Randomization Approaches for Network Revenue Management with Customer Choice Behavior


 Shon Chase
 1 years ago
 Views:
Transcription
1 Randomization Approaches for Network Revenue Management with Customer Choice Behavior Sumit Kunnumkal Indian School of Business, Gachibowli, Hyderabad, , India sumit March 9, 2011 Abstract In this paper, we present new approximation methods for the network revenue management problem with customer choice behavior. Our methods are samplingbased and so we require only minimal assumptions regarding the underlying customer choice model. The starting point for our methods is a dynamic program that allows randomization. An attractive feature of this dynamic program is that the size of its action space is linear in the number of itineraries, as opposed to exponential. It turns out that this dynamic program has a structure that is similar to the dynamic program for the network revenue management problem under the so called independent demand setting. Our approximation methods exploit this similarity and build on ideas developed for the independent demand setting. We present two approximation methods. The first one is based on relaxing the flight leg capacity constraints using Lagrange multipliers, whereas the second method involves solving a perfect hindsight relaxation problem. We show that both methods yield upper bounds on the optimal expected total revenue. Computational experiments indicate that our methods can generate tighter upper bounds and higher expected revenues when compared with the standard deterministic linear program that appears in the literature.
2 Network revenue management with customer choice behavior is wellstudied and has many applications in the airline, hotel and car rental industries. In the context of airlines, a representative example, it involves controlling the sale of itineraries over a flight network. Customers arrive over the booking period to purchase itineraries. The airline has to decide which itineraries to make available for sale at each point in time taking into account the remaining capacities on the flight legs. This is a crucial decision to make since the customer s purchasing decision is influenced by the set of itineraries that are offered. Depending on the offer set, the customer may purchase one of the offered itineraries, or may not purchase anything and simply leave. The airline s goal is to determine the set of itineraries to offer at each point in time that maximizes the expected total revenues over the booking period. The airline s decision problem can be formulated as a dynamic program. However, computing the value functions and the optimal policy quickly become intractable and one has to resort to approximation methods. Many of the approximation methods for the network revenue management problem with customer choice build on methods developed for network revenue management under the assumption that the customer s purchasing decision is not influenced by the set of offered itineraries. This is the so called independent demand setting, where we assume that customers arrive with the intention of purchasing a fixed itinerary. If the itinerary is available, they make the purchase. Otherwise, they leave without making any purchase. Even with the independent demand assumption, the network revenue management problem becomes intractable as the size of the state space increases exponentially with the number of flight legs. Consequently, the approximation methods for the network revenue management problem with independent demand have mainly been concerned with reducing the dimensionality of the state space. Incorporating customer choice behavior adds another layer of complexity since the size of the action space also increases exponentially with the number of itineraries. This is because of the combinatorial nature of the problem of deciding which subset of itineraries to offer for sale from the set of all possible itineraries. So, while many of the approximation methods for the network revenue management problem with customer choice are able to handle the dimensionality of the state space quite well, they are less effective in dealing with the complexity of the action space. As a result, the tractability of many of the existing methods depends on the underlying model of customer choice. It is usually assumed that the customer choices are governed by the multinomial logit model and that the consideration sets, the sets of itineraries of interest to the different customer segments, are disjoint. In this paper, we propose new approximation methods that remain tractable for a large class of choice models. We assume that a customer s choice decision is governed by a simple utility maximization principle. That is, a customer has a utility for purchasing each of the itineraries and to not purchasing anything. Of the available alternatives, the customer chooses the one with the highest utility. The starting point for our methods is a dynamic program that allows randomization. We generate a sample path of customer arrivals along with their utilities for the different itineraries and formulate a dynamic program in order to compute the optimal offer sets. We show that it is possible to reformulate this problem as a dynamic program where the number of decision variables is linear in the number of itineraries. As a result, the size of the action space becomes manageable. In fact, the resulting formulation is similar to the dynamic programming formulation of the network revenue management problem with independent demand. Consequently, we use ideas from the independent demand setting to reduce the 2
3 size of the state space. We particularly focus on two approximation methods. One is based on the Lagrangian relaxation idea developed in Kunnumkal and Topaloglu (2010a) and the second is based on the randomized linear programming approach developed in Talluri and van Ryzin (1999). The methods that we propose have a number of appealing features. Since they are samplingbased, they can handle a much broader class of choice models. In particular, we do not require the assumption that customer choices come from a multinomial logit choice model with disjoint consideration sets. Our methods yield upper bounds on the optimal expected revenue and estimates of the expected marginal values of capacity on the flight legs. The marginal value of capacity on a flight leg, referred to as its bid price, is useful in constructing control policies. On the other hand, upper bounds are useful when assessing the suboptimality of heuristic control policies. Another useful feature of our approach is that the randomized dynamic program we propose has a similar structure to the dynamic program for the network revenue management problem with independent demand. This allows us to draw upon the rich literature around the network revenue management problem with independent demand. The two approximation methods that we propose require solving only linear programs, which most commercial optimization packages are capable of. Moreover, since the linear programs we solve have only a polynomial number of variables and constraints, it minimizes the need for customized coding in the way of column generation techniques. This may further enhance the practical appeal of our methods. Our work builds on previous research. Liu and van Ryzin (2008) propose a deterministic linear program for the network revenue management problem with customer choice. Zhang and Adelman (2009), Meissner and Strauss (2008) and Zhang (2011) use the linear programming approach to approximate dynamic programming to come up with different value function approximations, where as Kunnumkal and Topaloglu (2008) and Kunnumkal and Topaloglu (2010b) use Lagrangian relaxation ideas. Kunnumkal and Topaloglu (2010c) propose a linear integer program that allows randomization and show how it can be used to compute bid prices. The tractability of the above mentioned methods depends on the assumptions that the customers choices are governed by the multinomial logit model and that the consideration sets of the different customer segments are disjoint. Bront, MendezDiaz and Vulcano (2009) analyze the case where the consideration sets overlap and show that the column generation subproblem in the deterministic linear program of Liu and van Ryzin (2008) is NPhard. Bront et al. (2009) and Meissner and Strauss (2010) propose heuristic methods for column generation. Talluri (2010) proposes a concave program for general choice models and describes a way to randomize it. Meissner, Strauss and Talluri (2011) build on this concave program and show how it can be strengthened by adding additional constraints. van Ryzin and Vulcano (2008) and Chaneton and Vulcano (2009) use stochastic approximation to respectively compute protection levels and bid prices for general choice models. The utility maximization criterion to model customer choice behavior has appeared in the literature. For example, van Ryzin and Vulcano (2008) and Chaneton and Vulcano (2009) use it to model customer choice in network revenue management while Mahajan and van Ryzin (2001) use it in the context of optimizing retail assortments. Our work also builds on approximation methods for the network 3
