1 |
#! /usr/bin/env python |
2 |
|
3 |
"""Dipolar Lattice Builder |
4 |
|
5 |
Creates cubic lattices of dipoles to test the dipole-dipole |
6 |
interaction code. |
7 |
|
8 |
Usage: buildDipolarArray |
9 |
|
10 |
Options: |
11 |
-h, --help show this help |
12 |
-x, --array-type-X use one of the basic "X" arrays |
13 |
-y, --array-type-Y use one of the basic "Y" arrays |
14 |
-z, --array-type-Z use one of the basic "Z" arrays |
15 |
-a, --array-type-A use array type "A" (default) |
16 |
-b, --array-type-B use array type "B" |
17 |
-l, --lattice=... use the specified lattice ( SC, FCC, or BCC ) |
18 |
-d, --direction=... use dipole orientation (001, 111, or 011) |
19 |
-c, --constant=... use the specified lattice constant |
20 |
-n use the specified number of unit cells |
21 |
-o, --output-file=... use specified output (.xyz) file |
22 |
|
23 |
Type "A" arrays have nearest neighbor strings of antiparallel dipoles. |
24 |
|
25 |
Type "B" arrays have nearest neighbor strings of antiparallel dipoles |
26 |
if the dipoles are contained in a plane perpendicular to the dipole |
27 |
direction that passes through the dipole. |
28 |
|
29 |
Example: |
30 |
buildDipolarArray -a -l fcc -d 001 -c 5 -n 3 -o A_fcc_001.xyz |
31 |
|
32 |
""" |
33 |
|
34 |
__author__ = "Dan Gezelter (gezelter@nd.edu)" |
35 |
__version__ = "$Revision: 1639 $" |
36 |
__date__ = "$Date: 2011-09-24 16:18:07 -0400 (Sat, 24 Sep 2011) $" |
37 |
|
38 |
__copyright__ = "Copyright (c) 2013 by the University of Notre Dame" |
39 |
__license__ = "OpenMD" |
40 |
|
41 |
import sys |
42 |
import getopt |
43 |
import string |
44 |
import math |
45 |
import numpy |
46 |
|
47 |
def usage(): |
48 |
print __doc__ |
49 |
|
50 |
|
51 |
|
52 |
def createLattice(latticeType, latticeNumber, latticeConstant, dipoleDirection, arrayType, outputFileName): |
53 |
# The following section creates 24 basic arrays from Luttinger and |
54 |
# Tisza: |
55 |
|
56 |
# The six unit vectors are: 3 spatial and 3 to describe the |
57 |
# orientation of the dipole. |
58 |
|
59 |
e1 = numpy.array([1.0,0.0,0.0,0.0,0.0,0.0]) |
60 |
e2 = numpy.array([0.0,1.0,0.0,0.0,0.0,0.0]) |
61 |
e3 = numpy.array([0.0,0.0,1.0,0.0,0.0,0.0]) |
62 |
e4 = numpy.array([0.0,0.0,0.0,1.0,0.0,0.0]) |
63 |
e5 = numpy.array([0.0,0.0,0.0,0.0,1.0,0.0]) |
64 |
e6 = numpy.array([0.0,0.0,0.0,0.0,0.0,1.0]) |
65 |
|
66 |
# Parameters describing the 8 basic arrays: |
67 |
cell = numpy.array([[0,0,0],[0,0,1],[1,0,0],[0,1,0], |
68 |
[1,1,0],[0,1,1],[1,0,1],[1,1,1]]) |
69 |
|
70 |
X = numpy.zeros(384).reshape((8,8,6)) |
71 |
Y = numpy.zeros(384).reshape((8,8,6)) |
72 |
Z = numpy.zeros(384).reshape((8,8,6)) |
73 |
# mX, mY, and mZ arrays have dipole direction flipped |
74 |
mX = numpy.zeros(384).reshape((8,8,6)) |
75 |
