3 * Calculating %$n$%th roots
5 * (c) 2020 Straylight/Edgeware
8 /*----- Licensing notice --------------------------------------------------*
10 * This file is part of Catacomb.
12 * Catacomb is free software: you can redistribute it and/or modify it
13 * under the terms of the GNU Library General Public License as published
14 * by the Free Software Foundation; either version 2 of the License, or
15 * (at your option) any later version.
17 * Catacomb is distributed in the hope that it will be useful, but
18 * WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Library General Public License for more details.
22 * You should have received a copy of the GNU Library General Public
23 * License along with Catacomb. If not, write to the Free Software
24 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
28 /*----- Header files ------------------------------------------------------*/
33 #include "primeiter.h"
35 /*----- Main code ---------------------------------------------------------*/
37 /* --- @mp_nthrt@ --- *
39 * Arguments: @mp *d@ = fake destination
40 * @mp *a@ = an integer
41 * @mp *n@ = a strictly positive integer
42 * @int *exectp_out@ = set nonzero if an exact solution is found
44 * Returns: The integer %$\bigl\lfloor \sqrt[n]{a} \bigr\rfloor$%.
46 * Use: Return an approximation to the %$n$%th root of %$a$%.
47 * Specifically, it returns the largest integer %$x$% such that
48 * %$x^n \le a$%. If %$x^n = a$% then @*exactp_out@ is set
49 * nonzero; otherwise it is set zero. (If @exactp_out@ is null
50 * then this information is discarded.)
52 * The exponent %$n$% must be strictly positive: it's not clear
53 * to me what the right answer is for %$n \le 0$%. If %$a$% is
54 * negative then %$n$% must be odd; otherwise there is no real
58 mp *mp_nthrt(mp *d, mp *a, mp *n, int *exactp_out)
60 /* We want to find %$x$% such that %$x^n = a$%. Newton--Raphson says we
61 * start with a guess %$x_i$%, and refine it to a better guess %$x_{i+1}$%
62 * which is where the tangent to the curve %$y = x^n - a$% at %$x = x_i$%
63 * meets the %$x$%-axis. The tangent meets the curve at
64 * %($x_i, x_i^n - a)$%, and has gradient %$n x_i^{n-1}$%, and hence meets
65 * the %$x$%-axis at %$x_{i+1} = x_i - (x_i^n - a)/(n x_i^{n-1})$%.
67 * We're working with plain integers here, and we actually want
68 * %$\lfloor x \rfloor$%. We can check that we're close enough because
69 * %$x^n$% is monotonic for positive %$x$%: if %$x_i^n > a$% then we're too
70 * large; if %$(xi + i)^n \le a$% then we're too small; otherwise we're
74 mp *ai = MP_NEW, *bi = MP_NEW, *xi = MP_NEW,
75 *nm1 = MP_NEW, *delta = MP_NEW;
80 /* Firstly, deal with a zero or negative %$xa%. */
86 a = mp_neg(MP_NEW, a);
89 /* Secondly, reject %$n \le 0$%.
91 * Clearly %$x^0 = 1$% for nonzero %$x$%, so if %$a$% is zero then we could
92 * return anything, but maybe %$1$% for concreteness; but what do we return
93 * for %$a \ne 1$%? For %$n < 0$% there's just no hope.
97 /* Pick a starting point. This is rather important to get right. In
98 * particular, for large %$n$%, if we our initial guess too small, then the
99 * next iteration is a wild overestimate and it takes a long time to
102 nm1 = mp_sub(nm1, n, MP_ONE);
103 xi = mp_fromulong(xi, mp_bits(a));
104 xi = mp_add(xi, xi, nm1); mp_div(&xi, 0, xi, n);
105 assert(MP_CMP(xi, <=, MP_ULONG_MAX));
106 xi = mp_setbit(xi, MP_ZERO, mp_toulong(xi));
108 /* The main iteration. */
110 ai = mp_exp(ai, xi, n);
111 bi = mp_add(bi, xi, MP_ONE); bi = mp_exp(bi, bi, n);
113 if (cmp <= 0 && MP_CMP(a, <, bi)) break;
114 ai = mp_sub(ai, ai, a);
115 bi = mp_exp(bi, xi, nm1); bi = mp_mul(bi, bi, n);
116 mp_div(&delta, 0, ai, bi);
117 if (MP_ZEROP(delta) && cmp > 0) xi = mp_sub(xi, xi, MP_ONE);
118 else xi = mp_sub(xi, xi, delta);
121 /* Final sorting out of negative inputs.
123 * If the input %$a$% is not an exact %$n$%th root, then we must round
124 * %%\emph{up}%% so that, after negation, we implement the correct floor
128 if (cmp) xi = mp_add(xi, xi, MP_ONE);
132 /* Free up the temporary things. */
133 MP_DROP(a); MP_DROP(ai); MP_DROP(bi);
134 MP_DROP(nm1); MP_DROP(delta); mp_drop(d);
137 if (exactp_out) *exactp_out = (cmp == 0);
143 /* --- @mp_perfect_power_p@ --- *
145 * Arguments: @mp **x@ = where to write the base
146 * @mp **n@ = where to write the exponent
147 * @mp *a@ = an integer
149 * Returns: Nonzero if %$a$% is a perfect power.
