Econometric Institute Report No. 9703/A A NOTE ON A PROFIT MAXIMIZING LOCATION MODEL Shuzhong Zhang January 6, 1997 ABSTRACT In this paper we discuss a locational model with a pro t-maximizing objective. The model can be illustrated by the following situation. There is a set of potential customers in a given region. A rm enters the market and wants to sell a certain product to this set of customers. The location and demand of each potential customer are assumed to be known. In order to maximize its total pro t, the rm has to decide: 1) where to locate its distribution warehouse to serve the customers; 2) the price for its product. Due to existence of competition, each customer holds a reservation price for the product. This reservation price is a decreasing function in the distance to the warehouse. If the actual price is higher than the reservation price, then the customer will turn to some other supplier and hence is lost from the rm's market. The problem of the rm is to nd the best location for its warehouse and the best price for its product at the same time in order to maximize the total pro t. We show that this problem can be solved in polynomial time. Key words: location, pricing, pro t maximization. AMS subject classi cation: 90B60, 90B85.  Econometric Institute, Erasmus University Rotterdam, The Netherlands. 1 Introduction Traditionally the literature on locational models under the microeconomic marketing environment is quite limited. See [1] and [2] for a survey. One reason for this could be that most locational models have an accent of minimizing costs of some type. One of the most celebrated models in location science is certainly the Weber model introduced by A. Weber in 1909 [5]. This model can be brie y described as follows. There is a set of customers to be served. Each customer is known by its location and the quantity of the demand. The problem is to nd a location for the distribution warehouse to serve these customers, where the objective is to minimize the total transportation costs. This model has been extensively studied in the literature, because it captures an essence common in many locational problems. At the same time, it receives criticisms as well. One of the criticisms was raised by Losch in [4]: \Weber's solution for the problem of location proves to be incorrect as soon as not only cost but also sales possibilities are considered. His fundamental error consists in seeking the place of lowest cost. This is as absurd as to consider the point of largest sales as the proper location. Every such one-sided orientation is wrong. Only search for the place of greatest pro t is right". In Chapter 15 of [1] Peeters and Thisse discussed several operational models which combine location problems with pro t-maximization objectives. Moreover, it is assumed that there is no competition in the market and that the demand is a decreasing ane linear function in the price, as in many other microeconomic pricing models. We refer the reader to Chapter 4 of [3] for a survey of various pricing models. The emphasis of the model to be discussed in this paper is di erent. We shall take competition into account. Consider the following situation. There is a rm which produces a certain product. This product will be brought to a warehouse for distribution. There are potential cutomers who are interested in this product. The locations of these customers are known to be with = 1 . Moreover, it is known that the demand quantity of customer is for = 1 . For each customer, it is attractive if the warehouse is located in his/her close neighborhood since the customer bares the transportation costs. Due to the existence of competition in the market, the reservation price (the maximum price up to which a customer is willing to pay) of each customer is a strictly decreasing function in the distance to the warehouse. As soon as the actual price goes beyond a customer's reservation price, he/she will turn to some other supplier, and thus is lost completely from the market of the rm. The problem of the rm is to choose the location for its warehouse n n ai i Qi i i ;


;n ;


