/* fp2rat.c (convert floating-point number to rational number) */ /*********************************************************************** * This code is part of GLPK (GNU Linear Programming Kit). * * Copyright (C) 2000-2013 Andrew Makhorin, Department for Applied * Informatics, Moscow Aviation Institute, Moscow, Russia. All rights * reserved. E-mail: . * * GLPK 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. * * GLPK 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 GLPK. If not, see . ***********************************************************************/ #include "env.h" #include "misc.h" /*********************************************************************** * NAME * * fp2rat - convert floating-point number to rational number * * SYNOPSIS * * #include "misc.h" * int fp2rat(double x, double eps, double *p, double *q); * * DESCRIPTION * * Given a floating-point number 0 <= x < 1 the routine fp2rat finds * its "best" rational approximation p / q, where p >= 0 and q > 0 are * integer numbers, such that |x - p / q| <= eps. * * RETURNS * * The routine fp2rat returns the number of iterations used to achieve * the specified precision eps. * * EXAMPLES * * For x = sqrt(2) - 1 = 0.414213562373095 and eps = 1e-6 the routine * gives p = 408 and q = 985, where 408 / 985 = 0.414213197969543. * * BACKGROUND * * It is well known that every positive real number x can be expressed * as the following continued fraction: * * x = b[0] + a[1] * ------------------------ * b[1] + a[2] * ----------------- * b[2] + a[3] * ---------- * b[3] + ... * * where: * * a[k] = 1, k = 0, 1, 2, ... * * b[k] = floor(x[k]), k = 0, 1, 2, ... * * x[0] = x, * * x[k] = 1 / frac(x[k-1]), k = 1, 2, 3, ... * * To find the "best" rational approximation of x the routine computes * partial fractions f[k] by dropping after k terms as follows: * * f[k] = A[k] / B[k], * * where: * * A[-1] = 1, A[0] = b[0], B[-1] = 0, B[0] = 1, * * A[k] = b[k] * A[k-1] + a[k] * A[k-2], * * B[k] = b[k] * B[k-1] + a[k] * B[k-2]. * * Once the condition * * |x - f[k]| <= eps * * has been satisfied, the routine reports p = A[k] and q = B[k] as the * final answer. * * In the table below here is some statistics obtained for one million * random numbers uniformly distributed in the range [0, 1). * * eps max p mean p max q mean q max k mean k * ------------------------------------------------------------- * 1e-1 8 1.6 9 3.2 3 1.4 * 1e-2 98 6.2 99 12.4 5 2.4 * 1e-3 997 20.7 998 41.5 8 3.4 * 1e-4 9959 66.6 9960 133.5 10 4.4 * 1e-5 97403 211.7 97404 424.2 13 5.3 * 1e-6 479669 669.9 479670 1342.9 15 6.3 * 1e-7 1579030 2127.3 3962146 4257.8 16 7.3 * 1e-8 26188823 6749.4 26188824 13503.4 19 8.2 * * REFERENCES * * W. B. Jones and W. J. Thron, "Continued Fractions: Analytic Theory * and Applications," Encyclopedia on Mathematics and Its Applications, * Addison-Wesley, 1980. */ int fp2rat(double x, double eps, double *p, double *q) { int k; double xk, Akm1, Ak, Bkm1, Bk, ak, bk, fk, temp; xassert(0.0 <= x && x < 1.0); for (k = 0; ; k++) { xassert(k <= 100); if (k == 0) { /* x[0] = x */ xk = x; /* A[-1] = 1 */ Akm1 = 1.0; /* A[0] = b[0] = floor(x[0]) = 0 */ Ak = 0.0; /* B[-1] = 0 */ Bkm1 = 0.0; /* B[0] = 1 */ Bk = 1.0; } else { /* x[k] = 1 / frac(x[k-1]) */ temp = xk - floor(xk); xassert(temp != 0.0); xk = 1.0 / temp; /* a[k] = 1 */ ak = 1.0; /* b[k] = floor(x[k]) */ bk = floor(xk); /* A[k] = b[k] * A[k-1] + a[k] * A[k-2] */ temp = bk * Ak + ak * Akm1; Akm1 = Ak, Ak = temp; /* B[k] = b[k] * B[k-1] + a[k] * B[k-2] */ temp = bk * Bk + ak * Bkm1; Bkm1 = Bk, Bk = temp; } /* f[k] = A[k] / B[k] */ fk = Ak / Bk; #if 0 print("%.*g / %.*g = %.*g", DBL_DIG, Ak, DBL_DIG, Bk, DBL_DIG, fk); #endif if (fabs(x - fk) <= eps) break; } *p = Ak; *q = Bk; return k; } /* eof */