Existing literature on cost sharing games predominantly focuses on the design and analysis of specific distri- bution rules. As such, there are a wide variety of distribution rules that are known to guarantee the existence of an equilibrium. Table3.1summarizes several well-known distribution rules (both budget-balanced and non-budget-balanced) from existing literature on cost sharing, and we discuss their salient features in the following.
Table 3.1: Example distribution rules
NAME PARAMETER FORMULA
Equal share None fW
EQ(i, S) = W(S) |S| Proportional share ω=(ω1,...,ωn) whereωi>0 for all1≤i≤n fP RW [ω](i, S) = ωi P j∈SωjW(S) Shapley value None fW SV(i, S) = P T⊆S\{i}(|T|)!(|S|S|−||!T|−1)!(W(T∪ {i})−W(T)) Marginal contribution fW M C(i, S) =W(S)−W(S− {i})
Weighted Shapley value
ω=(ω1,...,ωn) whereωi>0 for all1≤i≤n fW W SV[ω](i, S) = X T⊆S:i∈T ωi P j∈Tωj X R⊆T (−1)|T|−|R|W(R) Weighted marginal contribution fW M CW [ω](i, S) =ωi(W(S)−W(S− {i}))
Generalized weighted Shapley value ω=(λ,Σ) λ=(λ1,...,λn) Σ=(S1,...,Sm) whereλi>0 for all1≤i≤n andSi∩Sj=∅ fori6=j and∪Σ=N fW GW SV[ω](i,S)= X T⊆S:i∈T λi P j∈Tλj X R⊆T (−1)|T|−|R|W(R)
whereT=T∩Skandk=min{j|Sj∩T6=∅}
Generalized weighted marginal contribution fGW M CW [ω](i,S)=λi(W(Sk)−W(Sk−{i})) whereSk=S− k−1 [ ℓ=1 Sℓandi∈Sk
3.2.1
Equal/Proportional Share Distribution Rules
Most prior work in network cost sharing (Anshelevich et al. [7], Corbo and Parkes [18], Fiat et al. [29], Chekuri et al. [14], Christodoulou et al. [16]) deals with the equal share distribution rule,fW
EQ, defined in Table3.1. Here, the welfare is divided equally among the players. The proportional share distribution rule,
fW
P R[ω], is a generalization, parameterized (exogenously) byω ∈R |N|
++, a vector of strictly positive player-
specific weights, and the welfare is divided among the players in proportion to their weights. BothfW
EQandf W
P Rare budget-balanced distribution rules. However, for general welfare functions, they do not guarantee an equilibrium for all games.1
3.2.2
The Shapley Value Family of Distribution Rules
One of the oldest and most commonly studied distribution rules in the cost sharing literature is the Shapley value (Shapley [78]). Its extensions include theweightedShapley value and thegeneralized weightedShapley value, as defined in Table3.1.
1When the local welfare functions{W
r}are ‘anonymous’, i.e., whenWr(S)is purely a function of|S|for allS⊆Nandr∈R, n
fEQWroguarantees an equilibrium for all games. This is a consequence of it being identical to the Shapley value distribution rule (Section3.2.2) in this case. However, the analogous property forfW
The Shapley value family of distribution rules can be interpreted as follows. For any given subset of playersS, imagine the players ofS arriving one at a time to the resource, according to some orderπ. Each playerican be thought of as contributingW(Pπ
i ∪ {i})−W(Piπ)to the welfareW(S), wherePiπdenotes the set of players inSthat arrived beforeiinπ. This is the ‘marginal contribution’ of playerito the welfare, according to the orderπ. The Shapley value, fW
SV(i, S), is simply theaverage marginal contribution of playeritoW(S), under the assumption that all|S|!orders are equally likely. The weighted Shapley value,
fW
W SV[ω](i, S), is then a weighted average of the marginal contributions, according to a distribution with full support on all the|S|!orders, determined by the parameterω∈R|N|
++, a strictly positive vector of player
weights. The (symmetric) Shapley value is recovered when all weights are equal. The generalized weighted Shapley value,fW
GW SV[ω], generalizes the weighted Shapley value to allow for the possibility of player weights being zero. It is parameterized by aweight systemgiven byω = (λ,Σ), whereλ ∈ R|++N| is a vector of strictly positive player weights, andΣ = (S1, S2, . . . , Sm)is an ordered partition of the set of playersN. Once again, players get a weighted average of their marginal contributions, but according to a distribution determined by λ, with support only on orders that conform toΣ, i.e., for
1≤k < ℓ≤m, players inSℓarrive before players inSk. Note that the weighted Shapley value is recovered
when|Σ|= 1, i.e., whenΣis the trivial partition,(N).
The importance of the Shapley value family of distribution rules is that all distribution rules are budget- balanced, guarantee equilibrium existence in any game, and also guarantee that the resulting games are so- called ‘potential games’ (Hart and Mas-Colell [37], Ui [83]).2 However, they have one key drawback—
computing them is often3intractable (Matsui and Matsui [55], Conitzer and Sandholm [17]), since it requires
computing the sum of exponentially many marginal contributions.4
3.2.3
The Marginal Contribution Family of Distribution Rules
Another classic and commonly studied distribution rule isfW
M C, the marginal contribution distribution rule (Wolpert and Tumer [89]), where each player’s share is simply his marginal contribution to the welfare, see Table3.1. Clearly,fW
M C is not always budget-balanced. However, an equilibrium is always guaranteed to exist, and the resulting game is an exact potential game, where the potential function is the same as the welfare function. Accordingly, the marginal contribution distribution rule always guarantees that the welfare maximizing allocation is an equilibrium, i.e., the ‘price of stability’ is one. Finally, unlike the Shapley value family of distribution rules, note that it is easy to compute, as only two calls to the welfare function are required.
2Shapley value distribution rules result in exact potential games, weighted Shapley value distribution rules result in weighted po-
tential games, and generalized weighted Shapley value distribution rules result in a slight variation of weighted potential games (see Appendix3.6for details).
3The Shapley value has been shown to be efficiently computable in several applications (Deng and Papadimitriou [20], Mishra and
Rangarajan [56], Aadithya et al. [1]), where specific welfare functions and special structures on the action sets enable simplifications of the general Shapley value formula.
4Technically, if the entire welfare function is taken as an input, then the input size is alreadyO(2n), and Shapley values can be
computed ‘efficiently’. However, if access to the welfare function is by means of an oracle (Liben-Nowell et al. [47]), than the actual input size is stillO(n), and the hardness is exposed.
Note that, it is natural to consider weighted and generalized weighted marginal contribution distribution rules which parallel those for the Shapley value described above. These are defined formally in Table3.1, and they inherit the equilibrium existence and potential game properties offW
M C, in an analogous manner to their Shapley value counterparts. These rules have, to the best of our knowledge, not been considered previously in the literature; however, they are crucial to the characterizations provided in this article.