From 79b805a6aad9dce6d5acc38d9c2d7fad01b9e494 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Marek=20Ne=C4=8Dada?= Date: Wed, 11 Mar 2020 15:02:04 +0200 Subject: [PATCH] Remove pre-Ewald lattice sum related "split-Bessel" functions Former-commit-id: 18d1c77460bf1563cab50b3c55d481f8dc696364 --- qpms/bessels.c | 65 -------- qpms/bessels.h | 61 ------- qpms/lrhankel_recspace_dirty.c | 293 --------------------------------- 3 files changed, 419 deletions(-) delete mode 100644 qpms/bessels.c delete mode 100644 qpms/bessels.h delete mode 100644 qpms/lrhankel_recspace_dirty.c diff --git a/qpms/bessels.c b/qpms/bessels.c deleted file mode 100644 index fc5a21d..0000000 --- a/qpms/bessels.c +++ /dev/null @@ -1,65 +0,0 @@ -#include "bessels.h" -#include -#include -#include -#include - -static const double ln2 = 0.693147180559945309417; - - -// general; gives an array of size xxx with TODODESC -complex double * hankelcoefftable_init(size_t maxn) { - complex double *hct = malloc((maxn+1)*(maxn+2)/2 * sizeof(complex double)); - for(size_t n = 0; n <= maxn; ++n) { - complex double *hcs = hankelcoeffs_get(hct,n); - for (size_t k = 0; k <= n; ++k) { - double lcoeff = lgamma(n+k+1) - lgamma(n-k+1) - lgamma(k+1) - k*ln2; - // for some reason, casting k-n to double does not work,so - // cpow (I, k-n-1) cannot be used... - complex double ifactor; - switch ((n+1-k) % 4) { - case 0: - ifactor = 1; - break; - case 1: - ifactor = -I; - break; - case 2: - ifactor = -1; - break; - case 3: - ifactor = I; - break; - default: - abort(); - } - // the result should be integer, so round to remove inaccuracies - hcs[k] = round(exp(lcoeff)) * ifactor; - } - } - return hct; -} - -void hankelparts_fill(complex double *lrt, complex double *srt, size_t maxn, - size_t lrk_cutoff, complex double const * const hct, - unsigned kappa, double c, double x) { - if (lrt) memset(lrt, 0, (maxn+1)*sizeof(complex double)); - memset(srt, 0, (maxn+1)*sizeof(complex double)); - double regularisator = pow(1. - exp(-c * x), (double) kappa); - double antiregularisator = 1. - regularisator; - double xfrac = 1.; // x ** (-1-k) - for (size_t k = 0; k <= maxn; ++k) { - xfrac /= x; - for(size_t n = k; n <= maxn; ++n) // TODO Kahan sums here - srt[n] += ((k -#include - -complex double *hankelcoefftable_init(size_t maxn); - - -// For navigating in the coefficients, maybe not for public use -static inline complex double * -trindex_cd(complex double const * const arr, size_t n){ - return (complex double *)(arr + n*(n+1)/2); -} - -// general, gives the offset such that result[ql] is -// the coefficient corresponding to the e**(I * x) * x**(-ql-1) -// term of the n-th Hankel function; no boundary checks! -static inline complex double * -hankelcoeffs_get(complex