Orca is a free software suit (Abinitio) that is very reliable and fast and has a very good manual and a very welcoming community that answers your questions online. But if you are going to do molecular dynamics type of calculations I am not an expert in that area.

Define "best"...

It is not really my area of expertise, but your question seems extremely general. To hopefully make it more explicit:

1. What level of theory do you want to do? Wavefunction-based methods? DFT? Semi-empirical models? QM-MM? Force fields?

2. What systems do you want to study? How many and what atoms, what types of systems?

3. What do you want to calculate? Only the melting point or also other properties?

Thanks Lorenz for your answer. Actually i am new in this field so i have to learn more.

by using the best i  mean which method is more precise, easy to use and frequently used by researcher.

1. I am not familiar with those methods but as i read articles most of authors used the MM+ molecular mechanics force field and AM1 methods.

2. The dataset link is: http://pubs.acs.org/doi/suppl/10.1021/ie051277g/suppl_file/ie051277gsi20060215_020437.pdf.

3. I would like to calculate only the melting point.

Everything depends on the system that you are going to study. How many atoms you have in your sampe ? what types of atoms you have in the system ? how you will describe the interaction between these atoms ? you properly need a good potential energy function for this purpose that describes your system well. Then you can run an MD program for this purpose to calculate the melting point in wich you can control the temperature and the number of atoms by this classical method (i.e thousands of atoms !!) which is difficult by Quantum Methods (QM) where you consider only few atoms !! , in addition you can optimize the structure !! you can also cool your system (i.e quench ) the atoms and see what happens.

In my opinion specify your system first with regards to the above points. Then decide what you want to do. I wish you luck in your research.

There is no difference between free and open source for optimization. These are different when you want make changes in programs, which it is possible only in open sources.

Orca is a free software suit (Abinitio) that is very reliable and fast and has a very good manual and a very welcoming community that answers your questions online. But if you are going to do molecular dynamics type of calculations I am not an expert in that area.

All the packages have their proc and cons due to availability or unavailability of methods and algorithms.    

Try to optimize your geometry using different algorithms (like quadratic approximation, Berny algorithm etc.) at suitable convergence criterion with same level of theory by any package. Compare the SCF energies of all sets and choose the geometry that have minimum SCF energy. 

You should check also whether the optimized geometry at local/global minimum or not by real values of harmonic frequencies. 

Dear Pezhman

I think you have to use MD simulation with large number of atoms with PBC to get a reliable and precise melting point. So you dont need a high level qunatum mechanics optimization on this matter. As you mentioned a light pre-optimization using semi-empirical methods (like PM6, PM3, or AM1) is nice to do. I recommend the gaussian 09 program package of course is not free. Also you can use the Hyperchem which also is not free. If you really need a free software, then Orca is a good choice. After a pre-optimization you can use LAMMPS MD simulation code (free :)) with a large simulation box with PBC and an accurate force field (i suggest AIREBO, Tersoff or Reactive force fields all are implemented in LAMMPS). You can use a thermostat  to increase temperature gradually with time. Finally, monitoring some properties like density, RDF and MSD lead you to the melting point.

Be happy.

try also Avogadro