I want to compare between DFT and method of semiempirical but I do not know which calculation is better to use (single point energy, or calculation of geometry optimization)
Molecule to be optimized after doing conformational analysis using DFT methods such B3LYP using some basis sets. Conformational analysis can be done using potential energy surface scan by rotating any dihederal angle in that molecule
Single point calculation (Sp) does NOT do any good in the optimization of a structure, which means it gives you the same structure as before running the single point calculation. Generally single point calculations are done to get an output file which show some parameters of concern.
Full optimization (Opt) is the method of choice when are looking to find the global minimum structure (GM). GM is the most stable structure with the less energy among all other conformations. Generally, you have to search for the GM, especially if the structure has many flexible bonds. This is done by running reaction coordinate for a torsion angle or around any concerned bond. It is very important to find the GM since there are local minima which are higher in energy than the GM. Therefore, if you do not get the GM your calculations will be incorrect.
Generally, I use the key frequency when I run full optimization.
Regarding the methods, you mentioned DFT and semi-empirical; please note that we have many semi-empirical methods such as AM1. PM3. MINDO/3, MNDO, RM1 and etc and regarding the DFT method, there are many basis sets. Therefore, to compare between the two methods, you should know which method to use and what the basis set to be utilized. Of course, you need the experimental data for comparing the calculated values with the experimental value in order to judge which is the better method to use.
The best way is to consider the calculation for optimization. Because optimization will give the ground state structure of the molecule in its more stable form (less energy).
The short and useful answer to your question is the following:
If you want the quantitative precise results you need the DFT method. But for the qualitative description of your results you can use the semiempirical method.
I think it would depend on the properties that you would like to base the comparison. For example, total energies calculated using DFT methods are meaningless, unless you use it to calculate some properties such as cohesive energy, formation energy, etc. One good example is, for a certain solid, you might want to compare the bulk cohesive energy as estimated via DFT and via some semi-empirical models. From there, you could also estimate the lattice constant, and compare results obtained from DFT and the semi-empirical model.