Head-to-head

Measured comparisons of d-SEAMS 2.14.0 with other structure codes. The cage census is read out of the agreement text of the cage comparison.

A time in the summary and cage tables is the median of n repeats. The ratio is the other median divided by the d-SEAMS median. In those two tables every stored ratio is greater than 1: the other median is the larger time. Where a table has an agreement column, that column is the stored agreement text through the first semicolon. The cage timing table’s last column is the cage name. The census under it is the rest of the agreement text. Notes about serial calls are in the section prose.

The three comparisons

In the summary and cage tables, the version column is the version field. A slash, when present, separates the d-SEAMS version from the other code. Every summary row records cores 8. A one-thread summary row records cpus 0. An eight-thread summary row records cpus 0-7.

In the SQQ comparison, each timing row records d-SEAMS 2.14.0 and SQQ 0.5.8. Those rows store means. The summary tables store medians. The SQQ ratio is ratio_sqq_over_dseams.

The 884736-molecule comparison of ice and a TIP4P/ice liquid is on Performance. On the TIP4P/ice liquid the table states a freud \(q_6\) gap of 1.525e-04, called hard-cutoff rounding there, and a PLUMED \(q_6\) gap of 1.679e-01, also written 0.168, called the rational switch there. CHILL+ agrees on 830736 of 884736 molecules. The \(q_6\) table on this page is the summary comparison. Its agreement column is the maximum absolute gap stored for that size.

Neighbour lists

vesin 0.6.2 and freud 3.6.1. The agreement text is the matching pair count.

comparator

version

operation

molecules

threads

cores

n

d-SEAMS ms

other ms

ratio

agreement

freud

2.14.0/3.6.1

neighbours

4096

1

8

8

0.2809

2.9465

10.4886

pairs 8192 equal

vesin

2.14.0/0.6.2

neighbours

4096

1

8

8

0.2524

1.7137

6.7902

pairs 8192 equal

freud

2.14.0/3.6.1

neighbours

4096

8

8

8

0.3054

0.8688

2.8444

pairs 8192 equal

vesin

2.14.0/0.6.2

neighbours

4096

8

8

8

0.2509

0.5761

2.2961

pairs 8192 equal

freud

2.14.0/3.6.1

neighbours

110592

1

8

8

6.4788

91.8476

14.1766

pairs 221184 equal

vesin

2.14.0/0.6.2

neighbours

110592

1

8

8

6.4962

45.9557

7.0742

pairs 221184 equal

freud

2.14.0/3.6.1

neighbours

110592

8

8

8

3.8882

31.5433

8.1125

pairs 221184 equal

vesin

2.14.0/0.6.2

neighbours

110592

8

8

8

4.0501

14.9654

3.6950

pairs 221184 equal

freud

2.14.0/3.6.1

neighbours

884736

1

8

8

102.7091

930.3372

9.0580

pairs 1769472 equal

vesin

2.14.0/0.6.2

neighbours

884736

1

8

8

82.4432

1140.5612

13.8345

pairs 1769472 equal

freud

2.14.0/3.6.1

neighbours

884736

8

8

8

49.5928

343.2560

6.9215

pairs 1769472 equal

vesin

2.14.0/0.6.2

neighbours

884736

8

8

8

41.3136

209.6075

5.0736

pairs 1769472 equal

Local q6

mdapy 2.0.0, pyscal3 4.1.0, and PLUMED 2.10.0. For a given code and size the stored gap is the same at one thread and at eight. mdapy and pyscal3 also store the Lechner-Dellago average, q6bar_max_abs, equal to the local gap on these rows. PLUMED stores only q6_max_abs. These gaps are \(10^{-16}\) and \(10^{-15}\). The liquid comparison’s 1.679e-01 PLUMED gap is described above.

