File r38/doc/manual2/linalg.tex artifact c6f4a8fa55 part of check-in trunk


\chapter{LINALG: Linear algebra package}
\label{LINALG}
\typeout{{LINALG: Linear algebra package}}

{\footnotesize
\begin{center}
Matt Rebbeck \\
Konrad--Zuse--Zentrum f\"ur Informationstechnik Berlin \\
Takustra\"se 7 \\
D--14195 Berlin--Dahlem, Germany \\[0.05in]
\end{center}
}
\ttindex{LINALG}

\section{Introduction}

This package provides a selection of functions that are useful
in the world of linear algebra.  They can be classified into four
sections:

\subsection{Basic matrix handling}

\begin{center}
\begin{tabular}{l l l l}
add\_columns\ttindex{ADD\_COLUMNS}           &
add\_rows\ttindex{ADD\_ROWS}                 &
add\_to\_columns\ttindex{ADD\_TO\_COLUMNS}   &
add\_to\_rows\ttindex{ADD\_TO\_ROWS}         \\
augment\_columns\ttindex{AUGMENT\_COLUMNS}   &
char\_poly\ttindex{CHAR\_POLY}               &
column\_dim\ttindex{COLUMN\_DIM}             &
copy\_into\ttindex{COPY\_INTO}               \\
diagonal\ttindex{DIAGONAL}                   &
extend\ttindex{EXTEND}                       &
find\_companion\ttindex{FIND\_COMPANION}     &
get\_columns\ttindex{GET\_COLUMNS}           \\
get\_rows\ttindex{GET\_ROWS}                 &
hermitian\_tp\ttindex{HERMITIAN\_TP}         &
matrix\_augment\ttindex{MATRIX\_AUGMENT}     &
matrix\_stack\ttindex{MATRIX\_STACK}         \\
minor\ttindex{MINOR}                         &
mult\_columns\ttindex{MULT\_COLUMNS}         &
mult\_rows\ttindex{MULT\_ROWS}               &
pivot\ttindex{PIVOT}                         \\
remove\_columns\ttindex{REMOVE\_COLUMNS}     &
remove\_rows\ttindex{REMOVE\_ROWS}           &
row\_dim\ttindex{ROW\_DIM}                   &
rows\_pivot\ttindex{ROWS\_PIVOT}             \\
stack\_rows\ttindex{STACK\_ROWS}             &
sub\_matrix\ttindex{SUB\_MATRIX}             &
swap\_columns\ttindex{SWAP\_COLUMNS}         &
swap\_entries\ttindex{SWAP\_ENTRIES}         \\
swap\_rows\ttindex{SWAP\_ROWS}         & & &
\end{tabular}
\end{center}

\subsection{Constructors}

Functions that create matrices.

\begin{center}
\begin{tabular}{l l l l}
band\_matrix\ttindex{BAND\_MATRIX}        &
block\_matrix\ttindex{BLOCK\_MATRIX}       &
char\_matrix\ttindex{CHAR\_MATRIX}        &
coeff\_matrix\ttindex{COEFF\_MATRIX}       \\
companion\ttindex{COMPANION}           &
hessian\ttindex{HESSIAN}             &
hilbert\ttindex{HILBERT}             &
jacobian\ttindex{JACOBIAN}            \\
jordan\_block\ttindex{JORDAN\_BLOCK}       &
make\_identity\ttindex{MAKE\_IDENTITY}      &
random\_matrix\ttindex{RANDOM\_MATRIX}      &
toeplitz\ttindex{TOEPLITZ}            \\
vandermonde\ttindex{VANDERMONDE}         &
Kronecker\_Product\ttindex{KRONECKER\_PRODUCT}  &
\end{tabular}
\end{center}

