comparison man/calc.texi @ 90066:fb79180b618d

Revision: [e-mail]--gnu-2004/emacs--unicode--0--patch-78 Merge from emacs--cvs-trunk--0 Patches applied: * [e-mail]--gnu-2004/emacs--cvs-trunk--0--patch-719 - [e-mail]--gnu-2004/emacs--cvs-trunk--0--patch-732 Update from CVS * [e-mail]--gnu-2004/emacs--cvs-trunk--0--patch-733 Update from CVS: man/calc.texi: Fix some TeX definitions. * [e-mail]--gnu-2004/emacs--cvs-trunk--0--patch-734 Merge from gnus--rel--5.10 * [e-mail]--gnu-2004/gnus--rel--5.10--patch-75 Merge from emacs--cvs-trunk--0 * [e-mail]--gnu-2004/gnus--rel--5.10--patch-76 - [e-mail]--gnu-2004/gnus--rel--5.10--patch-77 Update from CVS

90065:c26eb15cd14a

@c @texline foo
@c    `foo' will appear only in TeX output
@c @infoline foo
@c    `foo' will appear only in non-TeX output

@c In TeX output, @tmath{expr} will typeset expr in  math mode.
@c In any output, @expr{expr} will typeset an expression;
@c $x$ in TeX, @samp{x} otherwise.

@iftex
@macro texline{stuff}
\stuff\
@end macro
@macro tmath{stuff}
@tex
$\stuff\$
@end tex
@end macro
@alias infoline=comment
@c @alias expr=math
@tex
\gdef\expr#1{\tex 
             \let\t\ttfont 
             \turnoffactive 
             $#1$
             \endgroup}
@end tex
@macro cpi{}
@math{@pi{}}
@end macro
@macro cpiover{den}
@math{@pi/\den\}

@alias texline=comment
@macro infoline{stuff}
\stuff\
@end macro
@alias expr=samp
@macro cpi{}
@expr{pi}
@end macro
@macro cpiover{den}
@expr{pi/\den\}

@strong{RPN calculation.}  In RPN, you type the input number(s) first,
then the command to operate on the numbers.

@noindent
Type @kbd{2 @key{RET} 3 + Q} to compute 
@texline @tmath{\sqrt{2+3} = 2.2360679775}.
@infoline the square root of 2+3, which is 2.2360679775.

@noindent
Type @kbd{P 2 ^} to compute 
@texline @tmath{\pi^2 = 9.86960440109}.
@infoline the value of `pi' squared, 9.86960440109.

@noindent
Type @key{TAB} to exchange the order of these two results.

conventional ``algebraic'' notation.  To enter an algebraic formula,
use the apostrophe key.

@noindent
Type @kbd{' sqrt(2+3) @key{RET}} to compute 
@texline @tmath{\sqrt{2+3}}.
@infoline the square root of 2+3.

@noindent
Type @kbd{' pi^2 @key{RET}} to enter 
@texline @tmath{\pi^2}.
@infoline `pi' squared.  
To evaluate this symbolic formula as a number, type @kbd{=}.

@noindent
Type @kbd{' arcsinh($ - $$) @key{RET}} to subtract the second-most-recent

the upper-leftmost @samp{1} and set the mark, then move to just after
the lower-right @samp{8} and press @kbd{M-# r}.

@noindent
Type @kbd{v t} to transpose this 
@texline @tmath{3\times2}
@infoline 3x2 
matrix into a 
@texline @tmath{2\times3}
@infoline 2x3
matrix.  Type @w{@kbd{v u}} to unpack the rows into two separate
vectors.  Now type @w{@kbd{V R + @key{TAB} V R +}} to compute the sums
of the two original columns. (There is also a special
grab-and-sum-columns command, @kbd{M-# :}.)

commands like @kbd{M-x calc} (for the standard user interface) or
@w{@kbd{M-x calc-keypad}} (for Keypad Mode).  First type @kbd{M-x}
(that's Meta with the letter @kbd{x}), then, at the prompt,
type the full command (like @kbd{calc-keypad}) and press Return.

If you type @kbd{M-x calc} and Emacs still doesn't recognize the
command (it will say @samp{[No match]} when you try to press
@key{RET}), then Calc has not been properly installed.

The same commands (like @kbd{M-# c} or @kbd{M-# M-#}) that start
the Calculator also turn it off if it is already on.

@node The Standard Interface, Quick Mode Overview, Starting Calc, Using Calc
@subsection The Standard Calc Interface

a printer, the trail corresponds to the paper tape that records what
you do.

In this case, the trail shows that four numbers (17.3, 3, 2, and 4)
were first entered into the Calculator, then the 2 and 4 were
multiplied to get 8, then the 3 and 8 were subtracted to get @i{-5}.
(The @samp{>} symbol shows that this was the most recent calculation.)
The net result is the two numbers 17.3 and @i{-5} sitting on the stack.

Most Calculator commands deal explicitly with the stack only, but
there is a set of commands that allow you to search back through
the trail and retrieve any previous result.

full complexity of the stack and trail.  To use it, type @kbd{M-# q}
(@code{quick-calc}) in any regular editing buffer.

Quick Mode is very simple:  It prompts you to type any formula in
standard algebraic notation (like @samp{4 - 2/3}) and then displays
the result at the bottom of the Emacs screen (@i{3.33333333333}
in this case).  You are then back in the same editing buffer you
were in before, ready to continue editing or to type @kbd{M-# q}
again to do another quick calculation.  The result of the calculation
will also be in the Emacs ``kill ring'' so that a @kbd{C-y} command
at this point will yank the result into your editing buffer.

@noindent
@dfn{Keypad Mode} is a mouse-based interface to the Calculator.
It is designed for use with terminals that support a mouse.  If you
don't have a mouse, you will have to operate keypad mode with your
arrow keys (which is probably more trouble than it's worth).  Keypad
mode is currently not supported under Emacs 19.

Type @kbd{M-# k} to turn Keypad Mode on or off.  Once again you
get two new windows, this time on the righthand side of the screen
instead of at the bottom.  The upper window is the familiar Calc
Stack; the lower window is a picture of a typical calculator keypad.

@noindent
Calc was originally started as a two-week project to occupy a lull
in the author's schedule.  Basically, a friend asked if I remembered
the value of 
@texline @tmath{2^{32}}.
@infoline @expr{2^32}.  
I didn't offhand, but I said, ``that's easy, just call up an
@code{xcalc}.''  @code{Xcalc} duly reported that the answer to our
question was @samp{4.294967e+09}---with no way to see the full ten
digits even though we knew they were there in the program's memory!  I

@key{RET} 3 @key{RET} 4 + * -} compute?  (@samp{*} is the symbol for
multiplication.)  Figure it out by hand, then try it with Calc to see
if you're right.  @xref{RPN Answer 1, 1}. (@bullet{})

(@bullet{}) @strong{Exercise 2.}  Compute 
@texline @tmath{(2\times4) + (7\times9.4) + {5\over4}}
@infoline @expr{2*4 + 7*9.5 + 5/4} 
using the stack.  @xref{RPN Answer 2, 2}. (@bullet{})

The @key{DEL} key is called Backspace on some keyboards.  It is
whatever key you would use to correct a simple typing error when

$$ 2 + { 3 \times 4 \times 5 \over 6 \times 7^8 } - 9 $$
\afterdisplay
@end tex

@noindent
The result of this expression will be the number @i{-6.99999826533}.

Calc's order of evaluation is the same as for most computer languages,
except that @samp{*} binds more strongly than @samp{/}, as the above
example shows.  As in normal mathematical notation, the @samp{*} symbol
can often be omitted:  @samp{2 a} is the same as @samp{2*a}.

Operators at the same level are evaluated from left to right, except
that @samp{^} is evaluated from right to left.  Thus, @samp{2-3-4} is
equivalent to @samp{(2-3)-4} or @i{-5}, whereas @samp{2^3^4} is equivalent
to @samp{2^(3^4)} (a very large integer; try it!).

If you tire of typing the apostrophe all the time, there is an
``algebraic mode'' you can select in which Calc automatically senses
when you are about to type an algebraic expression.  To enter this

intermediate results of a calculation as you go along.  You can
accomplish this in Calc by performing your calculation as a series
of algebraic entries, using the @kbd{$} sign to tie them together.
In an algebraic formula, @kbd{$} represents the number on the top
of the stack.  Here, we perform the calculation 
@texline @tmath{\sqrt{2\times4+1}},
@infoline @expr{sqrt(2*4+1)},
which on a traditional calculator would be done by pressing
@kbd{2 * 4 + 1 =} and then the square-root key.

@smallexample

@end smallexample

@noindent
The shift-@kbd{S} command computes the sine of an angle.  The sine
of 45 degrees is 
@texline @tmath{\sqrt{2}/2};
@infoline @expr{sqrt(2)/2}; 
squaring this yields @expr{2/4 = 0.5}.  However, there has been a slight
roundoff error because the representation of 
@texline @tmath{\sqrt{2}/2}
@infoline @expr{sqrt(2)/2} 
wasn't exact.  The @kbd{c 1} command is a handy way to clean up numbers
in this case; it temporarily reduces the precision by one digit while it
re-rounds the number on the top of the stack.

@end group
@end smallexample

@noindent
Here we compute the Inverse Sine of 
@texline @tmath{\sqrt{0.5}},
@infoline @expr{sqrt(0.5)}, 
first in radians, then in degrees.

Use @kbd{c d} and @kbd{c r} to convert a number from radians to degrees
and vice-versa.

logarithm).  These can be modified by the @kbd{I} (inverse) and
@kbd{H} (hyperbolic) prefix keys.

Let's compute the sine and cosine of an angle, and verify the
identity 
@texline @tmath{\sin^2x + \cos^2x = 1}.
@infoline @expr{sin(x)^2 + cos(x)^2 = 1}.  
We'll arbitrarily pick @i{-64} degrees as a good value for @expr{x}.
With the angular mode set to degrees (type @w{@kbd{m d}}), do:

@smallexample
@group
2:  -64        2:  -64        2:  -0.89879   2:  -0.89879   1:  1.

Remember, @kbd{f h} is the @code{calc-hypot}, or square-root of sum
of squares, command.

Another identity is 
@texline @tmath{\displaystyle\tan x = {\sin x \over \cos x}}.
@infoline @expr{tan(x) = sin(x) / cos(x)}.
@smallexample
@group

2:  -0.89879   1:  -2.0503    1:  -64.

@end group
@end smallexample

A physical interpretation of this calculation is that if you move
@expr{0.89879} units downward and @expr{0.43837} units to the right,
your direction of motion is @i{-64} degrees from horizontal.  Suppose
we move in the opposite direction, up and to the left:

@smallexample
@group
2:  -0.89879   2:  0.89879    1:  -2.0503    1:  -64.

the top two stack elements right after the @kbd{U U}, then a pair of
@kbd{M-@key{TAB}} commands to cycle the 116 up around the duplicates.

A similar identity is supposed to hold for hyperbolic sines and cosines,
except that it is the @emph{difference}
@texline @tmath{\cosh^2x - \sinh^2x}
@infoline @expr{cosh(x)^2 - sinh(x)^2} 
that always equals one.  Let's try to verify this identity.

@smallexample
@group

exact integer, you get an exact integer result (158 digits long
in this case).

If you take the factorial of a non-integer, Calc uses a generalized
factorial function defined in terms of Euler's Gamma function
@texline @tmath{\Gamma(n)}
@infoline @expr{gamma(n)}
(which is itself available as the @kbd{f g} command).

@smallexample
@group

@end group
@end smallexample

@noindent
Here we verify the identity 
@texline @tmath{n! = \Gamma(n+1)}.
@infoline @expr{@var{n}!@: = gamma(@var{n}+1)}.

The binomial coefficient @var{n}-choose-@var{m}
@texline or @tmath{\displaystyle {n \choose m}}
is defined by
@texline @tmath{\displaystyle {n! \over m! \, (n-m)!}}
@infoline @expr{n!@: / m!@: (n-m)!}
for all reals @expr{n} and @expr{m}.  The intermediate results in this
formula can become quite large even if the final result is small; the
@kbd{k c} command computes a binomial coefficient in a way that avoids
large intermediate values.

rows in the matrix is different from the number of elements in the
vector.

(@bullet{}) @strong{Exercise 1.}  Use @samp{*} to sum along the rows
of the above 
@texline @tmath{2\times3}
@infoline 2x3 
matrix to get @expr{[6, 15]}.  Now use @samp{*} to sum along the columns
to get @expr{[5, 7, 9]}. 
@xref{Matrix Answer 1, 1}. (@bullet{})

the matrix and vector.  If we multiplied in the other order, Calc would
assume the vector was a row vector in order to make the dimensions
come out right, and the answer would be incorrect.  If you
don't feel safe letting Calc take either interpretation of your
vectors, use explicit 
@texline @tmath{N\times1}
@infoline Nx1
or
@texline @tmath{1\times N}
@infoline 1xN
matrices instead.  In this case, you would enter the original column
vector as @samp{[[6], [2], [3]]} or @samp{[6; 2; 3]}.

(@bullet{}) @strong{Exercise 2.}  Algebraic entry allows you to make

@tex
\turnoffactive
$A^T A \, X = A^T B$, where $A^T$ is the transpose \samp{trn(A)}.
@end tex
Now 
@texline @tmath{A^T A}
@infoline @expr{trn(A)*A} 
is a square matrix so a solution is possible.  It turns out that the
@expr{X} vector you compute in this way will be a ``least-squares''
solution, which can be regarded as the ``closest'' solution to the set
of equations.  Use Calc to solve the following over-determined

other a plain number.)  In the final step, we take the square root
of each element.

(@bullet{}) @strong{Exercise 1.}  Compute a vector of powers of two
from 
@texline @tmath{2^{-4}}
@infoline @expr{2^-4} 
to @expr{2^4}.  @xref{List Answer 1, 1}. (@bullet{})

You can also @dfn{reduce} a binary operator across a vector.
For example, reducing @samp{*} computes the product of all the

\afterdisplay
@end tex

@noindent
where 
@texline @tmath{\sum x}
@infoline @expr{sum(x)} 
represents the sum of all the values of @expr{x}.  While there is an
actual @code{sum} function in Calc, it's easier to sum a vector using a
simple reduction.  First, let's compute the four different sums that
this formula uses.

   r 5            r 8 r 3 *       -              r 7 /   t 9
@end group
@end smallexample

Let's ``plot'' this straight line approximation, 
@texline @tmath{y \approx m x + b},
@infoline @expr{m x + b}, 
and compare it with the original data.

@smallexample
@group

@xref{List Answer 7, 7}. (@bullet{})

@cindex Maximizing a function over a list of values
@c [fix-ref Numerical Solutions]
(@bullet{}) @strong{Exercise 8.}  Compute a list of values of Bessel's
@texline @tmath{J_1(x)}
@infoline @expr{J1} 
function @samp{besJ(1,x)} for @expr{x} from 0 to 5 in steps of 0.25.
Find the value of @expr{x} (from among the above set of values) for
which @samp{besJ(1,x)} is a maximum.  Use an ``automatic'' method,
i.e., just reading along the list by hand to find the largest value

of thing automatically; @pxref{Numerical Solutions}.)
@xref{List Answer 8, 8}. (@bullet{})

@cindex Digits, vectors of
(@bullet{}) @strong{Exercise 9.}  You are given an integer in the range
@texline @tmath{0 \le N < 10^m}
@infoline @expr{0 <= N < 10^m} 
for @expr{m=12} (i.e., an integer of less than
twelve digits).  Convert this integer into a vector of @expr{m}
digits, each in the range from 0 to 9.  In vector-of-digits notation,
add one to this integer to produce a vector of @expr{m+1} digits

@kbd{V R a =} to test if all numbers in a list were equal.  What
happened?  How would you do this test?  @xref{List Answer 10, 10}. (@bullet{})

(@bullet{}) @strong{Exercise 11.}  The area of a circle of radius one
is @cpi{}.  The area of the 
@texline @tmath{2\times2}
@infoline 2x2
square that encloses that circle is 4.  So if we throw @var{n} darts at
random points in the square, about @cpiover{4} of them will land inside
the circle.  This gives us an entertaining way to estimate the value of 
@cpi{}.  The @w{@kbd{k r}}
command picks a random number between zero and the value on the stack.
We could get a random floating-point number between @i{-1} and 1 by typing
@w{@kbd{2.0 k r 1 -}}.  Build a vector of 100 random @expr{(x,y)} points in
this square, then use vector mapping and reduction to count how many
points lie inside the unit circle.  Hint:  Use the @kbd{v b} command.
@xref{List Answer 11, 11}. (@bullet{})

(@bullet{}) @strong{Exercise 12.}  The @dfn{matchstick problem} provides
another way to calculate @cpi{}.  Say you have an infinite field
of vertical lines with a spacing of one inch.  Toss a one-inch matchstick
onto the field.  The probability that the matchstick will land crossing
a line turns out to be 
@texline @tmath{2/\pi}.
@infoline @expr{2/pi}.  
Toss 100 matchsticks to estimate @cpi{}.  (If you want still more fun,
the probability that the GCD (@w{@kbd{k g}}) of two large integers is
one turns out to be 
@texline @tmath{6/\pi^2}.
@infoline @expr{6/pi^2}.
That provides yet another way to estimate @cpi{}.)
@xref{List Answer 12, 12}. (@bullet{})

(@bullet{}) @strong{Exercise 13.}  An algebraic entry of a string in

commands do nested function evaluations.  @kbd{H V U} takes a starting
value and a number of steps @var{n} from the stack; it then applies the
function you give to the starting value 0, 1, 2, up to @var{n} times
and returns a vector of the results.  Use this command to create a
``random walk'' of 50 steps.  Start with the two-dimensional point
@expr{(0,0)}; then take one step a random distance between @i{-1} and 1
in both @expr{x} and @expr{y}; then take another step, and so on.  Use the
@kbd{g f} command to display this random walk.  Now modify your random
walk to walk a unit distance, but in a random direction, at each step.
(Hint:  The @code{sincos} function returns a vector of the cosine and
sine of an angle.)  @xref{List Answer 14, 14}. (@bullet{})

    9 n        Q             c p            2 *            Q
@end group
@end smallexample

@noindent
The square root of @i{-9} is by default rendered in rectangular form
(@w{@expr{0 + 3i}}), but we can convert it to polar form (3 with a
phase angle of 90 degrees).  All the usual arithmetic and scientific
operations are defined on both types of complex numbers.

Another generalized kind of number is @dfn{infinity}.  Infinity

@end group
@end smallexample

@noindent
Since infinity is infinitely large, multiplying it by any finite
number (like @i{-17}) has no effect, except that since @i{-17}
is negative, it changes a plus infinity to a minus infinity.
(``A huge positive number, multiplied by @i{-17}, yields a huge
negative number.'')  Adding any finite number to infinity also
leaves it unchanged.  Taking an absolute value gives us plus
infinity again.  Finally, we add this plus infinity to the minus
infinity we had earlier.  If you work it out, you might expect
the answer to be @i{-72} for this.  But the 72 has been completely
lost next to the infinities; by the time we compute @w{@samp{inf - inf}}
the finite difference between them, if any, is undetectable.
So we say the result is @dfn{indeterminate}, which Calc writes
with the symbol @code{nan} (for Not A Number).

original error estimates were valid standard deviations, there is about
a 60% chance that the result is correct within 0.59 degrees.

@cindex Torus, volume of
(@bullet{}) @strong{Exercise 7.}  The volume of a torus (a donut shape) is
@texline @tmath{2 \pi^2 R r^2}
@infoline @w{@expr{2 pi^2 R r^2}} 
where @expr{R} is the radius of the circle that
defines the center of the tube and @expr{r} is the radius of the tube
itself.  Suppose @expr{R} is 20 cm and @expr{r} is 4 cm, each known to
within 5 percent.  What is the volume and the relative uncertainty of

that arises in the second one.

@cindex Fermat, primality test of
(@bullet{}) @strong{Exercise 10.}  A theorem of Pierre de Fermat
says that 
@texline @w{@tmath{x^{n-1} \bmod n = 1}}
@infoline @expr{x^(n-1) mod n = 1}
if @expr{n} is a prime number and @expr{x} is an integer less than
@expr{n}.  If @expr{n} is @emph{not} a prime number, this will
@emph{not} be true for most values of @expr{x}.  Thus we can test
informally if a number is prime by trying this formula for several

@noindent
This calculation tells me it is six hours and 22 minutes until midnight.

(@bullet{}) @strong{Exercise 11.}  A rule of thumb is that one year
is about 
@texline @tmath{\pi \times 10^7}
@infoline @w{@expr{pi * 10^7}} 
seconds.  What time will it be that many seconds from right now?
@xref{Types Answer 11, 11}. (@bullet{})

(@bullet{}) @strong{Exercise 12.}  You are preparing to order packaging

@end group
@end smallexample

@noindent
Calc has invented the variable @samp{s1} to represent an unknown sign;
it is supposed to be either @i{+1} or @i{-1}.  Here we have used
the ``let'' command to evaluate the expression when the sign is negative.
If we plugged this into our second derivative we would get the same,
negative, answer, so @expr{x = -1.19023} is also a maximum.

To find the actual maximum value, we must plug our two values of @expr{x}

@end group
@end smallexample

(@bullet{}) @strong{Exercise 3.}  Find the integral from 1 to @expr{y}
of 
@texline @tmath{x \sin \pi x}
@infoline @w{@expr{x sin(pi x)}} 
(where the sine is calculated in radians).  Find the values of the
integral for integers @expr{y} from 1 to 5.  @xref{Algebra Answer 3,
3}. (@bullet{})

Calc's integrator can do many simple integrals symbolically, but many
others are beyond its capabilities.  Suppose we wish to find the area
under the curve 
@texline @tmath{\sin x \ln x}
@infoline @expr{sin(x) ln(x)} 
over the same range of @expr{x}.  If you entered this formula and typed
@kbd{a i x @key{RET}} (don't bother to try this), Calc would work for a
long time but would be unable to find a solution.  In fact, there is no
closed-form solution to this integral.  Now what do we do?

           + f(a+(n-2)h) + f(a+(n-1)h)) $$
\afterdisplay
@end tex

Compute the integral from 1 to 2 of 
@texline @tmath{\sin x \ln x}
@infoline @expr{sin(x) ln(x)} 
using Simpson's rule with 10 slices.  
@xref{Algebra Answer 4, 4}. (@bullet{})

Calc has a built-in @kbd{a I} command for doing numerical integration.

@ignore
@starindex
@end ignore
@tindex Si
(@bullet{}) @strong{Exercise 1.}  The ``sine integral'' function
@texline @tmath{{\rm Si}(x)}
@infoline @expr{Si(x)} 
is defined as the integral of @samp{sin(t)/t} for
@expr{t = 0} to @expr{x} in radians.  (It was invented because this
integral has no solution in terms of basic functions; if you give it
to Calc's @kbd{a i} command, it will ponder it for a long time and then

the following functions:

@enumerate
@item
Compute 
@texline @tmath{\displaystyle{\sin x \over x}},
@infoline @expr{sin(x) / x}, 
where @expr{x} is the number on the top of the stack.

@item
Compute the base-@expr{b} logarithm, just like the @kbd{B} key except

@cindex Golden ratio
@cindex Phi, golden ratio
A fascinating property of the Fibonacci numbers is that the @expr{n}th
Fibonacci number can be found directly by computing 
@texline @tmath{\phi^n / \sqrt{5}}
@infoline @expr{phi^n / sqrt(5)}
and then rounding to the nearest integer, where 
@texline @tmath{\phi} (``phi''),
@infoline @expr{phi}, 
the ``golden ratio,'' is 
@texline @tmath{(1 + \sqrt{5}) / 2}.
@infoline @expr{(1 + sqrt(5)) / 2}. 
(For convenience, this constant is available from the @code{phi}
variable, or the @kbd{I H P} command.)

@smallexample

@end smallexample

@cindex Continued fractions
(@bullet{}) @strong{Exercise 5.}  The @dfn{continued fraction}
representation of 
@texline @tmath{\phi}
@infoline @expr{phi} 
is 
@texline @tmath{1 + 1/(1 + 1/(1 + 1/( \ldots )))}.
@infoline @expr{1 + 1/(1 + 1/(1 + 1/( ...@: )))}.
We can compute an approximate value by carrying this however far
and then replacing the innermost 
@texline @tmath{1/( \ldots )}
@infoline @expr{1/( ...@: )} 
by 1.  Approximate
@texline @tmath{\phi}
@infoline @expr{phi} 
using a twenty-term continued fraction.
@xref{Programming Answer 5, 5}. (@bullet{})

(@bullet{}) @strong{Exercise 6.}  Linear recurrences like the one for

@cindex Bernoulli numbers, approximate
The @dfn{Bernoulli numbers} are a sequence with the interesting
property that all of the odd Bernoulli numbers are zero, and the
even ones, while difficult to compute, can be roughly approximated
by the formula 
@texline @tmath{\displaystyle{2 n! \over (2 \pi)^n}}.
@infoline @expr{2 n!@: / (2 pi)^n}.  
Let's write a keyboard macro to compute (approximate) Bernoulli numbers.
(Calc has a command, @kbd{k b}, to compute exact Bernoulli numbers, but
this command is very slow for large @expr{n} since the higher Bernoulli
numbers are very large fractions.)

@end tex

@noindent
where @expr{f'(x)} is the derivative of @expr{f}.  The @expr{x}
values will quickly converge to a solution, i.e., eventually
@texline @tmath{x_{\rm new}}
@infoline @expr{new_x} 
and @expr{x} will be equal to within the limits
of the current precision.  Write a program which takes a formula
involving the variable @expr{x}, and an initial guess @expr{x_0},
on the stack, and produces a value of @expr{x} for which the formula
is zero.  Use it to find a solution of 
@texline @tmath{\sin(\cos x) = 0.5}
@infoline @expr{sin(cos(x)) = 0.5}
near @expr{x = 4.5}.  (Use angles measured in radians.)  Note that
the built-in @w{@kbd{a R}} (@code{calc-find-root}) command uses Newton's
method when it is able.  @xref{Programming Answer 8, 8}. (@bullet{})

@cindex Digamma function
@cindex Gamma constant, Euler's
@cindex Euler's gamma constant
(@bullet{}) @strong{Exercise 9.}  The @dfn{digamma} function 
@texline @tmath{\psi(z) (``psi'')}
@infoline @expr{psi(z)}
is defined as the derivative of 
@texline @tmath{\ln \Gamma(z)}.
@infoline @expr{ln(gamma(z))}.  
For large values of @expr{z}, it can be approximated by the infinite sum

@ifinfo
@example

\afterdisplay
@end tex

@noindent
where 
@texline @tmath{\sum}
@infoline @expr{sum} 
represents the sum over @expr{n} from 1 to infinity
(or to some limit high enough to give the desired accuracy), and
the @code{bern} function produces (exact) Bernoulli numbers.
While this sum is not guaranteed to converge, in practice it is safe.
An interesting mathematical constant is Euler's gamma, which is equal
to about 0.5772.  One way to compute it is by the formula,
@texline @tmath{\gamma = -\psi(1)}.
@infoline @expr{gamma = -psi(1)}.  
Unfortunately, 1 isn't a large enough argument
for the above formula to work (5 is a much safer value for @expr{z}).
Fortunately, we can compute 
@texline @tmath{\psi(1)}
@infoline @expr{psi(1)} 
from 
@texline @tmath{\psi(5)}
@infoline @expr{psi(5)} 
using the recurrence 
@texline @tmath{\psi(z+1) = \psi(z) + {1 \over z}}.
@infoline @expr{psi(z+1) = psi(z) + 1/z}.  
Your task:  Develop a program to compute 
@texline @tmath{\psi(z)};
@infoline @expr{psi(z)}; 
it should ``pump up'' @expr{z}
if necessary to be greater than 5, then use the above summation
formula.  Use looping commands to compute the sum.  Use your function
to compute 
@texline @tmath{\gamma}
@infoline @expr{gamma} 
to twelve decimal places.  (Calc has a built-in command
for Euler's constant, @kbd{I P}, which you can use to check your answer.)
@xref{Programming Answer 9, 9}. (@bullet{})

@noindent
@kbd{1 @key{RET} 2 @key{RET} 3 @key{RET} 4 + * -}

The result is 
@texline @tmath{1 - (2 \times (3 + 4)) = -13}.
@infoline @expr{1 - (2 * (3 + 4)) = -13}.

@node RPN Answer 2, RPN Answer 3, RPN Answer 1, Answers to Exercises
@subsection RPN Tutorial Exercise 2

@noindent
@texline @tmath{2\times4 + 7\times9.5 + {5\over4} = 75.75}
@infoline @expr{2*4 + 7*9.5 + 5/4 = 75.75}

After computing the intermediate term 
@texline @tmath{2\times4 = 8},
@infoline @expr{2*4 = 8}, 
you can leave that result on the stack while you compute the second
term.  With both of these results waiting on the stack you can then
compute the final term, then press @kbd{+ +} to add everything up.

Dividing two integers that are larger than the current precision may
give a floating-point result that is inaccurate even when rounded
down to an integer.  Consider @expr{123456789 / 2} when the current
precision is 6 digits.  The true answer is @expr{61728394.5}, but
with a precision of 6 this will be rounded to 
@texline @tmath{12345700.0/2.0 = 61728500.0}.
@infoline @expr{12345700.@: / 2.@: = 61728500.}.
The result, when converted to an integer, will be off by 106.

Here are two solutions:  Raise the precision enough that the
floating-point round-off error is strictly to the right of the

@node Matrix Answer 3, List Answer 1, Matrix Answer 2, Answers to Exercises
@subsection Matrix Tutorial Exercise 3

@noindent
To solve 
@texline @tmath{A^T A \, X = A^T B},
@infoline @expr{trn(A) * A * X = trn(A) * B}, 
first we compute
@texline @tmath{A' = A^T A}
@infoline @expr{A2 = trn(A) * A} 
and 
@texline @tmath{B' = A^T B};
@infoline @expr{B2 = trn(A) * B}; 
now, we have a system 
@texline @tmath{A' X = B'}
@infoline @expr{A2 * X = B2} 
which we can solve using Calc's @samp{/} command.

@ifinfo
@example

\afterdisplayh
@end tex

The first step is to enter the coefficient matrix.  We'll store it in
quick variable number 7 for later reference.  Next, we compute the
@texline @tmath{B'}
@infoline @expr{B2} 
vector.

@smallexample
@group

@end group
@end smallexample

@noindent
Now we compute the matrix 
@texline @tmath{A'}
@infoline @expr{A2} 
and divide.

@smallexample
@group

@noindent
(The actual computed answer will be slightly inexact due to
round-off error.)

Notice that the answers are similar to those for the 
@texline @tmath{3\times3}
@infoline 3x3
system solved in the text.  That's because the fourth equation that was 
added to the system is almost identical to the first one multiplied
by two.  (If it were identical, we would have gotten the exact same
answer since the 
@texline @tmath{4\times3}
@infoline 4x3
system would be equivalent to the original 
@texline @tmath{3\times3}
@infoline 3x3
system.)

Since the first and fourth equations aren't quite equivalent, they
can't both be satisfied at once.  Let's plug our answers back into

$$ m \times x + b \times 1 = y $$
\afterdisplay
@end tex

Thus we want a 
@texline @tmath{19\times2}
@infoline 19x2
matrix with our @expr{x} vector as one column and
ones as the other column.  So, first we build the column of ones, then
we combine the two columns to form our @expr{A} matrix.

