void
_arb_poly_sinh_cosh_series_exponential(arb_ptr s, arb_ptr c,
const arb_srcptr h, slong hlen, slong len, slong prec)
{
arb_ptr t, u, v;
arb_t s0, c0;
hlen = FLINT_MIN(hlen, len);
if (hlen == 1)
{
arb_sinh_cosh(s, c, h, prec);
_arb_vec_zero(s + 1, len - 1);
_arb_vec_zero(c + 1, len - 1);
return;
}
arb_init(s0);
arb_init(c0);
t = _arb_vec_init(3 * len);
u = t + len;
v = u + len;
arb_sinh_cosh(s0, c0, h, prec);
_arb_vec_set(t + 1, h + 1, hlen - 1);
_arb_poly_exp_series(t, t, len, len, prec);
/* todo: part of the inverse could be avoided since exp computes
it internally to half the length */
_arb_poly_inv_series(u, t, len, len, prec);
/* hyperbolic sine */
_arb_vec_sub(s, t, u, len, prec);
_arb_vec_scalar_mul_2exp_si(s, s, len, -1);
/* hyperbolic cosine */
_arb_vec_add(c, t, u, len, prec);
_arb_vec_scalar_mul_2exp_si(c, c, len, -1);
/* sinh(h0 + h1) = cosh(h0) sinh(h1) + sinh(h0) cosh(h1)
cosh(h0 + h1) = cosh(h0) cosh(h1) + sinh(h0) sinh(h1) */
if (!arb_is_zero(s0))
{
_arb_vec_scalar_mul(t, s, len, c0, prec);
_arb_vec_scalar_mul(u, c, len, s0, prec);
_arb_vec_scalar_mul(v, s, len, s0, prec);
_arb_vec_add(s, t, u, len, prec);
_arb_vec_scalar_mul(t, c, len, c0, prec);
_arb_vec_add(c, t, v, len, prec);
}
_arb_vec_clear(t, 3 * len);
arb_clear(s0);
arb_clear(c0);
}
void _arb_poly_divrem(arb_ptr Q, arb_ptr R,
arb_srcptr A, slong lenA,
arb_srcptr B, slong lenB, slong prec)
{
const slong lenQ = lenA - lenB + 1;
_arb_poly_div(Q, A, lenA, B, lenB, prec);
if (lenB > 1)
{
if (lenQ >= lenB - 1)
_arb_poly_mullow(R, Q, lenQ, B, lenB - 1, lenB - 1, prec);
else
_arb_poly_mullow(R, B, lenB - 1, Q, lenQ, lenB - 1, prec);
_arb_vec_sub(R, A, R, lenB - 1, prec);
}
}
/* with inverse=1 simultaneously computes g = exp(-x) to length n
with inverse=0 uses g as scratch space, computing
g = exp(-x) only to length (n+1)/2 */
static void
_arb_poly_exp_series_newton(arb_ptr f, arb_ptr g,
arb_srcptr h, slong len, slong prec, int inverse, slong cutoff)
{
slong alloc;
arb_ptr T, U, hprime;
alloc = 3 * len;
T = _arb_vec_init(alloc);
U = T + len;
hprime = U + len;
_arb_poly_derivative(hprime, h, len, prec);
arb_zero(hprime + len - 1);
NEWTON_INIT(cutoff, len)
/* f := exp(h) + O(x^m), g := exp(-h) + O(x^m2) */
NEWTON_BASECASE(n)
_arb_poly_exp_series_basecase(f, h, n, n, prec);
_arb_poly_inv_series(g, f, (n + 1) / 2, (n + 1) / 2, prec);
NEWTON_END_BASECASE
/* extend from length m to length n */
NEWTON_LOOP(m, n)
slong m2 = (m + 1) / 2;
slong l = m - 1; /* shifted for derivative */
/* g := exp(-h) + O(x^m) */