double const * const hankelcoefftable, size_t n){ - return trindex_cd(hankelcoefftable, n); -} - -/* Compute the untransformed long- (optional) and short-range parts of spherical Hankel functions */ -// general; target_longrange and target_shortrange are of size (maxn+1) -// if target_longrange is NULL, only the short-range part is calculated -void hankelparts_fill(complex double *target_longrange, /* Not needed for the actual calculations - (only for testing and error estimates) - as summed in the reciprocal space: - pass NULL to omit */ - complex double *target_shortrange, - size_t maxn, size_t longrange_order_cutoff, /* terms e**(I x)/x**(k+1), - k>= longrange_order_cutoff go - completely to short-range part */ - complex double const * const hankelcoefftable, - unsigned kappa, double c, /* regularisation "slope", dimensionless */ - double x); // dimensionless x = k0 * r - - - -/* Hankel transforms of the long-range parts of the spherical Hankel functions */ -// this declaration is general; however, the implementation -// is so far only for kappa == 5, maxp <= 5, longrange_order_cutoff <= 1 -void lrhankel_recpart_fill(complex double *target_longrange_kspace /*Must be of size maxn*(maxn+1)/2*/, - size_t maxp /* Max. order of the Hankel transform */, - size_t longrange_order_cutoff /* terms e**(I x)/x**(k+1), k>= longrange_order_cutoff go - completely to the shortrange part - index with hankelcoeffs_get(target,p)l[delta_m] */, - complex double const * const hankelcoefftable, - unsigned kappa, - // These are dimensionFUL (inverse lengths): - double cv, double k0, double k); - -#endif //BESSELS_H diff --git a/qpms/lrhankel_recspace_dirty.c b/qpms/lrhankel_recspace_dirty.c deleted file mode 100644 index 22953b5..0000000 --- a/qpms/lrhankel_recspace_dirty.c +++ /dev/null @@ -1,293 +0,0 @@ -/* - * This is a dirty implementation of lrhankel_recpart_fill() that calculates - * the (cylindrical) Hankel transforms of the regularised part of the spherical Hankel - * functions that are to be summed in the reciprocal space. - * - * For now, only the regularisation with κ == 5 && q <= 2 && n <= 5 is implemented - * by writing down the explicit formula for each q,n pair and k>k0 vs k= 3, probably due to catastrophic cancellation (hopefully not due - * to an error in the formula!). On the other hand, - * these numbers are tiny in their absolute value, so their contribution to the - * lattice sum should be negligible. - * - * Therefore TODO use kahan summation. - * - */ - -#define MAXQM 1 -#define MAXN 5 -#define MAXKAPPA 5 - -#include "bessels.h" -//#include "mdefs.h" -#include -#include -#include -#define SQ(x) ((x)*(x)) - -#define P4(x) (((x)*(x))*((x)*(x))) - - -/* - * General form of the