pith. sign in

arxiv: 1907.02456 · v1 · pith:C2ZSPLKYnew · submitted 2019-07-04 · 🧮 math.PR

Precise large deviation asymptotics for products of random matrices

Hui Xiao , Ion Grama , Quansheng Liu This is my paper

Pith reviewed 2026-05-25 09:00 UTC · model grok-4.3

classification 🧮 math.PR
keywords large deviationsrandom matrix productsLyapunov exponentlocal limit theoremoperator normmultiplicative ergodic theorempositive matrices
0
0 comments X

The pith

Precise asymptotics hold for the probability that the norm of a random matrix product exceeds its typical Lyapunov growth by a vanishing amount.

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

The paper establishes exact asymptotic expressions for the large-deviation probability that log of the norm of the product G_n of i.i.d. random matrices exceeds n times (q plus a small l), where q lies above the top Lyapunov exponent γ and l tends to zero with n. The same asymptotics are obtained for the joint law of the normalized direction X_n^x and the log-norm. The results apply both to invertible matrices and to positive matrices. They directly improve earlier large-deviation principles for the operator norm of G_n and produce a precise local limit theorem that accounts for the large-deviation regime.

Core claim

For an i.i.d. sequence of d by d real random matrices possessing top Lyapunov exponent γ, the probability P(log |G_n x| ≥ n(q + l)) with q > γ fixed and l vanishing as n → ∞ admits precise asymptotics, and parallel formulas hold for the pair (X_n^x, log |G_n x|) with X_n^x = G_n x / |G_n x|; the same statements are proved for positive matrices.

What carries the argument

The large-deviation rate function of the top Lyapunov exponent together with a renewal argument that tracks the small overshoot l when the deviation window shrinks with n.

If this is right

  • The large-deviation principle for the operator norm ||G_n|| holds with an explicit rate function and sharper prefactors.
  • A local limit theorem for the distribution of (X_n^x, log |G_n x|) remains valid inside the large-deviation zone q > γ.
  • The same asymptotic statements apply uniformly to every starting vector x on the unit sphere.
  • The results for positive matrices are obtained by the same argument used for invertible matrices.

Where Pith is reading between the lines

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

  • The same renewal technique may produce analogous asymptotics when the deviation l shrinks at any polynomial rate in 1/n.
  • The joint local-limit statement could be used to obtain moderate-deviation principles between the central-limit and large-deviation regimes.
  • Applications to the growth of norms in random linear dynamical systems become quantitative once the prefactors are known explicitly.

Load-bearing premise

The matrices are i.i.d. and satisfy the moment and irreducibility conditions that guarantee a finite top Lyapunov exponent γ.

What would settle it

Numerical Monte-Carlo estimation, for a concrete matrix distribution such as i.i.d. Gaussian entries, of the probability P(log |G_n x| ≥ n(q + l)) for large n and successively smaller l, compared against the explicit asymptotic formula given by the paper.

read the original abstract

Let $(g_{n})_{n\geq 1}$ be a sequence of independent identically distributed $d\times d$ real random matrices with Lyapunov exponent $\gamma$. For any starting point $x$ on the unit sphere in $\mathbb R^d$, we deal with the norm $ | G_n x | $, where $G_{n}:=g_{n} \ldots g_{1}$. The goal of this paper is to establish precise asymptotics for large deviation probabilities $\mathbb P(\log | G_n x | \geq n(q+l))$, where $q>\gamma $ is fixed and $l$ is vanishing as $n\to \infty$. We study both invertible matrices and positive matrices and give analogous results for the couple $(X_n^x,\log | G_n x |)$ with target functions, where $X_n^x= G_n x /| G_n x |$. As applications we improve previous results on the large deviation principle for the matrix norm $\|G_n\|$ and obtain a precise local limit theorem with large deviations.

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 establishes precise large-deviation asymptotics for P(log |G_n x| ≥ n(q + l)) with q > γ fixed and l → 0 as n → ∞, for the norm of products G_n of i.i.d. d × d random matrices acting on a unit vector x. Separate statements are given for the invertible and positive-matrix cases; analogous results are derived for the joint process (X_n^x, log |G_n x|) where X_n^x is the normalized direction. Applications include an improved large-deviation principle for ||G_n|| and a precise local limit theorem under large deviations.

