aboutsummaryrefslogtreecommitdiff
path: root/libcrystfel/src/rational.c
diff options
context:
space:
mode:
Diffstat (limited to 'libcrystfel/src/rational.c')
-rw-r--r--libcrystfel/src/rational.c675
1 files changed, 675 insertions, 0 deletions
diff --git a/libcrystfel/src/rational.c b/libcrystfel/src/rational.c
new file mode 100644
index 00000000..65b209ff
--- /dev/null
+++ b/libcrystfel/src/rational.c
@@ -0,0 +1,675 @@
+/*
+ * rational.c
+ *
+ * A small rational number library
+ *
+ * Copyright © 2019 Deutsches Elektronen-Synchrotron DESY,
+ * a research centre of the Helmholtz Association.
+ *
+ * Authors:
+ * 2019 Thomas White <taw@physics.org>
+ *
+ * This file is part of CrystFEL.
+ *
+ * CrystFEL is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * CrystFEL is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with CrystFEL. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <assert.h>
+#include <locale.h>
+
+#include "rational.h"
+#include "integer_matrix.h"
+#include "utils.h"
+
+
+/**
+ * SECTION:rational
+ * @short_description: Rational numbers
+ * @title: Rational numbers
+ * @section_id:
+ * @see_also:
+ * @include: "rational.h"
+ * @Image:
+ *
+ * A rational number library
+ */
+
+
+/* Eucliden algorithm for finding greatest common divisor */
+static signed int gcd(signed int a, signed int b)
+{
+ while ( b != 0 ) {
+ signed int t = b;
+ b = a % b;
+ a = t;
+ }
+ return a;
+}
+
+
+static void squish(Rational *rt)
+{
+ signed int g;
+
+ if ( rt->num == 0 ) {
+ rt->den = 1;
+ return;
+ }
+
+ g = gcd(rt->num, rt->den);
+ assert(g != 0);
+ rt->num /= g;
+ rt->den /= g;
+
+ if ( rt->den < 0 ) {
+ rt->num = -rt->num;
+ rt->den = -rt->den;
+ }
+}
+
+
+Rational rtnl_zero()
+{
+ Rational r;
+ r.num = 0;
+ r.den = 1;
+ return r;
+}
+
+
+Rational rtnl(signed long long int num, signed long long int den)
+{
+ Rational r;
+ r.num = num;
+ r.den = den;
+ squish(&r);
+ return r;
+}
+
+
+double rtnl_as_double(Rational r)
+{
+ return (double)r.num/r.den;
+}
+
+
+static void overflow(long long int c, long long int a, long long int b)
+{
+ setlocale(LC_ALL, "");
+ ERROR("Overflow detected in rational number library.\n");
+ ERROR("%'lli < %'lli * %'lli\n", c, a, b);
+ abort();
+}
+
+
+static void check_overflow(long long int c, long long int a, long long int b)
+{
+ if ( (a==0) || (b==0) ) {
+ if ( c != 0 ) overflow(c,a,b);
+ } else if ( (llabs(c) < llabs(a)) || (llabs(c) < llabs(b)) ) {
+ overflow(c,a,b);
+ }
+}
+
+
+Rational rtnl_mul(Rational a, Rational b)
+{
+ Rational r;
+ r.num = a.num * b.num;
+ r.den = a.den * b.den;
+ check_overflow(r.num, a.num, b.num);
+ check_overflow(r.den, a.den, b.den);
+ squish(&r);
+ return r;
+}
+
+
