Questaal Home

The mcx matrix calculator

Table of Contents


You should be aware of the standard Questaal format (which, among other formats, mcx can read) which has programming language capabilities.

mcx is required and is assumed to be in your path.

This manual is written for version 1.058. To see what version you are using, do:

$ mcx --version

1. Introduction

mcx is a matrix extension of an ordinary calculator. It is mainly designed to work with 2D arrays; it operates on numerical arrays. Usually the arrays are stored in ASCII format, though mcx has the ability to read binary files.

mcx is command-line driven and can efficiently manipulate arrays for convenient analysis. Some Questaal testing scripts make use of this calculator to analyze whether a test passes a certain tolerance.

mcx is stack-based: each time you read an array from a file it is pushed onto the stack. You can assign the top-level array to a name, in which case an independent copy is made.

The ASCII format of the file is quite flexible; you can specify the number of rows nr and number of columns nc in a number of ways:

  1. You can specify the number of columns on the command line, with −nc=#; similarly with the number of rows (−nr=#). −nc=#; this switch is used in Example 2.2.

  2. You can include a directive at the start of the file, before any data is read. mcx passes the input through a file preprocessor first; in addition a directive specifying some combination nr and nc, e.g.

     % rows #1  cols #2

    supplies the needed information to mcx.

  3. In the absence of an explicit specification, mcx will infer nc by counting how many numbers (more generally expressions) are on the first line.

If nc is obtained somehow but not nr, mcx determines it by counting the number of expressions in the entire file and dividing by nc.

Note: : if the −nc=# or −nr=# switches are not used, mcx determines them from the standard Questaal algorithm for reading 2D arrays, described in more detail here.

2. Examples

This section develops a few examples to provide an intuitive feel for how mcx works and to illustrate some features.
A systematic description of mcx’s features and arguments is given in Section 3.

Cut and paste the following data into array into file mat1

1.1 2.2
3.3 4.4

and this data into file mat2.

5 6
7 8

Example 2.1.   Add mat1 to mat2

If you do the following:

$ mcx mat1 mat2 -+

you should see

% rows 2 cols 2 real
    6.100000    8.200000
   10.300000   12.400000

Any argument that begins with  “−“  is a switch or an operator (as distinct from data). The string following  “−“  names the operator.

  • mat1 : reads file mat1 from disk and pushes it onto the stack; call it s0.

  • mat2 : reads file mat2 and pushes it onto the stack so s0s1 and the contents of mat2s0.

  • −+ : is the binary operation “add.” If s0 and s1 exist, they are summed and popped off the stack. Their sum is pushed onto top-level array s0.

Note: if operations occur before arrays are given, they are push on an operations stack and execute when arrays become available. The following commands accomplish the same thing

$ mcx -+ mat1 mat2
$ mcx mat1 -+ mat2
$ mcx mat1 mat2 -+

Use −show to see the stack. Try this

$ mcx -f2f8.1 mat1 mat2 -show  -+

You should get

# 0 named arrays, 2 on stack; pending 0 unops 0 bops (vsn 1.057)
# stack             nr    nc   cast
#    2               2     2   real
#    1               2     2   real
% rows 2 cols 2 real
     6.1     8.2
    10.3    12.4

−show prints out what arrays are in the calculator; the addition is subsequently performed and the result printed.

“−f…“  is a formatting statement; 2f8.1 follows the fortran convention for formatting real numbers.

Example 2.2   Standard input and standard output

After all the command-line arguments are parsed, usually mcx prints the top-level array s0, and exits silently.

  • If s0 is not present, mcx waits for you to enter an array from standard input.
    For example try

    $ mcx -f2f8.1
    $ 1 2
    $ 3
    $ 4 <ctrl-D>

    mcx should print out

    % rows 2 cols 2 real
         1.0     2.0
         3.0     4.0
  • If the last command is −show, mcx just prints out information about named arrays and the stack and exits (even if the stack has no arrays). Thus the following command

    $ mcx -f2f8.1 mat1 mat2 -show  -+ -show

    does the same as it did in Example 2.1 but prints out stack contents a second time instead of s0.

