Function: idealred
Section: number_fields
C-Name: idealred0
Prototype: GGDG
Help: idealred(nf,I,{v=0}): LLL reduction of the ideal I in the number
 field nf along direction v, in HNF.
Doc: \idx{LLL} reduction of
 the ideal $I$ in the number field $K$ attached to \var{nf}, along the
 direction $v$. The $v$ parameter is best left omitted, but if it is present,
 it must be an $\kbd{nf.r1} + \kbd{nf.r2}$-component vector of
 \emph{nonnegative} integers. (What counts is the relative magnitude of the
 entries: if all entries are equal, the effect is the same as if the vector
 had been omitted.)

 This function finds an $a\in K^{*}$ such that $J = (a)I$ is
 ``small'' and integral (see the end for technical details).
 The result is the Hermite normal form of
 the ``reduced'' ideal $J$.
 \bprog
 ? K = nfinit(y^2+1);
 ? P = idealprimedec(K,5)[1];
 ? idealred(K, P)
 %3 =
 [1 0]

 [0 1]
 @eprog\noindent More often than not, a \idx{principal ideal} yields the unit
 ideal as above. This is a quick and dirty way to check if ideals are principal,
 but it is not a necessary condition: a nontrivial result does not prove that
 the ideal is nonprincipal. For guaranteed results, see \kbd{bnfisprincipal},
 which requires the computation of a full \kbd{bnf} structure.

 If the input is an extended ideal $[I,s]$, the output is $[J, sa]$; in
 this way, one keeps track of the principal ideal part:
 \bprog
 ? idealred(K, [P, 1])
 %5 = [[1, 0; 0, 1], [2, -1]~]
 @eprog\noindent
 meaning that $P$ is generated by $[2, -1]~$. The number field element in the
 extended part is an algebraic number in any form \emph{or} a factorization
 matrix (in terms of number field elements, not ideals!). In the latter case,
 elements stay in factored form, which is a convenient way to avoid
 coefficient explosion; see also \tet{idealpow}.

 \misctitle{Technical note} The routine computes an LLL-reduced
 basis for the lattice $I^{-1}$ equipped with the quadratic
 form
 $$|| x ||_{v}^{2} = \sum_{i=1}^{r_{1}+r_{2}}
   2^{v_{i}}\varepsilon_{i}|\sigma_{i}(x)|^{2},$$
 where as usual the $\sigma_{i}$ are the (real and) complex embeddings and
 $\varepsilon_{i} = 1$, resp.~$2$, for a real, resp.~complex place. The element
 $a$ is simply the first vector in the LLL basis. The only reason you may want
 to try to change some directions and set some $v_{i}\neq 0$ is to randomize
 the elements found for a fixed ideal, which is heuristically useful in index
 calculus algorithms like \tet{bnfinit} and \tet{bnfisprincipal}.

 \misctitle{Even more technical note} In fact, the above is a white lie.
 We do not use $||\cdot||_{v}$ exactly but a rescaled rounded variant which
 gets us faster and simpler LLLs. There's no harm since we are not using any
 theoretical property of $a$ after all, except that it belongs to $I^{-1}$
 and that $a I$ is ``expected to be small''.
