🤖 AI Summary
This paper addresses the logical characterization of k-EXPSPACE queries in descriptive complexity. While the classical PSPACE characterization via partial fixed-point logic (PFP) resists generalization to higher exponential space hierarchies, we establish, for the first time, an exact equivalence between k-EXPSPACE and the extension of (k+1)-th-order logic with partial fixed-point operators (HO^{k+1}[PFP]), for all k ≥ 0. Crucially, this characterization requires no order assumption on input structures—overcoming a fundamental limitation of prior approaches reliant on linear orders. Technically, we deploy higher-order model theory and structured induction to rigorously delineate the expressive boundaries of each order of logic for exponential-space computation. Our result unifies and generalizes known characterizations for PSPACE (k = 0) and EXPSPACE (k = 1), fills a long-standing gap in the logical characterization of exponential space hierarchies, and completes the correspondence spectrum between complexity classes and higher-order logics in descriptive complexity.
📝 Abstract
The characterization of PSPACE-queries over ordered structures as exactly those expressible in first-order logic with partial fixpoints (Vardi'82) is one of the classical results in the field of descriptive complexity. In this paper, we extend this result to characterizations of k-EXPSPACE-queries for arbitrary k, characterizing them as exactly those expressible in order-k+1-higher-order logic with partial fixpoints. For k>1, the restriction to ordered structures is no longer necessary due to the high expressive power of higher-order logic.