@c DO NOT EDIT!  Generated automatically by munge-texi.pl.

@c Copyright (C) 1996-2025 The Octave Project Developers
@c
@c This file is part of Octave.
@c
@c Octave is free software: you can redistribute it and/or modify it
@c under the terms of the GNU General Public License as published by
@c the Free Software Foundation, either version 3 of the License, or
@c (at your option) any later version.
@c
@c Octave is distributed in the hope that it will be useful, but
@c WITHOUT ANY WARRANTY; without even the implied warranty of
@c MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
@c GNU General Public License for more details.
@c
@c You should have received a copy of the GNU General Public License
@c along with Octave; see the file COPYING.  If not, see
@c <https://www.gnu.org/licenses/>.

@node Matrix Manipulation
@chapter Matrix Manipulation

There are a number of functions available for checking to see if the
elements of a matrix meet some condition, and for rearranging the
elements of a matrix.  For example, Octave can easily tell you if all
the elements of a matrix are finite, or are less than some specified
value.  Octave can also rotate the elements, extract the upper- or
lower-triangular parts, or sort the columns of a matrix.

@menu
* Finding Elements and Checking Conditions::
* Rearranging Matrices::
* Special Utility Matrices::
* Famous Matrices::
@end menu

@node Finding Elements and Checking Conditions
@section Finding Elements and Checking Conditions

The functions @code{any} and @code{all} are useful for determining
whether any or all of the elements of a matrix satisfy some condition.
The @code{find} function is also useful in determining which elements of
a matrix meet a specified condition.

@c any libinterp/corefcn/data.cc
@anchor{XREFany}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{tf} =} any (@var{x})
@deftypefnx {} {@var{tf} =} any (@var{x}, @var{dim})
For a vector argument, return true (logical 1) if any element of the vector
is nonzero.

For a matrix argument, return a row vector of logical ones and zeros with each
element indicating whether any of the elements of the corresponding column of
the matrix are nonzero.  For example:

@example
@group
any (eye (2, 4))
 @result{} [ 1, 1, 0, 0 ]
@end group
@end example

If the optional argument @var{dim} is supplied, work along dimension @var{dim}.
For example:

@example
@group
any (eye (2, 4), 2)
 @result{} [ 1; 1 ]
@end group
@end example
@xseealso{@ref{XREFall,,all}}
@end deftypefn


@c all libinterp/corefcn/data.cc
@anchor{XREFall}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{tf} =} all (@var{x})
@deftypefnx {} {@var{tf} =} all (@var{x}, @var{dim})
For a vector argument, return true (logical 1) if all elements of the vector
are nonzero.

For a matrix argument, return a row vector of logical ones and zeros with each
element indicating whether all of the elements of the corresponding column of
the matrix are nonzero.  For example:

@example
@group
all ([2, 3; 1, 0])
    @result{} [ 1, 0 ]
@end group
@end example

If the optional argument @var{dim} is supplied, work along dimension @var{dim}.
@xseealso{@ref{XREFany,,any}}
@end deftypefn


Since the comparison operators (@pxref{Comparison Ops}) return matrices
of ones and zeros, it is easy to test a matrix for many things, not just
whether the elements are nonzero.  For example,

@example
@group
all (all (rand (5) < 0.9))
     @result{} 0
@end group
@end example

@noindent
tests a random 5 by 5 matrix to see if all of its elements are less
than 0.9.

Note that in conditional contexts (like the test clause of @code{if} and
@code{while} statements) Octave treats the test as if you had typed
@code{all (all (condition))}.

@c xor scripts/general/xor.m
@anchor{XREFxor}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{z} =} xor (@var{x}, @var{y})
@deftypefnx {} {@var{z} =} xor (@var{x1}, @var{x2}, @dots{})
Return the @dfn{exclusive or} of @var{x} and @var{y}.

For boolean expressions @var{x} and @var{y},
@code{xor (@var{x}, @var{y})} is true if and only if one of @var{x} or
@var{y} is true.  Otherwise, if @var{x} and @var{y} are both true or both
false, @code{xor} returns false.

The truth table for the xor operation is

@multitable @columnfractions 0.41 .044 .06 .042 0.42
@item @tab @var{x} @tab @var{y} @tab @var{z} @tab
@item @tab - @tab - @tab - @tab
@item @tab 0 @tab 0 @tab 0 @tab
@item @tab 1 @tab 0 @tab 1 @tab
@item @tab 0 @tab 1 @tab 1 @tab
@item @tab 1 @tab 1 @tab 0 @tab
@end multitable

If more than two arguments are given the xor operation is applied
cumulatively from left to right:

@example
(@dots{}((x1 XOR x2) XOR x3) XOR @dots{})
@end example

@xseealso{@ref{XREFand,,and}, @ref{XREFor,,or}, @ref{XREFnot,,not}}
@end deftypefn


@c diff libinterp/corefcn/data.cc
@anchor{XREFdiff}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{y} =} diff (@var{x})
@deftypefnx {} {@var{y} =} diff (@var{x}, @var{k})
@deftypefnx {} {@var{y} =} diff (@var{x}, @var{k}, @var{dim})
If @var{x} is a vector of length @math{n}, @w{@code{diff (@var{x})}}@ is the
vector of first differences
@tex
 $x_2 - x_1, \ldots{}, x_n - x_{n-1}$.
@end tex
@ifnottex
 @var{x}(2) - @var{x}(1), @dots{}, @var{x}(n) - @var{x}(n-1).
@end ifnottex

If @var{x} is a matrix, @w{@code{diff (@var{x})}}@ is the matrix of column
differences along the first non-singleton dimension.

The second argument is optional.  If supplied,
@w{@code{diff (@var{x}, @var{k})}}, where @var{k} is a non-negative integer,
returns the @var{k}-th differences.  It is possible that @var{k} is larger than
the first non-singleton dimension of the matrix.  In this case, @code{diff}
continues to take the differences along the next non-singleton dimension.

The dimension along which to take the difference can be explicitly stated with
the optional variable @var{dim}.  In this case the @var{k}-th order differences
are calculated along this dimension.  In the case where @var{k} exceeds
@w{@code{size (@var{x}, @var{dim})}}@ an empty matrix is returned.
@xseealso{@ref{XREFsort,,sort}, @ref{XREFmerge,,merge}}
@end deftypefn


@c isinf libinterp/corefcn/mappers.cc
@anchor{XREFisinf}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{tf} =} isinf (@var{x})
Return a logical array which is true where the elements of @var{x} are
infinite and false where they are not.

For example:

@example
@group
isinf ([13, Inf, NA, NaN])
      @result{} [ 0, 1, 0, 0 ]
@end group
@end example
@xseealso{@ref{XREFisfinite,,isfinite}, @ref{XREFisnan,,isnan}, @ref{XREFisna,,isna}}
@end deftypefn


@c isnan libinterp/corefcn/mappers.cc
@anchor{XREFisnan}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{tf} =} isnan (@var{x})
Return a logical array which is true where the elements of @var{x} are
NaN values and false where they are not.

NA values are also considered NaN values.  For example:

@example
@group
isnan ([13, Inf, NA, NaN])
      @result{} [ 0, 0, 1, 1 ]
@end group
@end example
@xseealso{@ref{XREFisna,,isna}, @ref{XREFisinf,,isinf}, @ref{XREFisfinite,,isfinite}}
@end deftypefn


@c isfinite libinterp/corefcn/mappers.cc
@anchor{XREFisfinite}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{tf} =} isfinite (@var{x})
Return a logical array which is true where the elements of @var{x} are
finite values and false where they are not.

For example:

@example
@group
isfinite ([13, Inf, NA, NaN])
     @result{} [ 1, 0, 0, 0 ]
@end group
@end example
@xseealso{@ref{XREFisinf,,isinf}, @ref{XREFisnan,,isnan}, @ref{XREFisna,,isna}}
@end deftypefn


@c common_size scripts/general/common_size.m
@anchor{XREFcommon_size}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {[@var{err}, @var{yi}, @dots{}] =} common_size (@var{xi}, @dots{})
Determine if all input arguments are either scalar or of common size.

If true, @var{err} is zero, and @var{yi} is a matrix of the common size
with all entries equal to @var{xi} if this is a scalar or @var{xi}
otherwise.  If the inputs cannot be brought to a common size, @var{err} is
1, and @var{yi} is @var{xi}.  For example:

@example
@group
[err, a, b] = common_size ([1 2; 3 4], 5)
     @result{} err = 0
     @result{} a = [ 1, 2; 3, 4 ]
     @result{} b = [ 5, 5; 5, 5 ]
@end group
@end example

@noindent
This is useful for implementing functions where arguments can either be
scalars or of common size.
@xseealso{@ref{XREFsize,,size}, @ref{XREFsize_equal,,size_equal}, @ref{XREFnumel,,numel}, @ref{XREFndims,,ndims}}
@end deftypefn


@c find libinterp/corefcn/find.cc
@anchor{XREFfind}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{idx} =} find (@var{x})
@deftypefnx {} {@var{idx} =} find (@var{x}, @var{n})
@deftypefnx {} {@var{idx} =} find (@var{x}, @var{n}, @var{direction})
@deftypefnx {} {[i, j] =} find (@dots{})
@deftypefnx {} {[i, j, v] =} find (@dots{})
Return a vector of indices of nonzero elements of a matrix, as a row if
@var{x} is a row vector or as a column otherwise.

To obtain a single index for each matrix element, Octave pretends that the
columns of a matrix form one long vector (like Fortran arrays are stored).
For example:

@example
@group
find (eye (2))
  @result{} [ 1; 4 ]
@end group
@end example

If two inputs are given, @var{n} indicates the maximum number of elements to
find from the beginning of the matrix or vector.

If three inputs are given, @var{direction} should be one of
@qcode{"first"} or @qcode{"last"}, requesting only the first or last
@var{n} indices, respectively.  However, the indices are always returned in
ascending order.

If two outputs are requested, @code{find} returns the row and column
indices of nonzero elements of a matrix.  For example:

@example
@group
[i, j] = find (2 * eye (2))
    @result{} i = [ 1; 2 ]
    @result{} j = [ 1; 2 ]
@end group
@end example

If three outputs are requested, @code{find} also returns a vector
containing the nonzero values.  For example:

@example
@group
[i, j, v] = find (3 * eye (2))
       @result{} i = [ 1; 2 ]
       @result{} j = [ 1; 2 ]
       @result{} v = [ 3; 3 ]
@end group
@end example

If @var{x} is a multi-dimensional array of size m x n x p x @dots{}, @var{j}
contains the column locations as if @var{x} was flattened into a
two-dimensional matrix of size m x (n + p + @dots{}).

Note that this function is particularly useful for sparse matrices, as
it extracts the nonzero elements as vectors, which can then be used to
create the original matrix.  For example:

@example
@group
sz = size (a);
[i, j, v] = find (a);
b = sparse (i, j, v, sz(1), sz(2));
@end group
@end example
@xseealso{@ref{XREFnonzeros,,nonzeros}}
@end deftypefn


@c lookup libinterp/corefcn/lookup.cc
@anchor{XREFlookup}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{idx} =} lookup (@var{table}, @var{y})
@deftypefnx {} {@var{idx} =} lookup (@var{table}, @var{y}, @var{opt})
Lookup values in a @strong{sorted} table.

This function is usually used as a prelude to interpolation.

If table is increasing, of length N and @code{idx = lookup (table, y)}, then
@code{table(idx(i)) <= y(i) < table(idx(i+1))} for all @code{y(i)} within the
table.  If @code{y(i) < table(1)} then @code{idx(i)} is 0.  If
@code{y(i) >= table(end)} or @code{isnan (y(i))} then @code{idx(i)} is N.

If the table is decreasing, then the tests are reversed.  For non-strictly
monotonic tables, empty intervals are always skipped.  The result is undefined
if @var{table} is not monotonic, or if @var{table} contains a NaN.

The complexity of the lookup is O(M*log(N)) where M is the size of @var{y}.
In the special case when @var{y} is also sorted, the complexity is
O(min (M*log(N), M+N)).

@var{table} and @var{y} can also be cell arrays of strings (or @var{y} can be a
single string).  In this case, string lookup is performed using lexicographical
comparison.

If @var{opts} is specified, it must be a string with letters indicating
additional options.

@table @code
@item m
Match.  @code{table(idx(i)) == y(i)} if @code{y(i)} occurs in table;
otherwise, @code{idx(i)} is zero.

