pith. sign in

arxiv: 2606.13453 · v1 · pith:ECY2JAEYnew · submitted 2026-06-11 · 🧮 math-ph · math.MP· stat.ML

Rapid mixing for Gibbs measures in Riemannian manifolds

\'Angela Capel , Marco Castrill\'on-L\'opez , Sofyan Iblisdir , Angelo Lucia , Pablo P\'aez-Velasco , David P\'erez-Garc\'ia This is my paper

Pith reviewed 2026-06-27 05:27 UTC · model grok-4.3

classification 🧮 math-ph math.MPstat.ML
keywords Langevin dynamicsRiemannian manifoldsGibbs measureslogarithmic Sobolev inequalityrapid mixingRiemannian submersionmixing times
0
0 comments X

The pith

Langevin dynamics on Riemannian manifolds achieves rapid mixing to the Gibbs measure under curvature and temperature conditions that yield a logarithmic Sobolev inequality.

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

The paper identifies conditions on manifold curvature, inverse temperature, escaping directions from saddle points, and the exclusion of barren plateaus and spurious local minima that ensure a logarithmic Sobolev inequality for the Langevin process. This inequality guarantees that the dynamics converges to the target Gibbs measure in time polynomial in the manifold dimension. The proof proceeds by establishing a relation between the Langevin processes on a manifold and on the image of a Riemannian submersion, which transfers the inequality. A sympathetic reader cares because this supplies a route to efficient high-dimensional sampling on curved spaces when the geometric and temperature hypotheses hold.

Core claim

When the manifold curvature satisfies suitable bounds, the inverse temperature lies in an appropriate regime, saddle points admit escaping directions, and barren plateaus together with spurious local minima are absent, the Langevin dynamics on the Riemannian manifold satisfies a logarithmic Sobolev inequality obtained via a Riemannian submersion relation and therefore mixes to the Gibbs measure in time polynomial in dimension.

What carries the argument

The relation between Langevin processes in the domain and in the image of a Riemannian submersion, which transfers the logarithmic Sobolev inequality from one to the other.

Load-bearing premise

The stated conditions on curvature, temperature, and saddle properties are sufficient to produce a logarithmic Sobolev inequality via the submersion relation.

What would settle it

A concrete Riemannian manifold and potential satisfying the curvature bounds, temperature regime, and saddle-escape properties for which the mixing time of the Langevin dynamics to the Gibbs measure grows superpolynomially in dimension.

Figures

Figures reproduced from arXiv: 2606.13453 by \'Angela Capel, Angelo Lucia, David P\'erez-Garc\'ia, Marco Castrill\'on-L\'opez, Pablo P\'aez-Velasco, Sofyan Iblisdir.

Figure 1
Figure 1. Figure 1: Sketch of the subsets defined in Eq. (2.20). We have used stripped filling to denote that some region is included in two subsets. The critical points are isolated by Assumption 2.14, after choosing a sufficiently small value for √a β . for every x ∈ U  √a β , S˜  . Furthermore, by the choice of β, it holds that √a β ≤ i(B) and so, as we saw in Remark 2.26 it holds without loss of generality that ∂U √a β… view at source ↗
Figure 2
Figure 2. Figure 2: Sketch describing the Markov triples, functions and Riemannian submersions [PITH_FULL_IMAGE:figures/full_fig_p061_2.png] view at source ↗
read the original abstract

Langevin dynamics on Riemannian manifolds is analyzed. Conditions ensuring the existence of a suitable logarithmic Sobolev inequality (rapid mixing to the Gibbs measure) are identified. These conditions involve the curvature of the manifold, the inverse temperature, escaping directions from saddle points, and exclude barren plateaus and spurious local minima. We show that when these conditions are met, mixing times polynomial in the dimension of the manifold are achievable. This result is obtained through a relation between Langevin processes in the domain and in the image of a Riemannian submersion. Such a relation can be of independent interest.

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

1 major / 1 minor

Summary. The manuscript analyzes Langevin dynamics on Riemannian manifolds and identifies conditions on manifold curvature, inverse temperature, escaping directions from saddle points, and the exclusion of barren plateaus and spurious local minima under which a logarithmic Sobolev inequality holds for the Gibbs measure. Rapid (dimension-polynomial) mixing then follows from a transfer relation between the Langevin process on the domain and its image under a Riemannian submersion.