Significance. If the derivations hold under the stated moment and irreducibility hypotheses, the results supply sharper tail asymptotics than the standard LDP for random matrix products. Such refinements are useful for quantitative estimates in random dynamical systems and for deriving local limit theorems with large-deviation corrections. The paper supplies explicit error terms and handles both the invertible and positive settings, which broadens applicability.

major comments (2)
  1. [§3, Theorem 3.2] §3, Theorem 3.2: the claimed asymptotic equivalence P(log |G_n x| ≥ n(q + l)) ∼ C(x,q) exp(−n I(q)) l^{α−1} (or the precise form given) relies on the existence of a spectral gap for the transfer operator on a suitable Banach space; the proof sketch invokes this gap but does not verify the required Doeblin-type condition or the moment assumption (E[||g||^p] < ∞ for p > 1) explicitly for the positive-matrix case.
  2. [§4.1, Eq. (4.3)] §4.1, Eq. (4.3): the local limit theorem with large deviations is stated for the couple (X_n^x, log |G_n x|), yet the error term O(l^β) with β > 0 is derived only under the additional assumption that the distribution of log |g x| is non-lattice; this non-lattice condition is used without being listed among the standing hypotheses of the main theorems.
minor comments (2)
  1. [§2.1] Notation: the symbol γ is introduced as the top Lyapunov exponent in the abstract but is redefined in §2.1 without cross-reference; a single consistent definition would improve readability.
  2. [Figure 1] Figure 1: the caption does not specify the matrix dimension d or the distribution used for the numerical illustration, making it difficult to reproduce the plotted curves.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and the positive recommendation. We respond to each major comment below.

read point-by-point responses

  1. Referee: [§3, Theorem 3.2] §3, Theorem 3.2: the claimed asymptotic equivalence P(log |G_n x| ≥ n(q + l)) ∼ C(x,q) exp(−n I(q)) l^{α−1} (or the precise form given) relies on the existence of a spectral gap for the transfer operator on a suitable Banach space; the proof sketch invokes this gap but does not verify the required Doeblin-type condition or the moment assumption (E[||g||^p] < ∞ for p > 1) explicitly for the positive-matrix case.

    Authors: We agree that an explicit verification strengthens the presentation. The moment condition E[||g||^p] < ∞ for p > 1 is already part of the standing hypotheses in Section 2 for both the invertible and positive cases. For positive matrices, the Doeblin condition follows directly from the strict positivity, irreducibility, and the moment assumption via standard arguments for positive transfer operators (as in the references cited in §2). We will add a short clarifying paragraph in the revised §3 making this verification explicit. revision: yes

  2. Referee: [§4.1, Eq. (4.3)] §4.1, Eq. (4.3): the local limit theorem with large deviations is stated for the couple (X_n^x, log |G_n x|), yet the error term O(l^β) with β > 0 is derived only under the additional assumption that the distribution of log |g x| is non-lattice; this non-lattice condition is used without being listed among the standing hypotheses of the main theorems.

    Authors: The referee is correct; the non-lattice assumption on the law of log |g x| is required for the stated error term but was omitted from the hypotheses of the local limit theorem in §4.1. We will add it explicitly to the standing assumptions for that result in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper derives precise large-deviation asymptotics for P(log |G_n x| ≥ n(q + l)) directly from the i.i.d. assumption on the matrices, the existence of the top Lyapunov exponent γ, and standard moment/irreducibility conditions. These inputs are external to the claimed asymptotics; the results for the norm, the couple (X_n^x, log |G_n x|), the improved LDP for ||G_n||, and the local limit theorem are presented as consequences rather than reductions by definition, fitted parameters renamed as predictions, or load-bearing self-citations. The derivation chain is self-contained within the probabilistic framework for random matrix products.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the existence of the Lyapunov exponent and standard regularity conditions from multiplicative ergodic theory; no free parameters or invented entities are mentioned.

axioms (1)
  • domain assumption Existence of a finite top Lyapunov exponent γ for the i.i.d. matrix sequence
    Explicitly invoked in the abstract as the baseline growth rate.

pith-pipeline@v0.9.0 · 5709 in / 1112 out tokens · 34168 ms · 2026-05-25T09:00:29.438726+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

36 extracted references · 36 canonical work pages · 1 internal anchor

  1. [1]

    R., Rao R

    Bahadur R. R., Rao R. R.: On deviations of the sample mean. The Annals of Mathe- matical Statistics, 31(4), 1015-1027, 1960

    work page 1960

  2. [2]

