Coverage for /builds/kinetik161/ase/ase/vibrations/albrecht.py: 89.82%
285 statements
« prev ^ index » next coverage.py v7.2.7, created at 2023-12-10 11:04 +0000
« prev ^ index » next coverage.py v7.2.7, created at 2023-12-10 11:04 +0000
1import sys
2from itertools import combinations_with_replacement
4import numpy as np
6import ase.units as u
7from ase.parallel import paropen, parprint
8from ase.vibrations.franck_condon import (FranckCondonOverlap,
9 FranckCondonRecursive)
10from ase.vibrations.resonant_raman import ResonantRaman
13class Albrecht(ResonantRaman):
14 def __init__(self, *args, **kwargs):
15 """
16 Parameters
17 ----------
18 all from ResonantRaman.__init__
19 combinations: int
20 Combinations to consider for multiple excitations.
21 Default is 1, possible 2
22 skip: int
23 Number of first transitions to exclude. Default 0,
24 recommended: 5 for linear molecules, 6 for other molecules
25 nm: int
26 Number of intermediate m levels to consider, default 20
27 """
28 self.combinations = kwargs.pop('combinations', 1)
29 self.skip = kwargs.pop('skip', 0)
30 self.nm = kwargs.pop('nm', 20)
31 approximation = kwargs.pop('approximation', 'Albrecht')
33 ResonantRaman.__init__(self, *args, **kwargs)
35 self.set_approximation(approximation)
37 def set_approximation(self, value):
38 approx = value.lower()
39 if approx in ['albrecht', 'albrecht b', 'albrecht c', 'albrecht bc']:
40 if not self.overlap:
41 raise ValueError('Overlaps are needed')
42 elif not approx == 'albrecht a':
43 raise ValueError('Please use "Albrecht" or "Albrecht A/B/C/BC"')
44 self._approx = value
46 def calculate_energies_and_modes(self):
47 if hasattr(self, 'im_r'):
48 return
50 ResonantRaman.calculate_energies_and_modes(self)
52 # single transitions and their occupation
53 om_Q = self.om_Q[self.skip:]
54 om_v = om_Q
55 ndof = len(om_Q)
56 n_vQ = np.eye(ndof, dtype=int)
58 l_Q = range(ndof)
59 ind_v = list(combinations_with_replacement(l_Q, 1))
61 if self.combinations > 1:
62 if not self.combinations == 2:
63 raise NotImplementedError
65 for c in range(2, self.combinations + 1):
66 ind_v += list(combinations_with_replacement(l_Q, c))
68 nv = len(ind_v)
69 n_vQ = np.zeros((nv, ndof), dtype=int)
70 om_v = np.zeros((nv), dtype=float)
71 for j, wt in enumerate(ind_v):
72 for i in wt:
73 n_vQ[j, i] += 1
74 om_v = n_vQ.dot(om_Q)
76 self.ind_v = ind_v
77 self.om_v = om_v
78 self.n_vQ = n_vQ # how many of each
79 self.d_vQ = np.where(n_vQ > 0, 1, 0) # do we have them ?
81 def get_energies(self):
82 self.calculate_energies_and_modes()
83 return self.om_v
85 def _collect_r(self, arr_ro, oshape, dtype):
86 """Collect an array that is distributed."""
87 if len(self.myr) == self.ndof: # serial
88 return arr_ro
89 data_ro = np.zeros([self.ndof] + oshape, dtype)
90 if len(arr_ro):
91 data_ro[self.slize] = arr_ro
92 self.comm.sum(data_ro)
93 return data_ro
95 def Huang_Rhys_factors(self, forces_r):
96 """Evaluate Huang-Rhys factors derived from forces."""