@item b
Boolean.  @code{idx(i)} is a logical 1 or 0, indicating whether @code{y(i)}
is contained in table or not.

@item l
Left.  For numeric lookups the leftmost subinterval shall be extended to
minus infinity (i.e., all indices at least 1).

@item r
Right.  For numeric lookups the rightmost subinterval shall be extended to
infinity (i.e., all indices at most N-1).
@end table

@strong{Note}: If @var{table} is not sorted the results from @code{lookup}
will be unpredictable.
@end deftypefn


If you wish to check if a variable exists at all, instead of properties
its elements may have, consult @ref{Status of Variables}.

@node Rearranging Matrices
@section Rearranging Matrices

@c fliplr scripts/general/fliplr.m
@anchor{XREFfliplr}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{B} =} fliplr (@var{A})
Flip array left to right.

Return a copy of @var{A} with the order of the columns reversed.  In other
words, @var{A} is flipped left-to-right about a vertical axis.  For example:

@example
@group
fliplr ([1, 2; 3, 4])
     @result{}  2  1
         4  3
@end group
@end example

@xseealso{@ref{XREFflipud,,flipud}, @ref{XREFflip,,flip}, @ref{XREFrot90,,rot90}, @ref{XREFrotdim,,rotdim}}
@end deftypefn


@c flipud scripts/general/flipud.m
@anchor{XREFflipud}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{B} =} flipud (@var{A})
Flip array upside down.

Return a copy of @var{A} with the order of the rows reversed.  In other
words, @var{A} is flipped upside-down about a horizontal axis.  For example:

@example
@group
flipud ([1, 2; 3, 4])
     @result{}  3  4
         1  2
@end group
@end example

@xseealso{@ref{XREFfliplr,,fliplr}, @ref{XREFflip,,flip}, @ref{XREFrot90,,rot90}, @ref{XREFrotdim,,rotdim}}
@end deftypefn


@c flip scripts/general/flip.m
@anchor{XREFflip}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{B} =} flip (@var{A})
@deftypefnx {} {@var{B} =} flip (@var{A}, @var{dim})
Return a copy of array @var{A} flipped across dimension @var{dim}.

If @var{dim} is unspecified it defaults to the first non-singleton
dimension.

Examples:

@example
## row vector
flip ([1  2  3  4])
      @result{}  4  3  2  1

## column vector
flip ([1; 2; 3; 4])
      @result{}  4
          3
          2
          1

## 2-D matrix along dimension 1
flip ([1 2; 3 4])
      @result{}  3  4
          1  2

## 2-D matrix along dimension 2
flip ([1 2; 3 4], 2)
      @result{}  2  1
          4  3
@end example

@xseealso{@ref{XREFfliplr,,fliplr}, @ref{XREFflipud,,flipud}, @ref{XREFrot90,,rot90}, @ref{XREFrotdim,,rotdim}, @ref{XREFpermute,,permute}, @ref{XREFtranspose,,transpose}}
@end deftypefn


@c rot90 scripts/general/rot90.m
@anchor{XREFrot90}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{B} =} rot90 (@var{A})
@deftypefnx {} {@var{B} =} rot90 (@var{A}, @var{k})
Rotate array by 90 degree increments.

Return a copy of @var{A} with the elements rotated counterclockwise in
90-degree increments.

The second argument is optional, and specifies how many 90-degree rotations
are to be applied (the default value is 1).  Negative values of @var{k}
rotate the matrix in a clockwise direction.
For example,

@example
@group
rot90 ([1, 2; 3, 4], -1)
    @result{}  3  1
        4  2
@end group
@end example

@noindent
rotates the given matrix clockwise by 90 degrees.  The following are all
equivalent statements:

@example
@group
rot90 ([1, 2; 3, 4], -1)
rot90 ([1, 2; 3, 4], 3)
rot90 ([1, 2; 3, 4], 7)
@end group
@end example

The rotation is always performed on the plane of the first two dimensions,
i.e., rows and columns.  To perform a rotation on any other plane, use
@code{rotdim}.

@xseealso{@ref{XREFrotdim,,rotdim}, @ref{XREFfliplr,,fliplr}, @ref{XREFflipud,,flipud}, @ref{XREFflip,,flip}}
@end deftypefn


@c rotdim scripts/general/rotdim.m
@anchor{XREFrotdim}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{B} =} rotdim (@var{A})
@deftypefnx {} {@var{B} =} rotdim (@var{A}, @var{n})
@deftypefnx {} {@var{B} =} rotdim (@var{A}, @var{n}, @var{plane})
Return a copy of @var{A} with the elements rotated counterclockwise in
90-degree increments.

The second argument @var{n} is optional, and specifies how many 90-degree
rotations are to be applied (the default value is 1).  Negative values of
@var{n} rotate the matrix in a clockwise direction.

The third argument is also optional and defines the plane of the rotation.
If present, @var{plane} is a two element vector containing two different
valid dimensions of the matrix.  When @var{plane} is not given the first two
non-singleton dimensions are used.

For example,

@example
@group
rotdim ([1, 2; 3, 4], -1, [1, 2])
     @result{}  3  1
         4  2
@end group
@end example

@noindent
rotates the given matrix clockwise by 90 degrees.  The following are all
equivalent statements:

@example
@group
rotdim ([1, 2; 3, 4], -1, [1, 2])
rotdim ([1, 2; 3, 4], 3, [1, 2])
rotdim ([1, 2; 3, 4], 7, [1, 2])
@end group
@end example
@xseealso{@ref{XREFrot90,,rot90}, @ref{XREFfliplr,,fliplr}, @ref{XREFflipud,,flipud}, @ref{XREFflip,,flip}}
@end deftypefn


@c cat libinterp/corefcn/data.cc
@anchor{XREFcat}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{A} =} cat (@var{dim}, @var{array1}, @var{array2}, @dots{}, @var{arrayN})
Return the concatenation of N-D array objects, @var{array1}, @var{array2},
@dots{}, @var{arrayN} along dimension @var{dim}.

@example
@group
A = ones (2, 2);
B = zeros (2, 2);
cat (2, A, B)
  @result{} 1 1 0 0
     1 1 0 0
@end group
@end example

Alternatively, we can concatenate @var{A} and @var{B} along the second
dimension in the following way:

@example
@group
[A, B]
@end group
@end example

@var{dim} can be larger than the dimensions of the N-D array objects and the
result will thus have @var{dim} dimensions as the following example shows:

@example
@group
cat (4, ones (2, 2), zeros (2, 2))
  @result{} ans(:,:,1,1) =

       1 1
       1 1

     ans(:,:,1,2) =

       0 0
       0 0
@end group
@end example
@xseealso{@ref{XREFhorzcat,,horzcat}, @ref{XREFvertcat,,vertcat}}
@end deftypefn


@c horzcat libinterp/corefcn/data.cc
@anchor{XREFhorzcat}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{A} =} horzcat (@var{array1}, @var{array2}, @dots{}, @var{arrayN})
Return the horizontal concatenation of N-D array objects, @var{array1},
@var{array2}, @dots{}, @var{arrayN} along dimension 2.

Arrays may also be concatenated horizontally using the syntax for creating
new matrices.  For example:

@example
@var{A} = [ @var{array1}, @var{array2}, @dots{} ]
@end example

This syntax is slightly more efficient because the Octave parser can
concatenate the arrays without the overhead of a function call.
@xseealso{@ref{XREFcat,,cat}, @ref{XREFvertcat,,vertcat}}
@end deftypefn


@c vertcat libinterp/corefcn/data.cc
@anchor{XREFvertcat}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{A} =} vertcat (@var{array1}, @var{array2}, @dots{}, @var{arrayN})
Return the vertical concatenation of N-D array objects, @var{array1},
@var{array2}, @dots{}, @var{arrayN} along dimension 1.

Arrays may also be concatenated vertically using the syntax for creating
new matrices.  For example:

@example
@var{A} = [ @var{array1}; @var{array2}; @dots{} ]
@end example

This syntax is slightly more efficient because the Octave parser can
concatenate the arrays without the overhead of a function call.
@xseealso{@ref{XREFcat,,cat}, @ref{XREFhorzcat,,horzcat}}
@end deftypefn


@c permute libinterp/corefcn/data.cc
@anchor{XREFpermute}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{B} =} permute (@var{A}, @var{perm})
Return the generalized transpose for an N-D array object @var{A}.

The permutation vector @var{perm} must contain the elements
@w{@code{1:ndims (A)}}@ (in any order, but each element must appear only
once).  The @var{N}th dimension of @var{A} gets remapped to dimension
@code{@var{PERM}(@var{N})}.  For example:

@example
@group
@var{x} = zeros ([2, 3, 5, 7]);
size (@var{x})
   @result{}  2   3   5   7

size (permute (@var{x}, [2, 1, 3, 4]))
   @result{}  3   2   5   7

size (permute (@var{x}, [1, 3, 4, 2]))
   @result{}  2   5   7   3

## The identity permutation
size (permute (@var{x}, [1, 2, 3, 4]))
   @result{}  2   3   5   7
@end group
@end example
@xseealso{@ref{XREFipermute,,ipermute}}
@end deftypefn


@c ipermute libinterp/corefcn/data.cc
@anchor{XREFipermute}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{A} =} ipermute (@var{B}, @var{iperm})
The inverse of the @code{permute} function.

The expression

@example
ipermute (permute (A, perm), perm)
@end example

@noindent
returns the original array @var{A}.
@xseealso{@ref{XREFpermute,,permute}}
@end deftypefn


@c reshape libinterp/corefcn/data.cc
@anchor{XREFreshape}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{B} =} reshape (@var{A}, @var{m}, @var{n}, @dots{})
@deftypefnx {} {@var{B} =} reshape (@var{A}, [@var{m} @var{n} @dots{}])
@deftypefnx {} {@var{B} =} reshape (@var{A}, @dots{}, [], @dots{})
@deftypefnx {} {@var{B} =} reshape (@var{A}, @var{size})
Return a matrix with the specified dimensions (@var{m}, @var{n}, @dots{})
whose elements are taken from the matrix @var{A}.

The elements of the matrix are accessed in column-major order (like Fortran
arrays are stored).

The following code demonstrates reshaping a 1x4 row vector into a 2x2 square
matrix.

@example
@group
reshape ([1, 2, 3, 4], 2, 2)
      @result{}  1  3
          2  4
@end group
@end example

@noindent
Note that the total number of elements in the original matrix
(@code{prod (size (@var{A}))}) must match the total number of elements
in the new matrix (@code{prod ([@var{m} @var{n} @dots{}])}).

A single dimension of the return matrix may be left unspecified and Octave
will determine its size automatically.  An empty matrix ([]) is used to flag
the unspecified dimension.
@xseealso{@ref{XREFresize,,resize}, @ref{XREFvec,,vec}, @ref{XREFpostpad,,postpad}, @ref{XREFcat,,cat}, @ref{XREFsqueeze,,squeeze}}
@end deftypefn


@c resize libinterp/corefcn/data.cc
@anchor{XREFresize}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{B} =} resize (@var{A}, @var{m})
@deftypefnx {} {@var{B} =} resize (@var{A}, @var{m}, @var{n}, @dots{})
@deftypefnx {} {@var{B} =} resize (@var{A}, [@var{m} @var{n} @dots{}])
Resize @var{A} cutting off elements as necessary.

In the result, element with certain indices is equal to the corresponding
element of @var{A} if the indices are within the bounds of @var{A}; otherwise,
the element is set to zero.

In other words, the statement

@example
B = resize (A, dv)
@end example

@noindent
is equivalent to the following code:

@example
@group
B = zeros (dv, class (A));
sz = min (dv, size (A));
for i = 1:length (sz)
  idx@{i@} = 1:sz(i);
endfor
B(idx@{:@}) = A(idx@{:@});
@end group
@end example

@noindent
but is performed more efficiently.

If only @var{m} is supplied, and it is a scalar, the dimension of the result is
@var{m}-by-@var{m}.  If @var{m}, @var{n}, @dots{} are all scalars, then the
dimensions of the result are @var{m}-by-@var{n}-by-@enddots{}  If given a vector
as input, then the dimensions of the result are given by the elements of that
vector.

