New 4-step CPS verification guide tames abstraction soundness pitfalls

Formal verification of cyber-physical systems (CPS) often relies on symbolic model checking, which requires discretizing continuous dynamics into a finite-state abstraction. But a sound abstraction must conservatively approximate the concrete state space—a task riddled with pitfalls like under-approximating states or transitions, unsound pruning of degenerate behaviors, and improper specification lifting. In this arXiv tutorial (2608.10254), Jordan Peper, Krish Kapadia, James Gast, Ethan Howes, and Ivan Ruchkin offer a pragmatic workflow that ensures conservativeness by construction, breaking the process into four modular steps with interchangeable subroutines.

The workflow begins with state-space partition and abstraction-function design, then builds conservative transitions using axis-aligned bounding boxes, polytopes, or sampling with PAC (Probably Approximately Correct) coverage certificates. To combat spurious transitions and self-loops, it employs certified erasure and counterexample-guided abstraction refinement (CEGAR). Finally, LTL (Linear Temporal Logic) properties are lifted soundly using may-must semantics. The authors demonstrate the end-to-end pipeline on three case studies, detailing how each design choice impacts abstraction size, runtime, and verification results. This guide gives practitioners a defensible, repeatable method for producing trusted verifiable models of closed-loop dynamical systems.