Abstract
We study a stylized queueing model motivated by paid express lanes on highways. There are two parallel, observable first-come, first-served queues with finitely many servers: one queue has a faster service rate, but charges a fee to join, and the other is free but slow. Upon arrival, customers see the state of each queue and choose between them by comparing the respective disutility of time spent waiting, subject to random shocks. This framework encompasses both the multinomial logit and exponomial customer choice models. Using a fluid limit approximation, we give a detailed characterization of the equilibrium in this system. We show that social welfare is optimized when the express queue is exactly at (but not over) full capacity; however, in some cases, revenue is maximized by artificially creating congestion in the free queue. The latter behavior is caused by changes in the price elasticity of demand as the service capacity of the free queue fills up.
Similar content being viewed by others
References
Afèche, P.: Delay performance in stochastic processing networks with priority service. Oper. Res. Lett. 31(5), 390–400 (2003)
Afèche, P., Ata, B.: Bayesian dynamic pricing in queueing systems with unknown delay cost characteristics. Manuf. Serv. Oper. Manag. 15(2), 292–304 (2013)
Afèche, P., Baron, O., Milner, J., Roet-Green, R.: Pricing and prioritizing time-sensitive customers with heterogeneous demand rates. Oper. Res. 67(4), 1184–1208 (2019)
Afèche, P., Pavlin, J.M.: Optimal price/lead-time menus for queues with customer choice: segmentation, pooling, and strategic delay. Manag. Sci. 62(8), 2412–2436 (2016)
Afèche, P., Sarhangian, V.: Rational Abandonment from Priority Queues: Equilibrium Strategy and Pricing Implications. Technical report, Columbia University (2015)
Alptekinoglu, A., Semple, J.H.: The exponomial choice model: a new alternative for assortment and price optimization. Oper. Res. 64(1), 79–93 (2016)
Armony, M., Maglaras, C.: On customer contact centers with a call-back option: customer decisions, routing rules, and system design. Oper. Res. 52(2), 271–292 (2004)
Aveklouris, A., Vlasiou, M., Zwart, B.: Bounds and limit theorems for a layered queueing model in electric vehicle charging. Queueing Syst. 93(1–2), 83–137 (2019)
Besbes, O., Maglaras, C.: Revenue optimization for a make-to-order queue in an uncertain market environment. Oper. Res. 57(6), 1438–1450 (2009)
Boyd, S., Vandenberghe, L.: Convex Optimization. Cambridge University Press, Cambridge (2004)
Cao, P., Wang, Y., **e, J.: Priority service pricing with heterogeneous customers: impact of delay cost distribution. Prod. Oper. Manag. 28(11), 2854–2876 (2019)
Cui, S., Wang, Z., Yang, L.: The economics of line-sitting. Manag. Sci. 66(1), 227–242 (2020)
Frolkova, M., Zwart, B.: Fluid limit of a PS-queue with multistage service. Probab. Eng. Inf. Sci. 33(1), 1–27 (2019)
Gavirneni, S., Kulkarni, V.G.: Self-selecting priority queues with Burr distributed waiting costs. Prod. Oper. Manag. 25(6), 979–992 (2016)
Göçmen, F.C.: Infrastructure scaling and pricing. Ph.D. thesis, Columbia University (2014)
Göçmen, F.C., Phillips, R., van Ryzin, G.: Revenue Maximizing Dynamic Tolls for Managed Lanes: A Simulation Study. Technical report, Columbia University (2015)
Guo, P., Zhang, Z.G.: Strategic queueing behavior and its impact on system performance in service systems with the congestion-based staffing policy. Manuf. Serv. Oper. Manag. 15(1), 118–131 (2013)
Hale, J.K.: Ordinary Differential Equations. Dover Publications Inc., Mineola (2009)
Hassin, R.: Rational Queueing. Chapman and Hall/CRC, Boca Raton (2016)
Hassin, R., Roet-Green, R.: The impact of inspection cost on equilibrium, revenue, and social welfare in a single-server queue. Oper. Res. 65(3), 804–820 (2017)
Hu, M., Li, Y., Wang, J.: Efficient ignorance: information heterogeneity in a queue. Manag. Sci. 64(6), 2650–2671 (2017)