\subsection{High level algorithms}

\begin{center}
\begin{tabular}{l l l l}
char\_poly\ttindex{CHAR\_POLY}        &
cholesky\ttindex{CHOLESKY}          &
gram\_schmidt\ttindex{GRAM\_SCHMIDT}     &
lu\_decom\ttindex{LU\_DECOM}         \\
pseudo\_inverse\ttindex{PSEUDO\_INVERSE}   &
simplex\ttindex{SIMPLEX}           &
svd\ttindex{SVD}               &
triang\_adjoint\ttindex{TRIANG\_ADJOINT} \\
\end{tabular}
\end{center}

\vspace*{5mm}
There is a separate {\small NORMFORM} package (chapter~\ref{NORMFORM})
for computing the matrix normal forms smithex, smithex\_int,
frobenius, ratjordan, jordansymbolic and jordan in \REDUCE. 

\subsection{Predicates}

\begin{center}
\begin{tabular}{l l l}
matrixp\ttindex{MATRIXP}      &
squarep\ttindex{SQUAREP}      &
symmetricp\ttindex{SYMMETRICP}
\end{tabular}
\end{center}

\section{Explanations}

In the examples the matrix ${\cal A}$ will be 

\begin{flushleft}
\begin{math}
{\cal A} = \left( \begin{array}{ccc} 1 & 2 & 3 \\ 4 & 5 & 6 \\ 7 & 8 & 9
\end{array} \right)
\end{math}
\end{flushleft}

Throughout ${\cal I}$ is used to indicate the identity matrix and 
${\cal A}^T$ to indicate the transpose of the matrix ${\cal A}$.

Many of the functions have a fairly obvious meaning.  Others need a
little explanation.  

\section{Basic matrix handling}

The functions \f{ADD\_COLUMNS}\ttindex{ADD\_COLUMNS} and \f{ADD\_ROWS}
provide basic operations between rows and columns.  The form is 

\noindent {\tt add\_columns(${\cal A}$,c1,c2,expr);}

and it replaces column c2 of the matix by expr $*$ column(${\cal
A}$,c1) $+$ column(${\cal A}$,c2). 

\f{ADD\_TO\_COLUMNS}\ttindex{ADD\_TO\_COLUMNS} and
\f{ADD\_TO\_ROWS}\ttindex{ADD\_TO\_ROWS} do a similar task, adding an
expression to each of a number of columns (or rows) specified by a
list.

\begin{math}
\begin{array}{ccc}
{\tt add\_to\_columns}({\cal A},\{1,2\},10) & = & 
\left( \begin{array}{ccc} 11 & 12 & 3 \\ 14 & 15 & 6 \\ 17 & 18 & 9 
\end{array} \right)  
\end{array}
\end{math}

The functions \f{MULT\_COLUMNS}\ttindex{MULT\_COLUMNS} and
\f{MULT\_ROW}\ttindex{MULT\_ROW} are equivalent to multiply columns
and rows.


\f{COLUMN\_DIM}\ttindex{COLUMN\_DIM} and
\f{ROW\_DIM}\ttindex{ROW\_DIM} find the column dimension and row
dimension of their argument.

Parts of a matrix can be replaced from another by using
\f{COPY\_INTO}\ttindex{COPY\_INTO}; the last two arguments are row and
column counters for to where to copy the matrix.

\begin{flushleft}
\hspace*{0.175in}
\begin{math}  
{\cal G} = \left( \begin{array}{cccc} 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0
\end{array} \right)
\end{math}  
\end{flushleft}

\begin{flushleft}
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt copy\_into}({\cal A,G},1,2) & = & 
\left( \begin{array}{cccc} 0 & 1 & 2 & 3 \\ 0 & 4 & 5 & 6 \\ 0 & 7 & 8 
& 9 \\ 0 & 0 & 0 & 0  
\end{array} \right)
\end{array}
\end{math}  
\end{flushleft}

A diagonal matrix can be created with \f{DIAGONAL}\ttindex{DIAGONAL}.
The argument is a list of expressions of matrices which form the
diagonal.

An existing matrix can be extended; the call \f{EXTEND}(A,r,c,exp)\ttindex{EXTEND}
returns the matrix A extended by r rows and c columns, with the new
entries all exp.

