Enhancing automated loop invariant generation for complex programs with large language models

Automated program verification has always been an important component of building trustworthy software. While the analysis of loops remains a theoretical challenge, the automation of loop invariant analysis has effectively resolved the problem. However, existing invariant generation tools are predominantly effective for programs with purely numerical or purely pointer-based structures. Real-world programs often mix complex data structures and control flows. These structures can include arrays, pointers, and recursive definitions, while control flows may involve multiple nested or concurrent loops. Traditional methods generally only generate invariants for simple numerical programs or specific segments, lacking broad applicability. In order to automatically generate loop invariants for real-world programs, we proposed ACInv, an Automated Complex program loop Invariant generation tool, which combines static analysis with prompting with Large Language Models (LLM) to generate the proper loop invariants. We employ static analysis to systematically decompose the program's data structures and loops. This involves layer-by-layer transmission of structural information about variables, numerical data, and the complete loop structure to the LLM, enabling the generation of corresponding invariants. In comparison to prior work on AutoSpec, we delve deeper into the variable information within each loop. We conducted experiments on ACInv, which showed that ACInv outperformed previous tools on data sets with data structures and maintained similar performance to the state-of-the-art tool AutoSpec on numerical programs without data structures. For the total data set, ACInv can solve 21% more examples than AutoSpec, and can generate reference data structure templates.