4 revenue management problem with independent demand. The papers closest to ours are Kunnumkal and Topaloglu (2010a) and Talluri and van Ryzin (1999). Kunnumkal and Topaloglu (2010a) propose a linear program that yields time dependent bid prices. Talluri and van Ryzin (1999) propose a randomized linear program that works with samples of the demand random variables. We refer the reader to Talluri and van Ryzin (2004) for a comprehensive review of the revenue management literature. We make the following research contributions in this paper. 1) We present a new dynamic programming approximation for the network revenue management problem with customer choice behavior. This dynamic programming formulation is attractive because it allows randomization and the size of its action space is linear in the number of itineraries. 2) We further build on this randomized dynamic program to obtain tractable approximation methods. As our methods are samplingbased, we are not constrained by the underlying customer choice model. We are able to handle a variety of choice models; all we require is the ability to generates samples of the customers utilities for the different alternatives. 3) We show that our approximation methods generate upper bounds on the optimal expected revenues. Upper bounds are useful when assessing the suboptimality of heuristic control policies. We also show how our methods can be used to obtain bid prices. 4) Computational experiments indicate that our methods can yield significantly tighter upper bounds and higher revenues than the standard deterministic linear program. The rest of the paper is organized as follows. Section 1 describes the network revenue management problem with customer choice behavior and formulates it as a dynamic program. In Section 2, we describe the linear program proposed by Liu and van Ryzin (2008). In Section 3, we present the randomized dynamic program and in Section 4 we describe two tractable approximation methods based on it. The first method is based on relaxing the flight leg capacity constraints whereas the second method solves a perfect hindsight relaxation. Section 5 presents our computational experiments. The proofs of all the propositions and lemmas are deferred to the Appendix. 1 Problem Formulation We have an airline network consisting of a set of flight legs that we can use to serve the customers that arrive over time with the intention of purchasing itineraries. We use L to denote the set of flight legs in the airline network. The initial capacity on flight leg i is c i. We use J to denote the set of all itineraries. An itinerary j has a revenue associated with it, which we denote by r j. If we accept a request for itinerary j, then we consume capacity on one or more flight legs. We use a ij to denote the number of units of capacity consumed by itinerary j on flight leg i. Naturally, we have a ij = 0 if itinerary j does not include flight leg i. We discretize the planning horizon into a finite number of time periods T = {1,..., τ} and assume that the discretization is fine enough so that there is at most one customer arrival at each time period. The probability of a customer arrival at time period t is λ. The fact that the arrival probability is constant over time is only for ease of exposition and it is straightforward to allow the arrival probability to depend on the time period t. We assume that customer choice is governed by a simple utility maximization principle. That is, 4
5 the customer s utilities for the different alternatives are random variables and the customer chooses the alternative with the highest utility. We note that the utility maximization principle is essentially equivalent to a choice model where customers have an ordered list of preferences and pick the most preferred alternative from the ones available. We let U jt be the random variable which denotes the utility for purchasing itinerary j at time period t and let U φt be the random variable which denotes the utility for not purchasing any itinerary at time period t. We let U J = {U jt : j J, t T } and U φ = {U φt : t T }. We allow the random variables {U jt : j J {φ}} to be dependent within each time period, but assume that they are independent across time periods. In other words, the purchasing decisions of the different customers are assumed to be independent of each other. Given an offer set S, the customer chooses the alternative j t = argmax j S {φ} {U jt } with the highest utility. We assume that there are no ties with probability 1. The probability that the customer chooses itinerary j at time period t given the offer set S is Pr{j t = j S} = Pr{U jt = max {U kt}} for j S. k S {φ} We have Pr{j t = j S} = 0 for j / S and the probability of purchasing nothing is Pr{j t = φ S} = Pr{U φt = max {U kt}} = 1 Pr{j t = j S}. (1) k S {φ} Here we emphasize that the customer s utilities for the different alternatives do not depend on the set of itineraries made available for sale. While there are choice models that do not satisfy this assumption, it covers many of the commonly used choice models in the literature; see Mahajan and van Ryzin (2001) and Zhang and Cooper (2005). At each time period, we have to decide which itineraries to make available for sale taking into account the state of the remaining leg capacities. Using x it to denote the remaining capacity on flight leg i at time period t, x t = {x it : i L} captures the state of the remaining leg capacities. We let Q(x t ) = {j J : a ij x it i L}, (2) denote the itineraries that can be potentially offered given the remaining leg capacities. The decision problem is to determine the set of itineraries to offer to the customers at each time period so as to maximize the expected total revenue over the planning horizon. Under the assumption that the customer arrivals in the different time periods and the purchasing decisions of the different customers are independent of each other, we can obtain the value functions {V t ( ) : t T } through the optimality equation V t (x t ) = max S Q(x t) = max S Q(x t) { { λ Pr{j t = j S} r j + V t+1 (x t } ] ] i L a ij e i ) + 1 λ + λ Pr{j t = φ S} V t+1 (x t ) λ Pr{j t = j S} r j + V t+1 (x t i L a ij e i ) V t+1 (x t )] } + V t+1 (x t ), (3) where e i is the L dimensional unit vector with a one in the element corresponding to i L and the second equality follows from (1). The boundary condition for the optimality equation above is V τ+1 ( ) = 5
6 0. Throughout the rest of the paper, we assume that λ = 1 for notational brevity. We note that this is equivalent to letting Pr{j t = j S} = λ Pr{j t = j S} and Pr{j t = φ S} = 1 λ + λ Pr{j t = j S} and working with the probabilities { Pr{j t = j S} : j J {φ}}. Solving the above dynamic program for practical problem instances becomes difficult for two reasons. One is that the size of the state space increases exponentially with the number of flight legs in the airline network. For, if we let C i = {0,..., c i }, then the state space of the above dynamic program is i L C i, which is exponential in the number of flight legs. Secondly, the size of the action space also increases exponentially with the number of itineraries in the flight network since the number of potential offer sets is of the order of 2 J. In the following sections, we look at relaxations of problem (3) that are computationally tractable. 2 Choice Based Deterministic Linear Program The choice based deterministic linear program, proposed by Liu and van Ryzin (2008), is an approximation that replaces all random quantities by their expected values. If set S is offered at time period t, then the expected revenue obtained is r j Pr{j t = j S}, while the expected capacity consumed on flight leg i is a ij Pr{j t = j S}. The choice based deterministic linear program assumes that the revenue generated and the capacities consumed by offering set S take on their expected values. It determines the optimal choice of offer sets at each time period by solving z CDLP = max r j Pr{j t = j S}h t (S) (4) t T subject to a ij Pr{j t = j S}h t (S) c i i L (5) t T h t (S) 1 t T (6) S J h t (S) 0 t T. (7) In the above linear program, the decision variable h t (S) denotes the frequency with which set S is offered at time period t. The first set of constraints ensure that the expected capacity consumed on each flight leg does not exceed the available capacity. The second set of constraints ensure that the total frequency with which we offer the sets at each time period is at most one. Note that the number of decision variables in the above linear program is exponential in the number of itineraries. So in general, one has to resort to column generation to solve the problem (4)(7). Liu and van Ryzin (2008) show that column generation can be efficiently carried out provided the choice probabilities come from the multinomial logit model. Gallego, Ratliff and Shebalov (2010) show that problem (4)(7) can be reformulated as a linear program with only a polynomial number of variables provided the choice probabilities come from a general attraction model, of which the multinomial logit is a special case. There are two main uses of the choice based deterministic linear program. First, Liu and van Ryzin (2008) show that its optimal objective value gives an upper bound on the optimal expected total 6
7 revenue. That is, we have V 1 (c) z CDLP. Second, we can use the dual solution of the choice based deterministic linear program to construct heuristic control policies. Let ˆπ = {ˆπ i : i L} denote the optimal values of the dual variables associated with constraints (5). Noting that ˆπ i approximates the marginal value of capacity on flight leg i, we use ˆπ i as its bid price. We can use these bid prices to come up with different control policies. Zhang and Adelman (2009) propose approximating the value function V t (x t ) by i L ˆπ ix it and solving problem (3) using this value function approximation to decide on the offer set. That is, we solve the problem max S Q(x t) Pr{j t = j S} r j ] a ij ˆπ i i L to decide on the set of itineraries to offer at time period t. (8) We note that the above maximization problem is combinatorial in nature and can be potentially difficult to solve for a general choice model. Bront et al. (2009) and Meissner and Strauss (2010) propose heuristic methods for solving problem (8). Chaneton and Vulcano (2009) propose a simpler alternative, where we make an itinerary available for sale provided its fare exceeds the sum of the bid prices on the flight legs it uses and there is sufficient capacity. 3 Randomized Dynamic Program In this section, we present a randomized dynamic program for the network revenue management problem with customer choice behavior. Letting U J = {U jt : j J, t T } be a sample of the customers utilities for the different itineraries at the different time periods, we solve the optimization problem { V t (x t U J ) = max Pr{j t = j S, U J } r j + V t+1 (x t ] } i L a ij e i U J ) V t+1 (x t U J ) S Q(x t ) +V t+1 (x t U J ), (9) with the boundary condition that V τ+1 ( U J ) = 0. We use the argument U J to emphasize that the solution to the above optimality equation depends on the sampled utilities U J and therefore is a random variable. We also note that U J only specifies the utilities for purchasing the itineraries; the utilities for not purchasing anything U φ = {U φt : t T } are still random. The following proposition shows that E{V t (x t U J )} is an upper bound on V t (x t ), where the expectation is with respect to U J. Proposition 1 We have V t (x t ) E{V t (x t U J )} for all t T. Note that Proposition 1 implies that V 1 (c) E{V 1 (c U J )} and so we get an upper bound on the optimal expected revenue by solving problem (9). Besides giving an upper bound on the value function, the randomized dynamic program also simplifies the optimization problem by reducing the size of the action space. We show below that instead of optimizing over subsets of itineraries, it is sufficient to optimize over the individual itineraries. We introduce some notation first. We let p jt (U J ) = Pr{j t = j {j}, U J } = Pr{U jt > U φt U J } 7
