# Gaussian09 and Path Integral Monte Carlo which one predicts the total internal energy more accurately?

I'm learning PIMC with primitive action and currently just testing the code with single methane molecule, I did run a Gaussian ccsd(t) calculation, I notice that at 298.15K & 1atm, the energy predicted by PIMC(0.0476a.u.) is slightly lower than the one by Gaussian(0.0477a.u.) both errors start from 5th decimal number. I'm just wondering which result could be more reliable? And why one is more reliable than the other?
It might be a dumb question, but I don't see anywhere made such comparison and I'm just an undergraduate student with no experience with either of two computational techniques. Would be greatly appreciated if can use some expertise!

• what energy are you calculating? what are the parameters of your pimc calculation (grid)? what basis set do you use in your cc calculation? convergence criteria?
– user37142
Nov 23 '16 at 13:28
• Total internal energy is the one I'm calculating. Grid approach to PIMC... cc-PVTZ for H aug-cc-PVTZ for C I'm not sure what you mean by convergence criteria, I ran an extremely long simulation then I applied the "Flyvbjerg-Petersen"blocking algorithm to work out the number of samples required to reach convergence Nov 23 '16 at 14:08
• Every time you look for convergence you stop if the change between two (or more) points is less than a certain value. For SCF convergence in Gaussian use the keyword SCF=(Conver=N) which sets the SCF convergence criteria to 10^-N, see gaussian.com/g_tech/g_ur/k_scf.htm. I guess you are doing single point calculations?
– DSVA
Nov 23 '16 at 15:50
• Can you add your G09 input to your post? And yes, both methods do that but the question is what the convergence criteria are! If for MC it's extremly tight and for G09 it's not then it doesn't make sense to compare them! Make the convergence criteria tighter and look if the results change.
– DSVA
Nov 23 '16 at 23:21
• One problem in comparing the two methods is also that MC uses statistics and CCSD(T) is an ab-initio method. CCSD(T) is often used as a reference/benchmark value for other calculations since it is highly accurate.
– user37142
Nov 24 '16 at 9:43