@end group
@end smallexample

@noindent
Now we compute 
@texline @tmath{A^T y}
@infoline @expr{trn(A) * y} 
and 
@texline @tmath{A^T A}
@infoline @expr{trn(A) * A} 
and divide.

@smallexample
@group

    /
@end group
@end smallexample

Since we were solving equations of the form 
@texline @tmath{m \times x + b \times 1 = y},
@infoline @expr{m*x + b*1 = y}, 
these numbers should be @expr{m} and @expr{b}, respectively.  Sure
enough, they agree exactly with the result computed using @kbd{V M} and
@kbd{V R}!

@node List Answer 4, List Answer 5, List Answer 3, Answers to Exercises
@subsection List Tutorial Exercise 4

@noindent
A number @expr{j} is a divisor of @expr{n} if 
@texline @tmath{n \mathbin{\hbox{\code{\%}}} j = 0}.
@infoline @samp{n % j = 0}.  
The first step is to get a vector that identifies the divisors.

@smallexample
@group

so that the mapping operation works; no prime factor will ever be
zero, so adding zeros on the left and right is safe.  From then on
the job is pretty straightforward.

Incidentally, Calc provides the 
@texline @dfn{M@"obius} @tmath{\mu}
@infoline @dfn{Moebius mu} 
function which is zero if and only if its argument is square-free.  It
would be a much more convenient way to do the above test in practice.

@node List Answer 6, List Answer 7, List Answer 5, Answers to Exercises

The numbers down the lefthand edge of the list we desire are called
the ``triangular numbers'' (now you know why!).  The @expr{n}th
triangular number is the sum of the integers from 1 to @expr{n}, and
can be computed directly by the formula 
@texline @tmath{n (n+1) \over 2}.
@infoline @expr{n * (n+1) / 2}.

@smallexample
@group
2:  [ [0], [0, 1], ... ]    2:  [ [0], [0, 1], ... ]

@end smallexample

@noindent
It's a good idea to verify, as in the last step above, that only
one value is equal to the maximum.  (After all, a plot of 
@texline @tmath{\sin x}
@infoline @expr{sin(x)}
might have many points all equal to the maximum value, 1.)

The vector we have now has a single 1 in the position that indicates
the maximum value of @expr{x}.  Now it is a simple matter to convert

@noindent
This problem can be made a lot easier by taking advantage of some
symmetries.  First of all, after some thought it's clear that the
@expr{y} axis can be ignored altogether.  Just pick a random @expr{x}
component for one end of the match, pick a random direction 
@texline @tmath{\theta},
@infoline @expr{theta},
and see if @expr{x} and 
@texline @tmath{x + \cos \theta}
@infoline @expr{x + cos(theta)} 
(which is the @expr{x} coordinate of the other endpoint) cross a line.
The lines are at integer coordinates, so this happens when the two
numbers surround an integer.

In fact, since the field of lines is infinite we can choose the
coordinates 0 and 1 for the lines on either side of the leftmost
endpoint.  The rightmost endpoint will be between 0 and 1 if the
match does not cross a line, or between 1 and 2 if it does.  So:
Pick random @expr{x} and 
@texline @tmath{\theta},
@infoline @expr{theta}, 
compute
@texline @tmath{x + \cos \theta},
@infoline @expr{x + cos(theta)},
and count how many of the results are greater than one.  Simple!

We can make this go a bit faster by using the @kbd{v .} and @kbd{t .}
commands.

@end smallexample

@noindent
Aha!  It's unlikely that an irrational number would equal a fraction
this simple to within ten digits, so our original number was probably
@texline @tmath{\sqrt{27 \pi / 53}}.
@infoline @expr{sqrt(27 pi / 53)}.

Notice that we didn't need to re-round the number when we reduced the
precision.  Remember, arithmetic operations always round their inputs
to the current precision before they begin.

as a possible value.

The fourth calculation, @samp{1 / (-10 .. 10)}, has the same problem.
Zero is buried inside the interval, but it's still a possible value.
It's not hard to see that the actual result of @samp{1 / (-10 .. 10)}
will be either greater than @i{0.1}, or less than @i{-0.1}.  Thus
the interval goes from minus infinity to plus infinity, with a ``hole''
in it from @i{-0.1} to @i{0.1}.  Calc doesn't have any way to
represent this, so it just reports @samp{[-inf .. inf]} as the answer.
It may be disappointing to hear ``the answer lies somewhere between
minus infinity and plus infinity, inclusive,'' but that's the best
that interval arithmetic can do in this case.

    [ 3 n .. 3 ]        @key{RET} 2 ^           @key{TAB} @key{RET} *
@end group
@end smallexample

@noindent
In the first case the result says, ``if a number is between @i{-3} and
3, its square is between 0 and 9.''  The second case says, ``the product
of two numbers each between @i{-3} and 3 is between @i{-9} and 9.''

An interval form is not a number; it is a symbol that can stand for
many different numbers.  Two identical-looking interval forms can stand
for different numbers.

Each of these functions can be computed using the stack, or using
algebraic entry, whichever way you prefer:

@noindent
Computing 
@texline @tmath{\displaystyle{\sin x \over x}}:
@infoline @expr{sin(x) / x}:

Using the stack:  @kbd{C-x (  @key{RET} S @key{TAB} /  C-x )}.

Using algebraic entry:  @kbd{C-x (  ' sin($)/$ @key{RET}  C-x )}.

@end example

@noindent
This program is quite efficient because Calc knows how to raise a
matrix (or other value) to the power @expr{n} in only 
@texline @tmath{\log_2 n}
@infoline @expr{log(n,2)}
steps.  For example, this program can compute the 1000th Fibonacci
number (a 209-digit integer!) in about 10 steps; even though the
@kbd{Z < ... Z >} solution had much simpler steps, it would have
required so many steps that it would not have been practical.

@subsection Programming Tutorial Exercise 8

@noindent
The first step is to compute the derivative @expr{f'(x)} and thus
the formula 
@texline @tmath{\displaystyle{x - {f(x) \over f'(x)}}}.
@infoline @expr{x - f(x)/f'(x)}.

(Because this definition is long, it will be repeated in concise form
below.  You can use @w{@kbd{M-# m}} to load it from there.  While you are
entering a @kbd{Z ` Z '} body in a macro, Calc simply collects

@noindent
The first step is to adjust @expr{z} to be greater than 5.  A simple
``for'' loop will do the job here.  If @expr{z} is less than 5, we
reduce the problem using 
@texline @tmath{\psi(z) = \psi(z+1) - 1/z}.
@infoline @expr{psi(z) = psi(z+1) - 1/z}.  We go
on to compute 
@texline @tmath{\psi(z+1)},
@infoline @expr{psi(z+1)}, 
and remember to add back a factor of @expr{-1/z} when we're done.  This
step is repeated until @expr{z > 5}.

(Because this definition is long, it will be repeated in concise form

  @key{RET} 5        a <    Z [  5 Z (  & s + 2  1 s + 1  1 Z ) r 1  Z ]
@end group
@end smallexample

Now we compute the initial part of the sum:  
@texline @tmath{\ln z - {1 \over 2z}}
@infoline @expr{ln(z) - 1/2z}
minus the adjustment factor.

@smallexample
@group

  @key{TAB} @key{RET} M-@key{TAB}       - @key{RET} M-@key{TAB}      a =     Z /    2  Z )  Z ' C-x )
@end group
@end smallexample

This is the value of 
@texline @tmath{-\gamma},
@infoline @expr{- gamma}, 
with a slight bit of roundoff error.  To get a full 12 digits, let's use
a higher precision:

@smallexample

@subsection Programming Tutorial Exercise 10

@noindent
Taking the derivative of a term of the form @expr{x^n} will produce
a term like 
@texline @tmath{n x^{n-1}}.
@infoline @expr{n x^(n-1)}.  
Taking the derivative of a constant
produces zero.  From this it is easy to see that the @expr{n}th
derivative of a polynomial, evaluated at @expr{x = 0}, will equal the
coefficient on the @expr{x^n} term times @expr{n!}.

(@key{RET} corresponds to the @key{ENTER} key on most calculators.)
The first three keystrokes ``push'' the numbers 1 and 2 onto the stack.  The
@kbd{+} key always ``pops'' the top two numbers from the stack, adds them,
and pushes the result (3) back onto the stack.  This number is ready for
further calculations:  @kbd{5 -} pushes 5 onto the stack, then pops the
3 and 5, subtracts them, and pushes the result (@i{-2}).

Note that the ``top'' of the stack actually appears at the @emph{bottom}
of the buffer.  A line containing a single @samp{.} character signifies
the end of the buffer; Calculator commands operate on the number(s)
directly above this line.  The @kbd{d t} (@code{calc-truncate-stack})

@kbd{-}, @kbd{n}, and @kbd{_} (the underscore).  The @kbd{-} key subtracts
the two numbers on the top of the stack.  The @kbd{n} key changes the sign
of the number on the top of the stack or the number currently being entered.
The @kbd{_} key begins entry of a negative number or changes the sign of
the number currently being entered.  The following sequences all enter the
number @i{-5} onto the stack:  @kbd{0 @key{RET} 5 -}, @kbd{5 n @key{RET}},
@kbd{5 @key{RET} n}, @kbd{_ 5 @key{RET}}, @kbd{5 _ @key{RET}}.

Some other keys are active during numeric entry, such as @kbd{#} for
non-decimal numbers, @kbd{:} for fractions, and @kbd{@@} for HMS forms.
These notations are described later in this manual with the corresponding

@cindex Algebraic notation
@cindex Formulas, entering
Calculations can also be entered in algebraic form.  This is accomplished
by typing the apostrophe key, @kbd{'}, followed by the expression in
standard format:  @kbd{@key{'} 2+3*4 @key{RET}} computes
@texline @tmath{2+(3\times4) = 14}
@infoline @expr{2+(3*4) = 14} 
and pushes that on the stack.  If you wish you can
ignore the RPN aspect of Calc altogether and simply enter algebraic
expressions in this way.  You may want to use @key{DEL} every so often to
clear previous results off the stack.

is greater than this, it will recompute @cpi{} using a series
approximation.  This value will not need to be recomputed ever again
unless you raise the precision still further.  Many operations such as
logarithms and sines make use of similarly cached values such as
@cpiover{4} and 
@texline @tmath{\ln 2}.
@infoline @expr{ln(2)}.  
The visible effect of caching is that
high-precision computations may seem to do extra work the first time.
Other things cached include powers of two (for the binary arithmetic
functions), matrix inverses and determinants, symbolic integrals, and

@cindex Floating-point numbers
A floating-point number or @dfn{float} is a number stored in scientific
notation.  The number of significant digits in the fractional part is
governed by the current floating precision (@pxref{Precision}).  The
range of acceptable values is from 
@texline @tmath{10^{-3999999}}
@infoline @expr{10^-3999999} 
(inclusive) to 
@texline @tmath{10^{4000000}}
@infoline @expr{10^4000000}
(exclusive), plus the corresponding negative values and zero.

Calculations that would exceed the allowable range of values (such
as @samp{exp(exp(20))}) are left in symbolic form by Calc.  The

@var{imag} is the imaginary part, each of which may be any real number.
Rectangular complex numbers can also be displayed in @samp{@var{a}+@var{b}i}
notation; @pxref{Complex Formats}.

Polar complex numbers are displayed in the form 
@texline `@t{(}@var{r}@t{;}@tmath{\theta}@t{)}'
@infoline `@t{(}@var{r}@t{;}@var{theta}@t{)}'
where @var{r} is the nonnegative magnitude and 
@texline @tmath{\theta}
@infoline @var{theta} 
is the argument or phase angle.  The range of 
@texline @tmath{\theta}
@infoline @var{theta} 
depends on the current angular mode (@pxref{Angular Modes}); it is
generally between @i{-180} and @i{+180} degrees or the equivalent range
in radians. 

Complex numbers are entered in stages using incomplete objects.
@xref{Incomplete Objects}.

really mean is that @expr{1 / x}, as @expr{x} becomes larger and
larger, becomes arbitrarily close to zero.  So you can imagine
that if @expr{x} got ``all the way to infinity,'' then @expr{1 / x}
would go all the way to zero.  Similarly, when they say that
@samp{exp(inf) = inf}, they mean that 
@texline @tmath{e^x}
@infoline @expr{exp(x)} 
grows without bound as @expr{x} grows.  The symbol @samp{-inf} likewise
stands for an infinitely negative real value; for example, we say that
@samp{exp(-inf) = 0}.  You can have an infinity pointing in any
direction on the complex plane:  @samp{sqrt(-inf) = i inf}.

@starindex
@end ignore
@tindex vec
Algebraic functions for building vectors include @samp{vec(a, b, c)}
to build @samp{[a, b, c]}, @samp{cvec(a, n, m)} to build an 
@texline @tmath{n\times m}
@infoline @var{n}x@var{m}
matrix of @samp{a}s, and @samp{index(n)} to build a vector of integers
from 1 to @samp{n}.

@node Strings, HMS Forms, Vectors and Matrices, Data Types

The @var{hours} value is an integer (or integer-valued float).
The @var{mins} value is an integer or integer-valued float between 0 and 59.
The @var{secs} value is a real number between 0 (inclusive) and 60
(exclusive).  A positive HMS form is interpreted as @var{hours} +
@var{mins}/60 + @var{secs}/3600.  A negative HMS form is interpreted
as @i{- @var{hours}} @i{-} @var{mins}/60 @i{-} @var{secs}/3600.
Display format for HMS forms is quite flexible.  @xref{HMS Formats}.

HMS forms can be added and subtracted.  When they are added to numbers,
the numbers are interpreted according to the current angular mode.  HMS
forms can also be multiplied and divided by real numbers.  Dividing

between, say, @samp{<12:00am Mon Jan 1, 1900>} and
@samp{<12:00am Sat Jan 1, 2000>}.)

Calc uses the Julian calendar for all dates before the year 1752,
including dates BC when the Julian calendar technically had not
yet been invented.  Thus the claim that day number @i{-10000} is
called ``August 16, 28 BC'' should be taken with a grain of salt.

Please note that there is no ``year 0''; the day before
@samp{<Sat Jan 1, +1>} is @samp{<Fri Dec 31, -1>}.  These are
days 0 and @i{-1} respectively in Calc's internal numbering scheme.

@cindex Julian day counting
Another day counting system in common use is, confusingly, also
called ``Julian.''  It was invented in 1583 by Joseph Justus
Scaliger, who named it in honor of his father Julius Caesar
Scaliger.  For obscure reasons he chose to start his day
numbering on Jan 1, 4713 BC at noon, which in Calc's scheme
is @i{-1721423.5} (recall that Calc starts at midnight instead
of noon).  Thus to convert a Calc date code obtained by
unpacking a date form into a Julian day number, simply add
1721423.5.  The Julian code for @samp{6:00am Jan 9, 1991}
is 2448265.75.  The built-in @kbd{t J} command performs
this conversion for you.

A @dfn{modulo form} is a real number which is taken modulo (i.e., within
an integer multiple of) some value @var{M}.  Arithmetic modulo @var{M}
often arises in number theory.  Modulo forms are written
`@var{a} @t{mod} @var{M}',
where @var{a} and @var{M} are real numbers or HMS forms, and
@texline @tmath{0 \le a < M}.
@infoline @expr{0 <= a < @var{M}}.
In many applications @expr{a} and @expr{M} will be
integers but this is not required.

Modulo forms are not to be confused with the modulo operator @samp{%}.

@expr{M} is non-prime), or if any of the arguments are non-integers, the
division is left in symbolic form.  Other operations, such as square
roots, are not yet supported for modulo forms.  (Note that, although
@w{`@t{(}@var{a} @t{mod} @var{M}@t{)^.5}'} will compute a ``modulo square root''
in the sense of reducing 
@texline @tmath{\sqrt a}
@infoline @expr{sqrt(a)} 
modulo @expr{M}, this is not a useful definition from the
number-theoretical point of view.)

@ignore

@noindent
@cindex Error forms
@cindex Standard deviations
An @dfn{error form} is a number with an associated standard
deviation, as in @samp{2.3 +/- 0.12}.  The notation
@texline `@var{x} @t{+/-} @tmath{\sigma}' 
@infoline `@var{x} @t{+/-} sigma' 
stands for an uncertain value which follows
a normal or Gaussian distribution of mean @expr{x} and standard
deviation or ``error'' 
@texline @tmath{\sigma}.
@infoline @expr{sigma}.
Both the mean and the error can be either numbers or
formulas.  Generally these are real numbers but the mean may also be
complex.  If the error is negative or complex, it is changed to its
absolute value.  An error form with zero error is converted to a
regular number by the Calculator.

All arithmetic and transcendental functions accept error forms as input.
Operations on the mean-value part work just like operations on regular
numbers.  The error part for any function @expr{f(x)} (such as 
@texline @tmath{\sin x}
@infoline @expr{sin(x)})
is defined by the error of @expr{x} times the derivative of @expr{f}
evaluated at the mean value of @expr{x}.  For a two-argument function
@expr{f(x,y)} (such as addition) the error is the square root of the sum
of the squares of the errors due to @expr{x} and @expr{y}.

Consult a good text on error analysis for a discussion of the proper use
of standard deviations.  Actual errors often are neither Gaussian-distributed
nor uncorrelated, and the above formulas are valid only when errors
are small.  As an example, the error arising from
@texline `@t{sin(}@var{x} @t{+/-} @tmath{\sigma}@t{)}' 
@infoline `@t{sin(}@var{x} @t{+/-} @var{sigma}@t{)}' 
is 
@texline `@tmath{\sigma} @t{abs(cos(}@var{x}@t{))}'.  
@infoline `@var{sigma} @t{abs(cos(}@var{x}@t{))}'.  
When @expr{x} is close to zero,
@texline @tmath{\cos x}
@infoline @expr{cos(x)} 
is close to one so the error in the sine is close to 
@texline @tmath{\sigma};
@infoline @expr{sigma};
this makes sense, since 
@texline @tmath{\sin x}
@infoline @expr{sin(x)} 
is approximately @expr{x} near zero, so a given error in @expr{x} will
produce about the same error in the sine.  Likewise, near 90 degrees
@texline @tmath{\cos x}
@infoline @expr{cos(x)} 
is nearly zero and so the computed error is
small:  The sine curve is nearly flat in that region, so an error in @expr{x}
has relatively little effect on the value of 
@texline @tmath{\sin x}.
@infoline @expr{sin(x)}.  
However, consider @samp{sin(90 +/- 1000)}.  The cosine of 90 is zero, so
Calc will report zero error!  We get an obviously wrong result because
we have violated the small-error approximation underlying the error
analysis.  If the error in @expr{x} had been small, the error in
@texline @tmath{\sin x}
@infoline @expr{sin(x)} 
would indeed have been negligible.

@ignore
@mindex p

@samp{1 / (-2 .. 2) = [-inf .. inf]}.

While it may seem that intervals and error forms are similar, they are
based on entirely different concepts of inexact quantities.  An error
form 
@texline `@var{x} @t{+/-} @tmath{\sigma}' 
@infoline `@var{x} @t{+/-} @var{sigma}' 
means a variable is random, and its value could
be anything but is ``probably'' within one 
@texline @tmath{\sigma} 
@infoline @var{sigma} 
of the mean value @expr{x}. An interval 
`@t{[}@var{a} @t{..@:} @var{b}@t{]}' means a
variable's value is unknown, but guaranteed to lie in the specified
range.  Error forms are statistical or ``average case'' approximations;

@samp{=>} [@code{evalto}].

Note that, unlike in usual computer notation, multiplication binds more
strongly than division:  @samp{a*b/c*d} is equivalent to 
@texline @tmath{a b \over c d}.
@infoline @expr{(a*b)/(c*d)}.

@cindex Multiplication, implicit
@cindex Implicit multiplication
The multiplication sign @samp{*} may be omitted in many cases.  In particular,

element down to level @var{n}, making room for it by pulling all the
intervening stack elements toward the top.  @kbd{M-@key{TAB}} moves the
element at level @var{n} up to the top.  (Compare with @key{LFD},
which copies instead of moving the element in level @var{n}.)

With a negative argument @i{-@var{n}}, @key{TAB} rotates the stack
to move the object in level @var{n} to the deepest place in the
stack, and the object in level @i{@var{n}+1} to the top.  @kbd{M-@key{TAB}}
rotates the deepest stack element to be in level @i{n}, also
putting the top stack element in level @i{@var{n}+1}.

@xref{Selecting Subformulas}, for a way to apply these commands to
any portion of a vector or formula on the stack.

@node Editing Stack Entries, Trail Commands, Stack Manipulation, Stack and Trail

If the file name you give contains the string @samp{.emacs} anywhere
inside it, @kbd{m F} will not automatically load the new file.  This
is because you are presumably switching to your @file{~/.emacs} file,
which may contain other things you don't want to reread.  You can give
a numeric prefix argument of 1 to @kbd{m F} to force it to read the
file no matter what its name.  Conversely, an argument of @i{-1} tells
@kbd{m F} @emph{not} to read the new file.  An argument of 2 or @i{-2}
tells @kbd{m F} not to reset the modes to their defaults beforehand,
which is useful if you intend your new file to have a variant of the
modes present in the file you were using before.

@kindex m x

Likewise, the @kbd{H} key (@code{calc-hyperbolic}) sets or clears the
Hyperbolic Flag, which transforms @code{calc-sin} into @code{calc-sinh}.
If both of these flags are set at once, the effect will be
@code{calc-arcsinh}.  (The Hyperbolic flag is also used by some
non-trigonometric commands; for example @kbd{H L} computes a base-10,
instead of base-@i{e}, logarithm.)

Command names like @code{calc-arcsin} are provided for completeness, and
may be executed with @kbd{x} or @kbd{M-x}.  Their effect is simply to
toggle the Inverse and/or Hyperbolic flags and then execute the
corresponding base command (@code{calc-sin} in this case).

The @kbd{m i} command with a numeric prefix argument of zero,
i.e., @kbd{C-u 0 m i}, turns on a ``positive infinite mode'' in
which zero is treated as positive instead of being directionless.
Thus, @samp{1 / 0 = inf} and @samp{-1 / 0 = -inf} in this mode.
Note that zero never actually has a sign in Calc; there are no
separate representations for @i{+0} and @i{-0}.  Positive
infinite mode merely changes the interpretation given to the
single symbol, @samp{0}.  One consequence of this is that, while
you might expect @samp{1 / -0 = -inf}, actually @samp{1 / -0}
is equivalent to @samp{1 / 0}, which is equal to positive @code{inf}.

@end table

Calc uses this information to determine when certain simplifications
of formulas are safe.  For example, @samp{(x^y)^z} cannot be
simplified to @samp{x^(y z)} in general; for example,
@samp{((-3)^2)^1:2} is 3, but @samp{(-3)^(2*1:2) = (-3)^1} is @i{-3}.
However, this simplification @emph{is} safe if @code{z} is known
to be an integer, or if @code{x} is known to be a nonnegative
real number.  If you have given declarations that allow Calc to
deduce either of these facts, Calc will perform this simplification
of the formula.

@item const
The value is a constant with respect to other variables.
@end table

Calc does not check the declarations for a variable when you store
a value in it.  However, storing @i{-3.5} in a variable that has
been declared @code{pos}, @code{int}, or @code{matrix} may have
unexpected effects; Calc may evaluate @samp{sqrt(x^2)} to @expr{3.5}
if it substitutes the value first, or to @expr{-3.5} if @code{x}
was declared @code{pos} and the formula @samp{sqrt(x^2)} is
simplified to @samp{x} before the value is substituted.  Before

represent the integer and no more.  The @kbd{d z} (@code{calc-leading-zeros})
command causes integers to be padded out with leading zeros according to the
current binary word size.  (@xref{Binary Functions}, for a discussion of
word size.)  If the absolute value of the word size is @expr{w}, all integers
are displayed with at least enough digits to represent 
@texline @tmath{2^w-1}
@infoline @expr{(2^w)-1} 
in the current radix.  (Larger integers will still be displayed in their
entirety.) 

@node Grouping Digits, Float Formats, Radix Modes, Display Modes

by the arguments in parentheses.  However, functions for which @TeX{} has
special names (like @code{\sin}) will use curly braces instead of
parentheses for very simple arguments.  During input, curly braces and
parentheses work equally well for grouping, but when the document is
formatted the curly braces will be invisible.  Thus the printed result is
@texline @tmath{\sin{2 x}}
@infoline @expr{sin 2x} 
but 
@texline @tmath{\sin(2 + x)}.
@infoline @expr{sin(2 + x)}.

Function and variable names not treated specially by @TeX{} are simply
written out as-is, which will cause them to come out in italic letters
in the printed document.  If you invoke @kbd{d T} with a positive numeric

@item
Polar mode.  Value is 0 (rectangular) or 1 (polar); default is 0.
Command is @kbd{m p}.

@item
Matrix/scalar mode.  Default value is @i{-1}.  Value is 0 for scalar
mode, @i{-2} for matrix mode, or @var{N} for 
@texline @tmath{N\times N}
@infoline @var{N}x@var{N} 
matrix mode.  Command is @kbd{m v}.

@item
Simplification mode.  Default is 1.  Value is @i{-1} for off (@kbd{m O}),
0 for @kbd{m N}, 2 for @kbd{m B}, 3 for @kbd{m A}, 4 for @kbd{m E},
or 5 for @w{@kbd{m U}}.  The @kbd{m D} command accepts these prefixes.

@item
Infinite mode.  Default is @i{-1} (off).  Value is 1 if the mode is on,
or 0 if the mode is on with positive zeros.  Command is @kbd{m i}.
@end enumerate

For example, the sequence @kbd{M-1 m g @key{RET} 2 + ~ p} increases the
precision by two, leaving a copy of the old precision on the stack.

@mindex @null
@end ignore
@tindex -
The @kbd{-} (@code{calc-minus}) command subtracts two values.  The top
number on the stack is subtracted from the one behind it, so that the
computation @kbd{5 @key{RET} 2 -} produces 3, not @i{-3}.  All options
available for @kbd{+} are available for @kbd{-} as well.

@kindex *
@pindex calc-times
@ignore

@kindex f s
@pindex calc-sign
@tindex sign
The @kbd{f s} (@code{calc-sign}) [@code{sign}] command returns 1 if its
argument is positive, @i{-1} if its argument is negative, or 0 if its
argument is zero.  In algebraic form, you can also write @samp{sign(a,x)}
which evaluates to @samp{x * sign(a)}, i.e., either @samp{x}, @samp{-x}, or
zero depending on the sign of @samp{a}.

@kindex &

@tindex xpon
The @kbd{f M} (@code{calc-mant-part}) [@code{mant}] function extracts
the ``mantissa'' part @expr{m} of its floating-point argument; @kbd{f X}
(@code{calc-xpon-part}) [@code{xpon}] extracts the ``exponent'' part
@expr{e}.  The original number is equal to 
@texline @tmath{m \times 10^e},
@infoline @expr{m * 10^e},
where @expr{m} is in the interval @samp{[1.0 ..@: 10.0)} except that
@expr{m=e=0} if the original number is zero.  For integers
and fractions, @code{mant} returns the number unchanged and @code{xpon}
returns zero.  The @kbd{v u} (@code{calc-unpack}) command can also be

floating-point numbers, the change is by one unit in the last place.
For example, incrementing @samp{12.3456} when the current precision
is 6 digits yields @samp{12.3457}.  If the current precision had been
8 digits, the result would have been @samp{12.345601}.  Incrementing
@samp{0.0} produces 
@texline @tmath{10^{-p}},
@infoline @expr{10^-p}, 
where @expr{p} is the current
precision.  These operations are defined only on integers and floats.
With numeric prefix arguments, they change the number by @expr{n} units.

@end ignore
@kindex H F
The @kbd{F} (@code{calc-floor}) [@code{floor} or @code{ffloor}] command
truncates a real number to the next lower integer, i.e., toward minus
infinity.  Thus @kbd{3.6 F} produces 3, but @kbd{_3.6 F} produces
@i{-4}.

@kindex I F
@pindex calc-ceiling
@tindex ceil
@tindex fceil

@mindex @null
@end ignore
@kindex H I F
The @kbd{I F} (@code{calc-ceiling}) [@code{ceil} or @code{fceil}]
command truncates toward positive infinity.  Thus @kbd{3.6 I F} produces
4, and @kbd{_3.6 I F} produces @i{-3}.

@kindex R
@pindex calc-round
@tindex round
@tindex fround

@kindex H R
The @kbd{R} (@code{calc-round}) [@code{round} or @code{fround}] command
rounds to the nearest integer.  When the fractional part is .5 exactly,
this command rounds away from zero.  (All other rounding in the
Calculator uses this convention as well.)  Thus @kbd{3.5 R} produces 4
but @kbd{3.4 R} produces 3; @kbd{_3.5 R} produces @i{-4}.

@kindex I R
@pindex calc-trunc
@tindex trunc
@tindex ftrunc

@end ignore
@kindex H I R
The @kbd{I R} (@code{calc-trunc}) [@code{trunc} or @code{ftrunc}]
command truncates toward zero.  In other words, it ``chops off''
everything after the decimal point.  Thus @kbd{3.6 I R} produces 3 and
@kbd{_3.6 I R} produces @i{-3}.

These functions may not be applied meaningfully to error forms, but they
do work for intervals.  As a convenience, applying @code{floor} to a
modulo form floors the value part of the form.  Applied to a vector,
these functions operate on all elements of the vector one by one.

@pindex calc-argument
@tindex arg
The @kbd{G} (@code{calc-argument}) [@code{arg}] command computes the
``argument'' or polar angle of a complex number.  For a number in polar
notation, this is simply the second component of the pair
@texline `@t{(}@var{r}@t{;}@tmath{\theta}@t{)}'.
@infoline `@t{(}@var{r}@t{;}@var{theta}@t{)}'.
The result is expressed according to the current angular mode and will
be in the range @i{-180} degrees (exclusive) to @i{+180} degrees
(inclusive), or the equivalent range in radians.

@pindex calc-imaginary
The @code{calc-imaginary} command multiplies the number on the
top of the stack by the imaginary number @expr{i = (0,1)}.  This

@end ignore
@kindex v p (complex)
@pindex calc-pack
The @kbd{v p} (@code{calc-pack}) command can pack the top two numbers on
the stack into a composite object such as a complex number.  With
a prefix argument of @i{-1}, it produces a rectangular complex number;
with an argument of @i{-2}, it produces a polar complex number.
(Also, @pxref{Building Vectors}.)

@ignore
@mindex v u
@end ignore

@pindex calc-clean
@tindex pclean
The @kbd{c c} (@code{calc-clean}) [@code{pclean}] command ``cleans'' the
number on the top of the stack.  Floating point numbers are re-rounded
according to the current precision.  Polar numbers whose angular
components have strayed from the @i{-180} to @i{+180} degree range
are normalized.  (Note that results will be undesirable if the current
angular mode is different from the one under which the number was
produced!)  Integers and fractions are generally unaffected by this
operation.  Vectors and formulas are cleaned by cleaning each component
number (i.e., pervasively).

the year specified by the input.  The month, day, and time
of the input date form are lost.  With a numeric prefix argument
@var{n} in the range from 1 to 366, @kbd{t Y} computes the
@var{n}th day of the year (366 is treated as 365 in non-leap
years).  A prefix argument of 0 computes the last day of the
year (December 31).  A negative prefix argument from @i{-1} to
@i{-12} computes the first day of the @var{n}th month of the year.

@kindex t W
@pindex calc-new-week
@tindex newweek
The @kbd{t W} (@code{calc-new-week}) [@code{newweek}] command

The Lisp variable @code{math-daylight-savings-hook} holds the
name of a function that is used to compute the daylight savings
adjustment for a given date.  The default is
@code{math-std-daylight-savings}, which computes an adjustment
(either 0 or @i{-1}) using the North American rules given above.

