extracted references · 8 linked inside Pith

  1. [1]

    and Maciejowski, Jan and Chopin, Nicolas , JOURNAL =

    Kantas, Nikolas and Doucet, Arnaud and Singh, Sumeetpal S. and Maciejowski, Jan and Chopin, Nicolas , JOURNAL =. On particle methods for parameter estimation in state-space models , VOLUME =

  2. [2]

    IEEE Transactions on Signal Processing , volume=

    A basic convergence result for particle filtering , author=. IEEE Transactions on Signal Processing , volume=. 2008 , publisher=

    2008

  3. [3]

    The Annals of Statistics , volume=

    On resampling schemes for particle filters with weakly informative observations , author=. The Annals of Statistics , volume=. 2022 , publisher=

    2022

  4. [4]

    Journal of Computational and Graphical Statistics , number=

    Conditional particle filters with bridge backward sampling , author=. Journal of Computational and Graphical Statistics , number=. 2023 , publisher=

    2023

  5. [5]

    The Annals of Statistics , volume=

    Negative association, ordering and convergence of resampling methods , author=. The Annals of Statistics , volume=. 2019 , publisher=

    2019

  6. [6]

    Sequential

    Del Moral, Pierre and Doucet, Arnaud and Jasra, Ajay , JOURNAL =. Sequential

  7. [7]

    On adaptive resampling strategies for sequential

    Del Moral, Pierre and Doucet, Arnaud and Jasra, Ajay , JOURNAL =. On adaptive resampling strategies for sequential

  8. [8]

    Gordon, Neil James and Salmond, D. J. and Smith, A. F. M. , optiOPTIDOI =. Novel approach to nonlinear/non-

  9. [9]

    Del Moral, Pierre , LOCATION =

  10. [10]

    Mean Field Simulation for

    Del Moral, Pierre , LOCATION =. Mean Field Simulation for

  11. [11]

    Journal of the American Statistical Association , volume=

    Blind deconvolution via sequential imputations , author=. Journal of the American Statistical Association , volume=. 1995 , publisher=

    1995

  12. [12]

    and Rosenthal, Jeffrey S

    Roberts, Gareth O. and Rosenthal, Jeffrey S. , optiOPTIDOI =. General state space. Probability Surveys , PAGES =

  13. [13]

    Kalman, R. E. , JOURNAL =. A new approach to linear filtering and prediction problems , VOLUME =

  14. [14]

    Jazwinski, A. H. , EDITION =. Stochastic Processes and Filtering Theory , YEAR =

  15. [15]

    and Wong, Wing Hung , JOURNAL =

    Kong, Augustine and Liu, Jun S. and Wong, Wing Hung , JOURNAL =. Sequential imputations and

  16. [16]

    Genshiro Kitagawa , JOURNAL =

  17. [17]

    A Self-Organizing State-Space Model , VOLUME =

    Genshiro Kitagawa , JOURNAL =. A Self-Organizing State-Space Model , VOLUME =

  18. [18]

    Comparison of resampling schemes for particle filtering , YEAR =

    Douc, Randal and Cappé, Olivier and Moulines, Eric , BOOKTITLE =. Comparison of resampling schemes for particle filtering , YEAR =

  19. [19]

    and Chen, Rong , JOURNAL =

    Liu, Jun S. and Chen, Rong , JOURNAL =. Sequential

  20. [20]

    A note on auxiliary particle filters , VOLUME =

    Adam Michael Johansen and Arnaud Doucet , JOURNAL =. A note on auxiliary particle filters , VOLUME =

  21. [21]

    An improved particle filter for nonlinear problems , VOLUME =

    Carpenter, James and Clifford, Peter and Fearnhead, Paul , ISSN =. An improved particle filter for nonlinear problems , VOLUME =

  22. [22]

    Particle

    Andrieu, Christophe and Doucet, Arnaud and Holenstein, Roman , optiOPTIDOI =. Particle. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , NOTE =

  23. [23]

    Efficient block sampling strategies for sequential

    Doucet, Arnaud and Briers, Mark and Sénécal, Stéphane , JOURNAL =. Efficient block sampling strategies for sequential

  24. [24]

    , LOCATION =

    Liu, Jun S. , LOCATION =

  25. [25]

    Weighting a resampled particle in Sequential

    Martino, Luca and Elvira, Victor and Louzada, Francisco , booktitle=. Weighting a resampled particle in Sequential. 2016 , organization=

    2016

  26. [26]

    Rubin, D. B. , booktitle=. Using the. 1988 , organization=

    1988

  27. [27]

    Contribution to the discussion on `The calculation of posterior distributions by data augmentation' by Tanner, M. A. and Wong, W. H. , author=. Journal of the American Statistical Association , volume=. 1987 , publisher=

    1987

  28. [28]

    A Tutorial on Particle Filtering and Smoothing: Fifteen Years Later , YEAR =

    Doucet, Arnaud and Johansen, Adam Michael , BOOKTITLE =. A Tutorial on Particle Filtering and Smoothing: Fifteen Years Later , YEAR =