κ == 5 transforms. One usually has to put a (-1) factor - * to some part of the zeroth term of one of the cases k < k0 or k > k0 in order - * to stay on the correct branch of complex square root... - */ -#define KAPPA5SUM(form) (\ - (form(0, 1)) \ - - 5*(form(1, 1)) \ - +10*(form(2, 1)) \ - -10*(form(3, 1)) \ - + 5*(form(4, 1)) \ - - (form(5, 1)) \ -) - -#define KAPPA5SUMFF(form) (\ - (form(0, (-1))) \ - - 5*(form(1, 1)) \ - +10*(form(2, 1)) \ - -10*(form(3, 1)) \ - + 5*(form(4, 1)) \ - - (form(5, 1)) \ -) - -/* - * Prototype for the individual (per q,n) Bessel transform calculating functions. - * a, b, d, e, ash are recurring pre-calculated intermediate results, see the definition - * of lrhankel_recpart_fill() below to see their meaning - */ -#define LRHANKELDEF(fname) complex double fname(const double c, const double k0, const double k, \ - const complex double *a, const complex double *b, const complex double *d, \ - const complex double *e, const complex double *ash) - -typedef complex double (*lrhankelspec)(const double, const double, const double, - const complex double *, - const complex double *, - const complex double *, - const complex double *, - const complex double *); -// complex double fun(double c, double k0, double k, ccd *a, ccd *b, ccd *d, ccd *e) - -#define FORMK5Q1N0(i,ff) (ff*e[i]) -LRHANKELDEF(fk5q1n0l){ - return (KAPPA5SUMFF(FORMK5Q1N0))/k0; -} -LRHANKELDEF(fk5q1n0s){ - return (KAPPA5SUM(FORMK5Q1N0))/k0; -} -#undef FORMK5Q1N0 - -#define FORMK5Q1N1(i,ff) (-ff*d[i]) -LRHANKELDEF(fk5q1n1l){ - return (KAPPA5SUMFF(FORMK5Q1N1))/(k0*k); -} -LRHANKELDEF(fk5q1n1s){ - return (KAPPA5SUM(FORMK5Q1N1))/(k0*k); -} -#undef FORMK5Q1N1 - -#define FORMK5Q1N2(i,ff) (ff*e[i] - t*a[i] + ff*t*d[i]*a[i]) -LRHANKELDEF(fk5q1n2l){ - double t = 2/(k*k); - return (KAPPA5SUMFF(FORMK5Q1N2))/k0; -} -LRHANKELDEF(fk5q1n2s){ - double t = 2/(k*k); - return (KAPPA5SUM(FORMK5Q1N2))/k0; -} -#undef FORMK5Q1N2 - -#define FORMK5Q1N3(i,ff) (-ff*d[i] * (kk3 + 4*a[i]*a[i])) -LRHANKELDEF(fk5q1n3l){ - double kk3 = 3*k*k; - return (KAPPA5SUMFF(FORMK5Q1N3))/(k0*k*k*k); -} -LRHANKELDEF(fk5q1n3s){ - double kk3 = 3*k*k; - return (KAPPA5SUM(FORMK5Q1N3))/(k0*k*k*k); -} -#undef FORMK5Q1N3 - -#define FORMK5Q1N4(i,ff) (ff*e[i] * (kkkk + kk8*a[i]*a[i] + 8*P4(a[i]))) -LRHANKELDEF(fk5q1n4l){ - double kk8 = k*k*8, kkkk = P4(k); - return (KAPPA5SUMFF(FORMK5Q1N4))/(k0*kkkk); -} -LRHANKELDEF(fk5q1n4s){ - double kk8 = k*k*8, kkkk = P4(k); - return (KAPPA5SUM(FORMK5Q1N4))/(k0*kkkk); -} -#undef FORMK5Q1N4 - -#define FORMK5Q1N5(i,ff) (d[i]*(kkkk*(-5*ff+b[i])-ff*kk20*a[i]*a[i]-ff*16*P4(a[i]))) -LRHANKELDEF(fk5q1n5l){ - double kk20 = k*k*20, kkkk = P4(k); - return (KAPPA5SUMFF(FORMK5Q1N5))/(k0*kkkk*k); -} -LRHANKELDEF(fk5q1n5s){ - double kk20 = k*k*20, kkkk = P4(k); - return (KAPPA5SUM(FORMK5Q1N5))/(k0*kkkk*k); -} -#undef FORMK5Q1N5 - -#define FORMK5Q2N0(i,ff) (-ash[i]) -LRHANKELDEF(fk5q2n0){ - return (KAPPA5SUM(FORMK5Q2N0)) / (k0*k0); -} -const lrhankelspec fk5q2n0s = fk5q2n0, fk5q2n0l = fk5q2n0; -#undef FORMK5Q2N0 - -#define FORMK5Q2N1(i,ff) (ff*b[i]*a[i]) -LRHANKELDEF(fk5q2n1l){ - return (KAPPA5SUMFF(FORMK5Q2N1))/(k*k0*k0); -} -LRHANKELDEF(fk5q2n1s){ - return (KAPPA5SUM(FORMK5Q2N1))/(k*k0*k0); -} -#undef FORMK5Q2N1 - -#define FORMK5Q2N2(i,ff) (-ff*b[i]*a[i]*a[i]) -LRHANKELDEF(fk5q2n2l){ - return (KAPPA5SUMFF(FORMK5Q2N2)) / (k*k*k0*k0); -} -LRHANKELDEF(fk5q2n2s){ - return (KAPPA5SUM(FORMK5Q2N2)) / (k*k*k0*k0); -} - -#if 0 -complex double fk5q3n0l(double c, double k0, double k, - const complex double *a, const complex double *b, const complex double *d, const complex double *e, const complex double *ash) { // FIXME - return ( /* FIXME */ - - k*b[0] + a[0] * ash[0] - + 5 * k*b[1] + a[1] * ash[1] - -10 * k*b[2] + a[2] * ash[2] - +10 * k*b[3] + a[3] * ash[3] - - 5 * k*b[4] + a[4] * ash[4] - + k*b[5] + a[5] * ash[5] - )/(k0*k0*k0); -} -#endif - -#define FORMK5Q2N3(i,ff) (ff*a[i]*b[i]*(kk + 4*a[i]*a[i])) -LRHANKELDEF(fk5q2n3l){ - double kk = k*k; - return (KAPPA5SUMFF(FORMK5Q2N3))/(3*k0*k0*kk*k); -} -LRHANKELDEF(fk5q2n3s){ - double kk = k*k; - return (KAPPA5SUM(FORMK5Q2N3))/(3*k0*k0*kk*k); -} -#undef FORMK5Q2N3 - -#define FORMK5Q2N4(i,ff) (-ff*b[i]*a[i]*a[i]*(kk+2*a[i]*a[i])) -LRHANKELDEF(fk5q2n4l){ - double kk = k*k; - return (KAPPA5SUMFF(FORMK5Q2N4))/(k0*k0*kk*kk); -} -LRHANKELDEF(fk5q2n4s){ - double kk = k*k; - return (KAPPA5SUM(FORMK5Q2N4))/(k0*k0*kk*kk); -} -#undef FORMK5Q2N4 - -#define FORMK5Q2N5(i,ff) (ff*a[i]*b[i]*(kkkk+12*kk*(a[i]*a[i])+16*P4(a[i])) ) -LRHANKELDEF(fk5q2n5l){ - double kk = k*k; - double kkkk = kk * kk; - return ( - KAPPA5SUMFF(FORMK5Q2N5) - +16*120*P4(c)*c // Stirling S2(5,5) is no longer zero - )/(5*k0*k0*kkkk*k); -} -LRHANKELDEF(fk5q2n5s){ - double kk = k*k; - double kkkk = kk * kk; - return ( - KAPPA5SUM(FORMK5Q2N5) - +16*120*P4(c)*c // Stirling S2(5,5) is no longer zero - )/(5*k0*k0*kkkk*k); -} -#undef FORMK5Q2N5 - - -static const lrhankelspec transfuns_f[MAXKAPPA+1][MAXQM+1][MAXN+1] = { - {{NULL,NULL,NULL,NULL,NULL,NULL},{NULL,NULL,NULL,NULL,NULL,NULL}}, - {{NULL,NULL,NULL,NULL,NULL,NULL},{NULL,NULL,NULL,NULL,NULL,NULL}}, - {{NULL,NULL,NULL,NULL,NULL,NULL},{NULL,NULL,NULL,NULL,NULL,NULL}}, - {{NULL,NULL,NULL,NULL,NULL,NULL},{NULL,NULL,NULL,NULL,NULL,NULL}}, - {{NULL,NULL,NULL,NULL,NULL,NULL},{NULL,NULL,NULL,NULL,NULL,NULL}}, - {{fk5q1n0l,fk5q1n1l,fk5q1n2l,fk5q1n3l,fk5q1n4l,fk5q1n5l},{fk5q2n0,fk5q2n1l,fk5q2n2l,fk5q2n3l,fk5q2n4l,fk5q2n5l}} -}; - -static const lrhankelspec