FractionField

© 2005 John Abbott
GNU Free Documentation License, Version 1.2



CoCoALib Documentation Index

User documentation for FractionField

A FractionField in the CoCoA Library is a fraction field of an effective GCD domain. A FractionField offers all the operations of a ring plus a few new specific operations.

To create a new FractionField from a GCDDomain use NewFractionField:

   FractionField QR = NewFractionField(R);

This creates a new object of type FractionField whose elements are formal fractions of elements of R. CoCoA::FractionField is derived from CoCoA::ring, so a FractionField is a ring.

Given a FractionField the function BaseRing tells us which ring it is the FractionField of:

   R == BaseRing(QR); // yields true

BaseRing(QR) produces an identical copy of R (not merely an isomorphic ring).

Given just a plain ring S we can enquire whether S is in fact a FractionField, and if so view it as such:

    if (IsFractionField(S))
    {
       FractionField FrF = AsFractionField(S);
       ... code using FrF ...
    }

Calling AsFractionField(S) when IsFractionField(S) is false will throw an exception of type CoCoAError with code ERR::NotFracField

Let r be a RingElem belonging to a FractionField QR. Then

Note that the numerator and denominator are defined only upto multiples by a unit: so it is possible to have two elements of QR which are equal but which have different numerators and denominators, for instance, (x-1)/(x-2) = (1-x)/(2-x)

Homomorphisms

Maintainer documentation for FractionField, FractionFieldBase, FractionFieldImpl

The class FractionField is wholly analogous to the class ring, i.e. a reference counting smart pointer. The only difference is that FractionField knows that the myRingPtr data member actually points to an instance of a class derived from FractionFieldBase (and so can safely apply a static_cast when the pointer value is accessed).

FractionFieldBase is an abstract class derived from RingBase. It adds a few pure virtual functions to those contained in RingBase:

  virtual const ring& myBaseRing() const;
  virtual void myGetNum(RawPtr rawn, ConstRawPtr rawq) const; // n is elem of BaseRing
  virtual void myGetDen(RawPtr rawd, ConstRawPtr rawq) const; // d is elem of BaseRing
  virtual const RingHom& myEmbeddingHom() const;
  virtual RingHom myInducedHomCtor(const RingHom& phi) const;

myBaseRing returns a reference to the ring (effective GCD domain) supplied to the constructor.

num (resp. den) produces a copy of the numerator (resp. denominator) of q: NOTE that the first arg belongs to BaseRing and NOT to the FractionField! EmbeddingHom returns the embedding homomorphism from the BaseRing into the FractionField; it actually returns a reference to a fixed homomorpism held internallly. InducedHom creates a new homomorpism from the FractionField to another ring S given a homomorpism from the BaseRing to S.

FractionFieldImpl implements a general fraction field. Its elements are just pairs of RawValues belonging to the BaseRing (see the struct FractionFieldElem). For this implementation the emphasis was clearly on simplicity over speed (at least partly because we do not expect FractionFieldImpl to be used much). For polynomials whose coefficients lie in a FractionField we plan to implement a specific ring which uses a common denominator representation for the whole polynomial. If you want to make this code faster, see some of the comments in the bugs section.

Important: while fractions are guaranteed to be reduced (i.e. no common factor exists between numerator and denominator), it is rather hard to ensure that they are canonical since in general we can multiply numerator and denominator by any unit. See a bug comment about normalizing units.

Bugs, Shortcomings and other ideas

The functions myNew are not exception safe: memory would be leaked if space for the numerator were successfully allocated while allocation for the denominator failed -- nobody would clean up the resources consumed by the numerator. Probably need a sort of auto_ptr for holding temporary bits of a value.

I don't like the names myGetNum and myGetDen. Hope to change them soon.

Should partial homomorphisms be allowed: e.g. from Q to Z/(3)? Mathematically it is a bit dodgy, but in practice all works out fine provided you don't divide by zero. I think it would be too useful (e.g. for chinese remaindering methods) to be banned. Given phi:Z->Z[x] it might be risky to induce Q->Z[x]; note that ker(phi)=0, so this is not a sufficient criterion!

In fact one could create a FractionFieldImpl of any integral domain (it just wouldn't be possible to cancel factors without a GCD). I'll wait until someone really needs it before allowing it.

It is not clear how to make the denominator positive when the GCD domain is Z (so the fraction field is Q). In general we would need the GCD domain to supply a normalizing unit: such a function could return 1 unless we have some special desire to normalize the denominator in a particular way. HERE'S A CONUNDRUM: FractionField(Q[x]) -- all rationals are units, and so we could end up with horrible representations like (22/7)/(22/7) instead of just 1. MUST FIX THIS!!

The printing function is TERRIBLE!

FASTER + and -
Addition and subtraction can be done better: let h be the GCD of the two denominators, suppose we want to compute a/bh + c/dh (where gcd(a,bh) = gcd(c, dh) = gcd(b,d) = 1 i.e. h = gcd(B,D) where B,D are the denoms) If h = 1 then there is no cancellation, o/w gcd(ad+bc, bdh) = gcd(ad+bc, h), so we can use a simpler gcd computation to find the common factor.

FASTER * and /
Multiplication and division can also be made a little faster by simplifying the GCD computation a bit. The two cases are essentially the same, so I shall consider just multiplication. Assuming inputs are already reduced (i.e. there is no common factor between numerator and denominator). To compute (a/b)*(c/d), first calculate h1 = gcd(a, d) and h2 = gcd(b, c). The result is then: num = (a/h1)*(c/h2) & den = (b/h2)*(d/h1) and this is guaranteed to be in reduced form.

myIsRational is incomplete: it will fail to recognise rationals whose numer and denom have been multiplied by non-trivial units. BAD BUG! Ironically myIsInteger does work correctly.