You can also tell mcx to read the next array from standard input by using a full stop (“ . “) in lieu of file name. Thus

$ echo 1 -2 3 -4 | mcx .

pushes a 1×4 array onto s0, while

$ echo 1 -2 3 -4 | mcx -nc=2 .  mat2 -+

pushes a 2×2 array onto s0, then pushes mat2, and adds the two so that a single array remains on the stack. −nc=2 tells mcx to treat the line from stdin as an array with 2 columns.

Example 2.3   Manipulations of eigenvalues and eigenvectors of an array

This example finds the eigenvalues e and eigenvectors z of mat2, and shows that mat2 z = z e.

First try

$ mcx mat2 -evl

You should see a 2×1 complex array

% rows 2 cols 1 complex


The eigenvalues are complex because the matrix is not hermitian. The imaginary part follows the real part; this is how mcx displays and reads complex arrays in ASCII format.

You can force mat2 to be hermitian (symmetric since mat2 is real) with −herm. Now do

$ mcx mat2 -herm -a h h -evl h -evc -ap z -tog -v2dia -x  h z -x --

These instructions do the following:

  • mat2 : reads file mat2 and pushes it onto s0.

  • −herm : symmetrizes s0.

  • −a h : copies s0 to a named array  h  and pops it from the stack. The stack is now empty.

  • h −evl : pushes h onto s0 and replaces s0 with its eigenvalues. Because h is hermitian, s0 is real.

  • h −evc : pushes h onto the stack and replaces s0 with its eigenvectors. Now the stack has two arrays, s0 = z and s1 = e.

  • −ap z : copies s0 to a named array  z.

  • −tog : toggles s0 and s1, making s1 = z and s0 = e.

  • −v2dia : Turns s0 (a vector or 2×1 array of eigenvalues) into a diagonal 2×2 array.

  • −x : multiplies s1 × s0. One array remains on the stack, s0 = z × e.

  • h z −x : pushes h and z onto the stack and multiplies them. Now s0 = h × z, while s1 = z × e

  • −− : Adds s1 to −s0.  s1 and s1 are mathematically identical so the difference should be zero.

Note: this formula should still work even if h is not hermitian.

See Table of Contents

Example 2.4   Numerical integration, differentiation, and interpolation of a function

Integrate and differentiate the function , by tabulating it on a mesh and evaluating integrals and derivatives numerically. This example also uses the tabulated data to interpolate it to another mesh.

Copy the contents of the box below into file dat.

% const n=100 p=1 lam=2
% save
% macro iy(z) exp(-lam*z)*(-lam*z-1)/lam^2
% repeat i= 0:n
% var x=10*i/n
{x} {x^p*exp(-lam*x)}
% end

This creates 101 rows of xy pairs with x ranging between 0 and 10.

Note: : The number of points n+1, and also p and λ are declared in the file with the const preprocessor directive. You can override the values assigned there with command-line arguments, e.g. −vp=#. The save directive retains the variables declared in this file after it is read. The macro directive will be used for the indefinite integral, below.


The derivative is readily found to be

Try some of the following commands

$ mcx -vp=2 dat -diff
$ mcx -vp=2 dat -diff -e3 x1 x2 'x1^(p-1)*(p-lam*x1)*exp(-lam*x1)'
$ mcx -vp=2 -f3f15.10 dat -diff:nord=5 -e3 x1 x2 'x1^(p-1)*(p-lam*x1)*exp(-lam*x1)' -e3 x1 x2 x2-x3

All of them differentiate the second column with respect to the first, using p=2.

  • The first returns two columns with x and y′.
  • The second returns three columns with x, y′, and the analytic derivative of y′.
  • The third returns three columns (more decimals) with x, y′, and the error in the numerical estimate for y′.