8 be the probability that the customer purchases itinerary j when it is the only itinerary that is offered at time period t. Note that the last equality follows from the fact that the customer will purchase the itinerary only if its utility exceeds the utility of not purchasing anything. We use the argument U J to emphasize that this probability is conditional on the sampled utilities. The following lemma shows that at each time period, we can solve an optimization problem involving J decision variables as opposed to 2 J decision variables. Lemma 2 Consider the optimization problem { V t (x t U J ) = max p jt (U J )y jt r j + V t+1 (x t i L a ij e i U J ) V t+1 (x t U J )] } + V t+1 (x t U J ), subject to a ij y jt x it i L (10) (11) y jt 1 (12) y jt {0, 1} j J, (13) with the boundary condition V τ+1 ( U J ) = 0. We have V t (x t U J ) = V t (x t U J ) for all x t, t T. Although the number of decision variables in problem (10)(13) is manageable, the size of the state space is still exponential in the capacities of the flight legs. On the other hand, noting that the decision variables in problem (10)(13) are only over the itineraries, this problem has a similar structure to the network revenue management problem with independent demand. This allows us to use approximation ideas developed for the independent demand setting to reduce the complexity of the state space. We present two approximation methods in the following section. 4 Relaxations of the Randomized Dynamic Program In this section, we describe two tractable relaxations of problem (10)(13). The first method is based on relaxing the flight leg capacity constraints using Lagrange multipliers. This yields an upper bound on the value function of the randomized dynamic program. We find the set of Lagrange multipliers which yields the tightest upper bound by solving a linear program. This idea is similar to that pursued in Kunnumkal and Topaloglu (2010a). The second method we propose is based on solving a perfect hindsight relaxation, where we have access to the customers utilities for not purchasing anything also. This method is similar to the randomized linear programming method of Talluri and van Ryzin (1999). We note that other approximation methods developed for the network revenue management problem with independent demand can also be applied to problem (10)(13). In this paper, we particularly focus on the above mentioned two methods because they involve solving linear programs, which can be done quickly and efficiently. Speed is an important factor since we have to resolve the problems for many different samples. 8
9 4.1 Capacity Relaxation Letting Y = {y {0, 1} J : y j 1} and y t = {y jt : j J }, we consider relaxing constraints (11) by introducing Lagrange multipliers λ it and solve the optimization problem { V t (x t U J, λ) = max p jt (U J )y jt r j + V t+1 (x t ] i L a ij e i U J, λ) V t+1 (x t U J, λ) y t Y } + i L λ it (x it a ij y jt ) + V t+1 (x t U J, λ) (14) with the boundary condition that V τ+1 ( U J, λ) = 0. The following proposition shows that as long as the Lagrange multipliers are nonnegative, V t (x t U J, λ) is an upper bound on V t (x t U J ). Proposition 3 If λ = {λ it : i L, t T } 0, then we have V t (x t U J ) V t (x t U J, λ). Note that Propositions 1 and 3 together imply that as long as the Lagrange multipliers are nonnegative, we have V 1 (c) E{V 1 (c U J, λ)}. So we are naturally interested in finding the set of Lagrange multipliers that gives the tightest upper bound. solving the problem That is, for each sample U J, we are interested in min V 1(c U J, λ). λ 0 We next show that the above minimization problem reduces to solving a linear program and therefore is tractable. We begin with the following result, which gives a closed form expression for V t (x t U J, λ). Lemma 4 We have V t (x t U J, λ) = τ s=t Λ s + i L ( τ λ is )x it, where Λ t = max { p jt (U J ) r j i L a ij( τ s=t+1 λ is) ] i L a ijλ it } + and we use { } + = max{0, }. s=t Using the result in Lemma 4, we have that the problem min λ 0 V 1 (c U J, λ) can be solved as the linear program z CR (U J ) = min t T Λ t + ( i L t T λ it )c i subject to Λ t + ] a ij λ it + p jt (U J ) a ij (λ i,t λ iτ ) r j p jt (U J ) j J, t T i L Λ t 0 t T λ it 0 i L, t T, i L 9
10 with the understanding that λ i,τ+1 = 0. Taking the dual of this linear program, we get z CR (U J ) = max t T subject to r j p jt (U J )y jt (15) a ij y jt + ] a ij pj1 (U J )y j p j,t 1 (U J )y j,t 1 ci i L, t T(16) y jt 1 t T (17) y jt 0 j J, t T, (18) with the understanding that y j0 = 0. In the above linear program, we can interpret the decision variable y jt as the frequency with which we offer itinerary j for sale at time period t. Since p jt (U J )y jt represents the expected sales of itinerary j at time period t, we can interpret the first set of constraints as saying that the capacity consumed by the itineraries offered at time period t should not exceed the expected capacity consumed up to time period t, which is c i a ij pj1 (U J )y j p j,t 1 (U J )y j,t 1 ]. We emphasize that the expectations are conditional on the sampled utilities U J. The second set of constraints ensure that the total frequency with which we offer the individual itineraries at each time period is at most one. Letting ˆλ(U J ) = argmin λ 0 V 1 (c U J, λ), we have that E{V 1 (c U J, ˆλ(U J ))} = E{z CR (U J )} is an upper bound on the optimal expected revenue. Letting ˆλ it = E{ˆλ it (U J )}, we use τ s=t ˆλ is as the bid price of flight leg i at time period t. We approximate V t (x t ) by i L τ s=t ˆλ is x it and solve the problem max S Q(x t) Pr{j t = j S} r j a ij i L τ ] ˆλ is s=t (19) to decide on the set of itineraries to offer at time period t. As it becomes difficult to analytically compute the expectations E{z CR (U J )} and E{ˆλ it (U J )}, we resort to Monte Carlo simulation to estimate these quantities. In particular, we generate K samples of the customers utilities for the different itineraries UJ 1,..., U J K where U J k = {U jt k : j J, t T } are the utilities generated in the kth sample. We solve linear program (15)(18) for each sample. Letting z CR (UJ k ) denote the optimal objective value and {ˆλ k it : i L, t T } denote optimal values of the dual variables corresponding to constraints (16), we use K k=1 z CR(UJ k )/K and K ˆλ k=1 k it /K as the sample estimates of E{z CR(U J )} and ˆλ it = E{ˆλ it (U J )}, respectively. 4.2 Perfect Hindsight Relaxation We consider another relaxation of problem (9) where we allow access to the customers utilities for not purchasing anything as well. In particular, letting U J = {U jt : j J, t T } be a sample of the customers utilities for the different itineraries at the different time periods and U φ = {U φt : t T } be a sample of the customers utilities for not purchasing anything and U = U J U φ, we solve the 10
11 optimization problem V t (x t U) = max S Q(x t ) { Pr{j t = j S, U} r j + V t+1 (x t ] } i L a ij e i U) V t+1 (x t U) +V t+1 (x t U). (20) Note that we can interpret problem (20) as determining the set of itineraries to offer at each time period after knowing the entire sample path: the customers utilities for the different itineraries as well as for not purchasing anything. Not surprisingly, it also gives an upper bound on problem (9). Proposition 5 We have V t (x t U J ) E{V t (x t U) U J } for all t T. Proposition 5 together with Proposition 1 imply that V 1 (c) E{V 1 (c U)}. Therefore, we obtain another upper bound on the optimal expected revenue by solving problem (20). Noting that Pr{j t = j S, U} = 1(U jt > U φt ) if j = argmax k S {U kt } and is zero otherwise, we have { V t (x t U) = 1(U jt > U φt ) r j + V t+1 (x t i L a ij e i U) V t+1 (x t U)] } + V t+1 (x t U). max j Q(x t) It follows that we can solve problem (20) as the following linear binary integer program: z P H (U) = max t T subject to r j 1(U jt > U φt )y jt (21) a ij 1(U jt > U φt )y jt c i i L (22) t T y jt 1 t T (23) y jt {0, 1} j J, t T. (24) In the above problem, the decision variable y jt indicates whether we offer itinerary j at time period t. The first set of constraints ensure that the total capacity consumed by the itinerary requests on each flight leg does not exceed its available capacity. The second set of constraints ensure that we offer at most one itinerary at each time period. We use E{V 1 (c U)} = E{z P H (U)} as an upper bound on the optimal expected revenue. In order to obtain a control policy, we solve the linear programming relaxation of problem (21)(24). Letting ˆρ(U) = {ˆρ i (U) : i L} denote the optimal values of the dual variables corresponding to constraints (22), we use ˆρ i = E{ˆρ i (U)} as the bid price of flight leg i. We approximate V t (x t ) by i L ˆρ ix it and solve the problem max S Q(x t ) Pr{j t = j S} r j ] a ij ˆρ i i L (25) 11