Ibrahim, R.: Managing queueing systems where capacity is random and customers are impatient. Prod. Oper. Manag. 27(2), 234–250 (2018)
Jiménez, T., Koole, G.: Scaling and comparison of fluid limits of queues applied to call centers with time-varying parameters. OR Spectrum 26(3), 413–422 (2004)
Kurtz, T.G.: Strong approximation theorems for density dependent Markov chains. Stoch. Process. Appl. 6(3), 223–240 (1978)
Maglaras, C.: Revenue management for a multiclass single-server queue via a fluid model analysis. Oper. Res. 54(5), 914–932 (2006)
Maglaras, C., Yao, J., Zeevi, A.: Optimal price and delay differentiation in large-scale queueing systems. Manag. Sci. 64(5), 2427–2444 (2017)
Mandelbaum, A., Massey, W.A., Reiman, M.I.: Strong approximations for Markovian service networks. Queueing Syst. 30(1), 149–201 (1998)
Naor, P.: The regulation of queue size by levying tolls. Econometrica 37, 15–24 (1969)
Pender, J., Rand, R.H., Wesson, E.: A stochastic analysis of queues with customer choice and delayed information. Math. Oper. Res. 45(3), 1104–1126 (2020)
Pender, J., Rand, R.H., Wesson, E.: Queues with choice via delay differential equations. Int. J. Bifurc. Chaos 27(4), 1730016–1 (2017)
Pender, J., Rand, R.H., Wesson, E.: An analysis of queues with delayed information and time-varying arrival rates. Nonlinear Dyn. 91(4), 2411–2427 (2018)
Ratliff, L.J., Dowling, C., Mazumdar, E., Zhang, B.: To observe or not to observe: queuing game framework for urban parking. In: Proceedings of the 55th IEEE Conference on Decision and Control, pp. 5286–5291 (2016)
Seydel, R.: Practical Bifurcation and Stability Analysis, 3rd edn. Springer, Berlin (2010)
Taylor, T.A.: On-demand service platforms. Manuf. Serv. Oper. Manag. 20(4), 704–720 (2018)
Wang, J., Baron, O., Scheller-Wolf, A.: M/M/c queue with two priority classes. Oper. Res. 63(3), 733–749 (2015)
Wang, J., Cui, S., Wang, Z.: Equilibrium strategies in M/M/1 priority queues with balking. Prod. Oper. Manag. 28(1), 43–62 (2019)
Yang, L., Debo, L.G., Gupta, V.: Search among queues under quality differentiation. Manag. Sci. 65(8), 3605–3623 (2019)
Zhang, Y., Wang, J., Wang, F.: Equilibrium pricing strategies in retrial queueing systems with complementary services. Appl. Math. Model. 40(11–12), 5775–5792 (2016)
Acknowledgements
The authors are grateful to **aoshan Peng for many helpful discussions, and to the Associate Editor and referee for their thoughtful comments on the paper.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Proofs
Proofs
Below, we give the complete proofs of all results that were stated in the main text.
1.1 Proof of Theorem 1
Again, we assume \(Q_i\left( 0\right) = Q^n_i\left( 0\right) = q_i\left( 0\right) = 0\). We follow [29] in using the following result from [24]:
Lemma 1
A standard Poisson process \(\varPi \) and a standard Brownian motion B can be constructed on the same probability space in such a way that the limit
is a.s. finite and has finite mean and finite moment-generating function in the neighborhood of zero.
Lemma 1 allows us to rewrite \(Q^n_i\) in terms of two standard Brownian motions \(B^\mathrm{arr}_i,B^\mathrm{dep}_i\) via
and
Now, we calculate the difference between the scaled length process for queue \(i \in \left\{ e,r\right\} \) and its fluid limit, given by
whence
Substituting (44) into the first term of (46), we obtain
with the Brownian term satisfying
Substituting (45) into the second term of (46) yields
with the Brownian term satisfying
Thus, (46) has become
The choice probability \(p_i\) and the departure function both have uniformly bounded derivatives by assumption P1, so there exist constants C and \(\varepsilon \) such that, for large enough n, we have
Applying Gronwall’s lemma [18], we obtain
We can then take \(\varepsilon \rightarrow 0\) to obtain the desired result.