The function \f{GET\_COLUMNS}\ttindex{GET\_COLUMNS} extracts from a
matrix a list of the specified columns as matrices. 
\f{GET\_ROWS}\ttindex{GET\_ROWS} does the equivalent for rows.

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt get\_columns}({\cal A},\{1,3\}) & = & 
\left\{ 
        \left( \begin{array}{c} 1 \\ 4 \\ 7 \end{array} \right),
        \left( \begin{array}{c} 3 \\ 6 \\ 9 \end{array} \right) 
\right\} 
\end{array}
\end{math}  
\end{flushleft}

The Hermitian transpose, that is a matrix in which the (i,$\,$j) entry is the conjugate of 
the (j,$\,$i) entry of the input is returned by \f{HERMITIAN\_TP}\ttindex{HERMITIAN\_TP}.

\f{MATRIX\_AUGMENT}(\{mat$_{1}$,mat$_{2}$, \ldots ,mat$_{n}$\})\ttindex{MATRIX\_AUGMENT}
produces a new matrix from the list joined as new columns.
\ttindex{MATRIX\_STACK}\f{MATRIX\_STACK} joins a list of matrices by
stacking them.

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt matrix\_stack}(\{{\cal A,A}\}) & = & 
        \left( \begin{array}{ccc} 1 & 2 & 3 \\ 4 & 5 & 6 \\ 7 & 8 & 9 
\\ 1 & 2 & 3 \\ 4 & 5 & 6 \\ 7 & 8 & 9 
 \end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}

\f{MINOR}(A,r,c)\ttindex{MINOR} calculates the (r,c) minor of A.

\f{PIVOT}\ttindex{PIVOT} pivots a matrix about its (r,c) entry.
To do this, multiples of the $r^{th}$ row are added to every other row in
the matrix.  This means that the $c^{th}$ column will be 0 except for
the (r,c) entry.

A variant on this operation is provided by
\f{ROWS\_PIVOT}\ttindex{ROWS\_PIVOT}.  It applies the pivot only to the
rows specified as the last argument.

A sub matrix can be extracted, giving a list or the rows and columns
to keep.

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt sub\_matrix}({\cal A},\{1,3\},\{2,3\}) & = & 
        \left( \begin{array}{cc} 2 & 3 \\ 8 & 9
 \end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}

The basic operation of swapping rows or columns is provided by
\f{SWAP\_ROWS}\ttindex{SWAP\_ROWS} and
\f{SWAP\_COLUMNS}\ttindex{SWAP\_COLUMNS}.  Individual entries can be
swapped with \f{SWAP\_ENTRIES}\ttindex{SWAP\_ENTRIES}.
\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt swap\_columns}({\cal A},2,3) & = & 
        \left( \begin{array}{ccc} 1 & 3 & 2 \\ 4 & 6 & 5 \\ 7 & 9 & 8
 \end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt swap\_entries}({\cal A},\{1,1\},\{3,3\}) & = & 
        \left( \begin{array}{ccc} 9 & 2 & 3 \\ 4 & 5 & 6 \\ 7 & 8 & 1
 \end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}


\section{Constructors}

\f{AUGMENT\_COLUMNS}\ttindex{AUGMENT\_COLUMNS} allows just specified
columns to be selected; \f{STACK\_ROWS}\ttindex{STACK\_ROWS} does
a similar job for rows.

\begin{math}  
\begin{array}{ccc}
{\tt stack\_rows}({\cal A},\{1,3\}) & = & 
\left( \begin{array}{ccc} 1 & 2 & 3 \\ 7 & 8 & 9
\end{array} \right)  
\end{array}  
\end{math}

Rows or columns can be removed with
\f{REMOVE\_COLUMNS}\ttindex{REMOVE\_COLUMNS} and
\f{REMOVE\_ROWS}\ttindex{REMOVE\_ROWS}.