;n 1 and the price for the product in order to maximize the total pro t. The goal of this note is to show that in fact there is a simple polynomial time solution method for solving the problem. Certainly, the model is not supposed to be immediately operational due to its simplicity in nature. However, many extensions of the model are possible. 2 The model Mathematically, the problem discussed at the beginning of the previous section can be formulated as maximize p Pni Qifpr kx,a k g subject to p 2 < and x 2 < where ri is the reservation price of the customer i given the location of the warehouse at x. In this expression s stands for the characteristic function of a statement s: i( =1 1 + i ) 2 8 < 1; if s is true s = : 0; otherwise. According to our model description, the function ri is a strictly decreasing function. This model looks quite messy in the sense that it is neither convex or concave in its decision variables p and x. Even for xed p, the objective is in general not continuous in x. It is interesting to see, however, that the problem can be solved when p is xed as a parameter. To illustrate this, let us introduce Fp(x) = p n X i=1 Qifpr (kx,a k)g : i i The original model can be rewritten as max max Fp(x): p2<1+ x2<2 The rst question we pose is: Can we eciently evaluate maxx2<2 Fp(x) when p is xed as a parameter? To keep analysis simple, we consider the case where the norm is Euclidean. In that case, the above problem is equivalent to the following combinatorial circle covering problem: 2 We are given circles i 1 on the plane. Circle i is centered at i with radius i, = 1 . It is assumed that if a point is in i then this point receives a weight i. The total weight of a point is denoted by ( ) and is simply the summation of all weights from the circles that the point is contained in, i.e. (Circle Covering) n a R i fC j  n  ng C ;


;n C w x ( )= w x X w x i w : fijx2Cig The problem is to nd a point on the plane with the maximum total weight. Namely, we 2 that maximizes ( ). wish to nd In the picture below we consider an example of the circle covering problem. x 2 < w x C2 C3 C4 C1  x )6 C5 Figure 1. Circle covering. In this instance, we let 1 = 7, 2 = 6, 3 = 10, 4 = 8 and 5 = 9. Therefore, an optimal solution  lies in the intersection of 3, 4 and 5. This problem is combinatorial in nature. Although the circles may form a partition of the plane in a very complex way, we will see in the rest of this paper that the problem in general allows an easy solution method. Observe that for each pair of circles i and j there can be maximally two intersection w x w w C w C C C C 3 w points. Let them form a set f g. To be more precise, we de ne 8 >< The two intersection points; if and intersect, if and tangentially touch at a point, f g = > The touching point; : The center of ; if contains . I I i;j i;j Cj Ci Cj Ci Cj Ci Let the set of all intersection points be = [1 I i<j Cj  If g : n i;j We observe that the set must contain at least one optimal solution, as the following lemma shows. I Lemma 2.1 For the circle covering problem, it holds that max ( ) = max () 2 2<2 w x w x : x x I Proof. No matter how complex the partition of the plane formed by the circles may look like, the function ( ) remains constant within each region given by this partition. Hence it is sucient to check representatives from each region to nd a maximum point for ( ). Clearly, each region of the partition will have a representative in the set . Therefore, checking all the points in yields an optimal solution. w x w x I I Q.E.D. Note that j j ( , 1) and that checking the weight of a given point amounts to ( ) operations. Hence, checking all the intersection points in yields an optimal solution in ( 3) operations. This computational complexity can be further improved by a more careful sorting. Let the circles be ordered by their natural indices: I n n O n I O n C1 ; C2 ;


; Cn : First, take the circle 1. Sort all of its intersection points with other circles, i.e. sort the points in the set 1 with 1 := [ =2 f1 g C I I ;:::;n I j 4 ;j in the clockwise direction with respect to 1. For simplicity, one may start sorting with an intersection point between 1 and 2. Take one intersection point in 1 in that order. Let be a -dimensional vector de ned as C C C I d n di 8 < = : 1 if the current point is contained in , 0 otherwise. ; Ci ; Adding 's with = 1 gives the total weight for the point under consideration. Now, take the next point in 1 according to the ordering described previously. Note that the vector of the next point di ers only in one position from the previous one. Hence, it takes only constant time to update the total weight. Remark that it involves a sorting for the points with respect to their positions in 1, and so this gives an computational complexity of order ( log ). After we are done with the points in 1, delete 1 from and continue the procedure with 2 log ) basic 2 . This will be repeated for all circles. In total the procedure will require ( operations to nd a point with maximum circle covering weight. wi di I d C O n n I I I C O n n 3 A procedure for solving the pricing/location model In this section we discuss the original problem in which the price plays as a part of the decision variable. For ease of exposition, assume that the reservation price of a customer is ane, i.e. (k , k) = maxf , k , k 0g where 0 and 0. This formula can be explained as follows. Suppose that apart from the current rm, the best price customer gets elsewhere is (including the transportation cost). Moreover, if customer sticks to the current rm, then the transportation cost is assumed to be k , k. Therefore, customer will stay as long as  (k , k). Let the inverse function of be , . In particular, p ri ui > x ai ui vi x ai ; vi > i ui i vi x ai i ri ri , (t) = u =v ri for 0   t ui =vi p i . 