IV.H Tuning and Debugging

The (T)olerances/debugging menu selection gives the user access to a number of internal parameters that control the quantity of output and diagnostic information generated and the way that the solver approaches the iterations it will perform. An explanation of these parameters is given below:

Nprint

If Nprint <= 0, the .dat file will contain only the final converged iteration of the model. Otherwise, DELTAE saves every Nprintth intermediate iteration. If N >= 0, intermediate steps in every stack integration are included in the data file. For Nprint > 0, every segment is displayed to the screen (equivalent to typing the .out file. This can be useful in finding model errors that cause DELTAE to crash before the first converged data point is ever stored. If Nprint < 0, a concise iteration summary line is printed, containing only the iteration number and the root-mean-square sum of the errors (targets - results), and the line will overstrike itself. If Nprint >= 0, the iteration count and the complete guess and (target - result) vectors are displayed on sequential lines. Default: -1.

PlotDat

This variable controls output generated during plots, where multiple solutions are processed sequentially. If PlotDat >= 0, all error messages are that occur when DELTAE has doubts about the convergence of a datapoint are announced (on a MacIntosh, `OK' must be clicked in the alert box before calculations will continue. For other values of PlotDat, DELTAE will continue silently, but will still write the messages to the .dat file and mark the lines in the .plt file with a `*.' If PlotDat >= 1, all converged endpoints are written to the .dat file (it can become quite large!). For PlotDat = 0, only the most recent is kept. Default: 0.

tolerance

Recommended range: 1.5e-7-1.e-2. This value governs the point at which DELTAE considers its iterations finished. The default value is close to the limit that can be reached using single-precision arithmetic (all DELTAE calculations are double precision). This value does not relate directly to errors between any particular result and its target value; it concerns changes in the norm of the error vector. Default: 3.00E-04.

Runge-Kutta steps

This is an even integer that determines the number of integration steps used to span each stack-type segment (slab, circular, or duct). It does not affect other segment types. Larger values will cause a slower, more accurate computation. Small values will increase speed at the price of integration accuracy, but may cause convergence problems if the specified tolerance is too small. Output from every other step can be enabled with the Nprint parameter. Default: 10.

Normalization mode

In a numerical problem where all of the input variables in the guess vector and all of the output variables used in the target vector are of wildly different magnitudes, a difficulty arises in choosing how much to change each variable and how much to weigh the errors between the target and the result values. Particularly, this affects HARD- and SOFTEnd segments. A 0.01 K error in a heat exchanger temperature is fairly benign to us, but in the complex end impedance, an error of 0.01 could leave us with hundreds of watts of power flow where there must be zero. In the standard mode (1), DELTAE uses the solver's internal method to normalize the solution vector, which usually does a reasonable job. For pathological cases, DELTAE has a special mode (2) that tries to normalize all input variables and output variable differences to unity. This can present its own problems, however, since we do not know how to normalize zero input variables (phases are a special case, automatically normalized by 360 degrees). In normalizing outputs, problems can occur if the model is very far from being converged, giving large initial error values; if it is very close to being converged, the errors could be near zero, presenting the other problem. Use mode 1 whenever possible, and mode 2 when you must. It may sometimes help to specify a zero input variable as some tolerably small real number when using this mode. Default: 1.

Step bound factor

recommended range: 0.01-100. This value regulates the size of initial excursions DELTAE makes from initial guesses to find the directions in which it must iterate. Some difficult cases can benefit by reducing this value. Default: 100.0.

FCNerr

There is a limit to the accuracy with which a computer can calculate the `function' that represents one complete pass through a model. The assumed value of this error affects the increments between iterations that the solver will choose; if the increments are too small, the effect on the result will be unpredictable. Larger values of FCNerr can speed iterations, with a less accurate endpoint. Too small a value can cause the solver to lose its way completely. This quantity is system-dependent, and it may be necessary to increase it slightly for very complex models. Recommended range: >5.e-15. Default: 1.00E-10.