void
_acb_poly_interpolate_fast_precomp(acb_ptr poly,
acb_srcptr ys, acb_ptr * tree, acb_srcptr weights,
slong len, slong prec)
{
acb_ptr t, u, pa, pb;
slong i, pow, left;
if (len == 0)
return;
t = _acb_vec_init(len);
u = _acb_vec_init(len);
for (i = 0; i < len; i++)
acb_mul(poly + i, weights + i, ys + i, prec);
for (i = 0; i < FLINT_CLOG2(len); i++)
{
pow = (WORD(1) << i);
pa = tree[i];
pb = poly;
left = len;
while (left >= 2 * pow)
{
_acb_poly_mul(t, pa, pow + 1, pb + pow, pow, prec);
_acb_poly_mul(u, pa + pow + 1, pow + 1, pb, pow, prec);
_acb_vec_add(pb, t, u, 2 * pow, prec);
left -= 2 * pow;
pa += 2 * pow + 2;
pb += 2 * pow;
}
if (left > pow)
{
_acb_poly_mul(t, pa, pow + 1, pb + pow, left - pow, prec);
_acb_poly_mul(u, pb, pow, pa + pow + 1, left - pow + 1, prec);
_acb_vec_add(pb, t, u, left, prec);
}
}
_acb_vec_clear(t, len);
_acb_vec_clear(u, len);
}
void
_acb_poly_powsum_one_series_sieved(acb_ptr z, const acb_t s, slong n, slong len, slong prec)
{
slong * divisors;
slong powers_alloc;
slong i, j, k, ibound, kprev, power_of_two, horner_point;
int critical_line, integer;
acb_ptr powers;
acb_ptr t, u, x;
acb_ptr p1, p2;
arb_t logk, v, w;
critical_line = arb_is_exact(acb_realref(s)) &&
(arf_cmp_2exp_si(arb_midref(acb_realref(s)), -1) == 0);
integer = arb_is_zero(acb_imagref(s)) && arb_is_int(acb_realref(s));
divisors = flint_calloc(n / 2 + 1, sizeof(slong));
powers_alloc = (n / 6 + 1) * len;
powers = _acb_vec_init(powers_alloc);
ibound = n_sqrt(n);
for (i = 3; i <= ibound; i += 2)
if (DIVISOR(i) == 0)
for (j = i * i; j <= n; j += 2 * i)
DIVISOR(j) = i;
t = _acb_vec_init(len);
u = _acb_vec_init(len);
x = _acb_vec_init(len);
arb_init(logk);
arb_init(v);
arb_init(w);
power_of_two = 1;
while (power_of_two * 2 <= n)
power_of_two *= 2;
horner_point = n / power_of_two;
_acb_vec_zero(z, len);
kprev = 0;
COMPUTE_POWER(x, 2, kprev);
for (k = 1; k <= n; k += 2)
{
/* t = k^(-s) */
if (DIVISOR(k) == 0)
{
COMPUTE_POWER(t, k, kprev);
}
else
{
p1 = POWER(DIVISOR(k));
p2 = POWER(k / DIVISOR(k));
if (len == 1)
acb_mul(t, p1, p2, prec);
else
_acb_poly_mullow(t, p1, len, p2, len, len, prec);
}
if (k * 3 <= n)
_acb_vec_set(POWER(k), t, len);
_acb_vec_add(u, u, t, len, prec);
while (k == horner_point && power_of_two != 1)
{
_acb_poly_mullow(t, z, len, x, len, len, prec);
_acb_vec_add(z, t, u, len, prec);
power_of_two /= 2;
horner_point = n / power_of_two;
horner_point -= (horner_point % 2 == 0);
}
}
_acb_poly_mullow(t, z, len, x, len, len, prec);
_acb_vec_add(z, t, u, len, prec);
flint_free(divisors);
_acb_vec_clear(powers, powers_alloc);
_acb_vec_clear(t, len);
_acb_vec_clear(u, len);
_acb_vec_clear(x, len);
arb_clear(logk);
