We wanted to make Spacer more efficient for reasoning about Software programs (with fixed-precision numbers) and complicated hardware designs. Such problems are conveniently expressed in the first order theory of Fixed-Size Bit-Vectors (BV). To instantiate Spacer for any theory, you would need three ingredients: a good SMT solver, an interpolation algorithm (for generalizing lemmas), and a Model Based Projection (MBP) algorithm (for computing predecessors to Bad states). For example, for the theory of Linear Integer Arithmetic, Spacer uses a simplex based SMT solver inside Z3, the Farkas interpolation algorithm, and a Fourier-Motzkin based MBP. In BV, Spacer has a choice of SMT solvers inside Z3 (and Nikolaj is working on a new one!!) but very limited support for interpolation and MBP. In our most recent work, we develop a new MBP algorithm for BV and extend global guidance techniques to BV. Here is the full paper.
The proof of correctness for our rewrite rules can be found here.