Continuous Run for Dimer Method
Moderator: moderators
-
- Posts: 28
- Joined: Fri Aug 03, 2018 7:06 am
Continuous Run for Dimer Method
Hi,
I am trying to find T.S. for CH+H on Ni-Cu catalyst.I have a some of questions.
1-I have run cNEB for NSW=50 for two time (using vfin.pl for continuous run) ,then use neb2dim.pl to utilize dimer method for NSW=100.Still calculation did not converge and wavefunction have been written after iterations are finished.How I can resume calculation (continuous run) from latest dimer calculation?
2-What is the best force convergence to switch from cNEB to Dimer ,knowing that EDIFFG= -0.01?
3-What is recommended optimizer for such case, currently I am using IOPT=3 and start initially with NSW=50 up to 150 iterations, but convergence is too slow and it goes away from my EDIFFG setting after 2nd run?
Thank you
A.Omran
I am trying to find T.S. for CH+H on Ni-Cu catalyst.I have a some of questions.
1-I have run cNEB for NSW=50 for two time (using vfin.pl for continuous run) ,then use neb2dim.pl to utilize dimer method for NSW=100.Still calculation did not converge and wavefunction have been written after iterations are finished.How I can resume calculation (continuous run) from latest dimer calculation?
2-What is the best force convergence to switch from cNEB to Dimer ,knowing that EDIFFG= -0.01?
3-What is recommended optimizer for such case, currently I am using IOPT=3 and start initially with NSW=50 up to 150 iterations, but convergence is too slow and it goes away from my EDIFFG setting after 2nd run?
Thank you
A.Omran
Re: Continuous Run for Dimer Method
1. If you run the vfin.pl script in the directory of a dimer calculation, it will move the results to the subdirectory that you specify and the calculate will be ready for continuation.
2. You want to make sure that your cNEB calculation has stably found a reaction path before switching to the dimer method. A force certainly lower than 0.5 eV/Ang on the climbing image is recommended.
3. IOPT=3 is a stable optimizer that can be used when the forces are high (above 1 eV/Ang). L-BFGS is generally the fastest when you are closer to convergence.
2. You want to make sure that your cNEB calculation has stably found a reaction path before switching to the dimer method. A force certainly lower than 0.5 eV/Ang on the climbing image is recommended.
3. IOPT=3 is a stable optimizer that can be used when the forces are high (above 1 eV/Ang). L-BFGS is generally the fastest when you are closer to convergence.
-
- Posts: 28
- Joined: Fri Aug 03, 2018 7:06 am
Re: Continuous Run for Dimer Method
Thank you Dr.Graeme for your valuable support
-
- Posts: 28
- Joined: Fri Aug 03, 2018 7:06 am
Re: Continuous Run for Dimer Method
Hi
I stil face some problems when switching from QM optimizer to Dimer.For instance,I check of CH dissociation on catalyst and I got the following results from 8 images using nebef.dat:After 50 iterations,I got
0 0.000000 -187.570500 0.000000
1 0.056474 -187.526100 0.044400
2 0.129705 -187.352300 0.218200
3 0.102515 -186.971000 0.599500
4 0.045191 -186.507800 1.062700
5 0.026198 -186.166800 1.403700
6 0.094932 -186.483800 1.086700
7 0.174650 -186.627700 0.942800
8 0.096999 -186.663800 0.906700
9 0.009391 -186.877700 0.692800
As force become <0.5 eV,I switched to Dimer I got some positive curvature and forces are not converging.
Any recommendations? Should I switch to IOPT=1 in that case?
Thank you
I stil face some problems when switching from QM optimizer to Dimer.For instance,I check of CH dissociation on catalyst and I got the following results from 8 images using nebef.dat:After 50 iterations,I got
0 0.000000 -187.570500 0.000000
1 0.056474 -187.526100 0.044400
2 0.129705 -187.352300 0.218200
3 0.102515 -186.971000 0.599500
4 0.045191 -186.507800 1.062700
5 0.026198 -186.166800 1.403700
6 0.094932 -186.483800 1.086700
7 0.174650 -186.627700 0.942800
8 0.096999 -186.663800 0.906700
9 0.009391 -186.877700 0.692800
As force become <0.5 eV,I switched to Dimer I got some positive curvature and forces are not converging.
