How to obtain accurate frequencies from CP2K? (inside the post the frequencies are compared between CP2K and Gaussian)

[J. Ye]

Hi all, 

I calculate frequency for small molecules and try to compare with Gaussian 09 and experimental results. I show the results from CP2K (PBE functional-BASIS_MOPLOT, GTH potental, nonperiodic) and  Gaussian09 (M062X/6-311g(d,p)) . 

I have several questions:

1. You will see the total number of modes obtained from CP2K is 15, however 12 modes from Gaussian. Are the first three modes are translation modes or rotation modes for C2H4?

2. If we only compare the 12 vibration modes, the value for most of  modes from Gaussian is larger than CP2K. What's the reason for getting smaller frequencies from CP2K?  

3. What is the keyword can strongly affect the frequencies?  Affect the accuracy of frequencies? 

Thanks very much! 

CP2K	Gaussian
c2h4
15 modes
-19.0703
-5.64472
19.50384
807.9613	831.8134
942.954	988.1607
995.6397	996.8044
1045.988	1071.741
1206.11	1239.784
1373.488	1386.642
1446.925	1470.955
1646.778	1717.766
3158.382	3156.212
3171.654	3172.992
3269.88	3235.274
3270.754	3260.871

[Matt W]

Hi,

you are not consistently running a non-periodic system - you need a nonperiodic poisson section too. And 

1. You have 

&VIBRATIONAL_ANALYSIS

    DX 1.0E-03

    NPROC_REP  24

    FULLY_PERIODIC TRUE

the fully periodic keyword means that CP2K won't remove rotations (because they aren't there for a crystal with pbc) - so you get an extra 3 modes.

2. You are running a totally different 'model chemistry' in Gaussian terminology. You can't compare M062X with PBE directly. PBE (GGAs) typically underbind, bonds too long, frequencies too small. M06 maybe is better on average.

3 So you are mixing two things - accurate, i.e. properly performed calculation at a given level of theory - and agreement with experiment. Changing the model will change the frequencies. Changing settings (CUTOFF mainly, probably) will just ensure you get the correct numbers for the model. You should run PBE in Gaussian to compare to, if this is what you want.

Matt

[S Ling]

Hi

In addition to what Matt said, if you really want to compare results from CP2K and Gaussian, you should: (1) use the all-electron method, i.e. GAPW, in CP2K; (2) use the same basis sets; (3) use the same method to calculate frequencies. For (3), Gaussian calculates frequencies analytically by default, while CP2K calculates frequencies numerically. To calculate frequencies numerically in Gaussian, you should specify Freq=Numer. And also the calculated frequencies may be sensitive to the the step size for numerical differentiation, therefore, you should use the same step size in both calculations. In Gaussian, this is controlled by the Step keyword, see http://www.gaussian.com/g_tech/g_ur/k_freq.htm; in CP2K, this is controlled by the DX keyword.

SL

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe@googlegroups.com.
To post to this group, send email to cp...@googlegroups.com.
Visit this group at https://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.

[J. Ye]

Hi Matt, 

Thanks very much!. Actually, I found the poisson section will not affect the total energy of the molecule. I think this is reasonable for small molecule, but may affect the total energy of the system which is larger. Am I right? 

What does the poisson section do in the calculation?

Thanks very much!

[J. Ye]

Hi S Ling, 

Thanks very much! 

How to choose  a correct or reasonable step size (DX in CP2K)? I think DX value should be as close as the displacement of atoms. However, the displacement of atoms in real condition should be different. In the calculation, we are forced to use the same displacement (DX) to calculate the force constant. How does this affect the frequencies. Just curious? Maybe it is very minor. 

Why there's many other output generate other than f.out? 

For example, I have f.inp, f.out and other output files (xxr-1.out, xx-r-2.out, xx-r-3.out, .......xx-r-9.out). Each xx-r-x.out file looks like a single point calculations. 

Best,   

To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.

[Satish Kumar]

You can some tests with your preferred values of DX and see how your vibrational frequencies compare with available experimental data for example. 

You see multiple *.out files based on the value of NPROC_REP. If you run your calculation on 48 processors for example with NPROC_REP of 24, you will see 48/24 =2 output files.