An object can be resized to more dimensions than it has; in such case the
missing dimensions are assumed to be 1.  Resizing an object to fewer dimensions
is not possible.
@xseealso{@ref{XREFreshape,,reshape}, @ref{XREFpostpad,,postpad}, @ref{XREFprepad,,prepad}, @ref{XREFcat,,cat}}
@end deftypefn


@c circshift scripts/general/circshift.m
@anchor{XREFcircshift}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{y} =} circshift (@var{x}, @var{n})
@deftypefnx {} {@var{y} =} circshift (@var{x}, @var{n}, @var{dim})
Circularly shift the values of the array @var{x}.

@var{n} must be a vector of integers no longer than the number of dimensions
in @var{x}.  The values of @var{n} can be either positive or negative, which
determines the direction in which the values of @var{x} are shifted.  If an
element of @var{n} is zero, then the corresponding dimension of @var{x} will
not be shifted.  If @var{n} is a scalar and no @var{dim} is specified then
the shift is applied to the first non-singular dimension.

If a scalar @var{dim} is given then operate along the specified dimension.
In this case @var{n} must be a scalar as well.

Examples:

@example
x = [1, 2, 3;
     4, 5, 6;
     7, 8, 9];
## positive shift on rows (1st non-singular dim)
circshift (x, 1)
  @result{}
       7   8   9
       1   2   3
       4   5   6
## negative shift on rows (1st non-singular dim)
circshift (x, -2)
  @result{}
       7   8   9
       1   2   3
       4   5   6
## no shift of rows, shift columns by 1 (2nd dimension)
circshift (x, [0,1])
  @result{}
       3   1   2
       6   4   5
       9   7   8
## shift columns (2nd dimension)
circshift (x, 1, 2)
  @result{}
       3   1   2
       6   4   5
       9   7   8
@end example
@xseealso{@ref{XREFpermute,,permute}, @ref{XREFipermute,,ipermute}, @ref{XREFshiftdim,,shiftdim}}
@end deftypefn


@c shiftdim scripts/general/shiftdim.m
@anchor{XREFshiftdim}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{y} =} shiftdim (@var{x}, @var{n})
@deftypefnx {} {[@var{y}, @var{ns}] =} shiftdim (@var{x})
Shift the dimensions of @var{x} by @var{n}, where @var{n} must be
an integer scalar.

When @var{n} is positive, the dimensions of @var{x} are shifted to the left,
with the leading dimensions circulated to the end.  If @var{n} is negative,
then the dimensions of @var{x} are shifted to the right, with @var{n}
leading singleton dimensions added.

Called with a single argument, @code{shiftdim}, removes the leading
singleton dimensions, returning the number of dimensions removed in the
second output argument @var{ns}.

For example:

@example
@group
x = ones (1, 2, 3);
size (shiftdim (x, -1))
  @result{}   1   1   2   3
size (shiftdim (x, 1))
  @result{}   2   3
[b, ns] = shiftdim (x)
  @result{} b =
        1   1   1
        1   1   1
  @result{} ns = 1
@end group
@end example
@xseealso{@ref{XREFreshape,,reshape}, @ref{XREFpermute,,permute}, @ref{XREFipermute,,ipermute}, @ref{XREFcircshift,,circshift}, @ref{XREFsqueeze,,squeeze}}
@end deftypefn


@c sort libinterp/corefcn/data.cc
@anchor{XREFsort}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {[@var{s}, @var{i}] =} sort (@var{x})
@deftypefnx {} {[@var{s}, @var{i}] =} sort (@var{x}, @var{dim})
@deftypefnx {} {[@var{s}, @var{i}] =} sort (@var{x}, @var{mode})
@deftypefnx {} {[@var{s}, @var{i}] =} sort (@var{x}, @var{dim}, @var{mode})
Return a copy of @var{x} with the elements arranged in increasing order.

For matrices, @code{sort} orders the elements within columns

For example:

@example
@group
sort ([1, 2; 2, 3; 3, 1])
   @result{}  1  1
       2  2
       3  3
@end group
@end example

If the optional argument @var{dim} is given, then the matrix is sorted
along the dimension defined by @var{dim}.  The optional argument @var{mode}
defines the order in which the values will be sorted.  Valid values of
@var{mode} are @qcode{"ascend"} or @qcode{"descend"}.

The @code{sort} function may also be used to produce a matrix
containing the original row indices of the elements in the sorted
matrix.  For example:

@example
@group
[s, i] = sort ([1, 2; 2, 3; 3, 1])
  @result{} s = 1  1
         2  2
         3  3
  @result{} i = 1  3
         2  1
         3  2
@end group
@end example

For equal elements, the indices are such that equal elements are listed
in the order in which they appeared in the original list.

Sorting of complex entries is done first by magnitude
(@w{@code{abs (@var{z})}})@ and for any ties by phase angle
(@w{@code{angle (z)}}).  For example:

@example
@group
sort ([1+i; 1; 1-i])
    @result{} 1 + 0i
       1 - 1i
       1 + 1i
@end group
@end example

NaN values are treated as being greater than any other value and are sorted
to the end of the list.

The @code{sort} function may also be used to sort strings and cell arrays
of strings, in which case ASCII dictionary order (uppercase 'A' precedes
lowercase 'a') of the strings is used.

The algorithm used in @code{sort} is optimized for the sorting of partially
ordered lists.
@xseealso{@ref{XREFsortrows,,sortrows}, @ref{XREFissorted,,issorted}}
@end deftypefn


@c sortrows scripts/general/sortrows.m
@anchor{XREFsortrows}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {[@var{s}, @var{i}] =} sortrows (@var{A})
@deftypefnx {} {[@var{s}, @var{i}] =} sortrows (@var{A}, @var{c})
Sort the rows of the matrix @var{A} according to the order of the columns
specified in @var{c}.

By default (@var{c} omitted, or a particular column unspecified in @var{c})
an ascending sort order is used.  However, if elements of @var{c} are
negative then the corresponding column is sorted in descending order.  If
the elements of @var{A} are strings then a lexicographical sort is used.

Example: sort by column 2 in descending order, then 3 in ascending order

@example
@group
x = [ 7, 1, 4;
      8, 3, 5;
      9, 3, 6 ];
sortrows (x, [-2, 3])
   @result{} 8  3  5
      9  3  6
      7  1  4
@end group
@end example

@xseealso{@ref{XREFsort,,sort}}
@end deftypefn


@c issorted libinterp/corefcn/data.cc
@anchor{XREFissorted}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{tf} =} issorted (@var{A})
@deftypefnx {} {@var{tf} =} issorted (@var{A}, @var{mode})
@deftypefnx {} {@var{tf} =} issorted (@var{A}, "rows", @var{mode})
Return true if the vector @var{A} is sorted according to @var{mode}, which
may be either @qcode{"ascend"}, @qcode{"descend"}, @qcode{"either"}, or
@qcode{"monotonic"} (@qcode{"either"} and @qcode{"monotonic"} are
equivalent).

By default, @var{mode} is @qcode{"ascend"}.  NaNs are treated in the same
manner as @code{sort}.

If the optional argument @qcode{"rows"} is supplied, check whether the matrix
is sorted by rows as output by the function @code{sortrows} (with no options).

This function does not support sparse matrices.
@xseealso{@ref{XREFsort,,sort}, @ref{XREFsortrows,,sortrows}}
@end deftypefn


@c nth_element libinterp/corefcn/data.cc
@anchor{XREFnth_element}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{nel} =} nth_element (@var{x}, @var{n})
@deftypefnx {} {@var{nel} =} nth_element (@var{x}, @var{n}, @var{dim})
Select the n-th smallest element of a vector, using the ordering defined by
@code{sort}.

The result is equivalent to @code{sort(@var{x})(@var{n})}.

@var{n} can also be a contiguous range, either ascending @code{l:u}
or descending @code{u:-1:l}, in which case a range of elements is returned.

If @var{x} is an array, @code{nth_element} operates along the dimension
defined by @var{dim}, or the first non-singleton dimension if @var{dim} is
not given.

Programming Note: nth_element encapsulates the C++ standard library
algorithms nth_element and partial_sort.  On average, the complexity of the
operation is O(M*log(K)), where @w{@code{M = size (@var{x}, @var{dim})}}@ and
@w{@code{K = length (@var{n})}}.  This function is intended for cases where
the ratio K/M is small; otherwise, it may be better to use @code{sort}.
@xseealso{@ref{XREFsort,,sort}, @ref{XREFmin,,min}, @ref{XREFmax,,max}}
@end deftypefn


@c tril libinterp/corefcn/tril.cc
@anchor{XREFtril}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{A_LO} =} tril (@var{A})
@deftypefnx {} {@var{A_LO} =} tril (@var{A}, @var{k})
@deftypefnx {} {@var{A_LO} =} tril (@var{A}, @var{k}, @var{pack})
Return a new matrix formed by extracting the lower triangular part of the
matrix @var{A}, and setting all other elements to zero.

The optional second argument specifies how many diagonals above or below the
main diagonal should also be set to zero.  The default value of @var{k} is
zero which includes the main diagonal as part of the result.  If the value of
@var{k} is a nonzero integer then the selection of elements starts at an offset
of @var{k} diagonals above the main diagonal for positive @var{k} or below the
main diagonal for negative @var{k}.  The absolute value of @var{k} may not be
greater than the number of subdiagonals or superdiagonals.

Example 1 : exclude main diagonal

@example
@group
tril (ones (3), -1)
     @result{}  0  0  0
         1  0  0
         1  1  0
@end group
@end example

@noindent

Example 2 : include first superdiagonal

@example
@group
tril (ones (3), 1)
     @result{}  1  1  0
         1  1  1
         1  1  1
@end group
@end example

If the optional third argument @qcode{"pack"} is given then the extracted
elements are not inserted into a matrix, but instead stacked column-wise one
above another, and returned as a column vector.
@xseealso{@ref{XREFtriu,,triu}, @ref{XREFistril,,istril}, @ref{XREFdiag,,diag}}
@end deftypefn


@c triu libinterp/corefcn/tril.cc
@anchor{XREFtriu}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{A_UP} =} triu (@var{A})
@deftypefnx {} {@var{A_UP} =} triu (@var{A}, @var{k})
@deftypefnx {} {@var{A_UP} =} triu (@var{A}, @var{k}, @var{pack})
Return a new matrix formed by extracting the upper triangular part of the
matrix @var{A}, and setting all other elements to zero.

The optional second argument specifies how many diagonals above or below the
main diagonal should also be set to zero.  The default value of @var{k} is
zero which includes the main diagonal as part of the result.  If the value of
@var{k} is a nonzero integer then the selection of elements starts at an offset
of @var{k} diagonals above the main diagonal for positive @var{k} or below the
main diagonal for negative @var{k}.  The absolute value of @var{k} may not be
greater than the number of subdiagonals or superdiagonals.

Example 1 : exclude main diagonal

@example
@group
triu (ones (3), 1)
     @result{}  0  1  1
         0  0  1
         0  0  0
@end group
@end example

@noindent

Example 2 : include first subdiagonal

@example
@group
triu (ones (3), -1)
     @result{}  1  1  1
         1  1  1
         0  1  1
@end group
@end example

If the optional third argument @qcode{"pack"} is given then the extracted
elements are not inserted into a matrix, but instead stacked column-wise one
above another, and returned as a column vector.
@xseealso{@ref{XREFtril,,tril}, @ref{XREFistriu,,istriu}, @ref{XREFdiag,,diag}}
@end deftypefn


@c vec libinterp/corefcn/data.cc
@anchor{XREFvec}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{v} =} vec (@var{x})
@deftypefnx {} {@var{v} =} vec (@var{x}, @var{dim})
Return the vector obtained by stacking the columns of the matrix @var{x}
one above the other.

Without @var{dim} this is equivalent to @code{@var{x}(:)}.

If @var{dim} is supplied, the dimensions of @var{v} are set to @var{dim}
with all elements along the last dimension.  This is equivalent to
@code{shiftdim (@var{x}(:), 1-@var{dim})}.
@xseealso{@ref{XREFvech,,vech}, @ref{XREFresize,,resize}, @ref{XREFcat,,cat}}
@end deftypefn


@c vech scripts/linear-algebra/vech.m
@anchor{XREFvech}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{v} =} vech (@var{x})
Return the vector obtained by eliminating all superdiagonal elements of
the square matrix @var{x} and stacking the result one column above the
other.