mY = numpy.zeros(384).reshape((8,8,6)) |
76 |
mZ = numpy.zeros(384).reshape((8,8,6)) |
77 |
|
78 |
# create the 24 basic arrays using Eq. 12 in Luttinger & Tisza: |
79 |
for i in range(8): |
80 |
which = 0 |
81 |
for l1 in range(2): |
82 |
for l2 in range(2): |
83 |
for l3 in range(2): |
84 |
lvals = numpy.array([l1,l2,l3]) |
85 |
value = math.pow(-1, numpy.dot(cell[i], lvals)) |
86 |
Xvec = (l1*e1 + l2*e2 + l3*e3) + value * e4 |
87 |
Yvec = (l1*e1 + l2*e2 + l3*e3) + value * e5 |
88 |
Zvec = (l1*e1 + l2*e2 + l3*e3) + value * e6 |
89 |
mXvec = (l1*e1 + l2*e2 + l3*e3) - value * e4 |
90 |
mYvec = (l1*e1 + l2*e2 + l3*e3) - value * e5 |
91 |
mZvec = (l1*e1 + l2*e2 + l3*e3) - value * e6 |
92 |
X[i][which] = Xvec |
93 |
Y[i][which] = Yvec |
94 |
Z[i][which] = Zvec |
95 |
mX[i][which] = mXvec |
96 |
mY[i][which] = mYvec |
97 |
mZ[i][which] = mZvec |
98 |
which = which + 1 |
99 |
|
100 |
# The simple cubic array has only one site at the lattice point: |
101 |
lp = numpy.array([0.0,0.0,0.0,0.0,0.0,0.0]) |
102 |
|
103 |
# The body-centered cubic array also has a body-centerered site: |
104 |
bc = numpy.array([0.5,0.5,0.5,0.0,0.0,0.0]) |
105 |
|
106 |
# The face-centered cubic array also has 3 face-centered sites: |
107 |
xy = numpy.array([0.5,0.5,0.0,0.0,0.0,0.0]) |
108 |
yz = numpy.array([0.0,0.5,0.5,0.0,0.0,0.0]) |
109 |
xz = numpy.array([0.5,0.0,0.5,0.0,0.0,0.0]) |
110 |
|
111 |
sc = numpy.array([[0.0,0.0,0.0]]) |
112 |
|
113 |
bcc = numpy.array([[0.0,0.0,0.0], |
114 |
[0.5,0.5,0.5]]) |
115 |
|
116 |
fcc = numpy.array([[0.0,0.0,0.0], |
117 |
[0.5,0.5,0.0], |
118 |
[0.0,0.5,0.5], |
119 |
[0.5,0.0,0.5]]) |
120 |
|
121 |
known_case = False |
122 |
|
123 |
if (arrayType == 'X'): |
124 |
if (int(latticeType)): |
125 |
which = int(latticeType) - 1 |
126 |
basic_array = X[which] |
127 |
known_case = True |
128 |
if (arrayType == 'Y'): |
129 |
if (int(latticeType)): |
130 |
which = int(latticeType) - 1 |
131 |
basic_array = Y[which] |
132 |
known_case = True |
133 |
if (arrayType == 'Z'): |
134 |
if (int(latticeType)): |
135 |
which = int(latticeType) - 1 |
136 |
basic_array = Z[which] |
137 |
known_case = True |
138 |
if (arrayType == 'A'): |
139 |
if (latticeType.lower() == 'sc'): |
140 |
basic_array = Z[5]+lp |
141 |
known_case = True |
142 |
if (latticeType.lower() == 'bcc'): |
143 |
if (dipoleDirection.lower() == '001'): |
144 |
basic_array = numpy.append(Z[1]+lp, mZ[1]+bc, axis=0) |
145 |
known_case = True |
146 |
if (dipoleDirection.lower() == '111'): |
147 |
basic_array = numpy.append(Z[5]+X[7]+Y[6]+lp, |
148 |
Z[5]+X[7]+Y[6]+bc, axis=0) |
149 |
known_case = True |
150 |
if (latticeType.lower() == 'fcc'): |
151 |
if (dipoleDirection.lower() == '001'): |
152 |
basic_array = numpy.append(Z[1]+lp, Z[1]+xy, axis=0) |
