Software Engineering
Researchers verify correctness of semidefinite programming solver using AI and formal methods
A team has built a Rust-based optimization tool and used AI-assisted formal verification to prove its mathematical soundness.
Researchers have successfully created and formally verified a Rust implementation of a semidefinite programming solver — a mathematically complex optimization tool — using an automated theorem prover. This work demonstrates a pragmatic approach to ensuring correctness in demanding numerical software.
The Challenge#
Semidefinite programming solvers are notoriously difficult to implement correctly. They involve intricate numerical computations and must maintain mathematical guarantees even when working with floating-point approximations. Getting such systems right is traditionally labour-intensive and error-prone; verifying their correctness formally is even harder.
The Approach#
The team hand-wrote the solver in Rust, then enlisted Claude (an AI assistant) to generate a full formal verification using Verus, a theorem prover designed specifically for Rust programs. Rather than trying to prove every line manually, the researchers adopted a pragmatic strategy: identifying the key algorithmic properties that matter most and focusing verification effort there.
Why It Matters#
This work bridges the gap between mathematical rigor and practical software engineering. By coupling a human-written implementation with AI-assisted formal proof generation, the approach becomes more tractable than purely manual verification—yet still delivers the kind of machine-checkable guarantees that numerical code demands. The result offers a template for verifying other nontrivial pieces of computational software where correctness is critical.