+Rational rtnl_div(Rational a, Rational b)
+{
+ signed int t = b.num;
+ b.num = b.den;
+ b.den = t;
+ return rtnl_mul(a, b);
+}
+
+
+Rational rtnl_add(Rational a, Rational b)
+{
+ Rational r, trt1, trt2;
+
+ trt1.num = a.num * b.den;
+ trt2.num = b.num * a.den;
+ check_overflow(trt1.num, a.num, b.den);
+ check_overflow(trt2.num, b.num, a.den);
+
+ trt1.den = a.den * b.den;
+ trt2.den = trt1.den;
+ check_overflow(trt1.den, a.den, b.den);
+
+ r.num = trt1.num + trt2.num;
+ r.den = trt1.den;
+ squish(&r);
+ return r;
+}
+
+
+Rational rtnl_sub(Rational a, Rational b)
+{
+ b.num = -b.num;
+ return rtnl_add(a, b);
+}
+
+
+/* -1, 0 +1 respectively for a<b, a==b, a>b */
+signed int rtnl_cmp(Rational a, Rational b)
+{
+ Rational trt1, trt2;
+
+ trt1.num = a.num * b.den;
+ trt2.num = b.num * a.den;
+
+ trt1.den = a.den * b.den;
+ trt2.den = a.den * b.den;
+
+ if ( trt1.num > trt2.num ) return +1;
+ if ( trt1.num < trt2.num ) return -1;
+ return 0;
+}
+
+
+Rational rtnl_abs(Rational a)
+{
+ Rational r = a;
+ squish(&r);
+ if ( r.num < 0 ) r.num = -r.num;
+ return r;
+}
+
+
+/**
+ * rtnl_format
+ * @rt: A %Rational
+ *
+ * Formats @rt as a string
+ *
+ * Returns: a string which should be freed by the caller
+ */
+char *rtnl_format(Rational rt)
+{
+ char *v = malloc(32);
+ if ( v == NULL ) return NULL;
+ if ( rt.den == 1 ) {
+ snprintf(v, 31, "%lli", rt.num);
+ } else {
+ snprintf(v, 31, "%lli/%lli", rt.num, rt.den);
+ }
+ return v;
+}
+
+
+Rational *rtnl_list(signed int num_min, signed int num_max,
+ signed int den_min, signed int den_max,
+ int *pn)
+{
+ signed int num, den;
+ Rational *list;
+ int n = 0;
+
+ list = malloc((1+num_max-num_min)*(1+den_max-den_min)*sizeof(Rational));
+ if ( list == NULL ) return NULL;
+
+ for ( num=num_min; num<=num_max; num++ ) {
+ for ( den=den_min; den<=den_max; den++ ) {
+
+ Rational r = rtnl(num, den);
+
+ /* Denominator zero? */
+ if ( den == 0 ) continue;
+
+ /* Same as last entry? */
+ if ( (n>0) && (rtnl_cmp(list[n-1], r)==0) ) continue;
+
+ /* Can be reduced? */
+ if ( gcd(num, den) != 1 ) continue;
+
+ list[n++] = r;
+ }
+ }
+ *pn = n;
+ return list;
+}
+
+
+/**
+ * SECTION:rational_matrix
+ * @short_description: Rational matrices
+ * @title: Rational matrices
+ * @section_id:
+ * @see_also:
+ * @include: "rational.h"
+ * @Image:
+ *
+ * A rational matrix library
+ */
+
+
+struct _rationalmatrix
+{
+ unsigned int rows;
+ unsigned int cols;
+ Rational *v;
+};
+
+
+/**
+ * rtnl_mtx_new:
+ * @rows: Number of rows that the new matrix is to have
+ * @cols: Number of columns that the new matrix is to have
+ *
+ * Allocates a new %RationalMatrix with all elements set to zero.
+ *
+ * Returns: a new %RationalMatrix, or NULL on error.