151 * Use: Returns whether an integer %$a$% is a perfect power, i.e.,
152 * whether it can be written in the form %$a = x^n$% where
153 * %$|x| > 1$% and %$n > 1$% are integers. If this is possible,
154 * then (a) store %$x$% and the largest such %$n$% in @*x@ and
155 * @*n@, and return nonzero; otherwise, store %$x = a$% and
156 * %$n = 1$% and return zero. (Either @x@ or @n@, or both, may
157 * be null to discard these outputs.)
159 * Note that %$-1$%, %$0$% and %$1$% are not considered perfect
160 * powers by this definition. (The exponent is not well-defined
161 * in these cases, but it seemed better to implement a function
162 * which worked for all integers.) Note also that %$-4$% is not
163 * a perfect power since it has no real square root.
166 int mp_perfect_power_p(mp **x, mp **n, mp *a)
168 mp *r = MP_ONE, *p = MP_NEW, *t = MP_NEW;
175 if (MP_ZEROP(a)) goto done;
176 primeiter_create(&pi, MP_NEGP(a) ? MP_THREE : MP_TWO);
177 if (MP_NEGP(a)) { a = mp_neg(a, a); f |= f_neg; }
178 p = primeiter_next(&pi, p);
180 t = mp_nthrt(t, a, p, &exactp);
181 if (MP_EQ(t, MP_ONE))
184 if (MP_EQ(t, MP_ONE)) break;
185 p = primeiter_next(&pi, p);
188 MP_DROP(a); a = t; t = MP_NEW;
192 primeiter_destroy(&pi);
194 if (x) { mp_drop(*x); *x = f&f_neg ? mp_neg(a, a) : a; a = 0; }
195 if (n) { mp_drop(*n); *n = r; r = 0; }
196 mp_drop(a); mp_drop(p); mp_drop(r); mp_drop(t);
202 /*----- Test rig ----------------------------------------------------------*/
206 #include <mLib/testrig.h>
208 static int vrf_nthrt(dstr *v)
210 mp *a = *(mp **)v[0].buf;
211 mp *n = *(mp **)v[1].buf;
212 mp *r0 = *(mp **)v[2].buf;
213 int exactp0 = *(int *)v[3].buf, exactp;
214 mp *r = mp_nthrt(MP_NEW, a, n, &exactp);
217 if (!MP_EQ(r, r0) || exactp != exactp0) {
219 fputs("\n*** nthrt failed", stderr);
220 fputs("\n*** a = ", stderr); mp_writefile(a, stderr, 10);
221 fputs("\n*** n = ", stderr); mp_writefile(n, stderr, 10);
222 fputs("\n*** r calc = ", stderr); mp_writefile(r, stderr, 10);
223 fprintf(stderr, "\n*** ex calc = %d", exactp);
224 fputs("\n*** r exp = ", stderr); mp_writefile(r0, stderr, 10);
225 fprintf(stderr, "\n*** ex exp = %d", exactp0);
229 mp_drop(a); mp_drop(n); mp_drop(r); mp_drop(r0);
230 assert(mparena_count(MPARENA_GLOBAL) == 0);
235 static int vrf_ppp(dstr *v)
237 mp *a = *(mp **)v[0].buf;
238 int ret0 = *(int *)v[1].buf;
239 mp *x0 = *(mp **)v[2].buf;
240 mp *n0 = *(mp **)v[3].buf;
241 mp *x = MP_NEW, *n = MP_NEW;
242 int ret = mp_perfect_power_p(&x, &n, a);
245 if (ret != ret0 || !MP_EQ(x, x0) || !MP_EQ(n, n0)) {
247 fputs("\n*** perfect_power_p failed", stderr);
248 fputs("\n*** a = ", stderr); mp_writefile(a, stderr, 10);
249 fprintf(stderr, "\n*** r calc = %d", ret);
250 fputs("\n*** x calc = ", stderr); mp_writefile(x, stderr, 10);
251 fputs("\n*** n calc = ", stderr); mp_writefile(n, stderr, 10);
252 fprintf(stderr, "\n*** r exp = %d", ret0);
253 fputs("\n*** x exp = ", stderr); mp_writefile(x0, stderr, 10);
254 fputs("\n*** n exp = ", stderr); mp_writefile(n0, stderr, 10);
258 mp_drop(a); mp_drop(x); mp_drop(n); mp_drop(x0); mp_drop(n0);
259 assert(mparena_count(MPARENA_GLOBAL) == 0);
264 static test_chunk tests[] = {
265 { "nthrt", vrf_nthrt, { &type_mp, &type_mp, &type_mp, &type_int, 0 } },
266 { "perfect-power-p", vrf_ppp, { &type_mp, &type_int, &type_mp, &type_mp, 0 } },
270 int main(int argc, char *argv[])
273 test_run(argc, argv, tests, SRCDIR "/t/mp");
279 /*----- That's all, folks -------------------------------------------------*/