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt remove\_columns}({\cal A},2) & = & 
        \left( \begin{array}{cc} 1 & 3 \\ 4 & 6 \\ 7 & 9  
 \end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}


{\tt BAND\_MATRIX}\ttindex{BAND\_MATRIX} creates a square matrix of
dimension its second argument.  The diagonal consists of the middle
expressions of the first argument, which is an expression list. The
expressions to the left of this fill the required number of
sub\_diagonals and the expressions to the right the super\_diagonals. 

\begin{math}  
\begin{array}{ccc}
{\tt band\_matrix}(\{x,y,z\},6) & = & 
\left( \begin{array}{cccccc} y & z & 0 & 0 & 0 & 0 \\ x & y & z & 0 & 0
& 0 \\ 0 & x & y & z & 0 & 0 \\ 0 & 0 & x & y & z & 0 \\ 0 & 0 & 0 & x &
 y & z \\ 0 & 0 & 0 & 0 & x & y 
\end{array} \right)
\end{array}  
\end{math}  

Related to the band matrix is a block matrix, which can be created by

\noindent {\tt BLOCK\_MATRIX(r,c,matrix\_list)}.\ttindex{BLOCK\_MATRIX}

The resulting matrix consists of r by c matrices filled from the
matrix\_list row wise. 

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\cal B} = \left( \begin{array}{cc} 1 & 0 \\ 0 & 1
\end{array} \right), & 
{\cal C} = \left( \begin{array}{c} 5 \\ 5
\end{array} \right), &
{\cal D} = \left( \begin{array}{cc} 22 & 33 \\ 44 & 55
\end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}

\vspace*{0.175in}

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt block\_matrix}(2,3,\{{\cal B,C,D,D,C,B}\}) & = & 
\left( \begin{array}{ccccc} 1 & 0 & 5 & 22 & 33 \\ 0 & 1 & 5 & 44 & 55 
\\
22 & 33 & 5 & 1 & 0 \\ 44 & 55 & 5 & 0 & 1
\end{array} \right)  
\end{array}  
\end{math}  
\end{flushleft}

Characteristic polynomials and characteristic matrices are created by 
the functions
{\tt CHAR\_POLY}\ttindex{CHAR\_POLY} and
\f{CHAR\_MATRIX}\ttindex{CHAR\_MATRIX}. 

A set of linear equations can be turned into the associated
coefficient matrix and vector of unknowns and the righthandside.  
\f{COEFF\_MATRIX} returns a list \{${\cal C,X,B}$\} such that ${\cal
CX} = {\cal B}$. 

\begin{math}
\hspace*{0.175in}
{\tt coeff\_matrix}(\{x+y+4*z=10,y+x-z=20,x+y+4\}) =  
\end{math}

\vspace*{0.1in}

\begin{flushleft}
\hspace*{0.175in}
\begin{math}  
\left\{ \left( \begin{array}{ccc} 4 & 1 & 1 \\ -1 & 1 & 1 \\ 0 & 1 & 1 
\end{array} \right), \left( \begin{array}{c} z \\ y \\ x \end{array} 
\right), \left( \begin{array}{c} 10 \\ 20 \\ -4 
\end{array} \right) \right\} 
\end{math}  
\end{flushleft}

\f{COMPANION}(poly,x)  creates the companion matrix ${\cal C}$ of a
polynomial.  That is the square matrix of dimension n, where n is the
degree of polynomial with respect to x, and the entries of ${\cal C}$ are: 
${\cal C}$(i,n) = -coeffn(poly,x,i-1) for i = 1 \ldots n, ${\cal
C}$(i,i-1) = 1 for i = 2 \ldots n and the rest are 0.

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt companion}(x^4+17*x^3-9*x^2+11,x) & = & 
\left( \begin{array}{cccc} 0 & 0 & 0 & -11 \\ 1 & 0 & 0 & 0 \\ 
0 & 1 & 0 & 9 \\ 0 & 0 & 1 & -17 
\end{array} \right)
\end{array}
\end{math}  
\end{flushleft}

The polynomial associated with a companion matrix can be recovered by
calling \f{FIND\_COMPANION}\ttindex{FIND\_COMPANION}.