12 to decide on the set of itineraries to offer at time period t. It again becomes difficult to analytically compute E{z P H (U)} and E{ˆρ i (U)} and so we resort to Monte Carlo simulation. In particular, we generate K samples of the customers utilities for the different itineraries as well as not purchasing anything U 1,..., U K where U k = {Ujt k : j J {φ}, t T } are the utilities generated in the kth sample. We solve problem (21)(24) for each sample. Letting z P H (U k ) denote the optimal objective value, we use K k=1 z P H(U k )/K as the sample estimate of E{z P H (U)}. Letting {ˆρ k i : i L} denote the optimal values of the dual variables corresponding to constraints (22) in the linear programming relaxation of problem (21)(24), we use K k=1 ˆρk i /K as an estimate of ˆρ i = E{ˆρ i (U)}. We close this section with a comment on the upper bounds obtained by problems (4)(7), (15)(18) and (21)(24). It turns out that none of the upper bounds uniformly dominates the other. For example, consider a revenue management problem on a single flight leg for a single time period where we have two itineraries with r 1 = 11 and r 2 = 5. We have a single unit of capacity on the flight leg and each itinerary consumes one unit of capacity. Furthermore, we have U φ1 = 0 with probability 1. On the other hand, we have (1, 1) with probability 1/3 (U 11, U 21 ) = ( 1, 1) with probability 1/3 (1, 2) with probability 1/3. It is easy to verify that we have V 1 (c) = z CDLP = 22/3, while we have E{z P H (U)} = E{z CR (U J )} = 9. So, in this case we have that z CDLP < E{z P H (U)} = E{z CR (U J )}. It is also possible to come up with examples where the direction of the inequalities is reversed. On the other hand, in our computational experiments that we present next, we find that the capacity relaxation method consistently generates tighter upper bounds than the perfect hindsight relaxation method which in turn is tighter than the choice based deterministic linear program. 5 Computational Experiments In this section, we numerically compare the performance of the choice based deterministic linear program, the capacity relaxation method and the perfect hindsight relaxation method. We first describe the benchmark solution methods. After that we present our experimental setup and the results of the numerical study. Choice Based Deterministic Linear Program (CDLP): This is the solution method that we describe in Section 2. In our practical implementation, we divide the booking horizon into five equal segments. At the beginning of each segment, we solve problem (4)(7) after replacing the right hand side of equation (5) with the remaining capacities on the flight legs and the set of time periods T with the current set of remaining time periods. We get a fresh set of optimal dual values {ˆπ i : i L} and we plug them into decision rule (8) to decide on the set of itineraries to offer. We continue to use this decision rule until the beginning of the next segment, where we resolve problem (4)(7). Capacity Relaxation (CR): This is the solution method that we describe in Section 4.1. In our practical implementation, we divide the booking horizon into five equal segments. At the beginning of each 12
13 segment, we solve problem (15)(18) after replacing the right hand side of equation (16) with the remaining capacities on the flight legs and the set of time periods T with the current set of remaining time periods. We repeat this for K samples to get a fresh set of dual values {ˆλ k it : i L, t T, k K} and use these in decision rule (19) to decide on the set of itineraries to offer. We continue to use this decision rule until the beginning of the next segment, where we resolve problem (15)(18). We use K = 100 in our computational experiments. Increasing the value of K further did not result in any noticeable changes in performance. Perfect Hindsight Relaxation (PH): This is the solution method that we describe in Section 4.2. As with CDLP and CR, in our practical implementation, we divide the booking horizon into five equal segments. At the start of each segment, we refresh our bid prices by solving the linear programming relaxation of problem (21)(24) after replacing the right hand side of equation (22) with the remaining capacities on the flight legs and the set of time periods T with the current set of remaining time periods. We repeat this for K samples and use the fresh set of optimal dual values {ˆρ k i : i L, k K} in decision rule (25) to decide on the set of itineraries to offer. We continue to use this decision rule until the beginning of the next segment, where we again resolve problem (21)(24). As in CR, we use K = 100 in our computational experiments. We note that all of the above mentioned benchmark methods obtain bid prices that are capacity independent, in that they do not naturally change with the capacities on the flight legs. It is possible to obtain capacity dependent bid prices by using the optimal dual values obtained by the benchmark methods in a dynamic programming decomposition scheme as suggested by Liu and van Ryzin (2008) or Zhang (2011). We do not pursue that here for a number of reasons. Capacity dependent bid prices typically come with a higher overhead, both in terms of computation and implementation. Numerical studies also indicate that the performance gap between capacity independent bid prices remains intact when we use them in a dynamic programming decomposition scheme; see for example Kunnumkal and Topaloglu (2010c). We test the performance of the benchmark solution methods on two groups of test problems. The first group involves an airline network with a single hub serving multiple spokes, while the second group of test problems have an airline network with two hubs serving multiple spokes. Our test problems closely parallel those in Kunnumkal and Topaloglu (2010b). 5.1 Airline Network with a Single Hub We consider an airline network with a single hub that serves N spokes. Half of the spokes have two flights to the hub, while the remaining half have two flights from the hub. The total number of flights is 2N. Figure 1 shows the structure of the airline network with N = 8. There are four itineraries between each spoketohub and hubtospoke origin destination pair. On the other hand, we have eight itineraries between each spoketospoke origin destination pair, so that the total number of itineraries is 2N(N + 2). Half of these itineraries are high fare itineraries while the other half are low fare itineraries. We let γ denote the ratio between the high fare and the low fare. 13
14 Each origin destination pair is associated with a customer segment. We let K denote the set of customer segments. At each time period a customer from segment l K arrives with probability λ l. An arriving customer is interested only in the set of itineraries connecting the origin destination pair that it is associated with. Therefore, the consideration sets of the different customer segments are disjoint. Customer choice is governed by the multinomial logit model. In the multinomial logit model, the utility for purchasing itinerary j that is in the consideration set of customer segment l is given by U ljt = u ljt + ξ ljt, where u ljt is a constant called the nominal utility and ξ ljt is a Gumbel random variable with mean zero and scale parameter one. The utility for not purchasing anything for customer segment l is U lφt = u lφt + ξ lφt, where u lφt is the nominal utility for not purchasing anything and ξ lφt is a Gumbel random variable with mean zero and scale parameter one. The random variables {ξ ljt : j J {φ}, t T } are independent; see BenAkiva and Lerman (1994). We measure the tightness of the leg capacities in the same manner as Zhang and Adelman (2009). Letting S t = argmax S J r j Pr{j t = j S} be the offer set that maximizes expected revenue at time period t when there is ample capacity on all the flight legs, we use l K α = λ l t T i L a ij Pr{j t = j St } i L c, i to measure the tightness of the leg capacities. We have T = 200 time periods in all of our test problems. We vary N, γ and α to obtain different test problems. We label our test problems by the triplet (N, γ, α) {8, 10, 12} {1.5, 3} {1.3, 1.6}, where N is the number of spokes, γ is the ratio between the high and low fare itineraries and α measures the tightness of the leg capacities. This gives us a total of twelve test problems. Table 1 compares the upper bounds obtained by CR, PH and CDLP. The first column in this table gives the characteristics of the problem by using (N, γ, α). The second, third and fourth columns, respectively, give the upper bounds obtained by CR, PH and CDLP. The fifth column gives the percentage gap between the upper bounds obtained by PH and CR, while the last column gives the percentage gap between the upper bounds obtained by CDLP and CR. CR performs consistently well in our computational experiments and we use CR as a benchmark. In the last two columns, a indicates that the gap is significant at the 95% level, while a indicates that the gap is not significant at the 95% level. We observe that CR generates significantly tighter upper bounds than PH and CDLP. On average, the upper bounds obtained by CR are about 2% tighter than PH and 8% tighter than CDLP. Table 2 compares the total expected revenues obtained by CR, PH and CDLP. We evaluate the expected revenues by simulation and use common random numbers in our simulations. The columns have a similar interpretation as in Table 1 except that they give the expected revenues obtained by the three methods. The last two columns include a if CR does better than the respective solution method at the 95% level, a otherwise and a if there does not exist a statistically significant difference between the two. The average gap between the total expected revenues obtained by CR and CDLP is around 2%. The performance gaps are statistically significant in ten out of the twelve test problems. The performance gap between CR and CDLP seems to increase with the fare ratio and the tightness of the leg capacities. The performance gaps between CR and PH are small in most 14
15 cases, although we observe one instance where PH performs about 1% better than CR. PH performs significantly better than CDLP. The average gap between the total expected revenues obtained by PH and CDLP is around 2%. 5.2 Airline Network with Two Hubs We consider an airline network with two hubs that serve N spokes in total. Half of the spokes have two flights to the first hub, while the other half have two flights from the second hub. In addition, there are four flights from the first to the second hub. The total number of flights is 2N + 4. Figure 2 shows the structure of the airline network with N = 8. We randomly sample from the set of all the possible itineraries so that the total number of itineraries is around 4N 2. Half of these itineraries are high fare itineraries while the other half are low fare itineraries. Similar to the test problems with a single hub, each origin destination pair is associated with a customer segment. An arriving customer belongs to one of the segments and is interested only in the set of itineraries connecting the origin destination pair that it is associated with. We continue to assume that customer choice is governed by the multinomial logit model with disjoint consideration sets. We label the test problems by the triplet (N, γ, α) {4, 6, 8} {1.5, 3} {1.3, 1.6}, which gives us a total of twelve test problems. Table 3 compares the upper bounds obtained by CR, PH and CDLP. The columns have the same interpretation as in Table 1. The results display the same trends that we observed for the airline network with a single hub. CR consistently generates the tightest upper bounds, followed by PH and CDLP. On average, the upper bounds obtained by CR are about 2% tighter than PH and 9% tighter than CDLP. Table 4 compares the total expected revenues obtained by CR, PH and CDLP. CR generates significantly higher revenues than CDLP. The average gap between the total expected revenues obtained by CR and CDLP is around 2%, although we observe test problems where the gap is as high as 5%. We find one test problem where the performance gap between PH and CR is around 2.5%, but the gaps are quite small and insignificant in the remaining cases. The average performance gap between PH and CDLP is around 2%. The ratio between the high and low fares and the tightness of the leg capacities seem to be two factors which contribute to increasing the performance gaps between CDLP and the other two solution methods. Problems with large differences between the high and low fares and tight leg capacities tend to be more difficult to solve, because the consequences of offering the wrong set of itineraries tend to be more severe. It is therefore encouraging that CR and PH provide good performance for such test problems. All of the computational experiments are carried out on a Pentium Core 2 Duo desktop with 3 GHz CPU and 3 GB RAM running Windows XP. The running time of CDLP is of the order of seconds. For K = 100 samples, the running time of PH is of the order of seconds, while that of CR is in minutes. CR takes about a minute and a half to solve the largest test problem. 15