+ **/
+RationalMatrix *rtnl_mtx_new(unsigned int rows, unsigned int cols)
+{
+ RationalMatrix *m;
+ int i;
+
+ m = malloc(sizeof(RationalMatrix));
+ if ( m == NULL ) return NULL;
+
+ m->v = calloc(rows*cols, sizeof(Rational));
+ if ( m->v == NULL ) {
+ free(m);
+ return NULL;
+ }
+
+ m->rows = rows;
+ m->cols = cols;
+
+ for ( i=0; i<m->rows*m->cols; i++ ) {
+ m->v[i] = rtnl_zero();
+ }
+
+ return m;
+}
+
+
+RationalMatrix *rtnl_mtx_identity(int rows)
+{
+ int i;
+ RationalMatrix *m = rtnl_mtx_new(rows, rows);
+ for ( i=0; i<rows; i++ ) {
+ rtnl_mtx_set(m, i, i, rtnl(1,1));
+ }
+ return m;
+}
+
+
+RationalMatrix *rtnl_mtx_copy(const RationalMatrix *m)
+{
+ RationalMatrix *n;
+ int i;
+
+ n = rtnl_mtx_new(m->rows, m->cols);
+ if ( n == NULL ) return NULL;
+
+ for ( i=0; i<m->rows*m->cols; i++ ) {
+ n->v[i] = m->v[i];
+ }
+
+ return n;
+}
+
+
+Rational rtnl_mtx_get(const RationalMatrix *m, int i, int j)
+{
+ assert(m != NULL);
+ return m->v[j+m->cols*i];
+}
+
+
+void rtnl_mtx_set(const RationalMatrix *m, int i, int j, Rational v)
+{
+ assert(m != NULL);
+ m->v[j+m->cols*i] = v;
+}
+
+
+RationalMatrix *rtnl_mtx_from_intmat(const IntegerMatrix *m)
+{
+ RationalMatrix *n;
+ unsigned int rows, cols;
+ int i, j;
+
+ intmat_size(m, &rows, &cols);
+ n = rtnl_mtx_new(rows, cols);
+ if ( n == NULL ) return NULL;
+
+ for ( i=0; i<rows; i++ ) {
+ for ( j=0; j<cols; j++ ) {
+ rtnl_mtx_set(n, i, j, rtnl(intmat_get(m, i, j), 1));
+ }
+ }
+
+ return n;
+}
+
+
+IntegerMatrix *intmat_from_rtnl_mtx(const RationalMatrix *m)
+{
+ IntegerMatrix *n;
+ int i, j;
+
+ n = intmat_new(m->rows, m->cols);
+ if ( n == NULL ) return NULL;
+
+ for ( i=0; i<m->rows; i++ ) {
+ for ( j=0; j<m->cols; j++ ) {
+ Rational v = rtnl_mtx_get(m, i, j);
+ squish(&v);
+ if ( v.den != 1 ) {
+ ERROR("Rational matrix can't be converted to integers\n");
+ intmat_free(n);
+ return NULL;
+ }
+ intmat_set(n, i, j, v.num);
+ }
+ }
+
+ return n;
+}
+
+
+void rtnl_mtx_free(RationalMatrix *mtx)
+{
+ if ( mtx == NULL ) return;
+ free(mtx->v);
+ free(mtx);
+}
+
+
+/* rtnl_mtx_solve:
+ * @P: A %RationalMatrix
+ * @vec: An array of %Rational
+ * @ans: An array of %Rational in which to store the result
+ *
+ * Solves the matrix equation m*ans = vec, where @ans and @vec are
+ * vectors of %Rational.
+ *
+ * In this version, @m must be square.
+ *
+ * The number of columns in @m must equal the length of @ans, and the number
+ * of rows in @m must equal the length of @vec, but note that there is no way
+ * for this function to check that this is the case.
+ *
+ * Given that P is transformation of the unit cell axes (see ITA chapter 5.1),
+ * this function calculates the fractional coordinates of a point in the
+ * transformed axes, given its fractional coordinates in the original axes.
