Lanczos question

[barryhyc]

Dear all,
I download the VTST tools and recompile the vasp (4.6.35). When I use Lanzcos, I found the force is always high(use vef script, the force is more than 2). I am

looking forward the decline of the force, but when I take the intermediate configuration, unfortunately, the structure looks like IS (or FS). I tried two systems, the

same things happened.
This is my INCAR
SYSTEM=25percentzn-223
ENAUG=650
ENCUT = 400.0
NELMIN = 3
EDIFF = 1E-05 #stopping-criterion for ELM
LREAL = Auto #real-space projection
LMAXPAW= 0 #max onsite density
VOSKOWN= 1 #Vosko Wilk Nusair interpolation
EDIFFG = -0.03 #stopping-criterion for IOM
NSW = 500 #number of steps for IOM
IBRION = 3 #ionic relax: 0-MD 1-quasi-New 2-CG
POTIM = 0 #time-step for ionic-motion
ISMEAR = 1
SIGMA = 0.10
IDIPOL = 3 #1-x, 2-y, 3-z
IALGO = 48
GGA = 91
LPLANE = .TRUE.
ICHAIN = 3
SNL = 20 #Maximum size of Lanczos matrix
SdR = 5.0E-3 #Finite difference step length
SLTOL = 1.0E-2 #Tolerance for eigenvalue convergence
IOPT = 2
MAXMOVE = 0.2
FDSTEP = 5.0E-3

So, is the problem of my INCAR(increase the parameters of SdR?) or other reasons? Please give me some suggestions, many thanks.

[graeme]

If you could upload a tar.gz of your calculation, we'll take a look at it. I've increased the maximum file size so that you should be able upload it here.

[barryhyc]

Thanks for your promptly response. I upload as an attachment here, please help me analysis the reason. Also, I still trying with the dimer method which I use optimized H2COO on substate as an initial state, and give the direction of C-O bond breaking to produce CH2O and O with MODECAR file. It looks like the curvature is going down(from 13.89 at the first step to 1.76 at the fourth step), however, the force is still relatively large(6 to 8). Is it necessary to continue? How to improve the efficiency of dimer?

Sorry, the maximum upload size is only 2MB and my compressed file is about 6MB. So, do you mind I send you as an e-mail? Thank you again.

[graeme]

Ok, the posting issue should be fixed now, with a limit of 200MB. You can also email it to me.

[barryhyc]

Attached here. Please check it.

How about my dimer question, thanks again.

[graeme]

The biggest problem appears to be that the initial structure has a very short O-Zn bond length of ~1 Ang. This leads to high forces, on the order of 150 eV/Ang. While the calculation may eventually converge, it will be slow, especially with this large system.

If you know the reaction mechanism that you want to study, I would recommend using the NEB (even with 1 or 3 images) to find a rough path that is physically reasonable. Then you can run the neb2dim.pl or neb2lan.pl to initiate a lanczos/dimer calculation closer to the saddle.

I would also suggest using less expensive parameters to get close to the saddle and then increase the accuracy and refine your calculation. For example, initially use gamma point k-point sampling and a frozen (even 3 layer) Pd slab. Use the soft O_s and C_s potential and a cutoff ~270 eV. Combined, these changes could reduce the cost by an order of magnitude.

[barryhyc]

Thank Prof. graeme.

The reason for so high forces is because I use the script of nebmake.pl to produce one immediate configuration which I think it has the proper C-O distance. I ignore the O-Zn bond length. Indeed, when I search for TSs, I always use cneb method to obtain likeness TS and then optimized with quasi-Newton algorithm. However, this system I cannot locate TS for our previous method. I will adopt your suggestion of using neb2dim(lan).pl to setup TS calculation.

Your suggestions of reducing computation cost seems very efficient. Thank for your so good strategies.