Significance. If the stated conditions are rigorously shown to imply the LSI and the submersion transfer is verified, the result would supply a concrete set of geometric and temperature hypotheses guaranteeing polynomial mixing times on manifolds. The submersion relation itself may be reusable for other stochastic processes on quotient spaces and therefore carries independent technical value.

major comments (1)
  1. [Abstract / Main theorem statement] The central claim that the listed conditions suffice for an LSI (and hence for polynomial mixing) is load-bearing, yet the manuscript states the existence of such an inequality without supplying the derivation or the verification that the curvature, temperature, and saddle-escape hypotheses actually produce the required Poincaré or log-Sobolev constant.
minor comments (1)
  1. [Abstract] The abstract asserts that the submersion relation 'can be of independent interest' but provides no comparison with existing literature on stochastic processes under Riemannian submersions or any indication of other potential applications.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and for highlighting the need for greater clarity in the main theorem statement. We address the comment below.

read point-by-point responses

  1. Referee: [Abstract / Main theorem statement] The central claim that the listed conditions suffice for an LSI (and hence for polynomial mixing) is load-bearing, yet the manuscript states the existence of such an inequality without supplying the derivation or the verification that the curvature, temperature, and saddle-escape hypotheses actually produce the required Poincaré or log-Sobolev constant.

    Authors: The manuscript verifies the LSI under the stated hypotheses by combining the Bakry-Émery criterion (applied to the given curvature and inverse-temperature bounds) with the saddle-escape and barren-plateau assumptions to obtain an explicit dimension-polynomial constant; this is carried out in the proof of the main theorem via the Riemannian submersion transfer. The abstract is deliberately concise, but we agree it would benefit from an explicit pointer to this verification and will revise the theorem statement and abstract accordingly. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained

full rationale

The paper identifies conditions (curvature bounds, inverse temperature, saddle escape properties) claimed to ensure a logarithmic Sobolev inequality on the manifold, then transfers rapid mixing via a Riemannian submersion relation between domain and image processes. No quoted step reduces a prediction or LSI to a fitted parameter, self-definition, or self-citation chain; the submersion relation is explicitly noted as potentially independent. The abstract and reader's assessment show the central claim rests on external verification of the LSI conditions rather than internal reduction to inputs. This is the normal case of a self-contained argument against standard manifold probability benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract only; limited visibility into assumptions.

axioms (1)
  • domain assumption A logarithmic Sobolev inequality holds under the stated curvature and temperature conditions.
    Invoked to obtain rapid mixing; standard in analysis but taken as given here.

pith-pipeline@v0.9.1-grok · 5649 in / 1097 out tokens · 15522 ms · 2026-06-27T05:27:44.108618+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

97 extracted references · 6 canonical work pages

  1. [1]

    M. Creutz. Quarks, Gluons and Lattices . Cambridge Monographs on Mathematical Physics. Cambridge University Press, 2023

    2023

  2. [2]

    N. Boumal. An introduction to optimization on smooth manifolds . To appear with Cambridge University Press. 2022

    2022

  3. [3]

    Absil, R

    P.-A. Absil, R. Mahony, and R. Sepulchre. Optimization algorithms on matrix man- ifolds. Princeton University Press, 2008

    2008

  4. [4]

    Differential privacy dynamics of langevin dif- fusion and noisy gradient descent

    R. Chourasia, J. Ye, and R. Shokri. “Differential privacy dynamics of langevin dif- fusion and noisy gradient descent”. Advances in Neural Information Processing Sys- tems 34 (2021), pp. 14771–14781

    2021

  5. [5]

    Can stochastic gradient Langevin dynamics provide differ- ential privacy for deep learning?

