LBFGS optimizer

[kai]

Dear all,

I am trying to use the LBFGS optimizer to do a climbing NEB calculation, in hope of getting a better convergence, ie 0.05 ev/A; all the images have been pre converged to 0.1 ev/A using the quasi-Newton built in the original vasp4.6.

The calculation is still running; but the energies so far from first ~50 force calls are fluctuate....is this a bit worrisome? I know for qusi-Newton, the energy is consistently dropping down.
1 F= -.72521407E+03 E0= -.72521407E+03 d E =-.725214E+03
2 F= -.72521774E+03 E0= -.72521774E+03 d E =-.367350E-02
3 F= -.72521756E+03 E0= -.72521756E+03 d E =0.179849E-03
4 F= -.72521830E+03 E0= -.72521830E+03 d E =-.742751E-03
5 F= -.72521981E+03 E0= -.72521981E+03 d E =-.151147E-02
6 F= -.72522059E+03 E0= -.72522059E+03 d E =-.778508E-03
7 F= -.72522122E+03 E0= -.72522122E+03 d E =-.622596E-03
8 F= -.72522205E+03 E0= -.72522205E+03 d E =-.831171E-03
9 F= -.72522233E+03 E0= -.72522233E+03 d E =-.279010E-03
10 F= -.72522080E+03 E0= -.72522080E+03 d E =0.153022E-02
11 F= -.72521376E+03 E0= -.72521376E+03 d E =0.703187E-02
12 F= -.72520248E+03 E0= -.72520248E+03 d E =0.112876E-01
13 F= -.72520759E+03 E0= -.72520759E+03 d E =-.510913E-02
14 F= -.72521775E+03 E0= -.72521775E+03 d E =-.101594E-01
15 F= -.72522142E+03 E0= -.72522142E+03 d E =-.367648E-02
16 F= -.72522213E+03 E0= -.72522213E+03 d E =-.707092E-03
17 F= -.72522185E+03 E0= -.72522185E+03 d E =0.277974E-03
18 F= -.72522104E+03 E0= -.72522104E+03 d E =0.811703E-03
19 F= -.72522014E+03 E0= -.72522014E+03 d E =0.900058E-03
20 F= -.72521919E+03 E0= -.72521919E+03 d E =0.944893E-03
21 F= -.72521789E+03 E0= -.72521789E+03 d E =0.130542E-02
22 F= -.72521738E+03 E0= -.72521738E+03 d E =0.505560E-03
23 F= -.72521685E+03 E0= -.72521685E+03 d E =0.535098E-03
24 F= -.72521601E+03 E0= -.72521601E+03 d E =0.840789E-03
25 F= -.72521477E+03 E0= -.72521477E+03 d E =0.123851E-02
26 F= -.72521372E+03 E0= -.72521372E+03 d E =0.104572E-02
27 F= -.72521350E+03 E0= -.72521350E+03 d E =0.224343E-03
28 F= -.72521400E+03 E0= -.72521400E+03 d E =-.499669E-03
29 F= -.72521456E+03 E0= -.72521456E+03 d E =-.563591E-03
30 F= -.72521496E+03 E0= -.72521496E+03 d E =-.397593E-03
31 F= -.72521533E+03 E0= -.72521533E+03 d E =-.367134E-03
32 F= -.72521554E+03 E0= -.72521554E+03 d E =-.215583E-03
33 F= -.72521513E+03 E0= -.72521513E+03 d E =0.407397E-03
34 F= -.72521385E+03 E0= -.72521385E+03 d E =0.128313E-02
35 F= -.72521207E+03 E0= -.72521207E+03 d E =0.177956E-02
36 F= -.72520982E+03 E0= -.72520982E+03 d E =0.225202E-02
37 F= -.72520703E+03 E0= -.72520703E+03 d E =0.279016E-02
38 F= -.72520225E+03 E0= -.72520225E+03 d E =0.478050E-02
39 F= -.72518941E+03 E0= -.72518941E+03 d E =0.128421E-01
40 F= -.72518460E+03 E0= -.72518460E+03 d E =0.480915E-02
41 F= -.72517808E+03 E0= -.72517808E+03 d E =0.651893E-02
42 F= -.72517966E+03 E0= -.72517966E+03 d E =-.158499E-02
43 F= -.72518352E+03 E0= -.72518352E+03 d E =-.386019E-02
44 F= -.72518508E+03 E0= -.72518508E+03 d E =-.156061E-02
45 F= -.72518448E+03 E0= -.72518448E+03 d E =0.601359E-03
46 F= -.72518070E+03 E0= -.72518070E+03 d E =0.378569E-02
47 F= -.72516758E+03 E0= -.72516758E+03 d E =0.131199E-01
48 F= -.72515821E+03 E0= -.72515821E+03 d E =0.936562E-02
49 F= -.72513905E+03 E0= -.72513905E+03 d E =0.191575E-01
50 F= -.72513128E+03 E0= -.72513128E+03 d E =0.777015E-02
51 F= -.72511927E+03 E0= -.72511927E+03 d E =0.120178E-01
52 F= -.72512287E+03 E0= -.72512287E+03 d E =-.360319E-02
53 F= -.72512451E+03 E0= -.72512451E+03 d E =-.164140E-02
54 F= -.72512411E+03 E0= -.72512411E+03 d E =0.404288E-03
55 F= -.72512139E+03 E0= -.72512138E+03 d E =0.272203E-02

