.TH SPLINE 1 "Dec 1998" "FSF" "GNU Plotting Utilities"
.SH NAME
spline \- interpolate datasets using splines under tension
.SH SYNOPSIS
.B spline
[
.I options
] [
.I files
]
.SH DESCRIPTION
.LP
.B spline
reads datasets from standard input or from one or more files, and
fits a smooth curve (a "spline") through each dataset.
An interpolated version of each dataset, consisting of points from the
smooth curve, is written to standard output.
.LP
Unless the
.B \-a
or
.B \-A
options are used (see below), each dataset should be a sequence of values
for a vector-valued function of a single scalar variable.
That is, each dataset should be a sequence of data points,
given as alternating \fIt\fP\^
and \fIy\fP\^ values.
\fIt\fP\^ is a scalar independent variable, and
\fIy\fP\^ is a vector-valued dependent variable.
The dimensionality of
\fIy\fP\^ is specified with the
.B \-d
option (the default dimensionality is 1).
Between each data point
and the next, \fIt\fP\^ should increase.
.LP
An input file may contain more than a single dataset.
If an input file is in
.SM ASCII
format (the default), its datasets should be separated by blank lines.
The \fIt\fP\^ and \fIy\fP\^ values of the data points in each dataset
may be arranged arbitrarily, so long as they are separated by white space.
Besides datasets, an input file may contain any number of
comment lines, which should begin with the comment character `#'.
Comment lines are ignored.
They are not treated as blank, i.e.,
they do not interrupt a dataset in progress.
.LP
Options and file names may be interspersed on the command line, but
the options are processed before the file names are read.
If
.B \-\-
is seen, it is interpreted as the end of the options.
If no file names are specified, or the file name
.B \-
is encountered, the standard input is read.
.LP
The type of interpolation, and the format of the input and output files,
may be selected by command-line options.
.SH OPTIONS
.SS "Interpolation-Related Options"
.TP
.B \-f
.br
.ns
.TP
.B \-\-filter
Use a local interpolation algorithm (the cubic Bessel algorithm), so that
.B spline
can be used as a real-time filter.
The slope of the interpolating curve at each point in a dataset will be
chosen by fitting a quadratic function through that point and the two
adjacent points in the dataset.
If
.B \-f
is specified then the
.B \-t
option, otherwise optional, must be used as well.
Also, if
.B \-f
is specified then the \fB\-k\fP, \fB\-p\fP, and \fB\-T\fP options
may not be used.
.IP ""
If
.BR \-f
is \fInot\fP\^ specified, then the default (global) interpolation algorithm
will be used.
.TP
.BI \-k " k"
.br
.ns
.TP
.BI \-\-boundary\-condition " k"
Set the boundary condition parameter for each constructed spline to be
.IR k .
(The default value is 1.0.)
In each of its components, the spline will
satisfy the two boundary conditions y"[0]=ky"[1] and y"[n]=ky"[n-1].
Here y[0] and y[1] signify the values of a specified component of the
vector-valued dependent variable \fIy\fP\^ at the first two points of a
dataset, and y[n-1] and y[n] the values at the last two points.
Setting \fIk\fP\^ to zero will yield a "natural" spline, i.e., one that has
zero curvature at the two ends of the dataset.
The \fB\-k\fP option may not be used if \fB\-f\fP or \fB\-p\fP is specified.
.TP
.BI \-n " n"
.br
.ns
.TP
.BI \-\-number\-of\-intervals " n"
Subdivide the interval over which interpolation occurs into \fIn\fP\^
subintervals.
The number of data points computed, and written to the
output, will be
.IR n+1 .
The default value for \fIn\fP\^ is 100.
.TP
.B \-p
.br
.ns
.TP
.B \-\-periodic
Construct a periodic spline.
If this option is specified, the \fIy\fP\^
values for the first and last points in each dataset must be equal.
The
\fB\-f\fP and \fB\-k\fP options may not be used if \fB\-p\fP is specified.
.TP
.BI \-T " tension"
.br
.ns
.TP
.BI \-\-tension " tension"
Each interpolating curve will be a spline under tension.