The solution of (6) and (7) is non-negative because, for \(i\in \left\{ r,e\right\} \), \(q_i=0\) implies \(q'_i>0\) (the choice probabilities are always strictly positive). Uniqueness of the solution is shown by letting \(q^{(1)}_i,q^{(2)}_i\) be two sets of solutions with the same initial values, and writing
Then, we again invoke the properties of \(p_i\) and Gronwall’s lemma, as above, to conclude that \(q^{(1)}_i=q^{(2)}_i\).
1.2 Proof of Theorem 2
Existence of the equilibrium We first show the existence of the equilibrium using Brouwer’s fixed point theorem, which states that, if f is a continuous function map** a compact convex set to itself, there exists a point \(x_0\) satisfying \(f\left( x_0\right) = x_0\).
We rewrite the equilibrium conditions (8) and (9) as
We can then express the system (47) and (48) as \(f\left( q\right) = q\), where \(q = \left( q_e,q_r\right) \). Because we have assumed continuity of \(p_e,p_r\) (assumption P1), it straightforwardly follows that f is continuous.
To show that \(f = \left( f_e,f_r\right) \) maps a compact convex set to itself, let us consider the first component \(f_e\) and suppose that \(q_e < {\bar{q}}_e\). In this case, we have the bound \(f_e\left( q_e\right) \le \lambda + {\bar{q}}_e\).
When \(q_e \ge {\bar{q}}_e\), we have \(f_e\left( q_e\right) = \lambda p_e\left( u\left( \frac{q_e}{\mu _e{\bar{q}}_e}\right) ,u_r,{\bar{u}}\right) - \mu _e {\bar{q}}_e + q_e\). Note that, if \({\bar{q}}_e \ge \frac{\lambda }{\mu _e}\), then \(f_e\left( q_e\right) < q_e\) and the codomain of \(f_e\) is automatically contained in the domain.
If \({\bar{q}}_e < \frac{\lambda }{\mu _e}\), let \({\tilde{q}}_e\) be a value satisfying
Then, for \(q_e \ge {\tilde{q}}_e\), we have
implying \(f\left( q_e\right) \le q_e\). Finally, for \({\bar{q}}_e< q_e < {\tilde{q}}_e\), we have
Denote by \({\hat{q}}_e\) the right-hand side of (49). Then, for any \(0 \le q_e \le \max \left\{ {\bar{q}}_e,{\tilde{q}}_e,{\hat{q}}_e,\frac{\lambda }{\mu _e}\right\} \), we have \(f\left( q_e\right) \) in the same interval, regardless of \(q_e\). Thus, the conditions for Brouwer’s fixed point theorem hold and the equilibrium exists.
Uniqueness of the equilibrium Let \(\lambda ,\mu _e,\mu _r\) and the disutility function u be given. Suppose that there are two non-identical equilibrium solutions \(\left( q^{(1)}_e,q^{(1)}_r\right) \) and \(\left( q^{(2)}_e,q^{(2)}_r\right) \). Let us focus on the case where \(q^{(1)}_e < q^{(2)}_e\) (the other case where we start with \(q^{(1)}_r < q^{(2)}_r\) is handled symmetrically).