The daylight savings hook function is called with four arguments:
The date, as a floating-point number in standard Calc format;
a six-element list of the date decomposed into year, month, day,
hour, minute, and second, respectively; a string which contains

@end smallexample

@noindent
The @code{bump} parameter is equal to zero when Calc is converting
from a date form in a generalized time zone into a GMT date value.
It is @i{-1} when Calc is converting in the other direction.  The
adjustments shown above ensure that the conversion behaves correctly
and reasonably around the 2 a.m.@: transition in each direction.

There is a ``missing'' hour between 2 a.m.@: and 3 a.m.@: at the
beginning of daylight savings time; converting a date/time form that

particular, negative arguments are converted to positive integers modulo
@expr{2^w} by all binary functions.

If the word size is negative, binary operations produce 2's complement
integers from 
@texline @tmath{-2^{-w-1}}
@infoline @expr{-(2^(-w-1))} 
to 
@texline @tmath{2^{-w-1}-1}
@infoline @expr{2^(-w-1)-1} 
inclusive.  Either mode accepts inputs in any range; the sign of
@expr{w} affects only the results produced.

@kindex b c

their results to the current word size.  Note that other operations like
addition do not use the current word size, since integer addition
generally is not ``binary.''  (However, @pxref{Simplification Modes},
@code{calc-bin-simplify-mode}.)  For example, with a word size of 8
bits @kbd{b c} converts a number to the range 0 to 255; with a word
size of @i{-8} @kbd{b c} converts to the range @i{-128} to 127.

@kindex b w
@pindex calc-word-size
The default word size is 32 bits.  All operations except the shifts and
rotates allow you to specify a different word size for that one

When the binary operations are written in symbolic form, they take an
optional second (or third) word-size parameter.  When a formula like
@samp{and(a,b)} is finally evaluated, the word size current at that time
will be used, but when @samp{and(a,b,-8)} is evaluated, a word size of
@i{-8} will always be used.  A symbolic binary function will be left
in symbolic form unless the all of its argument(s) are integers or
integer-valued floats.

If either or both arguments are modulo forms for which @expr{M} is a
power of two, that power of two is taken as the word size unless a

@cindex Golden ratio
One miscellaneous command is shift-@kbd{P} (@code{calc-pi}), which pushes
the value of @cpi{} (at the current precision) onto the stack.  With the
Hyperbolic flag, it pushes the value @expr{e}, the base of natural logarithms.
With the Inverse flag, it pushes Euler's constant 
@texline @tmath{\gamma}
@infoline @expr{gamma} 
(about 0.5772).  With both Inverse and Hyperbolic, it
pushes the ``golden ratio'' 
@texline @tmath{\phi}
@infoline @expr{phi} 
(about 1.618).  (At present, Euler's constant is not available
to unlimited precision; Calc knows only the first 100 digits.)
In Symbolic mode, these commands push the
actual variables @samp{pi}, @samp{e}, @samp{gamma}, and @samp{phi},

The @kbd{H L} (@code{calc-log10}) [@code{log10}] command computes the common
(base-10) logarithm of a number.  (With the Inverse flag [@code{exp10}],
it raises ten to a given power.)  Note that the common logarithm of a
complex number is computed by taking the natural logarithm and dividing
by 
@texline @tmath{\ln10}.
@infoline @expr{ln(10)}.

@kindex B
@kindex I B
@pindex calc-log
@tindex log
@tindex alog
The @kbd{B} (@code{calc-log}) [@code{log}] command computes a logarithm
to any base.  For example, @kbd{1024 @key{RET} 2 B} produces 10, since
@texline @tmath{2^{10} = 1024}.
@infoline @expr{2^10 = 1024}.  
In certain cases like @samp{log(3,9)}, the result
will be either @expr{1:2} or @expr{0.5} depending on the current Fraction
Mode setting.  With the Inverse flag [@code{alog}], this command is
similar to @kbd{^} except that the order of the arguments is reversed.

@kindex f E
@pindex calc-expm1
@tindex expm1
The @kbd{f E} (@code{calc-expm1}) [@code{expm1}] command computes
@texline @tmath{e^x - 1},
@infoline @expr{exp(x)-1}, 
but using an algorithm that produces a more accurate
answer when the result is close to zero, i.e., when 
@texline @tmath{e^x}
@infoline @expr{exp(x)} 
is close to one.

@kindex f L
@pindex calc-lnp1
@tindex lnp1
The @kbd{f L} (@code{calc-lnp1}) [@code{lnp1}] command computes
@texline @tmath{\ln(x+1)},
@infoline @expr{ln(x+1)}, 
producing a more accurate answer when @expr{x} is close to zero.

@node Trigonometric and Hyperbolic Functions, Advanced Math Functions, Logarithmic Functions, Scientific Functions
@section Trigonometric/Hyperbolic Functions

@kindex f T
@pindex calc-arctan2
@tindex arctan2
The @kbd{f T} (@code{calc-arctan2}) [@code{arctan2}] command takes two
numbers from the stack and computes the arc tangent of their ratio.  The
result is in the full range from @i{-180} (exclusive) to @i{+180}
(inclusive) degrees, or the analogous range in radians.  A similar
result would be obtained with @kbd{/} followed by @kbd{I T}, but the
value would only be in the range from @i{-90} to @i{+90} degrees
since the division loses information about the signs of the two
components, and an error might result from an explicit division by zero
which @code{arctan2} would avoid.  By (arbitrary) definition,
@samp{arctan2(0,0)=0}.

The @kbd{f g} (@code{calc-gamma}) [@code{gamma}] command computes the Euler
gamma function.  For positive integer arguments, this is related to the
factorial function:  @samp{gamma(n+1) = fact(n)}.  For general complex
arguments the gamma function can be defined by the following definite
integral:  
@texline @tmath{\Gamma(a) = \int_0^\infty t^{a-1} e^t dt}.
@infoline @expr{gamma(a) = integ(t^(a-1) exp(t), t, 0, inf)}.  
(The actual implementation uses far more efficient computational methods.)

@kindex f G
@tindex gammaP

@end ignore
@tindex gammaG
The @kbd{f G} (@code{calc-inc-gamma}) [@code{gammaP}] command computes
the incomplete gamma function, denoted @samp{P(a,x)}.  This is defined by
the integral, 
@texline @tmath{P(a,x) = \left( \int_0^x t^{a-1} e^t dt \right) / \Gamma(a)}.
@infoline @expr{gammaP(a,x) = integ(t^(a-1) exp(t), t, 0, x) / gamma(a)}.
This implies that @samp{gammaP(a,inf) = 1} for any @expr{a} (see the
definition of the normal gamma function).

Several other varieties of incomplete gamma function are defined.

@kindex f b
@pindex calc-beta
@tindex beta
The @kbd{f b} (@code{calc-beta}) [@code{beta}] command computes the
Euler beta function, which is defined in terms of the gamma function as
@texline @tmath{B(a,b) = \Gamma(a) \Gamma(b) / \Gamma(a+b)},
@infoline @expr{beta(a,b) = gamma(a) gamma(b) / gamma(a+b)}, 
or by
@texline @tmath{B(a,b) = \int_0^1 t^{a-1} (1-t)^{b-1} dt}.
@infoline @expr{beta(a,b) = integ(t^(a-1) (1-t)^(b-1), t, 0, 1)}.

@kindex f B
@kindex H f B
@pindex calc-inc-beta
@tindex betaI
@tindex betaB
The @kbd{f B} (@code{calc-inc-beta}) [@code{betaI}] command computes
the incomplete beta function @expr{I(x,a,b)}.  It is defined by
@texline @tmath{I(x,a,b) = \left( \int_0^x t^{a-1} (1-t)^{b-1} dt \right) / B(a,b)}.
@infoline @expr{betaI(x,a,b) = integ(t^(a-1) (1-t)^(b-1), t, 0, x) / beta(a,b)}.
Once again, the @kbd{H} (hyperbolic) prefix gives the corresponding
un-normalized version [@code{betaB}].

@kindex f e

@pindex calc-erf
@tindex erf
@tindex erfc
The @kbd{f e} (@code{calc-erf}) [@code{erf}] command computes the
error function 
@texline @tmath{\hbox{erf}(x) = {2 \over \sqrt{\pi}} \int_0^x e^{-t^2} dt}.
@infoline @expr{erf(x) = 2 integ(exp(-(t^2)), t, 0, x) / sqrt(pi)}.
The complementary error function @kbd{I f e} (@code{calc-erfc}) [@code{erfc}]
is the corresponding integral from @samp{x} to infinity; the sum
@texline @tmath{\hbox{erf}(x) + \hbox{erfc}(x) = 1}.
@infoline @expr{erf(x) + erfc(x) = 1}.

@kindex f j
@kindex f y
@pindex calc-bessel-J

   -, -eps      +eps, -      +eps, -
@end smallexample

For @samp{z1^z2}:  This is defined by @samp{exp(ln(z1)*z2)}.
One interesting consequence of this is that @samp{(-8)^1:3} does
not evaluate to @i{-2} as you might expect, but to the complex
number @expr{(1., 1.732)}.  Both of these are valid cube roots
of @i{-8} (as is @expr{(1., -1.732)}); Calc chooses a perhaps
less-obvious root for the sake of mathematical consistency.

For @samp{arcsin(z)}:  This is defined by @samp{-i*ln(i*z + sqrt(1-z^2))}.
The branch cuts are on the real axis, less than @i{-1} and greater than 1.

For @samp{arccos(z)}:  This is defined by @samp{-i*ln(z + i*sqrt(1-z^2))},
or equivalently by @samp{pi/2 - arcsin(z)}.  The branch cuts are on
the real axis, less than @i{-1} and greater than 1.

For @samp{arctan(z)}:  This is defined by
@samp{(ln(1+i*z) - ln(1-i*z)) / (2*i)}.  The branch cuts are on the
imaginary axis, below @expr{-i} and above @expr{i}.

For @samp{arccosh(z)}:  This is defined by
@samp{ln(z + (z+1)*sqrt((z-1)/(z+1)))}.  The branch cut is on the
real axis less than 1.

For @samp{arctanh(z)}:  This is defined by @samp{(ln(1+z) - ln(1-z)) / 2}.
The branch cuts are on the real axis, less than @i{-1} and greater than 1.

The following tables for @code{arcsin}, @code{arccos}, and
@code{arctan} assume the current angular mode is radians.  The
hyperbolic functions operate independently of the angular mode.

The @kbd{k r} (@code{calc-random}) [@code{random}] command produces
random numbers of various sorts.

Given a positive numeric prefix argument @expr{M}, it produces a random
integer @expr{N} in the range 
@texline @tmath{0 \le N < M}.
@infoline @expr{0 <= N < M}.  
Each of the @expr{M} values appears with equal probability.

With no numeric prefix argument, the @kbd{k r} command takes its argument
from the stack instead.  Once again, if this is a positive integer @expr{M}
the result is a random integer less than @expr{M}.  However, note that
while numeric prefix arguments are limited to six digits or so, an @expr{M}
taken from the stack can be arbitrarily large.  If @expr{M} is negative,
the result is a random integer in the range 
@texline @tmath{M < N \le 0}.
@infoline @expr{M < N <= 0}.

If the value on the stack is a floating-point number @expr{M}, the result
is a random floating-point number @expr{N} in the range 
@texline @tmath{0 \le N < M}
@infoline @expr{0 <= N < M}
or 
@texline @tmath{M < N \le 0},
@infoline @expr{M < N <= 0}, 
according to the sign of @expr{M}.

If @expr{M} is zero, the result is a Gaussian-distributed random real
number; the distribution has a mean of zero and a standard deviation
of one.  The algorithm used generates random numbers in pairs; thus,
every other call to this function will be especially fast.

If @expr{M} is an error form 
@texline @tmath{m} @code{+/-} @tmath{\sigma}
@infoline @samp{m +/- s} 
where @var{m} and 
@texline @tmath{\sigma}
@infoline @var{s} 
are both real numbers, the result uses a Gaussian distribution with mean
@var{m} and standard deviation 
@texline @tmath{\sigma}.
@var{s}.

If @expr{M} is an interval form, the lower and upper bounds specify the
acceptable limits of the random numbers.  If both bounds are integers,
the result is a random integer in the specified range.  If either bound

generators that are typically used to implement @code{random}.

If @code{RandSeed} contains an integer, Calc uses this integer to
seed an ``additive congruential'' method (Knuth's algorithm 3.2.2A,
computing 
@texline @tmath{X_{n-55} - X_{n-24}}.
@infoline @expr{X_n-55 - X_n-24}).  
This method expands the seed
value into a large table which is maintained internally; the variable
@code{RandSeed} is changed from, e.g., 42 to the vector @expr{[42]}
to indicate that the seed has been absorbed into this table.  When

or more, Calc throws it out and tries again until it gets a suitable
value.

To create a random floating-point number with precision @var{p}, Calc
simply creates a random @var{p}-digit integer and multiplies by
@texline @tmath{10^{-p}}.
@infoline @expr{10^-p}.  
The resulting random numbers should be very clean, but note
that relatively small numbers will have few significant random digits.
In other words, with a precision of 12, you will occasionally get
numbers on the order of 
@texline @tmath{10^{-9}}
@infoline @expr{10^-9} 
or 
@texline @tmath{10^{-10}},
@infoline @expr{10^-10}, 
but those numbers will only have two or three random digits since they
correspond to small integers times 
@texline @tmath{10^{-12}}.
@infoline @expr{10^-12}.

To create a random integer in the interval @samp{[0 .. @var{m})}, Calc
counts the digits in @var{m}, creates a random integer with three
additional digits, then reduces modulo @var{m}.  Unless @var{m} is a

@pindex calc-extended-gcd
@tindex egcd
The @kbd{k E} (@code{calc-extended-gcd}) [@code{egcd}] command computes
the GCD of two integers @expr{x} and @expr{y} and returns a vector
@expr{[g, a, b]} where 
@texline @tmath{g = \gcd(x,y) = a x + b y}.
@infoline @expr{g = gcd(x,y) = a x + b y}.

@kindex !
@pindex calc-factorial
@tindex fact

binomial coefficient @expr{N}-choose-@expr{M}, where @expr{M} is the number
on the top of the stack and @expr{N} is second-to-top.  If both arguments
are integers, the result is an exact integer.  Otherwise, the result is a
floating-point approximation.  The binomial coefficient is defined for all
real numbers by
@texline @tmath{N! \over M! (N-M)!\,}.
@infoline @expr{N! / M! (N-M)!}.

@kindex H k c
@pindex calc-perm
@tindex perm

@pindex calc-stirling-number
@tindex stir1
@tindex stir2
The @kbd{k s} (@code{calc-stirling-number}) [@code{stir1}] command
computes a Stirling number of the first 
@texline kind@tie{}@tmath{n \brack m},
@infoline kind,
given two integers @expr{n} and @expr{m} on the stack.  The @kbd{H k s}
[@code{stir2}] command computes a Stirling number of the second 
@texline kind@tie{}@tmath{n \brace m}.
@infoline kind.
These are the number of @expr{m}-cycle permutations of @expr{n} objects,
and the number of ways to partition @expr{n} objects into @expr{m}
non-empty sets, respectively.

to 25 million, the answer is exact although it may take some time.  The
result is a vector of the prime factors in increasing order.  For larger
inputs, prime factors above 5000 may not be found, in which case the
last number in the vector will be an unfactored integer greater than 25
million (with a warning message).  For negative integers, the first
element of the list will be @i{-1}.  For inputs @i{-1}, @i{0}, and
@i{1}, the result is a list of the same number.

@kindex k n
@pindex calc-next-prime
@ignore
@mindex nextpr@idots

@kindex k t
@pindex calc-totient
@tindex totient
The @kbd{k t} (@code{calc-totient}) [@code{totient}] command computes the
Euler ``totient'' 
@texline function@tie{}@tmath{\phi(n)},
@infoline function,
the number of integers less than @expr{n} which
are relatively prime to @expr{n}.

@kindex k m
@pindex calc-moebius
@tindex moebius
The @kbd{k m} (@code{calc-moebius}) [@code{moebius}] command computes the
@texline M@"obius @tmath{\mu}
@infoline Moebius ``mu''
function.  If the input number is a product of @expr{k}
distinct factors, this is @expr{(-1)^k}.  If the input number has any
duplicate factors (i.e., can be divided by the same prime more than once),
the result is zero.

@ignore
@mindex @null
@end ignore
@tindex ltpc
The @samp{utpc(x,v)} function uses the chi-square distribution with
@texline @tmath{\nu}
@infoline @expr{v} 
degrees of freedom.  It is the probability that a model is
correct if its chi-square statistic is @expr{x}.

@kindex k F

@mindex @null
@end ignore
@tindex ltpf
The @samp{utpf(F,v1,v2)} function uses the F distribution, used in
various statistical tests.  The parameters 
@texline @tmath{\nu_1}
@infoline @expr{v1} 
and 
@texline @tmath{\nu_2}
@infoline @expr{v2}
are the degrees of freedom in the numerator and denominator,
respectively, used in computing the statistic @expr{F}.

@kindex k N

@mindex @null
@end ignore
@tindex ltpn
The @samp{utpn(x,m,s)} function uses a normal (Gaussian) distribution
with mean @expr{m} and standard deviation 
@texline @tmath{\sigma}.
@infoline @expr{s}.  
It is the probability that such a normal-distributed random variable
would exceed @expr{x}.

@kindex k P

@mindex @null
@end ignore
@tindex ltpt
The @samp{utpt(t,v)} function uses the Student's ``t'' distribution
with 
@texline @tmath{\nu}
@infoline @expr{v} 
degrees of freedom.  It is the probability that a
t-distributed random variable will be greater than @expr{t}.
(Note:  This computes the distribution function 
@texline @tmath{A(t|\nu)}
@infoline @expr{A(t|v)}
where 
@texline @tmath{A(0|\nu) = 1}
@infoline @expr{A(0|v) = 1} 
and 
@texline @tmath{A(\infty|\nu) \to 0}.
@infoline @expr{A(inf|v) -> 0}.  
The @code{UTPT} operation on the HP-48 uses a different definition which
returns half of Calc's value:  @samp{UTPT(t,v) = .5*utpt(t,v)}.)

While Calc does not provide inverses of the probability distribution

The first is the mantissa; the second, which must be an
integer, is the exponent.  The result is the mantissa
times ten to the power of the exponent.

@item -12
This is treated the same as @i{-11} by the @kbd{v p} command.
When unpacking, @i{-12} specifies that a floating-point mantissa
is desired.

@item -13
A real number is converted into a date form.

returned in the form @samp{[@w{[a, b, c]}, [d, e, f]]}.

If any elements of the vector are negative, other kinds of
packing are done at that level as described above.  For
example, @samp{[2, 3, -4]} takes 12 objects and creates a
@texline @tmath{2\times3}
@infoline 2x3
matrix of error forms: @samp{[[a +/- b, c +/- d ... ]]}.
Also, @samp{[-4, -10]} will convert four integers into an
error form consisting of two fractions:  @samp{a:b +/- c:d}.

For example, given the vector @samp{[@w{a +/- b}, c^2, d +/- 7]},
the result of @kbd{C-u -4 v u} will be the two vectors
@samp{[a, c^2, d]} and @w{@samp{[b, 0, 7]}}.

Note that the prefix argument can have an effect even when the input is
not a vector.  For example, if the input is the number @i{-5}, then
@kbd{c-u -1 v u} yields @i{-5} and 0 (the components of @i{-5}
when viewed as a rectangular complex number); @kbd{C-u -2 v u} yields 5
and 180 (assuming degrees mode); and @kbd{C-u -10 v u} yields @i{-5}
and 1 (the numerator and denominator of @i{-5}, viewed as a rational
number).  Plain @kbd{v u} with this input would complain that the input
is not a composite object.

Unpacking mode @i{-11} converts a float into an integer mantissa and
an integer exponent, where the mantissa is not divisible by 10
(except that 0.0 is represented by a mantissa and exponent of 0).
Unpacking mode @i{-12} converts a float into a floating-point mantissa
and integer exponent, where the mantissa (for non-zero numbers)
is guaranteed to lie in the range [1 .. 10).  In both cases,
the mantissa is shifted left or right (and the exponent adjusted
to compensate) in order to satisfy these constraints.

prefix, if specified, must match the size of the vector.  If the value on
the stack is a scalar, it is used for each element on the diagonal, and
the prefix argument is required.

To build a constant square matrix, e.g., a 
@texline @tmath{3\times3}
@infoline 3x3
matrix filled with ones, use @kbd{0 M-3 v d 1 +}, i.e., build a zero
matrix first and then add a constant value to that matrix.  (Another
alternative would be to use @kbd{v b} and @kbd{v a}; see below.)

@tindex index
The @kbd{v x} (@code{calc-index}) [@code{index}] function builds a vector
of consecutive integers from 1 to @var{n}, where @var{n} is the numeric
prefix argument.  If you do not provide a prefix argument, you will be
prompted to enter a suitable number.  If @var{n} is negative, the result
is a vector of negative integers from @var{n} to @i{-1}.

With a prefix argument of just @kbd{C-u}, the @kbd{v x} command takes
three values from the stack: @var{n}, @var{start}, and @var{incr} (with
@var{incr} at top-of-stack).  Counting starts at @var{start} and increases
by @var{incr} for successive vector elements.  If @var{start} or @var{n}

@tindex mdims
With the Hyperbolic flag, @kbd{H v l} [@code{mdims}] computes a vector
of the dimensions of a vector, matrix, or higher-order object.  For
example, @samp{mdims([[a,b,c],[d,e,f]])} returns @samp{[2, 3]} since
its argument is a 
@texline @tmath{2\times3}
@infoline 2x3
matrix.

@kindex v f
@pindex calc-vector-find

If the number of columns does not evenly divide the number of elements
in the vector, the last row will be short and the result will not be
suitable for use as a matrix.  For example, with the matrix
@samp{[[1, 2], @w{[3, 4]}]} on the stack, @kbd{v a 4} produces
@samp{[[1, 2, 3, 4]]} (a 
@texline @tmath{1\times4}
@infoline 1x4
matrix), @kbd{v a 1} produces @samp{[[1], [2], [3], [4]]} (a 
@texline @tmath{4\times1}
@infoline 4x1
matrix), @kbd{v a 2} produces @samp{[[1, 2], [3, 4]]} (the original 
@texline @tmath{2\times2}
@infoline 2x2
matrix), @w{@kbd{v a 3}} produces @samp{[[1, 2, 3], [4]]} (not a
matrix), and @kbd{v a 0} produces the flattened list 
@samp{[1, 2, @w{3, 4}]}.

and only if it is in both of the input sets.  Thus if the input
sets are disjoint, i.e., if they share no common elements, the result
will be the empty vector @samp{[]}.  Note that the characters @kbd{V}
and @kbd{^} were chosen to be close to the conventional mathematical
notation for set 
@texline union@tie{}(@tmath{A \cup B})
@infoline union
and 
@texline intersection@tie{}(@tmath{A \cap B}).
@infoline intersection.

@kindex V -
@pindex calc-set-difference
@tindex vdiff

the same set.  The set may include positive infinity, but must
not include any negative numbers.  The input is interpreted as a
set of integers in the sense of @kbd{V F} (@code{vfloor}).  Beware
that a simple input like @samp{[100]} can result in a huge integer
representation 
@texline (@tmath{2^{100}}, a 31-digit integer, in this case).
@infoline (@expr{2^100}, a 31-digit integer, in this case).

@node Statistical Operations, Reducing and Mapping, Set Operations, Matrix Functions
@section Statistical Operations on Vectors

@tindex vmean
@cindex Mean of data values
The @kbd{u M} (@code{calc-vector-mean}) [@code{vmean}] command
computes the average (arithmetic mean) of the data values.
If the inputs are error forms 
@texline @tmath{x \pm \sigma},
@infoline @samp{x +/- s}, 
this is the weighted mean of the @expr{x} values with weights 
@texline @tmath{1 /\sigma^2}.
@infoline @expr{1 / s^2}.
@tex
\turnoffactive
$$ \mu = { \displaystyle \sum { x_i \over \sigma_i^2 } \over
           \displaystyle \sum { 1 \over \sigma_i^2 } } $$

If the inputs are not error forms, this is simply the sum of the
values divided by the count of the values.

Note that a plain number can be considered an error form with
error 
@texline @tmath{\sigma = 0}.
@infoline @expr{s = 0}.  
If the input to @kbd{u M} is a mixture of
plain numbers and error forms, the result is the mean of the
plain numbers, ignoring all values with non-zero errors.  (By the
above definitions it's clear that a plain number effectively

@tindex vsdev
@cindex Standard deviation
@cindex Sample statistics
The @kbd{u S} (@code{calc-vector-sdev}) [@code{vsdev}] command
computes the standard 
@texline deviation@tie{}@tmath{\sigma}
@infoline deviation
of the data values.  If the values are error forms, the errors are used
as weights just as for @kbd{u M}.  This is the @emph{sample} standard
deviation, whose value is the square root of the sum of the squares of
the differences between the values and the mean of the @expr{N} values,

This function also applies to distributions.  The standard deviation
of a single error form is simply the error part.  The standard deviation
of a continuous interval happens to equal the difference between the
limits, divided by 
@texline @tmath{\sqrt{12}}.
@infoline @expr{sqrt(12)}.  
The standard deviation of an integer interval is the same as the
standard deviation of a vector of those integers.

@kindex I u S

@cindex Variance of data values
The @kbd{H u S} (@code{calc-vector-variance}) [@code{vvar}] and
@kbd{H I u S} (@code{calc-vector-pop-variance}) [@code{vpvar}]
commands compute the variance of the data values.  The variance
is the 
@texline square@tie{}@tmath{\sigma^2}
@infoline square
of the standard deviation, i.e., the sum of the
squares of the deviations of the data values from the mean.
(This definition also applies when the argument is a distribution.)

The functions in this section take two arguments, which must be
vectors of equal size.  The vectors are each flattened in the same
way as by the single-variable statistical functions.  Given a numeric
prefix argument of 1, these functions instead take one object from
the stack, which must be an 
@texline @tmath{N\times2}
@infoline Nx2
matrix of data values.  Once again, variable names can be used in place
of actual vectors and matrices.

@kindex u C

If any argument to @kbd{V M} is a matrix, the operator is normally mapped
across all elements of the matrix.  For example, given the matrix
@expr{[[1, -2, 3], [-4, 5, -6]]}, @kbd{V M A} takes six absolute values to
produce another 
@texline @tmath{3\times2}
@infoline 3x2
matrix, @expr{[[1, 2, 3], [4, 5, 6]]}.

@tindex mapr
The command @kbd{V M _} [@code{mapr}] (i.e., type an underscore at the

If you give a numeric prefix argument of 2 to @kbd{a v}, it simplifies
as if in algebraic simplification mode.  This is equivalent to typing
@kbd{a s}; @pxref{Simplifying Formulas}.  If you give a numeric prefix
of 3 or more, it uses extended simplification mode (@kbd{a e}).

If you give a negative prefix argument @i{-1}, @i{-2}, or @i{-3},
it simplifies in the corresponding mode but only works on the top-level
function call of the formula.  For example, @samp{(2 + 3) * (2 + 3)} will
simplify to @samp{(2 + 3)^2}, without simplifying the sub-formulas
@samp{2 + 3}.  As another example, typing @kbd{V R +} to sum the vector
@samp{[1, 2, 3, 4]} produces the formula @samp{reduce(add, [1, 2, 3, 4])}

cases where the order of terms in a sum is changed by the default
simplifications.)

The distributive law is used to simplify sums in some cases:
@expr{a x + b x} to @expr{(a + b) x}, where @expr{a} represents
a number or an implicit 1 or @i{-1} (as in @expr{x} or @expr{-x})
and similarly for @expr{b}.  Use the @kbd{a c}, @w{@kbd{a f}}, or
@kbd{j M} commands to merge sums with non-numeric coefficients
using the distributive law.

The distributive law is only used for sums of two terms, or

@expr{(-a) (b - c)}, where @expr{-a} is a negative number, is
rewritten to @expr{a (c - b)}.

The distributive law of products and powers is used for adjacent
terms of the product: @expr{x^a x^b} goes to 
@texline @tmath{x^{a+b}}
@infoline @expr{x^(a+b)}
where @expr{a} is a number, or an implicit 1 (as in @expr{x}),
or the implicit one-half of @expr{@t{sqrt}(x)}, and similarly for
@expr{b}.  The result is written using @samp{sqrt} or @samp{1/sqrt}
if the sum of the powers is @expr{1/2} or @expr{-1/2}, respectively.
If the sum of the powers is zero, the product is simplified to
@expr{1} or to @samp{idn(1)} if matrix mode is enabled.

The product of a negative power times anything but another negative
power is changed to use division:  
@texline @tmath{x^{-2} y}
@infoline @expr{x^(-2) y} 
goes to @expr{y / x^2} unless matrix mode is
in effect and neither @expr{x} nor @expr{y} are scalar (in which
case it is considered unsafe to rearrange the order of the terms).

The quotient @expr{x / 0} is left unsimplified or changed to an
infinite quantity, as directed by the current infinite mode.
@xref{Infinite Mode}.

The expression 
@texline @tmath{a / b^{-c}}
@infoline @expr{a / b^(-c)} 
is changed to @expr{a b^c}, where @expr{-c} is any negative-looking
power.  Also, @expr{1 / b^c} is changed to 
@texline @tmath{b^{-c}}
@infoline @expr{b^(-c)} 
for any power @expr{c}.

Also, @expr{(-a) / b} and @expr{a / (-b)} go to @expr{-(a/b)};
@expr{(a/b) / c} goes to @expr{a / (b c)}; and @expr{a / (b/c)}

Powers of products or quotients @expr{(a b)^c}, @expr{(a/b)^c}
are distributed to @expr{a^c b^c}, @expr{a^c / b^c} only if @expr{c}
is an integer, or if either @expr{a} or @expr{b} are nonnegative
real numbers.  Powers of powers @expr{(a^b)^c} are simplified to
@texline @tmath{a^{b c}}
@infoline @expr{a^(b c)} 
only when @expr{c} is an integer and @expr{b c} also
evaluates to an integer.  Without these restrictions these simplifications
would not be safe because of problems with principal values.
(In other words, 
@texline @tmath{((-3)^{1/2})^2}
@infoline @expr{((-3)^1:2)^2} 
is safe to simplify, but
@texline @tmath{((-3)^2)^{1/2}}
@infoline @expr{((-3)^2)^1:2} 
is not.)  @xref{Declarations}, for ways to inform Calc that your
variables satisfy these requirements.

As a special case of this rule, @expr{@t{sqrt}(x)^n} is simplified to
@texline @tmath{x^{n/2}}
@infoline @expr{x^(n/2)} 
only for even integers @expr{n}.

If @expr{a} is known to be real, @expr{b} is an even integer, and
@expr{c} is a half- or quarter-integer, then @expr{(a^b)^c} is

Also, @expr{(-a)^b} is simplified to @expr{a^b} if @expr{b} is an
even integer, or to @expr{-(a^b)} if @expr{b} is an odd integer,
for any negative-looking expression @expr{-a}.

Square roots @expr{@t{sqrt}(x)} generally act like one-half powers
@texline @tmath{x^{1:2}}
@infoline @expr{x^1:2} 
for the purposes of the above-listed simplifications.

