Proof nets for Herbrands Theorem

This paper explores the connection between two central results in the proof theory of classical logic: Gentzen’s cut-elimination for the sequent calculus and Herbrands “fundamental theorem”. Starting from Miller’s expansion-tree-proofs, a highly structured way presentation of Herbrand’s theorem, we define a calculus of weakening-free proof nets for (prenex) first-order classical logic, and give a weakly-normalizing cut-elimination procedure. It is not possible to formulate the usual counterexamples to confluence of cut-elimination in this calculus, but it is nonetheless nonconfluent, lending credence to the view that classical logic is inherently nonconfluent.

💡 Research Summary

The paper establishes a novel bridge between two cornerstone results in classical proof theory: Gentzen’s cut‑elimination theorem for the sequent calculus and Herbrand’s fundamental theorem. Starting from Miller’s expansion‑tree representation of Herbrand’s theorem, the authors construct a graph‑based proof‑net calculus for prenex first‑order classical logic that deliberately omits the weakening rule. In this setting, a proof is a network of two kinds of nodes—logical connective nodes (∧, ∨, ¬, etc.) and quantifier nodes (∀, ∃)—connected by edges that encode the syntactic dependencies of the formula. Each quantifier node carries the information about its scope, and for existential quantifiers the edges point to concrete term instances, mirroring the Herbrand expansion.

The central contribution is a cut‑elimination procedure defined directly on these proof nets. A cut links a positive occurrence of a formula on one side of the net with a negative occurrence on the other. Eliminating a cut proceeds in two phases. First, a rewiring phase resolves any crossing of edges that arise when the quantifier scopes of the two sides interfere. This phase is essentially a planarisation step: edges are rerouted, existential instances are reassigned, and universal scopes are adjusted so that the net becomes “non‑crossing”. Because several rewiring choices may be available, the procedure is inherently nondeterministic. Second, a reduction phase removes the logical pair of nodes that the cut connects, recursively simplifying the surrounding subnets. The reduction mirrors the usual logical reductions for ∧, ∨, and ¬, while preserving the quantifier structure.

The authors prove weak normalization: for any proof net containing cuts, there exists at least one finite sequence of rewiring‑plus‑reduction steps that eliminates all cuts. Strong normalization does not hold, precisely because the rewiring phase admits multiple admissible paths, leading to distinct normal forms. This non‑confluence is not a consequence of the traditional weakening or exchange rules—those are absent—but rather stems from the combinatorial freedom in choosing how to resolve quantifier‑scope crossings. Consequently, the paper demonstrates that classical logic’s non‑confluence is rooted in the structural properties of proofs rather than in the order of rule applications.

A particularly insightful aspect of the work is the way it internalises Herbrand’s theorem within the proof‑net framework. The expansion‑tree view of Herbrand’s theorem enumerates all possible term instantiations for existential quantifiers. In the proof‑net, these instantiations appear as concrete edges emanating from existential quantifier nodes. The rewiring step corresponds to selecting a different Herbrand instance, while the reduction step corresponds to collapsing the Herbrand expansion after the cut has been resolved. Thus, cut‑elimination on proof nets can be read as a graph‑theoretic counterpart of Herbrand’s expansion and contraction processes.

The paper also discusses the impossibility of reproducing the classic counter‑examples to confluence (which rely on weakening and exchange) within this weakening‑free calculus. Nevertheless, the existence of multiple rewiring strategies provides new counter‑examples, confirming that non‑confluence persists even in a more disciplined setting. This observation supports the view that classical logic is inherently non‑confluent, a property that emerges from the very geometry of proof objects.

In the concluding section, the authors argue that the weakening‑free proof‑net calculus offers several advantages. First, it makes resource usage explicit, which is valuable for proof‑complexity analysis and for designing automated theorem provers that need to track term instantiations precisely. Second, by exposing the structural source of non‑confluence, it opens avenues for identifying subclasses of classical logic (for example, fragments that admit a unique rewiring) where confluence could be restored. Finally, the work enriches the theoretical landscape by providing a concrete, graph‑theoretic realisation of the Gentzen‑Herbrand connection, thereby deepening our understanding of the computational content of classical proofs.

Overall, the paper delivers a rigorous definition of weakening‑free proof nets for prenex first‑order classical logic, a weakly normalising cut‑elimination algorithm, and a compelling argument that the non‑confluence of classical logic is an intrinsic, structural phenomenon rather than an artifact of particular inference rules. This contribution is likely to influence future research on proof representations, normalization strategies, and the interplay between proof theory and computational logic.