    G. Heller and E. Fetaya. “Can stochastic gradient Langevin dynamics provide differ- ential privacy for deep learning?” 2023 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML). IEEE. 2023, pp. 68–106

    2023

  6. [6]

    Ryffel, F

    T. Ryffel, F. Bach, and D. Pointcheval. Differential Privacy Guarantees for Stochastic Gradient Langevin Dynamics. 2022. arXiv: 2201.11980

    work page arXiv 2022

  7. [7]

    Brooks, A

    S. Brooks, A. Gelman, G. Jones, and X.-L. Meng. Handbook of markov chain monte carlo. CRC press, 2011. 83

    2011

  8. [8]

    Gelman, J

    A. Gelman, J. B. Carlin, H. S. Stern, and D. B. Rubin. Bayesian data analysis . Chapman and Hall/CRC, 1995

    1995

  9. [9]

    D. J. MacKay. Information theory, inference and learning algorithms . Cambridge university press, 2003

    2003

  10. [10]

    C. P. Robert, G. Casella, and G. Casella. Monte Carlo statistical methods . Vol. 2. Springer, 2004

    2004

  11. [11]

    Efficient sampling on Riemannian manifolds via Langevin MCMC

    X. Cheng, J. Zhang, and S. Sra. “Efficient sampling on Riemannian manifolds via Langevin MCMC”. Advances in Neural Information Processing Systems 35 (2022), pp. 5995–6006

    2022

  12. [12]

    Bakry, I

    D. Bakry, I. Gentil, and M. Ledoux. Analysis and Geometry of Markov Diffusion Operators. Grundlehren der mathematischen Wissenschaften. Springer International Publishing, 2013

    2013

  13. [13]

    Poincar´ e and logarithmic Sobolev inequalities by decomposition of the energy landscape

    G. Menz and A. Schlichting. “Poincar´ e and logarithmic Sobolev inequalities by decomposition of the energy landscape”. The Annals of Probability 42.5 (2014), pp. 1809–1884

    2014

  14. [14]

    Analysis of langevin monte carlo from poincare to log-sobolev

    S. Chewi, M. A. Erdogdu, M. Li, R. Shen, and M. S. Zhang. “Analysis of langevin monte carlo from poincare to log-sobolev”. Foundations of Computational Mathe- matics 25.4 (2025), pp. 1345–1395

    2025

  15. [15]

    Rapid convergence of the unadjusted langevin algo- rithm: Isoperimetry suffices

    S. Vempala and A. Wibisono. “Rapid convergence of the unadjusted langevin algo- rithm: Isoperimetry suffices”. Advances in neural information processing systems 32 (2019)

    2019

  16. [16]

    Wibisono

    A. Wibisono. Proximal Langevin Algorithm: Rapid Convergence Under Isoperimetry

  17. [17]

    Riemannian Langevin algorithm for solving semidefinite programs

    M. Li and M. A. Erdogdu. “Riemannian Langevin algorithm for solving semidefinite programs”. Bernoulli 29.4 (2023), pp. 3093–3113

    2023

  18. [18]

    Li and M

    M. Li and M. A. Erdogdu. Supplement to ”Riemannian Langevin algorithm for solv- ing semidefinite programs”. 2023

    2023

  19. [19]

    S. Cao, R. Nissim, and S. Sheffield. Dynamical approach to area law for lattice Yang- Mills. 2025. arXiv: 2509.04688

    work page arXiv 2025

  20. [20]

    Matrix product states and projected entangled pair states: Concepts, symmetries, theorems

    J. I. Cirac, D. P´ erez-Garc´ ıa, N. Schuch, and F. Verstraete. “Matrix product states and projected entangled pair states: Concepts, symmetries, theorems”. Rev. Mod. Phys. 93 (4 2021), p. 045003

    2021

  21. [21]

    Montvay and G

    I. Montvay and G. M¨ unster.Quantum fields on a lattice. Cambridge University Press, 1994

    1994

  22. [22]

    Tensor network manifolds and Riemannian fundamental theorem for tensor networks

    P. P´ aez Velasco. “Tensor network manifolds and Riemannian fundamental theorem for tensor networks”. unpublished

  23. [23]

    E. Hsu. Stochastic Analysis on Manifolds . Contemporary Mathematics. American Mathematical Soc., 2002

    2002

  24. [24]

