Hello everyone, I am experiencing differences with phenix.refine versions 1.8.3 or 1.8.4 compared to 1.8.2. Performing comparable runs, Rs are rising several percents with the latest versions, while marginally decreasing using version 1.8.2. I am a very recent convert to Phenix, so maybe I'm missing something obvious, but here are the details: Refinement strategy is xyz+real space+tls+individual Bs at 3.5 A, using NCS and secondary structure restraints, and weight optimization. Starting R/Rfree are around 24.5/28.5 Part of the problem seems to come from secondary structure restraints, with (many) messages in the log file of the refinement going wild looking like WARNING: could not process nonbonded pair Original error: cctbx Internal Error: /net/clover/scratch1/phenix/phenix-1.8.4-1496/cctbx_project/cctbx/crystal/pair_tables.h(663): CCTBX_ASSERT(j_sym >= 0) failure. pdb=" N GLN B 130 " segid="B " pdb=" O VAL B 115 " segid="B " In log files of refinement with version 1.8.2, there are no such messages, and in either case the definition of beta strands were done automatically by Phenix But even secondary structure restraints are switched off, it starts ok but goes wrong at the beginning of cycle 2, during local real space refinement start: r_work=0.2271 r_free=0.2785 final: r_work=0.2418 r_free=0.2937 And then continue to worsen. Any help appreciated! Cheers, Hugues -- Hugues Nury Institut de Biologie Structurale 41 rue Jules Horowitz, F-38027 Grenoble Cedex 1, France +33 6 52 45 17 79