Download PDFOpen PDF in browser

Prover9 Unleashed: Automated Configuration for Enhanced Proof Discovery

10 pagesPublished: May 26, 2024


While many of the state-of-art Automated Theorem Provers (ATP) like E and Vampire, were subject to extensive tuning of strategy schedules in the last decade, the classical ATP prover Prover9 has never been optimized in this direction. Both E and Vampire provide the user with an automatic mode to select good proof search strategies based on the properties of the input problem, while Prover9 provides by default only a relatively weak auto mode. Interestingly, Prover9 provides more varied means for proof control than its competitors. These means, however, must be manually investigated and that is possible only by experienced Prover9 users with a good understanding of how Prover9 works.
In this paper, we investigate the possibilities of automatic configuration of Prover9 for user-specified benchmark problems. We employ the automated strategy invention system Grackle to generate Prover9 strategies with both basic and advanced proof search options which require sophisticated strategy space features for Grackle. We test the strategy invention on AIM train/test problem collection and we show that Prover9 can outperform both E and Vampire on these problems. To test the generality of our approach we train and evaluate strategies also on TPTP problems, showing that Prover9 can achieve reasonable complementarity with other ATPs.

Keyphrases: automated theorem proving, strategy invention, Strategy Scheduling

In: Nikolaj Bjorner, Marijn Heule and Andrei Voronkov (editors). Proceedings of 25th Conference on Logic for Programming, Artificial Intelligence and Reasoning, vol 100, pages 360--369

BibTeX entry
  author    = {Kristina Aleksandrova and Jan Jakubuv and Cezary Kaliszyk},
  title     = {Prover9 Unleashed: Automated Configuration for Enhanced Proof Discovery},
  booktitle = {Proceedings of 25th Conference on Logic for Programming, Artificial Intelligence and Reasoning},
  editor    = {Nikolaj Bj\{\textbackslash{}o\}rner and Marijn Heule and Andrei Voronkov},
  series    = {EPiC Series in Computing},
  volume    = {100},
  pages     = {360--369},
  year      = {2024},
  publisher = {EasyChair},
  bibsource = {EasyChair,},
  issn      = {2398-7340},
  url       = {},
  doi       = {10.29007/sd6t}}
Download PDFOpen PDF in browser