Abstract
We consider a discrete-time two-dimensional process \(\{(X_{1,n},X_{2,n})\}\) on \(\mathbb {Z}_+^2\) with a supplemental process \(\{J_n\}\) on a finite set, where the individual processes \(\{X_{1,n}\}\) and \(\{X_{2,n}\}\) are both skip-free. We assume that the joint process \(\{\varvec{Y}_n\}=\{(X_{1,n},X_{2,n},J_n)\}\) is Markovian and that the transition probabilities of the two-dimensional process \(\{(X_{1,n},X_{2,n})\}\) are modulated depending on the state of the supplemental process \(\{J_n\}\). This modulation is space homogeneous except for the boundaries of \(\mathbb {Z}_+^2\). We call this process a discrete-time two-dimensional quasi-birth-and-death process. Under several conditions, we obtain the exact asymptotic formulae of the stationary distribution in the coordinate directions.
We’re sorry, something doesn't seem to be working properly.
Please try refreshing the page. If that doesn't work, please contact support so we can address the problem.
References
Fayolle, G., Malyshev, V.A., Menshikov, M.V.: Topics in the Constructive Theory of Countable Markov Chains. Cambridge University Press, Cambridge (1995)
Flajolet, P., Sedgewick, R.: Analytic Combinatorics. Cambridge University Press, Cambridge (2009)
Foley, R.D., McDonald, D.R.: Bridges and networks: exact asymptotics. Ann. Appl. Probab. 15(1B), 542–586 (2005)
Gail, H.R., Hantler, S.L., Taylor, B.A.: Spectral analysis of M/G/1 and G/M/1 type Markov chain. Adv. Appl. Probab. 28, 114–165 (1996)
Gohberg, I., Lancaster, P., Rodman, L.: Matrix Polynomials. SIAM, Philadelphia (2009)
Horn, R.A., Johnson, C.R.: Topics in Matrix Analysis. Cambridge University Press, Cambridge (1991)
Kingman, J.F.C.: A convexity property of positive matrices. Q. J. Math. Oxf. 2(12), 283–284 (1961)
Kobayashi, M., Miyazawa, M.: Revisit to the tail asymptotics of the double QBD process: refinement and complete solutions for the coordinate and diagonal directions. In: Matrix-Analytic Methods in Stochastic Models, pp. 145–185 (2013)
Latouche, G., Ramaswami, V.: Introduction to Matrix Analytic Methods in Stochastic Modeling. SIAM, Philadelphia (1999)
Li, Q.-L., Zhao, Y.Q.: \(\beta \)-invariant measures for transition matrices of GI/M/1 type. Stoch. Models 19(2), 201–233 (2003)
Markushevich, A.I.: Theory of Functions of a Complex Variable. AMS Chelsea Publishing, Providence (2005)
Miyazawa, M., Zhao, Y.Q.: The stationary tail asymptotics in the GI/G/1 type queue with countably many background states. Adv. Appl. Probab. 36(4), 1231–1251 (2004)
Miyazawa, M.: Tail decay rates in double QBD processes and related reflected random walks. Math. Oper. Res. 34(3), 547–575 (2009)
Miyazawa, M.: Light tail asymptotics in multidimensional reflecting processes for queueing networks. TOP 19(2), 233–299 (2011)
Miyazawa, M.: Superharmonic vector for a nonnegative matrix with QBD block structure and its application to a Markov modulated two dimensional reflecting process. Queueing Syst. 81, 1–48 (2015)
Neuts, M.F.: Matrix-Geometric Solutions in Stochastic Models. Dover Publications, New York (1994)
Ozawa, T.: Asymptotics for the stationary distribution in a discrete-time two-dimensional quasi-birth-and-death process. Queueing Syst. 74, 109–149 (2013)
Ozawa, T.: Stability of multidimensional skip-free Markov modulated reflecting random walks: revisit to Malyshev and Menshikov’s results and application to queueing networks. Submitted (2015). arXiv:1208.3043
Ozawa, T., Kobayashi, M.: Exact asymptotic formulae of the stationary distribution of a discrete-time 2d-QBD process: an example and additional proofs (2018). arXiv:1805.04802
Seneta, E.: Non-negative Matrices and Markov Chains, Revised Printing. Springer, New York (2006)
Tweedie, R.L.: Operator-geometric stationary distributions for Markov chains with applications to queueing models. Adv. Appl. Probab. 14, 368–391 (1982)
Acknowledgements
We are grateful to Professor Masakiyo Miyazawa for valuable discussions with him about the convergence domain of the generating function \(\varvec{\varphi }(z,w)\). Also, the authors would like to thank the referees for their valuable comments and suggestions. This work was supported by JSPS KAKENHI Grant Number JP17K18126.
