Skip to main content
Log in

General Limit Distributions for Sums of Random Variables with a Matrix Product Representation

  • Published:
Journal of Statistical Physics Aims and scope Submit manuscript

Abstract

The general limit distributions of the sum of random variables described by a finite matrix product ansatz are characterized. Using a mapping to a Hidden Markov Chain formalism, non-standard limit distributions are obtained, and related to a form of ergodicity breaking in the underlying non-homogeneous Hidden Markov Chain. The link between ergodicity and limit distributions is detailed and used to provide a full algorithmic characterization of the general limit distributions.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3

Similar content being viewed by others

Notes

  1. The definition (2) is valid for any probability space, however in the present article, we restrict ourselves to real random variables.

  2. Note that \(\fancyscript{P}_{ij}(x)\) is uniquely defined only when \(\fancyscript{E}_{ij} \ne 0\). When \(\fancyscript{E}_{ij} = 0\), the distribution \(\fancyscript{P}_{ij}(x)\) plays no role and can thus be any arbitrary distribution.

  3. http://perso.quaesituri.org/florian.angeletti/Softwares/Scientific.

  4. Without overflow

  5. More precisely \(K_\fancyscript{C}(s)\) is diffeomorph to the Cartesian product of a \(|l-1|\)-simplex and a \(|r-1|\)-simplex.

References

  1. Angeletti, F., Bertin, E., Abry, P.: Matrix products for the synthesis of stationary time series with a priori prescribed joint distributions. In: Proceeding of the IEEE International Conference on Acoustics Speech and Signal Processing (ICASSP), pp. 3897–3900, (2012)

  2. Angeletti, F., Bertin, E., Abry, P.: Random vector and time series definition and synthesis from matrix product representations: from statistical physics to hidden markov models. IEEE Trans. Signal Process. 61, 5389–5400 (2013).

  3. Angeletti, F., Bertin, E., Abry, P.: Statistics of sums of correlated variables described by a matrix product ansatz. Eur. Phys. Lett. 104, 50009 (2013)

  4. Angeletti, F., Touchette, H., Bertin, E., Abry, P.: Large deviations for correlated random variables described by a matrix product ansatz. J. Stat. Mech 2014, P02003 (2014)

  5. Antal, T., Droz, M., Györgyi, G., Rácz, Z.: “Roughness distributions for \(1/f^\alpha \) signals”. Phys. Rev. E 65, 046140 (2002)

  6. Bardou, F., Bouchaud, J.P., Aspect, A., Cohen-Tannoudji, C.: Lévy Statistics and Laser Cooling. Cambridge University Press, Cambridge (2002)

    MATH  Google Scholar 

  7. Basu, U., Mohanty, P.K.: Active-absorbing-state phase transition beyond directed percolation: a class of exactly solvable models. Phys. Rev. E 79, 041143 (2009)

  8. Bertin, E., Györgyi, G.: Renormalization flow in extreme value statistics. J. Stat. Mech. 2010, P08022 (2010)

  9. Blythe, R.A., Evans, M.R.: Nonequilibrium steady states of matrix-product form: a solver’s guide. J. Phys. A 40, R333–R441 (2007)

  10. Bouchaud, J.P.: Weak ergodicity breaking and aging in disordered systems. J. Phys. I (France) 2, 1705–1713 (1992)

  11. Bouchaud, J.P., Georges, A.: Anomalous diffusion in disordered media: statistical mechanisms, models and physical applications. Phys. Rep. 195, 127–293 (1990)

  12. Bramwell, S.T., Fortin, J.Y., Holdsworth, P.C.W., Peysson, S., Pinton, J.F., Portelli, B., Sellitto, M.: Magnetic fluctuations in the classical XY model: the origin of an exponential tail in a complex system. Phys. Rev. E 63, 041106 (2001)

  13. Cappe, O., Moulines, E., Ryden, T.: Inference in Hidden Markov Models. Springer Series in Statistics. Springer, New York (2005)

    MATH  Google Scholar 

  14. Clusel, M., Bertin, E.: Global fluctuations in physical systems: a subtle interplay between sum and extreme value statistics. Int. J. Mod. Phys. B 22, 3311–3368 (2008)