This option sets the tension value (the default is 0.0).
.IP ""
If \fItension\fP\^ equals zero, the curve will be a piecewise cubic spline.
Increasing the tension above zero makes the curve "tighter", and
reduces the likelihood of spurious inflection points.
That is because between each pair of successive points in a dataset,
the curve will satisfy the fourth-order differential equation
y""=sgn(\fItension\fP\^)*(\fItension\fP\^^2)y" in each of its components.
As \fItension\fP\^ increases to positive infinity,
it will converge to a polygonal line.
The \fB\-T\fP option may not be used if \fB\-f\fP is specified.
.TP
.B \-t \fItmin tmax [tspacing]\fP
.br
.ns
.TP
.B \-\-t\-spacing \fItmin tmax [tspacing]\fP
For each dataset, set the interval over which interpolation occurs
to be the interval between
\fItmin\fP\^ and
.IR tmax .
If \fItspacing\fP\^ is not specified, the interval will be divided into the
number of subintervals specified by the \fB\-n\fP option.
.IP ""
If the \fB\-t\fP
option is not used, the interval over which interpolation occurs will be
the entire range of the independent variable in the dataset.
The \fB\-t\fP
option must always be used if the \fB\-f\fP option is used to request
filter-like behavior (see above).
.SS "Format-Related Options"
.TP
.BI \-d " dimension"
.br
.ns
.TP
.BI \-\-y\-dimension " dimension"
Set the dimensionality of the dependent variable
.IR y " in"
the input and output files to be
.IR dimension .
The default dimension is 1.
.TP
.BI \-I " data-format"
.br
.ns
.TP
.BI \-\-input\-format " data-format"
Set the data format for the input file(s) to be
.IR data-format ,
which may be one of the following.
.RS
.TP
.B a
.SM ASCII
format (the default).
Each file is a sequence of floating point numbers, interpreted as the
\fIt\fP\^ and \fIy\fP\^ coordinates of the successive data points in a dataset.
If \fIy\fP\^ is \fId\fP\^-dimensional,
there will be \fId+1\fP\^ numbers for each point.
The \fIt\fP\^ and
\fIy\fP\^ coordinates of a point need not appear on the same line, and
points need not appear on different lines.
But if a blank line occurs (i.e., two newlines in succession are seen), it
is interpreted as the end of a dataset, and the beginning of the next.
.TP
.B f
Single precision binary format.
Each file is a sequence of floating point
numbers, interpreted as the \fIt\fP\^
and \fIy\fP\^ coordinates of the successive data
points in a dataset.
If \fIy\fP\^ is \fId\fP\^-dimensional,
there will be \fId+1\fP\^ numbers for each point.
Successive datasets are separated by a single occurrence of the quantity
.SM FLT_MAX,
which is the largest possible single precision floating point number.
On most machines this is approximately 3.4x10^38.
.TP
.B d
Double precision binary format.
Each file is a sequence of double
precision floating point numbers, interpreted as the \fIt\fP\^
and \fIy\fP\^ coordinates of the successive data points in a dataset.
If \fIy\fP\^ is \fId\fP\^-dimensional, there will
be \fId+1\fP\^ numbers for each point.
Successive datasets are separated by a single occurrence of the quantity
.SM DBL_MAX,
which is the largest possible double precision floating point number.
On most machines this is approximately 1.8x10^308.
.TP
.B i
Integer binary format.
Each file is a sequence of integers, interpreted as
the \fIt\fP\^ and \fIy\fP\^ coordinates of the successive data points
in a dataset.
If \fIy\fP\^ is \fId\fP\^-dimensional,
there will be \fId+1\fP\^ numbers for each point.
Successive datasets are separated by a single occurrence of the quantity
.SM INT_MAX,
which is the largest possible integer.
On most machines this is 2^31\-1.
.RE
.TP
.B \-a \fI[step_size [lower_limit]]\fP
.br
.ns
.TP
.B \-\-auto\-abscissa \fI[step_size [lower_limit]]\fP
Automatically generate values for
.IR t ,
the independent variable (the default values of \fIstep_size\fP\^ and
\fIlower_limit\fP\^ are 1.0 and 0.0, respectively).