Unpacking the third command:

−vp=2Declares variable p and assigns the value 2. This overrides the assignment in dat.
−f3f15.10Formats output (fortran format 3f15.10)
datread s0 from dat.
−diff:nord=5Replace column 2 with a numerical estimate for y
Use a 5-point polynomial to interpolate the data; estimate is derivative of polynomial interpolation
−e3   x1   x2   ‘x1^(p−1)*(p−lam*x1)*exp(−lam*x1)’replace s0 with a three column array consisting of x, y′, and the analytic derivative of y
−e3   x1   x2   x2−x3replace s0 with a three column array consisting of x, y′, and difference between the numerical and analytical y

Some observations:

  • The largest error appears for x→0. The interpolation is less accurate when all the data lies on one side of the interpolating point.
  • The error improves when higher order polynomials (nord=#) are used, or when the mesh is made finer (−vn=#).
  • If you use p<1, y′ diverges at the origin. The numerical derivative cannot reproduce this.

y is small at x=10 if λ=−2, provided p is 3 or less; so we will use 10 in place of ∞.

Try some of the following commands

$ mcx -vp=1 dat -int 0 10
$ mcx -vp=2 dat -int 0 10
$ mcx -vp=3 dat -int 0 10

These commands calculate  I  between 0 and 10 for three values of  p.

You should find that  I  is close to , i.e. 1/4, 1/4, and 3/8 for p=1,2,3. Some numerical errors appear in the 6th digit. The integral is carried out by fitting the data to a polynomial of order nord−1, and integrating the polynomial. You can reduce the error by using a higher order than the default value of 4 for nord, viz

$ mcx -vp=3 dat -int:nord=6 0 10

Also for p>3 you should increase the upper bound beyond 10 since the integral from 10 to ∞ is on the order of 10−6.

Indefinite Integral

As noted  I  must be evaluated between definite limits. However, you can make mcx simulate an indefinite integral by integrating over a range of upper bounds.

To compare with exact results, note that when p=1 the indefinite integral is

The macro in dat evaluates this integral.

Try the following:

$ mcx dat -int 0 0:2:.2 -e3 x1 x2 'iy(x1)-iy(0)'

This evaluates  I  for a lower bound of 0 and a uniform mesh of points for the upper bound between 0 and 2. In the third column the analytic integral is evaluated from the macro iy at the upper and lower bounds.

You should that the numerical and analytic integrals agree to about 6 decimal places:

% rows 11 cols 3 real
    0.000000    0.000000    0.000000
    0.200000    0.015387    0.015388
    0.400000    0.047801    0.047802
    0.600000    0.084343    0.084343
    0.800000    0.118767    0.118767
    1.000000    0.148498    0.148499
    1.200000    0.172890    0.172890
    1.400000    0.192230    0.192230
    1.600000    0.207200    0.207200
    1.800000    0.218578    0.218578
    2.000000    0.227105    0.227105

Interpolation proceeds much in the same was as integration; only interpolation has lower bound. Try

$ mcx -vp=2  dat -intrp .5:1:.05 -e3 x1 x2 'x1^p*exp(-lam*x1)' -e3 x1 x2 x2-x3

This returns the abscissa on a mesh twice finer than the original mesh. Every odd point is perfectly interpolated (they lie on the original mesh); the even points reflect the error of the interpolation.

3. mcx manual

mcx is a stack-based, command-line driven calculator for matrices. Matrices reside on the stack, ordered as s0, s1, … . There are unary operators that operate on the top-level element s0, replacing it with some transformation, and binary operators that operate on s1 and s0 replacing both of them with the result of some operation, e.g. s1 × s0.


mcx [−switches] data-file -ops …

Arguments that do not begin with  “−“  or  ”[“  must be files, stored in the form of a 2D array. For ASCII files, data is read using the standard Questaal format which features programming language capabilities.

Any argument that begins with  “−“  is a switch, a unary operator, or a binary operator.

Any argument that begins with  ”[“  is a declaration of a command-line looping construct where command arguments between  ”[“  and  ”]”  are iterated over, as described below.

When a file is read, its contents (together with the number of rows nr and columns nc) is pushed onto the stack and becomes the top-level stack element s0. Elements already existing on the stack get pushed down one level. If there are n such elements, si-1si for i = n, n−1, …, and the new element becomes s0.