\f{HESSIAN}(expr, var\_list)\ttindex{HESSIAN} calculates the Hessian
matrix of the expressions with respect to the variables in the list,
or the single variable.  That is the matrix with the (i,$\,$j) element
the $j^{th}$ derivative of the expressions with respect to the
$i^{th}$ variable.

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}        
\begin{array}{ccc}
{\tt hessian}(x*y*z+x^2,\{w,x,y,z\}) & = & 
\left( \begin{array}{cccc} 0 & 0 & 0 & 0 \\ 0 & 2 & z & y \\ 0 & z & 0 
& x \\ 0 & y & x & 0
\end{array} \right)
\end{array}
\end{math}  
\end{flushleft}

Hilbert's matrix, that is where the (i,$\,$j) element is $1/(i+j-x)$
is constructed by \f{HILBERT}(n,x)\ttindex{HILBERT}.

The Jacobian of an expression list with respect to a variable list is
calculated by
\f{JACOBIAN}(expr\_list,variable\_list)\ttindex{JACOBIAN}.  This is a
matrix whose (i,$\,$j) entry is df(expr\_list(i),variable\_list(j)).


The square Jordan block matrix of dimension $n$ is calculated by the
function \f{JORDAN\_BLOCK}(exp,n).\ttindex{JORDAN\_BLOCK}  The entries
of the Jordan\_block matrix are ${\cal J}$(i,i) = expr for i=1 \ldots
n, ${\cal J}$(i,i+1) = 1 for i=1 \ldots n-1, and all other entries are 0.

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}        
\begin{array}{ccc}
{\tt jordan\_block(x,5)} & = & 
\left( \begin{array}{ccccc} x & 1 & 0 & 0 & 0 \\ 0 & x & 1 & 0 & 0 \\ 0 
& 0 & x & 1 & 0 \\ 0 & 0 & 0 & x & 1 \\ 0 & 0 & 0 & 0 & x
\end{array} \right)
\end{array}
\end{math}  
\end{flushleft}

\f{MAKE\_IDENTITY}(n)\ttindex{MAKE\_IDENTITY} generates the $n \times
n$ identity matrix. 

\f{RANDOM\_MATRIX}(r,c,limit)\ttindex{RANDOM\_MATRIX} generates and $r
\times c$ matrix with random values limited by {\tt limit}.  The type
of entries is controlled by a number of switches.

\begin{description}
\item[{\tt IMAGINARY}]\ttindex{IMAGINARY}
If on then matrix entries are $x+i*y$ where $-limit < x,y < limit$.
\item[{\tt NOT\_NEGATIVE}]\ttindex{NOT\_NEGATIVE}
If on then $0 < entry < limit$. In the imaginary case we have $0 < x,y
< limit$.
\item[{\tt ONLY\_INTEGER}]\ttindex{ONLY\_INTEGER}
If on then each entry is an integer.  In the imaginary case $x$ and $y$ are
integers.  If off the values are rounded.
\item[{\tt SYMMETRIC}]\ttindex{SYMMETRIC}
If on then the matrix is symmetric.
\item[{\tt UPPER\_MATRIX}]\ttindex{UPPER\_MATRIX}
If on then the matrix is upper triangular.
\item[{\tt LOWER\_MATRIX}]\ttindex{LOWER\_MATRIX}
If on then the matrix is lower triangular.
\end{description}

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt random\_matrix}(3,3,10) & = & 
        \left( \begin{array}{ccc} -4.729721 & 6.987047 & 7.521383 \\
- 5.224177 & 5.797709 & - 4.321952 \\
- 9.418455 & - 9.94318 & - 0.730980
 \end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}