transfuns_n[MAXKAPPA+1][MAXQM+1][MAXN+1] = { - {{NULL,NULL,NULL,NULL,NULL,NULL},{NULL,NULL,NULL,NULL,NULL,NULL}}, - {{NULL,NULL,NULL,NULL,NULL,NULL},{NULL,NULL,NULL,NULL,NULL,NULL}}, - {{NULL,NULL,NULL,NULL,NULL,NULL},{NULL,NULL,NULL,NULL,NULL,NULL}}, - {{NULL,NULL,NULL,NULL,NULL,NULL},{NULL,NULL,NULL,NULL,NULL,NULL}}, - {{NULL,NULL,NULL,NULL,NULL,NULL},{NULL,NULL,NULL,NULL,NULL,NULL}}, - {{fk5q1n0s,fk5q1n1s,fk5q1n2s,fk5q1n3s,fk5q1n4s,fk5q1n5s},{fk5q2n0,fk5q2n1s,fk5q2n2s,fk5q2n3s,fk5q2n4s,fk5q2n5s}} -}; - -void lrhankel_recpart_fill(complex double *target, - size_t maxp /* max. order of the Hankel transform */, - size_t lrk_cutoff, - complex double const *const hct, - unsigned kappa, double c, double k0, double k) -{ - assert(5 == kappa); // Only kappa == 5 implemented so far - assert(maxp <= MAXN); // only n <= 5 implemented so far - assert(lrk_cutoff <= MAXQM + 1); // only q <= 2 implemented so far; TODO shouldn't it be only MAXQM ??? - const lrhankelspec (*funarr)[MAXQM+1][MAXN+1] = (k>k0) ? transfuns_f : transfuns_n; - memset(target, 0, maxp*(maxp+1)/2*sizeof(complex double)); - complex double a[kappa+1], b[kappa+1], d[kappa+1], e[kappa+1], ash[kappa+1]; - for (size_t sigma = 0; sigma <= kappa; ++sigma) { - a[sigma] = (sigma * c - I * k0); - b[sigma] = csqrt(1+k*k/(a[sigma]*a[sigma])); - d[sigma] = 1/b[sigma]; - e[sigma] = d[sigma] / a[sigma]; - ash[sigma] = casinh(a[sigma]/k); - } - for (size_t ql = 0; (ql <= maxp) && (ql < lrk_cutoff); ++ql) // ql is q-1, i.e. corresponds to the hankel term power - for (size_t deltam = 0; deltam <= maxp; ++deltam){ - complex double result = funarr[kappa][ql][deltam](c,k0,k,a,b,d,e,ash); - for (size_t p = 0; p <= maxp; ++p) - trindex_cd(target,p)[deltam] += result * hankelcoeffs_get(hct,p)[ql]; - } -} - -#ifdef TESTING -#include -int main() { - double k0 = 0.7; - double c = 0.1324; - double kmin = 0.000; - double kmax = 20; - double kstep = 0.001; - size_t kappa = 5; - - for (double k = kmin; k <= kmax; k += kstep) { - printf("%f ", k); - complex double a[kappa+1], b[kappa+1], d[kappa+1], e[kappa+1], ash[kappa+1]; - for (size_t sigma = 0; sigma <= kappa; ++sigma) { - a[sigma] = (sigma * c - I * k0); - b[sigma] = csqrt(1+k*k/(a[sigma]*a[sigma])); - d[sigma] = 1/b[sigma]; - e[sigma] = d[sigma] / a[sigma]; - ash[sigma] = casinh(a[sigma]/k); - } - for (size_t qm = 0; qm <= MAXQM; ++qm) - for (size_t n = 0; n <= MAXN; ++n) { - //if (/*!*/((qm==1)&&(n==0))){ // not skip q==2, n=0 for now -// complex double fun(double c, double k0, double k, ccd *a, ccd *b, ccd *d, ccd *e) - complex double result = - (k < k0 ? transfuns_n : transfuns_f)[kappa][qm][n](c,k0,k,a,b,d,e,ash); - printf("%.16e %.16e ", creal(result), cimag(result)); - } - printf("\n"); - } - return 0; -} -#endif