comparator

version

operation

molecules

threads

cores

n

d-SEAMS ms

other ms

ratio

agreement

PLUMED

2.14.0/2.10.0

q6

4096

1

8

8

0.7230

121.5863

168.1621

q6_max_abs 2.220e-16

mdapy

2.14.0/2.0.0

q6

4096

1

8

8

0.7256

4.6527

6.4123

q6_max_abs 4.441e-16 q6bar_max_abs 4.441e-16

pyscal3

2.14.0/4.1.0

q6

4096

1

8

8

0.7415

3.5078

4.7306

q6_max_abs 5.551e-16 q6bar_max_abs 5.551e-16

PLUMED

2.14.0/2.10.0

q6

4096

8

8

8

0.6326

109.1420

172.5345

q6_max_abs 2.220e-16

mdapy

2.14.0/2.0.0

q6

4096

8

8

8

0.5273

0.8488

1.6098

q6_max_abs 4.441e-16 q6bar_max_abs 4.441e-16

pyscal3

2.14.0/4.1.0

q6

4096

8

8

8

0.3900

1.9000

4.8712

q6_max_abs 5.551e-16 q6bar_max_abs 5.551e-16

PLUMED

2.14.0/2.10.0

q6

110592

1

8

8

19.2551

3006.8135

156.1569

q6_max_abs 5.551e-16

mdapy

2.14.0/2.0.0

q6

110592

1

8

8

18.7583

108.5550

5.7870

q6_max_abs 1.776e-15 q6bar_max_abs 1.776e-15

pyscal3

2.14.0/4.1.0

q6

110592

1

8

8

18.7787

99.5513

5.3013

q6_max_abs 2.442e-15 q6bar_max_abs 2.442e-15

PLUMED

2.14.0/2.10.0

q6

110592

8

8

8

8.9858

3028.2786

337.0065

q6_max_abs 5.551e-16

mdapy

2.14.0/2.0.0

q6

110592

8

8

8

5.3935

15.2391

2.8254

q6_max_abs 1.776e-15 q6bar_max_abs 1.776e-15

pyscal3

2.14.0/4.1.0

q6

110592

8

8

8

5.3208

18.3291

3.4448

q6_max_abs 2.442e-15 q6bar_max_abs 2.442e-15

PLUMED

2.14.0/2.10.0

q6

884736

1

8

8

240.7382

23208.5792

96.4059

q6_max_abs 6.661e-16

mdapy

2.14.0/2.0.0

q6

884736

1

8

8

237.0502

1039.0684

4.3833

q6_max_abs 2.331e-15 q6bar_max_abs 2.331e-15

pyscal3

2.14.0/4.1.0

q6

884736

1

8

8

237.9653

927.0447

3.8957

q6_max_abs 3.220e-15 q6bar_max_abs 3.220e-15

PLUMED

2.14.0/2.10.0

q6

884736

8

8

8

91.1466

23631.1588

259.2655

q6_max_abs 6.661e-16

mdapy

2.14.0/2.0.0

q6

884736

8

8

8

71.0785

149.8983

2.1089

q6_max_abs 2.331e-15 q6bar_max_abs 2.331e-15

pyscal3

2.14.0/4.1.0

q6

884736

8

8

8

71.7527

177.4625

2.4733

q6_max_abs 3.220e-15 q6bar_max_abs 3.220e-15

CHILL+

mdapy 2.0.0 matches the label on every molecule and stores the same q6_max_abs as its q6 rows. ClavisCage 0.3.18 matches the label on every molecule. The agreement text on the ClavisCage rows states that this CHILL+ path is a serial loop. The table is the measured time: the eight-thread ClavisCage median stays with the one-thread median at each size.

pyscal3 4.1.0 has no finished CHILL+ row. The agreement text says that version exports no CHILL or CHILL+ name, that common_neighbor_analysis is adaptive CNA and returns fcc, hcp, bcc, ico, and other, and that diamond_structure is extended CNA. The same text says neither function is the Nguyen and Molinero CHILL+ rule, and that PyPI’s latest pyscal3 is this same 4.1.0. The rows are recorded at one thread and at eight, with an empty molecule count.