16 6 Conclusions We presented new methods to obtain upper bounds and bid prices for the network revenue management problem with customer choice behavior. The starting point for our methods is a dynamic programming approximation that we solve for a sample of the customers utilities for the different itineraries. An attractive feature of this randomized dynamic program is that the number of decision variables is linear in the number of itineraries. As a result, we are able to reduce the complexity of the action space. We build on this randomized dynamic program to obtain two tractable approximation methods. The first method that we propose involves relaxing the flight leg capacity constraints using Lagrange multipliers. The second method involves solving a perfect hindsight relaxation. We showed that both methods give upper bounds on the optimal expected total revenue. Our methods may also be appealing from a practical standpoint as they involve solving only linear programs. Computational experiments indicate that our methods can significantly improve upon the upper bounds and expected revenues obtained by the choice based deterministic linear program. Appendix Proof of Proposition 1 We show the result by induction over the time periods. It is easy to show that the result holds at time period τ. Assuming the result holds at time period t+1, we show that it holds at time period t. Letting ˆS be an optimal solution for problem (3), we note that ˆS is feasible for problem (9). We also note that for a given offer set S, Pr{j t = j S, U J } is a function of {U jt : j J }, while V t+1 (x t U J ) is a function of {U js : j J, s {t + 1,..., T }}. Since the random variables {U jt : j J } are independent across time, it follows that { } E Pr{j t = j S, U J }V t+1 (x t U J ) = E { Pr{j t = j S, U J } } E { V t+1 (x t U J ) }, (26) where the expectation is with respect to U J. Therefore, we have E { V t (x t U J ) } E { Pr{j t = j ˆS, U J } } r j + E { V t+1 (x t i L a ij e i U J ) }] + 1 E { Pr{j t = j ˆS, U J } }] E { V t+1 (x t U J ) } Pr{j t = j ˆS} r j + V t+1 (x t ] i L a ij e i ) + 1 Pr{j t = j ˆS} ] V t+1 (x t ) = V t (x t ), where the first inequality uses (26) and the fact that ˆS is a feasible but not necessarily optimal solution to problem (9) and the second inequality follows from the induction assumption and the fact that { } { Pr{j t = j S} = E 1(U jt = max {U kt}) = E E { 1(U jt = max {U } } kt}) U J = E { Pr{j t = j S, U J } }. k S {φ} k S {φ} 16
17 Proof of Lemma 2 We show the result by induction over the time periods. It is easy to show that the result holds at time period τ. Assuming the result holds at time period t + 1, we show that it holds at time period t. We first show that V t (x t U J ) V t (x t U J ). Let y t = { y jt : j J } be an optimal solution to problem (10)(13) and let S = {j J : y jt = 1}. Note that since y t satisfies (11)(13), the offer set S is feasible for problem (9). We have V t (x t U J ) = p jt (U J ) y jt r j + V t+1 (x t i L a ij e i U J ) V ] t+1 (x t U J ) + V t+1 (x t U J ) = Pr{j t = j S, U J } r j + V t+1 (x t ] i L a ij e i U J ) V t+1 (x t U J ) + V t+1 (x t U J ) V t (x t U J ), where the first equality follows from the optimality of y t, the second equality follows from the fact that S 1 and Pr{j t = j S, U J } = 0 for j / S and the induction assumption. The inequality holds since S is feasible for problem (9). This implies V t (x t ) V t (x t ). To show the reverse inequality, let ˆS be the optimal solution to problem (9), ˆȷ = argmax k ˆS{U kt } and ˆy t = {ˆy jt : j J } with ˆy jt = 1 for j = ˆȷ and ˆy jt = 0 for j J {ˆȷ}. Note that since ˆS Q(x t ), we have that ˆy t satisfies constraints (11)(13). We have Pr{j t = j ˆS, U J } = Pr{U jt > U φt U J } = p jt (U J ) for j = ˆȷ. On the other hand, since U jt < Uˆȷt for j ˆS {ˆȷ}, we have Pr{j t = j ˆS, U J } = 0 for j ˆS {ˆȷ}. Also, note that Pr{j t = j ˆS, U J } = 0 for j / ˆS. Using the above facts and the optimality of ˆS, we have V t (x t U J ) = Pr{j t = j ˆS, U J } r j + V t+1 (x t ] i L a ij e i U J ) V t+1 (x t U J ) + V t+1 (x t U J ) = pˆȷt (U J ) rˆȷ + V t+1 (x t ] i L a iˆȷ e i U J ) V t+1 (x t U J ) + V t+1 (x t U J ) = p jt (U J )ˆy jt r j + V t+1 (x t i L a ij e i U J ) V ] t+1 (x t U J ) + V t+1 (x t U J ) V t (x t U J ), where the last equality follows from the induction assumption and the fact that ˆy jt = 0 for j J {ˆȷ} and the inequality holds since ˆy t is feasible for problem (10)(13). Proof of Proposition 3 We show the result by induction over the time periods. It is easy to show that the result holds at time period τ. Assuming the result holds at time period t + 1, we show that it holds at time period t. Using the equivalent representation of V t (x t U J ) in Lemma 2 and letting y t = { y jt : j J } be an optimal 17
18 solution to problem (10)(13), we have V t (x t U J ) = p jt (U J ) y jt r j + V t+1 (x t ] i L a ij e i U J ) + i L + 1 ] p jt(u J ) y jt V t+1 (x t U J ) p jt (U J ) y jt r j + V t+1 (x t ] i L a ij e i U J, λ) + 1 ] p jt(u J ) y jt V t+1 (x t U J, λ) λ it (x it a ij y jt ) V t (x t U J, λ), where the first inequality uses the induction assumption and the facts that since y t satisfies constraints (11)(13) and λ 0, we have 1 p jt(u J ) y jt ] 0 and λ it (x it a ij y jt ) 0. The last inequality holds since y t is a feasible solution to problem (14). Proof of Lemma 4 We show the result by induction over the time periods. It is easy to show that the result holds at time period τ. Assuming the result holds at time period t + 1, we show that it holds at time period t. We have V t (x t U J, λ) = max y t Y + = max y t Y = { p jt (U J )y jt rj a ij ( i L τ Λ s + s=t+1 i L { τ s=t Λ s + i L ( τ s=t+1 λ is )x it y jt ( p jt (U J ) r j i L a ij ( ( τ λ is )x it, s=t τ s=t+1 λ is ) ] + i L λ it (x it τ s=t+1 λ is ) ] i L a ij λ it )} + } a ij y jt ) τ s=t+1 Λ s + i L ( τ λ is )x it where the first { equality follows from the induction assumption and the last equality uses the fact ( that max yt Y y jt p jt (U J ) r j i L a ij( τ s=t+1 λ is) ] )} i L a ijλ it = max {p jt (U J ) r j i L a ij( τ s=t+1 λ is) ] } + i L a ijλ it = Λt. s=t Proof of Proposition 5 We show the result by induction over the time periods. It is easy to show that the result holds at time period τ. Assuming the result holds at time period t+1, we show that it holds at time period t. Letting ˆS be an optimal solution for problem (9), we note that ˆS is feasible for problem (20). We also note that for a given offer set S, Pr{j t = j S, U} is a function of {U jt : j J {φ}}, while V t+1 (x t U) is a function of {U js : j J {φ}, s {t + 1,..., T }}. Since the random variables {U jt : j J {φ}} are independent across time, it follows that { } E Pr{j t = j S, U}V t+1 (x t U) U J = E { } { } Pr{j t = j S, U} U J E Vt+1 (x t U) U J, (27) 18
19 where the expectation is with respect to U φ and we recall that U = U J U φ. Therefore, we have E { } V t (x t U) U J E { Pr{j t = j ˆS, } U} U J r j + E { V t+1 (x t i L a } ] ij e i U) U J + 1 E { Pr{j t = j ˆS, } ] U} U J E { } V t+1 (x t U) U J Pr{j t = j ˆS, U J } r j + V t+1 (x t ] i L a ij e i U J ) + 1 Pr{j t = j ˆS, ] U J } V t+1 (x t U J ) = V t (x t U J ), where the first inequality follows from (27) and the fact that ˆS is a feasible but not necessarily optimal solution to problem (20) and the second inequality follows from the induction assumption and the fact that { Pr{j t = j S, U J } = E 1(U jt > } { max {U kt}) U J = E E { 1(U jt > k S {φ} max {U kt}) U }} k S {φ} = E { Pr{j t = j S, U} }. References BenAkiva, M. and Lerman, S. (1994), Discrete Choice Analysis: Theory and Applications to Travel Demand, The MIT Press, Cambridge, MA. Bront, J. J. M., MendezDiaz, I. and Vulcano, G. (2009), A column generation algorithm for choicebased network revenue management, Operations Research 57, Chaneton, J. and Vulcano, G. (2009), Computing bidprices for revenue management under customer choice behavior, Working paper, New York University, New York City, NY. Gallego, G., Ratliff, R. and Shebalov, S. (2010), A general attraction model and an efficient formulation for the network revenue management problem, Working paper, Columbia University, New York. Kunnumkal, S. and Topaloglu, H. (2008), A refined deterministic linear program for the network revenue management problem with customer choice behavior, Naval Research Logistics Quarterly 55, Kunnumkal, S. and Topaloglu, H. (2010a), Computing timedependent bid prices in network revenue management problems, Transportation Science 44(1), Kunnumkal, S. and Topaloglu, H. (2010b), A new dynamic programming decomposition method for the network revenue management problem with customer choice behavior, Production and Operations Management 19(5), Kunnumkal, S. and Topaloglu, H. (2010c), A randomized linear program for the network revenue management problem with customer choice behavior, Journal of Revenue and Pricing Management. (to appear). Liu, Q. and van Ryzin, G. (2008), On the choicebased linear programming model for network revenue management, M&SOM 10(2),
