HaskellForMaths-0.1: Math/Algebra/LinearAlgebra.hs
-- Copyright (c) David Amos, 2008. All rights reserved.
{-# OPTIONS_GHC -fglasgow-exts #-}
module Math.Algebra.LinearAlgebra where
import qualified Data.List as L
import Math.Algebra.Field.Base -- not actually used in this module
infixr 8 *>, *>>
infixr 7 <<*>
infixl 7 <.>, <*>, <<*>>, <*>>
infixl 6 <+>, <->, <<+>>, <<->>
-- The mnemonic for these operations is that the number of angle brackets on each side indicates the dimension of the argument on that side
u <+> v = zipWith (+) u v
u <-> v = zipWith (-) u v
-- scalar multiplication
k *> v = map (k*) v
k *>> m = (map . map) (k*) m
-- dot product of vectors (also called inner or scalar product)
u <.> v = sum (zipWith (*) u v)
-- tensor product of vectors (also called outer or matrix product)
u <*> v = [ [a*b | b <- v] | a <- u]
-- matrix operations
a <<+>> b = (zipWith . zipWith) (+) a b
a <<->> b = (zipWith . zipWith) (-) a b
a <<*>> b = [ [u <.> v | v <- L.transpose b] | u <- a]
-- action on the left
m <<*> v = map (<.> v) m
-- action on the right
v <*>> m = map (v <.>) (L.transpose m)
fMatrix n f = [[f i j | j <- [1..n]] | i <- [1..n]]
-- version with indices from zero
fMatrix' n f = [[f i j | j <- [0..n-1]] | i <- [0..n-1]]
-- idMx n = fMatrix n (\i j -> if i == j then 1 else 0)
idMx n = idMxs !! n where
idMxs = map snd $ iterate next (0,[])
next (j,m) = (j+1, (1 : replicate j 0) : map (0:) m)
jMx n = replicate n (replicate n 1)
zMx n = replicate n (replicate n 0)
{-
-- VECTORS
data Vector d k = V [k] deriving (Eq,Ord,Show)
instance (IntegerAsType d, Num k) => Num (Vector d k) where
V a + V b = V $ a <+> b
V a - V b = V $ a <-> b
negate (V a) = V $ map negate a
fromInteger 0 = V $ replicate d' 0 where d' = fromInteger $ value (undefined :: d)
V v <>> M m = V $ v <*>> m
M m <<> V v = V $ m <<*> v
k |> V v = V $ k *> v
-}
-- MATRICES
{-
-- Square matrices of dimension d over field k
data Matrix d k = M [[k]] deriving (Eq,Ord,Show)
instance (IntegerAsType d, Num k) => Num (Matrix d k) where
M a + M b = M $ a <<+>> b
M a - M b = M $ a <<->> b
negate (M a) = M $ (map . map) negate a
M a * M b = M $ a <<*>> b
fromInteger 0 = M $ zMx d' where d' = fromInteger $ value (undefined :: d)
fromInteger 1 = M $ idMx d' where d' = fromInteger $ value (undefined :: d)
instance (IntegerAsType d, Fractional a) => Fractional (Matrix d a) where
recip (M a) = case inverse a of
Nothing -> error "Matrix.recip: matrix is singular"
Just a' -> M a'
-}
inverse m =
let d = length m -- the dimension
i = idMx d
m' = zipWith (++) m i
i1 = inverse1 m'
i2 = inverse2 i1
in if length i1 == d
then Just i2
else Nothing
-- given (M|I), use row operations to get to (U|A), where U is upper triangular with 1s on diagonal
inverse1 [] = []
inverse1 ((x:xs):rs) =
if x /= 0
then let r' = (1/x) *> xs
in (1:r') : inverse1 [ys <-> y *> r' | (y:ys) <- rs]
else case filter (\r' -> head r' /= 0) rs of
[] -> [] -- early termination, which will be detected in calling function
r:_ -> inverse1 (((x:xs) <+> r) : rs)
-- This is basically row echelon form
-- given (U|A), use row operations to get to M^-1
inverse2 [] = []
inverse2 ((1:r):rs) = inverse2' r rs : inverse2 rs where
inverse2' xs [] = xs
inverse2' (x:xs) ((1:r):rs) = inverse2' (xs <-> x *> r) rs
-- This is basically reduced row echelon form
xs ! i = xs !! (i-1) -- ie, a 1-based list lookup instead of 0-based
rowEchelonForm [] = []
rowEchelonForm ((x:xs):rs) =
if x /= 0
then let r' = (1/x) *> xs
in (1:r') : map (0:) (rowEchelonForm [ys <-> y *> r' | (y:ys) <- rs])
else case filter (\r' -> head r' /= 0) rs of
[] -> map (0:) (rowEchelonForm $ xs : map tail rs)
r:_ -> rowEchelonForm (((x:xs) <+> r) : rs)
rowEchelonForm zs@([]:_) = zs
reducedRowEchelonForm m = reverse $ reduce $ reverse $ rowEchelonForm m where
reduce (r:rs) = let r':rs' = reduceStep (r:rs) in r' : reduce rs' -- is this scanl or similar?
reduce [] = []
reduceStep ((1:xs):rs) = (1:xs) : [ 0: (ys <-> y *> xs) | y:ys <- rs]
reduceStep rs@((0:_):_) = zipWith (:) (map head rs) (reduceStep $ map tail rs)
reduceStep rs = rs
-- kernel of a matrix
-- returns basis for vectors v s.t m <<*> v == 0
kernel m = kernelRRE $ reducedRowEchelonForm m
kernelRRE m =
let nc = length $ head m -- the number of columns
is = findLeadingCols 1 (L.transpose m) -- these are the indices of the columns which have a leading 1
js = [1..nc] L.\\ is
freeCols = let m' = take (length is) m -- discard zero rows
in zip is $ L.transpose [map (negate . (!j)) m' | j <- js]
boundCols = zip js (idMx $ length js)
in L.transpose $ map snd $ L.sort $ freeCols ++ boundCols
where
findLeadingCols i (c@(1:_):cs) = i : findLeadingCols (i+1) (map tail cs)
findLeadingCols i (c@(0:_):cs) = findLeadingCols (i+1) cs
findLeadingCols _ _ = []
m ^- n = recip m ^ n
-- t (M m) = M (L.transpose m)
det [[x]] = x
det ((x:xs):rs) =
if x /= 0
then let r' = (1/x) *> xs
in x * det [ys <-> y *> r' | (y:ys) <- rs]
else case filter (\r' -> head r' /= 0) rs of
[] -> 0
r:_ -> det (((x:xs) <+> r) : rs)
{-
class IntegerAsType a where
value :: a -> Integer
data Z
instance IntegerAsType Z where
value _ = 0
data S a
instance IntegerAsType a => IntegerAsType (S a) where
value _ = value (undefined :: a) + 1
-}