Any recommendations? Should I switch to IOPT=1 in that case?
Thank you
- Attachments
-
- INCAR.txt
- (930 Bytes) Downloaded 17778 times
-
- DIMCAR.txt
- (3.79 KiB) Downloaded 17754 times
Re: Continuous Run for Dimer Method
This looks strange - did you run the neb2dim.pl script to initialize your dimer calculation?
Your dimer calculation starts with a much higher energy and force as compared to image 5 in your NEB calculation.
You could try using the geometry from image 05 and a MODECAR generated from 04 and 06 as in: modemake.pl 04/POSCAR 06/POSCAR
Anyway, something does not make sense and it is not just an issue with the optimizer.
Your dimer calculation starts with a much higher energy and force as compared to image 5 in your NEB calculation.
You could try using the geometry from image 05 and a MODECAR generated from 04 and 06 as in: modemake.pl 04/POSCAR 06/POSCAR
Anyway, something does not make sense and it is not just an issue with the optimizer.
-
- Posts: 28
- Joined: Fri Aug 03, 2018 7:06 am
Re: Continuous Run for Dimer Method
Thank you for your support.
Yes,I have used neb2dim.pl as follow:From NEB directory I make
1)vfin.pl Results 2)~>Results/nebresults.pl 3)~>Results/neb2dim.pl .The last step created "dim" folder with (INCAR KPOINTS MODECAR POSCAR POTCAR) which I used for the next run.
I have already continued the cNEB calculation again (IOPT=3) for another 80 stepsand I got:
0 0.000000 -187.570500 0.000000
1 0.053464 -187.526300 0.044200
2 0.100702 -187.351000 0.219500
3 0.072140 -186.967600 0.602900
4 0.030645 -186.508100 1.062400
5 0.019374 -186.167300 1.403200
6 0.093806 -186.508400 1.062100
7 0.106740 -186.637800 0.932700
8 0.068833 -186.657200 0.913300
9 0.009391 -186.877700 0.692800
It seems O.K (Is n't it?) soI will try to switch to dimer now as mentioned above and see if it will work ,otherwise I will go modemake.pl.For that approach,
choosing image 5 is because the energy difference is highest?
Yes,I have used neb2dim.pl as follow:From NEB directory I make
1)vfin.pl Results 2)~>Results/nebresults.pl 3)~>Results/neb2dim.pl .The last step created "dim" folder with (INCAR KPOINTS MODECAR POSCAR POTCAR) which I used for the next run.
I have already continued the cNEB calculation again (IOPT=3) for another 80 stepsand I got:
0 0.000000 -187.570500 0.000000
1 0.053464 -187.526300 0.044200
2 0.100702 -187.351000 0.219500
3 0.072140 -186.967600 0.602900
4 0.030645 -186.508100 1.062400
5 0.019374 -186.167300 1.403200
6 0.093806 -186.508400 1.062100
7 0.106740 -186.637800 0.932700
8 0.068833 -186.657200 0.913300
9 0.009391 -186.877700 0.692800
It seems O.K (Is n't it?) soI will try to switch to dimer now as mentioned above and see if it will work ,otherwise I will go modemake.pl.For that approach,
choosing image 5 is because the energy difference is highest?
Re: Continuous Run for Dimer Method
Well, without having resolved the problems with the dimer calculation, I think that your continued ci-NEB calculation looks pretty good. You have a saddle point geometry at the 05 image with a barrier of 1.4 eV and a residual force of only 0.02 eV/Ang.
-
- Posts: 28
- Joined: Fri Aug 03, 2018 7:06 am
Re: Continuous Run for Dimer Method
Hi,Thanks for your patience and explanation.