97 return 0.5 * self.unitless_displacements(forces_r)**2
99 def unitless_displacements(self, forces_r, mineigv=1e-12):
100 """Evaluate unitless displacements from forces
102 Parameters
103 ----------
104 forces_r: array
105 Forces in cartesian coordinates
106 mineigv: float
107 Minimal Eigenvalue to consider in matrix inversion to handle
108 numerical noise. Default 1e-12
110 Returns
111 -------
112 Unitless displacements in Eigenmode coordinates
113 """
114 assert len(forces_r.flat) == self.ndof
116 if not hasattr(self, 'Dm1_q'):
117 self.eigv_q, self.eigw_rq = np.linalg.eigh(
118 self.im_r[:, None] * self.H * self.im_r)
119 # there might be zero or nearly zero eigenvalues
120 self.Dm1_q = np.divide(1, self.eigv_q,
121 out=np.zeros_like(self.eigv_q),
122 where=np.abs(self.eigv_q) > mineigv)
123 X_r = self.eigw_rq @ np.diag(self.Dm1_q) @ self.eigw_rq.T @ (
124 forces_r.flat * self.im_r)
126 d_Q = np.dot(self.modes_Qq, X_r)
127 s = 1.e-20 / u.kg / u.C / u._hbar**2
128 d_Q *= np.sqrt(s * self.om_Q)
130 return d_Q
132 def omegaLS(self, omega, gamma):
133 omL = omega + 1j * gamma
134 omS_Q = omL - self.om_Q
135 return omL, omS_Q
137 def init_parallel_excitations(self):
138 """Init for paralellization over excitations."""
139 n_p = len(self.ex0E_p)
141 # collect excited state forces
142 exF_pr = self._collect_r(self.exF_rp, [n_p], self.ex0E_p.dtype).T
144 # select your work load
145 myn = -(-n_p // self.comm.size) # ceil divide
146 rank = self.comm.rank
147 s = slice(myn * rank, myn * (rank + 1))
148 return n_p, range(n_p)[s], exF_pr
150 def meA(self, omega, gamma=0.1):
151 """Evaluate Albrecht A term.
153 Returns
154 -------
155 Full Albrecht A matrix element. Unit: e^2 Angstrom^2 / eV
156 """
157 self.read()
159 if not hasattr(self, 'fcr'):
160 self.fcr = FranckCondonRecursive()
162 omL = omega + 1j * gamma
163 omS_Q = omL - self.om_Q
165 n_p, myp, exF_pr = self.init_parallel_excitations()
166 exF_pr = np.where(np.abs(exF_pr) > 1e-2, exF_pr, 0)
168 m_Qcc = np.zeros((self.ndof, 3, 3), dtype=complex)
169 for p in myp:
170 energy = self.ex0E_p[p]
171 d_Q = self.unitless_displacements(exF_pr[p])
172 energy_Q = energy - self.om_Q * d_Q**2 / 2.
173 me_cc = np.outer(self.ex0m_pc[p], self.ex0m_pc[p].conj())
175 wm_Q = np.zeros((self.ndof), dtype=complex)
176 wp_Q = np.zeros((self.ndof), dtype=complex)
177 for m in range(self.nm):
178 fco_Q = self.fcr.direct0mm1(m, d_Q)
179 e_Q = energy_Q + m * self.om_Q
180 wm_Q += fco_Q / (e_Q - omL)
181 wp_Q += fco_Q / (e_Q + omS_Q)
182 m_Qcc += np.einsum('a,bc->abc', wm_Q, me_cc)
183 m_Qcc += np.einsum('a,bc->abc', wp_Q, me_cc.conj())
184 self.comm.sum(m_Qcc)
186 return m_Qcc # e^2 Angstrom^2 / eV
188 def meAmult(self, omega, gamma=0.1):
189 """Evaluate Albrecht A term.
