Skip to main content

Aggregation and Truncation of Reversible Markov Chains Modulo State Renaming

  • Conference paper
  • First Online:
Analytical and Stochastic Modelling Techniques and Applications (ASMTA 2017)

Part of the book series: Lecture Notes in Computer Science ((LNPSE,volume 10378))

  • 506 Accesses

Abstract

The theory of time-reversibility has been widely used to derive the expressions of the invariant measures and, consequently, of the equilibrium distributions for a large class of Markov chains which found applications in optimisation problems, computer science, physics, and bioinformatics. One of the key-properties of reversible models is that the truncation of a reversible Markov chain is still reversible. In this work we consider a more general notion of reversibility, i.e., the reversibility modulo state renaming, called \(\rho \)-reversibility, and show that some of the properties of reversible chains cannot be straightforwardly extended to \(\rho \)-reversible ones. Among these properties, we show that in general the truncation of the state space of a \(\rho \)-reversible chain is not \(\rho \)-reversible. Hence, we derive further conditions that allow the formulation of the well-known properties of reversible chains for \(\rho \)-reversible Markov chains. Finally, we study the properties of the state aggregation in \(\rho \)-reversible chains and prove that there always exists a state aggregation that associates a \(\rho \)-reversible process with a reversible one.

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

Access this chapter

Institutional subscriptions

References

  1. Akyildiz, I.F.: Exact analysis of queueing networks with rejection blocking. In: Perros, H.G., Atliok, T. (eds.) Proceedings of the 1st International Workshop on Queueing Networks with Blocking, pp. 19–29 (1989)

    Google Scholar 

  2. Balsamo, S., Marin, A.: Separable solutions for Markov processes in random environments. Eur. J. Oper. Res. 229(2), 391–403 (2013)

    Article  MathSciNet  Google Scholar 

  3. Balsamo, S., Dei Rossi, G., Marin, A.: Lumping and reversed processes in cooperating automata. Ann. Oper. Res. 239, 695–722 (2014)

    Article  MathSciNet  Google Scholar 

  4. Benaim, M., Le Boudec, J.-Y.: A class of mean field interaction models for computer and communication systems. Perform. Eval. 65(11–12), 823–838 (2008)

    Article  Google Scholar 

  5. Block, R., Van Houdt, B.: Spatial fairness in multi-channel CSMA line networks. In: Proceedings of the 8th International Conference on Performance Evaluation Methodologies and Tools, VALUETOOLS, pp. 1–8 (2014)

    Google Scholar 

  6. Buchholz, P.: Product form approximations for communicating Markov processes. Perform. Eval. 67(9), 797–815 (2010). Special Issue: QEST 2008

    Article  Google Scholar 

  7. Bujari, A., Marin, A., Palazzi, C.E., Rossi, S.: Analysis of ECN/RED and SAP-LAW with simultaneous TCP and UDP traffic. Comput. Netw. 108, 160–170 (2016)

    Article  Google Scholar 

  8. Comert, G.: Queue length estimation from probe vehicles at isolated intersections: estimators for primary parameters. Eur. J. Oper. Res. 252(2), 502–521 (2016)

    Article  MathSciNet  Google Scholar 

  9. Gates, D.J.: Growth and decrescence of two-dimensional crystals: a Markov rate process. J. Stat. Phys. 52(1/2), 245–257 (1988)

    Article  MathSciNet  Google Scholar 

  10. Gates, D.J., Westcott, M.: Kinetics of polymer crystallization I. Discrete and continuum models. Proc. R. Soc. Lond. 416, 443–461 (1988)

    Article  MathSciNet  Google Scholar 

  11. Gates, D.J., Westcott, M.: Markovian models of steady crystal growth. J. Appl. Prob. 3(2), 339–355 (1993)

    Article  Google Scholar 

  12. Gelenbe, E., Labed, A.: G-Networks with multiple classes of signals and positive customers. Eur. J. Oper. Res. 48(5), 293–305 (1998)

    Article  Google Scholar 

  13. Kelly, F.: Reversibility and Stochastic Networks. Wiley, New York (1979)

    MATH  Google Scholar 

  14. Kelly, F.: Loss networks. Ann. Appl. Probab. 1(3), 319–378 (1991)

    Article  MathSciNet  Google Scholar 

  15. Kemeny, J.G., Snell, J.L.: Finite Markov Chains. Springer, Heidelberg (1976)

    MATH  Google Scholar 

  16. Kozlowski, D., Worthington, D.: Use of queue modelling in the analysis of elective patient treatment governed by a maximum waiting time policy. Eur. J. Oper. Res. 244(1), 331–338 (2015)