This has uses in matrix calculus where the underlying matrix is symmetric
and it would be pointless to keep values above the main diagonal.
@xseealso{@ref{XREFvec,,vec}}
@end deftypefn


@c prepad scripts/general/prepad.m
@anchor{XREFprepad}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{B} =} prepad (@var{A}, @var{l})
@deftypefnx {} {@var{B} =} prepad (@var{A}, @var{l}, @var{c})
@deftypefnx {} {@var{B} =} prepad (@var{A}, @var{l}, @var{c}, @var{dim})
Prepend the scalar value @var{c} to the vector @var{A} until it is of length
@var{l}.  If @var{c} is not given, a value of 0 is used.

If @code{length (@var{A}) > @var{l}}, elements from the beginning of @var{A}
are removed until a vector of length @var{l} is obtained.

If @var{A} is a matrix, elements are prepended or removed from each row.

If the optional argument @var{dim} is given, operate along this dimension.

If @var{dim} is larger than the dimensions of @var{A}, the result will have
@var{dim} dimensions.
@xseealso{@ref{XREFpostpad,,postpad}, @ref{XREFcat,,cat}, @ref{XREFresize,,resize}}
@end deftypefn


@c postpad scripts/general/postpad.m
@anchor{XREFpostpad}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{B} =} postpad (@var{A}, @var{l})
@deftypefnx {} {@var{B} =} postpad (@var{A}, @var{l}, @var{c})
@deftypefnx {} {@var{B} =} postpad (@var{A}, @var{l}, @var{c}, @var{dim})
Append the scalar value @var{c} to the vector @var{A} until it is of length
@var{l}.  If @var{c} is not given, a value of 0 is used.

If @code{length (@var{A}) > @var{l}}, elements from the end of @var{A} are
removed until a vector of length @var{l} is obtained.

If @var{A} is a matrix, elements are appended or removed from each row.

If the optional argument @var{dim} is given, operate along this dimension.

If @var{dim} is larger than the dimensions of @var{A}, the result will have
@var{dim} dimensions.
@xseealso{@ref{XREFprepad,,prepad}, @ref{XREFcat,,cat}, @ref{XREFresize,,resize}}
@end deftypefn


@c diag libinterp/corefcn/data.cc
@anchor{XREFdiag}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{M} =} diag (@var{v})
@deftypefnx {} {@var{M} =} diag (@var{v}, @var{k})
@deftypefnx {} {@var{M} =} diag (@var{v}, @var{m}, @var{n})
@deftypefnx {} {@var{v} =} diag (@var{M})
@deftypefnx {} {@var{v} =} diag (@var{M}, @var{k})
Return a diagonal matrix with vector @var{v} on diagonal @var{k}.

The second argument is optional.  If it is positive, the vector is placed on
the @var{k}-th superdiagonal.  If it is negative, it is placed on the
@var{-k}-th subdiagonal.  The default value of @var{k} is 0, and the vector
is placed on the main diagonal.  For example:

@example
@group
diag ([1, 2, 3], 1)
   @result{}  0  1  0  0
       0  0  2  0
       0  0  0  3
       0  0  0  0
@end group
@end example

@noindent
The 3-input form returns a diagonal matrix with vector @var{v} on the main
diagonal and the resulting matrix being of size @var{m} rows x @var{n}
columns.

Given a matrix argument, instead of a vector, @code{diag} extracts the
@var{k}-th diagonal of the matrix.
@end deftypefn


@c blkdiag scripts/general/blkdiag.m
@anchor{XREFblkdiag}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{M} =} blkdiag (@var{A}, @var{B}, @var{C}, @dots{})
Build a block diagonal matrix from @var{A}, @var{B}, @var{C}, @enddots{}

All arguments must be numeric and either two-dimensional matrices or
scalars.  If any argument is of type sparse, the output will also be sparse.
@xseealso{@ref{XREFdiag,,diag}, @ref{XREFhorzcat,,horzcat}, @ref{XREFvertcat,,vertcat}, @ref{XREFsparse,,sparse}}
@end deftypefn


@node Special Utility Matrices
@section Special Utility Matrices

@c eye libinterp/corefcn/data.cc
@anchor{XREFeye}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{I} =} eye (@var{n})
@deftypefnx {} {@var{I} =} eye (@var{m}, @var{n})
@deftypefnx {} {@var{I} =} eye ([@var{m} @var{n}])
@deftypefnx {} {@var{I} =} eye (@dots{}, @var{class})
Return an identity matrix.

If invoked with a single scalar argument @var{n}, return a square
@nospell{NxN} identity matrix.

If supplied two scalar arguments (@var{m}, @var{n}), @code{eye} takes them
to be the number of rows and columns.  If given a vector with two elements,
@code{eye} uses the values of the elements as the number of rows and
columns, respectively.  For example:

@example
@group
eye (3)
 @result{}  1  0  0
     0  1  0
     0  0  1
@end group
@end example

The following expressions all produce the same result:

@example
@group
eye (2)
@equiv{}
eye (2, 2)
@equiv{}
eye (size ([1, 2; 3, 4]))
@end group
@end example

The optional argument @var{class}, allows @code{eye} to return an array of
the specified type, like

@example
val = zeros (n,m, "uint8")
@end example

Calling @code{eye} with no arguments is equivalent to calling it with an
argument of 1.  Any negative dimensions are treated as zero.  These odd
definitions are for compatibility with @sc{matlab}.
@xseealso{@ref{XREFspeye,,speye}, @ref{XREFones,,ones}, @ref{XREFzeros,,zeros}}
@end deftypefn


@c ones libinterp/corefcn/data.cc
@anchor{XREFones}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{val} =} ones (@var{n})
@deftypefnx {} {@var{val} =} ones (@var{m}, @var{n})
@deftypefnx {} {@var{val} =} ones (@var{m}, @var{n}, @var{k}, @dots{})
@deftypefnx {} {@var{val} =} ones ([@var{m} @var{n} @dots{}])
@deftypefnx {} {@var{val} =} ones (@dots{}, "like", @var{var})
@deftypefnx {} {@var{val} =} ones (@dots{}, @var{class})
Return a matrix or N-dimensional array whose elements are all 1.

If invoked with a single scalar integer argument @var{n}, return a square
@nospell{NxN} matrix.

If invoked with two or more scalar integer arguments, or a vector of integer
values, return an array with the given dimensions.

To create a constant matrix whose values are all the same use an expression
such as

@example
val_matrix = val * ones (m, n)
@end example

If a variable @var{var} is specified after @qcode{"like"}, the output @var{val}
will have the same data type, complexity, and sparsity as @var{var}.

The optional argument @var{class} specifies the class of the return array
and defaults to double.  For example:

@example
val = ones (m,n, "uint8")
@end example
@xseealso{@ref{XREFzeros,,zeros}}
@end deftypefn


@c zeros libinterp/corefcn/data.cc
@anchor{XREFzeros}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{val} =} zeros (@var{n})
@deftypefnx {} {@var{val} =} zeros (@var{m}, @var{n})
@deftypefnx {} {@var{val} =} zeros (@var{m}, @var{n}, @var{k}, @dots{})
@deftypefnx {} {@var{val} =} zeros ([@var{m} @var{n} @dots{}])
@deftypefnx {} {@var{val} =} zeros (@dots{}, "like", @var{var})
@deftypefnx {} {@var{val} =} zeros (@dots{}, @var{class})
Return a matrix or N-dimensional array whose elements are all 0.

If invoked with a single scalar integer argument, return a square
@nospell{NxN} matrix.

If invoked with two or more scalar integer arguments, or a vector of integer
values, return an array with the given dimensions.

If a variable @var{var} is specified after @qcode{"like"}, the output @var{val}
will have the same data type, complexity, and sparsity as @var{var}.

The optional argument @var{class} specifies the class of the return array
and defaults to double.  For example:

@example
val = zeros (m,n, "uint8")
@end example
@xseealso{@ref{XREFones,,ones}}
@end deftypefn


@c repmat scripts/general/repmat.m
@anchor{XREFrepmat}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{B} =} repmat (@var{A}, @var{m})
@deftypefnx {} {@var{B} =} repmat (@var{A}, @var{m}, @var{n})
@deftypefnx {} {@var{B} =} repmat (@var{A}, @var{m}, @var{n}, @var{p} @dots{})
@deftypefnx {} {@var{B} =} repmat (@var{A}, [@var{m} @var{n}])
@deftypefnx {} {@var{B} =} repmat (@var{A}, [@var{m} @var{n} @var{p} @dots{}])
Repeat matrix or N-D array.

Form a block matrix of size @var{m} by @var{n}, with a copy of matrix
@var{A} as each element.

If @var{n} is not specified, form an @var{m} by @var{m} block matrix.  For
copying along more than two dimensions, specify the number of times to copy
across each dimension @var{m}, @var{n}, @var{p}, @dots{}, in a vector in the
second argument.

@xseealso{@ref{XREFbsxfun,,bsxfun}, @ref{XREFkron,,kron}, @ref{XREFrepelems,,repelems}}
@end deftypefn


@c repelems libinterp/corefcn/data.cc
@anchor{XREFrepelems}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{y} =} repelems (@var{x}, @var{r})
Construct a vector of repeated elements from @var{x}.

@var{r} is a 2x@var{N} integer matrix specifying which elements to repeat
and how often to repeat each element.  Entries in the first row,
@var{r}(1,j), select an element to repeat.  The corresponding entry in the
second row, @var{r}(2,j), specifies the repeat count.  If @var{x} is a
matrix then the columns of @var{x} are imagined to be stacked on top of
each other for purposes of the selection index.  A row vector is always
returned.

Conceptually the result is calculated as follows:

@example
@group
y = [];
for i = 1:columns (@var{r})
  y = [y, @var{x}(@var{r}(1,i)*ones(1, @var{r}(2,i)))];
endfor
@end group
@end example
@xseealso{@ref{XREFrepmat,,repmat}, @ref{XREFcat,,cat}}
@end deftypefn


@c repelem scripts/general/repelem.m
@anchor{XREFrepelem}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{xxx} =} repelem (@var{x}, @var{R})
@deftypefnx {} {@var{xxx} =} repelem (@var{x}, @var{R_1}, @dots{}, @var{R_n})
Construct an array of repeated elements from @var{x} and repeat instructions
@var{R_1}, @enddots{}

@var{x} must be a scalar, vector, or N-dimensional array.

A repeat instruction @var{R_j} must either be a scalar or a vector.  If the
instruction is a scalar then each component of @var{x} in dimension @var{j}
is repeated @var{R_j} times.  If the instruction is a vector then it must
have the same number of elements as the corresponding dimension @var{j} of
@var{x}.  In this case, the @var{k}th component of dimension @var{j} is
repeated @code{@var{R_j}(@var{k})} times.

If @var{x} is a scalar or vector then @code{repelem} may be called with just
a single repeat instruction @var{R} and @code{repelem} will return a vector
with the same orientation as the input.

If @var{x} is a matrix then at least two @var{R_j}s must be specified.

Note: Using @code{repelem} with a vector @var{x} and a vector for @var{R_j}
is equivalent to Run Length Decoding.

Examples:

@example
@group
A = [1 2 3 4 5];
B = [2 1 0 1 2];
repelem (A, B)
  @result{}   1   1   2   4   5   5
@end group
@end example

@example
@group
A = magic (3)
  @result{} A =
       8   1   6
       3   5   7
       4   9   2
B1 = [1 2 3];
B2 = 2;
repelem (A, B1, B2)
  @result{}     8   8   1   1   6   6
         3   3   5   5   7   7
         3   3   5   5   7   7
         4   4   9   9   2   2
         4   4   9   9   2   2
         4   4   9   9   2   2
@end group
@end example

More @var{R_j} may be specified than the number of dimensions of @var{x}.
Any excess @var{R_j} must be scalars (because @var{x}'s size in those
dimensions is only 1), and @var{x} will be replicated in those dimensions
accordingly.

@example
@group
A = [1 2 3 4 5];
B1 = 2;
B2 = [2 1 3 0 2];
B3 = 3;
repelem (A, B1, B2, B3)
  @result{}    ans(:,:,1) =
           1   1   2   3   3   3   5   5
           1   1   2   3   3   3   5   5

        ans(:,:,2) =

           1   1   2   3   3   3   5   5
           1   1   2   3   3   3   5   5

        ans(:,:,3) =
           1   1   2   3   3   3   5   5
           1   1   2   3   3   3   5   5
@end group
@end example

@var{R_j} must be specified in order.  A placeholder of 1 may be used for
dimensions which do not need replication.