\vspace*{0.2in}
\hspace*{0.165in}
{\tt on only\_integer, not\_negative, upper\_matrix, imaginary;}
\begin{flushleft}  
%\hspace*{0.12in}
\begin{math}        
\begin{array}{ccc}
{\tt random\_matrix}(4,4,10) & = & 
\left( \begin{array}{cccc} 2*i+5 & 3*i+7 & 7*i+3 & 6 \\ 0 & 2*i+5 & 
5*i+1 & 2*i+1 \\ 0 & 0 & 8 & i \\ 0 & 0 & 0& 5*i+9 
\end{array} \right)
\end{array}
\end{math}  
\end{flushleft}

{\tt TOEPLITZ}\ttindex{TOEPLITZ} creates the Toeplitz matrix from the
given expression list.  This is a square symmetric matrix in which the
first expression is placed on the diagonal  and the $i^{th}$
expression is placed on the $(i-1)^{th}$ sub- and super-diagonals.
It has dimension equal to the number of expressions.

\begin{flushleft}  
\begin{math}  
\begin{array}{ccc}
{\tt toeplitz}(\{w,x,y,z\}) & = & 
        \left( \begin{array}{cccc} w & x & y & z \\ x & w & x & y \\
      y & x & w & x \\ z & y & x & w
 \end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}

\f{VANDERMONDE}\ttindex{VANDERMONDE} creates the Vandermonde matrix
from the expression list; the square matrix in which the (i,$\,$j)
entry is expr\_list(i) $^{(j-1)}$.

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt vandermonde}(\{x,2*y,3*z\}) & = & 
        \left( \begin{array}{ccc} 1 & x & x^2 \\ 1 & 2*y & 4*y^2 \\ 1 
& 3*z & 9*z^2 
 \end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}

The direct product\index{direct product} (or tensor
product\index{tensor product}) is created by the
\f{KRONECKER\_PRODUCT}\ttindex{KRONECKER\_PRODUCT} function.

\begin{verbatim}
a1 := mat((1,2),(3,4),(5,6))$
a2 := mat((1,1,1),(2,z,2),(3,3,3))$
kronecker_product(a1,a2);
\end{verbatim}
\begin{flushleft}
\hspace*{0.1in}
\begin{math}
\begin{array}{ccc}
\left( \begin{array}{cccccc} 1 & 1 & 1 & 2 & 2 & 2 \\
2 &  z & 2 & 4  &2*z &4 \\
3 &  3 & 3 & 6  & 6  &6 \\
3 &  3 & 3 & 4  & 4  &4 \\
6 & 3*z& 6 & 8  &4*z &8 \\
9 &  9 & 9 & 12 &12  &12\\
5 &  5 & 5 & 6  & 6  &6 \\
10 &5*z& 10& 12 &6*z &12 \\ 
15 &15 & 15& 18 &18  &18 \end{array} \right)
\end{array}
\end{math}
\end{flushleft}

\section{Higher Algorithms}

The Cholesky decomposition  of a matrix can be
calculated with the function \f{CHOLESKY}.  It returns \{${\cal
L,U}$\} where ${\cal L}$ is a lower matrix, ${\cal U}$ is an upper
matrix, and ${\cal A} = {\cal LU}$, and ${\cal U} = {\cal L}^T$. 

Gram--Schmidt orthonormalisation can be calculated by
\f{GRAM\_SCHMIDT}\ttindex{GRAM\_SCHMIDT}.  It accepts a list of
linearly independent vectors, written as lists, and returns a list of
orthogonal normalised vectors.

\begin{verbatim}
gram_schmidt({{1,0,0},{1,1,0},{1,1,1}});
 
{{1,0,0},{0,1,0},{0,0,1}}

gram_schmidt({{1,2},{3,4}});

      1         2        2*sqrt(5)    - sqrt(5)
{{---------,---------},{-----------,------------}}
   sqrt(5)   sqrt(5)         5           5
\end{verbatim}

The LU decomposition of a real or imaginary matrix with numeric
entries is performed by {\tt LU\_DECOM(${\cal A}$)}.\ttindex{LU\_DECOM}
It returns \{${\cal L,U}$\} where ${\cal L}$ is a lower diagonal
matrix, ${\cal U}$ an upper diagonal matrix and ${\cal A} = {\cal LU}$.

