Types

Introduction

The axis language is strongly and statically typed: the interpreter first analyses the expressions and definitions given by the user to verify that types can be attributed to all subexpressions, and all operations are defined for the types they are applied to; only if the expression passes this test does the system attempt to evaluate the expression. Thus many computations that would produce problems on execution are signaled. For instance, after loading “groups.at” which defined GL: (int->RootDatum), atlas will complain about:

atlas> block_sizes (GL(5))
Type error:
  Subexpression GL(5) at <standard input>:...
  has wrong type: found RootDatum while InnerClass was needed.
Type check failed

because the built-in function block_sizes requires an InnerClass value but GL produces a value of type RootDatum. In the following example:

atlas> Lie_type(Cartan_matrix(GL(5)))
Error in expression Lie_type(Cartan_matrix(GL(5))) at...
  Failed to match 'Lie_type' with argument type mat
Type check failed

the message is somewhat different, because the name Lie_type corresponds to several functions, and atlas can only report that none of them applies for the given expression. To find out which instances are known, one can enter:

atlas> whattype Lie_type ?
Overloaded instances of Lie_type
string->LieType
RootDatum->LieType
(int,int)->LieType

(note the question mark, which here means treat Lie_type as possibly overloaded function name rather than as a variable) showing that either a string or a RootDatum value would have worked instead of a (Cartan) matrix. For instance:

atlas> Lie_type(GL(5))
Value: Lie type 'A4.T1'
atlas> Lie_type("E6.T1.D4")
Value: Lie type 'E6.T1.D4'

All built-in operators and functions, and normally also all user defined ones, are defined as “overloaded” symbols (even if only one meaning is built-in, as is the case for block_sizes). This allows the user to add new definitions of those symbols without overriding those built in. On the other hand, the values of all identifiers that are not functions are stored in a different table, allowing only one such value at a time to be associated to a given identifier.

Primitive Types

The interpreter knows about several “primitive” types, which it distinguishes but for which (in contrast to “composite” types discussed below) it usually does not provide specific language constructions (although it does know for instance that a Boolean value can be tested in a conditional expression). These basic types are:

Primitive Type Represent
bool truth values
int machine integers (32 or 64 bits)
rat rational numbers (quotient of two machine integers)
string string of characters
vec vector of machine integers
mat matrix of machine integers
ratvec rational vector (vector numerator with common denominator)
LieType Lie type
RootDatum root datum, specifying a connected complex reductive group
InnerClass inner class of real forms (based root datum with involution)
RealForm real form within an inner class
CartanClass all the conjugacy classes of Cartan subgroups of real groups in an inner class
KGBElt element of the set K\G/B associated to some RealForm value
Block block for a pair of RealForm values (at dual inner classes)
Split “split integer” a + b.s where s is “split unit” with s^2=1
Param value representing a standard module or its irreducible quotient
ParamPol virtual module with signature (Param values with Split coefs)
T we use T to represent any type

Note

If you want to check the data type of something, for example id_mat(3). You can do whattype id_mat(3) in atlas and it will output type: mat.

bool

bool represents truth values. Values of bool are true, false:

atlas> whattype true
type: bool
atlas> whattype false
type: bool

string

string represents string of characters. String values can be entered using a double quote:

atlas> set s = "atlas"
Variable s: string

Strings denotations contain newline characters (but a constant ‘new_line’ is provided, containing a string with a single newline character). Values of other basic types can only be obtained by using appropriate operators and functions (for instance 22/7 has type ‘rat’ and GL(5) has type ‘RootDatum’), or sometimes via implicit conversions.

int, rat, & vec

  • int represents machine integers (32 or 64 bits);
  • rat represents rational numbers (quotient of two machine integers);
  • vec represents vector of machine integers.

As you might expect, the sum of a int type and rat type is a rat type:

atlas> set a = 2
Variable a: int
atlas> set b = 1/2
Variable b: rat
atlas> whattype a+b
type: rat

Similarly, if you add a vec type and an array of int, whenever possible, the result is of type vec:

atlas> set v = vec:[1,2,3]
Variable v: vec
atlas> set w = [3,4,5]
Variable w: [int]
atlas> whattype v+w
type: vec

mat

mat represents matrix of machine integers.

If you directly enter [[1,2],[3,4]], the type would be set to [[int]]. You need to specifically declare the type mat if you want [[1,2],[3,4]] to be a matrix:

atlas> set m = mat : [[1,2],[3,4]]
Variable m: mat

Identity matrix of dimension n is a build-in expression. Suppose \(n=3\):

atlas> id_mat(3)
Value:
| 1, 0, 0 |
| 0, 1, 0 |
| 0, 0, 1 |

ratvec

ratvec represents rational vector (vector numerator with common denominator).

There are two basic ways to declare a rational vector:

atlas> set v = [1,2,3]/5
Variable v: ratvec
atlas> set w = ratvec:[1/2,3/5]
Variable w: ratvec

You can also make array of rational numbers [rat]:

atlas> set w = [1/2,3/5]
Variable w: [rat]

Similar to int, if you add a ratvec to [rat], the result is ratvec:

atlas> set v = [1,2,3]/5
Variable v: ratvec
atlas> set w = [1/2,3/5, 5/7]
Variable w: [rat]
atlas> whattype v+w
type: ratvec

LieType

LieType represents Lie types.

