[ phenix 1.6.4-486 ] [ ALSO : tlsanl, refmac, etc. from ccp4 6.1.3 ] [ molscript/povscript ] the one-liner : i have some TLS tensors that sometimes do not compare at all well with refmac/anisoanl tensors. sometimes, the principal axes of a TLS tensor refines *very* far away from the center of mass(=center of reaction), or the axes from tlsanl are exceptionally small as if not scaled properly. the eigenvalues are really small too. see [*] for troubleshooting and strategy used. i suspect i am missing something. i also understand this might not be a phenix question. -bryan [*] i compared another .def that gave sensible TLS results with a different dataset/model, and my {} are all good AFAIK. anisoanl/refmac/parvati results are reasonable. the phenix refinement appears well behaved. these are all using tlsextract to get the eigenvalues/origins, and molscript to check the axes/centers of mass. input B factors are Beq unless i made an error. the tls group selections can vary - however, simple groups also can give these results. in general, using strategy (excerpt) : refine { strategy = *individual_sites individual_sites_real_space rigid_body \ *individual_adp group_adp *tls *occupancies *group_anomalous sites { individual = None torsion_angles = None rigid_body = None } adp { individual { isotropic = None anisotropic = None } group_adp_refinement_mode = *one_adp_group_per_residue \ two_adp_groups_per_residue group_selection group = None tls = "chain A and (resid 1:41 or resid 42:159 or resid 195:237 or resid 263:420)" tls = "chain A and resid 160:194" tls = "chain A and resid 245:252" } [...cut off ... ] example output of an ill-behaved S tensor from tlsanl : SCREW ROTATION AXIS ABSOLUTE POSITION (A) PITCH (A) X Y Z 1 70.667 154.197 308.627 -144.435 2 977.334 819.588******** 6222.947 3 640.7191431.084 471.341 269.382