@example
@group
repelem ([-1, 0; 0, 1], 1, 2, 1, 2)
  @result{}  ans(:,:,1,1) =
        -1  -1   0   0
         0   0   1   1

      ans(:,:,1,2) =
        -1  -1   0   0
         0   0   1   1
@end group
@end example

If fewer @var{R_j} are given than the number of dimensions in @var{x},
@code{repelem} will assume @var{R_j} is 1 for those dimensions.

@example
A = cat (3, [-1 0; 0 1], [-1 0; 0 1])
  @result{}  ans(:,:,1) =
        -1   0
         0   1

      ans(:,:,2) =
        -1   0
         0   1

repelem (A,2,3)
  @result{}  ans(:,:,1) =
        -1  -1  -1   0   0   0
        -1  -1  -1   0   0   0
         0   0   0   1   1   1
         0   0   0   1   1   1

      ans(:,:,2) =
        -1  -1  -1   0   0   0
        -1  -1  -1   0   0   0
         0   0   0   1   1   1
         0   0   0   1   1   1
@end example

@code{repelem} preserves the class of @var{x}, and works with strings,
cell arrays, NA, and NAN inputs.  If any @var{R_j} is 0 the output will
be an empty array.

@example
@group
repelem ("Octave", 2, 3)
  @result{}    OOOccctttaaavvveee
        OOOccctttaaavvveee

repelem ([1 2 3; 1 2 3], 2, 0)
  @result{}    [](4x0)
@end group
@end example

@xseealso{@ref{XREFcat,,cat}, @ref{XREFkron,,kron}, @ref{XREFrepmat,,repmat}}
@end deftypefn


The functions @code{linspace} and @code{logspace} make it very easy to
create vectors with evenly or logarithmically spaced elements.
@xref{Ranges}.

@c linspace libinterp/corefcn/data.cc
@anchor{XREFlinspace}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{y} =} linspace (@var{start}, @var{end})
@deftypefnx {} {@var{y} =} linspace (@var{start}, @var{end}, @var{n})
Return a row vector with @var{n} linearly spaced elements between @var{start}
and @var{end}.

If the number of elements @var{n} is greater than one, then the endpoints
@var{start} and @var{end} are always included in the range.  If @var{start} is
greater than @var{end}, the elements are stored in decreasing order.  If the
number of points @var{n} is not specified, a value of 100 is used.

The @code{linspace} function returns a row vector when both @var{start} and
@var{end} are scalars.  If one, or both, inputs are vectors, then
@code{linspace} transforms them to column vectors and returns a matrix where
each row is an independent sequence between
@w{@code{@var{start}(@var{row_n}), @var{end}(@var{row_n})}}.

Programming Notes: For compatibility with @sc{matlab}, return the second
argument (@var{end}) when a single value (@var{n} = 1) is requested.  If
@var{n} is not an integer then @code{floor (@var{n})} is used to round the
number of elements.  If @var{n} is zero or negative then an empty 1x0 matrix
is returned.
@xseealso{@ref{XREFcolon,,colon}, @ref{XREFlogspace,,logspace}}
@end deftypefn


@c logspace scripts/general/logspace.m
@anchor{XREFlogspace}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{y} =} logspace (@var{a}, @var{b})
@deftypefnx {} {@var{y} =} logspace (@var{a}, @var{b}, @var{n})
@deftypefnx {} {@var{y} =} logspace (@var{a}, pi)
@deftypefnx {} {@var{y} =} logspace (@var{a}, pi, @var{n})
Return a row vector with @var{n} elements logarithmically spaced from
@tex
$10^{a}$ to $10^{b}$.
@end tex
@ifnottex
10^@var{a} to 10^@var{b}.
@end ifnottex

If the number of elements @var{n} is unspecified it defaults to 50.

If @var{b} is equal to
@tex
$\pi$,
@end tex
@ifnottex
pi,
@end ifnottex
the points are between
@tex
$10^{a}$ and $\pi$,
@end tex
@ifnottex
10^@var{a} and pi,
@end ifnottex
@emph{not}
@tex
$10^{a}$ and $10^{\pi}$,
@end tex
@ifnottex
10^@var{a} and 10^pi,
@end ifnottex
which is useful in digital signal processing.

Programming Notes: For compatibility with @sc{matlab}, return the right-hand
side of the range
@tex
($10^{b}$)
@end tex
@ifnottex
(10^@var{b})
@end ifnottex
when a single value (@var{n} = 1) is requested.
If @var{n} is not an integer then @code{floor (@var{n})} is used to round
the number of elements.  If @var{n} is zero or negative then an empty 1x0
matrix is returned.
@xseealso{@ref{XREFlinspace,,linspace}}
@end deftypefn


@c rand libinterp/corefcn/rand.cc
@anchor{XREFrand}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{x} =} rand (@var{n})
@deftypefnx {} {@var{x} =} rand (@var{m}, @var{n}, @dots{})
@deftypefnx {} {@var{x} =} rand ([@var{m} @var{n} @dots{}])
@deftypefnx {} {@var{x} =} rand (@dots{}, "single")
@deftypefnx {} {@var{x} =} rand (@dots{}, "double")
@deftypefnx {} {@var{v} =} rand ("state")
@deftypefnx {} {} rand ("state", @var{v})
@deftypefnx {} {} rand ("state", "reset")
@deftypefnx {} {@var{v} =} rand ("seed")
@deftypefnx {} {} rand ("seed", @var{v})
@deftypefnx {} {} rand ("seed", "reset")
Return a matrix with random elements uniformly distributed on the
interval (0, 1).

The arguments are handled the same as the arguments for @code{eye}.

You can query the state of the random number generator using the form

@example
v = rand ("state")
@end example

This returns a column vector @var{v} of length 625.  Later, you can restore
the random number generator to the state @var{v} using the form

@example
rand ("state", v)
@end example

@noindent
You may also initialize the state vector from an arbitrary vector of length
@leq{} 625 for @var{v}.  This new state will be a hash based on the value of
@var{v}, not @var{v} itself.

By default, the generator is initialized by contributing entropy from the
wall clock time, the CPU time, the current fraction of a second, the process
ID and---if available---up to 1024 bits from the C++ random numbers source
@code{random_device}, which might be non-deterministic (implementation
specific).  Note that this differs from @sc{matlab}, which always initializes
the state to the same state at startup.  To obtain behavior comparable to
@sc{matlab}, initialize with a deterministic state vector in Octave's startup
files (@pxref{Startup Files}).

To compute the pseudo-random sequence, @code{rand} uses the Mersenne
Twister with a period of @math{2^{19937}-1}
(See @nospell{M. Matsumoto and T. Nishimura},
@cite{Mersenne Twister: A 623-dimensionally equidistributed uniform
pseudorandom number generator},
@nospell{ACM} Trans.@: on Modeling and Computer Simulation Vol.@: 8, No.@: 1,
pp.@: 3--30, January 1998,
@url{http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html}).
Do @strong{not} use for cryptography without securely hashing several
returned values together, otherwise the generator state can be learned after
reading 624 consecutive values.

Older versions of Octave used a different random number generator.
The new generator is used by default as it is significantly faster than the
old generator, and produces random numbers with a significantly longer cycle
time.  However, in some circumstances it might be desirable to obtain the
same random sequences as produced by the old generators.  To do this the
keyword @qcode{"seed"} is used to specify that the old generators should
be used, as in

@example
rand ("seed", val)
@end example

@noindent
which sets the seed of the generator to @var{val}.  The seed of the
generator can be queried with

@example
s = rand ("seed")
@end example

However, it should be noted that querying the seed will not cause
@code{rand} to use the old generators, only setting the seed will.  To cause
@code{rand} to once again use the new generators, the keyword
@qcode{"state"} should be used to reset the state of the @code{rand}.

The state or seed of the generator can be reset to a new random value using
the @qcode{"reset"} keyword.

The class of the value returned can be controlled by a trailing
@qcode{"double"} or @qcode{"single"} argument.  These are the only valid
classes.
@xseealso{@ref{XREFrandn,,randn}, @ref{XREFrande,,rande}, @ref{XREFrandg,,randg}, @ref{XREFrandp,,randp}}
@end deftypefn


@c randi scripts/general/randi.m
@anchor{XREFrandi}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{R} =} randi (@var{imax})
@deftypefnx {} {@var{R} =} randi (@var{imax}, @var{n})
@deftypefnx {} {@var{R} =} randi (@var{imax}, @var{m}, @var{n}, @dots{})
@deftypefnx {} {@var{R} =} randi ([@var{imin} @var{imax}], @dots{})
@deftypefnx {} {@var{R} =} randi (@dots{}, "@var{class}")
Return random integers in the range 1:@var{imax}.

Additional arguments determine the shape of the return matrix.  When no
arguments are specified a single random integer is returned.  If one
argument @var{n} is specified then a square matrix @w{(@var{n} x @var{n})}
is returned.  Two or more arguments will return a multi-dimensional matrix
@w{(@var{m} x @var{n} x @dots{})}.

The integer range may optionally be described by a two-element matrix with a
lower and upper bound in which case the returned integers will be on the
interval @w{[@var{imin}, @var{imax}]}.

The optional argument @var{class} will return a matrix of the requested
type.  The default is @qcode{"double"}.

The following example returns 150 integers in the range 1--10.

@example
ri = randi (10, 150, 1)
@end example

Implementation Note: @code{randi} relies internally on @code{rand} which
uses class @qcode{"double"} to represent numbers.  This limits the maximum
integer (@var{imax}) and range (@var{imax} - @var{imin}) to the value
returned by the @code{flintmax} function.  For IEEE@tie{}754 floating point
numbers this value is @w{@math{2^{53} - 1}}.

@xseealso{@ref{XREFrand,,rand}, @ref{XREFrandn,,randn}}
@end deftypefn


@c randn libinterp/corefcn/rand.cc
@anchor{XREFrandn}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{x} =} randn (@var{n})
@deftypefnx {} {@var{x} =} randn (@var{m}, @var{n}, @dots{})
@deftypefnx {} {@var{x} =} randn ([@var{m} @var{n} @dots{}])
@deftypefnx {} {@var{x} =} randn (@dots{}, "single")
@deftypefnx {} {@var{x} =} randn (@dots{}, "double")
@deftypefnx {} {@var{v} =} randn ("state")
@deftypefnx {} {} randn ("state", @var{v})
@deftypefnx {} {} randn ("state", "reset")
@deftypefnx {} {@var{v} =} randn ("seed")
@deftypefnx {} {} randn ("seed", @var{v})
@deftypefnx {} {} randn ("seed", "reset")
Return a matrix with normally distributed random elements having zero mean
and variance one.

The arguments are handled the same as the arguments for @code{rand}.

By default, @code{randn} uses the @nospell{Marsaglia and Tsang}
``Ziggurat technique'' to transform from a uniform to a normal distribution.

The class of the value returned can be controlled by a trailing
@qcode{"double"} or @qcode{"single"} argument.  These are the only valid
classes.

Reference: @nospell{G. Marsaglia and W.W. Tsang},
@cite{Ziggurat Method for Generating Random Variables},
J. Statistical Software, vol 5, 2000,
@url{https://www.jstatsoft.org/v05/i08/}

@xseealso{@ref{XREFrand,,rand}, @ref{XREFrande,,rande}, @ref{XREFrandg,,randg}, @ref{XREFrandp,,randp}}
@end deftypefn


@c rande libinterp/corefcn/rand.cc
@anchor{XREFrande}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{x} =} rande (@var{n})
@deftypefnx {} {@var{x} =} rande (@var{m}, @var{n}, @dots{})
@deftypefnx {} {@var{x} =} rande ([@var{m} @var{n} @dots{}])
@deftypefnx {} {@var{x} =} rande (@dots{}, "single")
@deftypefnx {} {@var{x} =} rande (@dots{}, "double")
@deftypefnx {} {@var{v} =} rande ("state")
@deftypefnx {} {} rande ("state", @var{v})
@deftypefnx {} {} rande ("state", "reset")
@deftypefnx {} {@var{v} =} rande ("seed")
@deftypefnx {} {} rande ("seed", @var{v})
@deftypefnx {} {} rande ("seed", "reset")
Return a matrix with exponentially distributed random elements.