153 |
basic_array = numpy.append(basic_array, mZ[1]+yz, axis=0) |
154 |
basic_array = numpy.append(basic_array, mZ[1]+xz, axis=0) |
155 |
known_case = True |
156 |
if (dipoleDirection.lower() == '011'): |
157 |
basic_array = numpy.append(Z[1]+Y[1]+lp, Z[1]+Y[1]+yz, axis=0) |
158 |
basic_array = numpy.append(basic_array, mZ[1]+mY[1]+xy, axis=0) |
159 |
basic_array = numpy.append(basic_array, mZ[1]+mY[1]+xz, axis=0) |
160 |
known_case = True |
161 |
else: |
162 |
if (latticeType.lower() == 'sc'): |
163 |
basic_array = Z[5]+lp |
164 |
known_case = True |
165 |
if (latticeType.lower() == 'bcc'): |
166 |
if (dipoleDirection.lower() == '001'): |
167 |
basic_array = numpy.append(Z[5]+lp, mZ[5]+bc, axis=0) |
168 |
known_case = True |
169 |
if (dipoleDirection.lower() == '111'): |
170 |
basic_array = numpy.append(Z[5]+X[7]+Y[6]+lp, |
171 |
Z[5]+X[7]+Y[6]+bc, axis=0) |
172 |
known_case = True |
173 |
if (latticeType.lower() == 'fcc'): |
174 |
if (dipoleDirection.lower() == '001'): |
175 |
basic_array = numpy.append(Z[1]+lp, Z[1]+yz, axis=0) |
176 |
basic_array = numpy.append(basic_array, mZ[1]+xy, axis=0) |
177 |
basic_array = numpy.append(basic_array, mZ[1]+xz, axis=0) |
178 |
known_case = True |
179 |
if (dipoleDirection.lower() == '011'): |
180 |
basic_array = numpy.append(Z[8]+Y[8]+lp, Z[8]+Y[8]+xy, axis=0) |
181 |
basic_array = numpy.append(basic_array, Z[8]+Y[8]+yz, axis=0) |
182 |
basic_array = numpy.append(basic_array, mZ[8]+mY[8]+xz, axis=0) |
183 |
known_case = True |
184 |
|
185 |
if (not known_case): |
186 |
print "unhandled combination of lattice and dipole direction" |
187 |
print __doc__ |
188 |
|
189 |
print basic_array |
190 |
|
191 |
bravais_lattice = [] |
192 |
for i in range(latticeNumber): |
193 |
for j in range(latticeNumber): |
194 |
for k in range(latticeNumber): |
195 |
for l in range(len(basic_array)): |
196 |
bravais_lattice.append(2*i*e1 + 2*j*e2 + 2*k*e3 + basic_array[l]) |
197 |
|
198 |
outputFile = open(outputFileName, 'w') |
199 |
|
200 |
outputFile.write('<OpenMD version=2>\n') |
201 |
outputFile.write(' <MetaData>\n') |
202 |
outputFile.write(' molecule{\n') |
203 |
outputFile.write(' name = \"D\";\n') |
204 |
outputFile.write(' \n') |
205 |
outputFile.write(' atom[0]{\n') |
206 |
outputFile.write(' type = \"D\";\n') |
207 |
outputFile.write(' position(0.0, 0.0, 0.0);\n') |
208 |
outputFile.write(' orientation(0.0, 0.0, 0.0);\n') |
209 |
outputFile.write(' }\n') |
210 |
outputFile.write(' }\n') |
211 |
outputFile.write(' component{\n') |
212 |
outputFile.write(' type = \"D\";\n') |
213 |
outputFile.write(' nMol = '+ repr(len(bravais_lattice)) + ';\n') |
214 |
outputFile.write(' }\n') |
215 |
|
216 |
outputFile.write(' ensemble = NVE;\n') |
217 |
outputFile.write(' forceField = \"Multipole\";\n') |