An example of a valid Lie type is “A1.T1”:

atlas> set l = LieType : "A1.T1"
Variable l: LieType

RootDatum

RootDatum represents root datum, specifying a connected complex reductive group.

In atlas, a root datum is a pair of \(m\times n\) (integral) matrices \((A,B)\) such that \(A^T*B\) is a Cartan matrix. The number m is rank (number of rows) and n is the semi-simple rank (number of columns). One way to define a RootDatum is to use LieType:

atlas> set rd =  simply_connected(LieType:"A1.T1")
Variable rd: RootDatum

InnerClass

InnerClass represents inner class of real forms (based root datum with involution).

One can think of an inner class as a set of real forms (of a certain complex Lie group) that share some properties. One can define an inner class in atlas as:

atlas> inner_class(SL(2,R))
Value: Complex reductive group of type A1, with involution defining
inner class of type 'c', with 2 real forms and 2 dual real forms
atlas> whattype inner_class(SL(2,R))
type: InnerClass

RealForm

RealForm represents real form within an inner class.

A simple way of specifying a real form is:

atlas> set G = Sp(4,R)
Variable G: RealForm

This is enabled by the various user-defined scripts in “atlas-scripts” folder.

If furthermore you want to see all real forms that are in the same inner class as \(Sp(4,R)\), do:

atlas> real_forms(G)
Value: [compact connected real group with Lie algebra 'sp(2)',
connected real group with Lie algebra 'sp(1,1)',
connected split real group with Lie algebra 'sp(4,R)']

CartanClass

CartanClass represents all the conjugacy classes of Cartan subgroups of real groups in an inner class.

For a specific real group \(G = Sp(4,R)\), one can ask atlas what are the Cartan classes that are in the same inner class:

atlas> set G = Sp(4,R)
Variable G: RealForm
atlas>
atlas> Cartan_classes(G)
Value: [Cartan class #0, occurring for 3 real forms and for 1 dual real form,
Cartan class #1, occurring for 2 real forms and for 1 dual real form,
Cartan class #2, occurring for 1 real form and for 2 dual real forms,
Cartan class #3, occurring for 1 real form and for 3 dual real forms]
atlas>
atlas> whattype Cartan_classes(G)[1]
type: CartanClass

The reference to ‘dual real forms’ concerns the various blocks of representations for which each Cartan class appears.

KGBElt

KGBElt represents element of the set \(K\backslash G/B\) associated to some RealForm value.

Given a group \(G\), for example \(G = SL(2,R)\). One can ask atlas to print out the KGB elements associated to different Cartan involutions:

atlas> set G = SL(2,R)
Variable G: RealForm (overriding previous instance, which had type RealForm)
atlas> KGB(G)
Value: [KGB element #0,KGB element #1,KGB element #2]
atlas> KGB(G,0)
Value: KGB element #0

Block

Block represents block for a pair of RealForm values (at dual inner classes).

atlas> set G = SL(2,R)
Variable G: RealForm (overriding previous instance, which had type RealForm)
atlas> blocks(G)
Value: [Block of 1 elements,Block of 1 elements,Block of 3 elements]
atlas> whattype blocks(G)[0]
type: Block

Split

Split represents “split integer” \(a + b.s\) where \(s\) is “split unit” with \(s^2=1\).

Param

Param represents value representing a standard module or its irreducible quotient.

ParamPol

ParamPol represents virtual module with signature (Param values with Split coefs).

Composite Types

Composite types are either array (list) types, tuple types or function types. Array and tuple types both construct aggregates by combining a sequence of component values. The difference is:

  • for array types all components must have the same type and there could be any number of them (including none at all);
  • for tuple types the type explicitly enumerates the (may-be-different) types of the components, so in particular the number of components is determined by the type.

A function type specifies zero or more argument and result types; for either, unless exactly one such type is specified, the argument or result type is actually a tuple type. Thus if t0,t1,t2,t3 are types, one has composite types like:

Composite Types Represents
[t0] array of elements all of which have type t0
[[t0]] array of elements all of which have type [t0] (a list of lists)
etc
(t0,t1) 2-tuple formed of components of types t0 and t1 respectively
(t0,t1,t2) 3-tuple, with components of types t0,t1,t2 respectively
(t0,t1,t2,t3) 4-tuple, with components of types t0,t1,t2,t3 respectively
etc
void 0-tuple (irrelevant value)
(t0->t1) function with argument of type t0 and result of type t1
(t0,t1->t2) function with argument of type (t0,t1) and result of type t2
(t0->t1,t2) function with argument of type t0 and result of type (t1,t2)
(t0,t1->t2,t3) function with argument of type (t0,t1), result of type (t2,t3)
(t0,t1->) function with argument of type (t0,t1) and no useful result
(->t0) function with 0 arguments with result of type t0
etc

Often the user does not have to write any types, and the system will take care of deriving primitive and composite types as implicitly specified by the expression. However when writing user defined functions, the types of the arguments must be specified, so that types can be checked for the function body; once this check succeeds, a type is attributed to the function, and it will henceforth be treated just like a built-in function of that type would be (and one can in fact for instance form an array that contains both built-in and user-defined functions, provided they all have the same (function) type).