Note: the algorithm used can swap the rows of ${\cal A}$ during
the calculation.  This means that ${\cal LU}$ does not equal ${\cal
A}$ but a row equivalent of it.  Due to this, {\tt lu\_decom} returns
\{${\cal L,U}$,vec\}.  The call {\tt CONVERT(${\cal
A}$,vec)}\ttindex{CONVERT} will return the matrix that has been
decomposed, {\em i.e.\ } ${\cal LU} = $  {\tt convert(${\cal A}$,vec)}.

\begin{flushleft}
\hspace*{0.175in}
\begin{math}  
{\cal K} = \left( \begin{array}{ccc} 1 & 3 & 5 \\ -4 & 3 & 7 \\ 8 & 6 & 
4
\end{array} \right)
\end{math}  
\end{flushleft}

\begin{flushleft}  
%\hspace*{0.1in}
\begin{math}  
\begin{array}{cccc}
$%   {\tt lu} :=
 {\tt lu\_decom}$({\cal K}) & = & 
\left\{ 
        \left( \begin{array}{ccc} 8 & 0 & 0 \\ -4 & 6 & 0 \\ 1 & 2.25 & 
1.125 1 \end{array} \right), 
        \left( \begin{array}{ccc} 1 & 0.75 & 0.5 \\ 0 & 1 & 1.5 \\ 0 & 
0 & 1 \end{array} \right), 
        [\; 3 \; 2 \; 3 \; ]
\right\} 
\end{array}
\end{math}  
\end{flushleft}

{\tt PSEUDO\_INVERSE}\ttindex{PSEUDO\_INVERSE}, also known as the
Moore--Penrose inverse\index{Moore--Penrose inverse}, computes
the pseudo inverse of ${\cal A}$.  
Given the singular value decomposition of ${\cal A}$, {\em i.e.\ }
${\cal A} =  {\cal U} \sum {\cal V}^T$, then the pseudo inverse ${\cal
A}^{-1}$ is defined by ${\cal A}^{-1} = {\cal V}^T \sum^{-1} {\cal U}$.

Thus ${\cal A}$ $ * $ {\tt pseudo\_inverse}$({\cal A}) = {\cal I}$.

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt pseudo\_inverse}({\cal A}) & = & 
        \left( \begin{array}{cc} -0.2 & 0.1 \\ -0.05 & 0.05 \\ 0.1 & 0 
\\ 0.25 & -0.05 
 \end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}

\label{simplex}
The simplex linear programming algorithm\index{Simplex Algorithm} for
maximising or minimising a function subject to lineal inequalities can
be used with the function \f{SIMPLEX}\ttindex{SIMPLEX}.  It requires
three arguments, the first indicates where the action is to maximising
or minimising, the second is the test expressions, and the last is a
list of linear inequalities.
It returns \{optimal value,\{ values of variables at this optimal\}\}.
The algorithm implies that all the variables are non-negative.

\begin{addtolength}{\leftskip}{0.22in}
%\begin{math}
{\tt simplex($max,x+y,\{x>=10,y>=20,x+y<=25\}$);}
%\end{math}

{\tt ***** Error in simplex: Problem has no feasible solution.}

\vspace*{0.2in}

\parbox[t]{0.96\linewidth}{\tt simplex($max,10x+5y+5.5z,\{5x+3z<=200,
x+0.1y+0.5z<=12$,\\
\hspace*{0.55in} $0.1x+0.2y+0.3z<=9, 30x+10y+50z<=1500\}$);}

\vspace*{0.1in}
{\tt $\{525.0,\{x=40.0,y=25.0,z=0\}$\}}

\end{addtolength}

{\tt SVD}\ttindex{SVD} computes the singular value decomposition of
${\cal A}$ with numeric entries.   It returns \{${\cal U},\sum,{\cal V}$\} where ${\cal A} = {\cal U} 
\sum {\cal V}^T$ and $\sum = diag(\sigma_{1}, \ldots ,\sigma_{n}). \; 
\sigma_{i}$ for $i= (1 \ldots n)$ are the singular values of ${\cal A}$.
The singular values of ${\cal A}$ are the non-negative square roots of 
the eigenvalues of ${\cal A}^T {\cal A}$. 