    F.: Central limit theorem for linear groups

    Benoist Y., Quint J. F.: Central limit theorem for linear groups. The Annals of Prob- ability, 44(2), 1308-1340, 2016. 32 HUI XIAO 1, ION GRAMA 1,2, AND QUANSHENG LIU 1

    work page 2016

  3. [3]

    F.: Random walks on reductive groups

    Benoist Y., Quint J. F.: Random walks on reductive groups . Springer International Publishing, 2016

    work page 2016

  4. [4]

    A., Borovkov K

    Borovkov A. A., Borovkov K. A.: Asymptotic analysis of ra ndom walks. Cambridge University Press , 2008

    work page 2008

  5. [5]

    Birkhäuser Boston, 1985

    Bougerol P., Lacroix J.: Products of random matrices wit h applications to Schrödinger operators. Birkhäuser Boston, 1985

    work page 1985

  6. [6]

    Probability Theory and Related Fields , 132(1): 13-38, 2005

    Breuillard E.: Distributions diophantiennes et théorè me limite local sur Rd. Probability Theory and Related Fields , 132(1): 13-38, 2005

    work page 2005

  7. [7]

    Journal of Difference Equations and Applications , 20(11), 1523-1567, 2014

    Buraczewski D., Damek E., Guivarc’h Y., Mentemeier S.: O n multidimensional Man- delbrot cascades. Journal of Difference Equations and Applications , 20(11), 1523-1567, 2014

    work page 2014

  8. [8]

    Annales de l’Institut Henri Poincaré, Probabilités et Stat istiques

    Buraczewski D., Mentemeier S.: Precise large deviation results for products of random matrices. Annales de l’Institut Henri Poincaré, Probabilités et Stat istiques. Vol. 52, No. 3, 1474-1513, 2016

    work page 2016

  9. [9]

    The Annals of Probability 44(6), 3688-3739, 2016

    Buraczewski D., Collamore J., Damek E., Zienkiewicz J.: Large deviation estimates for exceedance times of perpetuity sequences and their dual processes. The Annals of Probability 44(6), 3688-3739, 2016

    work page 2016

  10. [10]

    E.: Saddlepoint approximations in statisti cs

    Daniels H. E.: Saddlepoint approximations in statisti cs. The Annals of Mathematical Statistics, 631-650, 1954

    work page 1954

  11. [11]

    Springer Sci- ence and Business Media , 2009

    Dembo A., Zeitouni O.: Large deviations techniques and applications. Springer Sci- ence and Business Media , 2009

    work page 2009

  12. [12]

    V.: Asymptotic, Integrals and Series, Nauk a, 1987 (in Russian)

    Fedoryuk M. V.: Asymptotic, Integrals and Series, Nauk a, 1987 (in Russian)

    work page 1987

  13. [13]

    Handbook of dynamical systems, 1, 931-1014, 2002

    Furman A.: Random walks on groups and random transforma tions. Handbook of dynamical systems, 1, 931-1014, 2002

    work page 2002

  14. [14]

    Transactions of the American Mathematical Society, 108(3), 377-428, 1963

    Furstenberg H., Noncommuting random products. Transactions of the American Mathematical Society, 108(3), 377-428, 1963

    work page 1963

  15. [15]

    The Annals of Mathemat- ical Statistics , 31(2), 457-469, 1960

    Furstenberg H., Kesten H.: Products of random matrices . The Annals of Mathemat- ical Statistics , 31(2), 457-469, 1960

    work page 1960

  16. [16]

    V.: On a local limit theorem of the theory of p robability

    Gnedenko B. V.: On a local limit theorem of the theory of p robability. Uspekhi Matematicheskikh Nauk , 3(3):187-194, 1948

    work page 1948

  17. [17]

    Y., Guivarc’h Y.: Zariski closure and the d imension of the Gaussian law of the product of random matrices

    Goldsheid I. Y., Guivarc’h Y.: Zariski closure and the d imension of the Gaussian law of the product of random matrices. Probability Theory and Related Fields , 105(1), 109-142, 1996

    work page 1996

  18. [18]

    Conditioned local limit theorems for random walks defined on finite Markov chains

    Grama I., Lauvergnat R., Le Page É.: Conditioned local l imit theorems for random walks defined on finite Markov chains. arXiv preprint arXiv:1707.06129, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  19. [19]