    Article  MathSciNet  Google Scholar 

  17. Latouche, G.: Queues with paired customers. J. Appl. Probab. 18(3), 684–696 (1981)

    Article  MathSciNet  Google Scholar 

  18. Marin, A., Rossi, S.: On discrete time reversibility modulo state renaming and its applications. In: Proceedings of the 8th International Conference on Performance Evaluation Methodologies and Tools, VALUETOOLS, pp. 1–8 (2014)

    Google Scholar 

  19. Marin, A., Rossi, S.: On the relations between lumpability and reversibility. In: Proceedings of the IEEE 22nd International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems (MASCOTS 2014), pp. 427–432 (2014)

    Google Scholar 

  20. Marin, A., Rossi, S.: Dynamic control of the join-queue lengths in saturated fork-join stations. In: Agha, G., Houdt, B. (eds.) QEST 2016. LNCS, vol. 9826, pp. 123–138. Springer, Cham (2016). doi:10.1007/978-3-319-43425-4_8

    Chapter  MATH  Google Scholar 

  21. Marin, A., Rossi, S.: On the relations between Markov chain lumpability and reversibility. Acta Inform., 1–39 (2016). doi:10.1007/s00236-016-0266-1

    Article  MathSciNet  Google Scholar 

  22. Pan, R., Prabhakar, B., Psounis, K.: CHOKe, a stateless active queue management scheme for approximating fair bandwidth allocation. In: Proceedings of IEEE INFOCOM 2000, pp. 942–951. IEEE Computer Society Press, Washington, DC (2000)

    Google Scholar 

  23. Roy, D., Krishnamurthy, A., Heragu, S.S., Malmborg, C.J.: Queuing models to analyze dwell-point and cross-aisle location in autonomous vehicle-based warehouse systems. Eur. J. Oper. Res. 242(1), 72–87 (2015)

    Article  Google Scholar 

  24. Shone, R., Knight, V.A., Williams, J.E.: Comparisons between observable and unobservable M/M/1 queues with respect to optimal customer behavior. Eur. J. Oper. Res. 227(1), 133–141 (2013)

    Article  MathSciNet  Google Scholar 

  25. Whittle, P.: Systems in Stochastic Equilibrium. Wiley, New York (1986)

    MATH  Google Scholar 

Download references

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Andrea Marin .

Editor information

Editors and Affiliations

A Proofs of the Results

A Proofs of the Results

1.1 Proof of Theorem 1

Proof

By Proposition 1 and Definition 1, to prove that \(\widetilde{X}(t)\) is \(\widetilde{\rho }\)-reversible it is sufficient to show that for all \(S_i,S_j\in \mathcal S/{\sim }\) with \(i\not =j\),

$$ \widetilde{\pi }(S_i) \widetilde{q}(S_i,S_j) =\widetilde{\pi }({\widetilde{\rho }(S_j)}) \widetilde{q}({\widetilde{\rho }(S_j)},\widetilde{\rho }(S_i)). $$

By Eq. (3) and Proposition 4, this is equivalent to:

figure a

which can be written as:

$$\begin{aligned} \sum _{s\in S_i}\sum _{s'\in S_j}\pi (s)q(s,s')= \sum _{s'\in \widetilde{\rho }(S_j)}\sum _{s\in \widetilde{\rho }(S_i)}\pi (s')q(s',s). \end{aligned}$$
(7)

We now proceed by considering four cases.

  1. 1.

    Assume that \(S_i=\{s\}\) and \(S_j=\{s'\}\), then Eq. (7) becomes:

    $$ \pi (s)q(s,s')=\pi ({\rho (s')})q({\rho (s')},\rho (s)), $$

    where we have used the definition of \(\widetilde{\rho }\) for singletons. This is true since by hypothesis X(t) is \(\rho \)-reversible and hence satisfies the \(\rho \)-detailed balance equation.

  2. 2.

    Assume \(S_i=\{s\}\) and \(|S_j|>1\), and recall that \(\widetilde{\rho }(S_j)=S_j\) by definition. Then Eq. (7) can be rewritten as:

    $$ \sum _{s' \in S_j}\pi (s)q(s,s')=\sum _{s' \in S_j} \pi (s')q(s',\rho (s)). $$

    Since \(\sim \) respects \(\rho \) we have that \(\rho \) restricted to the elements of \(S_j\) is still a bijection and hence we can write:

    $$ \sum _{s' \in S_j}\pi (s)q(s,s')=\sum _{s' \in S_j} \pi ({\rho (s')})q({\rho (s')},\rho (s)), $$

    which is true by the hypothesis of \(\rho \)-reversibility of X(t).

  3. 3.

    Assume \(|S_i|>1\) and hence \(\widetilde{\rho }(S_i)=S_i\) and \(S_j=\{s'\}\), then Eq. (7) can be written as:

    $$ \sum _{s \in S_i}\pi (s)q(s,s')=\sum _{s \in S_i}\pi ({\rho (s')})q({\rho (s')},s). $$

    Since \(\rho \) restricted to the elements of \(S_i\) is a bijection, then we have:

    $$ \sum _{s \in S_i}\pi (s)q(s,s')=\sum _{s \in S_i}\pi ({\rho (s')})q({\rho (s')},\rho (s)), $$

    which is an identity.