+ *
+ * Returns: non-zero on error
+ **/
+int transform_fractional_coords_rtnl(const RationalMatrix *P,
+ const Rational *ivec, Rational *ans)
+{
+ RationalMatrix *cm;
+ Rational *vec;
+ int h, k;
+ int i;
+
+ if ( P->rows != P->cols ) return 1;
+
+ /* Copy the matrix and vector because the calculation will
+ * be done in-place */
+ cm = rtnl_mtx_copy(P);
+ if ( cm == NULL ) return 1;
+
+ vec = malloc(cm->rows*sizeof(Rational));
+ if ( vec == NULL ) return 1;
+ for ( h=0; h<cm->rows; h++ ) vec[h] = ivec[h];
+
+ /* Gaussian elimination with partial pivoting */
+ h = 0;
+ k = 0;
+ while ( h<=cm->rows && k<=cm->cols ) {
+
+ int prow = 0;
+ Rational pval = rtnl_zero();
+ Rational t;
+
+ /* Find the row with the largest value in column k */
+ for ( i=h; i<cm->rows; i++ ) {
+ if ( rtnl_cmp(rtnl_abs(rtnl_mtx_get(cm, i, k)), pval) > 0 ) {
+ pval = rtnl_abs(rtnl_mtx_get(cm, i, k));
+ prow = i;
+ }
+ }
+
+ if ( rtnl_cmp(pval, rtnl_zero()) == 0 ) {
+ k++;
+ continue;
+ }
+
+ /* Swap 'prow' with row h */
+ for ( i=0; i<cm->cols; i++ ) {
+ t = rtnl_mtx_get(cm, h, i);
+ rtnl_mtx_set(cm, h, i, rtnl_mtx_get(cm, prow, i));
+ rtnl_mtx_set(cm, prow, i, t);
+ }
+ t = vec[prow];
+ vec[prow] = vec[h];
+ vec[h] = t;
+
+ /* Divide and subtract rows below */
+ for ( i=h+1; i<cm->rows; i++ ) {
+
+ int j;
+ Rational dval;
+
+ dval = rtnl_div(rtnl_mtx_get(cm, i, k),
+ rtnl_mtx_get(cm, h, k));
+
+ for ( j=0; j<cm->cols; j++ ) {
+ Rational t = rtnl_mtx_get(cm, i, j);
+ Rational p = rtnl_mul(dval, rtnl_mtx_get(cm, h, j));
+ t = rtnl_sub(t, p);
+ rtnl_mtx_set(cm, i, j, t);
+ }
+
+ /* Divide the right hand side as well */
+ Rational t = vec[i];
+ Rational p = rtnl_mul(dval, vec[h]);
+ vec[i] = rtnl_sub(t, p);
+ }
+
+ h++;
+ k++;
+
+
+ }
+
+ /* Back-substitution */
+ for ( i=cm->rows-1; i>=0; i-- ) {
+ int j;
+ Rational sum = rtnl_zero();
+ for ( j=i+1; j<cm->cols; j++ ) {
+ Rational av;
+ av = rtnl_mul(rtnl_mtx_get(cm, i, j), ans[j]);
+ sum = rtnl_add(sum, av);
+ }
+ sum = rtnl_sub(vec[i], sum);
+ ans[i] = rtnl_div(sum, rtnl_mtx_get(cm, i, i));
+ }
+
+ free(vec);
+ rtnl_mtx_free(cm);
+
+ return 0;
+}
+
+
+/**
+ * rtnl_mtx_print
+ * @m: A %RationalMatrix
+ *
+ * Prints @m to stderr.
