Fig. 5
From: A continuous-time MaxSAT solver with high analog performance
Algorithm performance on competition MaxSAT problems. a Using the 454 random benchmark problems from the 2016 SAT competition49, we compared the overall best (lowest) energy found by the competition solvers (black + symbols) and the minimum found/predicted by our Max-CTDS. There are four categories of problems, separated by vertical dashed lines: “s2” (Abrame-Habet) are Max 2-SAT with N ∈ [100, 200], M ∈ [1200, 2600]; “s3” (Abrame-Habet) are Max 3-SAT with N ∈ [70, 110], M ∈ [700,1500]; “HG3” (high-girth) are Max 3-SAT problems with N ∈ [250, 300], M ∈ [1000, 1200], and “HG4” are Max 4-SAT problems with N ∈ [100, 150], M ∈ [900, 1350]. b There are some small energy differences especially in the “s2” category, where the ODEs are more stiff and the integration becomes very slow. c The energy differences normalized by the number of clauses, (Emin − Ebest)/M. Max-CTDS solves all problems within 0.8% of the best energy value