  15. Crampe, N., Ragoucy, E., Simon, D.: Matrix coordinate Bethe Ansatz: applications to XXZ and ASEP models. J. Phys. A 44, 405003 (2011)

  16. Derrida, B., Evans, M.R.: Exact correlation functions in an asymmetric exclusion model with open boundaries. J. Phys. I (France) 3, 311–322 (1993)

  17. Derrida, B., Evans, M.R., Hakim, V., Pasquier, V.: Exact solution of a 1D asymmetric exclusion model using a matrix formulation. J Phys. A 26, 1493–1517 (1993)

    Article  ADS  MATH  MathSciNet  Google Scholar 

  18. Essler, F.H.L., Rittenberg, V.: Representations of the quadratic algebra and partially asymmetric diffusion with open boundaries. J. Phys. A 29, 3375–3407 (1996)

  19. Feller, W.: An Introduction to Probability Theory and its Applications, vol. I. Wiley, New York (1966)

    MATH  Google Scholar 

  20. Feller, W.: An Introduction to Probability Theory and its Applications, vol. II. Wiley, New York (1966)

    MATH  Google Scholar 

  21. Ferrari, P.L., Sasamoto, T., Spohn, H.: Coupled Kardar-Parisi-Zhang equations in one dimension. J. Stat. Phys. 153, 377–399 (2013)

  22. Gnedenko, B.V., Kolmogorov, A.N.: Limit Distributions for Sums of Independent Random Variables. Addison-Wesley, Cambridge (1954)

    MATH  Google Scholar 

  23. Hakim, V., Nadal, J.P.: Exact results for 2D directed animals on a strip of finite width. J. Phys. A 16, L213–L218 (1983)

  24. Hieida, Y., Sasamoto, T.: Construction of a matrix product stationary state from solutions of a finite-size system. J. Phys. A 37, 9873–9889 (2004)

  25. Hinrichsen, H.: Non-equilibrium critical phenomena and phase transitions into absorbing states. Adv. Phys. 49, 815–918 (2000)

  26. Hinrichsen, H., Sandow, S., Peschel, I.: On matrix product ground states for reaction-diffusion models. J. Phys. A 29, 2643–2649 (1996)

  27. Jafarpour, F.H.: First-order phase transition in a reaction-diffusion model with open boundary: the Yang-Lee theory approach. J. Phys. A 36, 7497–7505 (2003)

  28. Jafarpour, F.H.: Matrix product states of three families of one-dimensional interacting particle systems. Physica A 339, 369–384 (2004)

  29. Lazarescu, A.: Matrix ansatz for the fluctuations of the current in the ASEP with open boundaries. J. Phys. A 46, 145003 (2013)

  30. Lazarescu, A., Mallick, K.: An exact formula for the statistics of the current in the TASEP with open boundaries. J. Phys. A 44, 315001 (2011)

  31. Mallick, K., Sandow, S.: Finite dimensional representations of the quadratic algebra: applications to the exclusion process. J. Phys. A 30, 4513–4526 (1997)

  32. Metzler, R., Klafter, J.: The random walk’s guide to anomalous diffusion: a fractional dynamics approach. Phys. Rep. 339, 1–77 (2000)

    Article  ADS  MATH  MathSciNet  Google Scholar 

  33. Petrov, V.V.: Limit Theorems of Probability Theory. Oxford University Press, Oxford (1995)

    MATH  Google Scholar 

  34. Prolhac, S., Evans, M.R., Mallick, K.: The matrix product solution of the multispecies partially asymmetric exclusion process. J. Phys. 42, 165004 (2009)

  35. Ramaswamy, S., Simha, R.A., Toner, J.: Active nematics on a substrate: Giant number fluctuations and long-time tails. Europhys. Lett. 62, 196–202 (2003)

  36. Rosenblatt, M.: Limit theorems for fourier transforms of functionals of gaussian sequences. Z. Wahrsch. Verw. Gebiete 55, 123–132 (1981)

  37. Seneta, E.: Non-negative Matrices and Markov Chains. Springer Series in Statistics. Springer, New York (2006)

    MATH  Google Scholar 

  38. Speer, E.R.: Finite-dimensional representations of a shock algebra. J. Stat. Phys 89, 169–175 (1997)

  39. Taqqu, M.S.: Convergence of integrated processes of arbitrary hermite rank. Z. Wahrsch. Verw. Gebiete 50, 53–83 (1979)

  40. Toner, J., Tu, Y., Ramaswamy, S.: Hydrodynamics and phases of flocks. Ann. Phys. (Amsterdam) 318, 170–240 (2005)

  41. Zeraati, S., Jafarpour, F.H., Hinrichsen, H.: Phase transition in an exactly solvable reaction-diffusion process. Phys. Rev. E 87, 062120 (2013)

Download references

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Florian Angeletti.