Author information
Authors and Affiliations
Corresponding author
Electronic supplementary material
Below is the link to the electronic supplementary material.
Proof of Lemma 3.1
Proof of Lemma 3.1
Recall that, for \(r>0\), we denote by \(\Delta _r\), \(\bar{\Delta }_r\) and \(\partial \Delta _r\) the open disk, closed disk and circle of center 0 and radius r on the complex plane, respectively. The generating function \(\varvec{\varphi }_+(z,w)\) is the function of two variables defined by the power series
which is also the Taylor series for the function \(\varvec{\varphi }_+(z,w)\) at \((z,w)=(0,0)\). Since \(\varvec{\varphi }_+(1,1)<\infty \), power series (A.1) converges absolutely on \(\bar{\Delta }_1\times \bar{\Delta }_1\) and \(\varvec{\varphi }_+(z,w)\) is analytic in \(\Delta _1\times \Delta _1\). The lemma asserts that, for every \((p,q)\in \mathcal {D}_0\), the power series (A.1) converges absolutely on \(\bar{\Delta }_p\times \bar{\Delta }_q\), and to prove it, it suffices to show that, for every \((p,q)\in \mathcal {D}_0\), the power series (A.1) converges absolutely at \((z,w)=(p,q)\), since any coefficient in the power series is nonnegative. Define the vector function \(\varvec{h}(z,w)\) by \(\varvec{h}(z,w) = \varvec{f}(z,w)/g(z,w)\), where
The function g(z, w) is analytic in \(\mathbb {C}^2\) and, by Proposition 3.1, each element of \(\varvec{f}(z,w)\) is analytic in \(\Delta _{r_1}\times \Delta _{r_2}\). For \((z,w)\in \bar{\Delta }_1\times \bar{\Delta }_1\), since the power series (A.1) is absolutely convergent, Eq. (3.3) holds and we have the \(\varvec{\varphi }_+(z,w)=\varvec{h}(z,w)\). Hence, by the identity theorem, we see that \(\varvec{h}(z,w)\) is an analytic extension of \(\varvec{\varphi }_+(z,w)\). Hereafter, we denote the analytic extension by the same notation \(\varvec{\varphi }_+(z,w)\). The proof of the lemma is given as follows.
Proof of Lemma 3.1
Under Assumption 2.3, either \(a_1^{\{1,2\}}\) or \(a_2^{\{1,2\}}\) is negative. Here, we assume \(a_1^{\{1,2\}}<0\); the proof for the case where \(a_2^{\{1,2\}}<0\) is analogous. Since \(\chi _{z_1}(1,1)=a_1^{\{1,2\}}<0\), where \(\chi _{z_1}(z_1,z_2)=(\partial /\partial \,z_1)\,\chi (z_1,z_2)\), we have the open interval \((1,\min \{\bar{\zeta _1}(1),r_1\})\subset \mathcal {D}_0\) (see Fig. 4). Let (p, q) be an arbitrary point in \(\mathcal {D}_0\), and consider a path on \(\mathcal {D}_0\cup \{(1,1)\}\) connecting different points \((p_0,q_0)=(1,1)\), \((p_1,q_0)\), \((p_1,q)\) and (p, q) by lines in this order, where we assume \(1<p_1<\min \{\bar{\zeta _1}(1),r_1\}\) (see Fig. 4). This is always possible because \(\mathcal {D}_0\) is given by
and \(\bar{\Gamma }\) is a convex set.
