G-RRM: Guiding Symbolic Solvers with Recurrent Reasoning Models

Paper Overview

Field: Machine Learning Authors: Timo Bertram, Sidhant Bhavnani, Richard Freinschlag Published: 2026-07-04 arXiv: 2507.03231

Summary

This work focuses on SE-RRMs, a symbol-equivariant instantiation of Recurrent Reasoning Models (RRMs) that exhibits improved extrapolation to larger problem sizes. The authors propose a neuro-symbolic method, Guiding with Recurrent Reasoning Models (G-RRM), which integrates SE-RRMs with symbolic solvers for constraint satisfaction problems. SE-RRMs act as neural solvers that generate full solution proposals and guide classical symbolic solvers — such as backtracking or SAT-based methods like Glucose 4.1 and CaDiCaL 3.0.0 — to produce globally correct solutions.

The central research question: when does neural guidance improve the search efficiency of symbolic solvers?

Key Findings

G-RRM's efficacy depends on two conditions:

• The problem instances must have an expansive combinatorial search space to expose potential gains.
• The solver architecture must be capable of dynamically overwriting its branching choices to recover when neural hints are imperfect.

When these conditions are met:

• Guidance can reduce the median number of conflicts to zero and deliver significant wall-clock speedups.
• On 9x9 Sudoku (where SE-RRM correctly solved 91.1% of instances), backtracking achieved a 33.3x speedup and Glucose 4.1 a 1.70x speedup (median, p<0.001).
• Glucose 4.1 retained a 1.17x speedup on perfectly hinted 25x25 grids.

Counter-example:

• CaDiCaL 3.0.0's runtime was dominated by overhead, and it consistently respected injected branching hints rather than overwriting them. It showed no significant speedup (median 1.02x, n.s.) and even a small but significant mean slowdown (0.90x) on 9x9.

These results clarify the scenarios in which neural guidance translates into practical solver acceleration.

Original Abstract

> In this work, we focus on SE-RRMs, a symbol-equivariant instantiation of RRMs that exhibits improved extrapolation to larger problem sizes. We propose a neuro-symbolic approach, "Guiding with Recurrent Reasoning Models" (G-RRM), which integrates SE-RRMs with symbolic solvers for constraint satisfaction problems. SE-RRMs act as neural solvers that generate full solution proposals and guide classical symbolic solvers, such as backtracking or SAT-based methods like Glucose 4.1 and CaDiCaL 3.0.0, that produce globally correct solutions. Centrally, we investigate when neural guidance with G-RRM improves the search efficiency of symbolic solvers. Our experiments show that the efficacy of G-RRM depends on two conditions: first, the problem instances must have an expansive combinatorial search space to expose potential gains, and second, the solver architecture must be capable of dynamically overwriting its branching choices to recover when neural hints are imperfect...

*Auto-collected on 2026-07-05*