Appendix 1: A Periodic Irreducible Model

Appendix 1: A Periodic Irreducible Model

Periodic irreducible blocks \(\fancyscript{B}(c)\) correspond to the case where all the loops of the digraph \(\mathrm G(\fancyscript{B}(c))\) have a length which is a multiple of a base period \(P>1\):

$$\begin{aligned} (\fancyscript{B}(c)^{k})_{ii} > 0 \iff k \in P \mathbb Z. \end{aligned}$$
(125)

Consequently, it is possible to partition the indices \(\Xi _c\) in \(P\) subsets \(\Theta _{o}\) with \(o \in \mathbb Z/P\mathbb Z\) such that the edges of \(\mathrm G(\fancyscript{B}(c))\) only link indices from \(\Theta _{o}\) to \(\Theta _{o+1}\). The chain \(\Gamma ^{|c}\) cycles over the set \(\Theta _{o}\) with a period \(P\) and therefore does not converge to a stationary state. However, our aim is not to obtain a convergence result for the chain \(\Gamma ^{|c}\) but for the transition frequencies \(\nu _{ij}\). The transition frequencies \(\nu _{ij}\) are a global quantity that should not be influenced by the local periodic oscillation of \(\Gamma ^{|c}\). In particular, we can consider \(\Gamma ^{|c,o}\), the \(P\) subchains obtained by jumping over a period

$$\begin{aligned} \Gamma ^{|c,o} = \left( \Gamma ^{|c}_{o}, \dots , \Gamma ^{|c}_{o+kP}, \dots , \Gamma ^{|c}_{o + P \lfloor (N t_c-o) /P \rfloor } \right) \!, \end{aligned}$$
(126)

where \(\lfloor n\rfloor \) denotes the integer part of \(n\). The chain \(\Gamma ^{|c,o}\) corresponds to the Hidden Markov Chain of a matrix representation with structure matrix \(\fancyscript{B}(c)^P\):

$$\begin{aligned} P(\Gamma ^{|c,o}) \!=\! \frac{ \fancyscript{L}\left( \fancyscript{B}(c)^{o-1} \left[ \prod _{k=1}^{\lfloor (N t_c-o)/P\rfloor } (\fancyscript{B}(c)^P)_{{\Gamma ^{|c,o}}_{k} {\Gamma ^{|c,o}}_{k+1}} \right] \fancyscript{B}(c)^{N t_c - P \lfloor (N t_c-o)/P\rfloor - o +1 } \right) }{ \fancyscript{L}\left( \fancyscript{B}(c)^{ N t_c } \right) }. \end{aligned}$$
(127)

Moreover, if we call \(\nu \left( c,o\right) \) the transition frequencies of the subchain \(\Gamma ^{|c,o}\) then

$$\begin{aligned} \nu (c) = \frac{1}{P} \sum _{o=1}^{P} \nu \left( c,o\right) . \end{aligned}$$
(128)

The structure matrix \(\fancyscript{B}(c)^P\) is no longer periodic. If we relabel the indices of \(\fancyscript{B}(c)\) in order to make the \(\Theta _{o}\) contiguous, i.e. to ensure that \(\Theta _{o} = {s_o, s_0+1, \dots , d_o-1, d_o}\), then the matrix \(\fancyscript{B}(c)^P\) reads

$$\begin{aligned} \fancyscript{B}(c)^P = \left( \begin{array}{lll} D_{c,1} &{} &{} 0 \\ &{} \ddots &{} \\ 0 &{} &{} D_{c,P} \\ \end{array}\right) \end{aligned}$$
(129)

