transition state vibrational frequencies

[mff7d]

I'm doing some nudged elastic band calculations (NEB) and taking the maximum from that as my transition state and running vibrational calculations on it so that I might determine a rate constant of sorts.

I'm running into some problems. I'm finding that some of my vibrational frequency calculations of my transition states are resulting in all positive numbers, too many or too few negative vibrational frequencies.

What's going on? Does anyone have some insight into this?

[admin]

negative frequencies simply say that the vibration becomes soft along the mode. However, please check whether your calculations are really well-converged if (some of) the vibration frequencies seem strange to you: ie. all displacement steps should be
--) well converged electronically,
--) small enough to make sure that they are in the harmonic limit (ie POTIM not larger than 0.015, in general)
--) EDIFF small (1e-06)

[mff7d]

I believe that the problem is not in my vibrational calculations, but in my nudged elastic band calculations. When I look at the vibrational freq output I see that some degrees of freedom are converged (well or peak) and that others are not (not a maxima or minima).

Key parameters in my INCAR for my vibrational calculations are as follows:

EDIFF = 1E-5
EDIFFG = 1E-4
NSW = 1000
IBRION = 5
POTIM = 0.04
NFREE = 2
NPAR = 1

When I look in the output files I find evidence that the site (from the POSCAR) being investigated is not fully relaxed. I'm finding that some degrees of freedom suggest a well, some a peak, and some an unconverged solution.

Below are the key INCAR parameters for my nudged elastic band algorthm.

Electronic minimization
PREC = NORMAL
GGA = 91
VOSKOWN = 1
LREAL = .TRUE.
IALGO = 48
EDIFF = 1e-4
ISYM = 0
ISPIN = 1
NELM = 400
## LVTOT = .TRUE.
Ionic relaxation
EDIFFG = -0.05
NSW = 200
IBRION = 1
POTIM = 0.5
DOS related values
ISMEAR = 2
SIGMA = 0.2
Parallelization
ENMAX = 395.994
NSIM = 4
NPAR = 12
LPLANE = .TRUE.
LSCALU = .FALSE.

IMAGES = 4
ICHAIN = 0
SPRING = -5
SPRING2 = -5
ISPRING = 1
SPOWER = 1
EFIRST = -20.848363
ELAST = -20.848363
LCLIMB = .TRUE.
LTANGENT = .TRUE.

As you can see, I'm using four images. I have not done one image and then four thereafter. What might I do to cause the nudged elastic band to converge more accurately?

[dcurulla]

Hi,

I think that your settings for vibrational calculations are not the best you can use. First, EDIFF should be at least 1e-6 (and not 1e-5 as you use). Think that to get a resonable accuracy for the frequencies, you need to have a very good accuracy for the forces. Second, displacements for the frequency calculations should be as small as 0.02 A (or 0.015 as admin suggests). The choice of such number is a balance between getting new forces that are significantly larger that the order of magnitude of your error in the forces (determined by EDIFF) and constraining the displacement within the harmonic region of the vibration; in this sense 0.02 is a good compromise.

On the other hand, frequency calculations are somewhat tricky. If you obtain more than one imaginary frequency, it is possible that your structure is 1) not well converged, 2) a higher order saddle point indeed or 3) the frequency calculation is not good enough and you'll have to repeat it. Only one suggestion, if you take the highest energy image structure as your TS, do first a standard geometry optimization before doing the frequency calculation.

Cheers

[askhetan]

dear dcurulla,

in the NEB calculations, does the algo not already geometrically optimise the intermediate states? why do we need another geometrical optimization?