    Quotient-Space Dif- fusion Model

    Y. Xu, Y. Wang, S. Luo, K. Gao, T. He, C. Liu, and D. He. “Quotient-Space Dif- fusion Model”. Submitted to The Fourteenth International Conference on Learning Representations. Under review. 2026

    2026

  25. [25]

    G. Menon. The geometry of the deep linear network . 2024. arXiv: 2411.09004. 84

    work page arXiv 2024

  26. [26]

    Menon, A

    G. Menon, A. J. Stromme, and A. Vacher. On the implicit regularization of Langevin dynamics with projected noise . 2026. arXiv: 2602.12257

    work page arXiv 2026

  27. [27]

    Motion by mean curvature and Dyson Brownian Motion

    C.-P. Huang, D. Inauen, and G. Menon. “Motion by mean curvature and Dyson Brownian Motion”. Electronic Communications in Probability 28 (2023), pp. 1–10

    2023

  28. [28]

    Jolliffe

    I. Jolliffe. Principal component analysis. Springer, 2011, pp. 1094–1096

    2011

  29. [29]

    Graph embedding: A general frame- work for dimensionality reduction

    S. Yan, D. Xu, B. Zhang, and H.-J. Zhang. “Graph embedding: A general frame- work for dimensionality reduction”. 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05). Vol. 2. IEEE. 2005, pp. 830– 837

    2005

  30. [30]

    A geometric revisit to the trace quotient problem

    H. Shen, K. Diepold, and K. H¨ uper. “A geometric revisit to the trace quotient problem”. Proceedings of the 19th International Symposium of Mathematical Theory of Networks and Systems (MTNS 2010) . 2010, p. 1

    2010

  31. [31]

    The effects of mild over-parameterization on the optimization landscape of shallow relu neural networks

    I. M. Safran, G. Yehudai, and O. Shamir. “The effects of mild over-parameterization on the optimization landscape of shallow relu neural networks”. Conference on Learning Theory. PMLR. 2021, pp. 3889–3934

    2021

  32. [32]

    Semi-flat minima and saddle points by embedding neural networks to overparameterization

    K. Fukumizu, S. Yamaguchi, Y.-i. Mototake, and M. Tanaka. “Semi-flat minima and saddle points by embedding neural networks to overparameterization”. Advances in neural information processing systems 32 (2019)

    2019

  33. [33]

    Cheeger, D

    J. Cheeger, D. G. Ebin, and D. G. Ebin. Comparison theorems in Riemannian ge- ometry. Vol. 9. North-Holland Amsterdam, 1975

    1975

  34. [34]

    M. W. Hirsch. Differential topology. Springer Science & Business Media, 2012

    2012

  35. [35]

    F. Wang. Functional inequalities Markov semigroups and spectral theory . Elsevier, 2006

    2006

  36. [36]

    Revuz and M

    D. Revuz and M. Yor. Continuous martingales and Brownian motion . Vol. 293. Springer Science & Business Media, 2013

    2013

  37. [37]

    A simple proof of the Poincar´ e inequality for a large class of probability measures

    D. Bakry, F. Barthe, P. Cattiaux, and A. Guillin. “A simple proof of the Poincar´ e inequality for a large class of probability measures”. Electronic Communications in Probability 13 (2008), pp. 60–66

    2008

  38. [38]

    Bovier and F

    A. Bovier and F. den Hollander. Metastability: A Potential-Theoretic Approach. Die Grundlehren der mathematischen Wissenschaften. Springer International Publish- ing, 2015

    2015

  39. [39]

    J. M. Lee. Introduction to Riemannian manifolds . Vol. 2. Springer, 2018

    2018

  40. [40]

    Wainwright

    M. Wainwright. High-Dimensional Statistics: A Non-Asymptotic Viewpoint . Cam- bridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press, 2019

    2019

  41. [41]

    F.-Y. Wang. Analysis for Diffusion Processes on Riemannian Manifolds . World Sci- entific, 2013

    2013

  42. [42]

    A gradient estimate on a manifold with convex boundary

    Z. Qian. “A gradient estimate on a manifold with convex boundary”. Proceedings of the Royal Society of Edinburgh: Section A Mathematics 127.1 (1997), pp. 171–179