where \(D_{c,o}\) are irreducible aperiodic square matrices of size \(d_{c,o}\). The block diagonal structure of \(\fancyscript{B}(c)^P\) derives from the fact that after \(P\) jumps, the periodic chain \(\Gamma ^{|c}\) goes back to its original set \(\Theta _{o}\). For two indices \(o\ne o'\), there cannot be any transition between \(\Theta _{o}\) and \(\Theta _{o'}\) in the matrix \(\fancyscript{B}(c)^P\). In particular, if the final state of the subchain \(\Gamma ^{|c}_{N t_c}=f\) belongs to the set \(\Theta _{\omega }\) then for a non-zero probability subchain \(\Gamma ^{|c}\), the chain \(\Gamma ^{|c,o}\) stays inside the block \(\Theta _{\omega +o-N t_c}\):

$$\begin{aligned} \forall k,\quad {\Gamma ^{|c,o}}_k \in \Theta _{\omega +o-N t_c}. \end{aligned}$$
(130)

Taking in account the property Eq. (130), Eq. (127) simplifies to

$$\begin{aligned} P(\Gamma ^{|c,o}) = \frac{ \fancyscript{L}\left( \fancyscript{B}(c)^{o-1} \left[ \prod _{k=1}^{\lfloor (N t_c-o)/P\rfloor } (D_{c,\omega +o-N t_c})_{{\Gamma ^{|c,o}}_{k} {\Gamma ^{|c,o}}_{k+1}} \right] \fancyscript{B}(c)^{N t_c - P \lfloor (N t_c-o)/P\rfloor - o +1 } \right) }{\fancyscript{L}\left( \fancyscript{B}(c)^{ N t_c } \right) }. \end{aligned}$$
(131)

The subchain \(\Gamma ^{|c,o}\) therefore converges to the stationary state \( p_{\mathrm {st}}(c, \omega +o-N t_c) \) associated with the structure matrix \(D_{c, \omega +o-N t_c}\). As in the aperiodic case, the transition frequencies are therefore

$$\begin{aligned} \nu \left( c,o\right) _{ij} \overset{\text {a.s}}{\rightarrow } \frac{ \rho (c, \omega +o-N t_c)_i \, \fancyscript{B}(c)_{ij} \, \eta (c , \omega +o-N t_c)_j }{\Lambda } \end{aligned}$$
(132)

where \(\rho (c, o)\) and \(\eta (c,o)\) are respectively the left- and right-eigenvectors of the block \(D_{c,o}\) (embedded in the whole vector space of \(\fancyscript{B}(c)\)). Combining Eqs. (128) and (132) yields

$$\begin{aligned} \nu (c) \overset{\text {a.s}}{\rightarrow } \frac{1}{P} \sum _{o=1}^{P} \frac{ \rho (c,o)_i \fancyscript{B}(c)_{ij} \eta (c,o)_j }{\Lambda }. \end{aligned}$$
(133)

The left and right eigenvectors of \(\fancyscript{B}(c)\) associated with \(\Lambda \), respectively \(\eta (c)\) and \(\rho (c)\), are exactly

$$\begin{aligned} \eta (c) = \sum _{o=1}^P \eta (c,o),\end{aligned}$$
(134)
$$\begin{aligned} \rho (c) = \sum _{o=1}^P \rho (c,o). \end{aligned}$$
(135)

Moreover, the support of the eigenvectors \(\rho (c,o)\) and \(\eta (c , o' )\) are disjoint if \(o \ne o' \), consequently

$$\begin{aligned} \rho (c) \eta (c)^T = \frac{1}{P} \sum _{o, o'} \rho (c, o) \eta (c, o')^T = \frac{1}{P} \sum _{o=1}^P \rho (c,o)\eta (c,o)^T . \end{aligned}$$
(136)

Equation (133) therefore reads

$$\begin{aligned} \nu (c)_{i,j} \overset{\text {a.s}}{\rightarrow } \frac{ \rho (c)_i \fancyscript{B}(c)_{ij} \eta (c)_j}{\Lambda }. \end{aligned}$$
(137)

Hence the transition frequencies are exactly the same as the transition frequencies for the aperiodic case derived in Eq. (62).

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Angeletti, F., Bertin, E. & Abry, P. General Limit Distributions for Sums of Random Variables with a Matrix Product Representation. J Stat Phys 157, 1255–1283 (2014). https://doi.org/10.1007/s10955-014-1111-y

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10955-014-1111-y

Keywords

Navigation