Alternative implementation of qpms_scatsys_scattered_E()

For testing purposes. Seems to work OK.


Former-commit-id: 897e687d27dbd81b2aaac17fb8f19bc4257dc887

qpms/qpms_c.pyx

@@ -710,25 +710,29 @@ cdef class ScatteringSystem:
 
         return retdict
 
-    def scattered_E(self, cdouble wavenumber, scatcoeffvector_full, evalpos):
+    def scattered_E(self, cdouble wavenumber, scatcoeffvector_full, evalpos, bint alt=False):
-        evalpos = np.array(evalpos, copy=False)
+        evalpos = np.array(evalpos, dtype=float, copy=False)
-        cdef np.ndarray evalpos_a = evalpos
-        cdef np.ndarray[dtype=complex] scv = np.array(scatcoeffvector_full, copy=False)
-        cdef cdouble[::1] scv_view = scv
         if evalpos.shape[-1] != 3:
             raise ValueError("Last dimension of evalpos has to be 3")
-        cdef np.ndarray[complex] results = np.empty(evalpos.shape, dtype=complex)
+        cdef np.ndarray[double,ndim=2] evalpos_a = evalpos.reshape(-1,3)
+        cdef np.ndarray[dtype=complex, ndim=1] scv = np.array(scatcoeffvector_full, copy=False)
+        cdef cdouble[::1] scv_view = scv
+        cdef np.ndarray[complex, ndim=2] results = np.empty((evalpos_a.shape[0],3), dtype=complex)
         cdef ccart3_t res
         cdef cart3_t pos
-        for i in np.ndindex(evalpos.shape[:-1]):
+        cdef size_t i
+        for i in range(evalpos_a.shape[0]):
             pos.x = evalpos_a[i,0]
             pos.y = evalpos_a[i,1]
             pos.z = evalpos_a[i,2]
+            if alt:
+                res = qpms_scatsys_scattered_E__alt(self.s, wavenumber, &scv_view[0], pos)
+            else:
                 res = qpms_scatsys_scattered_E(self.s, wavenumber, &scv_view[0], pos)
             results[i,0] = res.x
             results[i,1] = res.y
             results[i,2] = res.z
-        return results
+        return results.reshape(evalpos.shape)
 
 def empty_lattice_modes_xy(EpsMu epsmu, reciprocal_basis, wavevector, double maxomega):
     '''Empty (2D, xy-plane) lattice mode (diffraction order) frequencies of a non-dispersive medium.
@@ -863,25 +867,29 @@ cdef class _ScatteringSystemAtOmega:
     def translation_matrix_full(self, blochvector = None):
         return self.ss_pyref.translation_matrix_full(wavenumber=self.wavenumber, blochvector=blochvector)
 
-    def scattered_E(self, scatcoeffvector_full, evalpos):
+    def scattered_E(self, scatcoeffvector_full, evalpos, bint alt=False):
-        evalpos = np.array(evalpos, copy=False)
+        evalpos = np.array(evalpos, dtype=float, copy=False)
-        cdef np.ndarray evalpos_a = evalpos
-        cdef np.ndarray[dtype=complex] scv = np.array(scatcoeffvector_full, copy=False)
-        cdef cdouble[::1] scv_view = scv
         if evalpos.shape[-1] != 3:
             raise ValueError("Last dimension of evalpos has to be 3")
-        cdef np.ndarray[complex] results = np.empty(evalpos.shape, dtype=complex)
+        cdef np.ndarray[double,ndim=2] evalpos_a = evalpos.reshape(-1,3)
+        cdef np.ndarray[dtype=complex, ndim=1] scv = np.array(scatcoeffvector_full, copy=False)
+        cdef cdouble[::1] scv_view = scv
+        cdef np.ndarray[complex, ndim=2] results = np.empty((evalpos_a.shape[0],3), dtype=complex)
         cdef ccart3_t res
         cdef cart3_t pos
-        for i in np.ndindex(evalpos.shape[:-1]):
+        cdef size_t i
+        for i in range(evalpos_a.shape[0]):
             pos.x = evalpos_a[i,0]
             pos.y = evalpos_a[i,1]
             pos.z = evalpos_a[i,2]
+            if alt:
+                res = qpms_scatsysw_scattered_E__alt(self.ssw, &scv_view[0], pos)
+            else:
                 res = qpms_scatsysw_scattered_E(self.ssw, &scv_view[0], pos)
             results[i,0] = res.x
             results[i,1] = res.y
             results[i,2] = res.z
-        return results
+        return results.reshape(evalpos.shape)
 