    1997

  43. [43]

    Sulanke and P

    R. Sulanke and P. Wintgen. Differentialgeometrie und Faserb¨ undel. Vol. 48. Springer, 1972

    1972

  44. [44]

    On the estimate of first eigenvalue of a compact Riemannian manifold

    J. Q. Zhong. “On the estimate of first eigenvalue of a compact Riemannian manifold”. Sci. Sinica Ser. A 27 (1984), pp. 1265–1273. 85

    1984

  45. [45]

    Weak logarithmic Sobolev inequalities and entropic convergence

    P. Cattiaux, I. Gentil, and A. Guillin. “Weak logarithmic Sobolev inequalities and entropic convergence”. Probability theory and related fields 139.3 (2007), pp. 563– 603

    2007

  46. [46]

    Logarithmic Sobolev inequalities and spectral gaps

    E. Carlen and M. Loss. “Logarithmic Sobolev inequalities and spectral gaps”. Con- temporary Mathematics 353 (2004), pp. 53–60

    2004

  47. [47]

    A. B. Tsybakov. Introduction to Nonparametric Estimation. Springer New York, NY, 2008

    2008

  48. [48]

    C. Villani. Optimal Transport: Old and New. Grundlehren der mathematischen Wis- senschaften. Springer Berlin Heidelberg, 2008

    2008

  49. [49]

    Riemannian submersions with totally geodesic fibers

    R. H. Escobales Jr. “Riemannian submersions with totally geodesic fibers”. Journal of Differential Geometry 10.2 (1975), pp. 253–276

    1975

  50. [50]

    Some isoperimetric inequalities and eigenvalue estimates

    C. B. Croke. “Some isoperimetric inequalities and eigenvalue estimates”. Annales scientifiques de l’ ´Ecole Normale Sup´ erieureSer. 4, 13.4 (1980), pp. 419–435

    1980

  51. [51]

    Riemannian geometry as determined by the volumes of small geodesic balls

    A. Gray and L. Vanhecke. “Riemannian geometry as determined by the volumes of small geodesic balls”. Acta Mathematica 142 (1979), pp. 157–198

    1979

  52. [52]

    Inequalities for the gamma function

    N. Batir. “Inequalities for the gamma function”. Archiv der Mathematik 91.6 (2008), pp. 554–563

    2008

  53. [53]

    Y´ ang¨ uez, T

    ´A. Y´ ang¨ uez, T. A. Hahn, and J. Kochanowski.Efficient Quantum Measurements: Computational Max- and Measured R´ enyi Divergences and Applications. 2025. arXiv: 2509.21308

    work page arXiv 2025

  54. [54]

    Density Matrix Renormalization Group and Periodic Boundary Conditions: A Quantum Information Perspective

    F. Verstraete, D. Porras, and J. I. Cirac. “Density Matrix Renormalization Group and Periodic Boundary Conditions: A Quantum Information Perspective”. Phys. Rev. Lett. 93 (22 2004), p. 227205

    2004

  55. [55]

    R. A. Horn and C. R. Johnson. Matrix analysis. Cambridge university press, 2012

    2012

  56. [56]

    A Grassmann manifold handbook: Basic geometry and computational aspects

    T. Bendokat, R. Zimmermann, and P.-A. Absil. “A Grassmann manifold handbook: Basic geometry and computational aspects”. Advances in Computational Mathemat- ics 50.1 (2024), p. 6

    2024

  57. [57]

    Polynomial-time approximation algorithms for the Ising model

    M. Jerrum and A. Sinclair. “Polynomial-time approximation algorithms for the Ising model”. SIAM Journal on computing 22.5 (1993), pp. 1087–1116

    1993

  58. [58]

    Petersen

    P. Petersen. Riemannian Geometry. Graduate Texts in Mathematics. Springer New York, 2006

    2006

  59. [59]

    Fulton and J

    W. Fulton and J. Harris. Representation theory: a first course . Springer Science & Business Media, 2013

    2013

  60. [60]

    J. Q. Gallier and J. Quaintance. Differential geometry and lie groups . Vol. 12. Springer, 2020

    2020

  61. [61]

