Dimer from NEB...moving away from saddle point?

[kai]

Hello, all

I am starting to learn dimer method as a means to achieve a more accurate saddle point; my first trial was running, though showing an unhealthy DIMCAR as:

Step Force Torque Energy Curvature Angle
1 0.07200 9.95315 -761.18452 -0.20945 24.31446
1 0.07200 12.68468 -761.18452 1.34022 19.30569
1 0.07200 11.11699 -761.18452 -0.09875 12.53713
1 0.07200 11.28598 -761.18452 -0.08494 16.17633
2 0.11176 15.26268 -761.18452 1.45914 5.90628
2 0.11176 15.86004 -761.18452 1.52288 4.55730
2 0.11176 15.65949 -761.18452 1.59415 23.35985
2 0.11176 12.52592 -761.18452 -0.08010 3.89426
3 0.12213 15.00516 -761.18472 1.74450 15.54868
3 0.12213 14.98464 -761.18472 1.10719 11.45278
3 0.12213 14.31655 -761.18472 1.22160 8.20060
3 0.12213 13.91477 -761.18472 1.11897 6.29399
4 1.65860 11.22268 -761.15139 3.58126 -10.19078
4 1.65860 11.09354 -761.15139 3.22923 -4.63448
4 1.65860 11.01191 -761.15139 3.08150 1.62978
4 1.65860 10.78061 -761.15139 3.00331 4.10809
5 3.35792 14.69485 -760.99681 9.17810 -17.22740
5 3.35792 13.95845 -760.99681 7.86717 -5.94698
5 3.35792 14.06279 -760.99681 7.62851 1.71601
5 3.35792 13.92514 -760.99681 7.68135 -0.75636
6 5.94754 19.43142 -760.60527 12.89405 -27.17173
6 5.94754 17.92166 -760.60527 10.29255 -5.22524
6 5.94754 17.85908 -760.60527 10.18835 -3.11135
6 5.94754 17.64082 -760.60527 9.90490 0.65818

Note the dimer calculation starts from a successful climbing NEB (force down to 0.08 ev/A); but i dont understand why dimer is moving away from saddle point.

MY DIMER INCAR is quite standard (dimer part) as:

ICHAIN=2
DdR=0.005
DRotMax=4
DFNMin=0.01
DFNMax=1.0

IOPT = 2
EDIFF = 1E-7

could anyone suggest anything?

Thanks

[graeme]

Something is not working correctly. The key quantity is the curvature. If the curvature stays negative it indicates that you are near a saddle. The calculation starts with a negative curvature but then in the first four rotation steps, there are large rotations in which the curvature increases -- it should not do this.

As a test, I suggest allowing more rotations by setting DRotMax=10 or 15 to see if the rotation converges to a fixed curvature with a small rotation angle. If it does not, there is no point allowing the dimer to translate.

You might try ediff 1e-8, but my feeling is that there is something else going wrong. Is there anything unusual about this calculation? Are the electronic iterations converging? Are you using DFT+U or the hybrid functional? How about cell relaxation with ISIF?

If you are willing to post a .tar.gz file of the run, we can look to see if we can see a problem.

[kai]

Thx a lot, Graeme; here is the calculation; i removed the CHARGE file since its way too big;

The geometry of this saddle point should be alright and the climbing NEB managed to converge it down to a force of 0.02 ev/A lately;

There is something missing in my mind regarding the parameters in the dimer calculation; and I think the standard setting appears a bit too aggressive to me for this system.