arb_clear(v);
arb_clear(w);
}
void
_acb_poly_zeta_cpx_reflect(acb_ptr t, const acb_t h, const acb_t a, int deflate, slong len, slong prec)
{
/* use reflection formula */
if (arf_sgn(arb_midref(acb_realref(h))) < 0 && acb_is_one(a))
{
/* zeta(s) = (2*pi)**s * sin(pi*s/2) / pi * gamma(1-s) * zeta(1-s) */
acb_t pi, hcopy;
acb_ptr f, s1, s2, s3, s4, u;
slong i;
acb_init(pi);
acb_init(hcopy);
f = _acb_vec_init(2);
s1 = _acb_vec_init(len);
s2 = _acb_vec_init(len);
s3 = _acb_vec_init(len);
s4 = _acb_vec_init(len);
u = _acb_vec_init(len);
acb_set(hcopy, h);
acb_const_pi(pi, prec);
/* s1 = (2*pi)**s */
acb_mul_2exp_si(pi, pi, 1);
_acb_poly_pow_cpx(s1, pi, h, len, prec);
acb_mul_2exp_si(pi, pi, -1);
/* s2 = sin(pi*s/2) / pi */
acb_set(f, h);
acb_one(f + 1);
acb_mul_2exp_si(f, f, -1);
acb_mul_2exp_si(f + 1, f + 1, -1);
_acb_poly_sin_pi_series(s2, f, 2, len, prec);
_acb_vec_scalar_div(s2, s2, len, pi, prec);
/* s3 = gamma(1-s) */
acb_sub_ui(f, hcopy, 1, prec);
acb_neg(f, f);
acb_set_si(f + 1, -1);
_acb_poly_gamma_series(s3, f, 2, len, prec);
/* s4 = zeta(1-s) */
acb_sub_ui(f, hcopy, 1, prec);
acb_neg(f, f);
_acb_poly_zeta_cpx_series(s4, f, a, 0, len, prec);
for (i = 1; i < len; i += 2)
acb_neg(s4 + i, s4 + i);
_acb_poly_mullow(u, s1, len, s2, len, len, prec);
_acb_poly_mullow(s1, s3, len, s4, len, len, prec);
_acb_poly_mullow(t, u, len, s1, len, len, prec);
/* add 1/(1-(s+t)) = 1/(1-s) + t/(1-s)^2 + ... */
if (deflate)
{
acb_sub_ui(u, hcopy, 1, prec);
acb_neg(u, u);
acb_inv(u, u, prec);
for (i = 1; i < len; i++)
acb_mul(u + i, u + i - 1, u, prec);
_acb_vec_add(t, t, u, len, prec);
}
acb_clear(pi);
acb_clear(hcopy);
_acb_vec_clear(f, 2);
_acb_vec_clear(s1, len);
_acb_vec_clear(s2, len);
_acb_vec_clear(s3, len);
_acb_vec_clear(s4, len);
_acb_vec_clear(u, len);
}
else
{
_acb_poly_zeta_cpx_series(t, h, a, deflate, len, prec);
}
}
void
_acb_poly_zeta_em_tail_naive(acb_ptr sum, const acb_t s, const acb_t Na, acb_srcptr Nasx, slong M, slong d, slong prec)
{
acb_ptr u, term;
acb_t Na2, splus, rec;
arb_t x;
fmpz_t c;
int aint;
slong r;
BERNOULLI_ENSURE_CACHED(2 * M);
u = _acb_vec_init(d);
term = _acb_vec_init(d);
acb_init(splus);
acb_init(rec);
acb_init(Na2);
arb_init(x);
fmpz_init(c);
_acb_vec_zero(sum, d);
/* u = 1/2 * Nasx */
_acb_vec_scalar_mul_2exp_si(u, Nasx, d, -WORD(1));
/* term = u * (s+x) / (N+a) */
_acb_poly_mullow_cpx(u, u, d, s, d, prec);
_acb_vec_scalar_div(term, u, d, Na, prec);
/* (N+a)^2 or 1/(N+a)^2 */
acb_mul(Na2, Na, Na, prec);
aint = acb_is_int(Na2);
if (!aint)
acb_inv(Na2, Na2, prec);
for (r = 1; r <= M; r++)
{
/* flint_printf("sum 2: %wd %wd\n", r, M); */
/* sum += bernoulli number * term */