20 Mahajan, S. and van Ryzin, G. (2001), Stocking retail assortments under dynamic consumer substitution, Operations Research 49(3), Meissner, J. and Strauss, A. K. (2008), Network revenue management with inventory sensitive bid prices and customer choice, Working paper, Department of Management Science, Lancaster University. Meissner, J. and Strauss, A. K. (2010), Choicebased network revenue management under weak market segmentation, Working paper, Department of Management Science, Lancaster University. Meissner, J., Strauss, A. K. and Talluri, K. (2011), An enhanced concave program relaxation for choice network revenue management, Working paper, Universitat Pompeu Fabra, Barcelona, Spain. Talluri, K. (2010), A randomized concave programming method for choice network revenue management, Working paper, Universitat Pompeu Fabra, Barcelona, Spain. Talluri, K. and van Ryzin, G. (1999), A randomized linear programming method for computing network bid prices, Transportation Science 33(2), Talluri, K. and van Ryzin, G. (2004), The Theory and Practice of Revenue Management, Kluwer Academic Press. van Ryzin, G. and Vulcano, G. (2008), Computing virtual nesting controls for network revenue management under customer choice behavior, M&SOM 10, Zhang, D. (2011), An improved dynamic programming decomposition approach for network revenue management, M&SOM 13, Zhang, D. and Adelman, D. (2009), An approximate dynamic programming approach to network revenue management with customer choice, Transportation Science 43, Zhang, D. and Cooper, W. L. (2005), Revenue management for parallel flights with customerchoice behavior, Operations Research 53,
An Improved Dynamic Programming Decomposition Approach for Network Revenue Management
An Improved Dynamic Programming Decomposition Approach for Network Revenue Management Dan Zhang Leeds School of Business University of Colorado at Boulder May 21, 2012 Outline Background Network revenue
More informationRole of Stochastic Optimization in Revenue Management. Huseyin Topaloglu School of Operations Research and Information Engineering Cornell University
Role of Stochastic Optimization in Revenue Management Huseyin Topaloglu School of Operations Research and Information Engineering Cornell University Revenue Management Revenue management involves making
More informationA Randomized Linear Programming Method for Network Revenue Management with ProductSpecific NoShows
A Randomized Linear Programming Method for Network Revenue Management with ProductSpecific NoShows Sumit Kunnumkal Indian School of Business, Gachibowli, Hyderabad, 500032, India sumit kunnumkal@isb.edu
More informationAppointment Scheduling under Patient Preference and NoShow Behavior
Appointment Scheduling under Patient Preference and NoShow Behavior Jacob Feldman School of Operations Research and Information Engineering, Cornell University, Ithaca, NY 14853 jbf232@cornell.edu Nan
More informationRevenue Management for Transportation Problems
Revenue Management for Transportation Problems Francesca Guerriero Giovanna Miglionico Filomena Olivito Department of Electronic Informatics and Systems, University of Calabria Via P. Bucci, 87036 Rende
More informationRobust Assortment Optimization in Revenue Management Under the Multinomial Logit Choice Model
Robust Assortment Optimization in Revenue Management Under the Multinomial Logit Choice Model Paat Rusmevichientong Huseyin Topaloglu May 9, 2011 Abstract We study robust formulations of assortment optimization
More informationCargo Capacity Management with Allotments and Spot Market Demand
Submitted to Operations Research manuscript OPRE200808420.R3 Cargo Capacity Management with Allotments and Spot Market Demand Yuri Levin and Mikhail Nediak School of Business, Queen s University, Kingston,
More informationUpgrades, Upsells and Pricing in Revenue Management
Submitted to Management Science manuscript Upgrades, Upsells and Pricing in Revenue Management Guillermo Gallego IEOR Department, Columbia University, New York, NY 10027, gmg2@columbia.edu Catalina Stefanescu
More informationReSolving Stochastic Programming Models for Airline Revenue Management
ReSolving Stochastic Programming Models for Airline Revenue Management Lijian Chen Department of Industrial, Welding and Systems Engineering The Ohio State University Columbus, OH 43210 chen.855@osu.edu
More informationCargo Capacity Management with Allotments and Spot Market Demand
OPERATIONS RESEARCH Vol. 60, No. 2, March April 2012, pp. 351 365 ISSN 0030364X (print) ISSN 15265463 (online) http://dx.doi.org/10.1287/opre.1110.1023 2012 INFORMS Cargo Capacity Management with Allotments
More informationPrinciples of demand management Airline yield management Determining the booking limits. » A simple problem» Stochastic gradients for general problems
Demand Management Principles of demand management Airline yield management Determining the booking limits» A simple problem» Stochastic gradients for general problems Principles of demand management Issues:»
More informationIntroduction to Support Vector Machines. Colin Campbell, Bristol University
Introduction to Support Vector Machines Colin Campbell, Bristol University 1 Outline of talk. Part 1. An Introduction to SVMs 1.1. SVMs for binary classification. 1.2. Soft margins and multiclass classification.
More informationThe Trip Scheduling Problem
The Trip Scheduling Problem Claudia Archetti Department of Quantitative Methods, University of Brescia Contrada Santa Chiara 50, 25122 Brescia, Italy Martin Savelsbergh School of Industrial and Systems
More informationmax cx s.t. Ax c where the matrix A, cost vector c and right hand side b are given and x is a vector of variables. For this example we have x
Linear Programming Linear programming refers to problems stated as maximization or minimization of a linear function subject to constraints that are linear equalities and inequalities. Although the study
More informationBranchandPrice Approach to the Vehicle Routing Problem with Time Windows
TECHNISCHE UNIVERSITEIT EINDHOVEN BranchandPrice Approach to the Vehicle Routing Problem with Time Windows Lloyd A. Fasting May 2014 Supervisors: dr. M. Firat dr.ir. M.A.A. Boon J. van Twist MSc. Contents
More informationECON20310 LECTURE SYNOPSIS REAL BUSINESS CYCLE
ECON20310 LECTURE SYNOPSIS REAL BUSINESS CYCLE YUAN TIAN This synopsis is designed merely for keep a record of the materials covered in lectures. Please refer to your own lecture notes for all proofs.
More informationCHAPTER 9. Integer Programming
CHAPTER 9 Integer Programming An integer linear program (ILP) is, by definition, a linear program with the additional constraint that all variables take integer values: (9.1) max c T x s t Ax b and x integral
More information2.3 Convex Constrained Optimization Problems
42 CHAPTER 2. FUNDAMENTAL CONCEPTS IN CONVEX OPTIMIZATION Theorem 15 Let f : R n R and h : R R. Consider g(x) = h(f(x)) for all x R n. The function g is convex if either of the following two conditions
More informationMotivated by a problem faced by a large manufacturer of a consumer product, we
A Coordinated Production Planning Model with Capacity Expansion and Inventory Management Sampath Rajagopalan Jayashankar M. Swaminathan Marshall School of Business, University of Southern California, Los
More informationOffline sorting buffers on Line
Offline sorting buffers on Line Rohit Khandekar 1 and Vinayaka Pandit 2 1 University of Waterloo, ON, Canada. email: rkhandekar@gmail.com 2 IBM India Research Lab, New Delhi. email: pvinayak@in.ibm.com
More informationSupport Vector Machines with Clustering for Training with Very Large Datasets
Support Vector Machines with Clustering for Training with Very Large Datasets Theodoros Evgeniou Technology Management INSEAD Bd de Constance, Fontainebleau 77300, France theodoros.evgeniou@insead.fr Massimiliano
More informationprinceton univ. F 13 cos 521: Advanced Algorithm Design Lecture 6: Provable Approximation via Linear Programming Lecturer: Sanjeev Arora
princeton univ. F 13 cos 521: Advanced Algorithm Design Lecture 6: Provable Approximation via Linear Programming Lecturer: Sanjeev Arora Scribe: One of the running themes in this course is the notion of
More informationA Lagrangian relaxation approach for network inventory control of stochastic revenue management with perishable commodities
Journal of the Operational Research Society (2006), 19 2006 Operational Research Society Ltd. All rights reserved. 01605682/06 $30.00 www.palgravejournals.com/jors A Lagrangian relaxation approach
More informationMinimize subject to. x S R
Chapter 12 Lagrangian Relaxation This chapter is mostly inspired by Chapter 16 of [1]. In the previous chapters, we have succeeded to find efficient algorithms to solve several important problems such
More informationA central problem in network revenue management
A Randomized Linear Programming Method for Computing Network Bid Prices KALYAN TALLURI Universitat Pompeu Fabra, Barcelona, Spain GARRETT VAN RYZIN Columbia University, New York, New York We analyze a
More informationFairness in Routing and Load Balancing
Fairness in Routing and Load Balancing Jon Kleinberg Yuval Rabani Éva Tardos Abstract We consider the issue of network routing subject to explicit fairness conditions. The optimization of fairness criteria
More informationChapter 15 Introduction to Linear Programming
Chapter 15 Introduction to Linear Programming An Introduction to Optimization Spring, 2014 WeiTa Chu 1 Brief History of Linear Programming The goal of linear programming is to determine the values of
More informationThe Conference Call Search Problem in Wireless Networks
The Conference Call Search Problem in Wireless Networks Leah Epstein 1, and Asaf Levin 2 1 Department of Mathematics, University of Haifa, 31905 Haifa, Israel. lea@math.haifa.ac.il 2 Department of Statistics,
More informationTwoStage Stochastic Linear Programs
TwoStage Stochastic Linear Programs Operations Research Anthony Papavasiliou 1 / 27 TwoStage Stochastic Linear Programs 1 Short Reviews Probability Spaces and Random Variables Convex Analysis 2 Deterministic
More information1 Limiting distribution for a Markov chain
Copyright c 2009 by Karl Sigman Limiting distribution for a Markov chain In these Lecture Notes, we shall study the limiting behavior of Markov chains as time n In particular, under suitable easytocheck
More informationMincost flow problems and network simplex algorithm
Mincost flow problems and network simplex algorithm The particular structure of some LP problems can be sometimes used for the design of solution techniques more efficient than the simplex algorithm.