comparator

version

operation

molecules

threads

cores

n

d-SEAMS ms

other ms

ratio

agreement

ClavisCage

2.14.0/0.3.18

chill_plus

4096

1

8

8

0.7255

4.5952

6.3335

labels 4096 equal

mdapy

2.14.0/2.0.0

chill_plus

4096

1

8

8

0.7184

2.6670

3.7122

labels 4096 equal q6_max_abs 4.441e-16

ClavisCage

2.14.0/0.3.18

chill_plus

4096

8

8

8

0.4711

4.5363

9.6285

labels 4096 equal

mdapy

2.14.0/2.0.0

chill_plus

4096

8

8

8

0.5076

0.5913

1.1649

labels 4096 equal q6_max_abs 4.441e-16

ClavisCage

2.14.0/0.3.18

chill_plus

110592

1

8

8

18.4099

113.8680

6.1851

labels 110592 equal

mdapy

2.14.0/2.0.0

chill_plus

110592

1

8

8

18.5351

56.9564

3.0729

labels 110592 equal q6_max_abs 1.776e-15

ClavisCage

2.14.0/0.3.18

chill_plus

110592

8

8

8

5.2050

113.8816

21.8794

labels 110592 equal

mdapy

2.14.0/2.0.0

chill_plus

110592

8

8

8

7.5892

10.3904

1.3691

labels 110592 equal q6_max_abs 1.776e-15

ClavisCage

2.14.0/0.3.18

chill_plus

884736

1

8

8

208.1052

1007.4027

4.8408

labels 884736 equal

mdapy

2.14.0/2.0.0

chill_plus

884736

1

8

8

203.1011

585.3079

2.8819

labels 884736 equal q6_max_abs 2.331e-15

ClavisCage

2.14.0/0.3.18

chill_plus

884736

8

8

8

59.4890

1008.0481

16.9451

labels 884736 equal

mdapy

2.14.0/2.0.0

chill_plus

884736

8

8

8

66.4798

115.4407

1.7365

labels 884736 equal q6_max_abs 2.331e-15

Diamond structure

OVITO 3.17.2 identify_diamond. At each size the agreement text is hexagonal followed by the molecule count and equal: every molecule is hexagonal on both sides.

comparator

version

operation

molecules

threads

cores

n

d-SEAMS ms

other ms

ratio

agreement

OVITO

2.14.0/3.17.2

identify_diamond

4096

1

8

8

0.7306

4.7155

6.4542

hexagonal 4096 equal

OVITO

2.14.0/3.17.2

identify_diamond

4096

8

8

8

0.5566

1.9548

3.5122

hexagonal 4096 equal

OVITO

2.14.0/3.17.2

identify_diamond

110592

1

8

8

20.7730

136.5008

6.5711

hexagonal 110592 equal

OVITO

2.14.0/3.17.2

identify_diamond

110592

8

8

8

8.3309

29.5160

3.5430

hexagonal 110592 equal

OVITO

2.14.0/3.17.2

identify_diamond

884736

1

8

8

248.1369

1136.5466

4.5803

hexagonal 884736 equal

OVITO

2.14.0/3.17.2

identify_diamond

884736

8

8

8

62.1147

262.3612

4.2238

hexagonal 884736 equal

Rings

OVITO 3.17.2 FindRings, rings-statistics-matter 0.3.0, and cycless 0.7, on the rows that finished. The agreement text is the matching hexagon count: 8192, 221184, and 1769472 at 4096, 110592, and 884736 molecules.

The cycless rows state that cycles_iter has no thread argument. At 884736 molecules the cycless rows did not finish. The version field on those rows is 0.7, one sample exceeded 600 s, and the agreement text stores a fit labelled “Fit, not a measurement.” There is no measured time for that size.

freud has no ring row. The agreement text names freud 3.6.1, lists its modules, and states that there is no ring or cycle module. The citation in that text is Ramasubramani and others, Comput. Phys. Commun. 254, 107275 (2020). The version field on those rows is 2.14.0. The sizes are 4096, 110592, and 884736, at one thread and at eight.