  29. [29]

    Dynamic Programming , YEAR =

    Bellman, Richard Ernest , LOCATION =. Dynamic Programming , YEAR =

  30. [30]

    Inference in Hidden

    Cappé, Olivier and Moulines, Eric and Rydén, Tobias , ISBN =. Inference in Hidden

  31. [31]

    and Andrieu, Christophe , optiOPTIDOI =

    Doucet, Arnaud and Godsill, Simon J. and Andrieu, Christophe , optiOPTIDOI =. On sequential. Statistics and Computing , NUMBER =

  32. [32]

    Bayesian inference in econometric models using

    Geweke, John , JOURNAL =. Bayesian inference in econometric models using

  33. [33]

    Smoothing algorithms for state-space models , VOLUME =

    Briers, Mark and Doucet, Arnaud and Maskell, Simon , JOURNAL =. Smoothing algorithms for state-space models , VOLUME =

  34. [34]

    and Papaspiliopoulos, Omiros , optiOPTIDOI =

    Chopin, Nicolas and Jacob, Pierre E. and Papaspiliopoulos, Omiros , optiOPTIDOI =. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , NUMBER =

  35. [35]

    and Doucet, Arnaud and Singh, Sumeetpal S

    Poyiadjis, G. and Doucet, Arnaud and Singh, Sumeetpal S. , JOURNAL =. Particle approximations of the score and observed information matrix in state space models with application to parameter estimation , VOLUME =

  36. [36]

    and Tadic, V

    Andrieu, Christophe and Doucet, Arnaud and Singh, Sumeetpal S. and Tadic, V. B. , JOURNAL =. Particle methods for change detection, system identification, and control , VOLUME =

  37. [37]

    Singh , BOOKTITLE =

    George Poyiadjis and Arnaud Doucet and Sumeetpal S. Singh , BOOKTITLE =. Maximum likelihood parameter estimation in General State-Space Models using Particle Methods , YEAR =

  38. [38]

    A sequential particle filter method for static models , VOLUME =

    Nicolas Chopin , JOURNAL =. A sequential particle filter method for static models , VOLUME =

  39. [39]

    Bernoulli , author=

    On the role of interaction in sequential. Bernoulli , author=

  40. [40]

    Nonlinear Filtering using Random Particles , VOLUME =

    Del Moral, Pierre , FCONLY =. Nonlinear Filtering using Random Particles , VOLUME =. Theory of Probability & Its Applications , NUMBER =

  41. [41]

    Nonlinear filtering: interacting particle solution , VOLUME =

    Del Moral, Pierre , JOURNAL =. Nonlinear filtering: interacting particle solution , VOLUME =

  42. [42]

    On the utility of graphics cards to perform massively parallel simulation of advanced

    Lee, Anthony and Yau, Christopher and Giles, Michael B and Doucet, Arnaud and Holmes, Christopher C , JOURNAL =. On the utility of graphics cards to perform massively parallel simulation of advanced

  43. [43]

    Backward simulation methods for

    Lindsten, Fredrik and Schön, Thomas B , JOURNAL =. Backward simulation methods for

  44. [44]

    and Schön, Thomas B

    Lindsten, Fredrik and Jordan, Michael I. and Schön, Thomas B. , JOURNAL =. Particle

  45. [45]

    Filtering via simulation: Auxiliary particle filters , VOLUME =

    Pitt, Michael K and Shephard, Neil , JOURNAL =. Filtering via simulation: Auxiliary particle filters , VOLUME =

  46. [46]

    1997 , publisher=

    Higuchi, Tomoyuki , journal=. 1997 , publisher=

    1997

  47. [47]

    1995 , institution=

    On the resampling scheme in the filtering procedure of the Kitagawa Monte Carlo filter , author=. 1995 , institution=

    1995

  48. [48]

    Stewart, Leland and McCarty Jr, Perry , BOOKTITLE =. Use of

  49. [49]

    Particle filtering for partially observed

    Andrieu, Christophe and Doucet, Arnaud , JOURNAL =. Particle filtering for partially observed

  50. [50]

    A Stable Particle Filter in High-Dimensions , journal =

  51. [51]

    Central limit theorem for nonlinear filtering and interacting particle systems , YEAR =

    Del Moral, Pierre and Guionnet, Alice , JOURNAL =. Central limit theorem for nonlinear filtering and interacting particle systems , YEAR =

  52. [52]

    On the stability of interacting processes with applications to filtering and genetic algorithms , VOLUME =

    Del Moral, Pierre and Guionnet, Alice , JOURNAL =. On the stability of interacting processes with applications to filtering and genetic algorithms , VOLUME =

  53. [53]

    Limit Theorems for Weighted Samples with Applications to Sequential

    Douc, Randal and Moulines, Eric , JOURNAL =. Limit Theorems for Weighted Samples with Applications to Sequential