Also, note that 
@texline @tmath{1 / x^{1:2}}
@infoline @expr{1 / x^1:2} 
is changed to 
@texline @tmath{x^{-1:2}},
@infoline @expr{x^(-1:2)},
but @expr{1 / @t{sqrt}(x)} is left alone.

@tex
\bigskip

taken into account when terms of a product are being compared.
Thus @expr{(x + y) (y + x)} will be simplified to @expr{(x + y)^2}.
A subtle point is that @expr{(x - y) (y - x)} will @emph{not}
be simplified to @expr{-(x - y)^2}; Calc does not notice that
one term can be written as a constant times the other, even if
that constant is @i{-1}.

A fraction times any expression, @expr{(a:b) x}, is changed to
a quotient involving integers:  @expr{a x / b}.  This is not
done for floating-point numbers like @expr{0.5}, however.  This
is one reason why you may find it convenient to turn Fraction mode

Square roots of integer or rational arguments are simplified in
several ways.  (Note that these will be left unevaluated only in
Symbolic mode.)  First, square integer or rational factors are
pulled out so that @expr{@t{sqrt}(8)} is rewritten as
@texline @tmath{$2\,\t{sqrt}(2)$}.
@infoline @expr{2 sqrt(2)}.  
Conceptually speaking this implies factoring the argument into primes
and moving pairs of primes out of the square root, but for reasons of
efficiency Calc only looks for primes up to 29.

No simplifications for inverse trigonometric and hyperbolic
functions are known, except for negative arguments of @code{arcsin},
@code{arctan}, @code{arcsinh}, and @code{arctanh}.  Note that
@expr{@t{arcsin}(@t{sin}(x))} can @emph{not} safely change to
@expr{x}, since this only correct within an integer multiple of 
@texline @tmath{2 \pi}
@infoline @expr{2 pi} 
radians or 360 degrees.  However, @expr{@t{arcsinh}(@t{sinh}(x))} is
simplified to @expr{x} if @expr{x} is known to be real.

Several simplifications that apply to logarithms and exponentials
are that @expr{@t{exp}(@t{ln}(x))}, 
@texline @t{e}@tmath{^{\ln(x)}},
@infoline @expr{e^@t{ln}(x)}, 
and
@texline @tmath{10^{{\rm log10}(x)}}
@infoline @expr{10^@t{log10}(x)} 
all reduce to @expr{x}.  Also, @expr{@t{ln}(@t{exp}(x))}, etc., can
reduce to @expr{x} if @expr{x} is provably real.  The form
@expr{@t{exp}(x)^y} is simplified to @expr{@t{exp}(x y)}.  If @expr{x}
is a suitable multiple of 
@texline @tmath{\pi i} 
@infoline @expr{pi i}
(as described above for the trigonometric functions), then
@expr{@t{exp}(x)} or @expr{e^x} will be expanded.  Finally,
@expr{@t{ln}(x)} is simplified to a form involving @code{pi} and
@code{i} where @expr{x} is provably negative, positive imaginary, or

values of @expr{x} in a certain range; outside that range, values
are folded down to the 360-degree range that the inverse trigonometric
functions always produce.

Powers of powers @expr{(x^a)^b} are simplified to 
@texline @tmath{x^{a b}}
@infoline @expr{x^(a b)}
for all @expr{a} and @expr{b}.  These results will be valid only
in a restricted range of @expr{x}; for example, in 
@texline @tmath{(x^2)^{1:2}}
@infoline @expr{(x^2)^1:2}
the powers cancel to get @expr{x}, which is valid for positive values
of @expr{x} but not for negative or complex values.

Similarly, @expr{@t{sqrt}(x^a)} and @expr{@t{sqrt}(x)^a} are both
simplified (possibly unsafely) to 
@texline @tmath{x^{a/2}}.
@infoline @expr{x^(a/2)}.

Forms like @expr{@t{sqrt}(1 - sin(x)^2)} are simplified to, e.g.,
@expr{@t{cos}(x)}.  Calc has identities of this sort for @code{sin},
@code{cos}, @code{tan}, @code{sinh}, and @code{cosh}.

number for an answer, then the quotient simplifies to that number.

For powers and square roots, the ``unsafe'' simplifications
@expr{(a b)^c} to @expr{a^c b^c}, @expr{(a/b)^c} to @expr{a^c / b^c},
and @expr{(a^b)^c} to 
@texline @tmath{a^{b c}}
@infoline @expr{a^(b c)} 
are done if the powers are real numbers.  (These are safe in the context
of units because all numbers involved can reasonably be assumed to be
real.)

according to the previous paragraph.  For example, @samp{acre^1.5}
is simplified by noting that @expr{1.5 = 3:2}, that @samp{acre}
is defined in terms of @samp{m^2}, and that the 2 in the power of
@code{m} is a multiple of 2 in @expr{3:2}.  Thus, @code{acre^1.5} is
replaced by approximately 
@texline @tmath{(4046 m^2)^{1.5}}
@infoline @expr{(4046 m^2)^1.5}, 
which is then changed to 
@texline @tmath{4046^{1.5} \, (m^2)^{1.5}},
@infoline @expr{4046^1.5 (m^2)^1.5}, 
then to @expr{257440 m^3}.

The functions @code{float}, @code{frac}, @code{clean}, @code{abs},
as well as @code{floor} and the other integer truncation functions,

answer!

If you use the @code{deriv} function directly in an algebraic formula,
you can write @samp{deriv(f,x,x0)} which represents the derivative
of @expr{f} with respect to @expr{x}, evaluated at the point 
@texline @tmath{x=x_0}.
@infoline @expr{x=x0}.

If the formula being differentiated contains functions which Calc does
not know, the derivatives of those functions are produced by adding
primes (apostrophe characters).  For example, @samp{deriv(f(2x), x)}

respect to a variable.  The integrator is not guaranteed to work for
all integrable functions, but it is able to integrate several large
classes of formulas.  In particular, any polynomial or rational function
(a polynomial divided by a polynomial) is acceptable.  (Rational functions
don't have to be in explicit quotient form, however; 
@texline @tmath{x/(1+x^{-2})}
@infoline @expr{x/(1+x^-2)}
is not strictly a quotient of polynomials, but it is equivalent to
@expr{x^3/(x^2+1)}, which is.)  Also, square roots of terms involving
@expr{x} and @expr{x^2} may appear in rational functions being
integrated.  Finally, rational functions involving trigonometric or

@end tex

Please note that the current implementation of Calc's integrator sometimes
produces results that are significantly more complex than they need to
be.  For example, the integral Calc finds for 
@texline @tmath{1/(x+\sqrt{x^2+1})}
@infoline @expr{1/(x+sqrt(x^2+1))}
is several times more complicated than the answer Mathematica
returns for the same input, although the two forms are numerically
equivalent.  Also, any indefinite integral should be considered to have
an arbitrary constant of integration added to it, although Calc does not
write an explicit constant of integration in its result.  For example,
Calc's solution for 
@texline @tmath{1/(1+\tan x)}
@infoline @expr{1/(1+tan(x))} 
differs from the solution given in the @emph{CRC Math Tables} by a
constant factor of  
@texline @tmath{\pi i / 2}
@infoline @expr{pi i / 2},
due to a different choice of constant of integration.

The Calculator remembers all the integrals it has done.  If conditions
change in a way that would invalidate the old integrals, say, a switch

@end ignore
@tindex Ei
As a more serious example, the expression @samp{exp(x)/x} cannot be
integrated in terms of the standard functions, so the ``exponential
integral'' function 
@texline @tmath{{\rm Ei}(x)}
@infoline @expr{Ei(x)} 
was invented to describe it.
We can get Calc to do this integral in terms of a made-up @code{Ei}
function by adding the rule @samp{[integtry(exp(x)/x, x) := Ei(x)]}
to @code{IntegRules}.  Now entering @samp{exp(2x)/x} on the stack

form @expr{X = 0}.

This command also works for inequalities, as in @expr{y < 3x + 6}.
Some inequalities cannot be solved where the analogous equation could
be; for example, solving 
@texline @tmath{a < b \, c}
@infoline @expr{a < b c} 
for @expr{b} is impossible
without knowing the sign of @expr{c}.  In this case, @kbd{a S} will
produce the result 
@texline @tmath{b \mathbin{\hbox{\code{!=}}} a/c}
@infoline @expr{b != a/c} 
(using the not-equal-to operator) to signify that the direction of the
inequality is now unknown.  The inequality 
@texline @tmath{a \le b \, c}
@infoline @expr{a <= b c} 
is not even partially solved.  @xref{Declarations}, for a way to tell
Calc that the signs of the variables in a formula are in fact known.

Two useful commands for working with the result of @kbd{a S} are

Some equations have more than one solution.  The Hyperbolic flag
(@code{H a S}) [@code{fsolve}] tells the solver to report the fully
general family of solutions.  It will invent variables @code{n1},
@code{n2}, @dots{}, which represent independent arbitrary integers, and
@code{s1}, @code{s2}, @dots{}, which represent independent arbitrary
signs (either @i{+1} or @i{-1}).  If you don't use the Hyperbolic
flag, Calc will use zero in place of all arbitrary integers, and plus
one in place of all arbitrary signs.  Note that variables like @code{n1}
and @code{s1} are not given any special interpretation in Calc except by
the equation solver itself.  As usual, you can use the @w{@kbd{s l}}
(@code{calc-let}) command to obtain solutions for various actual values

value of the variable which minimizes the formula's value, along
with the minimum value itself.

Note that this command looks for a @emph{local} minimum.  Many functions
have more than one minimum; some, like 
@texline @tmath{x \sin x},
@infoline @expr{x sin(x)}, 
have infinitely many.  In fact, there is no easy way to define the
``global'' minimum of 
@texline @tmath{x \sin x}
@infoline @expr{x sin(x)} 
but Calc can still locate any particular local minimum
for you.  Calc basically goes downhill from the initial guess until it
finds a point at which the function's value is greater both to the left
and to the right.  Calc does not use derivatives when minimizing a function.

the @dfn{parameters} of the model.

The @kbd{a F} command takes the data set to be fitted from the stack.
By default, it expects the data in the form of a matrix.  For example,
for a linear or polynomial fit, this would be a 
@texline @tmath{2\times N}
@infoline 2xN
matrix where the first row is a list of @expr{x} values and the second
row has the corresponding @expr{y} values.  For the multilinear fit
shown above, the matrix would have four rows (@expr{x_1}, @expr{x_2},
@expr{x_3}, and @expr{y}, respectively).

If you happen to have an 
@texline @tmath{N\times2}
@infoline Nx2
matrix instead of a 
@texline @tmath{2\times N}
@infoline 2xN
matrix, just press @kbd{v t} first to transpose the matrix.

After you type @kbd{a F}, Calc prompts you to select a model.  For a
linear fit, press the digit @kbd{1}.

@noindent
which is clearly zero if @expr{a + b x} exactly fits all data points,
and increases as various @expr{a + b x_i} values fail to match the
corresponding @expr{y_i} values.  There are several reasons why the
summand is squared, one of them being to ensure that 
@texline @tmath{\chi^2 \ge 0}.
@infoline @expr{chi^2 >= 0}.
Least-squares fitting simply chooses the values of @expr{a} and @expr{b}
for which the error 
@texline @tmath{\chi^2}
@infoline @expr{chi^2} 
is as small as possible.

Other kinds of models do the same thing but with a different model
formula in place of @expr{a + b x_i}.

0.142857142855 x^2 + 1.34285714287 x + 3.59999999998
@end example

An important result from the theory of polynomial fitting is that it
is always possible to fit @var{n} data points exactly using a polynomial
of degree @i{@var{n}-1}, sometimes called an @dfn{interpolating polynomial}.
Using the modified (14) data matrix, a model number of 4 gives
a polynomial that exactly matches all five data points:

@example
0.04167 x^4 - 0.4167 x^3 + 1.458 x^2 - 0.08333 x + 4.

It is also possible for the @emph{input} to a fitting operation to
contain error forms.  The data values must either all include errors
or all be plain numbers.  Error forms can go anywhere but generally
go on the numbers in the last row of the data matrix.  If the last
row contains error forms
@texline `@var{y_i}@w{ @t{+/-} }@tmath{\sigma_i}', 
@infoline `@var{y_i}@w{ @t{+/-} }@var{sigma_i}', 
then the 
@texline @tmath{\chi^2}
@infoline @expr{chi^2}
statistic is now,

@ifinfo
@example

the fitting operation.

If there are error forms on other rows of the data matrix, all the
errors for a given data point are combined; the square root of the
sum of the squares of the errors forms the 
@texline @tmath{\sigma_i}
@infoline @expr{sigma_i} 
used for the data point.

Both @kbd{a F} and @kbd{H a F} can accept error forms in the input
matrix, although if you are concerned about error analysis you will
probably use @kbd{H a F} so that the output also contains error
estimates.

If the input contains error forms but all the 
@texline @tmath{\sigma_i}
@infoline @expr{sigma_i} 
values are the same, it is easy to see that the resulting fitted model
will be the same as if the input did not have error forms at all 
@texline (@tmath{\chi^2}
@infoline (@expr{chi^2}
is simply scaled uniformly by 
@texline @tmath{1 / \sigma^2},
@infoline @expr{1 / sigma^2}, 
which doesn't affect where it has a minimum).  But there @emph{will} be
a difference in the estimated errors of the coefficients reported by
@kbd{H a F}. 

will have length @expr{M = d+1} with the constant term first.

@item
The covariance matrix @expr{C} computed from the fit.  This is
an @var{m}x@var{m} symmetric matrix; the diagonal elements
@texline @tmath{C_{jj}}
@infoline @expr{C_j_j} 
are the variances 
@texline @tmath{\sigma_j^2}
@infoline @expr{sigma_j^2} 
of the parameters.  The other elements are covariances
@texline @tmath{\sigma_{ij}^2} 
@infoline @expr{sigma_i_j^2} 
that describe the correlation between pairs of parameters.  (A related
set of numbers, the @dfn{linear correlation coefficients} 
@texline @tmath{r_{ij}},
@infoline @expr{r_i_j},
are defined as 
@texline @tmath{\sigma_{ij}^2 / \sigma_i \, \sigma_j}.)
@infoline @expr{sigma_i_j^2 / sigma_i sigma_j}.)

@item
A vector of @expr{M} ``parameter filter'' functions whose
meanings are described below.  If no filters are necessary this
will instead be an empty vector; this is always the case for the
polynomial and multilinear fits described so far.

@item
The value of 
@texline @tmath{\chi^2}
@infoline @expr{chi^2} 
for the fit, calculated by the formulas shown above.  This gives a
measure of the quality of the fit; statisticians consider
@texline @tmath{\chi^2 \approx N - M}
@infoline @expr{chi^2 = N - M} 
to indicate a moderately good fit (where again @expr{N} is the number of
data points and @expr{M} is the number of parameters).

@item
A measure of goodness of fit expressed as a probability @expr{Q}.
This is computed from the @code{utpc} probability distribution
function using 
@texline @tmath{\chi^2}
@infoline @expr{chi^2} 
with @expr{N - M} degrees of freedom.  A
value of 0.5 implies a good fit; some texts recommend that often
@expr{Q = 0.1} or even 0.001 can signify an acceptable fit.  In
particular, 
@texline @tmath{\chi^2}
@infoline @expr{chi^2} 
statistics assume the errors in your inputs
follow a normal (Gaussian) distribution; if they don't, you may
have to accept smaller values of @expr{Q}.

The @expr{Q} value is computed only if the input included error
estimates.  Otherwise, Calc will report the symbol @code{nan}
for @expr{Q}.  The reason is that in this case the 
@texline @tmath{\chi^2}
@infoline @expr{chi^2}
value has effectively been used to estimate the original errors
in the input, and thus there is no redundant information left
over to use for a confidence test.
@end enumerate

Here is a complete list of the standard models recognized by @kbd{a F}:

@table @kbd
@item 1
Linear or multilinear.  @i{a + b x + c y + d z}.
@item 2-9
Polynomials.  @i{a + b x + c x^2 + d x^3}.
@item e
Exponential.  @i{a} @t{exp}@i{(b x)} @t{exp}@i{(c y)}.
@item E
Base-10 exponential.  @i{a} @t{10^}@i{(b x)} @t{10^}@i{(c y)}.
@item x
Exponential (alternate notation).  @t{exp}@i{(a + b x + c y)}.
@item X
Base-10 exponential (alternate).  @t{10^}@i{(a + b x + c y)}.
@item l
Logarithmic.  @i{a + b} @t{ln}@i{(x) + c} @t{ln}@i{(y)}.
@item L
Base-10 logarithmic.  @i{a + b} @t{log10}@i{(x) + c} @t{log10}@i{(y)}.
@item ^
General exponential.  @i{a b^x c^y}.
@item p
Power law.  @i{a x^b y^c}.
@item q
Quadratic.  @i{a + b (x-c)^2 + d (x-e)^2}.
@item g
Gaussian.  
@texline @tmath{{a \over b \sqrt{2 \pi}} \exp\left( -{1 \over 2} \left( x - c \over b \right)^2 \right)}.
@infoline @i{(a / b sqrt(2 pi)) exp(-0.5*((x-c)/b)^2)}.
@end table

All of these models are used in the usual way; just press the appropriate
letter at the model prompt, and choose variable names if you wish.  The
result will be a formula as shown in the above table, with the best-fit

Calc uses the principal values of inverse functions like @code{ln}
and @code{arcsin} when doing fits.  For example, when you enter
the model @samp{y = sin(a t + b)} Calc actually uses the easier
form @samp{arcsin(y) = a t + b}.  The @code{arcsin} function always
returns results in the range from @i{-90} to 90 degrees (or the
equivalent range in radians).  Suppose you had data that you
believed to represent roughly three oscillations of a sine wave,
so that the argument of the sine might go from zero to 
@texline @tmath{3\times360}
@infoline @i{3*360} 
degrees.
The above model would appear to be a good way to determine the
true frequency and phase of the sine wave, but in practice it
would fail utterly.  The righthand side of the actual model
@samp{arcsin(y) = a t + b} will grow smoothly with @expr{t}, but
the lefthand side will bounce back and forth between @i{-90} and 90.
No values of @expr{a} and @expr{b} can make the two sides match,
even approximately.

There is no good solution to this problem at present.  You could
restrict your data to small enough ranges so that the above problem

ln(y) = ln(a) + b ln(x)
@end example

@noindent
which matches the desired form with 
@texline @tmath{Y = \ln(y)},
@infoline @expr{Y = ln(y)}, 
@texline @tmath{A = \ln(a)},
@infoline @expr{A = ln(a)},
@expr{F = 1}, @expr{B = b}, and 
@texline @tmath{G = \ln(x)}.
@infoline @expr{G = ln(x)}.  
Calc thus computes the logarithms of your @expr{y} and @expr{x} values,
does a linear fit for @expr{A} and @expr{B}, then solves to get 
@texline @tmath{a = \exp(A)} 
@infoline @expr{a = exp(A)} 
and @expr{b = B}.

Another interesting example is the ``quadratic'' model, which can
be handled by expanding according to the distributive law.

y = a + b c^2 - 2 b c x + b x^2
@end example

@noindent
which matches with @expr{Y = y}, @expr{A = a + b c^2}, @expr{F = 1},
@expr{B = -2 b c}, @expr{G = x} (the @i{-2} factor could just as easily
have been put into @expr{G} instead of @expr{B}), @expr{C = b}, and
@expr{H = x^2}.

The Gaussian model looks quite complicated, but a closer examination
shows that it's actually similar to the quadratic model but with an

@end tex

A last desperate step would be to use the general-purpose
@code{minimize} function rather than @code{fit}.  After all, both
functions solve the problem of minimizing an expression (the 
@texline @tmath{\chi^2}
@infoline @expr{chi^2}
sum) by adjusting certain parameters in the expression.  The @kbd{a F}
command is able to use a vastly more efficient algorithm due to its
special knowledge about linear chi-square sums, but the @kbd{a N}
command can do the same thing by brute force.

A compromise would be to pick out a few parameters without which the
fit is linearizable, and use @code{minimize} on a call to @code{fit}
which efficiently takes care of the rest of the parameters.  The thing
to be minimized would be the value of 
@texline @tmath{\chi^2}
@infoline @expr{chi^2} 
returned as the fifth result of the @code{xfit} function:

@smallexample
minimize(xfit(gaus(a,b,c,d,x), x, [a,b,c], data)_5, d, guess)

of the sum of the squares of the errors.  It then changes @expr{x}
and @expr{y} to be plain numbers, and makes @expr{z} into an error
form with this combined error.  The @expr{Y(x,y,z)} part of the
linearized model is evaluated, and the result should be an error
form.  The error part of that result is used for 
@texline @tmath{\sigma_i}
@infoline @expr{sigma_i} 
for the data point.  If for some reason @expr{Y(x,y,z)} does not return 
an error form, the combined error from @expr{z} is used directly for 
@texline @tmath{\sigma_i}.
@infoline @expr{sigma_i}.  
Finally, @expr{z} is also stripped of its error
for use in computing @expr{F(x,y,z)}, @expr{G(x,y,z)} and so on;
the righthand side of the linearized model is computed in regular
arithmetic with no error forms.

(While these rules may seem complicated, they are designed to do
the most reasonable thing in the typical case that @expr{Y(x,y,z)}
depends only on the dependent variable @expr{z}, and in fact is
often simply equal to @expr{z}.  For common cases like polynomials
and multilinear models, the combined error is simply used as the
@texline @tmath{\sigma}
@infoline @expr{sigma} 
for the data point with no further ado.)

@tex
\bigskip

If the lower limit is greater than the upper limit (assuming a
positive step size), the result is generally zero.  However,
Calc only guarantees a zero result when the upper limit is
exactly one step less than the lower limit, i.e., if the number
of iterations is @i{-1}.  Thus @samp{sum(f(k), k, n, n-1)} is zero
but the sum from @samp{n} to @samp{n-2} may report a nonzero value
if Calc used a closed form solution.

Calc's logical predicates like @expr{a < b} return 1 for ``true''
and 0 for ``false.''  @xref{Logical Operations}.  This can be

this would be a division by zero.  But at @expr{k = k_0}, this
formula works out to the indeterminate form @expr{0 / 0}, which
Calc will not assume is zero.  Better would be to use
@samp{(k != k_0) ? 1/(k-k_0) : 0}; the @samp{? :} operator does
an ``if-then-else'' test:  This expression says, ``if 
@texline @tmath{k \ne k_0},
@infoline @expr{k != k_0},
then @expr{1/(k-k_0)}, else zero.''  Now the formula @expr{1/(k-k_0)}
will not even be evaluated by Calc when @expr{k = k_0}.

@cindex Alternating sums

f(a b) := a f(b) :: real(a)]} is stored in variable @samp{linearF},
then the rule set @samp{[f(0) := 0, import(linearF)]} will apply
all three rules.  It is possible to modify the imported rules
slightly:  @samp{import(x, v1, x1, v2, x2, @dots{})} imports
the rule set @expr{x} with all occurrences of 
@texline @tmath{v_1},
@infoline @expr{v1}, 
as either a variable name or a function name, replaced with 
@texline @tmath{x_1}
@infoline @expr{x1} 
and so on.  (If 
@texline @tmath{v_1}
@infoline @expr{v1} 
is used as a function name, then 
@texline @tmath{x_1}
@infoline @expr{x1}
must be either a function name itself or a @w{@samp{< >}} nameless
function; @pxref{Specifying Operators}.)  For example, @samp{[g(0) := 0,
import(linearF, f, g)]} applies the linearity rules to the function
@samp{g} instead of @samp{f}.  Imports can be nested, but the

the @kbd{u g} command gets the definition of one of these constants
in its normal terms, and @kbd{u b} expresses the definition in base
units.

Two units, @code{pi} and @code{fsc} (the fine structure constant,
approximately @i{1/137}) are dimensionless.  The units simplification
commands simply treat these names as equivalent to their corresponding
values.  However you can, for example, use @kbd{u c} to convert a pure
number into multiples of the fine structure constant, or @kbd{u b} to
convert this back into a pure number.  (When @kbd{u c} prompts for the
``old units,'' just enter a blank line to signify that the value

All the arithmetic stores accept the Inverse prefix to reverse the
order of the operands.  If @expr{v} represents the contents of the
variable, and @expr{a} is the value drawn from the stack, then regular
@w{@kbd{s -}} assigns 
@texline @tmath{v \coloneq v - a},
@infoline @expr{v := v - a}, 
but @kbd{I s -} assigns
@texline @tmath{v \coloneq a - v}.
@infoline @expr{v := a - v}.  
While @kbd{I s *} might seem pointless, it is
useful if matrix multiplication is involved.  Actually, all the
arithmetic stores use formulas designed to behave usefully both
forwards and backwards:

In the first case, ``x'' and ``y'' are each vectors (not necessarily of
the same length); either or both may instead be interval forms.  The
``z'' value must be a matrix with the same number of rows as elements
in ``x'', and the same number of columns as elements in ``y''.  The
result is a surface plot where 
@texline @tmath{z_{ij}}
@infoline @expr{z_ij} 
is the height of the point
at coordinate @expr{(x_i, y_j)} on the surface.  The 3D graph will
be displayed from a certain default viewpoint; you can change this
viewpoint by adding a @samp{set view} to the @samp{*Gnuplot Commands*}

that don't have common ``x'' values.  (Of course, the range of ``x''
values covered by all the curves ought to be roughly the same if
they are to look nice on the same graph.)

For example, to plot 
@texline @tmath{\sin n x}
@infoline @expr{sin(n x)} 
for integers @expr{n}
from 1 to 5, you could use @kbd{v x} to create a vector of integers
(@expr{n}), then @kbd{V M '} or @kbd{V M $} to map @samp{sin(n x)}
across this vector.  The resulting vector of formulas is suitable

@code{PointStyles} variables.  These variables initially have no stored
values, but if you store a vector of integers in one of these variables,
the @kbd{g a} and @kbd{g f} commands will use those style numbers
instead of the defaults for new curves that are added to the graph.
An entry should be a positive integer for a specific style, or 0 to let
the style be chosen automatically, or @i{-1} to turn off lines or points
altogether.  If there are more curves than elements in the vector, the
last few curves will continue to have the default styles.  Of course,
you can later use @kbd{g s} and @kbd{g S} to change any of these styles.

For example, @kbd{'[2 -1 3] @key{RET} s t LineStyles} causes the first curve

terminals with no special graphics facilities.  It writes a crude
picture of the graph composed of characters like @code{-} and @code{|}
to a buffer called @samp{*Gnuplot Trail*}, which Calc then displays.
The graph is made the same size as the Emacs screen, which on most
dumb terminals will be 
@texline @tmath{80\times24}
@infoline 80x24
characters.  The graph is displayed in
an Emacs ``recursive edit''; type @kbd{q} or @kbd{M-# M-#} to exit
the recursive edit and return to Calc.  Note that the @code{dumb}
device is present only in GNUPLOT 3.0 and later versions.

@w{@samp{2 +/- 3 4 +/- 5}}, then grabbing with a prefix argument of 8
would correctly split the line into two error forms.

@xref{Matrix Functions}, to see how to pull the matrix apart into its
constituent rows and columns.  (If it is a 
@texline @tmath{1\times1}
@infoline 1x1
matrix, just hit @kbd{v u} (@code{calc-unpack}) twice.)

@kindex M-# :
@kindex M-# _

same limit as last time.

@key{INV GCD} computes the LCM (least common multiple) function.

@key{INV FACT} is the gamma function.  
@texline @tmath{\Gamma(x) = (x-1)!}.
@infoline @expr{gamma(x) = (x-1)!}.

@key{PERM} is the number-of-permutations function, which is on the
@kbd{H k c} key in normal Calc.

If you give a numeric prefix argument of 1 to @kbd{Z (}, the loop is
forced to use upward-counting conventions.  In this case, if @var{initial}
is greater than @var{final} the body will not be executed at all.
Note that @var{step} may still be negative in this loop; the prefix
argument merely constrains the loop-finished test.  Likewise, a prefix
argument of @i{-1} forces downward-counting conventions.

@kindex Z @{
@kindex Z @}
@pindex calc-kbd-loop
@pindex calc-kbd-end-loop

@starindex
@end ignore
@tindex mysin
A somewhat limited sine function could be defined as follows, using the
well-known Taylor series expansion for 
@texline @tmath{\sin x}:
@infoline @samp{sin(x)}:

@smallexample
(defmath mysin ((float (anglep x)))
  (interactive 1 "mysn")

which is not a Lisp list.

Large integers are stored as lists of the form @samp{(bigpos @var{d0}
@var{d1} @var{d2} @dots{})} for positive integers 1000000 or more, or
@samp{(bigneg @var{d0} @var{d1} @var{d2} @dots{})} for negative integers
@i{-1000000} or less.  Each @var{d} is a base-1000 ``digit,'' a Lisp integer
from 0 to 999.  The least significant digit is @var{d0}; the last digit,
@var{dn}, which is always nonzero, is the most significant digit.  For
example, the integer @i{-12345678} is stored as @samp{(bigneg 678 345 12)}.

The distinction between small and large integers is entirely hidden from
the user.  In @code{defmath} definitions, the Lisp predicate @code{integerp}
returns true for either kind of integer, and in general both big and small
integers are accepted anywhere the word ``integer'' is used in this manual.

Floating-point numbers are stored in the form, @samp{(float @var{mant}
@var{exp})}, where @var{mant} (the ``mantissa'') is an integer less than
@samp{10^@var{p}} in absolute value (@var{p} represents the current
precision), and @var{exp} (the ``exponent'') is a fixnum.  The value of
the float is @samp{@var{mant} * 10^@var{exp}}.  For example, the number
@i{-3.14} is stored as @samp{(float -314 -2) = -314*10^-2}.  Other constraints
are that the number 0.0 is always stored as @samp{(float 0 0)}, and,
except for the 0.0 case, the rightmost base-10 digit of @var{mant} is
always nonzero.  (If the rightmost digit is zero, the number is
rearranged by dividing @var{mant} by ten and incrementing @var{exp}.)

is displayed.  If the prefix is one, the unary function @var{unary}
is applied to the top stack element, or, if @var{unary} is not
specified, nothing happens.  When the argument is two or more,
the binary function @var{func} is reduced across the top @var{arg}
stack elements; when the argument is negative, the function is
mapped between the next-to-top @i{-@var{arg}} stack elements and the
top element.
@end defun

@defun calc-stack-size
Return the number of elements on the stack as an integer.  This count

modulo forms are recursively floated.  If the argument is a variable
or formula, this calls @code{reject-arg}.
@end defun

@defun compare x y
Compare the numbers @var{x} and @var{y}, and return @i{-1} if
@samp{(lessp @var{x} @var{y})}, 1 if @samp{(lessp @var{y} @var{x})},
0 if @samp{(math-equal @var{x} @var{y})}, or 2 if the order is
undefined or cannot be determined.
@end defun

@samp{ceil(log10(@var{n}))}, but much more efficient.  Zero is
considered to have zero digits.
@end defun

@defun scale-int x n
Shift integer @var{x} left @var{n} decimal digits, or right @i{-@var{n}}
digits with truncation toward zero.
@end defun

@defun scale-rounding x n
Like @code{scale-int}, except that a right shift rounds to the nearest

@end defun

@defun quarter-integer n
If @var{n} is an integer or integer-valued float, this function
returns zero.  If @var{n} is a half-integer (i.e., an integer plus
@i{1:2} or 0.5), it returns 2.  If @var{n} is a quarter-integer,
it returns 1 or 3.  If @var{n} is anything else, this function
returns @code{nil}.
@end defun

@node Vector Lisp Functions, Symbolic Lisp Functions, Computational Lisp Functions, Internals

input data set.  Each entry may be a single value or a vector of values.

