The most common problem is failure of a brand-new model to converge, and the most common causes are order-of-magnitude typos in the input file and a premature attempt to run DELTAE with too many variables in the guess and target vectors. The easiest way to fix such a problem is to Clear everything from the guess and target vectors, run DELTAE, and display the .dat file. Often it is obvious where your typo is-one of the output variables will go wild in a supposedly innocuous segment. If not, examine the results more closely for reasonableness. Modify suspicious variables a little, to see what effect they have on results. Try to get the model close to converging on your desired targets just by modifying your desired guesses one at a time, manually. Then add one guess-target pair at a time, running DELTAE each time, examining the results, and manually modifying your other desired guess variables to try to keep your other desired target variables under control. For further diagnostic information, try using the Nprint variable, found under the (T)olerances/debugging menu described in the previous section and in Chapter IV.
It is also useful to keep the model as simple as possible. Examples:
The solver within DELTAE can sometimes become stuck around a local minimum, particularly if you are making incremental changes from a model that has already converged-and often, the internal representation of the `best guess' does not agree with what we would like for a given model. Try manually changing one of the guess vector members slightly and see if DELTAE will loose its fondness for this particular point. Or, change one of the members of the guess vector, if you can think of an appropriate alternate.
If these steps fail, consider some of the options in (T)olerances/debugging. Some pathologically difficult cases converge better with tighter tolerance, alternate normalization mode, or one of the other tuning options described at the end of Chapter IV.
Always check results carefully for reasonableness, particularly when calculating complicated models or using any of DELTAE's more elaborate features. While DELTAE is a useful tool, it is far from foolproof; for example, the shooting method can easily end up generating devices that are several wavelengths long, if initial convergence is slow. All INS-type segments, TBRANches and UNIONs containing thermoacoustic elements also deserve special skepticism.