More information10.1 Integer Programming and LP relaxation
CS787: Advanced Algorithms Lecture 10: LP Relaxation and Rounding In this lecture we will design approximation algorithms using linear programming. The key insight behind this approach is that the closely
More informationJUSTINTIME SCHEDULING WITH PERIODIC TIME SLOTS. Received December May 12, 2003; revised February 5, 2004
Scientiae Mathematicae Japonicae Online, Vol. 10, (2004), 431 437 431 JUSTINTIME SCHEDULING WITH PERIODIC TIME SLOTS Ondřej Čepeka and Shao Chin Sung b Received December May 12, 2003; revised February
More informationNotes from Week 1: Algorithms for sequential prediction
CS 683 Learning, Games, and Electronic Markets Spring 2007 Notes from Week 1: Algorithms for sequential prediction Instructor: Robert Kleinberg 2226 Jan 2007 1 Introduction In this course we will be looking
More informationThe MultiItem Capacitated LotSizing Problem With Safety Stocks In ClosedLoop Supply Chain
International Journal of Mining Metallurgy & Mechanical Engineering (IJMMME) Volume 1 Issue 5 (2013) ISSN 23204052; EISSN 23204060 The MultiItem Capacated LotSizing Problem Wh Safety Stocks In ClosedLoop
More informationA Production Planning Problem
A Production Planning Problem Suppose a production manager is responsible for scheduling the monthly production levels of a certain product for a planning horizon of twelve months. For planning purposes,
More informationProximal mapping via network optimization
L. Vandenberghe EE236C (Spring 234) Proximal mapping via network optimization minimum cut and maximum flow problems parametric minimum cut problem application to proximal mapping Introduction this lecture:
More informationLecture 3: Linear Programming Relaxations and Rounding
Lecture 3: Linear Programming Relaxations and Rounding 1 Approximation Algorithms and Linear Relaxations For the time being, suppose we have a minimization problem. Many times, the problem at hand can
More informationPreliminary Draft. January 2006. Abstract
Assortment Planning and Inventory Management Under Dynamic Stockoutbased Substitution Preliminary Draft Dorothée Honhon, Vishal Gaur, Sridhar Seshadri January 2006 Abstract We consider the problem of
More informationPractical Guide to the Simplex Method of Linear Programming
Practical Guide to the Simplex Method of Linear Programming Marcel Oliver Revised: April, 0 The basic steps of the simplex algorithm Step : Write the linear programming problem in standard form Linear
More informationA Branch and Bound Algorithm for Solving the Binary Bilevel Linear Programming Problem
A Branch and Bound Algorithm for Solving the Binary Bilevel Linear Programming Problem John Karlof and Peter Hocking Mathematics and Statistics Department University of North Carolina Wilmington Wilmington,
More informationThe Impact of Linear Optimization on Promotion Planning
The Impact of Linear Optimization on Promotion Planning Maxime C. Cohen Operations Research Center, MIT, Cambridge, MA 02139, maxcohen@mit.edu NgaiHang Zachary Leung Operations Research Center, MIT, Cambridge,
More informationInventory Management with Auctions and Other Sales Channels: Optimality of (s, S) Policies
Inventory Management with Auctions and Other Sales Channels: Optimality of (s, S) Policies Woonghee Tim Huh, Columbia University Ganesh Janakiraman, New York University Initial Version: December 17, 2004
More informationTHE DYING FIBONACCI TREE. 1. Introduction. Consider a tree with two types of nodes, say A and B, and the following properties:
THE DYING FIBONACCI TREE BERNHARD GITTENBERGER 1. Introduction Consider a tree with two types of nodes, say A and B, and the following properties: 1. Let the root be of type A.. Each node of type A produces
More informationTHEORY OF SIMPLEX METHOD
Chapter THEORY OF SIMPLEX METHOD Mathematical Programming Problems A mathematical programming problem is an optimization problem of finding the values of the unknown variables x, x,, x n that maximize
More informationOnline Network Revenue Management using Thompson Sampling
Online Network Revenue Management using Thompson Sampling Kris Johnson Ferreira David SimchiLevi He Wang Working Paper 16031 Online Network Revenue Management using Thompson Sampling Kris Johnson Ferreira
More informationA Statistical Modeling Approach to Airline Revenue. Management
A Statistical Modeling Approach to Airline Revenue Management Sheela Siddappa 1, Dirk Günther 2, Jay M. Rosenberger 1, Victoria C. P. Chen 1, 1 Department of Industrial and Manufacturing Systems Engineering
More informationOPRE 6201 : 2. Simplex Method
OPRE 6201 : 2. Simplex Method 1 The Graphical Method: An Example Consider the following linear program: Max 4x 1 +3x 2 Subject to: 2x 1 +3x 2 6 (1) 3x 1 +2x 2 3 (2) 2x 2 5 (3) 2x 1 +x 2 4 (4) x 1, x 2
More informationMarkov chains and Markov Random Fields (MRFs)
Markov chains and Markov Random Fields (MRFs) 1 Why Markov Models We discuss Markov models now. This is the simplest statistical model in which we don t assume that all variables are independent; we assume
More informationRecovery of primal solutions from dual subgradient methods for mixed binary linear programming; a branchandbound approach
MASTER S THESIS Recovery of primal solutions from dual subgradient methods for mixed binary linear programming; a branchandbound approach PAULINE ALDENVIK MIRJAM SCHIERSCHER Department of Mathematical
More informationDynamic Capacity Management with General Upgrading
Submitted to Operations Research manuscript (Please, provide the mansucript number!) Dynamic Capacity Management with General Upgrading Yueshan Yu Olin Business School, Washington University in St. Louis,
More information1 Portfolio mean and variance
Copyright c 2005 by Karl Sigman Portfolio mean and variance Here we study the performance of a oneperiod investment X 0 > 0 (dollars) shared among several different assets. Our criterion for measuring
More informationApplied Algorithm Design Lecture 5
Applied Algorithm Design Lecture 5 Pietro Michiardi Eurecom Pietro Michiardi (Eurecom) Applied Algorithm Design Lecture 5 1 / 86 Approximation Algorithms Pietro Michiardi (Eurecom) Applied Algorithm Design
More informationModern Optimization Methods for Big Data Problems MATH11146 The University of Edinburgh
Modern Optimization Methods for Big Data Problems MATH11146 The University of Edinburgh Peter Richtárik Week 3 Randomized Coordinate Descent With Arbitrary Sampling January 27, 2016 1 / 30 The Problem
More informationCHAPTER 17. Linear Programming: Simplex Method
CHAPTER 17 Linear Programming: Simplex Method CONTENTS 17.1 AN ALGEBRAIC OVERVIEW OF THE SIMPLEX METHOD Algebraic Properties of the Simplex Method Determining a Basic Solution Basic Feasible Solution 17.2
More informationApproximation Algorithms: LP Relaxation, Rounding, and Randomized Rounding Techniques. My T. Thai
Approximation Algorithms: LP Relaxation, Rounding, and Randomized Rounding Techniques My T. Thai 1 Overview An overview of LP relaxation and rounding method is as follows: 1. Formulate an optimization
More informationNan Kong, Andrew J. Schaefer. Department of Industrial Engineering, Univeristy of Pittsburgh, PA 15261, USA
A Factor 1 2 Approximation Algorithm for TwoStage Stochastic Matching Problems Nan Kong, Andrew J. Schaefer Department of Industrial Engineering, Univeristy of Pittsburgh, PA 15261, USA Abstract We introduce
More informationAppendix: Simple Methods for Shift Scheduling in MultiSkill Call Centers
MSOM.1070.0172 Appendix: Simple Methods for Shift Scheduling in MultiSkill Call Centers In Bhulai et al. (2006) we presented a method for computing optimal schedules, separately, after the optimal staffing
More informationModeling and Performance Evaluation of Computer Systems Security Operation 1
Modeling and Performance Evaluation of Computer Systems Security Operation 1 D. Guster 2 St.Cloud State University 3 N.K. Krivulin 4 St.Petersburg State University 5 Abstract A model of computer system
More informationRegression Using Support Vector Machines: Basic Foundations
Regression Using Support Vector Machines: Basic Foundations Technical Report December 2004 Aly Farag and Refaat M Mohamed Computer Vision and Image Processing Laboratory Electrical and Computer Engineering
More informationMessagepassing sequential detection of multiple change points in networks
Messagepassing sequential detection of multiple change points in networks Long Nguyen, Arash Amini Ram Rajagopal University of Michigan Stanford University ISIT, Boston, July 2012 Nguyen/Amini/Rajagopal
More informationMoral Hazard. Itay Goldstein. Wharton School, University of Pennsylvania
Moral Hazard Itay Goldstein Wharton School, University of Pennsylvania 1 PrincipalAgent Problem Basic problem in corporate finance: separation of ownership and control: o The owners of the firm are typically
More informationIncreasing for all. Convex for all. ( ) Increasing for all (remember that the log function is only defined for ). ( ) Concave for all.
1. Differentiation The first derivative of a function measures by how much changes in reaction to an infinitesimal shift in its argument. The largest the derivative (in absolute value), the faster is evolving.
More informationSchooling, Political Participation, and the Economy. (Online Supplementary Appendix: Not for Publication)
Schooling, Political Participation, and the Economy Online Supplementary Appendix: Not for Publication) Filipe R. Campante Davin Chor July 200 Abstract In this online appendix, we present the proofs for
More informationKey words. Mixedinteger programming, mixing sets, convex hull descriptions, lotsizing.
