Abstract:We discuss Nash equilibria in combinatorial auctions with item bidding. Specifically, we give a characterization for the existence of a Nash equilibrium in such a combinatorial auction when valuations by two bidders satisfy symmetric and subadditive properties. Based on this characterization, we can obtain an algorithm for deciding whether a Nash equilibrium exists in such a combinatorial auction.Keywords: Nash equilibrium, combinatorial auction, second-price auction, subadditivity, symmetric valuation, price of anarchyIn a combinatorial auction, m items M = {1, 2, . . . , m} are offered for sale to n bidders N = {1, 2, . . . , n}. Each bidder i has a valuation f i that assigns a nonnegative real number to every subset S of M. The objective is to find a partition S 1 , S 2 , . . . , S n of M among the bidders such that the social welfareis maximized. The combinatorial auction problem is sometimes called the social welfare problem when we disregard strategic issues on bidders' selfish concerns. VCG (Vickrey-Clarke-Groves) mechanisms optimize the social welfare in a combinatorial auction with selfish bidders. However, it may take exponential time in m and n. Actually, the social welfare problem is shown to be NP-hard by Lehmann, Lehmann and Nisan, even if every val-Therefore approximation algorithms have also been proposed for the social welfare problem (in a combinatorial auction). Since each valuation f i is defined by 2 m subsets of M, most proposed approximation algorithms are based on oracle models. Two oracle models, the value queries oracle model and the demand queries oracle model, are commonly used. Furthermore, in most proposed approximation algorithms, each valuation f i is restricted to satisfy some conditions. Two restrictions, submodularity and subadditivity, are commonly used. For the submodular social welfare problem (i.e., each valuation is submodular) with the value queries oracle model, the following are known. Lehmann, Lehmann and Nisan proposed a 1 2 -approximation algorithm [12]. Khot et al. showed that this problem cannot be approximated to a factor better than 1 − asano@ise. chuo-u.ac.jp and Schapira proposed an improved (1 − 1 e )-approximation algorithm for the submodular social welfare problem [6].For the more general subadditive social welfare problem (where each valuation is subadditive), Dobzinski, Nisan, and Schapira proposed an Ω(1/ log m)-approximation algorithm using the value queries oracle model [5]. Using the more powerful demand queries oracle model, Feige proposed a 1 2 -approximation algorithm for the subadditive social welfare problem and also showed that it is NP-hard to approximate to a factor better than 1 2 [8]. He also proposed a (1 −