@c 20
@item
With a prefix argument of 1, take a single 
@texline @tmath{@var{n}\times2}
@infoline @i{@var{N}x2} 
matrix from the stack instead of two separate data vectors.

@c 21
@item
The row or column number @expr{n} may be given as a numeric prefix

@c 47
@item
The variable is replaced by the formula shown on the right.  The
Inverse flag reverses the order of the operands, e.g., @kbd{I s - x}
assigns 
@texline @tmath{x \coloneq a-x}.
@infoline @expr{x := a-x}.

@c 48
@item
Press @kbd{?} repeatedly to see how to choose a model.  Answer the
variables prompt with @expr{iv} or @expr{iv;pv} to specify
independent and parameter variables.  A positive prefix argument
takes @i{@var{n}+1} vectors from the stack; a zero prefix takes a matrix
and a vector from the stack.

@c 49
@item
With a plain @kbd{C-u} prefix, replace the current region of the

90066:fb79180b618d

@c @texline foo
@c    `foo' will appear only in TeX output
@c @infoline foo
@c    `foo' will appear only in non-TeX output

@c @expr{expr} will typeset an expression;
@c $x$ in TeX, @samp{x} otherwise.

@iftex
@macro texline{stuff}
\stuff\
@end macro
@alias infoline=comment
@tex
\gdef\exprsetup{\tex \let\t\ttfont \turnoffactive}
\gdef\expr{\exprsetup$\exprfinish}
\gdef\exprfinish#1{#1$\endgroup}
@end tex
@alias mathit=expr
@macro cpi{}
@math{@pi{}}
@end macro
@macro cpiover{den}
@math{@pi/\den\}

@alias texline=comment
@macro infoline{stuff}
\stuff\
@end macro
@alias expr=samp
@alias mathit=i
@macro cpi{}
@expr{pi}
@end macro
@macro cpiover{den}
@expr{pi/\den\}

@strong{RPN calculation.}  In RPN, you type the input number(s) first,
then the command to operate on the numbers.

@noindent
Type @kbd{2 @key{RET} 3 + Q} to compute 
@texline @math{\sqrt{2+3} = 2.2360679775}.
@infoline the square root of 2+3, which is 2.2360679775.

@noindent
Type @kbd{P 2 ^} to compute 
@texline @math{\pi^2 = 9.86960440109}.
@infoline the value of `pi' squared, 9.86960440109.

@noindent
Type @key{TAB} to exchange the order of these two results.

conventional ``algebraic'' notation.  To enter an algebraic formula,
use the apostrophe key.

@noindent
Type @kbd{' sqrt(2+3) @key{RET}} to compute 
@texline @math{\sqrt{2+3}}.
@infoline the square root of 2+3.

@noindent
Type @kbd{' pi^2 @key{RET}} to enter 
@texline @math{\pi^2}.
@infoline `pi' squared.  
To evaluate this symbolic formula as a number, type @kbd{=}.

@noindent
Type @kbd{' arcsinh($ - $$) @key{RET}} to subtract the second-most-recent

the upper-leftmost @samp{1} and set the mark, then move to just after
the lower-right @samp{8} and press @kbd{M-# r}.

@noindent
Type @kbd{v t} to transpose this 
@texline @math{3\times2}
@infoline 3x2 
matrix into a 
@texline @math{2\times3}
@infoline 2x3
matrix.  Type @w{@kbd{v u}} to unpack the rows into two separate
vectors.  Now type @w{@kbd{V R + @key{TAB} V R +}} to compute the sums
of the two original columns. (There is also a special
grab-and-sum-columns command, @kbd{M-# :}.)

commands like @kbd{M-x calc} (for the standard user interface) or
@w{@kbd{M-x calc-keypad}} (for Keypad Mode).  First type @kbd{M-x}
(that's Meta with the letter @kbd{x}), then, at the prompt,
type the full command (like @kbd{calc-keypad}) and press Return.

The same commands (like @kbd{M-# c} or @kbd{M-# M-#}) that start
the Calculator also turn it off if it is already on.

@node The Standard Interface, Quick Mode Overview, Starting Calc, Using Calc
@subsection The Standard Calc Interface

a printer, the trail corresponds to the paper tape that records what
you do.

In this case, the trail shows that four numbers (17.3, 3, 2, and 4)
were first entered into the Calculator, then the 2 and 4 were
multiplied to get 8, then the 3 and 8 were subtracted to get @mathit{-5}.
(The @samp{>} symbol shows that this was the most recent calculation.)
The net result is the two numbers 17.3 and @mathit{-5} sitting on the stack.

Most Calculator commands deal explicitly with the stack only, but
there is a set of commands that allow you to search back through
the trail and retrieve any previous result.

full complexity of the stack and trail.  To use it, type @kbd{M-# q}
(@code{quick-calc}) in any regular editing buffer.

Quick Mode is very simple:  It prompts you to type any formula in
standard algebraic notation (like @samp{4 - 2/3}) and then displays
the result at the bottom of the Emacs screen (@mathit{3.33333333333}
in this case).  You are then back in the same editing buffer you
were in before, ready to continue editing or to type @kbd{M-# q}
again to do another quick calculation.  The result of the calculation
will also be in the Emacs ``kill ring'' so that a @kbd{C-y} command
at this point will yank the result into your editing buffer.

@noindent
@dfn{Keypad Mode} is a mouse-based interface to the Calculator.
It is designed for use with terminals that support a mouse.  If you
don't have a mouse, you will have to operate keypad mode with your
arrow keys (which is probably more trouble than it's worth).

Type @kbd{M-# k} to turn Keypad Mode on or off.  Once again you
get two new windows, this time on the righthand side of the screen
instead of at the bottom.  The upper window is the familiar Calc
Stack; the lower window is a picture of a typical calculator keypad.

@noindent
Calc was originally started as a two-week project to occupy a lull
in the author's schedule.  Basically, a friend asked if I remembered
the value of 
@texline @math{2^{32}}.
@infoline @expr{2^32}.  
I didn't offhand, but I said, ``that's easy, just call up an
@code{xcalc}.''  @code{Xcalc} duly reported that the answer to our
question was @samp{4.294967e+09}---with no way to see the full ten
digits even though we knew they were there in the program's memory!  I

@key{RET} 3 @key{RET} 4 + * -} compute?  (@samp{*} is the symbol for
multiplication.)  Figure it out by hand, then try it with Calc to see
if you're right.  @xref{RPN Answer 1, 1}. (@bullet{})

(@bullet{}) @strong{Exercise 2.}  Compute 
@texline @math{(2\times4) + (7\times9.4) + {5\over4}}
@infoline @expr{2*4 + 7*9.5 + 5/4} 
using the stack.  @xref{RPN Answer 2, 2}. (@bullet{})

The @key{DEL} key is called Backspace on some keyboards.  It is
whatever key you would use to correct a simple typing error when

$$ 2 + { 3 \times 4 \times 5 \over 6 \times 7^8 } - 9 $$
\afterdisplay
@end tex

@noindent
The result of this expression will be the number @mathit{-6.99999826533}.

Calc's order of evaluation is the same as for most computer languages,
except that @samp{*} binds more strongly than @samp{/}, as the above
example shows.  As in normal mathematical notation, the @samp{*} symbol
can often be omitted:  @samp{2 a} is the same as @samp{2*a}.

Operators at the same level are evaluated from left to right, except
that @samp{^} is evaluated from right to left.  Thus, @samp{2-3-4} is
equivalent to @samp{(2-3)-4} or @mathit{-5}, whereas @samp{2^3^4} is equivalent
to @samp{2^(3^4)} (a very large integer; try it!).

If you tire of typing the apostrophe all the time, there is an
``algebraic mode'' you can select in which Calc automatically senses
when you are about to type an algebraic expression.  To enter this

intermediate results of a calculation as you go along.  You can
accomplish this in Calc by performing your calculation as a series
of algebraic entries, using the @kbd{$} sign to tie them together.
In an algebraic formula, @kbd{$} represents the number on the top
of the stack.  Here, we perform the calculation 
@texline @math{\sqrt{2\times4+1}},
@infoline @expr{sqrt(2*4+1)},
which on a traditional calculator would be done by pressing
@kbd{2 * 4 + 1 =} and then the square-root key.

@smallexample

@end smallexample

@noindent
The shift-@kbd{S} command computes the sine of an angle.  The sine
of 45 degrees is 
@texline @math{\sqrt{2}/2};
@infoline @expr{sqrt(2)/2}; 
squaring this yields @expr{2/4 = 0.5}.  However, there has been a slight
roundoff error because the representation of 
@texline @math{\sqrt{2}/2}
@infoline @expr{sqrt(2)/2} 
wasn't exact.  The @kbd{c 1} command is a handy way to clean up numbers
in this case; it temporarily reduces the precision by one digit while it
re-rounds the number on the top of the stack.

@end group
@end smallexample

@noindent
Here we compute the Inverse Sine of 
@texline @math{\sqrt{0.5}},
@infoline @expr{sqrt(0.5)}, 
first in radians, then in degrees.

Use @kbd{c d} and @kbd{c r} to convert a number from radians to degrees
and vice-versa.

logarithm).  These can be modified by the @kbd{I} (inverse) and
@kbd{H} (hyperbolic) prefix keys.

Let's compute the sine and cosine of an angle, and verify the
identity 
@texline @math{\sin^2x + \cos^2x = 1}.
@infoline @expr{sin(x)^2 + cos(x)^2 = 1}.  
We'll arbitrarily pick @mathit{-64} degrees as a good value for @expr{x}.
With the angular mode set to degrees (type @w{@kbd{m d}}), do:

@smallexample
@group
2:  -64        2:  -64        2:  -0.89879   2:  -0.89879   1:  1.

Remember, @kbd{f h} is the @code{calc-hypot}, or square-root of sum
of squares, command.

Another identity is 
@texline @math{\displaystyle\tan x = {\sin x \over \cos x}}.
@infoline @expr{tan(x) = sin(x) / cos(x)}.
@smallexample
@group

2:  -0.89879   1:  -2.0503    1:  -64.

@end group
@end smallexample

A physical interpretation of this calculation is that if you move
@expr{0.89879} units downward and @expr{0.43837} units to the right,
your direction of motion is @mathit{-64} degrees from horizontal.  Suppose
we move in the opposite direction, up and to the left:

@smallexample
@group
2:  -0.89879   2:  0.89879    1:  -2.0503    1:  -64.

the top two stack elements right after the @kbd{U U}, then a pair of
@kbd{M-@key{TAB}} commands to cycle the 116 up around the duplicates.

A similar identity is supposed to hold for hyperbolic sines and cosines,
except that it is the @emph{difference}
@texline @math{\cosh^2x - \sinh^2x}
@infoline @expr{cosh(x)^2 - sinh(x)^2} 
that always equals one.  Let's try to verify this identity.

@smallexample
@group

exact integer, you get an exact integer result (158 digits long
in this case).

If you take the factorial of a non-integer, Calc uses a generalized
factorial function defined in terms of Euler's Gamma function
@texline @math{\Gamma(n)}
@infoline @expr{gamma(n)}
(which is itself available as the @kbd{f g} command).

@smallexample
@group

@end group
@end smallexample

@noindent
Here we verify the identity 
@texline @math{n! = \Gamma(n+1)}.
@infoline @expr{@var{n}!@: = gamma(@var{n}+1)}.

The binomial coefficient @var{n}-choose-@var{m}
@texline or @math{\displaystyle {n \choose m}}
is defined by
@texline @math{\displaystyle {n! \over m! \, (n-m)!}}
@infoline @expr{n!@: / m!@: (n-m)!}
for all reals @expr{n} and @expr{m}.  The intermediate results in this
formula can become quite large even if the final result is small; the
@kbd{k c} command computes a binomial coefficient in a way that avoids
large intermediate values.

rows in the matrix is different from the number of elements in the
vector.

(@bullet{}) @strong{Exercise 1.}  Use @samp{*} to sum along the rows
of the above 
@texline @math{2\times3}
@infoline 2x3 
matrix to get @expr{[6, 15]}.  Now use @samp{*} to sum along the columns
to get @expr{[5, 7, 9]}. 
@xref{Matrix Answer 1, 1}. (@bullet{})

the matrix and vector.  If we multiplied in the other order, Calc would
assume the vector was a row vector in order to make the dimensions
come out right, and the answer would be incorrect.  If you
don't feel safe letting Calc take either interpretation of your
vectors, use explicit 
@texline @math{N\times1}
@infoline Nx1
or
@texline @math{1\times N}
@infoline 1xN
matrices instead.  In this case, you would enter the original column
vector as @samp{[[6], [2], [3]]} or @samp{[6; 2; 3]}.

(@bullet{}) @strong{Exercise 2.}  Algebraic entry allows you to make

@tex
\turnoffactive
$A^T A \, X = A^T B$, where $A^T$ is the transpose \samp{trn(A)}.
@end tex
Now 
@texline @math{A^T A}
@infoline @expr{trn(A)*A} 
is a square matrix so a solution is possible.  It turns out that the
@expr{X} vector you compute in this way will be a ``least-squares''
solution, which can be regarded as the ``closest'' solution to the set
of equations.  Use Calc to solve the following over-determined

other a plain number.)  In the final step, we take the square root
of each element.

(@bullet{}) @strong{Exercise 1.}  Compute a vector of powers of two
from 
@texline @math{2^{-4}}
@infoline @expr{2^-4} 
to @expr{2^4}.  @xref{List Answer 1, 1}. (@bullet{})

You can also @dfn{reduce} a binary operator across a vector.
For example, reducing @samp{*} computes the product of all the

\afterdisplay
@end tex

@noindent
where 
@texline @math{\sum x}
@infoline @expr{sum(x)} 
represents the sum of all the values of @expr{x}.  While there is an
actual @code{sum} function in Calc, it's easier to sum a vector using a
simple reduction.  First, let's compute the four different sums that
this formula uses.

   r 5            r 8 r 3 *       -              r 7 /   t 9
@end group
@end smallexample

Let's ``plot'' this straight line approximation, 
@texline @math{y \approx m x + b},
@infoline @expr{m x + b}, 
and compare it with the original data.

@smallexample
@group

@xref{List Answer 7, 7}. (@bullet{})

@cindex Maximizing a function over a list of values
@c [fix-ref Numerical Solutions]
(@bullet{}) @strong{Exercise 8.}  Compute a list of values of Bessel's
@texline @math{J_1(x)}
@infoline @expr{J1} 
function @samp{besJ(1,x)} for @expr{x} from 0 to 5 in steps of 0.25.
Find the value of @expr{x} (from among the above set of values) for
which @samp{besJ(1,x)} is a maximum.  Use an ``automatic'' method,
i.e., just reading along the list by hand to find the largest value

of thing automatically; @pxref{Numerical Solutions}.)
@xref{List Answer 8, 8}. (@bullet{})

@cindex Digits, vectors of
(@bullet{}) @strong{Exercise 9.}  You are given an integer in the range
@texline @math{0 \le N < 10^m}
@infoline @expr{0 <= N < 10^m} 
for @expr{m=12} (i.e., an integer of less than
twelve digits).  Convert this integer into a vector of @expr{m}
digits, each in the range from 0 to 9.  In vector-of-digits notation,
add one to this integer to produce a vector of @expr{m+1} digits

@kbd{V R a =} to test if all numbers in a list were equal.  What
happened?  How would you do this test?  @xref{List Answer 10, 10}. (@bullet{})

(@bullet{}) @strong{Exercise 11.}  The area of a circle of radius one
is @cpi{}.  The area of the 
@texline @math{2\times2}
@infoline 2x2
square that encloses that circle is 4.  So if we throw @var{n} darts at
random points in the square, about @cpiover{4} of them will land inside
the circle.  This gives us an entertaining way to estimate the value of 
@cpi{}.  The @w{@kbd{k r}}
command picks a random number between zero and the value on the stack.
We could get a random floating-point number between @mathit{-1} and 1 by typing
@w{@kbd{2.0 k r 1 -}}.  Build a vector of 100 random @expr{(x,y)} points in
this square, then use vector mapping and reduction to count how many
points lie inside the unit circle.  Hint:  Use the @kbd{v b} command.
@xref{List Answer 11, 11}. (@bullet{})

(@bullet{}) @strong{Exercise 12.}  The @dfn{matchstick problem} provides
another way to calculate @cpi{}.  Say you have an infinite field
of vertical lines with a spacing of one inch.  Toss a one-inch matchstick
onto the field.  The probability that the matchstick will land crossing
a line turns out to be 
@texline @math{2/\pi}.
@infoline @expr{2/pi}.  
Toss 100 matchsticks to estimate @cpi{}.  (If you want still more fun,
the probability that the GCD (@w{@kbd{k g}}) of two large integers is
one turns out to be 
@texline @math{6/\pi^2}.
@infoline @expr{6/pi^2}.
That provides yet another way to estimate @cpi{}.)
@xref{List Answer 12, 12}. (@bullet{})

(@bullet{}) @strong{Exercise 13.}  An algebraic entry of a string in

commands do nested function evaluations.  @kbd{H V U} takes a starting
value and a number of steps @var{n} from the stack; it then applies the
function you give to the starting value 0, 1, 2, up to @var{n} times
and returns a vector of the results.  Use this command to create a
``random walk'' of 50 steps.  Start with the two-dimensional point
@expr{(0,0)}; then take one step a random distance between @mathit{-1} and 1
in both @expr{x} and @expr{y}; then take another step, and so on.  Use the
@kbd{g f} command to display this random walk.  Now modify your random
walk to walk a unit distance, but in a random direction, at each step.
(Hint:  The @code{sincos} function returns a vector of the cosine and
sine of an angle.)  @xref{List Answer 14, 14}. (@bullet{})

    9 n        Q             c p            2 *            Q
@end group
@end smallexample

@noindent
The square root of @mathit{-9} is by default rendered in rectangular form
(@w{@expr{0 + 3i}}), but we can convert it to polar form (3 with a
phase angle of 90 degrees).  All the usual arithmetic and scientific
operations are defined on both types of complex numbers.

Another generalized kind of number is @dfn{infinity}.  Infinity

@end group
@end smallexample

@noindent
Since infinity is infinitely large, multiplying it by any finite
number (like @mathit{-17}) has no effect, except that since @mathit{-17}
is negative, it changes a plus infinity to a minus infinity.
(``A huge positive number, multiplied by @mathit{-17}, yields a huge
negative number.'')  Adding any finite number to infinity also
leaves it unchanged.  Taking an absolute value gives us plus
infinity again.  Finally, we add this plus infinity to the minus
infinity we had earlier.  If you work it out, you might expect
the answer to be @mathit{-72} for this.  But the 72 has been completely
lost next to the infinities; by the time we compute @w{@samp{inf - inf}}
the finite difference between them, if any, is undetectable.
So we say the result is @dfn{indeterminate}, which Calc writes
with the symbol @code{nan} (for Not A Number).

original error estimates were valid standard deviations, there is about
a 60% chance that the result is correct within 0.59 degrees.

@cindex Torus, volume of
(@bullet{}) @strong{Exercise 7.}  The volume of a torus (a donut shape) is
@texline @math{2 \pi^2 R r^2}
@infoline @w{@expr{2 pi^2 R r^2}} 
where @expr{R} is the radius of the circle that
defines the center of the tube and @expr{r} is the radius of the tube
itself.  Suppose @expr{R} is 20 cm and @expr{r} is 4 cm, each known to
within 5 percent.  What is the volume and the relative uncertainty of

that arises in the second one.

@cindex Fermat, primality test of
(@bullet{}) @strong{Exercise 10.}  A theorem of Pierre de Fermat
says that 
@texline @w{@math{x^{n-1} \bmod n = 1}}
@infoline @expr{x^(n-1) mod n = 1}
if @expr{n} is a prime number and @expr{x} is an integer less than
@expr{n}.  If @expr{n} is @emph{not} a prime number, this will
@emph{not} be true for most values of @expr{x}.  Thus we can test
informally if a number is prime by trying this formula for several

@noindent
This calculation tells me it is six hours and 22 minutes until midnight.

(@bullet{}) @strong{Exercise 11.}  A rule of thumb is that one year
is about 
@texline @math{\pi \times 10^7}
@infoline @w{@expr{pi * 10^7}} 
seconds.  What time will it be that many seconds from right now?
@xref{Types Answer 11, 11}. (@bullet{})

(@bullet{}) @strong{Exercise 12.}  You are preparing to order packaging

@end group
@end smallexample

@noindent
Calc has invented the variable @samp{s1} to represent an unknown sign;
it is supposed to be either @mathit{+1} or @mathit{-1}.  Here we have used
the ``let'' command to evaluate the expression when the sign is negative.
If we plugged this into our second derivative we would get the same,
negative, answer, so @expr{x = -1.19023} is also a maximum.

To find the actual maximum value, we must plug our two values of @expr{x}

@end group
@end smallexample

(@bullet{}) @strong{Exercise 3.}  Find the integral from 1 to @expr{y}
of 
@texline @math{x \sin \pi x}
@infoline @w{@expr{x sin(pi x)}} 
(where the sine is calculated in radians).  Find the values of the
integral for integers @expr{y} from 1 to 5.  @xref{Algebra Answer 3,
3}. (@bullet{})

Calc's integrator can do many simple integrals symbolically, but many
others are beyond its capabilities.  Suppose we wish to find the area
under the curve 
@texline @math{\sin x \ln x}
@infoline @expr{sin(x) ln(x)} 
over the same range of @expr{x}.  If you entered this formula and typed
@kbd{a i x @key{RET}} (don't bother to try this), Calc would work for a
long time but would be unable to find a solution.  In fact, there is no
closed-form solution to this integral.  Now what do we do?

           + f(a+(n-2)h) + f(a+(n-1)h)) $$
\afterdisplay
@end tex

Compute the integral from 1 to 2 of 
@texline @math{\sin x \ln x}
@infoline @expr{sin(x) ln(x)} 
using Simpson's rule with 10 slices.  
@xref{Algebra Answer 4, 4}. (@bullet{})

Calc has a built-in @kbd{a I} command for doing numerical integration.

@ignore
@starindex
@end ignore
@tindex Si
(@bullet{}) @strong{Exercise 1.}  The ``sine integral'' function
@texline @math{{\rm Si}(x)}
@infoline @expr{Si(x)} 
is defined as the integral of @samp{sin(t)/t} for
@expr{t = 0} to @expr{x} in radians.  (It was invented because this
integral has no solution in terms of basic functions; if you give it
to Calc's @kbd{a i} command, it will ponder it for a long time and then

the following functions:

@enumerate
@item
Compute 
@texline @math{\displaystyle{\sin x \over x}},
@infoline @expr{sin(x) / x}, 
where @expr{x} is the number on the top of the stack.

@item
Compute the base-@expr{b} logarithm, just like the @kbd{B} key except

@cindex Golden ratio
@cindex Phi, golden ratio
A fascinating property of the Fibonacci numbers is that the @expr{n}th
Fibonacci number can be found directly by computing 
@texline @math{\phi^n / \sqrt{5}}
@infoline @expr{phi^n / sqrt(5)}
and then rounding to the nearest integer, where 
@texline @math{\phi} (``phi''),
@infoline @expr{phi}, 
the ``golden ratio,'' is 
@texline @math{(1 + \sqrt{5}) / 2}.
@infoline @expr{(1 + sqrt(5)) / 2}. 
(For convenience, this constant is available from the @code{phi}
variable, or the @kbd{I H P} command.)

@smallexample

@end smallexample

@cindex Continued fractions
(@bullet{}) @strong{Exercise 5.}  The @dfn{continued fraction}
representation of 
@texline @math{\phi}
@infoline @expr{phi} 
is 
@texline @math{1 + 1/(1 + 1/(1 + 1/( \ldots )))}.
@infoline @expr{1 + 1/(1 + 1/(1 + 1/( ...@: )))}.
We can compute an approximate value by carrying this however far
and then replacing the innermost 
@texline @math{1/( \ldots )}
@infoline @expr{1/( ...@: )} 
by 1.  Approximate
@texline @math{\phi}
@infoline @expr{phi} 
using a twenty-term continued fraction.
@xref{Programming Answer 5, 5}. (@bullet{})

(@bullet{}) @strong{Exercise 6.}  Linear recurrences like the one for

@cindex Bernoulli numbers, approximate
The @dfn{Bernoulli numbers} are a sequence with the interesting
property that all of the odd Bernoulli numbers are zero, and the
even ones, while difficult to compute, can be roughly approximated
by the formula 
@texline @math{\displaystyle{2 n! \over (2 \pi)^n}}.
@infoline @expr{2 n!@: / (2 pi)^n}.  
Let's write a keyboard macro to compute (approximate) Bernoulli numbers.
(Calc has a command, @kbd{k b}, to compute exact Bernoulli numbers, but
this command is very slow for large @expr{n} since the higher Bernoulli
numbers are very large fractions.)

@end tex

@noindent
where @expr{f'(x)} is the derivative of @expr{f}.  The @expr{x}
values will quickly converge to a solution, i.e., eventually
@texline @math{x_{\rm new}}
@infoline @expr{new_x} 
and @expr{x} will be equal to within the limits
of the current precision.  Write a program which takes a formula
involving the variable @expr{x}, and an initial guess @expr{x_0},
on the stack, and produces a value of @expr{x} for which the formula
is zero.  Use it to find a solution of 
@texline @math{\sin(\cos x) = 0.5}
@infoline @expr{sin(cos(x)) = 0.5}
near @expr{x = 4.5}.  (Use angles measured in radians.)  Note that
the built-in @w{@kbd{a R}} (@code{calc-find-root}) command uses Newton's
method when it is able.  @xref{Programming Answer 8, 8}. (@bullet{})

@cindex Digamma function
@cindex Gamma constant, Euler's
@cindex Euler's gamma constant
(@bullet{}) @strong{Exercise 9.}  The @dfn{digamma} function 
@texline @math{\psi(z) (``psi'')}
@infoline @expr{psi(z)}
is defined as the derivative of 
@texline @math{\ln \Gamma(z)}.
@infoline @expr{ln(gamma(z))}.  
For large values of @expr{z}, it can be approximated by the infinite sum

@ifinfo
@example

\afterdisplay
@end tex

@noindent
where 
@texline @math{\sum}
@infoline @expr{sum} 
represents the sum over @expr{n} from 1 to infinity
(or to some limit high enough to give the desired accuracy), and
the @code{bern} function produces (exact) Bernoulli numbers.
While this sum is not guaranteed to converge, in practice it is safe.
An interesting mathematical constant is Euler's gamma, which is equal
to about 0.5772.  One way to compute it is by the formula,
@texline @math{\gamma = -\psi(1)}.
@infoline @expr{gamma = -psi(1)}.  
Unfortunately, 1 isn't a large enough argument
for the above formula to work (5 is a much safer value for @expr{z}).
Fortunately, we can compute 
@texline @math{\psi(1)}
@infoline @expr{psi(1)} 
from 
@texline @math{\psi(5)}
@infoline @expr{psi(5)} 
using the recurrence 
@texline @math{\psi(z+1) = \psi(z) + {1 \over z}}.
@infoline @expr{psi(z+1) = psi(z) + 1/z}.  
Your task:  Develop a program to compute 
@texline @math{\psi(z)};
@infoline @expr{psi(z)}; 
it should ``pump up'' @expr{z}
if necessary to be greater than 5, then use the above summation
formula.  Use looping commands to compute the sum.  Use your function
to compute 
@texline @math{\gamma}
@infoline @expr{gamma} 
to twelve decimal places.  (Calc has a built-in command
for Euler's constant, @kbd{I P}, which you can use to check your answer.)
@xref{Programming Answer 9, 9}. (@bullet{})

@noindent
@kbd{1 @key{RET} 2 @key{RET} 3 @key{RET} 4 + * -}

The result is 
@texline @math{1 - (2 \times (3 + 4)) = -13}.
@infoline @expr{1 - (2 * (3 + 4)) = -13}.

@node RPN Answer 2, RPN Answer 3, RPN Answer 1, Answers to Exercises
@subsection RPN Tutorial Exercise 2

@noindent
@texline @math{2\times4 + 7\times9.5 + {5\over4} = 75.75}
@infoline @expr{2*4 + 7*9.5 + 5/4 = 75.75}

After computing the intermediate term 
@texline @math{2\times4 = 8},
@infoline @expr{2*4 = 8}, 
you can leave that result on the stack while you compute the second
term.  With both of these results waiting on the stack you can then
compute the final term, then press @kbd{+ +} to add everything up.

Dividing two integers that are larger than the current precision may
give a floating-point result that is inaccurate even when rounded
down to an integer.  Consider @expr{123456789 / 2} when the current
precision is 6 digits.  The true answer is @expr{61728394.5}, but
with a precision of 6 this will be rounded to 
@texline @math{12345700.0/2.0 = 61728500.0}.
@infoline @expr{12345700.@: / 2.@: = 61728500.}.
The result, when converted to an integer, will be off by 106.

Here are two solutions:  Raise the precision enough that the
floating-point round-off error is strictly to the right of the

@node Matrix Answer 3, List Answer 1, Matrix Answer 2, Answers to Exercises
@subsection Matrix Tutorial Exercise 3

@noindent
To solve 
@texline @math{A^T A \, X = A^T B},
@infoline @expr{trn(A) * A * X = trn(A) * B}, 
first we compute
@texline @math{A' = A^T A}
@infoline @expr{A2 = trn(A) * A} 
and 
@texline @math{B' = A^T B};
@infoline @expr{B2 = trn(A) * B}; 
now, we have a system 
@texline @math{A' X = B'}
@infoline @expr{A2 * X = B2} 
which we can solve using Calc's @samp{/} command.

@ifinfo
@example

\afterdisplayh
@end tex

The first step is to enter the coefficient matrix.  We'll store it in
quick variable number 7 for later reference.  Next, we compute the
@texline @math{B'}
@infoline @expr{B2} 
vector.

@smallexample
@group

@end group
@end smallexample

@noindent
Now we compute the matrix 
@texline @math{A'}
@infoline @expr{A2} 
and divide.

@smallexample
@group

@noindent
(The actual computed answer will be slightly inexact due to
round-off error.)

Notice that the answers are similar to those for the 
@texline @math{3\times3}
@infoline 3x3
system solved in the text.  That's because the fourth equation that was 
added to the system is almost identical to the first one multiplied
by two.  (If it were identical, we would have gotten the exact same
answer since the 
@texline @math{4\times3}
@infoline 4x3
system would be equivalent to the original 
@texline @math{3\times3}
@infoline 3x3
system.)

Since the first and fourth equations aren't quite equivalent, they
can't both be satisfied at once.  Let's plug our answers back into

$$ m \times x + b \times 1 = y $$
\afterdisplay
@end tex

Thus we want a 
@texline @math{19\times2}
@infoline 19x2
matrix with our @expr{x} vector as one column and
ones as the other column.  So, first we build the column of ones, then
we combine the two columns to form our @expr{A} matrix.

@end group
@end smallexample

@noindent
Now we compute 
@texline @math{A^T y}
@infoline @expr{trn(A) * y} 
and 
@texline @math{A^T A}
@infoline @expr{trn(A) * A} 
and divide.

@smallexample
@group

    /
@end group
@end smallexample

Since we were solving equations of the form 
@texline @math{m \times x + b \times 1 = y},
@infoline @expr{m*x + b*1 = y}, 
these numbers should be @expr{m} and @expr{b}, respectively.  Sure
enough, they agree exactly with the result computed using @kbd{V M} and
@kbd{V R}!