191 Returns
192 -------
193 Full Albrecht A matrix element. Unit: e^2 Angstrom^2 / eV
194 """
195 self.read()
197 if not hasattr(self, 'fcr'):
198 self.fcr = FranckCondonRecursive()
200 omL = omega + 1j * gamma
201 omS_v = omL - self.om_v
202 nv = len(self.om_v)
203 om_Q = self.om_Q[self.skip:]
204 nQ = len(om_Q)
206 # n_v:
207 # how many FC factors are involved
208 # nvib_ov:
209 # delta functions to switch contributions depending on order o
210 # ind_ov:
211 # Q indicees
212 # n_ov:
213 # # of vibrational excitations
214 n_v = self.d_vQ.sum(axis=1) # multiplicity
216 nvib_ov = np.empty((self.combinations, nv), dtype=int)
217 om_ov = np.zeros((self.combinations, nv), dtype=float)
218 n_ov = np.zeros((self.combinations, nv), dtype=int)
219 d_ovQ = np.zeros((self.combinations, nv, nQ), dtype=int)
220 for o in range(self.combinations):
221 nvib_ov[o] = np.array(n_v == (o + 1))
222 for v in range(nv):
223 try:
224 om_ov[o, v] = om_Q[self.ind_v[v][o]]
225 d_ovQ[o, v, self.ind_v[v][o]] = 1
226 except IndexError:
227 pass
228 # XXXX change ????
229 n_ov[0] = self.n_vQ.max(axis=1)
230 n_ov[1] = nvib_ov[1]
232 n_p, myp, exF_pr = self.init_parallel_excitations()
234 m_vcc = np.zeros((nv, 3, 3), dtype=complex)
235 for p in myp:
236 energy = self.ex0E_p[p]
237 d_Q = self.unitless_displacements(exF_pr[p])[self.skip:]
238 S_Q = d_Q**2 / 2.
239 energy_v = energy - self.d_vQ.dot(om_Q * S_Q)
240 me_cc = np.outer(self.ex0m_pc[p], self.ex0m_pc[p].conj())
242 fco1_mQ = np.empty((self.nm, nQ), dtype=float)
243 fco2_mQ = np.empty((self.nm, nQ), dtype=float)
244 for m in range(self.nm):
245 fco1_mQ[m] = self.fcr.direct0mm1(m, d_Q)
246 fco2_mQ[m] = self.fcr.direct0mm2(m, d_Q)
248 wm_v = np.zeros((nv), dtype=complex)
249 wp_v = np.zeros((nv), dtype=complex)
250 for m in range(self.nm):
251 fco1_v = np.where(n_ov[0] == 2,
252 d_ovQ[0].dot(fco2_mQ[m]),
253 d_ovQ[0].dot(fco1_mQ[m]))
255 em_v = energy_v + m * om_ov[0]
256 # multiples of same kind
257 fco_v = nvib_ov[0] * fco1_v
258 wm_v += fco_v / (em_v - omL)
259 wp_v += fco_v / (em_v + omS_v)
260 if nvib_ov[1].any():
261 # multiples of mixed type
262 for n in range(self.nm):
263 fco2_v = d_ovQ[1].dot(fco1_mQ[n])
264 e_v = em_v + n * om_ov[1]
265 fco_v = nvib_ov[1] * fco1_v * fco2_v
266 wm_v += fco_v / (e_v - omL)
267 wp_v += fco_v / (e_v + omS_v)
269 m_vcc += np.einsum('a,bc->abc', wm_v, me_cc)
270 m_vcc += np.einsum('a,bc->abc', wp_v, me_cc.conj())
271 self.comm.sum(m_vcc)
273 return m_vcc # e^2 Angstrom^2 / eV
275 def meBC(self, omega, gamma=0.1,
276 term='BC'):
277 """Evaluate Albrecht BC term.
