Dear Phenix users, I have sent this message to phenixbb before I subscribed to it and did not see it show up in the bulletin board. I am sending it again and apologize for the possible duplicated messages. I have been trying to understand the B factors generated by TLS refinements and appreciate any help in this subject. In my particular case, I was trying to analyze the dynamics of different portions of a protein molecule by looking at the B factors. The structure was refined with the isotropic B factor and TLS and the TLS groups were determined automatically. After the refinement, I tried to analyze the TLS matrices with phenix.tls_analysis. The analysis failed for many of the TLS groups, the program complains that some of the matrices are not positive semidefinite. I have several questions related to this: 1, in cases like this, can I tweak the initial values of the matrix elements or add constrains to make the TLS refinement results physically meaningful? i.e. the refined TLS matrices can be interpreted in terms of elemental motions. I noticed that Afonine et al mentioned in a recent paper (doi: 10.1107/S2059798318005764) that a large fractions of TLS matrices in the PDB databank can not be interpreted by elemental motions. I wonder if there is a way to make the refined matrices more meaningful in the current implementation of TLS refinement in phenix? 2, what is the "best" B-factor to look at in an analysis for protein dynamics? I am interested in the dynamics of portions of protein that are smaller than the TLS groups. I am guessing that if the TLS matrices are not physically meaningful, one should look at the overall B-factors? If these matrices are meaningful, is it a good practice to look at the residual B factors (overall B-factor minus the TLS contributions)? 3, how to calculate the residual B factors? I could not find a manual for phenix.tls mentioned in the phenix refinement instruction. I tried with the extract_tls=true option with my PDB and got a PDB file with some negative B factors. Is it another sign that the refined TLS matrices are not physically meaningful or am I using the phenix.tls program wrong? Your help in this subject is greatly appreciated. Song Xiang