5 i , t=vi ri x ai Now we reformulate the pricing/location model slightly: max1 max F (x) 2<2 2<+ p with F (x) = p p x X n Q fk , i p i=1 x ai k i, ( )g: r p For a given p value, construct an n  n matrix M = (m )  as follows. Let m = 0 for all i. On its (i; j ) (i 6= j ) position we make a link to two (n + 1)-dimensional vectors w w representing two possible intersection points between C and C . For each vector w (l = 1; 2) and k = 1; 2; :::; n, let ij n n ii (i;j ) 2 i (i;j ) 1 and j (i;j ) l 8< (w ) = : 1; if this intersection point is contained in C , 0; otherwise. (i;j ) k k l Finally, let (w ) store the current weight of the intersection point. Remark that in the case C and C do not intersect, then we let these two vectors be empty, and when C and C tangentially touch at a point, or one circle contains the other, then these two vectors reduces to only one vector. For initialization, we let p be a suciently small number. It takes O(n ) operations to install all the entries of M and the associated w vectors. As p increase, all the circles will shrink simultaneously at di erent rates. In particular, the intersection points between C and C can be computed analytically as (i;j ) l (n+1) i i j j 3 i j If g = fa + r, (p)eg j i;j j with kek = 1 and 6 (e; a , a ) = arccosf[(r,(p)) , (r, (p)) + ka , a k ]=[2r, (p) ka , a k]g: i j 2 j 2 i 2 i j j i j At a certain stage, the state of the positions will change, i.e. one of the following situations occurs:  For certain i with 1  i  n, r, (p) becomes zero; i 6  The intersection points of two circles converge to only one point;  One intersection point of two circles lies on a third or more circles;  One intersection point of two circles gets o from a third or more circles. Knowing the analytical formula of all intersection points and the formula for , , it is easy to detect when (the value ) any of the above situations will occur. If the rst situation occurs, i.e. one of , ( ) becomes zero for = 1 2 , then we simply need to delete from further consideration and continue the procedure for the rest of the circles. For all the other cases, since , remains ane, it is easy to detect which circles will rst change their relative positions. Should this change takes place, we need to update the matrix and its associated records. However, this update requires only constant time as we know precisely what has changed, i.e. which circles have changed their positions. Notice that as increase, the relative position of any two circles can only possibly evolve in the direction: ri p ri p i ; ; :::; n Ci ri M p one contains the other ! tangentially touch from inside ! intersect ! tangentially touch from outside ! non-intersect : This ensures that one needs in total ( 3) basic operations to keep track with the whole procedure. By simply keeping the highest weight among the intersection points at each update, we select the best one and hence the problem is solved in ( 3 ) time. To summarize we have shown the following result: O n O n Theorem 3.1 The pricing/location model can be solved in at most ( 3) basic operations. O n 4 Discussions The algorithm is based on a real-number computational model. In fact, one may apply the same arguments if any other explicit form of norms are used, including the -norms or polyhedral gauges. The form of can be quite general as well. It is important, however, that one is able to detect the timing when two circles change their relative positions. In a more general form we may consider a model in which several rms start competing in a given region for the same set of potential customers. It can be interesting to investigate, for instance, whether or not an equilibrium exists. Lq ri 7 Remark that the problem we discuss is a planar one. It remains a topic for further research how to solve the problem in polynomial time as the dimension becomes more than two. References [1] Drezner, Z., editor, Facility Location: A survey of applications and methods, Springer Series in Operations Research, New York, 1995. [2] Hansen, P., Labbe, M., Peeters, D. and Thisse, J.-F., Facility Location Analysis, Fundamentals of Pure and Applied Economics 22 (1987) 1-70. [3] Lilieu, G.L., Kotler, Ph., and Sridhar Moorthy, K., Marketing Models, Prentice-Hall, New Jersey, 1992. [4] Losch, A., The Economics of Location, Yale University Press, New Haven, 1954. [5] Weber, A., Uber Den Standort Der Industrien, 1. Teil: Reine Theorie Des Standortes, Tubingen, Germany, 1909. 8