My limited knowledge does not give me too many hints at present; the electronic structure converges well and i am using the standard PBE-PAW

[graeme]

I ran a couple of tests, and they all seem to be converging properly. I'll attach the DIMCAR for the calculation which uses the same parameters as yours. It reached the convergence criteria in 18 translation steps. I still don't know why your calculations are behaving differently. If you have a different machine to test this on, you might see if a different level of parallelization or math library make a difference. If you want to help debug this, I can also give you an account on a machine here which gave this proper behavior:


Step Force Torque Energy Curvature Angle
1 0.31072 3.51150 -761.36422 -1.24340 6.58193
1 0.31072 3.64324 -761.36422 -1.63631 21.13665
1 0.31072 4.50464 -761.36422 -2.49720 27.04897
1 0.31072 4.11735 -761.36422 -3.26151 20.90051
2 0.25332 3.24759 -761.36807 -3.50405 13.08561
2 0.25332 2.56584 -761.36807 -3.71151 11.59431
2 0.25332 2.37194 -761.36807 -3.69722 10.12387
2 0.25332 1.68766 -761.36807 -3.75172 6.28486
3 0.33324 2.28740 -761.36963 -3.38788 4.08628
3 0.33324 1.79129 -761.36963 -3.54600 2.72318
3 0.33324 1.41512 -761.36963 -3.63629 2.80441
3 0.33324 1.33000 -761.36963 -3.65506 1.30190
4 0.27319 1.59674 -761.37227 -3.95364 4.44551
4 0.27319 1.46002 -761.37227 -3.94095 1.51754
4 0.27319 0.84991 -761.37227 -3.99774 1.40049
5 0.27895 1.60090 -761.37387 -3.48754 2.21439
5 0.27895 1.28957 -761.37387 -3.50787 1.36401
5 0.27895 0.83582 -761.37387 -3.50937 1.40548
6 0.26613 1.54579 -761.37541 -3.37417 1.97003
6 0.26613 1.27387 -761.37541 -3.41141 1.76493
6 0.26613 0.95696 -761.37541 -3.45269 1.16686
7 0.18117 1.42749 -761.37658 -3.39673 2.01628
7 0.18117 1.18270 -761.37658 -3.43990 1.27065
7 0.18117 0.75544 -761.37658 -3.45417 1.03133
8 0.18913 1.39135 -761.37726 -3.47549 1.89070
8 0.18913 1.17345 -761.37726 -3.45602 1.33609
8 0.18913 0.75657 -761.37726 -3.51698 1.02809
9 0.16570 1.28124 -761.37811 -3.24445 1.38849
9 0.16570 0.90345 -761.37811 -3.30229 1.18938
10 0.17145 1.22050 -761.37882 -3.51585 1.93724
10 0.17145 1.16157 -761.37882 -3.52386 1.24754
10 0.17145 0.85935 -761.37882 -3.55936 1.65580
11 0.18632 1.35943 -761.37976 -2.95730 1.66438
11 0.18632 1.04646 -761.37976 -3.02781 1.19168
11 0.18632 0.76708 -761.37976 -3.02800 1.06831
12 0.15731 1.13327 -761.38031 -3.38870 1.05573
12 0.15731 0.70896 -761.38031 -3.40870 1.45175
13 0.13622 1.15162 -761.38089 -3.82917 1.73381
13 0.13622 1.10571 -761.38089 -3.84002 1.14203
13 0.13622 0.72696 -761.38089 -3.82286 1.01774
14 0.13639 1.25451 -761.38123 -3.30508 1.18858
14 0.13639 0.86258 -761.38123 -3.37195 1.63660
15 0.09893 1.25508 -761.38155 -3.54895 1.77549
15 0.09893 1.18463 -761.38155 -3.56193 1.29178
15 0.09893 0.79834 -761.38155 -3.51880 1.16422
16 0.10780 1.21212 -761.38179 -3.39522 2.48978
16 0.10780 1.23148 -761.38179 -3.39663 1.11134
16 0.10780 0.83189 -761.38179 -3.40702 1.99781
17 0.09947 1.24838 -761.38199 -3.30777 1.65645
17 0.09947 1.08831 -761.38199 -3.31279 1.30418
17 0.09947 0.73582 -761.38199 -3.33846 0.92835
18 0.09947 --- -761.38199 --- ---

[kai]

thanks a lot, Graeme....I will do a further look and update here if i find something interesting

[cenwanglai]

What does the DFT + U matter to the dimer calculation?

Is there any problem comparing to DFT without U correction?