First, we consider a vector function of one variable given by \(\varvec{\varphi }_+(z,1)\). The Taylor series for \(\varvec{\varphi }_+(z,1)\) at \(z=0\) is given by the power series (A.1), where w is set at 1, and \(\varvec{\varphi }_+(z,1)\) is identical to \(\varvec{h}(z,1)\) on a domain where \(\varvec{h}(z,1)\) is well defined. Let \(\varepsilon \) be a sufficiently small positive number. Since g(z, 1) and each element of \(\varvec{f}(z,1)\) are analytic as a function of one variable in \(\Delta _{p_1+\varepsilon }\), \(\varvec{h}(z,1)\) is meromorphic as a function of one variable in that domain. We have, for \(z\in \Delta _{p_1+\varepsilon }{\setminus }\bar{\Delta }_1\), \(\mathrm{spr}(C(z,1))\le \mathrm{spr}(C(|z|,1))<1\), and, by Proposition 4.1 in Sect. 4, we have, for \(z\in \partial \Delta _{1}{\setminus }\{1\}\), \(\mathrm{spr}(C(z,1))<\mathrm{spr}(C(|z|,1))=1\). Hence, \(g(z,1)\ne 0\) for any \(z\in \Delta _{p_1+\varepsilon }{\setminus }(\Delta _1\cup \{1\})\), and each element of \(\varvec{h}(z,1)\) is analytic on \(\Delta _{p_1+\varepsilon }{\setminus }(\Delta _1\cup \{1\})\). Furthermore, we have \(\varvec{\varphi }_+(1,1)=\varvec{h}(1,1)<\infty \), and this implies that the point \(z=1\) is not a pole of any element of \(\varvec{h}(z,1)\); hence, it is a removable singularity. From this and the fact that \(\varvec{\varphi }_+(z,1)\) is analytic in \(\Delta _{1}\), we see that \(\varvec{\varphi }_+(z,1)\) is analytic in \(\Delta _{p_1+\varepsilon }\). This implies that the radius of convergence of the Taylor series for \(\varvec{\varphi }_+(z,1)\) at \(z=0\) is greater than \(p_1\), and the power series (A.1) converges at \((z,w)=(p_1,1)\).
Next, we consider a vector function of one variable given by the \(\varvec{\varphi }_+(p_1,w)\). By the fact obtained above, the Taylor series for \(\varvec{\varphi }_+(p_1,w)\) at \(w=0\) is given by the power series (A.1), where z is set at \(p_1\), and \(\varvec{\varphi }_+(p_1,w)\) is analytic as a function of one variable in \(\Delta _{1}\). Furthermore, we know that \(\varvec{\varphi }_+(p_1,w)\) is identical to \(\varvec{h}(p_1,w)\) on a domain where \(\varvec{h}(p_1,w)\) is well defined. If \(q\le 1\), then it is obvious that the power series (A.1) converges at \((z,w)=(p_1,q)\). Therefore, we assume \(q>1\). Let \(\varepsilon \) be a sufficiently small positive number. For \(w\in \Delta _{q+\varepsilon }{\setminus }\bar{\Delta }_{1-\varepsilon }\), we have \(\mathrm{spr}(C(p_1,w))\le \mathrm{spr}(C(p_1,|w|))<1\). Hence, for the same reason used in the case of \(\varvec{h}(z,1)\), we see that \(\varvec{h}(p_1,w)\) is analytic in \(\Delta _{q+\varepsilon }{\setminus }\bar{\Delta }_{1-\varepsilon }\), and this implies that \(\varvec{\varphi }_+(p_1,w)\) is analytic in \(\Delta _{q+\varepsilon }\). Hence, the radius of convergence of the Taylor series for \(\varvec{\varphi }_+(p_1,w)\) at \(w=0\) is greater than q, and the power series (A.1) converges at \((z,w)=(p_1,q)\). Applying a similar procedure to the vector function \(\varvec{\varphi }_+(z,q)\), we see that the power series (A.1) converges at \((z,w)=(p,q)\), and this completes the proof. \(\square \)
Rights and permissions
About this article
Cite this article
Ozawa, T., Kobayashi, M. Exact asymptotic formulae of the stationary distribution of a discrete-time two-dimensional QBD process. Queueing Syst 90, 351–403 (2018). https://doi.org/10.1007/s11134-018-9586-x
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11134-018-9586-x
Keywords
- Quasi-birth-and-death process
- Stationary distribution
- Asymptotic property
- Matrix analytic method
- Two-dimensional reflecting random walk