[graeme]

The energy of images do not always drop in an NEB calculation. The climbing-image, for example, can climb up in energy to a saddle point. We would need to see the entire calculation to see if something is going wrong with this calculation.

[kai]

thx, prof. graeme, i will look at the result more carefully once the calculation is done...or hitting the wall time...

[kai]

I did a comparable clibming NEB calculation using both LBFGS and Quasi Newton calculation with 9 images. All images have already been pre-converged to 0.1 ev/Angtrom level before these two calculations started, aiming at a force at 0.05 ev/angstrom.

My question lies at the fact how to use LBGFGS better? I could use quasi netown but not LBFGS to converge the neb.

One extra point even puzzled me is the fact that quasi newton is giving image 4 as saddle point (as expected, this is the saddle point in the precoverged calculaltion) while LBGFGS is pushing image 3 to saddle point. Is that the reason why quasi works better? In pre-converging thoese images to 0.1 ev/ang, i also used the climibing giving saddle point at image 4.

The Quasi Newton works nicely and converged within ~ 586 ionic steps, giving:

0 0.00000000 -725.25933600 0.00000000
1 0.04798400 -725.16043800 0.09889800
2 0.03496300 -725.16043500 0.09890100
3 0.01429300 -725.17901700 0.08031900
4 0.01258700 -724.99759100 0.26174500
5 0.01369900 -726.53056900 -1.27123300
6 0.02620400 -726.76467100 -1.50533500
7 0.04924000 -726.94720400 -1.68786800
8 0.02405400 -727.13057500 -1.87123900
9 0.03171200 -727.18864600 -1.92931000
10 0.00000000 -727.24304000 -1.98370400

while LBFGS couldnt converged the system before hiting the wall time

0 0.00000000 -725.25933600 0.00000000
1 0.26148300 -725.12632800 0.13300800
2 0.59448300 -725.17887300 0.08046300
3 1.19119300 -724.94696500 0.31237100
4 0.52011200 -726.34235700 -1.08302100
5 0.20903200 -726.68181200 -1.42247600
6 0.54717400 -726.88811300 -1.62877700
7 0.29292300 -727.09436800 -1.83503200
8 0.11763400 -727.20357900 -1.94424300
9 0.44389100 -727.18227700 -1.92294100
10 0.00000000 -727.24304000 -1.98370400