MIXING SETS LINKED BY BIDIRECTED PATHS MARCO DI SUMMA AND LAURENCE A. WOLSEY Abstract. Recently there has been considerable research on simple mixedinteger sets, called mixing sets, and closely related
More informationLecture 11: 01 Quadratic Program and Lower Bounds
Lecture :  Quadratic Program and Lower Bounds (3 units) Outline Problem formulations Reformulation: Linearization & continuous relaxation Branch & Bound Method framework Simple bounds, LP bound and semidefinite
More informationMinimizing the Number of Machines in a UnitTime Scheduling Problem
Minimizing the Number of Machines in a UnitTime Scheduling Problem Svetlana A. Kravchenko 1 United Institute of Informatics Problems, Surganova St. 6, 220012 Minsk, Belarus kravch@newman.basnet.by Frank
More informationScheduling Realtime Tasks: Algorithms and Complexity
Scheduling Realtime Tasks: Algorithms and Complexity Sanjoy Baruah The University of North Carolina at Chapel Hill Email: baruah@cs.unc.edu Joël Goossens Université Libre de Bruxelles Email: joel.goossens@ulb.ac.be
More informationScheduling Home Health Care with Separating Benders Cuts in Decision Diagrams
Scheduling Home Health Care with Separating Benders Cuts in Decision Diagrams André Ciré University of Toronto John Hooker Carnegie Mellon University INFORMS 2014 Home Health Care Home health care delivery
More informationScheduling Jobs and Preventive Maintenance Activities on Parallel Machines
Scheduling Jobs and Preventive Maintenance Activities on Parallel Machines Maher Rebai University of Technology of Troyes Department of Industrial Systems 12 rue Marie Curie, 10000 Troyes France maher.rebai@utt.fr
More information5 INTEGER LINEAR PROGRAMMING (ILP) E. Amaldi Fondamenti di R.O. Politecnico di Milano 1
5 INTEGER LINEAR PROGRAMMING (ILP) E. Amaldi Fondamenti di R.O. Politecnico di Milano 1 General Integer Linear Program: (ILP) min c T x Ax b x 0 integer Assumption: A, b integer The integrality condition
More information24. The Branch and Bound Method
24. The Branch and Bound Method It has serious practical consequences if it is known that a combinatorial problem is NPcomplete. Then one can conclude according to the present state of science that no
More informationAn important class of codes are linear codes in the vector space Fq n, where F q is a finite field of order q.
Chapter 3 Linear Codes An important class of codes are linear codes in the vector space Fq n, where F q is a finite field of order q. Definition 3.1 (Linear code). A linear code C is a code in Fq n for
More informationR u t c o r Research R e p o r t. A Method to Schedule Both Transportation and Production at the Same Time in a Special FMS.
R u t c o r Research R e p o r t A Method to Schedule Both Transportation and Production at the Same Time in a Special FMS Navid Hashemian a Béla Vizvári b RRR 32011, February 21, 2011 RUTCOR Rutgers
More informationOn spare parts optimization
On spare parts optimization Krister Svanberg Optimization and Systems Theory, KTH, Stockholm, Sweden. krille@math.kth.se This manuscript deals with some mathematical optimization models for multilevel
More informationA Constraint Programming based Column Generation Approach to Nurse Rostering Problems
Abstract A Constraint Programming based Column Generation Approach to Nurse Rostering Problems Fang He and Rong Qu The Automated Scheduling, Optimisation and Planning (ASAP) Group School of Computer Science,
More information15 Kuhn Tucker conditions
5 Kuhn Tucker conditions Consider a version of the consumer problem in which quasilinear utility x 2 + 4 x 2 is maximised subject to x +x 2 =. Mechanically applying the Lagrange multiplier/common slopes
More informationSection Notes 4. Duality, Sensitivity, Dual Simplex, Complementary Slackness. Applied Math 121. Week of February 28, 2011
Section Notes 4 Duality, Sensitivity, Dual Simplex, Complementary Slackness Applied Math 121 Week of February 28, 2011 Goals for the week understand the relationship between primal and dual programs. know
More informationLinear Programming. March 14, 2014
Linear Programming March 1, 01 Parts of this introduction to linear programming were adapted from Chapter 9 of Introduction to Algorithms, Second Edition, by Cormen, Leiserson, Rivest and Stein [1]. 1
More informationUsing the Simplex Method in Mixed Integer Linear Programming
Integer Using the Simplex Method in Mixed Integer UTFSM Nancy, 17 december 2015 Using the Simplex Method in Mixed Integer Outline Mathematical Programming Integer 1 Mathematical Programming Optimisation
More informationLecture 7: Approximation via Randomized Rounding
Lecture 7: Approximation via Randomized Rounding Often LPs return a fractional solution where the solution x, which is supposed to be in {0, } n, is in [0, ] n instead. There is a generic way of obtaining
More informationALMOST COMMON PRIORS 1. INTRODUCTION
ALMOST COMMON PRIORS ZIV HELLMAN ABSTRACT. What happens when priors are not common? We introduce a measure for how far a type space is from having a common prior, which we term prior distance. If a type
More informationPermutation Betting Markets: Singleton Betting with Extra Information
Permutation Betting Markets: Singleton Betting with Extra Information Mohammad Ghodsi Sharif University of Technology ghodsi@sharif.edu Hamid Mahini Sharif University of Technology mahini@ce.sharif.edu
More informationCyberSecurity Analysis of State Estimators in Power Systems
CyberSecurity Analysis of State Estimators in Electric Power Systems André Teixeira 1, Saurabh Amin 2, Henrik Sandberg 1, Karl H. Johansson 1, and Shankar Sastry 2 ACCESS Linnaeus Centre, KTHRoyal Institute
More informationMathematical finance and linear programming (optimization)
Mathematical finance and linear programming (optimization) Geir Dahl September 15, 2009 1 Introduction The purpose of this short note is to explain how linear programming (LP) (=linear optimization) may
More informationARTICLE IN PRESS. European Journal of Operational Research xxx (2004) xxx xxx. Discrete Optimization. Nan Kong, Andrew J.
A factor 1 European Journal of Operational Research xxx (00) xxx xxx Discrete Optimization approximation algorithm for twostage stochastic matching problems Nan Kong, Andrew J. Schaefer * Department of
More informationNo: 10 04. Bilkent University. Monotonic Extension. Farhad Husseinov. Discussion Papers. Department of Economics
No: 10 04 Bilkent University Monotonic Extension Farhad Husseinov Discussion Papers Department of Economics The Discussion Papers of the Department of Economics are intended to make the initial results
More informationS = {1, 2,..., n}. P (1, 1) P (1, 2)... P (1, n) P (2, 1) P (2, 2)... P (2, n) P = . P (n, 1) P (n, 2)... P (n, n)
last revised: 26 January 2009 1 Markov Chains A Markov chain process is a simple type of stochastic process with many social science applications. We ll start with an abstract description before moving
More informationThis article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
IEEE/ACM TRANSACTIONS ON NETWORKING 1 A Greedy Link Scheduler for Wireless Networks With Gaussian MultipleAccess and Broadcast Channels Arun Sridharan, Student Member, IEEE, C Emre Koksal, Member, IEEE,
More informationWeek 5 Integral Polyhedra
Week 5 Integral Polyhedra We have seen some examples 1 of linear programming formulation that are integral, meaning that every basic feasible solution is an integral vector. This week we develop a theory
More informationA simple criterion on degree sequences of graphs
Discrete Applied Mathematics 156 (2008) 3513 3517 Contents lists available at ScienceDirect Discrete Applied Mathematics journal homepage: www.elsevier.com/locate/dam Note A simple criterion on degree
More informationInner Product Spaces
Math 571 Inner Product Spaces 1. Preliminaries An inner product space is a vector space V along with a function, called an inner product which associates each pair of vectors u, v with a scalar u, v, and
More informationA Programme Implementation of Several Inventory Control Algorithms
BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume, No Sofia 20 A Programme Implementation of Several Inventory Control Algorithms Vladimir Monov, Tasho Tashev Institute of Information
More informationLecture 4: BK inequality 27th August and 6th September, 2007
CSL866: Percolation and Random Graphs IIT Delhi Amitabha Bagchi Scribe: Arindam Pal Lecture 4: BK inequality 27th August and 6th September, 2007 4. Preliminaries The FKG inequality allows us to lower bound
More informationThe Multiplicative Weights Update method
Chapter 2 The Multiplicative Weights Update method The Multiplicative Weights method is a simple idea which has been repeatedly discovered in fields as diverse as Machine Learning, Optimization, and Game
More informationMechanisms for Fair Attribution
Mechanisms for Fair Attribution Eric Balkanski Yaron Singer Abstract We propose a new framework for optimization under fairness constraints. The problems we consider model procurement where the goal is
More informationFinal Report. to the. Center for Multimodal Solutions for Congestion Mitigation (CMS) CMS Project Number: 2010018
Final Report to the Center for Multimodal Solutions for Congestion Mitigation (CMS) CMS Project Number: 2010018 CMS Project Title: Impacts of Efficient Transportation Capacity Utilization via MultiProduct
More informationSupport Vector Machines Explained
March 1, 2009 Support Vector Machines Explained Tristan Fletcher www.cs.ucl.ac.uk/staff/t.fletcher/ Introduction This document has been written in an attempt to make the Support Vector Machines (SVM),
More informationInteger Programming: Algorithms  3
Week 9 Integer Programming: Algorithms  3 OPR 992 Applied Mathematical Programming OPR 992  Applied Mathematical Programming  p. 1/12 DantzigWolfe Reformulation Example Strength of the Linear Programming
More informationA network flow algorithm for reconstructing. binary images from discrete Xrays
A network flow algorithm for reconstructing binary images from discrete Xrays Kees Joost Batenburg Leiden University and CWI, The Netherlands kbatenbu@math.leidenuniv.nl Abstract We present a new algorithm
More information