  4. 4.

    Assume \(|S_i|>1\) and \(|S_j|>1\), and hence \(\widetilde{\rho }(S_i)=S_i\) and \(\widetilde{\rho }(S_j)=S_j\). Then we can rewrite Eq. (7) as:

    $$ \sum _{s \in S_i} \sum _{s' \in S_j}\pi (s)q(s,s') = \sum _{s \in S_i}\sum _{s' \in S_j} \pi (s')q(s',s). $$

    Since \(\rho \) restricted to \(S_i\) and to \(S_j\) is still a bijection because \(\sim \) respects \(\rho \), we can rewrite the previous equation as:

    $$ \sum _{s \in S_i} \sum _{s' \in S_j}\pi (s)q(s,s') = \sum _{s \in S_i}\sum _{s' \in S_j} \pi ({\rho (s')})q({\rho (s')},\rho (s)). $$

    which is true by hypothesis.    \(\square \)

1.2 Proof or Proposition 5

Proof

By the general aggregation Eq. (3), for any \(S_i, S_j \in \mathcal S/{\sim }\),

$$\begin{aligned} \widetilde{q}(S_i,S_j)=\frac{\sum _{s'\in S_i}\pi (s')\sum _{s\in S_j}q(s',s)}{\sum _{s'\in S_i}\pi (s')}, \end{aligned}$$
(8)

Since X(t) is \(\rho \)-reversible and each \(S_i\in S/{\sim }\) is an orbit for \(\rho \), it holds that \(\pi (s)=\pi (s')\) for all \(s,s'\in S_i\). Let us denote by \(\pi (S_i)\) the equilibrium probability of each s belonging to the orbit \(S_i\). Hence, \(\sum _{s'\in S_i}\pi (s')=|S_i|\pi (S_i)\) and Eq. (8) can be written

$$\begin{aligned} \widetilde{q}(S_i,S_j)=\pi (S_i)\frac{\sum _{s'\in S_i}\sum _{s\in S_j}q(s',s)}{|S_i|\pi (S_i)} \end{aligned}$$
(9)

proving the statement.    \(\square \)

1.3 Proof of Lemma 1

Proof

To prove the lemma we use Proposition 1. In fact, let us consider two states \(s,u \in \mathcal A\), then the corresponding \(\rho \)-detailed balance equation is \(B \pi (s) q(s,u)=B\pi ({\rho (u)})q({\rho (u)},\rho (s))\) since we have by assumption that the partition respects \(\rho \) and hence also \(\rho (t),\rho (s) \in \mathcal A\). This equation is satisfied because X(t) is \(\rho \)-reversible. If \(s,u \in \mathcal S\smallsetminus \mathcal A\) the corresponding detailed balance equation is \(Bc \pi (s) q(s,u)=Bc\pi ({\rho (u)})q({\rho (u)},\rho (s))\) that is also satisfied for the same reasons. Let us consider \(s \in \mathcal A\) and \(u \in \mathcal S \smallsetminus \mathcal A\), then we have that the transition rates are modified and hence \(B \pi (s) \left( c q(s,u)\right) =Bc\pi ({\rho (u)})q({\rho (u)},\rho (s))\) which is an identity since \(\sim \) respects \(\rho \). Finally, we have to consider the case of \(s \in \mathcal S\smallsetminus \mathcal A\) and \(u \in \mathcal A\). The corresponding detailed balance equation is \(B c \pi (s) q(s,u)=B\pi ({\rho (u)})\left( cq({\rho (u)},\rho (s))\right) \) which is satisfied by hypothesis. The fact that the residence times in the states belonging to the same orbits of \(\rho \) in \(X'(t)\) are identically distributed is an assumption of the lemma.   \(\square \)

Rights and permissions

Reprints and permissions

Copyright information

© 2017 Springer International Publishing AG

About this paper

Cite this paper

Marin, A., Rossi, S. (2017). Aggregation and Truncation of Reversible Markov Chains Modulo State Renaming. In: Thomas, N., Forshaw, M. (eds) Analytical and Stochastic Modelling Techniques and Applications. ASMTA 2017. Lecture Notes in Computer Science(), vol 10378. Springer, Cham. https://doi.org/10.1007/978-3-319-61428-1_11

Download citation

  • DOI: https://doi.org/10.1007/978-3-319-61428-1_11

  • Published:

  • Publisher Name: Springer, Cham

  • Print ISBN: 978-3-319-61427-4

  • Online ISBN: 978-3-319-61428-1

  • eBook Packages: Computer ScienceComputer Science (R0)

Publish with us

Policies and ethics