Data is normally read from a file; however if data-file is a full stop (“ . ”), data is read from standard input in lieu of a file. (It can occur only once). See Example 2.2.


  • −nc=# (nr=#)
    Stipulate that next matrix read has # columns (rows). See Example 2.2 for an illustration.
  • −vvar=#
    define variable −vvar assign value to  #. This is the standard way Questaal programs assign variables from the command line. Since data files are parsed by the preprocessor, such variables may enter into preprocessor directives or as part of expressions in the data itself.
  • −show
    show data stack and any operations pending. See Example 2.1.
  • −w[l=string] fname | −bw[l=string] fname
    write s0 to file fname.   −bw writes to a binary file.
  • −wap
    (complex arrays only) write s0 as (amplitude,phase) rather than (Re, Im).
  • −a[nr=#|:nc=#] nam | −ap nam
    assign s0 to nam, and pop s0 off the stack. −ap nam performs the assignment but does not pop s0 off the stack.
  • −av[ir,ic] var
    assigns scalar variable var to element from s0(ir,ic). If ir and ic are specified, the (1,1) element is used.
  • -r~switches
    switches are separated by  ’~’; you can use a different character instead. Switches are:

    ~qrread with fortran read (fast, no algebra)
    ~s=#skips # records before reading
    ~openleaves file open after reading; thus if you read the file again it will read the next array
    ~brread from binary file, using first record to read nr and nc
    ~spcload in sparse format
    ~br,nr,ncbinary file has no first record; user supplies nr,nc
  • −px[:nprec] | −pxc
    write in row (column) compressed storage format, displaying only elements larger than a tolerance By default the tolerance is 10−8. If nprec is supplied, the tolerance is By default the tolerance is 10nprec.

Unary operators

These operators act on the top level matrix, s0, and replace it with the result of the unop.

Where expr appears in the switches below, algebraic variables declared from directives may be used; also you can use xn for the contents of the nth column.

  • −p:
    Pushes top array s0 onto stack, duplicating it.
  • −p+n (-p-n):
    Pushes nth array (from bottom) onto stack
  • −pop:
    Pops s0 from stack, shifting the other elements up one level.
  • −csum[:list]
    Sums the columns of s0, replacing it with a matrix of one column.
  • −rsum[:list]
    Sums the rows of s0, replacing it with a matrix of one row.
  • −s#:
    Scale s0 by # (may use -s#real,#imag)
  • −shft=#:
    Adds a constant # to s0
  • −sort expr
    Sorts rows s0, ordering them by the result of expr
  • −i | − iq:
    Inverts s0
  • −1:n
    Pushes unit matrix, dimension n, onto stack
  • −tp [nc~]list
    Generates matrix from list. Elements in list are real numbers; see here for syntax.
  • −evl | −evc
    Replaces s0 by its eigenvalues (eigenvectors)
  • −t:
    Transposes s0
  • −cc:
    Take complex conjugate of s0
  • −herm: Replaces s0 with its hermitian part
  • −real:
    Replaces s0 with its real part
  • −v2dia:
    If s0 is a vector, it is expanded into diagonal matrix.
    If s0 is a square matrix, its diagonal elements are used to form a vector.
  • −split   nam   x1,x2…,xn   y1,y2,…,yn:
    Splits s0 into subblocks; assign them names namij.
  • −rep:n1,n2
    Concatenates replicas of s0 to create (nr×n1,nc×n2) array.
  • −roll:#1[,#2]
    Cyclically shifts rows (and columns, respectively) by #1 (and #2) elements.
  • −pwr=#:
    Raises (col-1) matrix to a power
Row and column Unary Operators

These unops treat s0 as an array of nr rows and nc columns.