279 Returns
280 -------
281 Full Albrecht BC matrix element.
282 Unit: e^2 Angstrom / eV / sqrt(amu)
283 """
284 self.read()
286 if not hasattr(self, 'fco'):
287 self.fco = FranckCondonOverlap()
289 omL = omega + 1j * gamma
290 omS_Q = omL - self.om_Q
292 # excited state forces
293 n_p, myp, exF_pr = self.init_parallel_excitations()
294 # derivatives after normal coordinates
295 exdmdr_rpc = self._collect_r(
296 self.exdmdr_rpc, [n_p, 3], self.ex0m_pc.dtype)
297 dmdq_qpc = (exdmdr_rpc.T * self.im_r).T # unit e / sqrt(amu)
298 dmdQ_Qpc = np.dot(dmdq_qpc.T, self.modes_Qq.T).T # unit e / sqrt(amu)
300 me_Qcc = np.zeros((self.ndof, 3, 3), dtype=complex)
301 for p in myp:
302 energy = self.ex0E_p[p]
303 S_Q = self.Huang_Rhys_factors(exF_pr[p])
304 # relaxed excited state energy
305 # # n_vQ = np.where(self.n_vQ > 0, 1, 0)
306 # # energy_v = energy - n_vQ.dot(self.om_Q * S_Q)
307 energy_Q = energy - self.om_Q * S_Q
309 # # me_cc = np.outer(self.ex0m_pc[p], self.ex0m_pc[p].conj())
310 m_c = self.ex0m_pc[p] # e Angstrom
311 dmdQ_Qc = dmdQ_Qpc[:, p] # e / sqrt(amu)
313 wBLS_Q = np.zeros((self.ndof), dtype=complex)
314 wBSL_Q = np.zeros((self.ndof), dtype=complex)
315 wCLS_Q = np.zeros((self.ndof), dtype=complex)
316 wCSL_Q = np.zeros((self.ndof), dtype=complex)
317 for m in range(self.nm):
318 f0mmQ1_Q = (self.fco.directT0(m, S_Q) +
319 np.sqrt(2) * self.fco.direct0mm2(m, S_Q))
320 f0Qmm1_Q = self.fco.direct(1, m, S_Q)
322 em_Q = energy_Q + m * self.om_Q
323 wBLS_Q += f0mmQ1_Q / (em_Q - omL)
324 wBSL_Q += f0Qmm1_Q / (em_Q - omL)
325 wCLS_Q += f0mmQ1_Q / (em_Q + omS_Q)
326 wCSL_Q += f0Qmm1_Q / (em_Q + omS_Q)
328 # unit e^2 Angstrom / sqrt(amu)
329 mdmdQ_Qcc = np.einsum('a,bc->bac', m_c, dmdQ_Qc.conj())
330 dmdQm_Qcc = np.einsum('ab,c->abc', dmdQ_Qc, m_c.conj())
331 if 'B' in term:
332 me_Qcc += np.multiply(wBLS_Q, mdmdQ_Qcc.T).T
333 me_Qcc += np.multiply(wBSL_Q, dmdQm_Qcc.T).T
334 if 'C' in term:
335 me_Qcc += np.multiply(wCLS_Q, mdmdQ_Qcc.T).T
336 me_Qcc += np.multiply(wCSL_Q, dmdQm_Qcc.T).T
338 self.comm.sum(me_Qcc)
339 return me_Qcc # unit e^2 Angstrom / eV / sqrt(amu)
341 def electronic_me_Qcc(self, omega, gamma):
342 self.calculate_energies_and_modes()
344 approx = self.approximation.lower()
345 assert self.combinations == 1
346 Vel_Qcc = np.zeros((len(self.om_Q), 3, 3), dtype=complex)
347 if approx == 'albrecht a' or approx == 'albrecht':
348 Vel_Qcc += self.meA(omega, gamma) # e^2 Angstrom^2 / eV
349 # divide through pre-factor
350 with np.errstate(divide='ignore'):
351 Vel_Qcc *= np.where(self.vib01_Q > 0,
352 1. / self.vib01_Q, 0)[:, None, None]
353 # -> e^2 Angstrom / eV / sqrt(amu)
354 if approx == 'albrecht bc' or approx == 'albrecht':
355 Vel_Qcc += self.meBC(omega, gamma) # e^2 Angstrom / eV / sqrt(amu)
356 if approx == 'albrecht b':
357 Vel_Qcc += self.meBC(omega, gamma, term='B')
358 if approx == 'albrecht c':
359 Vel_Qcc = self.meBC(omega, gamma, term='C')
361 Vel_Qcc *= u.Hartree * u.Bohr # e^2 Angstrom^2 / eV -> Angstrom^3
363 return Vel_Qcc # Angstrom^2 / sqrt(amu)
365 def me_Qcc(self, omega, gamma):
366 """Full matrix element"""
367 self.read()
368 approx = self.approximation.lower()
369 nv = len(self.om_v)
