30 degrees difference sounds like a lot. Maybe you should inform us what level of theory you are using to refine the structure. Do you properly account for the periodic boundary conditions?

Hi, Thomas! Thanks for your kind answer! I was using CASTEP with norm-conserving pseudopotential to optimize the atomic coordinates and cell parameters. The PBE exchange and correlation functional was employed. I involved PBC in my calculation. What do you mean "properly"?

Is the crystal ionic? If so you can calculate just the single molecule with BP86 (geometry and frequency) including empirical dispersion. If the crystal has periodic covalent bonds then I think you will have anyway a large margin of error on your geometry but the IR and Raman intensity shouldn't be much affected, while the frequency itself would require to apply a scaling factor in any of the above mentioned cases.

It is not ionic crystal. But there are some atomic coordinates (fractional) that are larger than 1. Does this mean that periodic covalent bonds are present in my system?

Where the distance is smaller than the sum of the atomic radius I would say the bond is covalent. For sure Above 3 Angstrom is ionic.

Mr. Li,

You can follow the discussion 'Is geometry optimization needed before running single point energy calculation in gaussian09 for high-resolution crystal structure coordinates?' (https://www.researchgate.net/post/Is_geometry_optimization_needed_before_running_single_point_energy_calculation_in_gaussian09_for_high-resolution_crystal_structure_coordinates)

Thanks, Bojidarka! According to your answers to the previous question, may I conclude that I can fix the cell parameters of my sample, and just optimize the atomic coordinates inside the unit cell? Because the optimization of cell parameters is not the precondition for the calculation of vibrational spectra. I should try my best to make the final optimized structure as close to the experimental result as possible.

Mr. Li,

As you have properly pointed out, you can predict vibrations in solid-state for systems at a fixed state, which, however, is far from equilibrium such as for example those states in glasses. Your scaling factor shall be different, of course, from the known range already reported for systems computed at equilibrium state and/or close to equilibrium. But this range can be evaluated comparing with experimental data sets.

You could also pay attention to a discussion 'Could anybody help me with Gaussian09 about optimising gold layer?' (http://www.researchgate.net/post/Could_anybody_help_me_with_Gaussian09_about_optimising_gold_layer)