For further details on consult "Thermal vibrations in crystallography", B.T.M. Williis and A.W. Pryor, Cambridge University Press (1975).

You must note that all systematic errors in your diffraction data gets into B values obtained by least squares refinement and makes them sort of garbage bin. I cannot help you unless you tell me what sort of diffraction data you have and up to what sin theta/lambda value do they extend to. Also the strategy should be varied according to the sample studied, whether it contains, disorder, split positions etc. If you are talking about perovskite manganites then you should consult sever papers we have written on LaMnO3, La1-xSrxMnO3, La1-xBaxMnO3 and La1-xCaxMnO3. There you will find B values for all relevant atoms. The O atoms have usually larger B values and often increases anomalously close to a phase transition and this is normal. You can have an idea of what the B values should be from these literature remembering that B values crucially depends on the environment of surrounding atoms and bonding. If you get negative B values then you are sure that your data are not very good or do not contain enough high-Q information.

When you are sure that your structure model is right and the structural parametrs are reasonable then you refine the B_iso of the atoms. At the begining you can give some reasonable positive values for B_iso. Then B_iso should be refined as soon as the refinement appears to be OK always checking whether the refined values stay positive. When the refinement converges and the agreement factors are reasonable, then you can attempt to refine B_anisotropic. For this purpose you must have very good quality data and preferably up to a high Q = 4\pi sin theta/lambda. To get high Q values you must use short wavelength and measure diffraction intensities up to high 2theta. The refinement of B_anisotropic works much better with neutron diffraction than with X-ray diffraction for the reason that the X-ray diffraction intensities decreases with higher Q following the square of the form factor wheras in the neutron diffraction the scattering factor remains constant with increasing Q. You must take care of the symmetry restriction for B_anisotropic, but I guess FullProf takes care of it (I have to check). All good crystallographic programs should take care of the symmetry restriction and only allows permissible refinement of B_anisotropic with contraints whenever necessary. But always check whether the refinement parameters stay resonable.

Dear Sir,

First of thank you for your response. Most of the time the Biso parameter become negative showing red in colour. when I refine it for the solid solution then the parameters become greater than 1 (e.g. 5, 10 like this...). should it be greater than one. most of the time it is very difficult to refine the Biso parameter for Oxygen (O1 and O2 in case of Pbnm/Pnma + R3c space group). what should be the initial values I could not understant. Can you please help me in this aspact?

For further details on consult "Thermal vibrations in crystallography", B.T.M. Williis and A.W. Pryor, Cambridge University Press (1975).

You must note that all systematic errors in your diffraction data gets into B values obtained by least squares refinement and makes them sort of garbage bin. I cannot help you unless you tell me what sort of diffraction data you have and up to what sin theta/lambda value do they extend to. Also the strategy should be varied according to the sample studied, whether it contains, disorder, split positions etc. If you are talking about perovskite manganites then you should consult sever papers we have written on LaMnO3, La1-xSrxMnO3, La1-xBaxMnO3 and La1-xCaxMnO3. There you will find B values for all relevant atoms. The O atoms have usually larger B values and often increases anomalously close to a phase transition and this is normal. You can have an idea of what the B values should be from these literature remembering that B values crucially depends on the environment of surrounding atoms and bonding. If you get negative B values then you are sure that your data are not very good or do not contain enough high-Q information.