_arb_poly_mullow(T, f, m, g, m2, m, prec);
_arb_poly_mullow(g + m2, g, m2, T + m2, m - m2, m - m2, prec);
_arb_vec_neg(g + m2, g + m2, m - m2);
/* U := h' + g (f' - f h') + O(x^(n-1))
Note: should replace h' by h' mod x^(m-1) */
_arb_vec_zero(f + m, n - m);
_arb_poly_mullow(T, f, n, hprime, n, n, prec); /* should be mulmid */
_arb_poly_derivative(U, f, n, prec);
arb_zero(U + n - 1); /* should skip low terms */
_arb_vec_sub(U + l, U + l, T + l, n - l, prec);
_arb_poly_mullow(T + l, g, n - m, U + l, n - m, n - m, prec);
_arb_vec_add(U + l, hprime + l, T + l, n - m, prec);
/* f := f + f * (h - int U) + O(x^n) = exp(h) + O(x^n) */
_arb_poly_integral(U, U, n, prec); /* should skip low terms */
_arb_vec_sub(U + m, h + m, U + m, n - m, prec);
_arb_poly_mullow(f + m, f, n - m, U + m, n - m, n - m, prec);
/* g := exp(-h) + O(x^n) */
/* not needed if we only want exp(x) */
if (n == len && inverse)
{
_arb_poly_mullow(T, f, n, g, m, n, prec);
_arb_poly_mullow(g + m, g, m, T + m, n - m, n - m, prec);
_arb_vec_neg(g + m, g + m, n - m);
}
NEWTON_END_LOOP
NEWTON_END
_arb_vec_clear(T, alloc);
}
void
_acb_poly_mullow_transpose_gauss(acb_ptr res,
acb_srcptr poly1, slong len1,
acb_srcptr poly2, slong len2, slong n, slong prec)
{
arb_ptr a, b, c, d, e, f, w;
arb_ptr t, u, v;
slong i;
len1 = FLINT_MIN(len1, n);
len2 = FLINT_MIN(len2, n);
w = flint_malloc(sizeof(arb_struct) * (2 * (len1 + len2 + n)));
a = w;
b = a + len1;
c = b + len1;
d = c + len2;
e = d + len2;
f = e + n;
t = _arb_vec_init(n);
u = _arb_vec_init(n);
v = _arb_vec_init(n);
for (i = 0; i < len1; i++)
{
a[i] = *acb_realref(poly1 + i);
b[i] = *acb_imagref(poly1 + i);
}
for (i = 0; i < len2; i++)
{
c[i] = *acb_realref(poly2 + i);
d[i] = *acb_imagref(poly2 + i);
}
for (i = 0; i < n; i++)
{
e[i] = *acb_realref(res + i);
f[i] = *acb_imagref(res + i);
}
_arb_vec_add(t, a, b, len1, prec);
_arb_vec_add(u, c, d, len2, prec);
_arb_poly_mullow(v, t, len1, u, len2, n, prec);
_arb_poly_mullow(t, a, len1, c, len2, n, prec);
_arb_poly_mullow(u, b, len1, d, len2, n, prec);
_arb_vec_sub(e, t, u, n, prec);
_arb_vec_sub(f, v, t, n, prec);
_arb_vec_sub(f, f, u, n, prec);
for (i = 0; i < n; i++)
{
*acb_realref(res + i) = e[i];
*acb_imagref(res + i) = f[i];
}
_arb_vec_clear(t, n);
_arb_vec_clear(u, n);
_arb_vec_clear(v, n);
flint_free(w);
}
void
_arb_poly_lgamma_series(arb_ptr res, arb_srcptr h, slong hlen, slong len, slong prec)
{
int reflect;
slong r, n, wp;
arb_t zr;
arb_ptr t, u;
if (!arb_is_positive(h))
{
_arb_vec_indeterminate(res, len);
return;
}
hlen = FLINT_MIN(hlen, len);
