The Stackelberg Minimum Spanning Tree Game

Transcription

1 The Stackelberg Minimum Spanning Tree Game J. Cardinal, E. Demaine, S. Fiorini, G. Joret, S. Langerman, I. Newman, O. Weimann, The Stackelberg Minimum Spanning Tree Game, WADS 07

2 Stackelberg Game 2 players: leader and follower The leader moves first, then the follower moves The follower optimizes his objective function knowing the leader s move The leader optimizes his objective function by anticipating the optimal response of the follower Our goal: to find a good strategy for the leader

3 Setting We have a graph G=(V,E), with E=R B each e R, has a fixed positive cost c(e) Leader owns B, and has to set a price p(e) for each e B function c and function p define a weight function w:e R + the follower buys an MST T of G (w.r.t. to w) Leader s revenue of T is: p(e) e E(T) B goal: find prices in order to maximize the revenue

4 There is a trade-off: Leader should not put too a high price on the edges otherwise the follower will not buy them But the leader needs to put sufficiently high prices to optimize revenue

5 Minimum Spanning Tree problem

6 Minimum Spanning Tree (MST) problem Input: undirected weighted graph G=(V,E,w) Solution: a spanning tree of G, i.e. a tree T=(V,F) with F E Measure (to minimize): Total weight of T: e F w(e)

7 A famous algorithm: Kruskal s algorithm (95) Start with an empty tree T consider the edges of G in non-decreasing order: add the current edge e to T iff e does not form a cycle with the previous selected edges

8 Example B F 7 2 A 4 C E 4 30 D 0 9 G

9 Example B F 7 2 A 4 C E 4 30 D 0 9 G

10 Example B F 7 2 A 4 C E 4 30 D 0 9 G

11 Example B F 7 2 A 4 C E 4 30 D 0 9 G

12 Example B F 7 2 A 4 C E 4 30 D 0 9 G

13 Example B F 7 2 A 4 C E 4 30 D 0 9 G

14 Example B F 7 2 A 4 C E 4 30 D 0 9 G

15 Example B F 7 2 A 4 C E 4 30 D 0 9 G

16 Example B F 7 2 A 4 C E 4 30 D 0 9 G

17 Example B F 7 2 A 4 C E 4 30 D 0 9 G

18 Example B F 7 2 A 4 C E 4 30 D 0 9 G

19 turning to the Stackelberg MST Game

20 Example 0 4 0

21 Example The revenue is

22 Example 0 4 A better pricing

23 Example 0 4 with revenue 2

24 One more example

25 One more example The revenue is 3

26 One more example

27 One more example The revenue is

28 Assumptions G contains a spanning tree whose edges are all red Otherwise the optimal revenue is unbounded Among all edges of the same weight, blue edges are always preferred to red edges If we can get revenue r with this assumption, then we can get revenue r-, for any >0 by decreasing prices suitably

29 The revenue of the leader depends on the price function p and not on the particular MST picked by the follower Let w <w 2 < <w h be the different edge weights The greedy(kruskal s) algorithm works in h phases In its phase i, it considers: 2 all blue edges of weight w i (if any) Then, all red edges of weight w i (if any) Number of selected blue edges of weight w i does not depend on the order on which red and blue edges are considered! This implies

30 Lemma In every optimal price function, the prices assigned to blue edges appearing in some MST belong to the set {c(e): e R}

31 Lemma 2 Let p be an optimal price function and T be the corresponding MST. Suppose that there exists a red edge e in T and a blue edge f not in T such that e belongs to the unique cycle C in T+f. Then there exists a blue edge f distinct to f in C such that c(e) < p(f ) p(f) proof X e f T c(e) < p(f) f : the heaviest blue edge in C (different to f) p(f ) p(f) if p(f ) c(e) V\X f p(f)=c(e) will imply a greater revenue

32 Theorem The Stackelberg MST game is NP-hard, even when c(e) {,2} for all e R reduction from Set cover problem

33 minimum Set Cover Problem INPUT: Set of objects U={u,,u n } S ={S,,S m }, S j U OUTPUT: A cover C S whose union is U and C is minimized

34 U={u,,u n } S ={S,,S m } w.l.o.g. we assume: u n S j, for every j We define the following graph: S m S m- S j S a blue edge (u i,s j ) iff u i S j 2 u u 2 u 3 u i u n- u n Claim: (U,S) has a cover of size at most t maximum revenue r* n+t-+2(m-t)= n+2m-t-

35 ( ) S m S m- S j S a blue edge (u i,s j ) iff u i S j u u 2 u 3 u i u n- u n We define the price function as follows: For every blue edge e=(u i,s j ), p(e)= if S j is in the cover, 2 otherwise revenue r= n+t-+2(m-t)

36 ( ) p: optimal price function p:b {,2, } such that the corresponding MST T minimizes the number of red edges We ll show that:. T has blue edges only 2. There exists a cover of size at most t Remark: If all red edges in T have cost, then for every blue edge e=(u i,s j ) in T with price 2, we have that S j is a leaf in T by contradiction e cannot belong to T blue red or blue? u h S j 2 u i path of red edges of cost

37 ( ), () e: heaviest red edge in T since (V,B) is connected, there exists blue edge f T Lemma 2: f f such that c(e)<p(f ) p(f) X u i c(e)= and p(f )=2 V\X e f T f 2 S j By previous remark all blue edges in C-{f,f } have price p(f)= and p(f )= leads to a new MST with same revenue and less red edges. A contradiction.

38 ( ), (2) S m S m- S j S Assume T contains no red edge We define: C ={S j : S j is linked to some blue edge in T with price } every u i must be incident in T to some blue edge of price C is a cover any S j C must be a leaf in T u u 2 u 3 u i u n- u n S j u i u i+ path in T between u i and u i+ revenue = n+ C -+2(m- C )=n+2m- C - C t n+2m -t-

39 The single price algorithm Let c <c 2 < <c k be the different fixed costs For i =,,k set p(e)=c i for every e B Look at the revenue obtained return the solution which gives the best revenue

40 Theorem Let r be the revenue of the single price algorithm; and let r* be the optimal revenue. Then, r*/r, where =+min{log B, log (n-), log(c k /c )}

41 T: MST corresponding to the optimal price function h i : number of blue edges in T with price c i c c k c k- A c f(x)=x A A /x c c c k x B = j h j min{n-, B } Notice: The revenue r of the single price algorithm is at least c h k h k- h k-2 h x A x B hence: r*/r +log x B x B r* c + c /x dx = c(+ log x B log )= c(+log x B )

42 T: MST corresponding to the optimal price function k i : number of blue edges in T with price c i y c k c k- A c f(y)=x A A /y c c c k x B = j h j min{n-, B } Notice: The revenue r of the single price algorithm is at least c h k h k- h k-2 h x A x B x hence: r*/r +log (c k /c ) c k r* c + c /y dy = c(+ log c k log c )= c(+log (c k /c )) c

43 An asymptotically tight example /2 /i /n The single price algorithm obtains revenue r= The optimal solution obtains revenue n r*= /j = H n = (log n) j=

44 Exercise: prove the following Let r be the revenue of the single price algorithm; and let r* be the optimal revenue. Then, r*/r k, where k is the number of distinct red costs

45 Exercise: Give a polynomial time algorithm that, given an acyclic subset F B, find a pricing p such that: (i) The corresponding MST T of p contains exactly F as set o blue edges, i.e. E(T) B=F (ii) The revenue is maximized