Types in Nemo

On this page we discuss the type hierarchy in Nemo and a concept known as parents. These details are quite technical and should be skipped or skimmed by new users of Julia/Nemo. Types are almost never dealt with directly when scripting Nemo to do mathematical computations.

In contrast, Nemo developers will certainly want to know how we model mathematical objects and the rings, fields, groups, etc. that they belong to in Nemo.

Introduction

Julia provides two levels of types that we make use of

  • abstract types
  • concrete types

Concrete types are just like the usual types everyone is familiar with from C or C++.

Abstract types can be thought of as collections of types. They are used when writing generic functions that should work for any type in the given collection.

To write a generic function that accepts any type in a given collection of types, we first create an abstract type. Then we create the individual concrete types that belong to that abstract type. A generic function can then be constructed with a type parameter, T say, similar to a template parameter in C++. The main difference is that we can specify which abstract type our type parameter T must belong to.

We use the symbol <: in Julia to determine that a given type belongs to a given abstract type. For example the built-in Julia type Int64 for 64 bit machine integers belongs to the Julia abstract type Integer. Thus Int <: Integer returns true.

Here is some Julia code illustrating this with a more complex example. We create an abstract type called Shape and two user defined concrete types square and circle belonging to Shape. We then show how to write methods that accept each of the concrete types and then show how to write a generic function for any type T belonging to the abstract type Shape.

Note that in the type definitions of square and circle we specify that those types belong to the abstract type Shape using the <: operator.

abstract Shape

type square <: Shape
   width::Int
   border_thickness::Int
end

type circle <: Shape
   centre::Tuple{Int, Int}
   radius::Int
   border_thickness::Int
end

function area(s::square)
   return s.width^2
end

function area(s::circle)
   return pi*s.radius^2
end

function border_thickness{T <: Shape}(s::T)
   return s.border_thickness
end

s = square(3, 1)
c = circle((3, 4), 2, 2)

area(s)
area(c)
border_thickness(s)
border_thickness(c)

The abstract type hierarchy in Nemo

Abstract types in Julia can also belong to one another in a hierarchy. For example, the Nemo.Field abstract type belongs to the Nemo.Ring abstract type. An object representing a field in Nemo has type belonging to Nemo.Field. But because we define the inclusion Nemo.Field <: Nemo.Ring in Nemo, the type of such an object also automatically belongs to Nemo.Ring. This means that any generic function in Nemo which is designed to work with ring objects will certainly also work with field objects.

In Nemo we also distinguish between the elements of a field, say, and the field itself, and similarly for groups and rings and all other kinds of domains in Nemo. For example, we have an object of type GenPolyRing to model a generic polynomial ring, and elements of that polynomial ring would have type GenPoly.

In order to model this distinction between elements and the domains they belong to, Nemo has two main branches in its abstract type hierarchy, as shown in the following diagram. One branch consists of the abstract types for the domains available in Nemo and the other branch is for the abstract types for elements of those domains.

alt text

All objects in Nemo, whether they represent rings, fields, groups, sets, etc. on the one hand, or ring elements, field elements, etc. on the other hand, have concrete types that belong to one of the abstract types shown above.

Why types aren't enough

Naively, one may expect that rings in Nemo can be modeled as types and their elements as objects with the given type. But there are various reasons why this is not a good model.

As an example, consider the ring for a multiprecision integer . If we were to model the ring as a type, then the type would somehow need to contain the modulus . This is not possible in Julia, and in fact it is not desirable either.

Julia dispatches on type, and each time we call a generic function with different types, a new version of the function is compiled at runtime for performance. But this would be a disaster if we were writing a multimodular algorithm, say. In such an algorithm many rings would be needed and every function we use would be recompiled over and over for each different . This would result in a huge delay as the compiler is invoked many times.

For this reason, the modulus needs to be attached to the elements of the ring, not to type associated with those elements.

But now we have a problem. How do we create new elements of the ring given only the type? Suppose all rings were represented by the same type Zmod say. How would we create ? We could not write a = Zmod(3) since the modulus is not contained in the type Zmod.