${\cal U}$ and ${\cal V}$ are such that ${\cal UU}^T = {\cal VV}^T = 
{\cal V}^T {\cal V} = {\cal I}_n$.
\begin{flushleft}
\hspace*{0.175in}
\begin{math}  
{\cal Q} = \left( \begin{array}{cc} 1 & 3 \\ -4 & 3 
\end{array} \right)
\end{math}  
\end{flushleft}

\begin{eqnarray}
\hspace*{0.1in}
{\tt svd({\cal Q})} & = & 
\left\{ 
        \left( \begin{array}{cc} 0.289784 & 0.957092 \\ -0.957092 & 
0.289784 \end{array} \right), \left( \begin{array}{cc} 5.149162 & 0 \\ 
0 & 2.913094 \end{array} \right), \right. \nonumber \\ & & \left. \: \; 
\, \left( \begin{array}{cc} -0.687215 & 0.726453 \\ -0.726453 & 
-0.687215 \end{array} \right)       
\right\} \nonumber
\end{eqnarray}

{\tt TRIANG\_ADJOINT}\ttindex{TRIANG\_ADJOINT} computes the trianglarizing adjoint of
the given matrix. The triangularizing adjoint is a lower triangular matrix. The
multiplication of the triangularizing adjoint with the given matrix results in an
upper triangular matrix. The i-th entry in the diagonal of this matrix is the
determinant of the principal i-th minor of the given matrix.

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{ccc}
{\tt triang\_adjoint}({\cal A}) & = & 
        \left( \begin{array}{ccc} 1 & 0 & 0 \\ -4 & 1 & 0 \\ -3 & 6 & -3 
 \end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}

The multiplication of this matrix with ${\cal A}$ results in an upper triangular matrix.

\begin{flushleft}  
\hspace*{0.1in}
\begin{math}  
\begin{array}{cccc}
 \left( \begin{array}{ccc} 1 & 0 & 0 \\ -4 & 1 & 0 \\ -3 & 6 & -3 
 \end{array} \right) &
 \left( \begin{array}{ccc} 1 & 2 & 3 \\ 4 & 5 & 6 \\ 7 & 8 & 9 
 \end{array} \right)
 & = & 
        \left( \begin{array}{ccc} 1 & 2 & 3 \\ 0 & -3 & -6 \\ 0 & 0 & 0 
 \end{array} \right) 
\end{array}
\end{math}  
\end{flushleft}

\section{Fast Linear Algebra}

By turning the {\tt FAST\_LA}\ttindex{FAST\_LA} switch on, the speed
of the following functions will be increased:

\begin{tabular}{l l l l}
   add\_columns    & add\_rows      & augment\_columns & column\_dim  \\
   copy\_into      & make\_identity & matrix\_augment  & matrix\_stack\\
   minor           & mult\_column   &  mult\_row       & pivot        \\
   remove\_columns & remove\_rows   & rows\_pivot      & squarep      \\
   stack\_rows     & sub\_matrix    & swap\_columns    & swap\_entries\\
   swap\_rows      & symmetricp                                     
\end{tabular}

The increase in speed will be insignificant unless you are making a 
thousands of calls.  When using this switch, 
error checking is minimised, and thus illegal input may give strange
error messages.



REDUCE Historical
REDUCE Sourceforge Project | Historical SVN Repository | GitHub Mirror | SourceHut Mirror | NotABug Mirror | Chisel Mirror | Chisel RSS ]