wp = prec + FLINT_BIT_COUNT(prec);
t = _arb_vec_init(len);
u = _arb_vec_init(len);
arb_init(zr);
/* use zeta values at small integers */
if (arb_is_int(h) && (arf_cmpabs_ui(arb_midref(h), prec / 2) < 0))
{
r = arf_get_si(arb_midref(h), ARF_RND_DOWN);
if (r <= 0)
{
_arb_vec_indeterminate(res, len);
goto cleanup;
}
else
{
_arb_poly_lgamma_series_at_one(u, len, wp);
if (r != 1)
{
arb_one(zr);
_log_rising_ui_series(t, zr, r - 1, len, wp);
_arb_vec_add(u, u, t, len, wp);
}
}
}
else if (len <= 2)
{
arb_lgamma(u, h, wp);
if (len == 2)
arb_digamma(u + 1, h, wp);
}
else
{
/* otherwise use Stirling series */
arb_gamma_stirling_choose_param(&reflect, &r, &n, h, 0, 0, wp);
arb_add_ui(zr, h, r, wp);
_arb_poly_gamma_stirling_eval(u, zr, n, len, wp);
if (r != 0)
{
_log_rising_ui_series(t, h, r, len, wp);
_arb_vec_sub(u, u, t, len, wp);
}
}
/* compose with nonconstant part */
arb_zero(t);
_arb_vec_set(t + 1, h + 1, hlen - 1);
_arb_poly_compose_series(res, u, len, t, hlen, len, prec);
cleanup:
arb_clear(zr);
_arb_vec_clear(t, len);
_arb_vec_clear(u, len);
}
void
_arb_bell_sum_taylor(arb_t res, const fmpz_t n,
const fmpz_t a, const fmpz_t b, const fmpz_t mmag, long tol)
{
fmpz_t m, r, R, tmp;
mag_t B, C, D, bound;
arb_t t, u;
long wp, k, N;
if (_fmpz_sub_small(b, a) < 5)
{
arb_bell_sum_bsplit(res, n, a, b, mmag, tol);
return;
}
fmpz_init(m);
fmpz_init(r);
fmpz_init(R);
fmpz_init(tmp);
/* r = max(m - a, b - m) */
/* m = a + (b - a) / 2 */
fmpz_sub(r, b, a);
fmpz_cdiv_q_2exp(r, r, 1);
fmpz_add(m, a, r);
fmpz_mul_2exp(R, r, RADIUS_BITS);
mag_init(B);
mag_init(C);
mag_init(D);
mag_init(bound);
arb_init(t);
arb_init(u);
if (fmpz_cmp(R, m) >= 0)
{
mag_inf(C);
mag_inf(D);
}
else
{
/* C = exp(R * |F'(m)| + (1/2) R^2 * (n/(m-R)^2 + 1/(m-R))) */
/* C = exp(R * (|F'(m)| + (1/2) R * (n/(m-R) + 1)/(m-R))) */
/* D = (1/2) R * (n/(m-R) + 1)/(m-R) */
fmpz_sub(tmp, m, R);
mag_set_fmpz(D, n);
mag_div_fmpz(D, D, tmp);
mag_one(C);
mag_add(D, D, C);
mag_div_fmpz(D, D, tmp);
mag_mul_fmpz(D, D, R);
mag_mul_2exp_si(D, D, -1);
/* C = |F'(m)| */
wp = 20 + 1.05 * fmpz_bits(n);
arb_set_fmpz(t, n);
arb_div_fmpz(t, t, m, wp);
fmpz_add_ui(tmp, m, 1);
arb_set_fmpz(u, tmp);
arb_digamma(u, u, wp);
arb_sub(t, t, u, wp);
arb_get_mag(C, t);
/* C = exp(R * (C + D)) */
mag_add(C, C, D);
mag_mul_fmpz(C, C, R);
mag_exp(C, C);
}
if (mag_cmp_2exp_si(C, tol / 4 + 2) > 0)
{
_arb_bell_sum_taylor(res, n, a, m, mmag, tol);
_arb_bell_sum_taylor(t, n, m, b, mmag, tol);
arb_add(res, res, t, 2 * tol);
}
else
{
arb_ptr mx, ser1, ser2, ser3;
/* D = T(m) */