+ *
+ */
+void rtnl_mtx_print(const RationalMatrix *m)
+{
+ unsigned int i, j;
+
+ for ( i=0; i<m->rows; i++ ) {
+
+ fprintf(stderr, "[ ");
+ for ( j=0; j<m->cols; j++ ) {
+ char *v = rtnl_format(rtnl_mtx_get(m, i, j));
+ fprintf(stderr, "%4s ", v);
+ free(v);
+ }
+ fprintf(stderr, "]\n");
+ }
+}
+
+
+void rtnl_mtx_mtxmult(const RationalMatrix *A, const RationalMatrix *B,
+ RationalMatrix *ans)
+{
+ int i, j;
+
+ assert(ans->cols == B->cols);
+ assert(ans->rows == A->rows);
+ assert(A->cols == B->rows);
+
+ for ( i=0; i<ans->rows; i++ ) {
+ for ( j=0; j<ans->cols; j++ ) {
+ int k;
+ Rational sum = rtnl_zero();
+ for ( k=0; k<A->rows; k++ ) {
+ Rational add;
+ add = rtnl_mul(rtnl_mtx_get(A, i, k),
+ rtnl_mtx_get(B, k, j));
+ sum = rtnl_add(sum, add);
+ }
+ rtnl_mtx_set(ans, i, j, sum);
+ }
+ }
+}
+
+
+/* Given a "P-matrix" (see ITA chapter 5.1), calculate the fractional
+ * coordinates of point "vec" in the original axes, given its fractional
+ * coordinates in the transformed axes.
+ * To transform in the opposite direction, we would multiply by Q = P^-1.
+ * Therefore, this direction is the "easy way" and needs just a matrix
+ * multiplication. */
+void transform_fractional_coords_rtnl_inverse(const RationalMatrix *P,
+ const Rational *vec,
+ Rational *ans)
+{
+ int i, j;
+
+ for ( i=0; i<P->rows; i++ ) {
+ ans[i] = rtnl_zero();
+ for ( j=0; j<P->cols; j++ ) {
+ Rational add;
+ add = rtnl_mul(rtnl_mtx_get(P, i, j), vec[j]);
+ ans[i] = rtnl_add(ans[i], add);
+ }
+ }
+}
+
+
+static RationalMatrix *delete_row_and_column(const RationalMatrix *m,
+ unsigned int di, unsigned int dj)
+{
+ RationalMatrix *n;
+ unsigned int i, j;
+
+ n = rtnl_mtx_new(m->rows-1, m->cols-1);
+ if ( n == NULL ) return NULL;
+
+ for ( i=0; i<n->rows; i++ ) {
+ for ( j=0; j<n->cols; j++ ) {
+
+ Rational val;
+ unsigned int gi, gj;
+
+ gi = (i>=di) ? i+1 : i;
+ gj = (j>=dj) ? j+1 : j;
+ val = rtnl_mtx_get(m, gi, gj);
+ rtnl_mtx_set(n, i, j, val);
+
+ }
+ }
+
+ return n;
+}
+
+
+static Rational cofactor(const RationalMatrix *m,
+ unsigned int i, unsigned int j)
+{
+ RationalMatrix *n;
+ Rational t, C;
+
+ n = delete_row_and_column(m, i, j);
+ if ( n == NULL ) {
+ fprintf(stderr, "Failed to allocate matrix.\n");
+ return rtnl_zero();
+ }
+
+ /* -1 if odd, +1 if even */
+ t = (i+j) & 0x1 ? rtnl(-1, 1) : rtnl(1, 1);
+
+ C = rtnl_mul(t, rtnl_mtx_det(n));
+ rtnl_mtx_free(n);
+
+ return C;
+}
+
+
+
+Rational rtnl_mtx_det(const RationalMatrix *m)
+{
+ unsigned int i, j;
+ Rational det;
+
+ assert(m->rows == m->cols); /* Otherwise determinant doesn't exist */
+
+ if ( m->rows == 2 ) {
+ Rational a, b;
+ a = rtnl_mul(rtnl_mtx_get(m, 0, 0), rtnl_mtx_get(m, 1, 1));
+ b = rtnl_mul(rtnl_mtx_get(m, 0, 1), rtnl_mtx_get(m, 1, 0));
+ return rtnl_sub(a, b);
+ }
+
+ i = 0; /* Fixed */
+ det = rtnl_zero();
+ for ( j=0; j<m->cols; j++ ) {
+ Rational a;
+ a = rtnl_mul(rtnl_mtx_get(m, i, j), cofactor(m, i, j));
+ det = rtnl_add(det, a);
+ }
+
+ return det;
+}