.IP ""
Irrespective of data format (`a', `f', `d', or `i'), this option specifies
that the values of \fIt\fP\^ are missing from the input file: the
dataset(s) to be read contain only values of
.IR y ,
the dependent variable.
So if \fIy\fP\^ is \fId\fP\^-dimensional, there will be
only \fId\fP\^ numbers for each point.
The increment from each \fIt\fP\^ value to the next will be
.IR step_size ,
and the first
\fIt\fP\^ value
will be
.IR lower_limit .
This option is useful, e.g., when interpolating curves rather than functions.
.TP
.B \-A
.br
.ns
.TP
.B \-\-auto\-dist\-abscissa
Automatically generate values for
.IR t ,
the independent variable.
This is a variant form of the \fB\-a\fP option.
The increment from each \fIt\fP\^ value to the next will be the distance in
\fId\fP\^-dimensional space between the corresponding \fIy\fP\^ values, and
the first \fIt\fP\^ value will be 0.0.
That is, \fIt\fP\^ will be "polygonal arclength".
This option is useful when interpolating curves rather than functions.
.TP
.BI \-O " data-format"
.br
.ns
.TP
.BI \-\-output\-format " data-format"
Set the data format for the output file to be
.IR data-format .
The interpretation of \fIdata-format\fP\^ is the same as for the
\fB\-I\fP option.
The default is `a', i.e.,
.SM ASCII
format.
.TP
.BI \-P " significant-digits"
.br
.ns
.TP
.BI \-\-precision " significant-digits"
Set the numerical precision for the \fIt\fP\^ and \fIy\fP\^ values in the
output file to be
.IR significant-digits .
This takes effect only if the output file is written in `a' format,
i.e., in
.SM ASCII.
\fIsignificant-digits\fP\^ must be a positive integer (the default is 6).
.TP
.B \-s
.br
.ns
.TP
.B \-\-suppress\-abscissa
Omit the independent variable \fIt\fP\^ from the output file; for each
point, supply only the dependent variable
.IR y .
If \fIy\fP\^ is \fId\fP\^-dimensional, there will be only \fId\fP\^ numbers for
each point, not
.IR d+1 .
This option is useful when interpolating curves rather than functions.
.SS Informational Options
.TP
.B \-\-help
Print a list of command-line options, and exit.
.TP
.B \-\-version
Print the version number of
.B spline
and the plotting utilities package, and exit.
.SH EXAMPLES
.LP
Typing
.LP
.RS
.B echo 0 0 1 1 2 0 \||\| spline
.RE
.LP
will produce on standard output an interpolated dataset consisting of
101 data points.
If graphed, this interpolated dataset will yield a parabola.
.LP
It is sometimes useful to interpolate between a sequence of
arbitrarily placed points in \fId\fP\^-dimensional space,
i.e., to "spline a curve" rather than a function.
The
.B \-a
and
.B \-s
options are used for this.
For example,
.LP
.RS
.B echo 0 0 1 0 1 1 0 1 \||\| spline \-d 2 \-a \-s
.RE
.LP
will produce on standard output a 101-point dataset that interpolates
between the four points (0,0), (1,0), (1,1), and (0,1).
The
.B \-d 2
option specifies that the dependent variable \fIy\fP\^ is two-dimensional.
The
.B \-a
option specifies that the \fIt\fP\^ values are missing from the input
and should be automatically generated.
The
.B \-s
option specifies
that the \fIt\fP\^ values should be stripped from the output.
.SH AUTHORS
.B spline
was written by Robert S. Maier (\fBrsm@math.arizona.edu\fP),
starting with an earlier version by Rich Murphey (\fBrich@freebsd.org\fP).
The algorithms for constructing splines under tension are similar
to those used in the FITPACK
subroutine library, and are ultimately due to
Alan K. Cline (\fBcline@cs.utexas.edu\fP).
.SH "SEE ALSO"
"The GNU Plotting Utilities Manual".
.SH BUGS
Email bug reports to
.BR bug\-gnu\-utils@gnu.org .