218 |
|
219 |
outputFile.write(' cutoffMethod = \"shifted_force\";\n') |
220 |
outputFile.write(' electrostaticScreeningMethod = \"damped\";\n') |
221 |
|
222 |
outputFile.write(' cutoffRadius = 9.0;\n') |
223 |
outputFile.write(' dampingAlpha = 0.18;\n') |
224 |
outputFile.write(' statFileFormat = \"TIME|TOTAL_ENERGY|POTENTIAL_ENERGY|KINETIC_ENERGY|TEMPERATURE|PRESSURE|VOLUME|CONSERVED_QUANTITY|ELECTROSTATIC_POTENTIAL\";\n') |
225 |
outputFile.write(' dt = 1.0;\n') |
226 |
outputFile.write(' runTime = 1.0;\n') |
227 |
outputFile.write(' sampleTime = 1.0;\n') |
228 |
outputFile.write(' statusTime = 1.0;\n') |
229 |
outputFile.write(' </MetaData>\n') |
230 |
outputFile.write(' <Snapshot>\n') |
231 |
outputFile.write(' <FrameData>\n'); |
232 |
outputFile.write(" Time: %.10g\n" % (0.0)) |
233 |
|
234 |
Hxx = 2.0 * latticeConstant * latticeNumber |
235 |
Hyy = 2.0 * latticeConstant * latticeNumber |
236 |
Hzz = 2.0 * latticeConstant * latticeNumber |
237 |
|
238 |
outputFile.write(' Hmat: {{%d, 0, 0}, {0, %d, 0}, {0, 0, %d}}\n' % (Hxx, Hyy, Hzz)) |
239 |
outputFile.write(' </FrameData>\n') |
240 |
outputFile.write(' <StuntDoubles>\n') |
241 |
sdFormat = 'pvqj' |
242 |
index = 0 |
243 |
|
244 |
for i in range(len(bravais_lattice)): |
245 |
xcart = latticeConstant*(bravais_lattice[i][0]) |
246 |
ycart = latticeConstant*(bravais_lattice[i][1]) |
247 |
zcart = latticeConstant*(bravais_lattice[i][2]) |
248 |
dx = bravais_lattice[i][3] |
249 |
dy = bravais_lattice[i][4] |
250 |
dz = bravais_lattice[i][5] |
251 |
|
252 |
uz = numpy.array([dx, dy, dz]) |
253 |
uz = uz/numpy.linalg.norm(uz) |
254 |
|
255 |
uy = numpy.array([0.0, 1.0, 0.0]) |
256 |
uy = uy - uz * numpy.vdot(uy, uz) / numpy.vdot(uz, uz) |
257 |
uy = uy/numpy.linalg.norm(uy) |
258 |
|
259 |
ux = numpy.cross(uy, uz) |
260 |
|
261 |
RotMat = [ux, uy, uz] |
262 |
|
263 |
q = [0.0, 0.0, 0.0, 0.0] |
264 |
|
265 |
# RotMat to Quat code is out of OpenMD's SquareMatrix3.hpp code: |
266 |
|
267 |
t = RotMat[0][0] + RotMat[1][1] + RotMat[2][2] + 1.0 |
268 |
|
269 |
if( t > 1e-6 ): |
270 |
s = 0.5 / math.sqrt( t ) |
271 |
q[0] = 0.25 / s |
272 |
q[1] = (RotMat[1][2] - RotMat[2][1]) * s |
273 |
q[2] = (RotMat[2][0] - RotMat[0][2]) * s |
274 |
q[3] = (RotMat[0][1] - RotMat[1][0]) * s |
275 |
else: |
276 |
ad1 = RotMat[0][0] |
277 |
ad2 = RotMat[1][1] |
278 |
ad3 = RotMat[2][2] |
279 |
|
280 |
if( ad1 >= ad2 and ad1 >= ad3 ): |
281 |
s = 0.5 / math.sqrt( 1.0 + RotMat[0][0] - RotMat[1][1] - RotMat[2][2] ) |
282 |
q[0] = (RotMat[1][2] - RotMat[2][1]) * s |
283 |
q[1] = 0.25 / s |
284 |
q[2] = (RotMat[0][1] + RotMat[1][0]) * s |
285 |
q[3] = (RotMat[0][2] + RotMat[2][0]) * s |
286 |
elif ( ad2 >= ad1 and ad2 >= ad3 ): |
287 |