wp = 20 + 1.05 * fmpz_bits(n);
arb_set_fmpz(t, m);
arb_pow_fmpz(t, t, n, wp);
fmpz_add_ui(tmp, m, 1);
arb_gamma_fmpz(u, tmp, wp);
arb_div(t, t, u, wp);
arb_get_mag(D, t);
/* error bound: (b-a) * C * D * B^N / (1 - B), B = r/R */
/* ((b-a) * C * D * 2) * 2^(-N*RADIUS_BITS) */
/* ((b-a) * C * D * 2) */
mag_mul(bound, C, D);
mag_mul_2exp_si(bound, bound, 1);
fmpz_sub(tmp, b, a);
mag_mul_fmpz(bound, bound, tmp);
/* N = (tol + log2((b-a)*C*D*2) - mmag) / RADIUS_BITS */
if (mmag == NULL)
{
/* estimate D ~= 2^mmag */
fmpz_add_ui(tmp, MAG_EXPREF(C), tol);
fmpz_cdiv_q_ui(tmp, tmp, RADIUS_BITS);
}
else
{
fmpz_sub(tmp, MAG_EXPREF(bound), mmag);
fmpz_add_ui(tmp, tmp, tol);
fmpz_cdiv_q_ui(tmp, tmp, RADIUS_BITS);
}
if (fmpz_cmp_ui(tmp, 5 * tol / 4) > 0)
N = 5 * tol / 4;
else if (fmpz_cmp_ui(tmp, 2) < 0)
N = 2;
else
N = fmpz_get_ui(tmp);
/* multiply by 2^(-N*RADIUS_BITS) */
mag_mul_2exp_si(bound, bound, -N * RADIUS_BITS);
mx = _arb_vec_init(2);
ser1 = _arb_vec_init(N);
ser2 = _arb_vec_init(N);
ser3 = _arb_vec_init(N);
/* estimate (this should work for moderate n and tol) */
wp = 1.1 * tol + 1.05 * fmpz_bits(n) + 5;
/* increase precision until convergence */
while (1)
{
/* (m+x)^n / gamma(m+1+x) */
arb_set_fmpz(mx, m);
arb_one(mx + 1);
_arb_poly_log_series(ser1, mx, 2, N, wp);
for (k = 0; k < N; k++)
arb_mul_fmpz(ser1 + k, ser1 + k, n, wp);
arb_add_ui(mx, mx, 1, wp);
_arb_poly_lgamma_series(ser2, mx, 2, N, wp);
_arb_vec_sub(ser1, ser1, ser2, N, wp);
_arb_poly_exp_series(ser3, ser1, N, N, wp);
/* t = a - m, u = b - m */
arb_set_fmpz(t, a);
arb_sub_fmpz(t, t, m, wp);
arb_set_fmpz(u, b);
arb_sub_fmpz(u, u, m, wp);
arb_power_sum_vec(ser1, t, u, N, wp);
arb_zero(res);
for (k = 0; k < N; k++)
arb_addmul(res, ser3 + k, ser1 + k, wp);
if (mmag != NULL)
{
if (_fmpz_sub_small(MAG_EXPREF(arb_radref(res)), mmag) <= -tol)
break;
}
else
{
if (arb_rel_accuracy_bits(res) >= tol)
break;
}
wp = 2 * wp;
}
/* add the series truncation bound */
arb_add_error_mag(res, bound);
_arb_vec_clear(mx, 2);
_arb_vec_clear(ser1, N);
_arb_vec_clear(ser2, N);
_arb_vec_clear(ser3, N);
}
mag_clear(B);
mag_clear(C);
mag_clear(D);
mag_clear(bound);
arb_clear(t);
arb_clear(u);
fmpz_clear(m);
fmpz_clear(r);
fmpz_clear(R);
fmpz_clear(tmp);
}
void
_arb_poly_sin_cos_series_tangent(arb_ptr s, arb_ptr c,
arb_srcptr h, slong hlen, slong len, slong prec, int times_pi)
{
arb_ptr t, u, v;
arb_t s0, c0;
hlen = FLINT_MIN(hlen, len);
if (hlen == 1)
{
if (times_pi)
arb_sin_cos_pi(s, c, h, prec);
else
arb_sin_cos(s, c, h, prec);
_arb_vec_zero(s + 1, len - 1);
