Title:A Fixed Point Framework for the Existence of EFX Allocations

Abstract:We consider the problem of the existence of an envy-free allocation up to any good (EFX) for linear valuations and establish new results by connecting this problem to a fixed point framework. Specifically, we first use randomized rounding to extend the discrete EFX constraints into a continuous space and show that an EFX allocation exists if and only if the optimal value of the continuously extended objective function is nonpositive. In particular, we demonstrate that this optimization problem can be formulated as an unconstrained difference of convex (DC) program, which can be further simplified to the minimization of a piecewise linear concave function over a polytope. Leveraging this connection, we show that the proposed DC program has a nonpositive optimal objective value if and only if a well-defined continuous vector map admits a fixed point. Crucially, we prove that the reformulated fixed point problem satisfies all the conditions of Brouwer's fixed point theorem, except that self-containedness is violated by an arbitrarily small positive constant. To address this, we propose a slightly perturbed continuous map that always admits a fixed point. This fixed point serves as a proxy for the fixed point (if it exists) of the original map, and hence for an EFX allocation through an appropriate transformation. Our results offer a new approach to establishing the existence of EFX allocations through fixed point theorems. Moreover, the equivalence with DC programming enables a more efficient and systematic method for computing such allocations (if one exists) using tools from nonlinear optimization. Our findings bridge the discrete problem of finding an EFX allocation with two continuous frameworks: solving an unconstrained DC program and identifying a fixed point of a continuous vector map.