simple mode

Here are five examples of running MemExp in simple mode. Parameters are specified on the command line in the following order, with those in parentheses needed only for certain values of a preceding parameter.

Parameters: mode data IGNORE IRF (LTSCAT TSHIFT (DTSHFT DTMULT) VARYTS) D0 %drop Nd PPLT Nb FAMIN IBIGF (CHI2up LTcut)

memexp-6.0.exe simple dir1/data.001 1 none 1. 1.0 2 40 1 5.0D-4 0

memexp-6.0.exe simple dir2/data.002 1 none 1. 1.0 1 40 0 5.0D-4 0

memexp-6.0.exe simple dir3/data.003 2 dir3/irf.003 0 99. 0.2 0.9 1 1. 1.0 1 40 0 5.0D-4 0

memexp-6.0.exe simple dir4/data.004 1 none 1. 1.0 2 40 0 5.0D-4 0

memexp-6.0.exe simple dir4b/data.004 1 none 1. 1.0 2 40 0 5.0D-4 3 0.813 -1.1

The only file(s) input to MemExp in simple mode are the kinetics data to be analyzed and the instrument response function, if applicable. Other than those specified on the command line, parameters that are used to invert the data and analyze the distribution are set automatically in simple mode. All parameters are then written to files named for the data, e.g., dir1/data.001.mem and dir1/data.001.ana, respectively. If the results obtained in simple mode need to be fine-tuned for some reason, the '.mem' file and '.ana' file can be edited and input to MemExp in auto or allin1 mode.

The first argument following the program name on the command line specifies the mode and is set to simple.

The second argument is the name of the data file to be inverted and may include the path of the file (the subdirectory where it is located).

The third argument (IGNORE) is an integer that specifies how to treat the errors in the data. IF IGNORE = 0, the standard errors of the mean specified in the third column of the data file are used. If IGNORE = 1, one preliminary MEM run is used to estimate the standard errors, and a second MEM run is performed using these errors (e.g., dir1/data.001_e1.out.ps, dir2/data.002_e1.out.ps, dir4/data.004_e1.out.ps, and dir4b/data.004_e1.out.ps). If IGNORE = 2, the analysis assumes the data are governed by Poisson statistics (e.g., dir3/data.003.out.ps). If IGNORE = 3, use time-independent errors; do not do a prelimnary MEM calculation to estimate errors.

The fourth argument is either set to 'none' or to the name of the known instrument response to be deconvolved from the kinetics. Only in the latter case are an integer value of LTSCAT and a real value of TSHIFT specified next. Set LTSCAT to 1 to use a scattered-light correction; set LTSCAT to 0 to omit this term. Set TSHIFT to zero to forego use of a zero-time shift ($\delta$ in equation 2) between the data and IRF. If TSHIFT $\ge 90$ the program determines $\delta$ using preliminary MEM calculations, and TSHIFT is followed by two additional real values, DTSHFT and DTMULT. DTSHFT is the fraction of the time resolution to be used as the step size to initiate the estimation of $\delta$ via Brent's method. DTMULT favors values of $\delta$ that result in the first appreciable MEM peak having a mean greater than DTMULT times the time resolution, i.e., DTMULT ($t_2 - t_1$). The next argument is VARYTS, an integer. Set VARYTS to 1 to vary $\delta$ in the discrete fits; set VARYTS to 0 to fix $\delta$ in the discrete fits.

The next argument is a real number, D0, the data's normalization constant (equations 1, 2, 10, and 11). The normalization of the fit is ultimately determined by the product of D0 and the integral of the lifetime distribution. When deconvolving an instrument response, an input value of D0 = -1.0 will be reset to the maximum data value. (Warning: Severely underestimating the normalization can induce spurious maxima in f [10].)

The next argument is PCDROP, the per-cent decrease in the goodness-of-fit statistic to be used to select the recommended discrete fit.

The next argument is NDIST, the number of distributions to be used. Set NDIST to 1 for monotonic kinetics or 2 for kinetics that rise and fall.

The next argument is PPLT, the number of points per log unit used in the distribution(s). Try PPLT = 40. Of course, the computational cost increases with PPLT.

The next argument is NBASLN, the number of baseline parameters to be used, ranging from 0 (for no baseline) to 4 (for a cubic baseline). In most cases, NBASLN will be set to 0 or 1 (for a constant baseline). NBASLN can also be set to -1 (new to version 5.0) to signify that a constant baseline is to be fixed at the value OFFSET, which is then read as the next argument. OFFSET is only specified when NBASLN = -1.

The next argument is FAMIN. If set to a non-negative value on the command line in simple mode, FAMIN is the minimum area of a MEM peak to be accounted for when initializing parameters in the fits by discrete exponentials. FAMIN is specified as a fraction of the area of the largest MEM peak, e.g., 0.0005. When set to a negative value, the program sets this threshold.

The next argument (IBIGF) is an integer that determines how the 'prior model' is to be treated in the definition of the entropy to be maximized (equation 4). A uniform model is used when IBIGF = 0. When IBIGF = 1, 2, 3, or 4, the prior model is derived from the lifetime distribution obtained using a uniform model when $C$ (equation 5 or 6) has reached CHI2UP (the next argument, needed only when IBIGF is nonzero). When IBIGF = 3 or 4, features peaked at log $\tau < LTCUT$ (the next argument, needed only when IBIGF = 3 or 4) are omitted from the prior model.

To see how this option can suppress unwarranted peaks at short lifetimes while improving the means of the remaining peaks [2], compare the two analyses of data.004 using a uniform prior (dir4/data.004_e1.out.ps) and a prior adapted from the kinetics with IBIGF = 3 (dir4b/data.004_e1.out.ps).

See the documentation below for more discussion of IBIGF and the other parameters.