The arguments are handled the same as the arguments for @code{rand}.

By default, @code{rande} uses the @nospell{Marsaglia and Tsang}
``Ziggurat technique'' to transform from a uniform to an exponential
distribution.

The class of the value returned can be controlled by a trailing
@qcode{"double"} or @qcode{"single"} argument.  These are the only valid
classes.

Reference: @nospell{G. Marsaglia and W.W. Tsang},
@cite{Ziggurat Method for Generating Random Variables},
J. Statistical Software, vol 5, 2000,
@url{https://www.jstatsoft.org/v05/i08/}

@xseealso{@ref{XREFrand,,rand}, @ref{XREFrandn,,randn}, @ref{XREFrandg,,randg}, @ref{XREFrandp,,randp}}
@end deftypefn


@c randp libinterp/corefcn/rand.cc
@anchor{XREFrandp}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{x} =} randp (@var{l}, @var{n})
@deftypefnx {} {@var{x} =} randp (@var{l}, @var{m}, @var{n}, @dots{})
@deftypefnx {} {@var{x} =} randp (@var{l}, [@var{m} @var{n} @dots{}])
@deftypefnx {} {@var{x} =} randp (@dots{}, "single")
@deftypefnx {} {@var{x} =} randp (@dots{}, "double")
@deftypefnx {} {@var{v} =} randp ("state")
@deftypefnx {} {} randp ("state", @var{v})
@deftypefnx {} {} randp ("state", "reset")
@deftypefnx {} {@var{v} =} randp ("seed")
@deftypefnx {} {} randp ("seed", @var{v})
@deftypefnx {} {} randp ("seed", "reset")
Return a matrix with Poisson distributed random elements with mean value
parameter given by the first argument, @var{l}.

The arguments are handled the same as the arguments for @code{rand}, except
for the argument @var{l}.

Five different algorithms are used depending on the range of @var{l} and
whether or not @var{l} is a scalar or a matrix.

@table @asis
@item For scalar @var{l} @leq{} 12, use direct method.
W.H. Press, et al., @cite{Numerical Recipes in C},
Cambridge University Press, 1992.

@item For scalar @var{l} > 12, use rejection method.[1]
W.H. Press, et al., @cite{Numerical Recipes in C},
Cambridge University Press, 1992.

@item For matrix @var{l} @leq{} 10, use inversion method.[2]
@nospell{E. Stadlober, et al., WinRand source code}, available via FTP.

@item For matrix @var{l} > 10, use patchwork rejection method.
@nospell{E. Stadlober, et al., WinRand source code}, available via FTP, or
@nospell{H. Zechner}, @cite{Efficient sampling from continuous and discrete
unimodal distributions}, Doctoral Dissertation, 156pp., Technical
University @nospell{Graz}, Austria, 1994.

@item For @var{l} > 1e8, use normal approximation.
@nospell{L. Montanet}, et al., @cite{Review of Particle Properties},
Physical Review D 50 p1284, 1994.
@end table

The class of the value returned can be controlled by a trailing
@qcode{"double"} or @qcode{"single"} argument.  These are the only valid
classes.
@xseealso{@ref{XREFrand,,rand}, @ref{XREFrandn,,randn}, @ref{XREFrande,,rande}, @ref{XREFrandg,,randg}}
@end deftypefn


@c randg libinterp/corefcn/rand.cc
@anchor{XREFrandg}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{x} =} randg (@var{a}, @var{n})
@deftypefnx {} {@var{x} =} randg (@var{a}, @var{m}, @var{n}, @dots{})
@deftypefnx {} {@var{x} =} randg (@var{a}, [@var{m} @var{n} @dots{}])
@deftypefnx {} {@var{x} =} randg (@dots{}, "single")
@deftypefnx {} {@var{x} =} randg (@dots{}, "double")
@deftypefnx {} {@var{v} =} randg ("state")
@deftypefnx {} {} randg ("state", @var{v})
@deftypefnx {} {} randg ("state", "reset")
@deftypefnx {} {@var{v} =} randg ("seed")
@deftypefnx {} {} randg ("seed", @var{v})
@deftypefnx {} {} randg ("seed", "reset")

Return a matrix with @code{gamma (@var{a},1)} distributed random elements.

The arguments are handled the same as the arguments for @code{rand}, except
for the argument @var{a}.

This can be used to generate many distributions:

@table @asis
@item @code{gamma (a, b)} for @code{a > -1}, @code{b > 0}

@example
r = b * randg (a)
@end example

@item @code{beta (a, b)} for @code{a > -1}, @code{b > -1}

@example
@group
r1 = randg (a, 1)
r = r1 / (r1 + randg (b, 1))
@end group
@end example

@item @code{Erlang (a, n)}

@example
r = a * randg (n)
@end example

@item @code{chisq (df)} for @code{df > 0}

@example
r = 2 * randg (df / 2)
@end example

@item @code{t (df)} for @code{0 < df < inf} (use randn if df is infinite)

@example
r = randn () / sqrt (2 * randg (df / 2) / df)
@end example

@item @code{F (n1, n2)} for @code{0 < n1}, @code{0 < n2}

@example
@group
## r1 equals 1 if n1 is infinite
r1 = 2 * randg (n1 / 2) / n1
## r2 equals 1 if n2 is infinite
r2 = 2 * randg (n2 / 2) / n2
r = r1 / r2
@end group
@end example

@item negative @code{binomial (n, p)} for @code{n > 0}, @code{0 < p <= 1}

@example
r = randp ((1 - p) / p * randg (n))
@end example

@item non-central @code{chisq (df, L)}, for @code{df >= 0} and @code{L > 0}
(use chisq if @code{L = 0})

@example
@group
r = randp (L / 2)
r(r > 0) = 2 * randg (r(r > 0))
r(df > 0) += 2 * randg (df(df > 0)/2)
@end group
@end example

@item @code{Dirichlet (a1, @dots{} ak)}

@example
@group
r = (randg (a1), @dots{}, randg (ak))
r = r / sum (r)
@end group
@end example

@end table

The class of the value returned can be controlled by a trailing
@qcode{"double"} or @qcode{"single"} argument.  These are the only valid
classes.
@xseealso{@ref{XREFrand,,rand}, @ref{XREFrandn,,randn}, @ref{XREFrande,,rande}, @ref{XREFrandp,,randp}}
@end deftypefn


@c rng scripts/general/rng.m
@anchor{XREFrng}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {} rng (@var{seed})
@deftypefnx {} {} rng (@var{seed}, "@var{generator}")
@deftypefnx {} {} rng ("shuffle")
@deftypefnx {} {} rng ("shuffle", "@var{generator}")
@deftypefnx {} {} rng ("default")
@deftypefnx {} {@var{s} =} rng ()
@deftypefnx {} {} rng (@var{s})
@deftypefnx {} {@var{s} =} rng (@dots{})
Set or query the seed of the random number generator used by @code{rand} and
@code{randn}.

The input @var{seed} is a scalar numeric value used to initialize the state
vector of the random number generator.

The optional string @var{generator} specifies the type of random number
generator to be used.  Its value can be @qcode{"twister"},
@qcode{"v5uniform"}, or @qcode{"v5normal"}.  The @qcode{"twister"} keyword
is described below.  @qcode{"v5uniform"} and @qcode{"v5normal"} refer to
older versions of Octave that used to use a different random number
generator.

The state or seed of the random number generator can be reset to a new
random value using the @qcode{"shuffle"} keyword.

The random number generator can be reset to default values using the
@qcode{"default"} keyword.  The default values are to use the Mersenne
Twister generator with a seed of 0.

The optional return value @var{s} contains the state of the random number
generator at the time the function is called (i.e., before it might be
modified according to the input arguments).  It is encoded as a structure
variable with three fields: @qcode{"Type"}, @qcode{"Seed"}, and
@qcode{"State"}.  The random number generator can be restored to the state
@var{s} using @code{rng (@var{s})}.  This is useful when the identical
sequence of pseudo-random numbers is required for an algorithm.

By default, and with the @qcode{"twister"} option, pseudo-random sequences
are computed using the Mersenne Twister with a period of @math{2^{19937}-1}
(See @nospell{M. Matsumoto and T. Nishimura},
@cite{Mersenne Twister: A 623-dimensionally equidistributed uniform
pseudorandom number generator},
@nospell{ACM} Trans.@: on Modeling and Computer Simulation Vol.@: 8, No.@: 1,
pp.@: 3--30, January 1998,
@url{http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html}).
Do @strong{not} use for cryptography without securely hashing several
returned values together, otherwise the generator state can be learned after
reading 624 consecutive values.

@xseealso{@ref{XREFrand,,rand}, @ref{XREFrandn,,randn}}
@end deftypefn


The generators operate in the new or old style together, it is not
possible to mix the two.  Initializing any generator with
@qcode{"state"} or @qcode{"seed"} causes the others to switch to the
same style for future calls.

The state of each generator is independent and calls to different
generators can be interleaved without affecting the final result.  For
example,

@example
@group
rand ("state", [11, 22, 33]);
randn ("state", [44, 55, 66]);
u = rand (100, 1);
n = randn (100, 1);
@end group
@end example

@noindent
and

@example
@group
rand ("state", [11, 22, 33]);
randn ("state", [44, 55, 66]);
u = zeros (100, 1);
n = zeros (100, 1);
for i = 1:100
  u(i) = rand ();
  n(i) = randn ();
end
@end group
@end example

@noindent
produce equivalent results.  When the generators are initialized in
the old style with @qcode{"seed"} only @code{rand} and @code{randn} are
independent, because the old @code{rande}, @code{randg} and
@code{randp} generators make calls to @code{rand} and @code{randn}.

The generators are initialized with random states at start-up, so
that the sequences of random numbers are not the same each time you run
Octave.@footnote{The old versions of @code{rand} and @code{randn}
obtain their initial seeds from the system clock.} If you really do
need to reproduce a sequence of numbers exactly, you can set the state
or seed to a specific value.

If invoked without arguments, @code{rand} and @code{randn} return a
single element of a random sequence.

The original @code{rand} and @code{randn} functions use Fortran code from
@sc{ranlib}, a library of Fortran routines for random number generation,
compiled by Barry W. Brown and @nospell{James Lovato} of the Department of
Biomathematics at The University of Texas, M.D. Anderson Cancer Center,
Houston, TX 77030.

@c randperm libinterp/corefcn/rand.cc
@anchor{XREFrandperm}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{v} =} randperm (@var{n})
@deftypefnx {} {@var{v} =} randperm (@var{n}, @var{m})
Return a row vector containing a random permutation of @code{1:@var{n}}.

If @var{m} is supplied, return @var{m} unique entries, sampled without
replacement from @code{1:@var{n}}.

The complexity is O(@var{n}) in memory and O(@var{m}) in time, unless
@var{m} < @var{n}/5, in which case O(@var{m}) memory is used as well.  The
randomization is performed using rand().  All permutations are equally
likely.
@xseealso{@ref{XREFperms,,perms}}
@end deftypefn


@node Famous Matrices
@section Famous Matrices

The following functions return famous matrix forms.

@c gallery scripts/special-matrix/gallery.m
@anchor{XREFgallery}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {} gallery (@var{name})
@deftypefnx {} {} gallery (@var{name}, @var{args})
Create interesting matrices for testing.

@end deftypefn

@deftypefn  {} {@var{c} =} gallery ("cauchy", @var{x})
@deftypefnx {} {@var{c} =} gallery ("cauchy", @var{x}, @var{y})
Create a Cauchy matrix.

@end deftypefn

@deftypefn  {} {@var{c} =} gallery ("chebspec", @var{n})
@deftypefnx {} {@var{c} =} gallery ("chebspec", @var{n}, @var{k})
Create a Chebyshev spectral differentiation matrix.

@end deftypefn

@deftypefn  {} {@var{c} =} gallery ("chebvand", @var{p})
@deftypefnx {} {@var{c} =} gallery ("chebvand", @var{m}, @var{p})
Create a @nospell{Vandermonde}-like matrix for the Chebyshev polynomials.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("chow", @var{n})
@deftypefnx {} {@var{a} =} gallery ("chow", @var{n}, @var{alpha})
@deftypefnx {} {@var{a} =} gallery ("chow", @var{n}, @var{alpha}, @var{delta})
Create a Chow matrix -- a singular Toeplitz lower Hessenberg matrix.

