Getting started with padic_radix

A padic_radix context is a generic ring (gr_ctx_t) for \mathbf{Q}_p. Its elements are balls p^v u + O(p^N) or exact elements, and every operation computes an error term that is guaranteed to contain the error of the result. The context chooses how much precision results keep: a relative precision r (digits after the leading one), an absolute precision a (digits below p^a), both, or neither.

A first program#

The program works in \mathbf{Q}_7 with relative precision 10 and no absolute bound.

#include <flint/flint.h>
#include <flint/gr.h>
#include <flint/padic_radix.h>

int main(void)
{
    gr_ctx_t ctx;
    padic_radix_t x, y, z;
    int status = GR_SUCCESS;

    /* Q_7, relative precision 10 digits, no absolute precision cap */
    gr_ctx_init_padic_radix(ctx, 7, 10, PADIC_RADIX_PREC_INF, 0);
    gr_ctx_println(ctx);
    padic_radix_init(x, ctx);
    padic_radix_init(y, ctx);
    padic_radix_init(z, ctx);

    status |= padic_radix_set_ui(x, 50, ctx);
    status |= padic_radix_set_ui(y, 1, ctx);
    status |= padic_radix_sub(z, x, y, ctx);       /* exact: 49 */
    flint_printf("50 - 1  = "); gr_println(z, ctx);
    status |= padic_radix_div(z, y, z, ctx);       /* exact: 1/49 */
    flint_printf("1/49    = "); gr_println(z, ctx);

    status |= padic_radix_set_ui(x, 3, ctx);
    status |= padic_radix_inv(x, x, ctx);          /* 1/3 to 10 digits */
    flint_printf("1/3     = "); gr_println(x, ctx);

    status |= padic_radix_set_ui(z, 3, ctx);
    status |= padic_radix_mul(z, z, x, ctx);       /* 3 * (1/3) */
    status |= padic_radix_sub(z, z, y, ctx);       /* ... - 1 */
    flint_printf("3*(1/3) - 1 = "); gr_println(z, ctx);
    flint_printf("is zero: %s\n",
        padic_radix_is_zero(z, ctx) == T_UNKNOWN ? "unknown" : "decided");

    status |= padic_radix_set_ui(x, 2, ctx);
    status |= padic_radix_sqrt(z, x, ctx);         /* 2 = 3^2 mod 7 */
    flint_printf("sqrt(2) = "); gr_println(z, ctx);
    status |= padic_radix_set_ui(x, 7, ctx);
    status |= padic_radix_exp(z, x, ctx);
    status |= padic_radix_log(y, z, ctx);
    flint_printf("log(exp(7)) = "); gr_println(y, ctx);
    flint_printf("status = %d\n", status);

    padic_radix_clear(x, ctx);
    padic_radix_clear(y, ctx);
    padic_radix_clear(z, ctx);
    gr_ctx_clear(ctx);
    return 0;
}

Radix 7-adic numbers (rel prec 10, abs prec inf)
50 - 1  = (1) * 7^2
1/49    = (1) * 7^-2
1/3     = 188316833 + O(7^10)
3*(1/3) - 1 = 0 + O(7^10)
is zero: unknown
sqrt(2) = 266983762 + O(7^10)
log(exp(7)) = (1) * 7^1 + O(7^10)
status = 0

The integers 50 and 1 fit in 10 digits, so they and their difference 49 = 7^2 are exact, and so is 1/49, whose expansion terminates. The inverse of 3 has an infinite expansion and is truncated to 10 digits, which the error term O(7^{10}) records. Multiplying it by 3 and subtracting 1 cancels every known digit: the result is 0 + O(7^{10}), a ball that contains zero and nonzero numbers, so padic_radix_is_zero answers T_UNKNOWN, and dividing by it would return GR_UNABLE. In padic the same computation would return an exact-looking zero. The square root of 2 and the exponential and logarithm are computed to the relative precision of the context.

Choosing the precision#

In every configuration, inexact inputs limit the precision of the result further, as in the example. The precision of a result can be read with padic_radix_get_error, and padic_radix_is_exact tells whether it is exact.

Status codes#

The element functions return GR_SUCCESS, or GR_DOMAIN when the operation is undefined (division by an exact zero, the square root of a non-square, the logarithm of an element not congruent to 1 modulo p), or GR_UNABLE when the answer is not determined by the inputs (division by a ball containing zero) or cannot be represented (an infinite expansion in an exact context, valuations beyond PADIC_RADIX_ERR_MAX). Generic code should check the status as for any other ring; GR_MUST_SUCCEED aborts on failure.

Generic structures#

Since the context is a gr_ctx_t, the generic modules work over it: gr_poly for polynomials and power series arithmetic, gr_mat for matrices (products use the dot products of this module, determinants the generic algorithms for fields), and gr_series for power series with error tracking in both the p-adic and the x-adic direction. Compared with padic_poly and padic_mat, every coefficient has its own valuation and precision instead of sharing one, and precision loss in division and elimination is detected.