arb_set_round_fmpz(x, fmpq_numref(bernoulli_cache + 2 * r), prec);
arb_div_fmpz(x, x, fmpq_denref(bernoulli_cache + 2 * r), prec);
_acb_vec_scalar_mul_arb(u, term, d, x, prec);
_acb_vec_add(sum, sum, u, d, prec);
/* multiply term by ((s+x)+2r-1)((s+x)+2r) / ((N+a)^2 * (2*r+1)*(2*r+2)) */
acb_set(splus, s);
arb_add_ui(acb_realref(splus), acb_realref(splus), 2*r-1, prec);
_acb_poly_mullow_cpx(term, term, d, splus, d, prec);
arb_add_ui(acb_realref(splus), acb_realref(splus), 1, prec);
_acb_poly_mullow_cpx(term, term, d, splus, d, prec);
/* TODO: combine with previous multiplication? */
if (aint)
{
arb_mul_ui(x, acb_realref(Na2), 2*r+1, prec);
arb_mul_ui(x, x, 2*r+2, prec);
_acb_vec_scalar_div_arb(term, term, d, x, prec);
}
else
{
fmpz_set_ui(c, 2*r+1);
fmpz_mul_ui(c, c, 2*r+2);
acb_div_fmpz(rec, Na2, c, prec);
_acb_vec_scalar_mul(term, term, d, rec, prec);
}
}
_acb_vec_clear(u, d);
_acb_vec_clear(term, d);
acb_clear(splus);
acb_clear(rec);
acb_clear(Na2);
arb_clear(x);
fmpz_clear(c);
}
void
_acb_poly_zeta_em_sum(acb_ptr z, const acb_t s, const acb_t a, int deflate, ulong N, ulong M, slong d, slong prec)
{
acb_ptr t, u, v, term, sum;
acb_t Na, one;
slong i;
t = _acb_vec_init(d + 1);
u = _acb_vec_init(d);
v = _acb_vec_init(d);
term = _acb_vec_init(d);
sum = _acb_vec_init(d);
acb_init(Na);
acb_init(one);
prec += 2 * (FLINT_BIT_COUNT(N) + FLINT_BIT_COUNT(d));
acb_one(one);
/* sum 1/(k+a)^(s+x) */
if (acb_is_one(a) && d <= 3)
_acb_poly_powsum_one_series_sieved(sum, s, N, d, prec);
else if (N > 50 && flint_get_num_threads() > 1)
_acb_poly_powsum_series_naive_threaded(sum, s, a, one, N, d, prec);
else
_acb_poly_powsum_series_naive(sum, s, a, one, N, d, prec);
/* t = 1/(N+a)^(s+x); we might need one extra term for deflation */
acb_add_ui(Na, a, N, prec);
_acb_poly_acb_invpow_cpx(t, Na, s, d + 1, prec);
/* sum += (N+a) * 1/((s+x)-1) * t */
if (!deflate)
{
/* u = (N+a)^(1-(s+x)) */
acb_sub_ui(v, s, 1, prec);
_acb_poly_acb_invpow_cpx(u, Na, v, d, prec);
/* divide by 1/((s-1) + x) */
acb_sub_ui(v, s, 1, prec);
acb_div(u, u, v, prec);
for (i = 1; i < d; i++)
{
acb_sub(u + i, u + i, u + i - 1, prec);
acb_div(u + i, u + i, v, prec);
}
_acb_vec_add(sum, sum, u, d, prec);
}
/* sum += ((N+a)^(1-(s+x)) - 1) / ((s+x) - 1) */
else
{
/* at s = 1, this becomes (N*t - 1)/x, i.e. just remove one coeff */
if (acb_is_one(s))
{
for (i = 0; i < d; i++)
acb_mul(u + i, t + i + 1, Na, prec);
_acb_vec_add(sum, sum, u, d, prec);
}
else
{
/* TODO: this is numerically unstable for large derivatives,
and divides by zero if s contains 1. We want a good
way to evaluate the power series ((N+a)^y - 1) / y where y has
nonzero constant term, without doing a division.