@end deftypefn

@deftypefn {} {@var{c} =} gallery ("circul", @var{v})
Create a circulant matrix.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("clement", @var{n})
@deftypefnx {} {@var{a} =} gallery ("clement", @var{n}, @var{k})
Create a tridiagonal matrix with zero diagonal entries.

@end deftypefn

@deftypefn  {} {@var{c} =} gallery ("compar", @var{a})
@deftypefnx {} {@var{c} =} gallery ("compar", @var{a}, @var{k})
Create a comparison matrix.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("condex", @var{n})
@deftypefnx {} {@var{a} =} gallery ("condex", @var{n}, @var{k})
@deftypefnx {} {@var{a} =} gallery ("condex", @var{n}, @var{k}, @var{theta})
Create a @nospell{"counterexample"} matrix to a condition estimator.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("cycol", [@var{m} @var{n}])
@deftypefnx {} {@var{a} =} gallery ("cycol", @var{n})
@deftypefnx {} {@var{a} =} gallery (@dots{}, @var{k})
Create a matrix whose columns repeat cyclically.

@end deftypefn

@deftypefn  {} {[@var{c}, @var{d}, @var{e}] =} gallery ("dorr", @var{n})
@deftypefnx {} {[@var{c}, @var{d}, @var{e}] =} gallery ("dorr", @var{n}, @var{theta})
@deftypefnx {} {@var{a} =} gallery ("dorr", @dots{})
Create a diagonally dominant, ill-conditioned, tridiagonal matrix.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("dramadah", @var{n})
@deftypefnx {} {@var{a} =} gallery ("dramadah", @var{n}, @var{k})
Create a (0, 1) matrix whose inverse has large integer entries.

@end deftypefn

@deftypefn {} {@var{a} =} gallery ("fiedler", @var{c})
Create a symmetric @nospell{Fiedler} matrix.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("forsythe", @var{n})
@deftypefnx {} {@var{a} =} gallery ("forsythe", @var{n}, @var{alpha})
@deftypefnx {} {@var{a} =} gallery ("forsythe", @var{n}, @var{alpha}, @var{lambda})
Create a @nospell{Forsythe} matrix (a perturbed Jordan block).

@end deftypefn

@deftypefn  {} {@var{f} =} gallery ("frank", @var{n})
@deftypefnx {} {@var{f} =} gallery ("frank", @var{n}, @var{k})
Create a Frank matrix (ill-conditioned eigenvalues).

@end deftypefn

@deftypefn {} {@var{c} =} gallery ("gcdmat", @var{n})
Create a greatest common divisor matrix.

@var{c} is an @var{n}-by-@var{n} matrix whose values correspond to the
greatest common divisor of its coordinate values, i.e., @var{c}(i,j)
correspond @code{gcd (i, j)}.
@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("gearmat", @var{n})
@deftypefnx {} {@var{a} =} gallery ("gearmat", @var{n}, @var{i})
@deftypefnx {} {@var{a} =} gallery ("gearmat", @var{n}, @var{i}, @var{j})
Create a Gear matrix.

@end deftypefn

@deftypefn  {} {@var{g} =} gallery ("grcar", @var{n})
@deftypefnx {} {@var{g} =} gallery ("grcar", @var{n}, @var{k})
Create a Toeplitz matrix with sensitive eigenvalues.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("hanowa", @var{n})
@deftypefnx {} {@var{a} =} gallery ("hanowa", @var{n}, @var{d})
Create a matrix whose eigenvalues lie on a vertical line in the complex
plane.

@end deftypefn

@deftypefn  {} {@var{v} =} gallery ("house", @var{x})
@deftypefnx {} {[@var{v}, @var{beta}] =} gallery ("house", @var{x})
Create a householder matrix.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("integerdata", @var{imax}, [@var{M} @var{N} @dots{}], @var{j})
@deftypefnx {} {@var{a} =} gallery ("integerdata", @var{imax}, @var{M}, @var{N}, @dots{}, @var{j})
@deftypefnx {} {@var{a} =} gallery ("integerdata", [@var{imin}, @var{imax}], [@var{M} @var{N} @dots{}], @var{j})
@deftypefnx {} {@var{a} =} gallery ("integerdata", [@var{imin}, @var{imax}], @var{M}, @var{N}, @dots{}, @var{j})
@deftypefnx {} {@var{a} =} gallery ("integerdata", @dots{}, "@var{class}")
Create a matrix with random integers in the range [1, @var{imax}].
If @var{imin} is given then the integers are in the range
[@var{imin}, @var{imax}].

The second input is a matrix of dimensions describing the size of the
output.  The dimensions can also be input as comma-separated arguments.

The input @var{j} is an integer index in the range [0, 2^32-1].  The values
of the output matrix are always exactly the same (reproducibility) for a
given size input and @var{j} index.

The final optional argument determines the class of the resulting matrix.
Possible values for @var{class}: @qcode{"uint8"}, @qcode{"uint16"},
@qcode{"uint32"}, @qcode{"int8"}, @qcode{"int16"}, int32", @qcode{"single"},
@qcode{"double"}.  The default is @qcode{"double"}.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("invhess", @var{x})
@deftypefnx {} {@var{a} =} gallery ("invhess", @var{x}, @var{y})
Create the inverse of an upper Hessenberg matrix.

@end deftypefn

@deftypefn {} {@var{a} =} gallery ("invol", @var{n})
Create an involutory matrix.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("ipjfact", @var{n})
@deftypefnx {} {@var{a} =} gallery ("ipjfact", @var{n}, @var{k})
Create a Hankel matrix with factorial elements.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("jordbloc", @var{n})
@deftypefnx {} {@var{a} =} gallery ("jordbloc", @var{n}, @var{lambda})
Create a Jordan block.

@end deftypefn

@deftypefn  {} {@var{u} =} gallery ("kahan", @var{n})
@deftypefnx {} {@var{u} =} gallery ("kahan", @var{n}, @var{theta})
@deftypefnx {} {@var{u} =} gallery ("kahan", @var{n}, @var{theta}, @var{pert})
Create a @nospell{Kahan} matrix (upper trapezoidal).

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("kms", @var{n})
@deftypefnx {} {@var{a} =} gallery ("kms", @var{n}, @var{rho})
Create a @nospell{Kac-Murdock-Szego} Toeplitz matrix.

@end deftypefn

@deftypefn  {} {@var{b} =} gallery ("krylov", @var{a})
@deftypefnx {} {@var{b} =} gallery ("krylov", @var{a}, @var{x})
@deftypefnx {} {@var{b} =} gallery ("krylov", @var{a}, @var{x}, @var{j})
Create a Krylov matrix.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("lauchli", @var{n})
@deftypefnx {} {@var{a} =} gallery ("lauchli", @var{n}, @var{mu})
Create a @nospell{Lauchli} matrix (rectangular).

@end deftypefn

@deftypefn {} {@var{a} =} gallery ("lehmer", @var{n})
Create a @nospell{Lehmer} matrix (symmetric positive definite).

@end deftypefn

@deftypefn {} {@var{t} =} gallery ("lesp", @var{n})
Create a tridiagonal matrix with real, sensitive eigenvalues.

@end deftypefn

@deftypefn {} {@var{a} =} gallery ("lotkin", @var{n})
Create a @nospell{Lotkin} matrix.

@end deftypefn

@deftypefn {} {@var{a} =} gallery ("minij", @var{n})
Create a symmetric positive definite matrix MIN(i,j).

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("moler", @var{n})
@deftypefnx {} {@var{a} =} gallery ("moler", @var{n}, @var{alpha})
Create a @nospell{Moler} matrix (symmetric positive definite).

@end deftypefn

@deftypefn {} {[@var{a}, @var{t}] =} gallery ("neumann", @var{n})
Create a singular matrix from the discrete @nospell{Neumann} problem
(sparse).

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("normaldata", [@var{M} @var{N} @dots{}], @var{j})
@deftypefnx {} {@var{a} =} gallery ("normaldata", @var{M}, @var{N}, @dots{}, @var{j})
@deftypefnx {} {@var{a} =} gallery ("normaldata", @dots{}, "@var{class}")
Create a matrix with random samples from the standard normal distribution
(mean = 0, std = 1).

The first input is a matrix of dimensions describing the size of the output.
The dimensions can also be input as comma-separated arguments.

The input @var{j} is an integer index in the range [0, 2^32-1].  The values
of the output matrix are always exactly the same (reproducibility) for a
given size input and @var{j} index.

The final optional argument determines the class of the resulting matrix.
Possible values for @var{class}: @qcode{"single"}, @qcode{"double"}.
The default is @qcode{"double"}.

@end deftypefn

@deftypefn  {} {@var{q} =} gallery ("orthog", @var{n})
@deftypefnx {} {@var{q} =} gallery ("orthog", @var{n}, @var{k})
Create orthogonal and nearly orthogonal matrices.

@end deftypefn

@deftypefn {} {@var{a} =} gallery ("parter", @var{n})
Create a @nospell{Parter} matrix (a Toeplitz matrix with singular values
near pi).

@end deftypefn

@deftypefn  {} {@var{p} =} gallery ("pei", @var{n})
@deftypefnx {} {@var{p} =} gallery ("pei", @var{n}, @var{alpha})
Create a Pei matrix.

@end deftypefn

@deftypefn {} {@var{a} =} gallery ("poisson", @var{n})
Create a block tridiagonal matrix from Poisson's equation (sparse).

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("prolate", @var{n})
@deftypefnx {} {@var{a} =} gallery ("prolate", @var{n}, @var{w})
Create a prolate matrix (symmetric, ill-conditioned Toeplitz matrix).

@end deftypefn

@deftypefn {} {@var{h} =} gallery ("randhess", @var{x})
Create a random, orthogonal upper Hessenberg matrix.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("rando", @var{n})
@deftypefnx {} {@var{a} =} gallery ("rando", @var{n}, @var{k})
Create a random matrix with elements -1, 0 or 1.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("randsvd", @var{n})
@deftypefnx {} {@var{a} =} gallery ("randsvd", @var{n}, @var{kappa})
@deftypefnx {} {@var{a} =} gallery ("randsvd", @var{n}, @var{kappa}, @var{mode})
@deftypefnx {} {@var{a} =} gallery ("randsvd", @var{n}, @var{kappa}, @var{mode}, @var{kl})
@deftypefnx {} {@var{a} =} gallery ("randsvd", @var{n}, @var{kappa}, @var{mode}, @var{kl}, @var{ku})
Create a random matrix with pre-assigned singular values.

@end deftypefn

@deftypefn {} {@var{a} =} gallery ("redheff", @var{n})
Create a zero and ones matrix of @nospell{Redheffer} associated with the
Riemann hypothesis.

@end deftypefn

@deftypefn {} {@var{a} =} gallery ("riemann", @var{n})
Create a matrix associated with the Riemann hypothesis.

@end deftypefn

@deftypefn {} {@var{a} =} gallery ("ris", @var{n})
Create a symmetric Hankel matrix.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("smoke", @var{n})
@deftypefnx {} {@var{a} =} gallery ("smoke", @var{n}, @var{k})
Create a complex matrix, with a @nospell{"smoke ring"} pseudospectrum.

@end deftypefn

@deftypefn  {} {@var{t} =} gallery ("toeppd", @var{n})
@deftypefnx {} {@var{t} =} gallery ("toeppd", @var{n}, @var{m})
@deftypefnx {} {@var{t} =} gallery ("toeppd", @var{n}, @var{m}, @var{w})
@deftypefnx {} {@var{t} =} gallery ("toeppd", @var{n}, @var{m}, @var{w}, @var{theta})
Create a symmetric positive definite Toeplitz matrix.

@end deftypefn

@deftypefn  {} {@var{p} =} gallery ("toeppen", @var{n})
@deftypefnx {} {@var{p} =} gallery ("toeppen", @var{n}, @var{a})
@deftypefnx {} {@var{p} =} gallery ("toeppen", @var{n}, @var{a}, @var{b})
@deftypefnx {} {@var{p} =} gallery ("toeppen", @var{n}, @var{a}, @var{b}, @var{c})
@deftypefnx {} {@var{p} =} gallery ("toeppen", @var{n}, @var{a}, @var{b}, @var{c}, @var{d})
@deftypefnx {} {@var{p} =} gallery ("toeppen", @var{n}, @var{a}, @var{b}, @var{c}, @var{d}, @var{e})
Create a pentadiagonal Toeplitz matrix (sparse).