    The Frobenius norm and the commutator

    A. B¨ ottcher and D. Wenzel. “The Frobenius norm and the commutator”. Linear Algebra and its Applications 429.8-9 (Oct. 2008), pp. 1864–1885

    2008

  62. [62]

    M. P. Do Carmo and J. Flaherty Francis. Riemannian geometry. Vol. 6. Springer, 1992

    1992

  63. [63]

    A. Besse. Einstein Manifolds. Classics in Mathematics. Springer Berlin Heidelberg, 2007

    2007

  64. [64]

    Contributions to Riemannian Geometry in the Large

    W. Klingenberg. “Contributions to Riemannian Geometry in the Large”. Annals of Mathematics 69.3 (1959), pp. 654–666. 86

    1959

  65. [65]

    M. Berger. A Panoramic View of Riemannian Geometry. Springer Berlin Heidelberg, 2007

    2007

  66. [66]

    The fundamental equations of a submersion

    B. O’Neill. “The fundamental equations of a submersion.” Michigan Mathematical Journal 13.4 (1966), pp. 459–469

    1966

  67. [67]

    Submersions and geodesics

    B. O’Neill. “Submersions and geodesics”. Duke Math. J. 34.1 (1967), pp. 363–373

    1967

  68. [68]

    The topology of fiber bundles lecture notes

    R. L. Cohen. “The topology of fiber bundles lecture notes”. Standford University (1998)

    1998

  69. [69]

    Sub-Riemannian geometry of Stiefel manifolds

    C. Autenried and I. Markina. “Sub-Riemannian geometry of Stiefel manifolds”. SIAM Journal on Control and Optimization 52.2 (2014), pp. 939–959

    2014

  70. [70]

    Algorithms for data fitting on some common homogeneous spaces

    Q. Rentmeesters et al. “Algorithms for data fitting on some common homogeneous spaces”. PhD thesis. Ph. D. thesis, Universit´ e Catholique de Louvain, Louvain, Bel- gium, 2013

    2013

  71. [71]

    The ultimate upper bound on the injectivity radius of the Stiefel manifold

    P.-A. Absil and S. Mataigne. “The ultimate upper bound on the injectivity radius of the Stiefel manifold”. SIAM Journal on Matrix Analysis and Applications 46.2 (2025), pp. 1145–1167

    2025

  72. [72]

    S. K. Lando and A. K. Zvonkin. Graphs on surfaces and their applications . Vol. 141. Springer Science & Business Media, 2013

    2013

  73. [73]

    Grigor’yan

    A. Grigor’yan. Lecture notes on Analysis on Manifolds . 2024

    2024

  74. [74]

    Log-Sobolev inequality on non-convex Riemannian manifolds

    F.-Y. Wang. “Log-Sobolev inequality on non-convex Riemannian manifolds”. Ad- vances in Mathematics 222.5 (2009), pp. 1503–1520

    2009

  75. [75]

    A brief introduction to Brownian motion on a Riemannian manifold

    E. P. Hsu. “A brief introduction to Brownian motion on a Riemannian manifold”. lecture notes (2008)

    2008

  76. [76]

    A theory of the term structure of interest rates

    J. C. Cox, J. E. Ingersoll, S. A. Ross, et al. “A theory of the term structure of interest rates”. Econometrica 53.2 (1985), pp. 385–407

    1985

  77. [77]

    Escaping time of a modified CIR process

    mathusername. Escaping time of a modified CIR process . Mathematics Stack Ex- change. 2025. https://math.stackexchange.com/q/5117199

    work page arXiv 2025

  78. [78]

    The Cox-Ingersoll-Ross process under volatility uncertainty

    B. Akhtari and H. Li. “The Cox-Ingersoll-Ross process under volatility uncertainty”. Journal of Mathematical Analysis and Applications 531.1 (2024), p. 127867

    2024

  79. [79]

    J. Steele. Stochastic Calculus and Financial Applications . Stochastic Modelling and Applied Probability. Springer New York, 2012

    2012

  80. [80]

    Oksendal

    B. Oksendal. Stochastic differential equations: an introduction with applications . Springer Science & Business Media, 2013

    2013

Showing first 80 references.