@node List Answer 4, List Answer 5, List Answer 3, Answers to Exercises
@subsection List Tutorial Exercise 4

@noindent
A number @expr{j} is a divisor of @expr{n} if 
@texline @math{n \mathbin{\hbox{\code{\%}}} j = 0}.
@infoline @samp{n % j = 0}.  
The first step is to get a vector that identifies the divisors.

@smallexample
@group

so that the mapping operation works; no prime factor will ever be
zero, so adding zeros on the left and right is safe.  From then on
the job is pretty straightforward.

Incidentally, Calc provides the 
@texline @dfn{M@"obius} @math{\mu}
@infoline @dfn{Moebius mu} 
function which is zero if and only if its argument is square-free.  It
would be a much more convenient way to do the above test in practice.

@node List Answer 6, List Answer 7, List Answer 5, Answers to Exercises

The numbers down the lefthand edge of the list we desire are called
the ``triangular numbers'' (now you know why!).  The @expr{n}th
triangular number is the sum of the integers from 1 to @expr{n}, and
can be computed directly by the formula 
@texline @math{n (n+1) \over 2}.
@infoline @expr{n * (n+1) / 2}.

@smallexample
@group
2:  [ [0], [0, 1], ... ]    2:  [ [0], [0, 1], ... ]

@end smallexample

@noindent
It's a good idea to verify, as in the last step above, that only
one value is equal to the maximum.  (After all, a plot of 
@texline @math{\sin x}
@infoline @expr{sin(x)}
might have many points all equal to the maximum value, 1.)

The vector we have now has a single 1 in the position that indicates
the maximum value of @expr{x}.  Now it is a simple matter to convert

@noindent
This problem can be made a lot easier by taking advantage of some
symmetries.  First of all, after some thought it's clear that the
@expr{y} axis can be ignored altogether.  Just pick a random @expr{x}
component for one end of the match, pick a random direction 
@texline @math{\theta},
@infoline @expr{theta},
and see if @expr{x} and 
@texline @math{x + \cos \theta}
@infoline @expr{x + cos(theta)} 
(which is the @expr{x} coordinate of the other endpoint) cross a line.
The lines are at integer coordinates, so this happens when the two
numbers surround an integer.

In fact, since the field of lines is infinite we can choose the
coordinates 0 and 1 for the lines on either side of the leftmost
endpoint.  The rightmost endpoint will be between 0 and 1 if the
match does not cross a line, or between 1 and 2 if it does.  So:
Pick random @expr{x} and 
@texline @math{\theta},
@infoline @expr{theta}, 
compute
@texline @math{x + \cos \theta},
@infoline @expr{x + cos(theta)},
and count how many of the results are greater than one.  Simple!

We can make this go a bit faster by using the @kbd{v .} and @kbd{t .}
commands.

@end smallexample

@noindent
Aha!  It's unlikely that an irrational number would equal a fraction
this simple to within ten digits, so our original number was probably
@texline @math{\sqrt{27 \pi / 53}}.
@infoline @expr{sqrt(27 pi / 53)}.

Notice that we didn't need to re-round the number when we reduced the
precision.  Remember, arithmetic operations always round their inputs
to the current precision before they begin.

as a possible value.

The fourth calculation, @samp{1 / (-10 .. 10)}, has the same problem.
Zero is buried inside the interval, but it's still a possible value.
It's not hard to see that the actual result of @samp{1 / (-10 .. 10)}
will be either greater than @mathit{0.1}, or less than @mathit{-0.1}.  Thus
the interval goes from minus infinity to plus infinity, with a ``hole''
in it from @mathit{-0.1} to @mathit{0.1}.  Calc doesn't have any way to
represent this, so it just reports @samp{[-inf .. inf]} as the answer.
It may be disappointing to hear ``the answer lies somewhere between
minus infinity and plus infinity, inclusive,'' but that's the best
that interval arithmetic can do in this case.

    [ 3 n .. 3 ]        @key{RET} 2 ^           @key{TAB} @key{RET} *
@end group
@end smallexample

@noindent
In the first case the result says, ``if a number is between @mathit{-3} and
3, its square is between 0 and 9.''  The second case says, ``the product
of two numbers each between @mathit{-3} and 3 is between @mathit{-9} and 9.''

An interval form is not a number; it is a symbol that can stand for
many different numbers.  Two identical-looking interval forms can stand
for different numbers.

Each of these functions can be computed using the stack, or using
algebraic entry, whichever way you prefer:

@noindent
Computing 
@texline @math{\displaystyle{\sin x \over x}}:
@infoline @expr{sin(x) / x}:

Using the stack:  @kbd{C-x (  @key{RET} S @key{TAB} /  C-x )}.

Using algebraic entry:  @kbd{C-x (  ' sin($)/$ @key{RET}  C-x )}.

@end example

@noindent
This program is quite efficient because Calc knows how to raise a
matrix (or other value) to the power @expr{n} in only 
@texline @math{\log_2 n}
@infoline @expr{log(n,2)}
steps.  For example, this program can compute the 1000th Fibonacci
number (a 209-digit integer!) in about 10 steps; even though the
@kbd{Z < ... Z >} solution had much simpler steps, it would have
required so many steps that it would not have been practical.

@subsection Programming Tutorial Exercise 8

@noindent
The first step is to compute the derivative @expr{f'(x)} and thus
the formula 
@texline @math{\displaystyle{x - {f(x) \over f'(x)}}}.
@infoline @expr{x - f(x)/f'(x)}.

(Because this definition is long, it will be repeated in concise form
below.  You can use @w{@kbd{M-# m}} to load it from there.  While you are
entering a @kbd{Z ` Z '} body in a macro, Calc simply collects

@noindent
The first step is to adjust @expr{z} to be greater than 5.  A simple
``for'' loop will do the job here.  If @expr{z} is less than 5, we
reduce the problem using 
@texline @math{\psi(z) = \psi(z+1) - 1/z}.
@infoline @expr{psi(z) = psi(z+1) - 1/z}.  We go
on to compute 
@texline @math{\psi(z+1)},
@infoline @expr{psi(z+1)}, 
and remember to add back a factor of @expr{-1/z} when we're done.  This
step is repeated until @expr{z > 5}.

(Because this definition is long, it will be repeated in concise form

  @key{RET} 5        a <    Z [  5 Z (  & s + 2  1 s + 1  1 Z ) r 1  Z ]
@end group
@end smallexample

Now we compute the initial part of the sum:  
@texline @math{\ln z - {1 \over 2z}}
@infoline @expr{ln(z) - 1/2z}
minus the adjustment factor.

@smallexample
@group

  @key{TAB} @key{RET} M-@key{TAB}       - @key{RET} M-@key{TAB}      a =     Z /    2  Z )  Z ' C-x )
@end group
@end smallexample

This is the value of 
@texline @math{-\gamma},
@infoline @expr{- gamma}, 
with a slight bit of roundoff error.  To get a full 12 digits, let's use
a higher precision:

@smallexample

@subsection Programming Tutorial Exercise 10

@noindent
Taking the derivative of a term of the form @expr{x^n} will produce
a term like 
@texline @math{n x^{n-1}}.
@infoline @expr{n x^(n-1)}.  
Taking the derivative of a constant
produces zero.  From this it is easy to see that the @expr{n}th
derivative of a polynomial, evaluated at @expr{x = 0}, will equal the
coefficient on the @expr{x^n} term times @expr{n!}.

(@key{RET} corresponds to the @key{ENTER} key on most calculators.)
The first three keystrokes ``push'' the numbers 1 and 2 onto the stack.  The
@kbd{+} key always ``pops'' the top two numbers from the stack, adds them,
and pushes the result (3) back onto the stack.  This number is ready for
further calculations:  @kbd{5 -} pushes 5 onto the stack, then pops the
3 and 5, subtracts them, and pushes the result (@mathit{-2}).

Note that the ``top'' of the stack actually appears at the @emph{bottom}
of the buffer.  A line containing a single @samp{.} character signifies
the end of the buffer; Calculator commands operate on the number(s)
directly above this line.  The @kbd{d t} (@code{calc-truncate-stack})

@kbd{-}, @kbd{n}, and @kbd{_} (the underscore).  The @kbd{-} key subtracts
the two numbers on the top of the stack.  The @kbd{n} key changes the sign
of the number on the top of the stack or the number currently being entered.
The @kbd{_} key begins entry of a negative number or changes the sign of
the number currently being entered.  The following sequences all enter the
number @mathit{-5} onto the stack:  @kbd{0 @key{RET} 5 -}, @kbd{5 n @key{RET}},
@kbd{5 @key{RET} n}, @kbd{_ 5 @key{RET}}, @kbd{5 _ @key{RET}}.

Some other keys are active during numeric entry, such as @kbd{#} for
non-decimal numbers, @kbd{:} for fractions, and @kbd{@@} for HMS forms.
These notations are described later in this manual with the corresponding

@cindex Algebraic notation
@cindex Formulas, entering
Calculations can also be entered in algebraic form.  This is accomplished
by typing the apostrophe key, @kbd{'}, followed by the expression in
standard format:  @kbd{@key{'} 2+3*4 @key{RET}} computes
@texline @math{2+(3\times4) = 14}
@infoline @expr{2+(3*4) = 14} 
and pushes that on the stack.  If you wish you can
ignore the RPN aspect of Calc altogether and simply enter algebraic
expressions in this way.  You may want to use @key{DEL} every so often to
clear previous results off the stack.

is greater than this, it will recompute @cpi{} using a series
approximation.  This value will not need to be recomputed ever again
unless you raise the precision still further.  Many operations such as
logarithms and sines make use of similarly cached values such as
@cpiover{4} and 
@texline @math{\ln 2}.
@infoline @expr{ln(2)}.  
The visible effect of caching is that
high-precision computations may seem to do extra work the first time.
Other things cached include powers of two (for the binary arithmetic
functions), matrix inverses and determinants, symbolic integrals, and

@cindex Floating-point numbers
A floating-point number or @dfn{float} is a number stored in scientific
notation.  The number of significant digits in the fractional part is
governed by the current floating precision (@pxref{Precision}).  The
range of acceptable values is from 
@texline @math{10^{-3999999}}
@infoline @expr{10^-3999999} 
(inclusive) to 
@texline @math{10^{4000000}}
@infoline @expr{10^4000000}
(exclusive), plus the corresponding negative values and zero.

Calculations that would exceed the allowable range of values (such
as @samp{exp(exp(20))}) are left in symbolic form by Calc.  The

@var{imag} is the imaginary part, each of which may be any real number.
Rectangular complex numbers can also be displayed in @samp{@var{a}+@var{b}i}
notation; @pxref{Complex Formats}.

Polar complex numbers are displayed in the form 
@texline `@t{(}@var{r}@t{;}@math{\theta}@t{)}'
@infoline `@t{(}@var{r}@t{;}@var{theta}@t{)}'
where @var{r} is the nonnegative magnitude and 
@texline @math{\theta}
@infoline @var{theta} 
is the argument or phase angle.  The range of 
@texline @math{\theta}
@infoline @var{theta} 
depends on the current angular mode (@pxref{Angular Modes}); it is
generally between @mathit{-180} and @mathit{+180} degrees or the equivalent range
in radians. 

Complex numbers are entered in stages using incomplete objects.
@xref{Incomplete Objects}.

really mean is that @expr{1 / x}, as @expr{x} becomes larger and
larger, becomes arbitrarily close to zero.  So you can imagine
that if @expr{x} got ``all the way to infinity,'' then @expr{1 / x}
would go all the way to zero.  Similarly, when they say that
@samp{exp(inf) = inf}, they mean that 
@texline @math{e^x}
@infoline @expr{exp(x)} 
grows without bound as @expr{x} grows.  The symbol @samp{-inf} likewise
stands for an infinitely negative real value; for example, we say that
@samp{exp(-inf) = 0}.  You can have an infinity pointing in any
direction on the complex plane:  @samp{sqrt(-inf) = i inf}.

@starindex
@end ignore
@tindex vec
Algebraic functions for building vectors include @samp{vec(a, b, c)}
to build @samp{[a, b, c]}, @samp{cvec(a, n, m)} to build an 
@texline @math{n\times m}
@infoline @var{n}x@var{m}
matrix of @samp{a}s, and @samp{index(n)} to build a vector of integers
from 1 to @samp{n}.

@node Strings, HMS Forms, Vectors and Matrices, Data Types

The @var{hours} value is an integer (or integer-valued float).
The @var{mins} value is an integer or integer-valued float between 0 and 59.
The @var{secs} value is a real number between 0 (inclusive) and 60
(exclusive).  A positive HMS form is interpreted as @var{hours} +
@var{mins}/60 + @var{secs}/3600.  A negative HMS form is interpreted
as @mathit{- @var{hours}} @mathit{-} @var{mins}/60 @mathit{-} @var{secs}/3600.
Display format for HMS forms is quite flexible.  @xref{HMS Formats}.

HMS forms can be added and subtracted.  When they are added to numbers,
the numbers are interpreted according to the current angular mode.  HMS
forms can also be multiplied and divided by real numbers.  Dividing

between, say, @samp{<12:00am Mon Jan 1, 1900>} and
@samp{<12:00am Sat Jan 1, 2000>}.)

Calc uses the Julian calendar for all dates before the year 1752,
including dates BC when the Julian calendar technically had not
yet been invented.  Thus the claim that day number @mathit{-10000} is
called ``August 16, 28 BC'' should be taken with a grain of salt.

Please note that there is no ``year 0''; the day before
@samp{<Sat Jan 1, +1>} is @samp{<Fri Dec 31, -1>}.  These are
days 0 and @mathit{-1} respectively in Calc's internal numbering scheme.

@cindex Julian day counting
Another day counting system in common use is, confusingly, also
called ``Julian.''  It was invented in 1583 by Joseph Justus
Scaliger, who named it in honor of his father Julius Caesar
Scaliger.  For obscure reasons he chose to start his day
numbering on Jan 1, 4713 BC at noon, which in Calc's scheme
is @mathit{-1721423.5} (recall that Calc starts at midnight instead
of noon).  Thus to convert a Calc date code obtained by
unpacking a date form into a Julian day number, simply add
1721423.5.  The Julian code for @samp{6:00am Jan 9, 1991}
is 2448265.75.  The built-in @kbd{t J} command performs
this conversion for you.

A @dfn{modulo form} is a real number which is taken modulo (i.e., within
an integer multiple of) some value @var{M}.  Arithmetic modulo @var{M}
often arises in number theory.  Modulo forms are written
`@var{a} @t{mod} @var{M}',
where @var{a} and @var{M} are real numbers or HMS forms, and
@texline @math{0 \le a < M}.
@infoline @expr{0 <= a < @var{M}}.
In many applications @expr{a} and @expr{M} will be
integers but this is not required.

Modulo forms are not to be confused with the modulo operator @samp{%}.

@expr{M} is non-prime), or if any of the arguments are non-integers, the
division is left in symbolic form.  Other operations, such as square
roots, are not yet supported for modulo forms.  (Note that, although
@w{`@t{(}@var{a} @t{mod} @var{M}@t{)^.5}'} will compute a ``modulo square root''
in the sense of reducing 
@texline @math{\sqrt a}
@infoline @expr{sqrt(a)} 
modulo @expr{M}, this is not a useful definition from the
number-theoretical point of view.)

@ignore

@noindent
@cindex Error forms
@cindex Standard deviations
An @dfn{error form} is a number with an associated standard
deviation, as in @samp{2.3 +/- 0.12}.  The notation
@texline `@var{x} @t{+/-} @math{\sigma}' 
@infoline `@var{x} @t{+/-} sigma' 
stands for an uncertain value which follows
a normal or Gaussian distribution of mean @expr{x} and standard
deviation or ``error'' 
@texline @math{\sigma}.
@infoline @expr{sigma}.
Both the mean and the error can be either numbers or
formulas.  Generally these are real numbers but the mean may also be
complex.  If the error is negative or complex, it is changed to its
absolute value.  An error form with zero error is converted to a
regular number by the Calculator.

All arithmetic and transcendental functions accept error forms as input.
Operations on the mean-value part work just like operations on regular
numbers.  The error part for any function @expr{f(x)} (such as 
@texline @math{\sin x}
@infoline @expr{sin(x)})
is defined by the error of @expr{x} times the derivative of @expr{f}
evaluated at the mean value of @expr{x}.  For a two-argument function
@expr{f(x,y)} (such as addition) the error is the square root of the sum
of the squares of the errors due to @expr{x} and @expr{y}.

Consult a good text on error analysis for a discussion of the proper use
of standard deviations.  Actual errors often are neither Gaussian-distributed
nor uncorrelated, and the above formulas are valid only when errors
are small.  As an example, the error arising from
@texline `@t{sin(}@var{x} @t{+/-} @math{\sigma}@t{)}' 
@infoline `@t{sin(}@var{x} @t{+/-} @var{sigma}@t{)}' 
is 
@texline `@math{\sigma} @t{abs(cos(}@var{x}@t{))}'.  
@infoline `@var{sigma} @t{abs(cos(}@var{x}@t{))}'.  
When @expr{x} is close to zero,
@texline @math{\cos x}
@infoline @expr{cos(x)} 
is close to one so the error in the sine is close to 
@texline @math{\sigma};
@infoline @expr{sigma};
this makes sense, since 
@texline @math{\sin x}
@infoline @expr{sin(x)} 
is approximately @expr{x} near zero, so a given error in @expr{x} will
produce about the same error in the sine.  Likewise, near 90 degrees
@texline @math{\cos x}
@infoline @expr{cos(x)} 
is nearly zero and so the computed error is
small:  The sine curve is nearly flat in that region, so an error in @expr{x}
has relatively little effect on the value of 
@texline @math{\sin x}.
@infoline @expr{sin(x)}.  
However, consider @samp{sin(90 +/- 1000)}.  The cosine of 90 is zero, so
Calc will report zero error!  We get an obviously wrong result because
we have violated the small-error approximation underlying the error
analysis.  If the error in @expr{x} had been small, the error in
@texline @math{\sin x}
@infoline @expr{sin(x)} 
would indeed have been negligible.

@ignore
@mindex p

@samp{1 / (-2 .. 2) = [-inf .. inf]}.

While it may seem that intervals and error forms are similar, they are
based on entirely different concepts of inexact quantities.  An error
form 
@texline `@var{x} @t{+/-} @math{\sigma}' 
@infoline `@var{x} @t{+/-} @var{sigma}' 
means a variable is random, and its value could
be anything but is ``probably'' within one 
@texline @math{\sigma} 
@infoline @var{sigma} 
of the mean value @expr{x}. An interval 
`@t{[}@var{a} @t{..@:} @var{b}@t{]}' means a
variable's value is unknown, but guaranteed to lie in the specified
range.  Error forms are statistical or ``average case'' approximations;

@samp{=>} [@code{evalto}].

Note that, unlike in usual computer notation, multiplication binds more
strongly than division:  @samp{a*b/c*d} is equivalent to 
@texline @math{a b \over c d}.
@infoline @expr{(a*b)/(c*d)}.

@cindex Multiplication, implicit
@cindex Implicit multiplication
The multiplication sign @samp{*} may be omitted in many cases.  In particular,

element down to level @var{n}, making room for it by pulling all the
intervening stack elements toward the top.  @kbd{M-@key{TAB}} moves the
element at level @var{n} up to the top.  (Compare with @key{LFD},
which copies instead of moving the element in level @var{n}.)

With a negative argument @mathit{-@var{n}}, @key{TAB} rotates the stack
to move the object in level @var{n} to the deepest place in the
stack, and the object in level @mathit{@var{n}+1} to the top.  @kbd{M-@key{TAB}}
rotates the deepest stack element to be in level @mathit{n}, also
putting the top stack element in level @mathit{@var{n}+1}.

@xref{Selecting Subformulas}, for a way to apply these commands to
any portion of a vector or formula on the stack.

@node Editing Stack Entries, Trail Commands, Stack Manipulation, Stack and Trail

If the file name you give contains the string @samp{.emacs} anywhere
inside it, @kbd{m F} will not automatically load the new file.  This
is because you are presumably switching to your @file{~/.emacs} file,
which may contain other things you don't want to reread.  You can give
a numeric prefix argument of 1 to @kbd{m F} to force it to read the
file no matter what its name.  Conversely, an argument of @mathit{-1} tells
@kbd{m F} @emph{not} to read the new file.  An argument of 2 or @mathit{-2}
tells @kbd{m F} not to reset the modes to their defaults beforehand,
which is useful if you intend your new file to have a variant of the
modes present in the file you were using before.

@kindex m x

Likewise, the @kbd{H} key (@code{calc-hyperbolic}) sets or clears the
Hyperbolic Flag, which transforms @code{calc-sin} into @code{calc-sinh}.
If both of these flags are set at once, the effect will be
@code{calc-arcsinh}.  (The Hyperbolic flag is also used by some
non-trigonometric commands; for example @kbd{H L} computes a base-10,
instead of base-@mathit{e}, logarithm.)

Command names like @code{calc-arcsin} are provided for completeness, and
may be executed with @kbd{x} or @kbd{M-x}.  Their effect is simply to
toggle the Inverse and/or Hyperbolic flags and then execute the
corresponding base command (@code{calc-sin} in this case).

The @kbd{m i} command with a numeric prefix argument of zero,
i.e., @kbd{C-u 0 m i}, turns on a ``positive infinite mode'' in
which zero is treated as positive instead of being directionless.
Thus, @samp{1 / 0 = inf} and @samp{-1 / 0 = -inf} in this mode.
Note that zero never actually has a sign in Calc; there are no
separate representations for @mathit{+0} and @mathit{-0}.  Positive
infinite mode merely changes the interpretation given to the
single symbol, @samp{0}.  One consequence of this is that, while
you might expect @samp{1 / -0 = -inf}, actually @samp{1 / -0}
is equivalent to @samp{1 / 0}, which is equal to positive @code{inf}.

@end table

Calc uses this information to determine when certain simplifications
of formulas are safe.  For example, @samp{(x^y)^z} cannot be
simplified to @samp{x^(y z)} in general; for example,
@samp{((-3)^2)^1:2} is 3, but @samp{(-3)^(2*1:2) = (-3)^1} is @mathit{-3}.
However, this simplification @emph{is} safe if @code{z} is known
to be an integer, or if @code{x} is known to be a nonnegative
real number.  If you have given declarations that allow Calc to
deduce either of these facts, Calc will perform this simplification
of the formula.

@item const
The value is a constant with respect to other variables.
@end table

Calc does not check the declarations for a variable when you store
a value in it.  However, storing @mathit{-3.5} in a variable that has
been declared @code{pos}, @code{int}, or @code{matrix} may have
unexpected effects; Calc may evaluate @samp{sqrt(x^2)} to @expr{3.5}
if it substitutes the value first, or to @expr{-3.5} if @code{x}
was declared @code{pos} and the formula @samp{sqrt(x^2)} is
simplified to @samp{x} before the value is substituted.  Before

represent the integer and no more.  The @kbd{d z} (@code{calc-leading-zeros})
command causes integers to be padded out with leading zeros according to the
current binary word size.  (@xref{Binary Functions}, for a discussion of
word size.)  If the absolute value of the word size is @expr{w}, all integers
are displayed with at least enough digits to represent 
@texline @math{2^w-1}
@infoline @expr{(2^w)-1} 
in the current radix.  (Larger integers will still be displayed in their
entirety.) 

@node Grouping Digits, Float Formats, Radix Modes, Display Modes

by the arguments in parentheses.  However, functions for which @TeX{} has
special names (like @code{\sin}) will use curly braces instead of
parentheses for very simple arguments.  During input, curly braces and
parentheses work equally well for grouping, but when the document is
formatted the curly braces will be invisible.  Thus the printed result is
@texline @math{\sin{2 x}}
@infoline @expr{sin 2x} 
but 
@texline @math{\sin(2 + x)}.
@infoline @expr{sin(2 + x)}.

Function and variable names not treated specially by @TeX{} are simply
written out as-is, which will cause them to come out in italic letters
in the printed document.  If you invoke @kbd{d T} with a positive numeric

@item
Polar mode.  Value is 0 (rectangular) or 1 (polar); default is 0.
Command is @kbd{m p}.

@item
Matrix/scalar mode.  Default value is @mathit{-1}.  Value is 0 for scalar
mode, @mathit{-2} for matrix mode, or @var{N} for 
@texline @math{N\times N}
@infoline @var{N}x@var{N} 
matrix mode.  Command is @kbd{m v}.

@item
Simplification mode.  Default is 1.  Value is @mathit{-1} for off (@kbd{m O}),
0 for @kbd{m N}, 2 for @kbd{m B}, 3 for @kbd{m A}, 4 for @kbd{m E},
or 5 for @w{@kbd{m U}}.  The @kbd{m D} command accepts these prefixes.

@item
Infinite mode.  Default is @mathit{-1} (off).  Value is 1 if the mode is on,
or 0 if the mode is on with positive zeros.  Command is @kbd{m i}.
@end enumerate

For example, the sequence @kbd{M-1 m g @key{RET} 2 + ~ p} increases the
precision by two, leaving a copy of the old precision on the stack.

@mindex @null
@end ignore
@tindex -
The @kbd{-} (@code{calc-minus}) command subtracts two values.  The top
number on the stack is subtracted from the one behind it, so that the
computation @kbd{5 @key{RET} 2 -} produces 3, not @mathit{-3}.  All options
available for @kbd{+} are available for @kbd{-} as well.

@kindex *
@pindex calc-times
@ignore

@kindex f s
@pindex calc-sign
@tindex sign
The @kbd{f s} (@code{calc-sign}) [@code{sign}] command returns 1 if its
argument is positive, @mathit{-1} if its argument is negative, or 0 if its
argument is zero.  In algebraic form, you can also write @samp{sign(a,x)}
which evaluates to @samp{x * sign(a)}, i.e., either @samp{x}, @samp{-x}, or
zero depending on the sign of @samp{a}.

@kindex &

@tindex xpon
The @kbd{f M} (@code{calc-mant-part}) [@code{mant}] function extracts
the ``mantissa'' part @expr{m} of its floating-point argument; @kbd{f X}
(@code{calc-xpon-part}) [@code{xpon}] extracts the ``exponent'' part
@expr{e}.  The original number is equal to 
@texline @math{m \times 10^e},
@infoline @expr{m * 10^e},
where @expr{m} is in the interval @samp{[1.0 ..@: 10.0)} except that
@expr{m=e=0} if the original number is zero.  For integers
and fractions, @code{mant} returns the number unchanged and @code{xpon}
returns zero.  The @kbd{v u} (@code{calc-unpack}) command can also be

floating-point numbers, the change is by one unit in the last place.
For example, incrementing @samp{12.3456} when the current precision
is 6 digits yields @samp{12.3457}.  If the current precision had been
8 digits, the result would have been @samp{12.345601}.  Incrementing
@samp{0.0} produces 
@texline @math{10^{-p}},
@infoline @expr{10^-p}, 
where @expr{p} is the current
precision.  These operations are defined only on integers and floats.
With numeric prefix arguments, they change the number by @expr{n} units.

@end ignore
@kindex H F
The @kbd{F} (@code{calc-floor}) [@code{floor} or @code{ffloor}] command
truncates a real number to the next lower integer, i.e., toward minus
infinity.  Thus @kbd{3.6 F} produces 3, but @kbd{_3.6 F} produces
@mathit{-4}.

@kindex I F
@pindex calc-ceiling
@tindex ceil
@tindex fceil

@mindex @null
@end ignore
@kindex H I F
The @kbd{I F} (@code{calc-ceiling}) [@code{ceil} or @code{fceil}]
command truncates toward positive infinity.  Thus @kbd{3.6 I F} produces
4, and @kbd{_3.6 I F} produces @mathit{-3}.

@kindex R
@pindex calc-round
@tindex round
@tindex fround

@kindex H R
The @kbd{R} (@code{calc-round}) [@code{round} or @code{fround}] command
rounds to the nearest integer.  When the fractional part is .5 exactly,
this command rounds away from zero.  (All other rounding in the
Calculator uses this convention as well.)  Thus @kbd{3.5 R} produces 4
but @kbd{3.4 R} produces 3; @kbd{_3.5 R} produces @mathit{-4}.

@kindex I R
@pindex calc-trunc
@tindex trunc
@tindex ftrunc

@end ignore
@kindex H I R
The @kbd{I R} (@code{calc-trunc}) [@code{trunc} or @code{ftrunc}]
command truncates toward zero.  In other words, it ``chops off''
everything after the decimal point.  Thus @kbd{3.6 I R} produces 3 and
@kbd{_3.6 I R} produces @mathit{-3}.

These functions may not be applied meaningfully to error forms, but they
do work for intervals.  As a convenience, applying @code{floor} to a
modulo form floors the value part of the form.  Applied to a vector,
these functions operate on all elements of the vector one by one.

@pindex calc-argument
@tindex arg
The @kbd{G} (@code{calc-argument}) [@code{arg}] command computes the
``argument'' or polar angle of a complex number.  For a number in polar
notation, this is simply the second component of the pair
@texline `@t{(}@var{r}@t{;}@math{\theta}@t{)}'.
@infoline `@t{(}@var{r}@t{;}@var{theta}@t{)}'.
The result is expressed according to the current angular mode and will
be in the range @mathit{-180} degrees (exclusive) to @mathit{+180} degrees
(inclusive), or the equivalent range in radians.

@pindex calc-imaginary
The @code{calc-imaginary} command multiplies the number on the
top of the stack by the imaginary number @expr{i = (0,1)}.  This

@end ignore
@kindex v p (complex)
@pindex calc-pack
The @kbd{v p} (@code{calc-pack}) command can pack the top two numbers on
the stack into a composite object such as a complex number.  With
a prefix argument of @mathit{-1}, it produces a rectangular complex number;
with an argument of @mathit{-2}, it produces a polar complex number.
(Also, @pxref{Building Vectors}.)

@ignore
@mindex v u
@end ignore

@pindex calc-clean
@tindex pclean
The @kbd{c c} (@code{calc-clean}) [@code{pclean}] command ``cleans'' the
number on the top of the stack.  Floating point numbers are re-rounded
according to the current precision.  Polar numbers whose angular
components have strayed from the @mathit{-180} to @mathit{+180} degree range
are normalized.  (Note that results will be undesirable if the current
angular mode is different from the one under which the number was
produced!)  Integers and fractions are generally unaffected by this
operation.  Vectors and formulas are cleaned by cleaning each component
number (i.e., pervasively).

the year specified by the input.  The month, day, and time
of the input date form are lost.  With a numeric prefix argument
@var{n} in the range from 1 to 366, @kbd{t Y} computes the
@var{n}th day of the year (366 is treated as 365 in non-leap
years).  A prefix argument of 0 computes the last day of the
year (December 31).  A negative prefix argument from @mathit{-1} to
@mathit{-12} computes the first day of the @var{n}th month of the year.

@kindex t W
@pindex calc-new-week
@tindex newweek
The @kbd{t W} (@code{calc-new-week}) [@code{newweek}] command

The Lisp variable @code{math-daylight-savings-hook} holds the
name of a function that is used to compute the daylight savings
adjustment for a given date.  The default is
@code{math-std-daylight-savings}, which computes an adjustment
(either 0 or @mathit{-1}) using the North American rules given above.

The daylight savings hook function is called with four arguments:
The date, as a floating-point number in standard Calc format;
a six-element list of the date decomposed into year, month, day,
hour, minute, and second, respectively; a string which contains

@end smallexample

@noindent
The @code{bump} parameter is equal to zero when Calc is converting
from a date form in a generalized time zone into a GMT date value.
It is @mathit{-1} when Calc is converting in the other direction.  The
adjustments shown above ensure that the conversion behaves correctly
and reasonably around the 2 a.m.@: transition in each direction.