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("tridiag", @var{x}, @var{y}, @var{z})
@deftypefnx {} {@var{a} =} gallery ("tridiag", @var{n})
@deftypefnx {} {@var{a} =} gallery ("tridiag", @var{n}, @var{c}, @var{d}, @var{e})
Create a tridiagonal matrix (sparse).

@end deftypefn

@deftypefn  {} {@var{t} =} gallery ("triw", @var{n})
@deftypefnx {} {@var{t} =} gallery ("triw", @var{n}, @var{alpha})
@deftypefnx {} {@var{t} =} gallery ("triw", @var{n}, @var{alpha}, @var{k})
Create an upper triangular matrix discussed by
@nospell{Kahan, Golub, and Wilkinson}.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("uniformdata", [@var{M} @var{N} @dots{}], @var{j})
@deftypefnx {} {@var{a} =} gallery ("uniformdata", @var{M}, @var{N}, @dots{}, @var{j})
@deftypefnx {} {@var{a} =} gallery ("uniformdata", @dots{}, "@var{class}")
Create a matrix with random samples from the standard uniform distribution
(range [0,1]).

The first input is a matrix of dimensions describing the size of the output.
The dimensions can also be input as comma-separated arguments.

The input @var{j} is an integer index in the range [0, 2^32-1].  The values
of the output matrix are always exactly the same (reproducibility) for a
given size input and @var{j} index.

The final optional argument determines the class of the resulting matrix.
Possible values for @var{class}: @qcode{"single"}, @qcode{"double"}.
The default is @qcode{"double"}.

@end deftypefn

@deftypefn  {} {@var{a} =} gallery ("wathen", @var{nx}, @var{ny})
@deftypefnx {} {@var{a} =} gallery ("wathen", @var{nx}, @var{ny}, @var{k})
Create the @nospell{Wathen} matrix.

@end deftypefn

@deftypefn {} {[@var{a}, @var{b}] =} gallery ("wilk", @var{n})
Create various specific matrices devised/discussed by Wilkinson.

@end deftypefn


@c hadamard scripts/special-matrix/hadamard.m
@anchor{XREFhadamard}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{h} =} hadamard (@var{n})
Construct a Hadamard matrix (@nospell{Hn}) of size @var{n}-by-@var{n}.

The size @var{n} must be of the form @math{2^k * p} in which p is one of
1, 12, 20 or 28.  The returned matrix is normalized, meaning
@w{@code{Hn(:,1) == 1}}@ and @w{@code{Hn(1,:) == 1}}.

Some of the properties of Hadamard matrices are:

@itemize @bullet
@item
@code{kron (Hm, Hn)} is a Hadamard matrix of size @var{m}-by-@var{n}.

@item
@code{Hn * Hn' = @var{n} * eye (@var{n})}.

@item
The rows of @nospell{Hn} are orthogonal.

@item
@code{det (@var{A}) <= abs (det (Hn))} for all @var{A} with
@w{@code{abs (@var{A}(i, j)) <= 1}}.

@item
Multiplying any row or column by -1 and the matrix will remain a Hadamard
matrix.
@end itemize
@xseealso{@ref{XREFcompan,,compan}, @ref{XREFhankel,,hankel}, @ref{XREFtoeplitz,,toeplitz}}
@end deftypefn


@c hankel scripts/special-matrix/hankel.m
@anchor{XREFhankel}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{h} =} hankel (@var{c})
@deftypefnx {} {@var{h} =} hankel (@var{c}, @var{r})
Return the Hankel matrix constructed from the first column @var{c}, and
(optionally) the last row @var{r}.

If the last element of @var{c} is not the same as the first element of
@var{r}, the last element of @var{c} is used.  If the second argument is
omitted, it is assumed to be a vector of zeros with the same size as
@var{c}.

A Hankel matrix formed from an m-vector @var{c}, and an n-vector @var{r},
has the elements
@tex
$$
H(i, j) = \cases{c_{i+j-1},&$i+j-1\le m$;\cr r_{i+j-m},&otherwise.\cr}
$$
@end tex
@ifnottex

@example
@group
H(i,j) = c(i+j-1),  i+j-1 <= m;
H(i,j) = r(i+j-m),  otherwise
@end group
@end example

@end ifnottex
@xseealso{@ref{XREFhadamard,,hadamard}, @ref{XREFtoeplitz,,toeplitz}}
@end deftypefn


@c hilb scripts/special-matrix/hilb.m
@anchor{XREFhilb}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{h} =} hilb (@var{n})
Return the Hilbert matrix of order @var{n}.

The @math{i,j} element of a Hilbert matrix is defined as
@tex
$$
H(i, j) = {1 \over (i + j - 1)}
$$
@end tex
@ifnottex

@example
H(i, j) = 1 / (i + j - 1)
@end example

@end ifnottex

Hilbert matrices are close to being singular which make them difficult to
invert with numerical routines.  Comparing the condition number of a random
matrix 5x5 matrix with that of a Hilbert matrix of order 5 reveals just how
difficult the problem is.

@example
@group
cond (rand (5))
   @result{} 14.392
cond (hilb (5))
   @result{} 4.7661e+05
@end group
@end example

@xseealso{@ref{XREFinvhilb,,invhilb}}
@end deftypefn


@c invhilb scripts/special-matrix/invhilb.m
@anchor{XREFinvhilb}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{hinv} =} invhilb (@var{n})
Return the inverse of the Hilbert matrix of order @var{n}.

This can be computed exactly using
@tex
$$\eqalign{
  A_{ij} &= -1^{i+j} (i+j-1)
             \left( \matrix{n+i-1 \cr n-j } \right)
             \left( \matrix{n+j-1 \cr n-i } \right)
             \left( \matrix{i+j-2 \cr i-2 } \right)^2 \cr
         &= { p(i)p(j) \over (i+j-1) }
}$$
where
$$
  p(k) = -1^k \left( \matrix{ k+n-1 \cr k-1 } \right)
              \left( \matrix{ n \cr k } \right)
$$
@end tex
@ifnottex

@example
@group

           (i+j)         /n+i-1\  /n+j-1\   /i+j-2\ 2
A(i,j) = -1      (i+j-1)(       )(       ) (       )
                         \ n-j /  \ n-i /   \ i-2 /

       = p(i) p(j) / (i+j-1)

@end group
@end example

@noindent
where

@example
@group
         k  /k+n-1\   /n\
p(k) = -1  (       ) (   )
            \ k-1 /   \k/
@end group
@end example

@end ifnottex
The validity of this formula can easily be checked by expanding the binomial
coefficients in both formulas as factorials.  It can be derived more
directly via the theory of Cauchy matrices.  See @nospell{J. W. Demmel},
@cite{Applied Numerical Linear Algebra}, p.@: 92.

Compare this with the numerical calculation of @code{inv (hilb (n))},
which suffers from the ill-conditioning of the Hilbert matrix, and the
finite precision of your computer's floating point arithmetic.
@xseealso{@ref{XREFhilb,,hilb}}
@end deftypefn


@c magic scripts/special-matrix/magic.m
@anchor{XREFmagic}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{M} =} magic (@var{n})

Create an @var{n}-by-@var{n} magic square.

A magic square is an arrangement of the integers @code{1:n^2} such that the
row sums, column sums, and diagonal sums are all equal to the same value.

Note: @var{n} must be a scalar greater than or equal to 3.  If you supply
@var{n} less than 3, magic returns either a nonmagic square, or else the
degenerate magic squares 1 and [].
@end deftypefn


@c pascal scripts/special-matrix/pascal.m
@anchor{XREFpascal}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{P} =} pascal (@var{n})
@deftypefnx {} {@var{P} =} pascal (@var{n}, @var{t})
Return the Pascal matrix of order @var{n} if @code{@var{t} = 0}.

The default value of @var{t} is 0.

When @code{@var{t} = 1}, return the pseudo-lower triangular
Cholesky@tie{}factor of the Pascal matrix (The sign of some columns may be
negative).  This matrix is its own inverse, that is
@code{pascal (@var{n}, 1) ^ 2 == eye (@var{n})}.

If @code{@var{t} = -1}, return the true Cholesky@tie{}factor with strictly
positive values on the diagonal.

If @code{@var{t} = 2}, return a transposed and permuted version of
@code{pascal (@var{n}, 1)}, which is the cube root of the identity matrix.
That is, @code{pascal (@var{n}, 2) ^ 3 == eye (@var{n})}.

@xseealso{@ref{XREFchol,,chol}}
@end deftypefn


@c rosser scripts/special-matrix/rosser.m
@anchor{XREFrosser}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{R} =} rosser ()
Return the @nospell{Rosser} matrix.

This is a difficult test case used to evaluate eigenvalue algorithms.
@xseealso{@ref{XREFwilkinson,,wilkinson}, @ref{XREFeig,,eig}}
@end deftypefn


@c toeplitz scripts/special-matrix/toeplitz.m
@anchor{XREFtoeplitz}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{T} =} toeplitz (@var{c})
@deftypefnx {} {@var{T} =} toeplitz (@var{c}, @var{r})
Return the Toeplitz matrix constructed from the first column @var{c},
and optionally the first row @var{r}.

If the second argument is omitted, the first row is taken to be the
same as the first column.  If the first element of @var{r} is not the same
as the first element of @var{c}, the first element of @var{c} is used.

A Toeplitz, or diagonal-constant, matrix has the same value along each
diagonal.  Although it need not be square, it often is.  An @nospell{MxN}
Toeplitz matrix has the form:
@tex
$$
\left[\matrix{c_1    & r_2     & r_3      & \cdots & r_n\cr
              c_2    & c_1     & r_2      & \cdots & r_{n-1}\cr
              c_3    & c_2     & c_1      & \cdots & r_{n-2}\cr
              \vdots & \vdots  & \vdots   & \ddots & \vdots\cr
              c_m    & c_{m-1} & c_{m-2} & \ldots & c{m-n+1}}\right]
$$
@end tex
@ifnottex

@example
@group
c(1)  r(2)   r(3)  @dots{}  r(n)
c(2)  c(1)   r(2)  @dots{} r(n-1)
c(3)  c(2)   c(1)  @dots{} r(n-2)
 .     .      .   .      .
 .     .      .     .    .
 .     .      .       .  .
c(m) c(m-1) c(m-2) @dots{} c(m-n+1)
@end group
@end example

@end ifnottex
@xseealso{@ref{XREFhankel,,hankel}}
@end deftypefn


@c vander scripts/special-matrix/vander.m
@anchor{XREFvander}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn  {} {@var{V} =} vander (@var{c})
@deftypefnx {} {@var{V} =} vander (@var{c}, @var{n})
Return the @nospell{Vandermonde} matrix whose next to last column is
@var{c}.

If @var{n} is specified, it determines the number of columns; otherwise,
@var{n} is taken to be equal to the length of @var{c}.

A @nospell{Vandermonde} matrix has the form:
@tex
$$
\left[\matrix{c_1^{n-1}  & \cdots & c_1^2  & c_1    & 1      \cr
              c_2^{n-1}  & \cdots & c_2^2  & c_2    & 1      \cr
              \vdots     & \ddots & \vdots & \vdots & \vdots \cr
              c_n^{n-1}  & \cdots & c_n^2  & c_n    & 1      }\right]
$$
@end tex
@ifnottex

@example
@group
c(1)^(n-1) @dots{} c(1)^2  c(1)  1
c(2)^(n-1) @dots{} c(2)^2  c(2)  1
    .     .      .      .    .
    .       .    .      .    .
    .         .  .      .    .
c(n)^(n-1) @dots{} c(n)^2  c(n)  1
@end group
@end example

@end ifnottex
@xseealso{@ref{XREFpolyfit,,polyfit}}
@end deftypefn


@c wilkinson scripts/special-matrix/wilkinson.m
@anchor{XREFwilkinson}
@html
<span style="display:block; margin-top:-4.5ex;">&nbsp;</span>
@end html


@deftypefn {} {@var{W} =} wilkinson (@var{n})
Return the Wilkinson matrix of order @var{n}.

Wilkinson matrices are symmetric and tridiagonal with pairs of nearly, but
not exactly, equal eigenvalues.  They are useful in testing the behavior and
performance of eigenvalue solvers.

@xseealso{@ref{XREFrosser,,rosser}, @ref{XREFeig,,eig}}
@end deftypefn