 
 cdef class ScatteringMatrix:

qpms/qpms_cdefs.pxd

@@ -629,6 +629,10 @@ cdef extern from "scatsystem.h":
             const cdouble *f_excitation_vector_full, cart3_t where)
     ccart3_t qpms_scatsysw_scattered_E(const qpms_scatsys_at_omega_t *ssw,
             const cdouble *f_excitation_vector_full, cart3_t where)
+    ccart3_t qpms_scatsys_scattered_E__alt(const qpms_scatsys_t *ss, cdouble wavenumber,
+            const cdouble *f_excitation_vector_full, cart3_t where)
+    ccart3_t qpms_scatsysw_scattered_E__alt(const qpms_scatsys_at_omega_t *ssw,
+            const cdouble *f_excitation_vector_full, cart3_t where)
 
 cdef extern from "ewald.h":
     struct qpms_csf_result:

qpms/scatsystem.c

@@ -2018,6 +2018,69 @@ ccart3_t qpms_scatsysw_scattered_E(const qpms_scatsys_at_omega_t *ssw,
       cvf, where);
 }
 
+// Alternative implementation, using translation operator and regular dipole waves at zero
+ccart3_t qpms_scatsys_scattered_E__alt(const qpms_scatsys_t *ss, 
+    const complex double k,
+    const complex double *cvf, 
+    const cart3_t where
+    ) {
+  QPMS_UNTESTED;
+  ccart3_t res = {0,0,0};
+  ccart3_t res_kc = {0,0,0}; // kahan sum compensation
+
+  // bspec containing only electric dipoles
+  const qpms_vswf_set_spec_t dipspec = {
+    .n = 3,
+    .ilist = (qpms_uvswfi_t[]){
+      qpms_tmn2uvswfi(QPMS_VSWF_ELECTRIC, -1, 1),
+      qpms_tmn2uvswfi(QPMS_VSWF_ELECTRIC,  0, 1),
+      qpms_tmn2uvswfi(QPMS_VSWF_ELECTRIC, +1, 1),
+    },
+    .lMax=1, .lMax_M=0, .lMax_N=1, .lMax_L=-1,
+    .capacity=0,
+    .norm = ss->c->normalisation,
+  };
+
+  ccart3_t regdipoles_0[3]; {
+    const sph_t origin_sph = {.r = 0, .theta = M_PI_2, .phi=0}; // Should work with any theta/phi (TESTWORTHY)
+    csphvec_t regdipoles_0_sph[3];
+    QPMS_ENSURE_SUCCESS(qpms_uvswf_fill(regdipoles_0_sph, &dipspec,
+          sph2csph(origin_sph), QPMS_BESSEL_REGULAR));
+    for(int i = 0; i < 3; ++i)
+      regdipoles_0[i] = csphvec2ccart(regdipoles_0_sph[i], origin_sph);
+  }
+
+  complex double *s; // Translation matrix
+  QPMS_CRASHING_MALLOC(s, ss->max_bspecn * sizeof(*s) * dipspec.n);
+
+  for (qpms_ss_pi_t pi = 0; pi < ss->p_count; ++pi) {
+    const qpms_vswf_set_spec_t *bspec = qpms_ss_bspec_pi(ss, pi);
+    const cart3_t particle_pos = ss->p[pi].pos;
+    const complex double *particle_cv = cvf + ss->fecv_pstarts[pi];
+
+    const cart3_t origin_cart = {0, 0, 0};
+
+    QPMS_ENSURE_SUCCESS(qpms_trans_calculator_get_trans_array_lc3p(
+          ss->c, s, &dipspec, 1, bspec, 3, k, particle_pos, where,
+          QPMS_HANKEL_PLUS));
+
+    for(size_t i = 0; i < bspec->n; ++i)
+      for(size_t j = 0; j < 3; ++j){
+        ccart3_t summand = ccart3_scale(particle_cv[i] * s[3*i+j], regdipoles_0[j]);
+        ckahanadd(&(res.x), &(res_kc.x), summand.x);
+        ckahanadd(&(res.y), &(res_kc.y), summand.y);
+        ckahanadd(&(res.z), &(res_kc.z), summand.z);
+      }
+  }
+  free(s);
+  return res;
+}
+
+ccart3_t qpms_scatsysw_scattered_E__alt(const qpms_scatsys_at_omega_t *ssw, 
+    const complex double *cvf,  const cart3_t where) {
+  return qpms_scatsys_scattered_E__alt(ssw->ss, ssw->wavenumber,
+      cvf, where);
+}
 
