Hi,

I do have a data is at very low resolution, so only possible to do rigid body refinement.

it depends what you call "low resolution". In small molecule crystallography ~1A resolution is "low resolution" (Acta Cryst. (2007). D63, 160-170). So, without knowing what the resolution of your data is I can't really tell whether you need to refine individual anisotropic B-factors or the whole model as just one TLS group.

Anyway, if by "low resolution" you mean what comes to my mind this very minute, you can still try:

Coordinates:
- Simulated Annealing refinement in torsion angle space;
- Highly restrained refinement of individual coordinates;
- Optimize weights against Rfree.
- You may need to use secondary-structure restrains (available in recent PHENIX versions).
ADP:
- combined refinement of TLS+individual or group B-factors;
- Highly restrained individual ADP refinement;
- Simple group ADP refinement (where the size of groups depends on resolution or/and data-to-parameters ratio)
- Optimize target weights.
NCS:
- use if available.
Bulk-solvent:
- optimize mask parameters (r_solvent and r_shrink) using optimize_mask=true (this is will done automatically in future versions of phenix.refine).
Multi-Start SA:
- people find it useful (Andrei Korostelev, Martin Laurberg, and Harry F. Noller "Multistart simulated annealing refinement of the crystal structure of the 70S ribosome". PNAS 2009 106:18195-18200).

Good luck!
Pavel.