_arb_vec_zero(c + 1, len - 1);
return;
}
/*
sin(x) = 2*tan(x/2)/(1+tan(x/2)^2)
cos(x) = (1-tan(x/2)^2)/(1+tan(x/2)^2)
*/
arb_init(s0);
arb_init(c0);
t = _arb_vec_init(3 * len);
u = t + len;
v = u + len;
/* sin, cos of h0 */
if (times_pi)
arb_sin_cos_pi(s0, c0, h, prec);
else
arb_sin_cos(s0, c0, h, prec);
/* t = tan((h-h0)/2) */
arb_zero(u);
_arb_vec_scalar_mul_2exp_si(u + 1, h + 1, hlen - 1, -1);
if (times_pi)
{
arb_const_pi(t, prec);
_arb_vec_scalar_mul(u + 1, u + 1, hlen - 1, t, prec);
}
_arb_poly_tan_series(t, u, hlen, len, prec);
/* v = 1 + t^2 */
_arb_poly_mullow(v, t, len, t, len, len, prec);
arb_add_ui(v, v, 1, prec);
/* u = 1/(1+t^2) */
_arb_poly_inv_series(u, v, len, len, prec);
/* sine */
_arb_poly_mullow(s, t, len, u, len, len, prec);
_arb_vec_scalar_mul_2exp_si(s, s, len, 1);
/* cosine */
arb_sub_ui(v, v, 2, prec);
_arb_vec_neg(v, v, len);
_arb_poly_mullow(c, v, len, u, len, len, prec);
/* sin(h0 + h1) = cos(h0) sin(h1) + sin(h0) cos(h1)
cos(h0 + h1) = cos(h0) cos(h1) - sin(h0) sin(h1) */
if (!arb_is_zero(s0))
{
_arb_vec_scalar_mul(t, s, len, c0, prec);
_arb_vec_scalar_mul(u, c, len, s0, prec);
_arb_vec_scalar_mul(v, s, len, s0, prec);
_arb_vec_add(s, t, u, len, prec);
_arb_vec_scalar_mul(t, c, len, c0, prec);
_arb_vec_sub(c, t, v, len, prec);
}
_arb_vec_clear(t, 3 * len);
arb_clear(s0);
arb_clear(c0);
}
void
_acb_poly_mullow_transpose(acb_ptr res,
acb_srcptr poly1, slong len1,
acb_srcptr poly2, slong len2, slong n, slong prec)
{
arb_ptr a, b, c, d, e, f, w;
arb_ptr t;
slong i;
len1 = FLINT_MIN(len1, n);
len2 = FLINT_MIN(len2, n);
w = flint_malloc(sizeof(arb_struct) * (2 * (len1 + len2 + n)));
a = w;
b = a + len1;
c = b + len1;
d = c + len2;
e = d + len2;
f = e + n;
/* (e+fi) = (a+bi)(c+di) = (ac - bd) + (ad + bc)i */
t = _arb_vec_init(n);
for (i = 0; i < len1; i++)
{
a[i] = *acb_realref(poly1 + i);
b[i] = *acb_imagref(poly1 + i);
}
for (i = 0; i < len2; i++)
{
c[i] = *acb_realref(poly2 + i);
d[i] = *acb_imagref(poly2 + i);
}
for (i = 0; i < n; i++)
{
e[i] = *acb_realref(res + i);
f[i] = *acb_imagref(res + i);
}
_arb_poly_mullow(e, a, len1, c, len2, n, prec);
_arb_poly_mullow(t, b, len1, d, len2, n, prec);
_arb_vec_sub(e, e, t, n, prec);
_arb_poly_mullow(f, a, len1, d, len2, n, prec);
/* squaring */
if (poly1 == poly2 && len1 == len2)
{
_arb_vec_scalar_mul_2exp_si(f, f, n, 1);
}
else
{
_arb_poly_mullow(t, b, len1, c, len2, n, prec);
_arb_vec_add(f, f, t, n, prec);
}
for (i = 0; i < n; i++)
{
*acb_realref(res + i) = e[i];
*acb_imagref(res + i) = f[i];
}
_arb_vec_clear(t, n);
flint_free(w);
}
