Proofs automatically generated by interactive or automated theorem provers are often several orders of magnitude larger than proofs written by hand. This implies significant challenges for processing such proofs efficiently. It turns out that the data structures used to encode terms have a high impact on performance. This article proposes several term data structures; in particular, heterogeneous terms for proof checking that distinguish long- and short-lived terms, and abstract terms for proof parsing. Both term data structures are implemented in the proof checker Kontroli, enabling it to parse and check proofs both sequentially and concurrently without overhead. The evaluation on three datasets exported from interactive theorem provers shows that the new term data structures significantly improve proof checking performance.

Several proof assistants, such as Isabelle or Coq, can concurrently check multiple proofs. In contrast, the vast majority of today's small proof checkers either does not support concurrency at all or only limited forms thereof, restricting the efficiency of proof checking on multi-core processors. This work shows the design of a small, memory- and thread-safe kernel that efficiently checks proofs both concurrently and non-concurrently. This design is implemented in a new proof checker called Kontroli for the lambda-Pi calculus modulo rewriting, which is an established framework to uniformly express a multitude of logical systems. Kontroli is faster than the reference proof checker for this calculus, Dedukti, on all of five evaluated datasets obtained from proof assistants and interactive theorem provers. Furthermore, Kontroli reduces the time of the most time-consuming part of proof checking using eight threads by up to 6.6x.

Connection calculi allow for very compact implementations of goal-directed proof search. We give an overview of our work related to connection tableaux calculi: first, we show optimised functional implementations of connection tableaux proof search, including a consistent Skolemisation procedure for machine learning. Then, we show two guidance methods based on machine learning, namely reordering of proof steps with Naive Bayesian probabilities, and expansion of a proof search tree with Monte Carlo Tree Search.