Parameters for Analysis of the Lifetime Distributions

A second .def file controls the analysis of a MEM image stored during a previous inversion. This analysis is one or more of the steps performed when MemExp is run in allin1 mode. The fit and weighted residuals corresponding to the image are written to files in free format (two columns of numbers without a header). Also, the means and areas of peaks are automatically determined.

Remember that running MemExp in simple mode produces the '.ana' file that contains what are likely to be acceptable values for these analysis parameters.

The first line, WRFIT, should be T for true. The fit and weighted residuals will then be written (format-free) to files.

EXTRAP data points are added to the beginning and to the end of the data when evaluating the fit. Use this option to view the kinetics predicted by $f(log \tau)$ that lies outside the experimentally sampled times.

The next several lines specify variables whose meanings are the same as for inversions. See above for descriptions of IOFORM, ITAKE, XSCALE and YSCALE, MLOG, IGNORE, INSIG, TBASLN, RESPON, t1 and t2, D1 and D2, and IBIGF. If IBIGF = 2, then the name of the file containing the prior model to be used is read from the next line. Otherwise, the value of the constant to be used as the uniform prior is read from the next line.

INTGRT is an integer specifying the number of log-tau intervals over which to integrate $f(log \tau)$. The next INTGRT lines give the upper and lower bounds for each interval.

NSMOTH and ISMOTH are integers that determine how the standard errors are to be estimated from the deviations of the MEM fit being analyzed from the data. These error estimates can then be used in a subsequent MEM inversion. If ISMOTH $< 0$, then all standard errors are set to the average of the rms deviations. If NSMOTH $> 0$, then the rms deviations between the data and the MEM fit are smoothed by convolution with a bell curve (if ISMOTH $=0$) or a uniform distribution (if ISMOTH $> 0$) that ranges over NSMOTH points. (See PFIT in the next paragraph).

The remaining parameters govern the estimation of the errors from the deviations between the data and the MEM fit being analyzed. At each data point, sigma is estimated as the root mean square deviation of 2 PFIT + 1 values. IF NSMOTH $\leq 0$, then these estimated sigmas are fit to a piecewise polynomial to obtain error estimates that vary smoothly with time. Otherwise, they are smoothed as described in the previous paragraph.

Next come the fit parameters (needed only if NSMOTH $\leq 0$), each followed by a 0 or 1 to indicate the parameter is to be fixed or varied, respectively. The errors are generally fit to a piecewise cubic in log time. A0, A1, A2, and A3, are the constant, linear, quadratic, and cubic coefficients in log time for log t $<$ T. B2 and B3 are the quadratic and cubic coefficients and times log t $>$ T. Note that the constant and linear coefficients at long times are determined by the requirement that the function and derivative of the fit be continuous. If A2 and A3 are set to zero, then the fit is to a piecewise linear function, B2 represents the linear coefficient at long times, and only the function is continuous.