Warnings, Error Messages, and Troubleshooting

Input/Output Problems MemExp will be aborted if the parameters input from a .def file are not as expected. If the number of parameters on a given line or a variable type is incorrect, an I/O error message will be printed to the screen. Check the output file to see which line of the .def file was read last, just before the program crashed. This line in the output file will begin with 'RDPARS' if the problem occured during an inversion and with 'ANALYZ' if the problem occured while analyzing a lifetime distribution. Correct this line in the .def file by referring to the documentation above and to the .def files provided as templates.

Optimization of Q is unnecessarily slow or it's inadequate. See the discussion of EPSL, EPSU, CHI2L, and CHI2U above to learn how adjusting these values speeds up or slows down the adjustment of the Lagrange multipliers. In particular, EPSU may need to be increased.

No fits by discrete exponentials. Ensure that MAXEXP $>$ 0 and that EXPMIN, EXPMAX, and EXPMN2 are compatible with the time scales of the data being analyzed.

Not enough fits by discrete exponentials. This may well indicate that your data are better described by distributed lifetimes than by discrete exponentials. Try decreasing DCHI2S to force the MEM calculation to proceed to lower values of CHI2.

Residuals not uniform in magnitude. Apparently, the iterative estimation of the standard errors has not yet converged. Increase the number of preliminary MEM inversions (the integer following 'allin1' on the command line).

The first residuals plotted are very large, and convergence is erratic. Presumably, the initial MEM inversion being done with all standard errors set to a constant should be using a larger constant. Increase the absolute value of INSIG.

ECHECK$>$ WARNING ... There are two different warnings issued from subroutine ECHECK. This subroutine checks the parameters obtained in the discrete-exponential fits. A warning is issued if the amplitude of any exponential is smaller than the noise in the data at time values near the exponential's lifetime. A different warning is issued if the lifetimes of two exponentials are very similar. Either warning implies that too many exponentials are being used to fit the data at this poin WIDTH="61" HEIGHT="34" ALIGN="MIDDLE" BORDER="0" SRC="img82.gif" ALT="$\vert A\vert < \sigma$">' or 'Log $\tau$ too close!'

GTBLUR$>$ WARNING! NBHMAX exceeded ... The blurring of $\vec{f}$ into $\vec{F}$ was attempted with a Gaussian too broad for the maximum array sizes. The FWHM of the blurring Gaussian is automatically reduced appropriately.

NEWLAG$>$ BB-4AC ... These messages from subroutine NEWLAG may or may not signify a problem. When updating the Lagrange multipliers during a MEM inversion, a quadratic equation is solved to limit the updates. If the solutions to this quadratic equation are not real, this error message is written to the output file and the Lagrange multipliers are not updated. One of these messages once in a while is probably fine. If a multitude of these messages is printed and convergence appears to be stalled, try reducing EPSL and/or EPSU. This slows the multiplier updating and should solve this problem.