  • −rowl list  |  −coll list
    Creates a new array from a list of rows (columns) of s0
  • −rowl:mirr  |  −coll:mirr
    Rearranges rows (columns) in reverse order
  • −rowl:pf=fnam  |  −coll:pf=fnam  |  −rowl:ipf=fnam  |  −coll:ipf=fnam
    Same as −rowl (−coll) but list is read from permutation file fnam. ipf reverses the sense of the permutation.
    Example: file perm contains 2 3 1 4. Then
    • −rowl:pf=perm returns s0 with rows 2,3,1,4 permuted into 1,2,3,4
    • −rowl:ipf=perm returns s0 w/ rows 3,1,2,4 permuted into 1,2,3,4
  • −inc expr
    Retains rows from s0 for which expr is nonzero.
  • −sub t,b,l,r  |  -sub t,l
    Extracts a subblock of s0. In second form, bottom right corner = (nr,nc).
  • −subs:# t,b,l,r
    Scales a subblock of s0 by #.
  • −subv:# t,b,l,r
    Copies # to subblock of s0.
  • −e# expr1   expr2   …   expr#
    Create new matrix of # columns with values expr1   expr2   … expr#  .
    Variables x1,   x2,   … can be used in the expressions. These variables refer to the elements of s0in columns 1, 2, …, and change with each row.
    This switch is used in Example 2.4.
    Note: : at present this switch does not work for complex elements.
Unary Operators that treat data as discretized continuous functions of the first column

In these unops, column 1 (x1) is treated as an the independent variable.

Three of these unops fit data in other columns with a polynomial in x1. The polynomial is used to differentiate (−diff), integrate (−int), or interpolate (−intrp) data in the remaining columns. There are some modifiers (called [:opts] in the documentation below):

  1. :nord=# will change the number of points in fitting polynomial to  # (polynomial order is #−1). The default value is #=4.
  2. :rat will use a rational polynomial, rather than an ordinary one. Good for data with singularities.
  3. :mesh (applies to −int only) replaces s0 with its integral on the same mesh as the original set of values x1. A trapezoidal rule is used at present.

You can string the options together. These three unops are demonstrated in Example 2.4.

  • −diff[:opts]
    differentiates columns 2…nc wrt column 1, evaluated at each x1.
    s0 is returned as a table of points with list in the first column and the result of the derivative in the remaining columns.
  • −int[:opts] xlo list
    integrates columns 2…nc wrt column 1 from the lower bound xlo to a set of upper bounds, given by list.
    s0 is returned as a table of points with list in the first column and the result of the integral in the remaining columns.
  • −intrp[:opts] list
    interpolates column 2 to points in list.
    s0 is returned as a table of points with list in the first column and the result of the interpolation in the remaining columns.

  • −smo width,xmin,xmax,dx:
    smooths vector of delta-functions with gaussians
  • −abs
    takes the absolute value of each element
  • −max[:i|g]
    puts the largest element into the first column.
    Optional g returns max value of the entire array
    Optional i returns index to max value.
  • −unx[:i1] #1,#2
    (uncross) exchanges points in columns #1 and #2 after their point of closest approach
  • −unx2[:i1] #1,#2
    (cross) exchanges points in columns #1 and #2 at their point of closest approach, if they do not cross.
  • −at[:i1] val expr
    find adjacent rows that bracket expr=val. Array contains linearly interpolated x1 and expr.
  • −nint[#1]
    Replaces each element with nearest integer.
    Optional #1: scale array by #1 before operation, and by 1/#1 after. Thus -nint:1000 rounds to nearest thousandth.

Binary Operators

These operate on the top two matrices in the stack, s1 and s0. s1 and s0 are popped from the stack, and the result of the binary operation is put into s0.

Dimensions for s0 and s1 are not independent; for example if they are added they must have the same number of rows and columns.