How is this best done? */
_acb_vec_scalar_mul(t, t, d, Na, prec);
acb_sub_ui(t + 0, t + 0, 1, prec);
acb_sub_ui(u + 0, s, 1, prec);
acb_inv(u + 0, u + 0, prec);
for (i = 1; i < d; i++)
acb_mul(u + i, u + i - 1, u + 0, prec);
for (i = 1; i < d; i += 2)
acb_neg(u + i, u + i);
_acb_poly_mullow(v, u, d, t, d, d, prec);
_acb_vec_add(sum, sum, v, d, prec);
_acb_poly_acb_invpow_cpx(t, Na, s, d, prec);
}
}
/* sum += u = 1/2 * t */
_acb_vec_scalar_mul_2exp_si(u, t, d, -WORD(1));
_acb_vec_add(sum, sum, u, d, prec);
/* Euler-Maclaurin formula tail */
if (d < 5 || d < M / 10)
_acb_poly_zeta_em_tail_naive(u, s, Na, t, M, d, prec);
else
_acb_poly_zeta_em_tail_bsplit(u, s, Na, t, M, d, prec);
_acb_vec_add(z, sum, u, d, prec);
_acb_vec_clear(t, d + 1);
_acb_vec_clear(u, d);
_acb_vec_clear(v, d);
_acb_vec_clear(term, d);
_acb_vec_clear(sum, d);
acb_clear(Na);
acb_clear(one);
}
void
_acb_poly_sin_cos_series_tangent(acb_ptr s, acb_ptr c,
const acb_srcptr h, slong hlen, slong len, slong prec, int times_pi)
{
acb_ptr t, u, v;
acb_t s0, c0;
hlen = FLINT_MIN(hlen, len);
if (hlen == 1)
{
if (times_pi)
acb_sin_cos_pi(s, c, h, prec);
else
acb_sin_cos(s, c, h, prec);
_acb_vec_zero(s + 1, len - 1);
_acb_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)
*/
acb_init(s0);
acb_init(c0);
t = _acb_vec_init(3 * len);
u = t + len;
v = u + len;
/* sin, cos of h0 */
if (times_pi)
acb_sin_cos_pi(s0, c0, h, prec);
else
acb_sin_cos(s0, c0, h, prec);
/* t = tan((h-h0)/2) */
acb_zero(u);
_acb_vec_scalar_mul_2exp_si(u + 1, h + 1, hlen - 1, -1);
if (times_pi)
{
acb_const_pi(t, prec);
_acb_vec_scalar_mul(u + 1, u + 1, hlen - 1, t, prec);
}
_acb_poly_tan_series(t, u, hlen, len, prec);
/* v = 1 + t^2 */
_acb_poly_mullow(v, t, len, t, len, len, prec);
acb_add_ui(v, v, 1, prec);
/* u = 1/(1+t^2) */
_acb_poly_inv_series(u, v, len, len, prec);
/* sine */
_acb_poly_mullow(s, t, len, u, len, len, prec);
_acb_vec_scalar_mul_2exp_si(s, s, len, 1);
/* cosine */
acb_sub_ui(v, v, 2, prec);
_acb_vec_neg(v, v, len);
_acb_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 (!acb_is_zero(s0))
{
_acb_vec_scalar_mul(t, s, len, c0, prec);
_acb_vec_scalar_mul(u, c, len, s0, prec);
_acb_vec_scalar_mul(v, s, len, s0, prec);
_acb_vec_add(s, t, u, len, prec);
_acb_vec_scalar_mul(t, c, len, c0, prec);
_acb_vec_sub(c, t, v, len, prec);
}
_acb_vec_clear(t, 3 * len);
acb_clear(s0);
acb_clear(c0);
}