Quaternions and Wigner matrix elements; untested.

Former-commit-id: 27dfd620a73f7ec246882863239e96a29c87a867
This commit is contained in:
Marek Nečada 2019-02-22 05:33:35 +00:00
parent c5600a8558
commit 83b8b72578
2 changed files with 145 additions and 0 deletions

57
qpms/wigner.c Normal file
View File

@ -0,0 +1,57 @@
#include "wigner.h"
#include "tiny_inlines.h"
#include "kahansum.h"
#define WIGNER_ATOL (1e-15)
#define MIN(a,b) ((a)<(b)?(a):b)
#define MAX(a,b) ((a)>(b)?(a):b)
complex double qpms_wignerD_elem(const qpms_quat_t R,
const qpms_l_t l, const qpms_m_t mp, const qpms_m_t m) {
// TODO do some optimisations... The combinatoric coeffs could be precomputed.
if (abs(m) > l || abs(mp) > l) abort();//return 0; // It's safer to crash. :D
const double mRa = cabs(R.a), mRb = cabs(R.b);
if (mRa < WIGNER_ATOL) {
if (m != -mp) return 0;
else return min1pow(l+m) * cpow(R.b, 2*m);
} else if (mRb < WIGNER_ATOL) {
if (m != mp) return 0;
else return cpow(R.a, 2*m);
} else if (mRa >= mRb) {
complex double prefac = exp(.5*(lgamma(l+m+1) + lgamma(l-m+1)
- lgamma(l+mp+1) - lgamma(l-mp+1)))
* pow(mRa, 2*l-2*m)
* cpow(R.a, mp+m)
* cpow(R.b, -mp+m);
double sum, sumc; kahaninit( &sum, &sumc);
// FIXME this is probably quite stupid way to calculate the sum
for(int rho=MAX(0, -m+mp); rho <= MIN(l-m, l+mp); ++rho)
kahanadd(&sum, &sumc,
exp(
lgamma(l+mp+1) - lgamma(rho+1) - lgamma(l+mp-rho+1)
+ lgamma(l-mp+1) - lgamma(l-rho-m+1) - lgamma(-mp+rho+m+1)
)
* pow(-mRb*mRb/(mRa*mRa), rho)
);
return prefac * sum;
} else { // (mRa < mRb)
complex double prefac = min1pow(l-m) * exp(.5*(lgamma(l+m+1) + lgamma(l-m+1)
- lgamma(l+mp+1) - lgamma(l-mp+1)))
* pow(mRb, 2*l-2*m)
* cpow(R.a, mp+m)
* cpow(R.b, -mp+m);
double sum, sumc; kahaninit( &sum, &sumc);
// FIXME this is probably quite stupid way to calculate the sum
for(int rho=MAX(0, -m-mp); rho <= MIN(l-mp, l-m); ++rho)
kahanadd(&sum, &sumc,
exp(
lgamma(l+mp+1) - lgamma(l-m-rho+1) - lgamma(mp+m+rho+1)
+ lgamma(l-mp+1) - lgamma(rho+1) - lgamma(l-mp-rho+1)
)
* pow(-mRa*mRa/(mRb*mRb), rho)
);
return prefac * sum;
}
}

88
qpms/wigner.h Normal file
View File

@ -0,0 +1,88 @@
/*! \file wigner.h
* \brief Quaternions and Wigner matrices
*/
#ifndef QPMS_WIGNER_H
#define QPMS_WIGNER_H
#include "qpms_types.h"
/// Quaternion type.
/**
* Internaly represented as a pair of complex numbers,
* \f$ Q_a = Q_1 + iQ_z, Q_b = Q_y + i Q_x\f$.
*/
typedef struct qpms_quat_t {
complex double a, b;
} qpms_quat_t;
/// Quaternion type as four doubles.
typedef struct qpms_quat4d_t {
double c1, ci, cj, ck;
} qpms_quat4d_t;
/// Conversion from the 4*double to the 2*complex quaternion.
// TODO is this really correct?
// I.e. do the axis from moble's text match this convention?
static inline qpms_quat_t qpms_quat_2c_from_4d (qpms_quat4d_t q) {
qpms_quat_t q2c = {q.c1 + I * q.ck, q.cj + I * q.ci};
return q2c;
}
/// Quaternion multiplication.
/**
* \f[ (P Q)_a = P_a Q_a - \bar P_b Q_b, \f]
* \f[ (P Q)_b = P_b Q_a + \bar P_a Q_b. \f]
*/
static inline qpms_quat_t qpms_quat_mult(qpms_quat_t p, qpms_quat_t q) {
qpms_quat_t r;
r.a = p.a * q.a - conj(p.b) * q.b;
r.b = p.b * q.a + conj(p.a) * q.b;
return r;
}
/// Quaternion addition.
static inline qpms_quat_t qpms_quat_add(qpms_quat_t p, qpms_quat_t q) {
qpms_quat_t r;
r.a = p.a+q.a;
r.b = p.b+q.b;
return r;
}
/// Quaternion scaling with a real number.
static inline qpms_quat_t qpms_quat_rscale(double s, qpms_quat_t q) {
qpms_quat_t r = {s * q.a, s * q.b};
return r;
}
// quaternion "basis"
static const qpms_quat_t qpms_quat_1 = {1, 0};
static const qpms_quat_t qpms_quat_i = {I, 0};
static const qpms_quat_t qpms_quat_j = {0, 1};
static const qpms_quat_t qpms_quat_k = {0, I};
/// Quaternion conjugation.
static inline qpms_quat_t qpms_quat_conj(qpms_quat_t q) {
qpms_quat_t r = {conj(q.a), -q.b};
return r;
}
/// Quaternion norm.
static inline double qpms_quat_norm(qpms_quat_t q) {
return sqrt(creal(q.a * conj(q.a) + q.b * conj(q.b)));
}
/// Quaternion normalisation to unit norm.
static inline qpms_quat_t qpms_quat_normalise(qpms_quat_t q) {
double n = qpms_quat_norm(q);
return qpms_quat_rscale(1/n, q);
}
/// Wigner D matrix element from a rotator quaternion for integer l.
/**
* The D matrix are calculated using formulae (3), (4), (6), (7) from
* http://moble.github.io/spherical_functions/WignerDMatrices.html
*/
complex double qpms_wignerD_elem(qpms_quat_t q, qpms_l_t l,
qpms_m_t mp, qpms_m_t m);
#endif //QPMS_WIGNER_H