comparator

version

operation

molecules

threads

cores

n

d-SEAMS ms

other ms

ratio

agreement

OVITO

2.14.0/3.17.2

rings

4096

1

8

8

4.5301

15.8206

3.4923

hexagons 8192 equal

cycless

2.14.0/0.7

rings

4096

1

8

8

5.0379

12678.2659

2516.5770

hexagons 8192 equal

rings-statistics-matter

2.14.0/0.3.0

rings

4096

1

8

8

4.5552

56.9215

12.4961

hexagons 8192 equal

OVITO

2.14.0/3.17.2

rings

4096

8

8

8

2.6129

17.6346

6.7491

hexagons 8192 equal

cycless

2.14.0/0.7

rings

4096

8

8

8

2.9191

12516.1838

4287.6594

hexagons 8192 equal

rings-statistics-matter

2.14.0/0.3.0

rings

4096

8

8

8

2.2476

12.5984

5.6052

hexagons 8192 equal

OVITO

2.14.0/3.17.2

rings

110592

1

8

8

119.4383

503.1571

4.2127

hexagons 221184 equal

cycless

2.14.0/0.7

rings

110592

1

8

8

136.3856

341558.2557

2504.3573

hexagons 221184 equal

rings-statistics-matter

2.14.0/0.3.0

rings

110592

1

8

8

116.2389

1327.4966

11.4204

hexagons 221184 equal

OVITO

2.14.0/3.17.2

rings

110592

8

8

8

48.3409

323.8505

6.6993

hexagons 221184 equal

cycless

2.14.0/0.7

rings

110592

8

8

8

65.8726

337108.4778

5117.5843

hexagons 221184 equal

rings-statistics-matter

2.14.0/0.3.0

rings

110592

8

8

8

47.2318

307.1093

6.5022

hexagons 221184 equal

OVITO

2.14.0/3.17.2

rings

884736

1

8

8

1046.5149

4542.0060

4.3401

hexagons 1769472 equal

rings-statistics-matter

2.14.0/0.3.0

rings

884736

1

8

8

1032.6158

13012.7347

12.6017

hexagons 1769472 equal

OVITO

2.14.0/3.17.2

rings

884736

8

8

8

484.2037

2917.3354

6.0250

hexagons 1769472 equal

rings-statistics-matter

2.14.0/0.3.0

rings

884736

8

8

8

476.9637

2967.7415

6.2222

hexagons 1769472 equal

FindRings and the primitive census

The hexagon counts above are one check. The subset result is another.

On K2,3, d-SEAMS reports three primitive 4-rings and OVITO FindRings reports two, with a 3.5 Å cutoff giving six edges. Primitive rings and OVITO FindRings lists the three cycles.

On the 864-atom frame input/traj/interface_f0.lammpstrj, cutoff 3.5 Å, the graph has 1929 bonds and this primitive census: 142 of size 3, 188 of size 4, 377 of size 5, 1143 of size 6, 257 of size 7, 54 of size 8, 52 of size 9, and 449 of size 10. Every one of those rings passes the Franzblau test, and the sum of those bins is 2662. On the same trajectory, the bond count matches on both sides, every listed d-SEAMS ring is primitive, and each FindRings bin from size 4 through 10 is at most the primitive bin.

Polyhedral template matching

OVITO 3.17.2 PTM and PTM. Both label every molecule hexagonal. The PTM rows also store rmsd_max 4.337e-08 at every size and both thread counts.