s = 0.5 / math.sqrt( 1.0 + RotMat[1][1] - RotMat[0][0] - RotMat[2][2] ) |
288 |
q[0] = (RotMat[2][0] - RotMat[0][2] ) * s |
289 |
q[1] = (RotMat[0][1] + RotMat[1][0]) * s |
290 |
q[2] = 0.25 / s |
291 |
q[3] = (RotMat[1][2] + RotMat[2][1]) * s |
292 |
else: |
293 |
s = 0.5 / math.sqrt( 1.0 + RotMat[2][2] - RotMat[0][0] - RotMat[1][1] ) |
294 |
q[0] = (RotMat[0][1] - RotMat[1][0]) * s |
295 |
q[1] = (RotMat[0][2] + RotMat[2][0]) * s |
296 |
q[2] = (RotMat[1][2] + RotMat[2][1]) * s |
297 |
q[3] = 0.25 / s |
298 |
|
299 |
qlen = math.sqrt(q[0]*q[0] + q[1]*q[1] + q[2]*q[2] + q[3]*q[3]) |
300 |
q[0] = q[0]/qlen |
301 |
q[1] = q[1]/qlen |
302 |
q[2] = q[2]/qlen |
303 |
q[3] = q[3]/qlen |
304 |
|
305 |
outputFile.write("%10d %7s %18.10g %18.10g %18.10g %13e %13e %13e %13e %13e %13e %13e %13e %13e %13e\n" % (index, sdFormat, xcart, ycart, zcart, 0.0, 0.0, 0.0, q[0], q[1], q[2], q[3], 0.0, 0.0, 0.0)) |
306 |
index = index+1 |
307 |
|
308 |
outputFile.write(" </StuntDoubles>\n") |
309 |
outputFile.write(" </Snapshot>\n") |
310 |
outputFile.write("</OpenMD>\n") |
311 |
outputFile.close() |
312 |
|
313 |
outputFile.close() |
314 |
|
315 |
def main(argv): |
316 |
|
317 |
arrayType = "A" |
318 |
haveOutputFileName = False |
319 |
latticeType = "fcc" |
320 |
dipoleDirection = "001" |
321 |
latticeNumber = 3 |
322 |
latticeConstant = 4 |
323 |
try: |
324 |
opts, args = getopt.getopt(argv, "hxyzabl:d:c:n:o:", ["help","array-type-X", "array-type-Y", "array-type-Z", "array-type-A", "array-type-B", "lattice=" "direction=", "constant=", "output-file="]) |
325 |
except getopt.GetoptError: |
326 |
usage() |
327 |
sys.exit(2) |
328 |
for opt, arg in opts: |
329 |
if opt in ("-h", "--help"): |
330 |
usage() |
331 |
sys.exit() |
332 |
elif opt in ("-x", "--array-type-X"): |
333 |
arrayType = "X" |
334 |
elif opt in ("-y", "--array-type-Y"): |
335 |
arrayType = "Y" |
336 |
elif opt in ("-z", "--array-type-Z"): |
337 |
arrayType = "Z" |
338 |
elif opt in ("-b", "--array-type-B"): |
339 |
arrayType = "B" |
340 |
elif opt in ("-l", "--lattice"): |
341 |
latticeType = arg |
342 |
elif opt in ("-d", "--direction"): |
343 |
dipoleDirection = arg |
344 |
elif opt in ("-c", "--constant"): |
345 |
latticeConstant = float(arg) |
346 |
elif opt in ("-n"): |
347 |
latticeNumber = int(arg) |
348 |
elif opt in ("-o", "--output-file"): |
349 |
outputFileName = arg |
350 |
haveOutputFileName = True |
351 |
if (not haveOutputFileName): |
352 |
usage() |
353 |
print "No output file was specified" |
354 |
sys.exit() |
355 |
|
356 |
createLattice(latticeType, latticeNumber, latticeConstant, dipoleDirection, arrayType, outputFileName); |
357 |
|
358 |
if __name__ == "__main__": |
359 |
if len(sys.argv) == 1: |
360 |
usage() |
361 |
sys.exit() |
362 |
main(sys.argv[1:]) |