We first show that, if \(q^{(1)}_e < q^{(2)}_e\), then \(q^{(1)}_r < q^{(2)}_r\) as well. To see this, let us assume the contrary, i.e., that \(q^{(1)}_r \ge q^{(2)}_r\). We derive
where (50) is obtained from \(q^{(1)}_e < q^{(2)}_e\) and the fact that \(u' > 0\) (assumption U3) while \(p_e\) is decreasing in \(u_e\); equation (51) follows from the assumption that \(q^{(1)}_r \ge q^{(2)}_r\) and the fact that \(u' > 0\) while \(p_e\) is increasing in \(u_r\); and (52) follows from \(q^{(1)}_e < q^{(2)}_e\). However, since the first and last line both equal zero due to the equilibrium conditions, (50)–(52) must all hold with strict equality. Consequently, (51) and (52) imply that
whence we conclude \({\bar{q}}_e \le q^{(1)}_e < q^{(2)}_e\). From that, however, (50) yields
and this is impossible since \(u' > 0\) with strict inequality. Therefore, \(q^{(1)}_e < q^{(2)}_e\) implies \(q^{(1)}_r < q^{(2)}_r\).
Next, we claim that \(q^{(2)}_r > {\bar{q}}_r\). To see this, let us assume the opposite, i.e., that \(q^{(2)}_r \le {\bar{q}}_r\), whence \(u_r\left( q^{(1)}_r\right) = u_r\left( q^{(2)}_r\right) = u\left( \frac{1}{\mu _r}\right) \). We then have
and \(q^{(1)}_e < q^{(2)}_e\) implies that (54) holds with strict equality. This again implies (53) and the same reasoning as before can be repeated to obtain a contradiction. Therefore, \(q^{(2)}_r > {\bar{q}}_r\). A symmetric argument can be used to show \(q^{(2)}_e > {\bar{q}}_e\).
Combining the previous facts, (10) yields
However, from \(q^{(1)}_e < q^{(2)}_e\) and \(q^{(1)}_r < q^{(2)}_r\) we obtain
regardless of whether \(q^{(1)}_e\) and \(q^{(1)}_r\) are under or over capacity, because \(p_o\) satisfies assumption P2. We conclude that it is impossible to have \(q^{(1)}_e < q^{(2)}_e\) and still satisfy the equilibrium conditions for both solutions.
1.3 Proof of Theorem 3
We examine each of regimes R1–R4 separately. In each regime, we write (6) and (7) as
obtain all of the first-order partial derivatives \(\frac{\partial f_i}{\partial q_i}\) for \(i \in \left\{ e,r\right\} \), put them into matrix form (the Jacobian) and evaluate this matrix at the equilibrium \(\left( q^*_e,q^*_r\right) \), which we know exists and is unique from the preceding. The equilibrium is locally stable if both eigenvalues of the Jacobian are negative [33].
Regime R1 We have \(q^*_e \ge {\bar{q}}_e\), \(q^*_r \ge {\bar{q}}_r\) and the Jacobian is given by
Letting \(e_1,e_2\) be the eigenvalues, the characteristic equation is given by
which can be rewritten as
We argue that
is positive, which would imply that the product \(e_1 e_2\) in (55) is also positive. We first observe that the product \(\frac{\partial u_e}{\partial q_e}\frac{\partial u_r}{\partial q_r}\) is positive since, for example,
by assumption U3. The same is true of \(\frac{\partial u_r}{\partial q_r}\).
Assumption P3 implies
since changing all the disutilities by the same amount does not change the probability of any choice. Since \(\frac{\partial p_e}{\partial {\bar{u}}} > 0\) by assumption P2, it follows that \(\frac{\partial p_e}{\partial u_e} + \frac{\partial p_e}{\partial u_r} < 0\), whence
and, symmetrically,
From this we obtain
where the first inequality is obtained from (56) and the fact that \(\frac{\partial p_e}{\partial u_e} < 0\), while the second inequality is obtained from (57) and the fact that \(\frac{\partial p_e}{\partial u_r} > 0\). Thus, we conclude that \(\det \left( J^{R1}\right) > 0\) and so both \(e_1,e_2\) have the same sign.
From the preceding, it follows that the left-hand side of (55) is positive. On the right-hand side of (55), suppose that \(e_1,e_2\) are both positive; then we have
since \(\frac{\partial p_r}{\partial u_r},\frac{\partial p_e}{\partial u_e} < 0\) while \(\frac{\partial u_r}{\partial q_r},\frac{\partial u_e}{\partial q_e} > 0\). Therefore, both \(e_1,e_2\) must be negative, as required.