There is a ``missing'' hour between 2 a.m.@: and 3 a.m.@: at the
beginning of daylight savings time; converting a date/time form that

particular, negative arguments are converted to positive integers modulo
@expr{2^w} by all binary functions.

If the word size is negative, binary operations produce 2's complement
integers from 
@texline @math{-2^{-w-1}}
@infoline @expr{-(2^(-w-1))} 
to 
@texline @math{2^{-w-1}-1}
@infoline @expr{2^(-w-1)-1} 
inclusive.  Either mode accepts inputs in any range; the sign of
@expr{w} affects only the results produced.

@kindex b c

their results to the current word size.  Note that other operations like
addition do not use the current word size, since integer addition
generally is not ``binary.''  (However, @pxref{Simplification Modes},
@code{calc-bin-simplify-mode}.)  For example, with a word size of 8
bits @kbd{b c} converts a number to the range 0 to 255; with a word
size of @mathit{-8} @kbd{b c} converts to the range @mathit{-128} to 127.

@kindex b w
@pindex calc-word-size
The default word size is 32 bits.  All operations except the shifts and
rotates allow you to specify a different word size for that one

When the binary operations are written in symbolic form, they take an
optional second (or third) word-size parameter.  When a formula like
@samp{and(a,b)} is finally evaluated, the word size current at that time
will be used, but when @samp{and(a,b,-8)} is evaluated, a word size of
@mathit{-8} will always be used.  A symbolic binary function will be left
in symbolic form unless the all of its argument(s) are integers or
integer-valued floats.

If either or both arguments are modulo forms for which @expr{M} is a
power of two, that power of two is taken as the word size unless a

@cindex Golden ratio
One miscellaneous command is shift-@kbd{P} (@code{calc-pi}), which pushes
the value of @cpi{} (at the current precision) onto the stack.  With the
Hyperbolic flag, it pushes the value @expr{e}, the base of natural logarithms.
With the Inverse flag, it pushes Euler's constant 
@texline @math{\gamma}
@infoline @expr{gamma} 
(about 0.5772).  With both Inverse and Hyperbolic, it
pushes the ``golden ratio'' 
@texline @math{\phi}
@infoline @expr{phi} 
(about 1.618).  (At present, Euler's constant is not available
to unlimited precision; Calc knows only the first 100 digits.)
In Symbolic mode, these commands push the
actual variables @samp{pi}, @samp{e}, @samp{gamma}, and @samp{phi},

The @kbd{H L} (@code{calc-log10}) [@code{log10}] command computes the common
(base-10) logarithm of a number.  (With the Inverse flag [@code{exp10}],
it raises ten to a given power.)  Note that the common logarithm of a
complex number is computed by taking the natural logarithm and dividing
by 
@texline @math{\ln10}.
@infoline @expr{ln(10)}.

@kindex B
@kindex I B
@pindex calc-log
@tindex log
@tindex alog
The @kbd{B} (@code{calc-log}) [@code{log}] command computes a logarithm
to any base.  For example, @kbd{1024 @key{RET} 2 B} produces 10, since
@texline @math{2^{10} = 1024}.
@infoline @expr{2^10 = 1024}.  
In certain cases like @samp{log(3,9)}, the result
will be either @expr{1:2} or @expr{0.5} depending on the current Fraction
Mode setting.  With the Inverse flag [@code{alog}], this command is
similar to @kbd{^} except that the order of the arguments is reversed.

@kindex f E
@pindex calc-expm1
@tindex expm1
The @kbd{f E} (@code{calc-expm1}) [@code{expm1}] command computes
@texline @math{e^x - 1},
@infoline @expr{exp(x)-1}, 
but using an algorithm that produces a more accurate
answer when the result is close to zero, i.e., when 
@texline @math{e^x}
@infoline @expr{exp(x)} 
is close to one.

@kindex f L
@pindex calc-lnp1
@tindex lnp1
The @kbd{f L} (@code{calc-lnp1}) [@code{lnp1}] command computes
@texline @math{\ln(x+1)},
@infoline @expr{ln(x+1)}, 
producing a more accurate answer when @expr{x} is close to zero.

@node Trigonometric and Hyperbolic Functions, Advanced Math Functions, Logarithmic Functions, Scientific Functions
@section Trigonometric/Hyperbolic Functions

@kindex f T
@pindex calc-arctan2
@tindex arctan2
The @kbd{f T} (@code{calc-arctan2}) [@code{arctan2}] command takes two
numbers from the stack and computes the arc tangent of their ratio.  The
result is in the full range from @mathit{-180} (exclusive) to @mathit{+180}
(inclusive) degrees, or the analogous range in radians.  A similar
result would be obtained with @kbd{/} followed by @kbd{I T}, but the
value would only be in the range from @mathit{-90} to @mathit{+90} degrees
since the division loses information about the signs of the two
components, and an error might result from an explicit division by zero
which @code{arctan2} would avoid.  By (arbitrary) definition,
@samp{arctan2(0,0)=0}.

The @kbd{f g} (@code{calc-gamma}) [@code{gamma}] command computes the Euler
gamma function.  For positive integer arguments, this is related to the
factorial function:  @samp{gamma(n+1) = fact(n)}.  For general complex
arguments the gamma function can be defined by the following definite
integral:  
@texline @math{\Gamma(a) = \int_0^\infty t^{a-1} e^t dt}.
@infoline @expr{gamma(a) = integ(t^(a-1) exp(t), t, 0, inf)}.  
(The actual implementation uses far more efficient computational methods.)

@kindex f G
@tindex gammaP

@end ignore
@tindex gammaG
The @kbd{f G} (@code{calc-inc-gamma}) [@code{gammaP}] command computes
the incomplete gamma function, denoted @samp{P(a,x)}.  This is defined by
the integral, 
@texline @math{P(a,x) = \left( \int_0^x t^{a-1} e^t dt \right) / \Gamma(a)}.
@infoline @expr{gammaP(a,x) = integ(t^(a-1) exp(t), t, 0, x) / gamma(a)}.
This implies that @samp{gammaP(a,inf) = 1} for any @expr{a} (see the
definition of the normal gamma function).

Several other varieties of incomplete gamma function are defined.

@kindex f b
@pindex calc-beta
@tindex beta
The @kbd{f b} (@code{calc-beta}) [@code{beta}] command computes the
Euler beta function, which is defined in terms of the gamma function as
@texline @math{B(a,b) = \Gamma(a) \Gamma(b) / \Gamma(a+b)},
@infoline @expr{beta(a,b) = gamma(a) gamma(b) / gamma(a+b)}, 
or by
@texline @math{B(a,b) = \int_0^1 t^{a-1} (1-t)^{b-1} dt}.
@infoline @expr{beta(a,b) = integ(t^(a-1) (1-t)^(b-1), t, 0, 1)}.

@kindex f B
@kindex H f B
@pindex calc-inc-beta
@tindex betaI
@tindex betaB
The @kbd{f B} (@code{calc-inc-beta}) [@code{betaI}] command computes
the incomplete beta function @expr{I(x,a,b)}.  It is defined by
@texline @math{I(x,a,b) = \left( \int_0^x t^{a-1} (1-t)^{b-1} dt \right) / B(a,b)}.
@infoline @expr{betaI(x,a,b) = integ(t^(a-1) (1-t)^(b-1), t, 0, x) / beta(a,b)}.
Once again, the @kbd{H} (hyperbolic) prefix gives the corresponding
un-normalized version [@code{betaB}].

@kindex f e

@pindex calc-erf
@tindex erf
@tindex erfc
The @kbd{f e} (@code{calc-erf}) [@code{erf}] command computes the
error function 
@texline @math{\hbox{erf}(x) = {2 \over \sqrt{\pi}} \int_0^x e^{-t^2} dt}.
@infoline @expr{erf(x) = 2 integ(exp(-(t^2)), t, 0, x) / sqrt(pi)}.
The complementary error function @kbd{I f e} (@code{calc-erfc}) [@code{erfc}]
is the corresponding integral from @samp{x} to infinity; the sum
@texline @math{\hbox{erf}(x) + \hbox{erfc}(x) = 1}.
@infoline @expr{erf(x) + erfc(x) = 1}.

@kindex f j
@kindex f y
@pindex calc-bessel-J

   -, -eps      +eps, -      +eps, -
@end smallexample

For @samp{z1^z2}:  This is defined by @samp{exp(ln(z1)*z2)}.
One interesting consequence of this is that @samp{(-8)^1:3} does
not evaluate to @mathit{-2} as you might expect, but to the complex
number @expr{(1., 1.732)}.  Both of these are valid cube roots
of @mathit{-8} (as is @expr{(1., -1.732)}); Calc chooses a perhaps
less-obvious root for the sake of mathematical consistency.

For @samp{arcsin(z)}:  This is defined by @samp{-i*ln(i*z + sqrt(1-z^2))}.
The branch cuts are on the real axis, less than @mathit{-1} and greater than 1.

For @samp{arccos(z)}:  This is defined by @samp{-i*ln(z + i*sqrt(1-z^2))},
or equivalently by @samp{pi/2 - arcsin(z)}.  The branch cuts are on
the real axis, less than @mathit{-1} and greater than 1.

For @samp{arctan(z)}:  This is defined by
@samp{(ln(1+i*z) - ln(1-i*z)) / (2*i)}.  The branch cuts are on the
imaginary axis, below @expr{-i} and above @expr{i}.

For @samp{arccosh(z)}:  This is defined by
@samp{ln(z + (z+1)*sqrt((z-1)/(z+1)))}.  The branch cut is on the
real axis less than 1.

For @samp{arctanh(z)}:  This is defined by @samp{(ln(1+z) - ln(1-z)) / 2}.
The branch cuts are on the real axis, less than @mathit{-1} and greater than 1.

The following tables for @code{arcsin}, @code{arccos}, and
@code{arctan} assume the current angular mode is radians.  The
hyperbolic functions operate independently of the angular mode.

The @kbd{k r} (@code{calc-random}) [@code{random}] command produces
random numbers of various sorts.

Given a positive numeric prefix argument @expr{M}, it produces a random
integer @expr{N} in the range 
@texline @math{0 \le N < M}.
@infoline @expr{0 <= N < M}.  
Each of the @expr{M} values appears with equal probability.

With no numeric prefix argument, the @kbd{k r} command takes its argument
from the stack instead.  Once again, if this is a positive integer @expr{M}
the result is a random integer less than @expr{M}.  However, note that
while numeric prefix arguments are limited to six digits or so, an @expr{M}
taken from the stack can be arbitrarily large.  If @expr{M} is negative,
the result is a random integer in the range 
@texline @math{M < N \le 0}.
@infoline @expr{M < N <= 0}.

If the value on the stack is a floating-point number @expr{M}, the result
is a random floating-point number @expr{N} in the range 
@texline @math{0 \le N < M}
@infoline @expr{0 <= N < M}
or 
@texline @math{M < N \le 0},
@infoline @expr{M < N <= 0}, 
according to the sign of @expr{M}.

If @expr{M} is zero, the result is a Gaussian-distributed random real
number; the distribution has a mean of zero and a standard deviation
of one.  The algorithm used generates random numbers in pairs; thus,
every other call to this function will be especially fast.

If @expr{M} is an error form 
@texline @math{m} @code{+/-} @math{\sigma}
@infoline @samp{m +/- s} 
where @var{m} and 
@texline @math{\sigma}
@infoline @var{s} 
are both real numbers, the result uses a Gaussian distribution with mean
@var{m} and standard deviation 
@texline @math{\sigma}.
@var{s}.

If @expr{M} is an interval form, the lower and upper bounds specify the
acceptable limits of the random numbers.  If both bounds are integers,
the result is a random integer in the specified range.  If either bound

generators that are typically used to implement @code{random}.

If @code{RandSeed} contains an integer, Calc uses this integer to
seed an ``additive congruential'' method (Knuth's algorithm 3.2.2A,
computing 
@texline @math{X_{n-55} - X_{n-24}}.
@infoline @expr{X_n-55 - X_n-24}).  
This method expands the seed
value into a large table which is maintained internally; the variable
@code{RandSeed} is changed from, e.g., 42 to the vector @expr{[42]}
to indicate that the seed has been absorbed into this table.  When

or more, Calc throws it out and tries again until it gets a suitable
value.

To create a random floating-point number with precision @var{p}, Calc
simply creates a random @var{p}-digit integer and multiplies by
@texline @math{10^{-p}}.
@infoline @expr{10^-p}.  
The resulting random numbers should be very clean, but note
that relatively small numbers will have few significant random digits.
In other words, with a precision of 12, you will occasionally get
numbers on the order of 
@texline @math{10^{-9}}
@infoline @expr{10^-9} 
or 
@texline @math{10^{-10}},
@infoline @expr{10^-10}, 
but those numbers will only have two or three random digits since they
correspond to small integers times 
@texline @math{10^{-12}}.
@infoline @expr{10^-12}.

To create a random integer in the interval @samp{[0 .. @var{m})}, Calc
counts the digits in @var{m}, creates a random integer with three
additional digits, then reduces modulo @var{m}.  Unless @var{m} is a

@pindex calc-extended-gcd
@tindex egcd
The @kbd{k E} (@code{calc-extended-gcd}) [@code{egcd}] command computes
the GCD of two integers @expr{x} and @expr{y} and returns a vector
@expr{[g, a, b]} where 
@texline @math{g = \gcd(x,y) = a x + b y}.
@infoline @expr{g = gcd(x,y) = a x + b y}.

@kindex !
@pindex calc-factorial
@tindex fact

binomial coefficient @expr{N}-choose-@expr{M}, where @expr{M} is the number
on the top of the stack and @expr{N} is second-to-top.  If both arguments
are integers, the result is an exact integer.  Otherwise, the result is a
floating-point approximation.  The binomial coefficient is defined for all
real numbers by
@texline @math{N! \over M! (N-M)!\,}.
@infoline @expr{N! / M! (N-M)!}.

@kindex H k c
@pindex calc-perm
@tindex perm

@pindex calc-stirling-number
@tindex stir1
@tindex stir2
The @kbd{k s} (@code{calc-stirling-number}) [@code{stir1}] command
computes a Stirling number of the first 
@texline kind@tie{}@math{n \brack m},
@infoline kind,
given two integers @expr{n} and @expr{m} on the stack.  The @kbd{H k s}
[@code{stir2}] command computes a Stirling number of the second 
@texline kind@tie{}@math{n \brace m}.
@infoline kind.
These are the number of @expr{m}-cycle permutations of @expr{n} objects,
and the number of ways to partition @expr{n} objects into @expr{m}
non-empty sets, respectively.

to 25 million, the answer is exact although it may take some time.  The
result is a vector of the prime factors in increasing order.  For larger
inputs, prime factors above 5000 may not be found, in which case the
last number in the vector will be an unfactored integer greater than 25
million (with a warning message).  For negative integers, the first
element of the list will be @mathit{-1}.  For inputs @mathit{-1}, @mathit{0}, and
@mathit{1}, the result is a list of the same number.

@kindex k n
@pindex calc-next-prime
@ignore
@mindex nextpr@idots

@kindex k t
@pindex calc-totient
@tindex totient
The @kbd{k t} (@code{calc-totient}) [@code{totient}] command computes the
Euler ``totient'' 
@texline function@tie{}@math{\phi(n)},
@infoline function,
the number of integers less than @expr{n} which
are relatively prime to @expr{n}.

@kindex k m
@pindex calc-moebius
@tindex moebius
The @kbd{k m} (@code{calc-moebius}) [@code{moebius}] command computes the
@texline M@"obius @math{\mu}
@infoline Moebius ``mu''
function.  If the input number is a product of @expr{k}
distinct factors, this is @expr{(-1)^k}.  If the input number has any
duplicate factors (i.e., can be divided by the same prime more than once),
the result is zero.

@ignore
@mindex @null
@end ignore
@tindex ltpc
The @samp{utpc(x,v)} function uses the chi-square distribution with
@texline @math{\nu}
@infoline @expr{v} 
degrees of freedom.  It is the probability that a model is
correct if its chi-square statistic is @expr{x}.

@kindex k F

@mindex @null
@end ignore
@tindex ltpf
The @samp{utpf(F,v1,v2)} function uses the F distribution, used in
various statistical tests.  The parameters 
@texline @math{\nu_1}
@infoline @expr{v1} 
and 
@texline @math{\nu_2}
@infoline @expr{v2}
are the degrees of freedom in the numerator and denominator,
respectively, used in computing the statistic @expr{F}.

@kindex k N

@mindex @null
@end ignore
@tindex ltpn
The @samp{utpn(x,m,s)} function uses a normal (Gaussian) distribution
with mean @expr{m} and standard deviation 
@texline @math{\sigma}.
@infoline @expr{s}.  
It is the probability that such a normal-distributed random variable
would exceed @expr{x}.

@kindex k P

@mindex @null
@end ignore
@tindex ltpt
The @samp{utpt(t,v)} function uses the Student's ``t'' distribution
with 
@texline @math{\nu}
@infoline @expr{v} 
degrees of freedom.  It is the probability that a
t-distributed random variable will be greater than @expr{t}.
(Note:  This computes the distribution function 
@texline @math{A(t|\nu)}
@infoline @expr{A(t|v)}
where 
@texline @math{A(0|\nu) = 1}
@infoline @expr{A(0|v) = 1} 
and 
@texline @math{A(\infty|\nu) \to 0}.
@infoline @expr{A(inf|v) -> 0}.  
The @code{UTPT} operation on the HP-48 uses a different definition which
returns half of Calc's value:  @samp{UTPT(t,v) = .5*utpt(t,v)}.)

While Calc does not provide inverses of the probability distribution

The first is the mantissa; the second, which must be an
integer, is the exponent.  The result is the mantissa
times ten to the power of the exponent.

@item -12
This is treated the same as @mathit{-11} by the @kbd{v p} command.
When unpacking, @mathit{-12} specifies that a floating-point mantissa
is desired.

@item -13
A real number is converted into a date form.

returned in the form @samp{[@w{[a, b, c]}, [d, e, f]]}.

If any elements of the vector are negative, other kinds of
packing are done at that level as described above.  For
example, @samp{[2, 3, -4]} takes 12 objects and creates a
@texline @math{2\times3}
@infoline 2x3
matrix of error forms: @samp{[[a +/- b, c +/- d ... ]]}.
Also, @samp{[-4, -10]} will convert four integers into an
error form consisting of two fractions:  @samp{a:b +/- c:d}.

For example, given the vector @samp{[@w{a +/- b}, c^2, d +/- 7]},
the result of @kbd{C-u -4 v u} will be the two vectors
@samp{[a, c^2, d]} and @w{@samp{[b, 0, 7]}}.

Note that the prefix argument can have an effect even when the input is
not a vector.  For example, if the input is the number @mathit{-5}, then
@kbd{c-u -1 v u} yields @mathit{-5} and 0 (the components of @mathit{-5}
when viewed as a rectangular complex number); @kbd{C-u -2 v u} yields 5
and 180 (assuming degrees mode); and @kbd{C-u -10 v u} yields @mathit{-5}
and 1 (the numerator and denominator of @mathit{-5}, viewed as a rational
number).  Plain @kbd{v u} with this input would complain that the input
is not a composite object.

Unpacking mode @mathit{-11} converts a float into an integer mantissa and
an integer exponent, where the mantissa is not divisible by 10
(except that 0.0 is represented by a mantissa and exponent of 0).
Unpacking mode @mathit{-12} converts a float into a floating-point mantissa
and integer exponent, where the mantissa (for non-zero numbers)
is guaranteed to lie in the range [1 .. 10).  In both cases,
the mantissa is shifted left or right (and the exponent adjusted
to compensate) in order to satisfy these constraints.

prefix, if specified, must match the size of the vector.  If the value on
the stack is a scalar, it is used for each element on the diagonal, and
the prefix argument is required.

To build a constant square matrix, e.g., a 
@texline @math{3\times3}
@infoline 3x3
matrix filled with ones, use @kbd{0 M-3 v d 1 +}, i.e., build a zero
matrix first and then add a constant value to that matrix.  (Another
alternative would be to use @kbd{v b} and @kbd{v a}; see below.)

@tindex index
The @kbd{v x} (@code{calc-index}) [@code{index}] function builds a vector
of consecutive integers from 1 to @var{n}, where @var{n} is the numeric
prefix argument.  If you do not provide a prefix argument, you will be
prompted to enter a suitable number.  If @var{n} is negative, the result
is a vector of negative integers from @var{n} to @mathit{-1}.

With a prefix argument of just @kbd{C-u}, the @kbd{v x} command takes
three values from the stack: @var{n}, @var{start}, and @var{incr} (with
@var{incr} at top-of-stack).  Counting starts at @var{start} and increases
by @var{incr} for successive vector elements.  If @var{start} or @var{n}

@tindex mdims
With the Hyperbolic flag, @kbd{H v l} [@code{mdims}] computes a vector
of the dimensions of a vector, matrix, or higher-order object.  For
example, @samp{mdims([[a,b,c],[d,e,f]])} returns @samp{[2, 3]} since
its argument is a 
@texline @math{2\times3}
@infoline 2x3
matrix.

@kindex v f
@pindex calc-vector-find

If the number of columns does not evenly divide the number of elements
in the vector, the last row will be short and the result will not be
suitable for use as a matrix.  For example, with the matrix
@samp{[[1, 2], @w{[3, 4]}]} on the stack, @kbd{v a 4} produces
@samp{[[1, 2, 3, 4]]} (a 
@texline @math{1\times4}
@infoline 1x4
matrix), @kbd{v a 1} produces @samp{[[1], [2], [3], [4]]} (a 
@texline @math{4\times1}
@infoline 4x1
matrix), @kbd{v a 2} produces @samp{[[1, 2], [3, 4]]} (the original 
@texline @math{2\times2}
@infoline 2x2
matrix), @w{@kbd{v a 3}} produces @samp{[[1, 2, 3], [4]]} (not a
matrix), and @kbd{v a 0} produces the flattened list 
@samp{[1, 2, @w{3, 4}]}.

and only if it is in both of the input sets.  Thus if the input
sets are disjoint, i.e., if they share no common elements, the result
will be the empty vector @samp{[]}.  Note that the characters @kbd{V}
and @kbd{^} were chosen to be close to the conventional mathematical
notation for set 
@texline union@tie{}(@math{A \cup B})
@infoline union
and 
@texline intersection@tie{}(@math{A \cap B}).
@infoline intersection.

@kindex V -
@pindex calc-set-difference
@tindex vdiff

the same set.  The set may include positive infinity, but must
not include any negative numbers.  The input is interpreted as a
set of integers in the sense of @kbd{V F} (@code{vfloor}).  Beware
that a simple input like @samp{[100]} can result in a huge integer
representation 
@texline (@math{2^{100}}, a 31-digit integer, in this case).
@infoline (@expr{2^100}, a 31-digit integer, in this case).

@node Statistical Operations, Reducing and Mapping, Set Operations, Matrix Functions
@section Statistical Operations on Vectors

@tindex vmean
@cindex Mean of data values
The @kbd{u M} (@code{calc-vector-mean}) [@code{vmean}] command
computes the average (arithmetic mean) of the data values.
If the inputs are error forms 
@texline @math{x \pm \sigma},
@infoline @samp{x +/- s}, 
this is the weighted mean of the @expr{x} values with weights 
@texline @math{1 /\sigma^2}.
@infoline @expr{1 / s^2}.
@tex
\turnoffactive
$$ \mu = { \displaystyle \sum { x_i \over \sigma_i^2 } \over
           \displaystyle \sum { 1 \over \sigma_i^2 } } $$

If the inputs are not error forms, this is simply the sum of the
values divided by the count of the values.

Note that a plain number can be considered an error form with
error 
@texline @math{\sigma = 0}.
@infoline @expr{s = 0}.  
If the input to @kbd{u M} is a mixture of
plain numbers and error forms, the result is the mean of the
plain numbers, ignoring all values with non-zero errors.  (By the
above definitions it's clear that a plain number effectively

@tindex vsdev
@cindex Standard deviation
@cindex Sample statistics
The @kbd{u S} (@code{calc-vector-sdev}) [@code{vsdev}] command
computes the standard 
@texline deviation@tie{}@math{\sigma}
@infoline deviation
of the data values.  If the values are error forms, the errors are used
as weights just as for @kbd{u M}.  This is the @emph{sample} standard
deviation, whose value is the square root of the sum of the squares of
the differences between the values and the mean of the @expr{N} values,

This function also applies to distributions.  The standard deviation
of a single error form is simply the error part.  The standard deviation
of a continuous interval happens to equal the difference between the
limits, divided by 
@texline @math{\sqrt{12}}.
@infoline @expr{sqrt(12)}.  
The standard deviation of an integer interval is the same as the
standard deviation of a vector of those integers.

@kindex I u S

@cindex Variance of data values
The @kbd{H u S} (@code{calc-vector-variance}) [@code{vvar}] and
@kbd{H I u S} (@code{calc-vector-pop-variance}) [@code{vpvar}]
commands compute the variance of the data values.  The variance
is the 
@texline square@tie{}@math{\sigma^2}
@infoline square
of the standard deviation, i.e., the sum of the
squares of the deviations of the data values from the mean.
(This definition also applies when the argument is a distribution.)

The functions in this section take two arguments, which must be
vectors of equal size.  The vectors are each flattened in the same
way as by the single-variable statistical functions.  Given a numeric
prefix argument of 1, these functions instead take one object from
the stack, which must be an 
@texline @math{N\times2}
@infoline Nx2
matrix of data values.  Once again, variable names can be used in place
of actual vectors and matrices.

@kindex u C

If any argument to @kbd{V M} is a matrix, the operator is normally mapped
across all elements of the matrix.  For example, given the matrix
@expr{[[1, -2, 3], [-4, 5, -6]]}, @kbd{V M A} takes six absolute values to
produce another 
@texline @math{3\times2}
@infoline 3x2
matrix, @expr{[[1, 2, 3], [4, 5, 6]]}.

@tindex mapr
The command @kbd{V M _} [@code{mapr}] (i.e., type an underscore at the

If you give a numeric prefix argument of 2 to @kbd{a v}, it simplifies
as if in algebraic simplification mode.  This is equivalent to typing
@kbd{a s}; @pxref{Simplifying Formulas}.  If you give a numeric prefix
of 3 or more, it uses extended simplification mode (@kbd{a e}).

If you give a negative prefix argument @mathit{-1}, @mathit{-2}, or @mathit{-3},
it simplifies in the corresponding mode but only works on the top-level
function call of the formula.  For example, @samp{(2 + 3) * (2 + 3)} will
simplify to @samp{(2 + 3)^2}, without simplifying the sub-formulas
@samp{2 + 3}.  As another example, typing @kbd{V R +} to sum the vector
@samp{[1, 2, 3, 4]} produces the formula @samp{reduce(add, [1, 2, 3, 4])}

cases where the order of terms in a sum is changed by the default
simplifications.)

The distributive law is used to simplify sums in some cases:
@expr{a x + b x} to @expr{(a + b) x}, where @expr{a} represents
a number or an implicit 1 or @mathit{-1} (as in @expr{x} or @expr{-x})
and similarly for @expr{b}.  Use the @kbd{a c}, @w{@kbd{a f}}, or
@kbd{j M} commands to merge sums with non-numeric coefficients
using the distributive law.

The distributive law is only used for sums of two terms, or

@expr{(-a) (b - c)}, where @expr{-a} is a negative number, is
rewritten to @expr{a (c - b)}.

The distributive law of products and powers is used for adjacent
terms of the product: @expr{x^a x^b} goes to 
@texline @math{x^{a+b}}
@infoline @expr{x^(a+b)}
where @expr{a} is a number, or an implicit 1 (as in @expr{x}),
or the implicit one-half of @expr{@t{sqrt}(x)}, and similarly for
@expr{b}.  The result is written using @samp{sqrt} or @samp{1/sqrt}
if the sum of the powers is @expr{1/2} or @expr{-1/2}, respectively.
If the sum of the powers is zero, the product is simplified to
@expr{1} or to @samp{idn(1)} if matrix mode is enabled.

The product of a negative power times anything but another negative
power is changed to use division:  
@texline @math{x^{-2} y}
@infoline @expr{x^(-2) y} 
goes to @expr{y / x^2} unless matrix mode is
in effect and neither @expr{x} nor @expr{y} are scalar (in which
case it is considered unsafe to rearrange the order of the terms).

The quotient @expr{x / 0} is left unsimplified or changed to an
infinite quantity, as directed by the current infinite mode.
@xref{Infinite Mode}.

The expression 
@texline @math{a / b^{-c}}
@infoline @expr{a / b^(-c)} 
is changed to @expr{a b^c}, where @expr{-c} is any negative-looking
power.  Also, @expr{1 / b^c} is changed to 
@texline @math{b^{-c}}
@infoline @expr{b^(-c)} 
for any power @expr{c}.

Also, @expr{(-a) / b} and @expr{a / (-b)} go to @expr{-(a/b)};
@expr{(a/b) / c} goes to @expr{a / (b c)}; and @expr{a / (b/c)}

Powers of products or quotients @expr{(a b)^c}, @expr{(a/b)^c}
are distributed to @expr{a^c b^c}, @expr{a^c / b^c} only if @expr{c}
is an integer, or if either @expr{a} or @expr{b} are nonnegative
real numbers.  Powers of powers @expr{(a^b)^c} are simplified to
@texline @math{a^{b c}}
@infoline @expr{a^(b c)} 
only when @expr{c} is an integer and @expr{b c} also
evaluates to an integer.  Without these restrictions these simplifications
would not be safe because of problems with principal values.
(In other words, 
@texline @math{((-3)^{1/2})^2}
@infoline @expr{((-3)^1:2)^2} 
is safe to simplify, but
@texline @math{((-3)^2)^{1/2}}
@infoline @expr{((-3)^2)^1:2} 
is not.)  @xref{Declarations}, for ways to inform Calc that your
variables satisfy these requirements.

As a special case of this rule, @expr{@t{sqrt}(x)^n} is simplified to
@texline @math{x^{n/2}}
@infoline @expr{x^(n/2)} 
only for even integers @expr{n}.

If @expr{a} is known to be real, @expr{b} is an even integer, and
@expr{c} is a half- or quarter-integer, then @expr{(a^b)^c} is

Also, @expr{(-a)^b} is simplified to @expr{a^b} if @expr{b} is an
even integer, or to @expr{-(a^b)} if @expr{b} is an odd integer,
for any negative-looking expression @expr{-a}.

Square roots @expr{@t{sqrt}(x)} generally act like one-half powers
@texline @math{x^{1:2}}
@infoline @expr{x^1:2} 
for the purposes of the above-listed simplifications.

Also, note that 
@texline @math{1 / x^{1:2}}
@infoline @expr{1 / x^1:2} 
is changed to 
@texline @math{x^{-1:2}},
@infoline @expr{x^(-1:2)},
but @expr{1 / @t{sqrt}(x)} is left alone.

@tex
\bigskip

taken into account when terms of a product are being compared.
Thus @expr{(x + y) (y + x)} will be simplified to @expr{(x + y)^2}.
A subtle point is that @expr{(x - y) (y - x)} will @emph{not}
be simplified to @expr{-(x - y)^2}; Calc does not notice that
one term can be written as a constant times the other, even if
that constant is @mathit{-1}.