comparator

version

operation

molecules

threads

cores

n

d-SEAMS ms

other ms

ratio

agreement

OVITO

2.14.0/3.17.2

ptm

4096

1

8

8

0.7264

105.0304

144.5805

hexagonal 4096 equal

PTM

2.14.0

ptm

4096

1

8

8

0.7334

189.4989

258.3907

hexagonal 4096 equal rmsd_max 4.337e-08

OVITO

2.14.0/3.17.2

ptm

4096

8

8

8

0.5204

28.3374

54.4549

hexagonal 4096 equal

PTM

2.14.0

ptm

4096

8

8

8

0.2783

24.0122

86.2705

hexagonal 4096 equal rmsd_max 4.337e-08

OVITO

2.14.0/3.17.2

ptm

110592

1

8

8

21.0405

2903.9205

138.0160

hexagonal 110592 equal

PTM

2.14.0

ptm

110592

1

8

8

21.8284

4855.0064

222.4168

hexagonal 110592 equal rmsd_max 4.337e-08

OVITO

2.14.0/3.17.2

ptm

110592

8

8

8

8.5196

387.5268

45.4867

hexagonal 110592 equal

PTM

2.14.0

ptm

110592

8

8

8

8.2662

613.5692

74.2262

hexagonal 110592 equal rmsd_max 4.337e-08

OVITO

2.14.0/3.17.2

ptm

884736

1

8

8

259.8032

23247.5896

89.4815

hexagonal 884736 equal

PTM

2.14.0

ptm

884736

1

8

8

255.8396

38381.2306

150.0207

hexagonal 884736 equal rmsd_max 4.337e-08

OVITO

2.14.0/3.17.2

ptm

884736

8

8

8

63.6214

3062.3280

48.1336

hexagonal 884736 equal

PTM

2.14.0

ptm

884736

8

8

8

59.6748

4823.9120

80.8367

hexagonal 884736 equal rmsd_max 4.337e-08

Shape matching

IRA 2.2.0 on 4096 molecules. The agreement text is hexagonal 4096 equal at one thread and at eight. The version field is 2.14.0/2.2.0.

At 110592 molecules one sample exceeded 600 s. The version field on that row, and on the 884736-molecule row, is 2.2.0. The 884736-molecule row was not started because 110592 molecules exceeded 600 s. Both unfinished rows store a fit labelled “Fit, not a measurement” and state that classify_serial has no thread argument.

comparator

version

operation

molecules

threads

cores

n

d-SEAMS ms

other ms

ratio

agreement

IRA

2.14.0/2.2.0

shape

4096

1

8

8

0.7481

1302.6695

1741.2294

hexagonal 4096 equal

IRA

2.14.0/2.2.0

shape

4096

8

8

8

1.2898

1299.2780

1007.3679

hexagonal 4096 equal

Adaptive common-neighbour analysis

Three OVITO rows, operation cna, version field 2.14.0, 4096 molecules, one thread, cores 8, n 1. The version field does not name an OVITO release. There is no stored time. The agreement text says the modifier’s structure types are Other, FCC, HCP, BCC, and ICO, and that an FCC control is 256 of 256 FCC.

Ice Ih: 4096 of 4096 molecules labelled hexagonal_ice by CHILL+ are Other. Coordination 4, 3 staggered and 1 eclipsed, on all 4096.

Ice Ic: 4096 of 4096 molecules labelled cubic_ice are Other. Coordination 4, 4 staggered and 0 eclipsed, on all 4096.

Liquid: 222 of 4096 disagree. One hexagonal_ice molecule is Other at coordination 4, with 3 staggered and 1 eclipsed. Forty-nine hydrate molecules are Other at coordination 4, with 4 eclipsed and 0 staggered. One hundred sixty-eight interfacial_hydrate molecules are Other at coordination 4, with 3 eclipsed. Four interfacial_ice molecules are Other at coordination 4, with 2 staggered, or 3 staggered and 0 eclipsed.

The same liquid text names four shifted-ice waters with 12 neighbours inside 3.5 Å: 690 at (4.490, 15.864, 50.345), 1673 at (16.865, 18.399, 11.795), 1675 at (15.210, 17.036, 12.464), and 2304 at (16.902, 34.179, 1.355). An independent minimum-image count in that text gives the same 12 ids. The nearest pair is 0.183 Å, from the Gaussian shift.

ClavisCage TRACE cages

ClavisCage 0.3.18 against the d-SEAMS cage census. The one-thread and eight-thread rows of a cell store the same census. The record writes 512, 51262, and 51264 for the \(5^{12}\), \(5^{12}6^{2}\), and \(5^{12}6^{4}\) cages. both is the count found by both codes. The other two columns are cages found by only one code. TRACE-only is 0 on every signature below.

On the 46-molecule sI cell, cage 512-0 has 18 of 30 bonds across the periodic cell, and the six 51262 cages have 10 to 15. The \(5^{12}\) cage TRACE returns has none. The 2x2x2 tile matches. The record’s conclusion on that cell is that d-SEAMS is right.

On the 110160-molecule rows the agreement text states that TRACE is serial and that analyze_waters has no thread argument. The TRACE median is 61962.4841 ms at one thread and 61956.9697 ms at eight. The d-SEAMS median is 955.8314 ms at one thread and 303.0753 ms at eight.