Regime R2 We have \(q^*_e < {\bar{q}}_e\), \(q^*_r < {\bar{q}}_r\) and the Jacobian is given by
from which the conclusion directly follows.
Regime R3 We have \(q^*_e \ge {\bar{q}}_e\), \(q^*_r < {\bar{q}}_r\) and the Jacobian is given by
which is a lower triangular matrix, meaning that its eigenvalues are on the diagonal. It is easy to see that both are negative.
Regime R4 The proof is very similar to the previous case and is omitted.
1.4 Proof of Theorem 4
The assumptions on v imply that, for any \(s_1 < s_2\), we have
This fact will be used to show the desired result in each of the relevant regimes.
Regime R2 Since both queues are under capacity, we have \(u_e\left( q^u_e\right) = u\left( \frac{1}{\mu _e}\right) \) and \(u_r\left( q^u_r\right) = u\left( \frac{1}{\mu _r}\right) \). From (58), we obtain
We will now show that \(q^u_e < q^v_e\) by contradiction. Suppose that \(q^u_e \ge q^v_e\). It follows that the express queue continues to be under capacity when we switch to v. We then derive
where (60) and (61) follow because both \(q^u_e,q^v_e < {\bar{q}}_e\). Next, we let \(\delta = v\left( \frac{1}{\mu _r}\right) - u\left( \frac{1}{\mu _r}\right) \), note that \(\delta > 0\), and observe that
Above, (62) is due to assumption P3, (63) follows from assumption P2 combined with (59), and (64) follows by (8).
To obtain the desired contradiction, we consider two cases, one where \(q^v_r < {\bar{q}}_r\) and one where \(q^v_r \ge {\bar{q}}_r\). Suppose that \(q^v_r < {\bar{q}}_r\). Then, both queues are under capacity with v as the disutility function, so (8) implies
On the other hand, if \(q^v_r \ge {\bar{q}}_r\), we have \(q^v_r > q^u_r\) and
by assumption P2. Either case, when combined with (65), yields \(q^v_e < q^v_e\), which is impossible; therefore, we must have \(q^u_e < q^v_e\).
Regime R4 In this regime, only the express queue is under capacity. There are two possible permutations
Applying (58) to both of these yields
First, let us suppose that permutation (66) is correct. Again, we proceed by contradiction and assume that \(q^u_e \ge q^v_e\). Since the express queue is under capacity with either disutility function, we have
Letting \(\delta = {\bar{v}} - {\bar{u}}\), we further derive
where, as before, (68) is due to assumption P3, while (69) follows from (66) combined with assumption P2. From (70) and assumption P2, we conclude that \(q^u_r > q^v_r\), otherwise there will be no way to satisfy (8).
Now, (10) yields
whence
At the same time, letting \(\delta = v\left( \frac{q^u_r}{\mu _r{\bar{q}}_r}\right) -u\left( \frac{q^u_r}{\mu _r{\bar{q}}_r}\right) \), we obtain
where the first equality is due to assumption P3, while (73) follows from (66) and assumption P2, while (74) follows from assumption P2 and \(q^u_r > q^v_r\). Clearly (72) and (74) contradict each other, whence we conclude that \(q^u_e < q^v_e\).
Finally, we suppose that permutation (67) is correct. In this case, however, the proof is nearly identical. The only difference is that, in order to obtain (70), we use \(\delta = v\left( \frac{q^u_r}{\mu _r{\bar{q}}_r}\right) -u\left( \frac{q^u_r}{\mu _r{\bar{q}}_r}\right) \), while (73) is obtained by using \(\delta = {\bar{v}} - {\bar{u}}\). The same contradiction then follows.
Rights and permissions
About this article
Cite this article
Zhou, J., Ryzhov, I.O. Equilibrium analysis of observable express service with customer choice. Queueing Syst 99, 243–281 (2021). https://doi.org/10.1007/s11134-021-09720-z
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11134-021-09720-z