- input layer sep
- 3.35 A; same as AB-stacked BLG
- input parameter a
- 2.46 A; same as graphene
- relaxed layer sep
- 3.3737 A (same for both pairs of adjacent layers)
- relaxed parameter a
- 2.417 A
- symmetry
- P1 (!!!!)
- negative frequency bump around gamma in the lowest freq single band
freqs of lowest band at small x near gamma
| THz | wavenumber |
|---|---|
| -0.0000009477 | -3.1611870e-5 |
| -0.0102594265 | -0.34221764 |
| -0.0186724439 | -0.62284569 |
| -0.0241636577 | -0.80601287 |
| -0.0236045708 | -0.78736374 |
| 0.0092987455 | 0.31017277 |
./files/abc-manual-vdw/eigenvalues.yaml
- Plot energy vs (a,c) WITHOUT relaxing
- Where is the min energy?
- How does it compare to what was found automatically?
- Relax that
- How are the frequencies now?
Computation: ./files/do-params/proj/params/abc
Results for ABC can be seen here:
Google sheets for ABC manual data
Computation: ./files/do-params/work/closeby-abc-fixed
- layer sep
- 3.376 A
- parameter a
- 2.4231 A (= 0.985 * 2.46)
- iterations to relax
- 4
- relaxed energy
- -10608.260 eV
- symmetry
- P-3m1 (164)
freqs of lowest band at small x near gamma (no negative bump!)
| THz | wavenumber |
|---|---|
| -0.0000002956 | -9.8601548e-6 |
| 0.0259468139 | 0.86549256 |
| 0.0525448303 | 1.7527069 |
| 0.0800437164 | 2.6699710 |
| 0.1089008127 | 3.6325402 |
| 0.1395173641 | 4.6537984 |
:'<,'>! grep -A3 band | grep frequency
Computation: ./files/do-params/work/closeby-abc-auto
NOTE: Input looks identical to the fixed one because golden search configuration is hardcoded in the Shakefile.
- initial layer sep
- 3.376 A
- initial parameter a
- 2.4231 A
- golden search range
- (0.975, 1.036) times initial a (2.363, 2.510)
- golden search tolerance
- 1e-3
- relaxed layer sep
- 3.3737
- relaxed parameter a
- 2.4158 A
- worst energy attained
- -10589.260
- approx iterations to relax, for each value of ‘a’ tested
- 70, 10, 25, 10, 35, 10, 95 (note: the counts of ~10 are always when it goes from a smaller value of ‘a’ to a larger value; the larger counts are when ‘a’ decreases)
- relaxed energy
- -10608.569 eV
- symmetry
- P-3m1
freqs of lowest band at small x near gamma (bad!)
| THz | wavenumber |
|---|---|
| -0.0000015145 | -5.0518283e-5 |
| -0.0170219320 | -0.56779054 |
| -0.0329636166 | -1.0995479 |
| -0.0473640229 | -1.5798938 |
| -0.0592021276 | -1.9747704 |
| -0.0671887257 | -2.2411747 |
| -0.0693825483 | -2.3143527 |
| -0.0620172616 | -2.0686732 |
| -0.0308961545 | -1.0305848 |
| 0.0679308172 | 2.2659282 |
Computation: ./files/do-params/work/closeby-abc-narrower
that previous one had a very large search interval compared to the interval we manually searched for the center point, which seems silly. This interval is more reasonably narrowed down.
- initial layer sep
- 3.376 A
- initial parameter a
- 2.4231 A
- golden search range
- (0.99, 1.01) times initial a; (2.399 A, 2.447 A), or (0.975, 0.995) times 2.46
- golden search tolerance
- 1e-3
- relaxed layer sep
- 3.3739 A
- relaxed parameter a
- 2.4147 A
- worst energy attained
- -10606.850000
- approx iterations to relax, for each value of ‘a’ tested
- (60, 10, 50, 10, (as before, the 10-iteration ones are precisely those where ‘a’ increases)
- relaxed energy
- 10608.463
- symmetry
- P-3m1
freqs of lowest band at small x near gamma
| THz | wavenumber |
|---|---|
| -0.0000008226 | -2.7438983e-5 |
| -0.0211821797 | -0.70656147 |
| -0.0415198502 | -1.3849531 |
| -0.0606400597 | -2.0227347 |
| -0.0777968169 | -2.5950225 |
| -0.0921453269 | -3.0736373 |
| -0.1026288422 | -3.4233297 |
| -0.1077867267 | -3.5953782 |
| -0.1052963341 | -3.5123077 |
| -0.0904623818 | -3.0175003 |
| -0.0453065519 | -1.5112639 |
| 0.0908479236 | 3.0303606 |
Computation: ./files/do-params/work/??????????????/
- initial layer sep
- 3.376 A
- initial parameter a
- 2.4231 A
- golden search range
- (0.99, 1.01) times initial a; (2.399 A, 2.447 A), or (0.975, 0.995) times 2.46
- golden search tolerance
- 1e-3
- relaxed layer sep
- ??????????
- relaxed parameter a
- ??????????
- worst energy attained
- ??????????
- approx iterations to relax, for each value of ‘a’ tested
- ??????????????
- relaxed energy
- ????????????
- symmetry
- ???????????
freqs of lowest band at small x near gamma
| THz | wavenumber |
|---|---|
Computation: ./files/do-params/proj/params/abc/1.000-1.000/
- layer sep
- 3.35 A
- parameter a
- 2.46 A
- golden search
- none
- relaxed layer sep
- ??????????
- iterations to relax
- 200 (exceeded limit)
- relaxed energy
- -10582.29 eV
- symmetry
- P-3m1
freqs of lowest band at small x near gamma
| THz | wavenumber |
|---|---|
| -0.0000022343 | -7.4528227e-5 |
| 0.0709829395 | 2.3677360 |
| 0.1421962173 | 4.7431553 |
Did scanning (without relaxing) for approximate ‘a’ and layer sep produce different values from what I’ve been using (YES)
- layer sep
- 3.35 -> 3.37
- parameter a
- 2.46 -> 2.42
Does that approximate ‘a’ suffice for band computation, or does relaxation further change it? (DATA IS TOO COARSE )
(see Google sheet. Automatic relaxation very consistently gets within a much tighter interval than the (0.005 x 2.46 A) steps I computed)
Did scanning (without relaxing) for approximate ‘a’ and layer sep result in faster relaxation? (YES)
Did scanning (without relaxing) for approximate ‘a’ and layer sep resolve the instability of parameter relaxation? (ONLY PARTIALLY; SYMMETRY IS NOW PRESERVED)
Did decreasing the search interval for golden search resolve the issues of parameter relaxation? (NOT EVEN CLOSE)
Does disabling guess-sharing for golden search resolve the issues of parameter relaxation? (TODO - HAVEN’T TRIED)
… although it obviously will since it would be equivalent to manual parameter search. (Note however that guess-sharing cannot be the cause of the issues observed in the rotated data, which did not use golden search.)