  54. [54]

    and Doucet, Arnaud and West, Mike , JOURNAL =

    Godsill, Simon J. and Doucet, Arnaud and West, Mike , JOURNAL =

  55. [55]

    Branching and interacting particle systems approximations of

    Del Moral, Pierre and Miclo, Laurent , BOOKTITLE =. Branching and interacting particle systems approximations of

  56. [56]

    On the Forward Filtering Backward Smoothing particle approximations of the smoothing distribution in general state spaces models , journal =

  57. [57]

    Douc, Randal and Garivier, Aurélien and Moulines, Eric and Olsson, Jimmy , JOURNAL =

  58. [58]

    Sequential

    Olsson, Jimmy and Cappé, Olivier and Douc, Randal and Moulines, Eric , JOURNAL =. Sequential

  59. [59]

    Can local particle filters beat the curse of dimensionality? , VOLUME =

    Rebeschini, Patrick and van Handel, Ramon , JOURNAL =. Can local particle filters beat the curse of dimensionality? , VOLUME =

  60. [60]

    Comparison theorems for

    Rebeschini, Patrick and van Handel, Ramon , JOURNAL =. Comparison theorems for

  61. [61]

    ArXiv e-prints , archivePrefix = "arXiv", eprint =

    Forward Smoothing using Sequential. ArXiv e-prints , archivePrefix = "arXiv", eprint =

  62. [62]

    and Petrie, Ted and Soules, George and Weiss, Norman , JOURNAL =

    Baum, Leonard E. and Petrie, Ted and Soules, George and Weiss, Norman , JOURNAL =. A maximization technique occurring in the statistical analysis of probabilistic functions of

  63. [63]

    and Striebel, C

    Rauch, Herbert E. and Striebel, C. T. and Tung, F. , JOURNAL =. Maximum likelihood estimates of linear dynamic systems , VOLUME =

  64. [64]

    ArXiv e-prints , archivePrefix = "arXiv", eprint =

    Efficient particle-based online smoothing in general hidden. ArXiv e-prints , archivePrefix = "arXiv", eprint =

  65. [65]

    Equilibrium free-energy differences from nonequilibrium measurements: A master-equation approach , VOLUME =

    Jarzynski, Christopher , JOURNAL =. Equilibrium free-energy differences from nonequilibrium measurements: A master-equation approach , VOLUME =

  66. [66]

    Nonequilibrium equality for free energy differences , VOLUME =

    Jarzynski, Christopher , JOURNAL =. Nonequilibrium equality for free energy differences , VOLUME =

  67. [67]

    and Lindsten, Fredrik and Schön, Thomas B

    Naesseth, Christian A. and Lindsten, Fredrik and Schön, Thomas B. , BOOKTITLE =. Nested sequential

  68. [68]

    The factored frontier algorithm for approximate inference in

    Murphy, Kevin and Weiss, Yair , BOOKTITLE =. The factored frontier algorithm for approximate inference in

  69. [69]

    arXiv:1511.06286 , primaryClass = "stat.CO", keywords =

    The iterated auxiliary particle filter , journal =. arXiv:1511.06286 , primaryClass = "stat.CO", keywords =

    Pith/arXiv arXiv

  70. [70]

    , JOURNAL =

    Del Moral, Pierre and Doucet, Arnaud and Singh, Sumeetpal S. , JOURNAL =. A backward particle interpretation of

  71. [71]

    Dirk Eddelbuettel and Romain François , JOURNAL =

  72. [72]

    Seamless

    Dirk Eddelbuettel , LOCATION =. Seamless

  73. [73]

    Eddelbuettel, Dirk and Sanderson, Conrad , JOURNAL =

  74. [74]

    Sanderson, Conrad and Curtin, Ryan , JOURNAL =

  75. [75]

    Long-term stability of sequential

    Douc, Randal and Moulines, Eric and Olsson, Jimmy , JOURNAL =. Long-term stability of sequential. 2014 , PUBLISHER =

    2014

  76. [76]

    The Annals of Applied Probability , VOLUME=

    Stability properties of some particle filters , AUTHOR=. The Annals of Applied Probability , VOLUME=. 2013 , PUBLISHER=

    2013

  77. [77]

    Towards automatic model comparison: an adaptive sequential

    Zhou, Yan and Johansen, Adam M and Aston, John AD , journal=. Towards automatic model comparison: an adaptive sequential. 2016 , publisher=

    2016

  78. [78]

    Drovandi, Christopher C and McCutchan, Roy A , journal=. Alive. 2015 , publisher=

    2015

  79. [79]

    Interacting sequential

    Jasra, Ajay and Doucet, Arnaud and Stephens, David A and Holmes, Christopher C , journal=. Interacting sequential. 2008 , publisher=

    2008

  80. [80]

    Density-Tempered Marginalized Sequential

    Duan, Jin-Chuan and Fulop, Andras , journal=. Density-Tempered Marginalized Sequential. 2015 , publisher=

    2015

Showing first 80 references.