A fraction times any expression, @expr{(a:b) x}, is changed to
a quotient involving integers:  @expr{a x / b}.  This is not
done for floating-point numbers like @expr{0.5}, however.  This
is one reason why you may find it convenient to turn Fraction mode

Square roots of integer or rational arguments are simplified in
several ways.  (Note that these will be left unevaluated only in
Symbolic mode.)  First, square integer or rational factors are
pulled out so that @expr{@t{sqrt}(8)} is rewritten as
@texline @math{2\,\t{sqrt}(2)}.
@infoline @expr{2 sqrt(2)}.  
Conceptually speaking this implies factoring the argument into primes
and moving pairs of primes out of the square root, but for reasons of
efficiency Calc only looks for primes up to 29.

No simplifications for inverse trigonometric and hyperbolic
functions are known, except for negative arguments of @code{arcsin},
@code{arctan}, @code{arcsinh}, and @code{arctanh}.  Note that
@expr{@t{arcsin}(@t{sin}(x))} can @emph{not} safely change to
@expr{x}, since this only correct within an integer multiple of 
@texline @math{2 \pi}
@infoline @expr{2 pi} 
radians or 360 degrees.  However, @expr{@t{arcsinh}(@t{sinh}(x))} is
simplified to @expr{x} if @expr{x} is known to be real.

Several simplifications that apply to logarithms and exponentials
are that @expr{@t{exp}(@t{ln}(x))}, 
@texline @t{e}@math{^{\ln(x)}},
@infoline @expr{e^@t{ln}(x)}, 
and
@texline @math{10^{{\rm log10}(x)}}
@infoline @expr{10^@t{log10}(x)} 
all reduce to @expr{x}.  Also, @expr{@t{ln}(@t{exp}(x))}, etc., can
reduce to @expr{x} if @expr{x} is provably real.  The form
@expr{@t{exp}(x)^y} is simplified to @expr{@t{exp}(x y)}.  If @expr{x}
is a suitable multiple of 
@texline @math{\pi i} 
@infoline @expr{pi i}
(as described above for the trigonometric functions), then
@expr{@t{exp}(x)} or @expr{e^x} will be expanded.  Finally,
@expr{@t{ln}(x)} is simplified to a form involving @code{pi} and
@code{i} where @expr{x} is provably negative, positive imaginary, or

values of @expr{x} in a certain range; outside that range, values
are folded down to the 360-degree range that the inverse trigonometric
functions always produce.

Powers of powers @expr{(x^a)^b} are simplified to 
@texline @math{x^{a b}}
@infoline @expr{x^(a b)}
for all @expr{a} and @expr{b}.  These results will be valid only
in a restricted range of @expr{x}; for example, in 
@texline @math{(x^2)^{1:2}}
@infoline @expr{(x^2)^1:2}
the powers cancel to get @expr{x}, which is valid for positive values
of @expr{x} but not for negative or complex values.

Similarly, @expr{@t{sqrt}(x^a)} and @expr{@t{sqrt}(x)^a} are both
simplified (possibly unsafely) to 
@texline @math{x^{a/2}}.
@infoline @expr{x^(a/2)}.

Forms like @expr{@t{sqrt}(1 - sin(x)^2)} are simplified to, e.g.,
@expr{@t{cos}(x)}.  Calc has identities of this sort for @code{sin},
@code{cos}, @code{tan}, @code{sinh}, and @code{cosh}.

number for an answer, then the quotient simplifies to that number.

For powers and square roots, the ``unsafe'' simplifications
@expr{(a b)^c} to @expr{a^c b^c}, @expr{(a/b)^c} to @expr{a^c / b^c},
and @expr{(a^b)^c} to 
@texline @math{a^{b c}}
@infoline @expr{a^(b c)} 
are done if the powers are real numbers.  (These are safe in the context
of units because all numbers involved can reasonably be assumed to be
real.)

according to the previous paragraph.  For example, @samp{acre^1.5}
is simplified by noting that @expr{1.5 = 3:2}, that @samp{acre}
is defined in terms of @samp{m^2}, and that the 2 in the power of
@code{m} is a multiple of 2 in @expr{3:2}.  Thus, @code{acre^1.5} is
replaced by approximately 
@texline @math{(4046 m^2)^{1.5}}
@infoline @expr{(4046 m^2)^1.5}, 
which is then changed to 
@texline @math{4046^{1.5} \, (m^2)^{1.5}},
@infoline @expr{4046^1.5 (m^2)^1.5}, 
then to @expr{257440 m^3}.

The functions @code{float}, @code{frac}, @code{clean}, @code{abs},
as well as @code{floor} and the other integer truncation functions,

answer!

If you use the @code{deriv} function directly in an algebraic formula,
you can write @samp{deriv(f,x,x0)} which represents the derivative
of @expr{f} with respect to @expr{x}, evaluated at the point 
@texline @math{x=x_0}.
@infoline @expr{x=x0}.

If the formula being differentiated contains functions which Calc does
not know, the derivatives of those functions are produced by adding
primes (apostrophe characters).  For example, @samp{deriv(f(2x), x)}

respect to a variable.  The integrator is not guaranteed to work for
all integrable functions, but it is able to integrate several large
classes of formulas.  In particular, any polynomial or rational function
(a polynomial divided by a polynomial) is acceptable.  (Rational functions
don't have to be in explicit quotient form, however; 
@texline @math{x/(1+x^{-2})}
@infoline @expr{x/(1+x^-2)}
is not strictly a quotient of polynomials, but it is equivalent to
@expr{x^3/(x^2+1)}, which is.)  Also, square roots of terms involving
@expr{x} and @expr{x^2} may appear in rational functions being
integrated.  Finally, rational functions involving trigonometric or

@end tex

Please note that the current implementation of Calc's integrator sometimes
produces results that are significantly more complex than they need to
be.  For example, the integral Calc finds for 
@texline @math{1/(x+\sqrt{x^2+1})}
@infoline @expr{1/(x+sqrt(x^2+1))}
is several times more complicated than the answer Mathematica
returns for the same input, although the two forms are numerically
equivalent.  Also, any indefinite integral should be considered to have
an arbitrary constant of integration added to it, although Calc does not
write an explicit constant of integration in its result.  For example,
Calc's solution for 
@texline @math{1/(1+\tan x)}
@infoline @expr{1/(1+tan(x))} 
differs from the solution given in the @emph{CRC Math Tables} by a
constant factor of  
@texline @math{\pi i / 2}
@infoline @expr{pi i / 2},
due to a different choice of constant of integration.

The Calculator remembers all the integrals it has done.  If conditions
change in a way that would invalidate the old integrals, say, a switch

@end ignore
@tindex Ei
As a more serious example, the expression @samp{exp(x)/x} cannot be
integrated in terms of the standard functions, so the ``exponential
integral'' function 
@texline @math{{\rm Ei}(x)}
@infoline @expr{Ei(x)} 
was invented to describe it.
We can get Calc to do this integral in terms of a made-up @code{Ei}
function by adding the rule @samp{[integtry(exp(x)/x, x) := Ei(x)]}
to @code{IntegRules}.  Now entering @samp{exp(2x)/x} on the stack

form @expr{X = 0}.

This command also works for inequalities, as in @expr{y < 3x + 6}.
Some inequalities cannot be solved where the analogous equation could
be; for example, solving 
@texline @math{a < b \, c}
@infoline @expr{a < b c} 
for @expr{b} is impossible
without knowing the sign of @expr{c}.  In this case, @kbd{a S} will
produce the result 
@texline @math{b \mathbin{\hbox{\code{!=}}} a/c}
@infoline @expr{b != a/c} 
(using the not-equal-to operator) to signify that the direction of the
inequality is now unknown.  The inequality 
@texline @math{a \le b \, c}
@infoline @expr{a <= b c} 
is not even partially solved.  @xref{Declarations}, for a way to tell
Calc that the signs of the variables in a formula are in fact known.

Two useful commands for working with the result of @kbd{a S} are

Some equations have more than one solution.  The Hyperbolic flag
(@code{H a S}) [@code{fsolve}] tells the solver to report the fully
general family of solutions.  It will invent variables @code{n1},
@code{n2}, @dots{}, which represent independent arbitrary integers, and
@code{s1}, @code{s2}, @dots{}, which represent independent arbitrary
signs (either @mathit{+1} or @mathit{-1}).  If you don't use the Hyperbolic
flag, Calc will use zero in place of all arbitrary integers, and plus
one in place of all arbitrary signs.  Note that variables like @code{n1}
and @code{s1} are not given any special interpretation in Calc except by
the equation solver itself.  As usual, you can use the @w{@kbd{s l}}
(@code{calc-let}) command to obtain solutions for various actual values

value of the variable which minimizes the formula's value, along
with the minimum value itself.

Note that this command looks for a @emph{local} minimum.  Many functions
have more than one minimum; some, like 
@texline @math{x \sin x},
@infoline @expr{x sin(x)}, 
have infinitely many.  In fact, there is no easy way to define the
``global'' minimum of 
@texline @math{x \sin x}
@infoline @expr{x sin(x)} 
but Calc can still locate any particular local minimum
for you.  Calc basically goes downhill from the initial guess until it
finds a point at which the function's value is greater both to the left
and to the right.  Calc does not use derivatives when minimizing a function.

the @dfn{parameters} of the model.

The @kbd{a F} command takes the data set to be fitted from the stack.
By default, it expects the data in the form of a matrix.  For example,
for a linear or polynomial fit, this would be a 
@texline @math{2\times N}
@infoline 2xN
matrix where the first row is a list of @expr{x} values and the second
row has the corresponding @expr{y} values.  For the multilinear fit
shown above, the matrix would have four rows (@expr{x_1}, @expr{x_2},
@expr{x_3}, and @expr{y}, respectively).

If you happen to have an 
@texline @math{N\times2}
@infoline Nx2
matrix instead of a 
@texline @math{2\times N}
@infoline 2xN
matrix, just press @kbd{v t} first to transpose the matrix.

After you type @kbd{a F}, Calc prompts you to select a model.  For a
linear fit, press the digit @kbd{1}.

@noindent
which is clearly zero if @expr{a + b x} exactly fits all data points,
and increases as various @expr{a + b x_i} values fail to match the
corresponding @expr{y_i} values.  There are several reasons why the
summand is squared, one of them being to ensure that 
@texline @math{\chi^2 \ge 0}.
@infoline @expr{chi^2 >= 0}.
Least-squares fitting simply chooses the values of @expr{a} and @expr{b}
for which the error 
@texline @math{\chi^2}
@infoline @expr{chi^2} 
is as small as possible.

Other kinds of models do the same thing but with a different model
formula in place of @expr{a + b x_i}.

0.142857142855 x^2 + 1.34285714287 x + 3.59999999998
@end example

An important result from the theory of polynomial fitting is that it
is always possible to fit @var{n} data points exactly using a polynomial
of degree @mathit{@var{n}-1}, sometimes called an @dfn{interpolating polynomial}.
Using the modified (14) data matrix, a model number of 4 gives
a polynomial that exactly matches all five data points:

@example
0.04167 x^4 - 0.4167 x^3 + 1.458 x^2 - 0.08333 x + 4.

It is also possible for the @emph{input} to a fitting operation to
contain error forms.  The data values must either all include errors
or all be plain numbers.  Error forms can go anywhere but generally
go on the numbers in the last row of the data matrix.  If the last
row contains error forms
@texline `@var{y_i}@w{ @t{+/-} }@math{\sigma_i}', 
@infoline `@var{y_i}@w{ @t{+/-} }@var{sigma_i}', 
then the 
@texline @math{\chi^2}
@infoline @expr{chi^2}
statistic is now,

@ifinfo
@example

the fitting operation.

If there are error forms on other rows of the data matrix, all the
errors for a given data point are combined; the square root of the
sum of the squares of the errors forms the 
@texline @math{\sigma_i}
@infoline @expr{sigma_i} 
used for the data point.

Both @kbd{a F} and @kbd{H a F} can accept error forms in the input
matrix, although if you are concerned about error analysis you will
probably use @kbd{H a F} so that the output also contains error
estimates.

If the input contains error forms but all the 
@texline @math{\sigma_i}
@infoline @expr{sigma_i} 
values are the same, it is easy to see that the resulting fitted model
will be the same as if the input did not have error forms at all 
@texline (@math{\chi^2}
@infoline (@expr{chi^2}
is simply scaled uniformly by 
@texline @math{1 / \sigma^2},
@infoline @expr{1 / sigma^2}, 
which doesn't affect where it has a minimum).  But there @emph{will} be
a difference in the estimated errors of the coefficients reported by
@kbd{H a F}. 

will have length @expr{M = d+1} with the constant term first.

@item
The covariance matrix @expr{C} computed from the fit.  This is
an @var{m}x@var{m} symmetric matrix; the diagonal elements
@texline @math{C_{jj}}
@infoline @expr{C_j_j} 
are the variances 
@texline @math{\sigma_j^2}
@infoline @expr{sigma_j^2} 
of the parameters.  The other elements are covariances
@texline @math{\sigma_{ij}^2} 
@infoline @expr{sigma_i_j^2} 
that describe the correlation between pairs of parameters.  (A related
set of numbers, the @dfn{linear correlation coefficients} 
@texline @math{r_{ij}},
@infoline @expr{r_i_j},
are defined as 
@texline @math{\sigma_{ij}^2 / \sigma_i \, \sigma_j}.)
@infoline @expr{sigma_i_j^2 / sigma_i sigma_j}.)

@item
A vector of @expr{M} ``parameter filter'' functions whose
meanings are described below.  If no filters are necessary this
will instead be an empty vector; this is always the case for the
polynomial and multilinear fits described so far.

@item
The value of 
@texline @math{\chi^2}
@infoline @expr{chi^2} 
for the fit, calculated by the formulas shown above.  This gives a
measure of the quality of the fit; statisticians consider
@texline @math{\chi^2 \approx N - M}
@infoline @expr{chi^2 = N - M} 
to indicate a moderately good fit (where again @expr{N} is the number of
data points and @expr{M} is the number of parameters).

@item
A measure of goodness of fit expressed as a probability @expr{Q}.
This is computed from the @code{utpc} probability distribution
function using 
@texline @math{\chi^2}
@infoline @expr{chi^2} 
with @expr{N - M} degrees of freedom.  A
value of 0.5 implies a good fit; some texts recommend that often
@expr{Q = 0.1} or even 0.001 can signify an acceptable fit.  In
particular, 
@texline @math{\chi^2}
@infoline @expr{chi^2} 
statistics assume the errors in your inputs
follow a normal (Gaussian) distribution; if they don't, you may
have to accept smaller values of @expr{Q}.

The @expr{Q} value is computed only if the input included error
estimates.  Otherwise, Calc will report the symbol @code{nan}
for @expr{Q}.  The reason is that in this case the 
@texline @math{\chi^2}
@infoline @expr{chi^2}
value has effectively been used to estimate the original errors
in the input, and thus there is no redundant information left
over to use for a confidence test.
@end enumerate

Here is a complete list of the standard models recognized by @kbd{a F}:

@table @kbd
@item 1
Linear or multilinear.  @mathit{a + b x + c y + d z}.
@item 2-9
Polynomials.  @mathit{a + b x + c x^2 + d x^3}.
@item e
Exponential.  @mathit{a} @t{exp}@mathit{(b x)} @t{exp}@mathit{(c y)}.
@item E
Base-10 exponential.  @mathit{a} @t{10^}@mathit{(b x)} @t{10^}@mathit{(c y)}.
@item x
Exponential (alternate notation).  @t{exp}@mathit{(a + b x + c y)}.
@item X
Base-10 exponential (alternate).  @t{10^}@mathit{(a + b x + c y)}.
@item l
Logarithmic.  @mathit{a + b} @t{ln}@mathit{(x) + c} @t{ln}@mathit{(y)}.
@item L
Base-10 logarithmic.  @mathit{a + b} @t{log10}@mathit{(x) + c} @t{log10}@mathit{(y)}.
@item ^
General exponential.  @mathit{a b^x c^y}.
@item p
Power law.  @mathit{a x^b y^c}.
@item q
Quadratic.  @mathit{a + b (x-c)^2 + d (x-e)^2}.
@item g
Gaussian.  
@texline @math{{a \over b \sqrt{2 \pi}} \exp\left( -{1 \over 2} \left( x - c \over b \right)^2 \right)}.
@infoline @mathit{(a / b sqrt(2 pi)) exp(-0.5*((x-c)/b)^2)}.
@end table

All of these models are used in the usual way; just press the appropriate
letter at the model prompt, and choose variable names if you wish.  The
result will be a formula as shown in the above table, with the best-fit

Calc uses the principal values of inverse functions like @code{ln}
and @code{arcsin} when doing fits.  For example, when you enter
the model @samp{y = sin(a t + b)} Calc actually uses the easier
form @samp{arcsin(y) = a t + b}.  The @code{arcsin} function always
returns results in the range from @mathit{-90} to 90 degrees (or the
equivalent range in radians).  Suppose you had data that you
believed to represent roughly three oscillations of a sine wave,
so that the argument of the sine might go from zero to 
@texline @math{3\times360}
@infoline @mathit{3*360} 
degrees.
The above model would appear to be a good way to determine the
true frequency and phase of the sine wave, but in practice it
would fail utterly.  The righthand side of the actual model
@samp{arcsin(y) = a t + b} will grow smoothly with @expr{t}, but
the lefthand side will bounce back and forth between @mathit{-90} and 90.
No values of @expr{a} and @expr{b} can make the two sides match,
even approximately.

There is no good solution to this problem at present.  You could
restrict your data to small enough ranges so that the above problem

ln(y) = ln(a) + b ln(x)
@end example

@noindent
which matches the desired form with 
@texline @math{Y = \ln(y)},
@infoline @expr{Y = ln(y)}, 
@texline @math{A = \ln(a)},
@infoline @expr{A = ln(a)},
@expr{F = 1}, @expr{B = b}, and 
@texline @math{G = \ln(x)}.
@infoline @expr{G = ln(x)}.  
Calc thus computes the logarithms of your @expr{y} and @expr{x} values,
does a linear fit for @expr{A} and @expr{B}, then solves to get 
@texline @math{a = \exp(A)} 
@infoline @expr{a = exp(A)} 
and @expr{b = B}.

Another interesting example is the ``quadratic'' model, which can
be handled by expanding according to the distributive law.

y = a + b c^2 - 2 b c x + b x^2
@end example

@noindent
which matches with @expr{Y = y}, @expr{A = a + b c^2}, @expr{F = 1},
@expr{B = -2 b c}, @expr{G = x} (the @mathit{-2} factor could just as easily
have been put into @expr{G} instead of @expr{B}), @expr{C = b}, and
@expr{H = x^2}.

The Gaussian model looks quite complicated, but a closer examination
shows that it's actually similar to the quadratic model but with an

@end tex

A last desperate step would be to use the general-purpose
@code{minimize} function rather than @code{fit}.  After all, both
functions solve the problem of minimizing an expression (the 
@texline @math{\chi^2}
@infoline @expr{chi^2}
sum) by adjusting certain parameters in the expression.  The @kbd{a F}
command is able to use a vastly more efficient algorithm due to its
special knowledge about linear chi-square sums, but the @kbd{a N}
command can do the same thing by brute force.

A compromise would be to pick out a few parameters without which the
fit is linearizable, and use @code{minimize} on a call to @code{fit}
which efficiently takes care of the rest of the parameters.  The thing
to be minimized would be the value of 
@texline @math{\chi^2}
@infoline @expr{chi^2} 
returned as the fifth result of the @code{xfit} function:

@smallexample
minimize(xfit(gaus(a,b,c,d,x), x, [a,b,c], data)_5, d, guess)

of the sum of the squares of the errors.  It then changes @expr{x}
and @expr{y} to be plain numbers, and makes @expr{z} into an error
form with this combined error.  The @expr{Y(x,y,z)} part of the
linearized model is evaluated, and the result should be an error
form.  The error part of that result is used for 
@texline @math{\sigma_i}
@infoline @expr{sigma_i} 
for the data point.  If for some reason @expr{Y(x,y,z)} does not return 
an error form, the combined error from @expr{z} is used directly for 
@texline @math{\sigma_i}.
@infoline @expr{sigma_i}.  
Finally, @expr{z} is also stripped of its error
for use in computing @expr{F(x,y,z)}, @expr{G(x,y,z)} and so on;
the righthand side of the linearized model is computed in regular
arithmetic with no error forms.

(While these rules may seem complicated, they are designed to do
the most reasonable thing in the typical case that @expr{Y(x,y,z)}
depends only on the dependent variable @expr{z}, and in fact is
often simply equal to @expr{z}.  For common cases like polynomials
and multilinear models, the combined error is simply used as the
@texline @math{\sigma}
@infoline @expr{sigma} 
for the data point with no further ado.)

@tex
\bigskip

If the lower limit is greater than the upper limit (assuming a
positive step size), the result is generally zero.  However,
Calc only guarantees a zero result when the upper limit is
exactly one step less than the lower limit, i.e., if the number
of iterations is @mathit{-1}.  Thus @samp{sum(f(k), k, n, n-1)} is zero
but the sum from @samp{n} to @samp{n-2} may report a nonzero value
if Calc used a closed form solution.

Calc's logical predicates like @expr{a < b} return 1 for ``true''
and 0 for ``false.''  @xref{Logical Operations}.  This can be

this would be a division by zero.  But at @expr{k = k_0}, this
formula works out to the indeterminate form @expr{0 / 0}, which
Calc will not assume is zero.  Better would be to use
@samp{(k != k_0) ? 1/(k-k_0) : 0}; the @samp{? :} operator does
an ``if-then-else'' test:  This expression says, ``if 
@texline @math{k \ne k_0},
@infoline @expr{k != k_0},
then @expr{1/(k-k_0)}, else zero.''  Now the formula @expr{1/(k-k_0)}
will not even be evaluated by Calc when @expr{k = k_0}.

@cindex Alternating sums

f(a b) := a f(b) :: real(a)]} is stored in variable @samp{linearF},
then the rule set @samp{[f(0) := 0, import(linearF)]} will apply
all three rules.  It is possible to modify the imported rules
slightly:  @samp{import(x, v1, x1, v2, x2, @dots{})} imports
the rule set @expr{x} with all occurrences of 
@texline @math{v_1},
@infoline @expr{v1}, 
as either a variable name or a function name, replaced with 
@texline @math{x_1}
@infoline @expr{x1} 
and so on.  (If 
@texline @math{v_1}
@infoline @expr{v1} 
is used as a function name, then 
@texline @math{x_1}
@infoline @expr{x1}
must be either a function name itself or a @w{@samp{< >}} nameless
function; @pxref{Specifying Operators}.)  For example, @samp{[g(0) := 0,
import(linearF, f, g)]} applies the linearity rules to the function
@samp{g} instead of @samp{f}.  Imports can be nested, but the

the @kbd{u g} command gets the definition of one of these constants
in its normal terms, and @kbd{u b} expresses the definition in base
units.

Two units, @code{pi} and @code{fsc} (the fine structure constant,
approximately @mathit{1/137}) are dimensionless.  The units simplification
commands simply treat these names as equivalent to their corresponding
values.  However you can, for example, use @kbd{u c} to convert a pure
number into multiples of the fine structure constant, or @kbd{u b} to
convert this back into a pure number.  (When @kbd{u c} prompts for the
``old units,'' just enter a blank line to signify that the value

All the arithmetic stores accept the Inverse prefix to reverse the
order of the operands.  If @expr{v} represents the contents of the
variable, and @expr{a} is the value drawn from the stack, then regular
@w{@kbd{s -}} assigns 
@texline @math{v \coloneq v - a},
@infoline @expr{v := v - a}, 
but @kbd{I s -} assigns
@texline @math{v \coloneq a - v}.
@infoline @expr{v := a - v}.  
While @kbd{I s *} might seem pointless, it is
useful if matrix multiplication is involved.  Actually, all the
arithmetic stores use formulas designed to behave usefully both
forwards and backwards:

In the first case, ``x'' and ``y'' are each vectors (not necessarily of
the same length); either or both may instead be interval forms.  The
``z'' value must be a matrix with the same number of rows as elements
in ``x'', and the same number of columns as elements in ``y''.  The
result is a surface plot where 
@texline @math{z_{ij}}
@infoline @expr{z_ij} 
is the height of the point
at coordinate @expr{(x_i, y_j)} on the surface.  The 3D graph will
be displayed from a certain default viewpoint; you can change this
viewpoint by adding a @samp{set view} to the @samp{*Gnuplot Commands*}

that don't have common ``x'' values.  (Of course, the range of ``x''
values covered by all the curves ought to be roughly the same if
they are to look nice on the same graph.)

For example, to plot 
@texline @math{\sin n x}
@infoline @expr{sin(n x)} 
for integers @expr{n}
from 1 to 5, you could use @kbd{v x} to create a vector of integers
(@expr{n}), then @kbd{V M '} or @kbd{V M $} to map @samp{sin(n x)}
across this vector.  The resulting vector of formulas is suitable

@code{PointStyles} variables.  These variables initially have no stored
values, but if you store a vector of integers in one of these variables,
the @kbd{g a} and @kbd{g f} commands will use those style numbers
instead of the defaults for new curves that are added to the graph.
An entry should be a positive integer for a specific style, or 0 to let
the style be chosen automatically, or @mathit{-1} to turn off lines or points
altogether.  If there are more curves than elements in the vector, the
last few curves will continue to have the default styles.  Of course,
you can later use @kbd{g s} and @kbd{g S} to change any of these styles.

For example, @kbd{'[2 -1 3] @key{RET} s t LineStyles} causes the first curve

terminals with no special graphics facilities.  It writes a crude
picture of the graph composed of characters like @code{-} and @code{|}
to a buffer called @samp{*Gnuplot Trail*}, which Calc then displays.
The graph is made the same size as the Emacs screen, which on most
dumb terminals will be 
@texline @math{80\times24}
@infoline 80x24
characters.  The graph is displayed in
an Emacs ``recursive edit''; type @kbd{q} or @kbd{M-# M-#} to exit
the recursive edit and return to Calc.  Note that the @code{dumb}
device is present only in GNUPLOT 3.0 and later versions.

@w{@samp{2 +/- 3 4 +/- 5}}, then grabbing with a prefix argument of 8
would correctly split the line into two error forms.

@xref{Matrix Functions}, to see how to pull the matrix apart into its
constituent rows and columns.  (If it is a 
@texline @math{1\times1}
@infoline 1x1
matrix, just hit @kbd{v u} (@code{calc-unpack}) twice.)

@kindex M-# :
@kindex M-# _

same limit as last time.

@key{INV GCD} computes the LCM (least common multiple) function.

@key{INV FACT} is the gamma function.  
@texline @math{\Gamma(x) = (x-1)!}.
@infoline @expr{gamma(x) = (x-1)!}.

@key{PERM} is the number-of-permutations function, which is on the
@kbd{H k c} key in normal Calc.

If you give a numeric prefix argument of 1 to @kbd{Z (}, the loop is
forced to use upward-counting conventions.  In this case, if @var{initial}
is greater than @var{final} the body will not be executed at all.
Note that @var{step} may still be negative in this loop; the prefix
argument merely constrains the loop-finished test.  Likewise, a prefix
argument of @mathit{-1} forces downward-counting conventions.

@kindex Z @{
@kindex Z @}
@pindex calc-kbd-loop
@pindex calc-kbd-end-loop

@starindex
@end ignore
@tindex mysin
A somewhat limited sine function could be defined as follows, using the
well-known Taylor series expansion for 
@texline @math{\sin x}:
@infoline @samp{sin(x)}:

@smallexample
(defmath mysin ((float (anglep x)))
  (interactive 1 "mysn")

which is not a Lisp list.

Large integers are stored as lists of the form @samp{(bigpos @var{d0}
@var{d1} @var{d2} @dots{})} for positive integers 1000000 or more, or
@samp{(bigneg @var{d0} @var{d1} @var{d2} @dots{})} for negative integers
@mathit{-1000000} or less.  Each @var{d} is a base-1000 ``digit,'' a Lisp integer
from 0 to 999.  The least significant digit is @var{d0}; the last digit,
@var{dn}, which is always nonzero, is the most significant digit.  For
example, the integer @mathit{-12345678} is stored as @samp{(bigneg 678 345 12)}.

The distinction between small and large integers is entirely hidden from
the user.  In @code{defmath} definitions, the Lisp predicate @code{integerp}
returns true for either kind of integer, and in general both big and small
integers are accepted anywhere the word ``integer'' is used in this manual.

Floating-point numbers are stored in the form, @samp{(float @var{mant}
@var{exp})}, where @var{mant} (the ``mantissa'') is an integer less than
@samp{10^@var{p}} in absolute value (@var{p} represents the current
precision), and @var{exp} (the ``exponent'') is a fixnum.  The value of
the float is @samp{@var{mant} * 10^@var{exp}}.  For example, the number
@mathit{-3.14} is stored as @samp{(float -314 -2) = -314*10^-2}.  Other constraints
are that the number 0.0 is always stored as @samp{(float 0 0)}, and,
except for the 0.0 case, the rightmost base-10 digit of @var{mant} is
always nonzero.  (If the rightmost digit is zero, the number is
rearranged by dividing @var{mant} by ten and incrementing @var{exp}.)

is displayed.  If the prefix is one, the unary function @var{unary}
is applied to the top stack element, or, if @var{unary} is not
specified, nothing happens.  When the argument is two or more,
the binary function @var{func} is reduced across the top @var{arg}
stack elements; when the argument is negative, the function is
mapped between the next-to-top @mathit{-@var{arg}} stack elements and the
top element.
@end defun

@defun calc-stack-size
Return the number of elements on the stack as an integer.  This count

modulo forms are recursively floated.  If the argument is a variable
or formula, this calls @code{reject-arg}.
@end defun

@defun compare x y
Compare the numbers @var{x} and @var{y}, and return @mathit{-1} if
@samp{(lessp @var{x} @var{y})}, 1 if @samp{(lessp @var{y} @var{x})},
0 if @samp{(math-equal @var{x} @var{y})}, or 2 if the order is
undefined or cannot be determined.
@end defun

@samp{ceil(log10(@var{n}))}, but much more efficient.  Zero is
considered to have zero digits.
@end defun

@defun scale-int x n
Shift integer @var{x} left @var{n} decimal digits, or right @mathit{-@var{n}}
digits with truncation toward zero.
@end defun

@defun scale-rounding x n
Like @code{scale-int}, except that a right shift rounds to the nearest

@end defun

@defun quarter-integer n
If @var{n} is an integer or integer-valued float, this function
returns zero.  If @var{n} is a half-integer (i.e., an integer plus
@mathit{1:2} or 0.5), it returns 2.  If @var{n} is a quarter-integer,
it returns 1 or 3.  If @var{n} is anything else, this function
returns @code{nil}.
@end defun

@node Vector Lisp Functions, Symbolic Lisp Functions, Computational Lisp Functions, Internals

input data set.  Each entry may be a single value or a vector of values.

@c 20
@item
With a prefix argument of 1, take a single 
@texline @var{n}@math{\times2}
@infoline @mathit{@var{N}x2} 
matrix from the stack instead of two separate data vectors.

@c 21
@item
The row or column number @expr{n} may be given as a numeric prefix

@c 47
@item
The variable is replaced by the formula shown on the right.  The
Inverse flag reverses the order of the operands, e.g., @kbd{I s - x}
assigns 
@texline @math{x \coloneq a-x}.
@infoline @expr{x := a-x}.

@c 48
@item
Press @kbd{?} repeatedly to see how to choose a model.  Answer the
variables prompt with @expr{iv} or @expr{iv;pv} to specify
independent and parameter variables.  A positive prefix argument
takes @mathit{@var{n}+1} vectors from the stack; a zero prefix takes a matrix
and a vector from the stack.

@c 49
@item
With a plain @kbd{C-u} prefix, replace the current region of the