 #if 0
 ccart3_t qpms_scatsys_scattered_E_irrep(const qpms_scatsys_t *ss,

qpms/scatsystem.h

@@ -727,6 +727,37 @@ ccart3_t qpms_scatsysw_scattered_E(
 		cart3_t evalpoint ///< A point \f$ \vect r \f$, at which the field is evaluated.
 		);
 
+/// Evaluates scattered electric field at a point, given a full vector of scattered field coefficients.
+/**
+ * This is an alternative implementation of qpms_scatsys_scattered_E(), and should give the same results
+ * up to rounding errors.
+ *
+ * \return Complex electric field at the point defined by \a evalpoint.
+ *
+ * \see qpms_scatsys_scattered_E()
+ */
+ccart3_t qpms_scatsys_scattered_E__alt(
+		const qpms_scatsys_t *ss,
+		complex double wavenumber, ///< Wavenumber of the background medium.
+		const complex double *scatcoeff_full, ///< Full vector of the scattered field coefficients \f$ \wckcout \f$.
+		cart3_t evalpoint ///< A point \f$ \vect r \f$, at which the field is evaluated.
+		);
+
+/// Evaluates scattered electric field at a point, given a full vector of scattered field coefficients.
+/**
+ * This is an alternative implementation of qpms_scatsys_scattered_E(), and should give the same results
+ * up to rounding errors.
+ *
+ * \return Complex electric field at the point defined by \a evalpoint.
+ *
+ * \see qpms_scatsysw_scattered_E()
+ */
+ccart3_t qpms_scatsysw_scattered_E__alt(
+		const qpms_scatsys_at_omega_t *ssw,
+		const complex double *scatcoeff_full, ///< Full vector of the scattered field coefficients \f$ \wckcout \f$.
+		cart3_t evalpoint ///< A point \f$ \vect r \f$, at which the field is evaluated.
+		);
+
 #if 0
 /** Evaluates partial scattered fields (corresponding to a given irrep-reduced excitation vector)
  * at a given point.

tests/test_vswf_transop_consistency.py

@@ -0,0 +1,45 @@
+#!/usr/bin/env python3
+"""
+Tests whether direct evaluation of VSWFs gives the same result as their
+representation in terms of translation operators and regular electric dipole
+waves at origin
+"""
+
+from qpms import Particle, CTMatrix, lorentz_drude, EpsMuGenerator, TMatrixGenerator, BaseSpec, FinitePointGroup, ScatteringSystem, TMatrixInterpolator, EpsMu, dbgmsg_enable, dbgmsg_disable, dbgmsg_active, BesselType,eV, hbar
+from qpms.symmetries import point_group_info
+import numpy as np
+eh = eV/hbar
+np.random.seed(666)
+dbgmsg_enable(2)
+
+part_radius = 80e-9
+p = 1580e-9
+
+sym = FinitePointGroup(point_group_info['D4h'])
+bspec1 = BaseSpec(lMax=3)
+medium=EpsMuGenerator(EpsMu(1.52**2))
+t1 = TMatrixGenerator.sphere(medium, EpsMuGenerator(lorentz_drude['Au']), r=part_radius)
+p1 = Particle((0,0,0),t1,bspec=bspec1)
+ss, ssw = ScatteringSystem.create([p1], EpsMuGenerator(EpsMu(1.52**2)), 1.4*eh, sym)
+
+
+points = np.random.random((100,3)) * p
+points = points[np.linalg.norm(points, axis=-1) > part_radius]
+t,l,m = bspec1.tlm()
+
+fails=0
+
+for i in range(ss.fecv_size):
+    fvc = np.zeros((ss.fecv_size,), dtype=complex)
+    fvc[i] = 1
+    
+    E = ssw.scattered_E(fvc, points)
+    E_alt = ssw.scattered_E(fvc, points,alt=True)
+    diff = abs(E-E_alt)
+    reldiffavg = np.average(diff/(abs(E)+abs(E_alt)))
+    fail = reldiffavg > 1e-3
+    fails += fail
+                                
+    print('E' if t[i] == 2 else 'M', l[i], m[i], np.amax(diff), reldiffavg, 'FAIL!' if fail else 'OK')
+
+exit(fails)