We could of course use the notation a = Zmod(3, 7), but this would make implementation of generic algorithms very difficult, as they would need to distinguish the case where constructors take a single argument, such as a = ZZ(7) and cases where they take a modulus, such as a = Zmod(3, 7).

The way we get around this in Nemo is to have special (singleton) objects that act like types, but are really just ordinary Julia objects. These objects, called parent objects can contain extra information, such as the modulus .

In order to create new elements of as above, we overload the call operator for the parent object, making it callable. Making a parent object callable is exactly analogous to writing a constructor for a type.

In the following Nemo example, we create the parent object R corresponding to the ring . We then create a new element a of this ring by calling the parent object R, just as though R were a type with a constructor accepting an Int parameter.

R = ResidueRing(ZZ, 7)
a = R(3)

This example creates the element .

The important point is that unlike a type, a parent object such as R can contain additional information that a type cannot contain, such as the modulus of the ring in this example, or context objects required by C libraries in other examples.

More complex example of parent objects

Here is some Julia/Nemo code which constructs a polynomial ring over the integers, a polynomial in that ring and then does some introspection to illustrate the various relations between the objects and types.

julia> using Nemo

julia> R, x = ZZ["x"]
(Univariate Polynomial Ring in x over Integer Ring,x)

julia> f = x^2 + 3x + 1
x^2+3*x+1

julia> typeof(R)
Nemo.FmpzPolyRing

julia> typeof(f)
Nemo.fmpz_poly

julia> parent(f)
Univariate Polynomial Ring in x over Integer Ring

julia> typeof(R) <: PolyRing
true

julia> typeof(f) <: PolyElem
true

julia> parent(f) == R
true

Concrete types in Nemo

Finally we come to all the concrete types in Nemo.

These are of two main kinds: those for generic constructions (e.g. generic polynomials over an arbitrary ring) and those for specific implementations, usually provided by a C library (e.g. polynomials over the integers, provided by Flint).

Below we give the type of each kind of element available in Nemo. In parentheses we list the types of their corresponding parent objects. Note that these are the types of the element objects and parent objects respectively, not the abstract types to which these types belong, which the reader can easily guess.

For example, fmpz belongs to the abstract type RingElem and FlintIntegerRing belongs to Ring. Similarly Poly{T} belongs to PolyElem whereas PolynomialRing{T} belongs to PolyRing. We also have that fmpz_poly belongs to PolyElem and FmpzPolyRing belongs to PolyRing, and so on.

All the generic types are parameterised by a type T which is the type of the elements of the ring they are defined over.

  • Generic

    • GenPoly{T} (GenPolyRing{T})
    • GenRelSeries{T} (GenRelSeriesRing{T})
    • GenRes{T} (GenResRing{T})
    • GenFrac{T} (GenFracField{T})
    • GenMat{T} (GenMatSpace{T})
    • Flint

    • fmpz (FlintIntegerRing)

    • fmpq (FlintRationalField)
    • fq_nmod (FqNmodFiniteField)
    • fq (FqFiniteField)
    • padic (FlintPadicField)
    • fmpz_poly (FmpzPolyRing)
    • fmpq_poly (FmpqPolyRing)
    • nmod_poly (NmodPolyRing)
    • fmpz_mod_poly (FmpzModPolyRing)
    • fq_poly (FqPolyRing)
    • fq_nmod_poly (FqNmodPolyRing)
    • fmpz_rel_series (FmpzRelSeriesRing)
    • fmpq_rel_series (FmpqRelSeriesRing)
    • fmpz_mod_rel_series (FmpzModRelSeriesRing)
    • fq_nmod_rel_series (FqNmodRelSeriesRing)
    • fq_rel_series (FqRelSeriesRing)
    • fmpz_mat (FmpzMatSpace)
    • nmod_mat (NmodMatSpace)
    • perm (PermGroup)
    • Antic

    • nf_elem (AnticNumberField)

    • Arb

    • arb (ArbField)

    • acb (AcbField)