ClavisCage TRACE is the same census.

comparator

version

operation

molecules

threads

cores

n

d-SEAMS ms

other ms

ratio

cage

ClavisCage

2.14.0/0.3.18

cages

46

1

8

8

0.5796

50.3899

86.9321

sI

ClavisCage

2.14.0/0.3.18

cages

46

8

8

8

0.5947

50.2893

84.5632

sI

ClavisCage

2.14.0/0.3.18

cages

136

1

8

8

1.4924

68.3007

45.7666

sII

ClavisCage

2.14.0/0.3.18

cages

136

8

8

8

1.5655

68.2571

43.6016

sII

ClavisCage

2.14.0/0.3.18

cages

368

1

8

8

3.7222

297.6070

79.9538

sI-222

ClavisCage

2.14.0/0.3.18

cages

368

8

8

8

2.9243

297.9639

101.8907

sI-222

ClavisCage

2.14.0/0.3.18

cages

1088

1

8

8

10.7150

697.6882

65.1132

sII-222

ClavisCage

2.14.0/0.3.18

cages

1088

8

8

8

4.7284

697.7931

147.5752

sII-222

ClavisCage

2.14.0/0.3.18

cages

110160

1

8

8

955.8314

61962.4841

64.8257

sII-9x9x10

ClavisCage

2.14.0/0.3.18

cages

110160

8

8

8

303.0753

61956.9697

204.4277

sII-9x9x10

Census parsed from the agreement text of those rows. The eight-thread text matches the one-thread text on each cell.

cell

molecules

signature

both

d-SEAMS only

TRACE only

sI

46

512

1

1

0

sI

46

51262

0

6

0

sI

46

51264

0

0

0

sII

136

512

16

0

0

sII

136

51262

0

0

0

sII

136

51264

8

0

0

sI-222

368

512

16

0

0

sI-222

368

51262

48

0

0

sI-222

368

51264

0

0

0

sII-222

1088

512

128

0

0

sII-222

1088

51262

0

0

0

sII-222

1088

51264

64

0

0

sII-9x9x10

110160

512

12960

0

0

sII-9x9x10

110160

51262

0

0

0

sII-9x9x10

110160

51264

6480

0

0

SQQ

SQQ cage rows in the summary file did not run. The version field is 0.5.8. At 4096 molecules the agreement text says 512 molecules exceeded 600 s, so that size was not started. At 110592 and at 884736 molecules it says 4096 molecules exceeded 600 s, so that size was not started. The rows are recorded at one thread and at eight. There is no cage time.

The ring and ice comparison is the SQQ file. Frames in that file are Ic-4096, Ih-4096, mw-liquid-4096, tip4p2005-4096, Ic-110592, Ih-110592, mw-liquid-110592, and tip4p-o-250. Operations are rings, ice, ice-primitive, and ice-chordless. The tables here are the 4096-molecule TIP4P/2005 frame. The other frames are in the CSV. SQQ prints the full set.

On tip4p2005-4096 the edge and ring counts match. The record’s resolution on the chordless 6-ring row says every primitive 6-ring is inside that count, and that the reference enumeration is 4045.

The merged ice-label row stores d-SEAMS 4, SQQ 1129, both 2969, agreement 0.724854. Its resolution text says CHILL+ reproduced 4096/4096 by scipy, the SQQ rule reproduced 4096/4096 from the reference graph, and the mismatches are 1127.

The primitive-rule row stores d-SEAMS 1075, SQQ 1129, both 4042, agreement 0.986816. Its resolution text says the rule reproduced 4096/4096 from the reference graph, and that each mismatch has a primitive and a chordless six-ring count that differ on the molecule or a neighbour: 54/54.

The chordless-rule row stores d-SEAMS 1129, SQQ 1129, both 4096, agreement 1.000000. Its resolution text says the mismatches on or next to an edge SQQ lacks are 0/0.

frame

molecules

quantity

d-SEAMS

SQQ

both

agreement

tip4p2005-4096

4096

O-O edges

10495

10495

10495

1.000000

tip4p2005-4096

4096

4-rings

1758

1758

1758

1.000000

tip4p2005-4096

4096

5-rings

2763

2763

2763

1.000000

tip4p2005-4096

4096

6-rings

2485

2485

2485

1.000000

tip4p2005-4096

4096

chordless 6-rings

4045

4045

4045

1.000000

tip4p2005-4096

4096

ice label (merged)