  • −tog
    Toggle s0 and s1.
  • −+
    Add s1 + s0
  • −−
    Add s1s0
  • −x
    Multiply s1 × s0
  • −xe
    Multiply s1 and s0 element by element
  • −de
    Divide s1 / s0 element by element
  • −x3
    Multiply s1 s0 are thou they are 3D arrays: s1 = s1(n11,n21,n31) s0 = s0(n10,n20,n30) where n10=nr(0)/n20,n20=nc(1),n30=nc(0); n11=n10,n21=nr(1)/n11,n31=n20 Result(i,j,k) = sum_m s1(i,j,m) s0(i,m,k) is condensed into 2D (nr(1),nc(0))
  • −gevl  |  −gevc
    Same as −evl  |  −evc, but for the generalized eigenvalue problem. s1 is the nonorthogonal matrix.
  • −orthos:
    Replace s0 with s1-1/2 s0 s1-1/2
  • −ccat:
    Concatenate columns of s1 and s0 into a single array
  • −rcat:
    Concatenate rows of s1 and s0 into a single array
  • −cross:
    Cross product s1(1,1..3) x s0(:,1..3)
  • −suba[#] t,b,l,r  |  -suba[#] t,l
    Copy s1 to subblock of s0. Conventions for subblock are the same as for −suba t,b,l,r  |  -sub t,l.
    Optional # copies #×s1 into s0.
  • −index:
    Use s0 as an row index to s1. s0(i) is overwritten by s1(s0(i)). s1 is preserved. New s0 has row dimensions of the original s0 and column dimensions of s1.

See Table of Contents

Repeated Iteration of Command Line Arguments

Command line arguments can be repeated by enclosing them in brackets, with the syntax

[ name=list arg1 arg2 arg3 ... argn ]

list is a standard Questaal integer list.

mcx executes the sequence of command line arguments arg1, arg2, arg3, …, argn for each element in list. Within this special construct, arg1, arg2, … are treated as strings that are parsed for expression substitution. In particular, variables name and i are loaded and recalculated each pass. name is the value of the current integer, and i is the index in the list. name= is not required; mcx will use irpt as the variable if it is omitted.

The scheme is best explained by a concrete illustratation. Suppose files   a0, a1 and a4   reside on disk, with a0 a 4×4 symmetrix matrix, a1 4×4 hermitian, and a4 1×3 real. The command

mcx [ k=0,1,4 'a{k}' ] -show

should get expanded to a sequence of four arguments

110a{k}a0 loaded onto stack
221a{k}a1 loaded onto stack
334a{k}a4 loaded onto stack
4---showdisplays stack

You should see the following:

# 0 named arrays, 3 on stack; pending 0 unops 0 bops (vsn 1.058)
# stack             nr    nc   cast
#    3               1     3   real   
#    2               4     4   herm   
#    1               4     4   symm   
Conditional evaluation of an argument

Within the  [ … ]  construct, an argument that begins with  ?  is treated as an expression. The result of that expression determines whether the subsequent argument should be evaluated or skipped. Thus the construct

mcx ... [ ... '?expr' argi ... ]

parses expr as an expression. If it evaluates nonzero, argi is executed. Otherwise, argi is passed over.


mcx [ ix=1:4 'a{ix}' '?ix>1' -rcat ]

pushes a1, a2, a3, a4, onto the stack. -rcat is evaluated after each of the last three arrays are loaded, thus appending that array to the prior one on the stack. The final stack consists of a single array which concatenates a1, … , a4.

Special handling of the last iteration

You can prevent the last argument (or arguments) in the loop from executing, by appending a  /  to the   [. A single slash suppresses the last argument, two slashes suppress the last two arguments, and so on.

Example: cut and paste the 2×2 matrix in the box below to file a.

1 x
x x*x

Then do:

mcx  [/ 1:3 '-vx={i}' a -+ ]

This sums three instances of  a  with x=1, x=2, and x=3. Note that three arrays are loaded but there are only two additions, so   −+   should be suppressed the last iteration.

You should see the following output:

% rows 2 cols 2 real
    3.000000    6.000000
    6.000000   14.000000

The (1,1) element is the sum of x0 for x=1…3, while (1,2) and (2,1) are the sum of x1 and (2,2) is the sum of x2.

To see the arguments being parsed in detail, use the -debug switch:

mcx -debug [/ 1:3 '-vx={i}' a -+ ]

Other resources

The source code to mcx can be found in the bitbucket repository.

If this page has any errors, there is something you think is missing or unclear, or for any other issues, you can create a post here letting us know.