This window is used to display the
results of line position, line intensity, energy level or contour fits and will also display
details of calculations if the PrintLevel
is positive. It will typically appear automatically, or bring it
up with View, Log. For line position and
intensity fits, a line list giving details of each fitted
rotational line, with an observed and calculated frequency or
intensity and obs-calc for each line, is also given. For a graph
of the residuals, see the Residuals
Window. Note that the information displayed applies to one
fit cycle; as a non-linear fitting process is used expect to fit
several times until convergence is obtained. Click on an item in
the picture below for more information:
|nobs = The number of
lines fitted or the number of experimental points for a
|npar = The number of parameters
|Initial Average Error
|This is [Σ[(obsi-calci)/wi]2/(nobs-npar)]½
with the calculated values, calci, obtained using
the parameters at their initial values. wi
are the estimated (relative) standard deviations of the
|Predicted New Error
||This is [Σ[(obsi-calci)/wi]2/(nobs-npar)]½
with the calculated values, calci, obtained from
the non-linear least squares fit. They are typically close
to the values that would be obtained from the new
parameters, but are only exact if the calculated values are
linear (or nearly linear) in the parameters. This is should
be the case near to convergence, but often not when starting
the fitting process. Fit again to find the true new error.
|The parameter value at the
start of the fit.
|The parameter value after the
|The estimated standard
deviation of the parameter, based on the quality of the fit.
|(New-Old)/Std Dev - the relative change in
parameter (J. K. G. Watson, J. Mol. Spectrosc. 66, 500 (1977)) for the
parameter. This is the change in the parameter that would
make the average error of the fit increase by a factor of
and provides a useful guide as to how many figures should be
quoted for the parameter to ensure that the calculation can
be reproduced. Where parameter correlation is high, this can
be many more figures than suggested by the standard
deviation of the parameter. See also R. J. Le Roy, J. Mol.
Spectrosc. 191, 223 (1998) for discussion of this
issue, and alternative approaches.
|This column and the next are only shown if
"Check Derivatives" is selected. This prompts for a
multiple, c, and then for each parameter calculates
a maximum difference in derivatives using the default
increment, and the increment multiplied by c.
Formally the derivative of a calculated value, y,
with respect to a parameter, p, is worked out using:
dy/dp = (y(p+i)-y(p))/i
where i is the increment for the parameter. The
value displayed here is the largest difference between
derivatives calculated with increments i and i×c,
looking over all observations. It is divided by the largest
derivative for that parameter, again looking at all
|The observation number of the largest
derivative difference shown in the previous column.
|The new parameter, with the
one standard deviation in units of the last figure in
brackets. The sensitivity is used to determine how many
figures are displayed for the standard deviation.
||The correlation between the
||Do fit of indicated type. For
fits other than a Contour fit, the experimental information
is taken from the file set at the top of the window.
Alternatively, you can take the information from the
clipboard by right clicking on the fit button. (This works
well if you prefer to use Excel or another spreadsheet
program to manage the experimental line list; before each
fit select the required area of the spreadsheet and copy it
to the clipboard.)
||This pops up a menu providing the following
- Fit Once -
Do one fit cycle (The fit button above does FitCyles fits,
though the default is one)
- Fit Five
Times - Do five fit cycles.
Derivatives - Do one fit cycle, showing extra
information - see "Deriv Diff" above.
- Do a complete fit cycle, including print-out, but do
not float any parameters.
||Step constants back one fit
cycle for each press
||Step constants forward one
||Click to browse for
experimental file name
||Edit given file; this button
will normally start the Slave
||Clear log window
||Copy log to clipboard.
||Save log to file.
|Select amount of log output on simulation or
fitting. See PrintLevel
for the possible values.