### 3 Polynomials

#### 3.1 The Floats pseudo-field

Polynomials with floating-point coefficients may be manipulated in **GAP**; though they behave, in subtle ways, quite differently than polynomials over rings.

The "pseudo-field" of floating-point numbers is an object in **GAP**, called `FLOAT_PSEUDOFIELD`

. (It is not really a field, e.g. because addition of floating-point numbers in not associative). It may be used to create indeterminates, for example as

gap> x := Indeterminate(FLOAT_PSEUDOFIELD,"x");
x
gap> 2*x^2+3;
2.0*x^2+3.0
gap> Value(last,10);
203.0

#### 3.2 Roots of polynomials

The Jenkins-Traub algorithm has been implemented, in arbitrary precision for MPFR and MPC.

Furthermore, CXSC can provide complex enclosures for the roots of a complex polynomial.

#### 3.3 Finding integer relations

The PSLQ algorithm has been implemented by Steve A. Linton, as an external contribution to **Float**. This algorithm receives as input a vector of floats x and a required precision ϵ, and seeks an integer vector v such that |x⋅ v|<ϵ. The implementation follows quite closely the original article [BB01].

##### 3.3-1 PSLQ

`‣ PSLQ` ( `x`, `epsilon`[, `gamma`] ) | ( function ) |

`‣ PSLQ_MP` ( `x`, `epsilon`[, `gamma`[, `beta`]] ) | ( function ) |

Returns: An integer vector v with |x⋅ v|<ϵ.

The PSLQ algorithm by Bailey and Broadhurst (see [BB01]) searches for an integer relation between the entries in x.

β and γ are algorithm tuning parameters, and default to 4/10 and 2/sqrt(3) respectively.

The second form implements the "Multi-pair" variant of the algorithm, which is better suited to parallelization.

gap> PSLQ([1.0,(1+Sqrt(5.0))/2],1.e-2);
[ 55, -34 ] # Fibonacci numbers
gap> RootsFloat([1,-4,2]*1.0);
[ 0.292893, 1.70711 ] # roots of 2x^2-4x+1
gap> PSLQ(List([0..2],i->last[1]^i),1.e-7);
[ 1, -4, 2 ] # a degree-2 polynomial fitting well

#### 3.4 LLL lattice reduction

A faster implementation of the LLL lattice reduction algorithm has also been implemented. It is accessible via the commands `FPLLLReducedBasis(m)`

and `FPLLLShortestVector(m)`

.