4

1129

2969

0.724854

tip4p2005-4096

4096

ice label, SQQ rule on d-SEAMS primitive 6-rings

1075

1129

4042

0.986816

tip4p2005-4096

4096

ice label, SQQ rule on d-SEAMS chordless 6-rings

1129

1129

4096

1.000000

The timing notes in the CSV, for this frame:

  • rings is d-SEAMS primitive rings of size 3 to 6 against SQQ shortest-path rings of size 4 to 6

  • ice is whole tools: d-SEAMS CHILL+ with no rings, against SQQ ice on chordless 6-rings

  • ice-primitive is the SQQ rule on d-SEAMS primitive 6-rings, against SQQ ice

  • ice-chordless is the SQQ rule on chordless 6-cycles of the d-SEAMS graph, against SQQ ice

The ice, ice-primitive, and ice-chordless notes say the SQQ samples are shared, which is why the SQQ mean matches across those three operations. One thread records cores 1 and cpus 0. Eight threads record cores 8 and cpus 0-7. n is 8. The ratio is the SQQ mean divided by the d-SEAMS mean.

frame

molecules

operation

threads

cores

n

d-SEAMS ms

SQQ ms

SQQ/d-SEAMS

agreement

tip4p2005-4096

4096

rings

1

1

8

22.6026

1054.3846

46.649

1.000000

tip4p2005-4096

4096

rings

8

8

8

8.2165

1056.0863

128.532

1.000000

tip4p2005-4096

4096

ice

1

1

8

1.7860

914.2565

511.905

0.724854

tip4p2005-4096

4096

ice

8

8

8

1.5802

916.2551

579.849

0.724854

tip4p2005-4096

4096

ice-primitive

1

1

8

23.3003

914.2565

39.238

0.986816

tip4p2005-4096

4096

ice-primitive

8

8

8

8.4130

916.2551

108.910

0.986816

tip4p2005-4096

4096

ice-chordless

1

1

8

28.5570

914.2565

32.015

1.000000

tip4p2005-4096

4096

ice-chordless

8

8

8

7.9694

916.2551

114.971

1.000000

Performance also prints an SQQ ring median: 18234.926 ms at one thread and 17907.744 ms at eight threads, 110592 molecules, n 6. That line is SQQ’s time alone.

Per-atom q4 and q6

Per-atom \(q_4\) and \(q_6\), the Lechner-Dellago averages, and the mean of each, are compared on three frames. Ideal ice Ih is input/traj/ice1h.lammpstrj. Ideal ice Ic is a diamond cell with a 2.75 Å bond, four cells on a side. The liquid is that cell plus Gaussian noise of 0.9 Å, seed 91001. The neighbour rule is the same in d-SEAMS, freud, and pyscal3: a pair counts when its minimum-image distance is strictly below 3.5 Å. The comparison runs at one thread and at eight.

pyscal3 is float64. Local \(q_l\), the Lechner-Dellago average, and the mean of each over atoms that have a shell agree to \(10^{-10}\). freud stores \(q_l\) as float32. On the two crystals the gap stays within 8 units in the last place of that float32 value. On the liquid the spherical-harmonic sum is accumulated in float32, and the residual stays under \(2 \times 10^{-5}\).

An empty shell is 0 in d-SEAMS and NaN in freud and pyscal3. 0 is the empty accumulator. Local q4 and q6 records why that 0 stays: a NaN would enter every mean and every CHILL+ label.

OVITO 3.15.5 has no SteinhardtParametersModifier. OVITO 3.17.2 \(q_4\) and \(q_6\) are timed in the 884736-molecule comparison on Performance.

PLUMED on the ice Ih dump

On input/traj/ice1h.lammpstrj, the PLUMED engine reports hexagonal 128, six-rings 256, and \(q_6\) within \(5 \times 10^{-7}\) of 0.625855. PLUMED is the engine page. The standalone PLUMED action (dseams-plumed v0.4.1) is indexed in PLUMED-NEST egg 26.018.

That check is the ice Ih dump. The summary \(q_6\) table is the median comparison with PLUMED 2.10.0, with maximum absolute gaps from \(2.220 \times 10^{-16}\) to \(6.661 \times 10^{-16}\). The liquid comparison is the 1.679e-01 rational-switch gap.