Here is result when I switch again DImer after 150 cNEB iterations
Step Force Torque Energy Curvature Angle
1 0.84923 3.62994 -185.70560 -2.23269 18.47678
2 0.96178 2.76292 -185.70971 -3.97754 17.11630
3 1.13131 3.61388 -185.74328 -6.45217 11.62957
4 1.40516 5.29049 -185.87136 -7.29901 11.06646
5 1.16680 3.36944 -185.85990 -6.60446 -1.15874
6 1.04269 3.57476 -185.90329 -4.17793 3.42256
7 1.12233 4.37275 -186.06740 -1.43998 2.10034
8 1.07367 3.48640 -186.10194 -3.53575 4.48217
9 1.16114 2.83955 -186.11139 -3.92176 2.99372
10 0.89190 2.84295 -186.14214 -4.28931 9.03458
11 0.68610 3.56411 -186.16052 -3.18313 10.00363
12 0.55203 3.87381 -186.16079 -3.05244 5.66424
13 0.41947 4.65872 -186.16722 -3.13473 8.45496
14 0.48925 4.92214 -186.17625 -3.67440 -0.07800
15 0.39781 9.78384 -186.18225 -3.67915 -12.62329
16 0.38417 4.90992 -186.18344 -4.46433 12.73036
17 0.30428 5.67121 -186.18372 -3.71304 1.99846
18 0.25898 6.02609 -186.18466 -4.69276 17.52249
19 0.15166 4.28136 -186.18670 -4.95530 7.61024
20 0.12615 6.67854 -186.18730 -3.70547 6.17960
21 0.14061 5.57767 -186.18767 -3.88517 -6.38080
22 0.13453 6.55104 -186.18807 -3.76668 3.85917
23 0.06497 5.35770 -186.18860 -4.08607 0.90924
24 0.06560 6.65373 -186.18904 -3.92010 0.22148
25 0.06025 5.15128 -186.18910 -4.46243 -4.96257
26 0.06066 5.30112 -186.18951 -5.15377 -5.78318
27 0.07921 6.05269 -186.18960 -3.86014 11.41628
28 0.05860 4.79630 -186.18985 -4.53556 -5.73090
29 0.05741 5.93140 -186.19011 -5.02139 2.19062
30 0.06985 5.67415 -186.19027 -4.23092 -1.35739
31 0.03365 4.32841 -186.19030 -5.15312 -0.96875
32 0.03800 5.07795 -186.19035 -4.98681 1.44224
33 0.03546 5.25979 -186.19042 -4.80138 1.80390
34 0.03537 5.77536 -186.19045 -4.23083 -2.61033
35 0.02822 5.14491 -186.19053 -4.86009 1.61267
36 0.02864 5.75525 -186.19054 -4.20691 0.87834
37 0.03811 6.08218 -186.19060 -4.23730 0.72277
38 0.02541 4.89108 -186.19062 -4.62217 -2.17549
39 0.02215 --- -186.19063 --- ---
However, I have not changed anything on my way or INCAR file.The dimer looks better from last time;at least Force are coming down with -ve curvature.
Is it because I started DIMER this time from lower Force ? or it is a problem in my INCAR (i.e EDIFF = 1.0E-04 should be change to 1.0E-07 or 1.0E-08?)
The same problem still persist with all other calculation too whenever I switch to Dimer from cNEB.The INCAR is attached.
Here is result when I switch again DImer after 150 cNEB iterations
Step Force Torque Energy Curvature Angle
1 0.84923 3.62994 -185.70560 -2.23269 18.47678
2 0.96178 2.76292 -185.70971 -3.97754 17.11630
3 1.13131 3.61388 -185.74328 -6.45217 11.62957
4 1.40516 5.29049 -185.87136 -7.29901 11.06646
5 1.16680 3.36944 -185.85990 -6.60446 -1.15874
6 1.04269 3.57476 -185.90329 -4.17793 3.42256
7 1.12233 4.37275 -186.06740 -1.43998 2.10034
8 1.07367 3.48640 -186.10194 -3.53575 4.48217
9 1.16114 2.83955 -186.11139 -3.92176 2.99372
10 0.89190 2.84295 -186.14214 -4.28931 9.03458
11 0.68610 3.56411 -186.16052 -3.18313 10.00363
12 0.55203 3.87381 -186.16079 -3.05244 5.66424
13 0.41947 4.65872 -186.16722 -3.13473 8.45496
14 0.48925 4.92214 -186.17625 -3.67440 -0.07800
15 0.39781 9.78384 -186.18225 -3.67915 -12.62329
16 0.38417 4.90992 -186.18344 -4.46433 12.73036
17 0.30428 5.67121 -186.18372 -3.71304 1.99846
18 0.25898 6.02609 -186.18466 -4.69276 17.52249
19 0.15166 4.28136 -186.18670 -4.95530 7.61024
20 0.12615 6.67854 -186.18730 -3.70547 6.17960