    In International Conference on Modern Problems of Stochast ic Analysis and Statistics

    Grama I., Le Page É.: Bounds in the local limit theorem fo r a random walk condi- tioned to stay positive. In International Conference on Modern Problems of Stochast ic Analysis and Statistics. 103-127, Springer 2017

    work page 2017

  20. [20]

    Guivarc’h Y.: Spectral gap properties and limit theore ms for some random walks and dynamical systems. Proc. Sympos. Pure Math . 89, 279-310, 2015

    work page 2015

  21. [21]

    Annales de l’Institut Henri Poincaré, Probabilités et Statistiques

    Guivarc’h Y., Le Page É.: Spectral gap properties for li near random walks and Pareto’s asymptotics for affine stochastic recursions. Annales de l’Institut Henri Poincaré, Probabilités et Statistiques . Vol. 52. No. 2, 503-574, 2016

    work page 2016

  22. [22]

    Probability Theory and Related Fields, 69(2): 187-242, 1985

    Guivarc’h Y., Raugi A.: Frontiere de Furstenberg, prop riétés de contraction et théorèmes de convergence. Probability Theory and Related Fields, 69(2): 187-242, 1985

    work page 1985

  23. [23]

    Studia Math

    Guivarc’h Y., Urban R.: Semigroup actions on tori and st ationary measures on pro- jective spaces. Studia Math . 171, no. 1, 33-66, 2005

    work page 2005

  24. [24]

    The Annals of Probability, 25(4): 1545-1587, 1997

    Hennion H.: Limit theorems for products of positive ran dom matrices. The Annals of Probability, 25(4): 1545-1587, 1997. LARGE DEVIATIONS FOR PRODUCTS OF RANDOM MATRICES 33

    work page 1997

  25. [25]

    Hennion H., Hervé L.: Limit theorems for Markov chains a nd stochastic properties of dynamical systems by quasi-compactness. Vol. 1766, Lecture Notes in Mathematics . Springer-Verlag, Berlin, 2001

    work page 2001

  26. [26]

    The Annals of Probability , 32: 1934-1984, 2004

    Hennion H., Hervé L.: Central limit theorems for iterat ed random Lipschitz map- pings. The Annals of Probability , 32: 1934-1984, 2004

    work page 1934

  27. [27]

    Wolters, Noordhoff Pub., 1975

    Ibragimov I.A., Linnik Yu.V.: Independent and station ary sequences of random vari- ables. Wolters, Noordhoff Pub., 1975

    work page 1975

  28. [28]

    Acta Mathematica, vol

    Kesten H.: Random difference equations and renewal theo ry for products of random matrices. Acta Mathematica, vol. 131(1): 207-248, 1973

    work page 1973

  29. [29]

    Kingman J. F. C.: Subadditive ergodic theory. The Annals of Probability , 883-899, 1973

    work page 1973

  30. [30]

    In Probability measures on groups

    Le Page É.: Théorèmes limites pour les produits de matri ces aléatoires. In Probability measures on groups . Springer Berlin Heidelberg, 258-303, 1982

    work page 1982

  31. [31]

    V.: On the probabilities of large deviations f or sums of independent random variables

    Petrov V. V.: On the probabilities of large deviations f or sums of independent random variables. Theory of Probability and its Applications , 10(2): 287-298, 1965

    work page 1965

  32. [32]

    V.: Sums of independent random variables

    Petrov V. V.: Sums of independent random variables. Springer, 1975

    work page 1975

  33. [33]

    Theory of Probability and its Applications, 2(2): 206-220, 1957

    Richter W.: Local limit theorems for large deviations. Theory of Probability and its Applications, 2(2): 206-220, 1957

    work page 1957

  34. [34]

    A.: A local limit theorem

    Sheep L. A.: A local limit theorem. The Annals of Mathematical Statistics , 35: 419- 423, 1964

    work page 1964

  35. [35]

    The Annals of Mathematical Statistics , 36(2): 546-551, 1965

    Stone C.: A local limit theorem for nonlattice multi-di mensional distribution func- tions. The Annals of Mathematical Statistics , 36(2): 546-551, 1965

    work page 1965

  36. [36]

    Xiao H., Grama I., Liu Q.: Berry-Esseen bound and precis e moderate deviations for products of random matrices, Submitted, 2019

    work page 2019