370 V_vcc = np.zeros((nv, 3, 3), dtype=complex)
371 if approx == 'albrecht a' or approx == 'albrecht':
372 if self.combinations == 1:
373 # e^2 Angstrom^2 / eV
374 V_vcc += self.meA(omega, gamma)[self.skip:]
375 else:
376 V_vcc += self.meAmult(omega, gamma)
377 if approx == 'albrecht bc' or approx == 'albrecht':
378 if self.combinations == 1:
379 vel_vcc = self.meBC(omega, gamma)
380 V_vcc += vel_vcc * self.vib01_Q[:, None, None]
381 else:
382 vel_vcc = self.meBCmult(omega, gamma)
383 V_vcc = 0
384 elif approx == 'albrecht b':
385 assert self.combinations == 1
386 vel_vcc = self.meBC(omega, gamma, term='B')
387 V_vcc = vel_vcc * self.vib01_Q[:, None, None]
388 if approx == 'albrecht c':
389 assert self.combinations == 1
390 vel_vcc = self.meBC(omega, gamma, term='C')
391 V_vcc = vel_vcc * self.vib01_Q[:, None, None]
393 return V_vcc # e^2 Angstrom^2 / eV
395 def summary(self, omega=0, gamma=0,
396 method='standard', direction='central',
397 log=sys.stdout):
398 """Print summary for given omega [eV]"""
399 if self.combinations > 1:
400 return self.extended_summary()
402 om_v = self.get_energies()
403 intensities = self.get_absolute_intensities(omega, gamma)[self.skip:]
405 if isinstance(log, str):
406 log = paropen(log, 'a')
408 parprint('-------------------------------------', file=log)
409 parprint(' excitation at ' + str(omega) + ' eV', file=log)
410 parprint(' gamma ' + str(gamma) + ' eV', file=log)
411 parprint(' approximation:', self.approximation, file=log)
412 parprint(' Mode Frequency Intensity', file=log)
413 parprint(' # meV cm^-1 [A^4/amu]', file=log)
414 parprint('-------------------------------------', file=log)
415 for n, e in enumerate(om_v):
416 if e.imag != 0:
417 c = 'i'
418 e = e.imag
419 else:
420 c = ' '
421 e = e.real
422 parprint('%3d %6.1f %7.1f%s %9.1f' %
423 (n, 1000 * e, e / u.invcm, c, intensities[n]),
424 file=log)
425 parprint('-------------------------------------', file=log)
426 parprint('Zero-point energy: %.3f eV' %
427 self.vibrations.get_zero_point_energy(),
428 file=log)
430 def extended_summary(self, omega=0, gamma=0,
431 method='standard', direction='central',
432 log=sys.stdout):
433 """Print summary for given omega [eV]"""
434 self.read(method, direction)
435 om_v = self.get_energies()
436 intens_v = self.intensity(omega, gamma)
438 if isinstance(log, str):
439 log = paropen(log, 'a')
441 parprint('-------------------------------------', file=log)
442 parprint(' excitation at ' + str(omega) + ' eV', file=log)
443 parprint(' gamma ' + str(gamma) + ' eV', file=log)
444 parprint(' approximation:', self.approximation, file=log)
445 parprint(' observation:', self.observation, file=log)
446 parprint(' Mode Frequency Intensity', file=log)
447 parprint(' # meV cm^-1 [e^4A^4/eV^2]', file=log)
448 parprint('-------------------------------------', file=log)
449 for v, e in enumerate(om_v):
450 parprint(self.ind_v[v], '{:6.1f} {:7.1f} {:9.1f}'.format(
451 1000 * e, e / u.invcm, 1e9 * intens_v[v]),
452 file=log)
453 parprint('-------------------------------------', file=log)
454 parprint('Zero-point energy: %.3f eV' %
455 self.vibrations.get_zero_point_energy(),
456 file=log)