21 0.14061 5.57767 -186.18767 -3.88517 -6.38080
22 0.13453 6.55104 -186.18807 -3.76668 3.85917
23 0.06497 5.35770 -186.18860 -4.08607 0.90924
24 0.06560 6.65373 -186.18904 -3.92010 0.22148
25 0.06025 5.15128 -186.18910 -4.46243 -4.96257
26 0.06066 5.30112 -186.18951 -5.15377 -5.78318
27 0.07921 6.05269 -186.18960 -3.86014 11.41628
28 0.05860 4.79630 -186.18985 -4.53556 -5.73090
29 0.05741 5.93140 -186.19011 -5.02139 2.19062
30 0.06985 5.67415 -186.19027 -4.23092 -1.35739
31 0.03365 4.32841 -186.19030 -5.15312 -0.96875
32 0.03800 5.07795 -186.19035 -4.98681 1.44224
33 0.03546 5.25979 -186.19042 -4.80138 1.80390
34 0.03537 5.77536 -186.19045 -4.23083 -2.61033
35 0.02822 5.14491 -186.19053 -4.86009 1.61267
36 0.02864 5.75525 -186.19054 -4.20691 0.87834
37 0.03811 6.08218 -186.19060 -4.23730 0.72277
38 0.02541 4.89108 -186.19062 -4.62217 -2.17549
39 0.02215 --- -186.19063 --- ---
However, I have not changed anything on my way or INCAR file.The dimer looks better from last time;at least Force are coming down with -ve curvature.
Is it because I started DIMER this time from lower Force ? or it is a problem in my INCAR (i.e EDIFF = 1.0E-04 should be change to 1.0E-07 or 1.0E-08?)
The same problem still persist with all other calculation too whenever I switch to Dimer from cNEB.The INCAR is attached.
- Attachments
-
- INCAR.txt
- (930 Bytes) Downloaded 17850 times
Re: Continuous Run for Dimer Method
Ah, you need a lower ediff for the dimer calculation. Using 1e-4 will result in too much noise in the forces so that the curvature calculation will be unreliable, as well as the dimer calculation. Use 1e-7 or 1e-8.
-
- Posts: 28
- Joined: Fri Aug 03, 2018 7:06 am
Re: Continuous Run for Dimer Method
Extending my gratitude for your help, I could manage to get convergence with Ediff=1e-4 (in this particular case below)with energy barrier 1.41 as you expected .I guess probably it worked as the convergence has already been close with the residual force 0.02 eV (ediffg=0.01).
For other Dimer calculations with problems,I changed the setting with Ediff=1e-7 instead of 1e-4 and up till now it showed improvement ;particularly curvature become consistently negative and forces are reducing although very slowly.
For other Dimer calculations with problems,I changed the setting with Ediff=1e-7 instead of 1e-4 and up till now it showed improvement ;particularly curvature become consistently negative and forces are reducing although very slowly.
-
- Posts: 28
- Joined: Fri Aug 03, 2018 7:06 am
Re: Continuous Run for Dimer Method
Hi,
For dimer continuous run,In addition to vfin.pl run2 files,Can I use CHGCAR ,WAVECAR from previous run with ISTART=1 ,ICHARG=1 as per regular VASP run? In that case ,the system will use CENTCAR or CONTCAR?
Can I do the same with cNEB (i.e gunzip all WAVECAR ,CHGCAR in images after using vfin.pl,then adjust INCAR setting ISTART=1 ,ICHARG=1 )?
For dimer continuous run,In addition to vfin.pl run2 files,Can I use CHGCAR ,WAVECAR from previous run with ISTART=1 ,ICHARG=1 as per regular VASP run? In that case ,the system will use CENTCAR or CONTCAR?
Can I do the same with cNEB (i.e gunzip all WAVECAR ,CHGCAR in images after using vfin.pl,then adjust INCAR setting ISTART=1 ,ICHARG=1 )?
Re: Continuous Run for Dimer Method
Yes, for sure. The dimer calculation should be continued with the CENTCAR file. If you use the vfin.pl script, the CENTCAR will be copied to the POSCAR, ready for another calculation.