Faster Certified Symmetry Breaking Using Orders With Auxiliary Variables¶
Conference: AAAI 2026 arXiv: 2511.16637 Code: satsuma, VeriPB, CakePB Area: Other Keywords: symmetry breaking, proof logging, SAT solving, pseudo-Boolean, VeriPB, certified solving
TL;DR¶
By introducing auxiliary variables to encode lexicographic order in place of large-integer encodings, this work fundamentally redesigns the VeriPB proof system, achieving order-of-magnitude speedups in both proof generation and verification for certified SAT symmetry breaking, both theoretically and empirically.
Background & Motivation¶
- Background: Symmetry breaking is a key technique in combinatorial solving to avoid redundant exploration of equivalent search spaces.
- Limitations of Prior Work: The SAT community emphasizes provable correctness—solvers in competitions must produce machine-verifiable proofs. Certified symmetry breaking has long been an open challenge: the DRAT format cannot handle multiple symmetries. Bogaerts et al. (2023) proposed a fully general method based on pseudo-Boolean inequalities, but it encodes lexicographic order using large integers \(\sum_{i=1}^n 2^{n-i}x_i \leq \sum_{i=1}^n 2^{n-i}y_i\), where coefficients require \(n^2\) bits, resulting in linear overhead in proof generation and costly arbitrary-precision arithmetic during verification.
Core Problem¶
How can the performance bottleneck of large-integer encodings in symmetry breaking proofs be eliminated while retaining full generality?
Method¶
Mechanism¶
Auxiliary variables \(s_1, \ldots, s_n\) are introduced to encode the lexicographic order (analogous to a clause-based encoding where \(s_i\) indicates equality of the first \(i\) variables), replacing a single large-coefficient pseudo-Boolean inequality.
Technical Challenge¶
Auxiliary variables do not appear in the premises, which breaks the "implication" invariant of the original proof system. This necessitates a fundamental redesign of the proof system.
Key Design: Separation of Specification and Order¶
The order encoding is split into two components:
- Specification \(\mathcal{S}_\preceq(\vec{x}, \vec{y}, \vec{a})\): Defines how auxiliary variables \(\vec{a}\) are computed from the primary variables (analogous to a circuit definition); can be placed in the premises.
- Order \(\mathcal{O}_\preceq(\vec{x}, \vec{y}, \vec{a})\): Order constraints expressed in terms of auxiliary variables.
Definition: \(\alpha \preceq \beta\) if and only if there exists an assignment \(\varrho\) satisfying both \(\mathcal{S}_\preceq\) and \(\mathcal{O}_\preceq\) simultaneously.
Modified Proof Rules¶
- Dominance rule: The proof obligation becomes \(\mathcal{C} \cup \mathcal{D} \cup \{\neg C\} \cup \mathcal{S}_\preceq(\vec{z}|_\omega, \vec{z}, \vec{a}) \vdash \mathcal{C}|_\omega \cup \mathcal{O}_\preceq(\vec{z}|_\omega, \vec{z}, \vec{a})\)
- Redundance rule: \(\mathcal{S}_\preceq\) is similarly incorporated as an additional premise.
- Reflexivity and transitivity proof obligations are adjusted accordingly.
Asymptotic Complexity Improvements¶
| Operation | Prior Method | Ours |
|---|---|---|
| Defining lexicographic order (\(n\) variables) | \(O(n^2)\) | \(O(n)\) |
| Symmetry breaking (support \(k\)) logging | \(O(nk)\) | \(O(k)\) |
| Symmetry breaking checking | \(O(n^2 + nk^2)\) | \(O(n)\) |
Key Experimental Results¶
Constructed Benchmarks (5 families)¶
- The ratio of proof generation time to solving time (without proof) approaches a constant for the proposed method.
- The prior method exhibits clearly degrading asymptotic behavior on PHP/RPHP families.
- For proof verification, the proposed method demonstrates better scaling across all 5 families.
SAT Competition Instances (2020–2024, 982 instances with symmetry)¶
| Metric | Ours | Prior Method |
|---|---|---|
| Successful proof generation | 982/982 | 930/982 |
| VeriPB verification passed | 893 | 806 |
| CakePB verification passed | 799 | 732 |
The proposed method is never worse than the prior method by more than a constant factor and is frequently faster by orders of magnitude.
Highlights & Insights¶
- Resolves a long-standing performance bottleneck in certified symmetry breaking.
- Theoretical improvements (at least a linear-factor speedup) are thoroughly validated experimentally.
- A complete end-to-end formally verified chain is established: satsuma → VeriPB → CakePB, providing formal correctness guarantees.
- Implemented in state-of-the-art tools, offering direct practical value.
Limitations & Future Work¶
- Proof verification may still be asymptotically slower than symmetry breaking itself (each symmetry requires reasoning over all order variables).
- Enabling proof generation still incurs a constant-factor overhead, primarily due to disk I/O.
- Only static symmetry breaking is addressed; dynamic symmetry exploitation during search is not considered.
- The efficiency of CakePB's formally verified arithmetic library still has room for improvement.
Related Work & Insights¶
| Dimension | Bogaerts et al. (2023) | Ours |
|---|---|---|
| Order encoding | Large-integer PB inequalities | Auxiliary variable clauses |
| Coefficient bit-width | \(O(n^2)\) | \(O(n)\) |
| Generality | Fully general | Fully general |
| Practical feasibility | Infeasible for large symmetries | Feasible for all 982 instances |
Insights: - "Replacing large coefficients with auxiliary variables" is a general optimization strategy in PB reasoning. - The Specification/Order separation pattern for proof system design may prove useful in other domains of certified solving.
Rating¶
⭐⭐⭐⭐ — Strong integration of theory and practice, resolves a long-standing open problem, complete and formally verified toolchain.