Vector spaces

Arjen Markus (7 september 2005) Inspired by Lars Hellstrom's "essay" On mathematical notation, I sat down last night and concocted a straightforward implementation of vector spaces in Tcl. The idea is described very clearly by Lars, so I will only explain a few implementation details.

First: the code below only builds procedures that takes vectors (lists) of whatever data object the underlying field represents. You can not easily do combinations of scalar and vector arguments this way (for instance, the underlying field may use a list to represent the "number", and that is indistinguishable by [list] from a list of such numbers - you could do that if there is an "isscalar" function ...

Second: th code assumes that all operations act on individual components of the vectors or two individual components from two vectors. There is no such thing as matrix multiplication or a dot product in the code.

Third: the code uses the very important fact that [namespace] and [proc] are completely ordinary commands in Tcl. You can therefore use them inside another procedure to dynamically create procedures and namespaces of your liking.

 # vectorspace.tcl --
 #     Define vector spaces over arbitrary fields
 #     Notes:
 #     Procedures of one and two arguments only are expanded
 #     No provision for scalar-vector operations or for reducing
 #     operations such as summing all components. All operations
 #     are considered to be component-by-component.

 # vectorspace --
 #     Set up the vector space of given dimension over a given field
 # Arguments:
 #     name      Name of the vector space
 #     field     Name of the algebraic construct (a field or some other
 #               construct like a ring)
 #     dim       Number of dimensions
 # Result:
 #     Name of the new vector space
 # Note:
 #     The field is represented by a namespace containing one or
 #     more exported procedures of one or two arguments. The
 #     vector space is represented by procedures based on these
 #     procedures that are expanded to take care of vector-valued
 #     arguments
 proc vectorspace {name field dim} {

     # Check that the namespace "field" exists, then construct
     # a namespace whose procedures are vector-based extensions
     # of the original
     if { ! [namespace exists $field] } {
         error "Field $field is unknown"

     namespace eval $name {}

     # A constructor procedure
     proc ${name}::vector {args} [string map [list DIM $dim] {
         if { [llength $args] != DIM } {
             error "Wrong number of arguments"
         return $args
     }] ;# End string map

     foreach op [namespace eval $field {namespace export}] {
         set args_op [namespace eval $field [list info args $op]]
         switch -- [llength $args_op] {
         "1" {
             namespace eval $name [list namespace export $op]
             proc ${name}::$op {v} [string map [list FIELD $field OP $op DIM $dim] {
                 set r {}
                 if { [llength $v] != DIM } {
                     error "Argument has wrong dimension"
                 foreach c $v {
                     lappend r [FIELD::OP $c]
                 return $r
             }] ;# End string map
         "2" {
             namespace eval $name [list namespace export $op]
             proc ${name}::$op {u v} [string map [list FIELD $field OP $op DIM $dim] {
                 set r {}
                 if { [llength $u] != DIM || [llength $v] != DIM } {
                     error "At least one argument has wrong dimension"
                 foreach c $u d $v {
                     lappend r [FIELD::OP $c $d]
                 return $r
             }] ;# End string map
         "default" {
             # Ignore

     return $name

 # Create a few functions to test this idea
 namespace eval float {
     proc cos {x}   {expr {cos($x)}}
     proc +   {x y} {expr {$x+$y}}
     proc -   {x y} {expr {$x-$y}}
     proc *   {x y} {expr {$x*$y}}
     namespace export cos + - *

 # main --
 #     Test the vectorspace procedure
 #     1. Create a three-dimensional vector space
 #     2. Create a space of 3x3 matrices
 vectorspace R^3 float 3
 vectorspace R3x3 R^3 3

 puts "3D vectors:"
 set a [R^3::vector 1 2 3]
 set b [R^3::vector 4 1 2]
 set c [R^3::vector 2 5 1]
 puts "Sum:        [R^3::+ $a $b]"
 puts "Difference: [R^3::- $a $b]"
 puts "Cosine:     [R^3::cos $c]"

 puts "3x3 matrices:"
 set A [R3x3::vector $a $b $c]
 set B [R3x3::vector $c $a $b]
 set C [R3x3::vector $b $c $a]
 puts "Sum:        [R3x3::+ $A $B]"
 puts "Difference: [R3x3::- $A $B]"
 puts "Cosine:     [R3x3::cos $C]"