Settings
Settings tab
On this page, settings, options, and tuning parameters for the fitting procedures can be modified.
(Note: the help here is stale and significant changes have been made recently. Refer to the context help for details (press ctrl+h and hover over any control/indicator))
Data Noise Determination
To determine the estimated data noise, a "quiet range" of the raw dipolar evolution is compared with the back-calculated Tikhonov fit. (Note that the noise after background correction is time-dependent and not useful.). The alpha used can be adjusted here. The noise variance and standard deviation based on the current data and noise alpha is shown.
Save Options
Pressing “save” will generate a set of output files as described here. If “Also Print” is checked, a hardcopy of the fitting results will be printed to the default printer. Make sure a printer is available. If there is no valid fit result, the Tikhonov regularization is saved.
Tikhonov Regularization
Tikhonov regularization is implemented in linear algebra. In the most simple form, it allows negative distance probabilities. At a slight computing cost, the regularization can be forced to produce all positive results. This often cleans up the result. The default is non-negative but can be changed if desired.
(Note that background determination requires the potentially negative values to help gauge the background quality. For the background determination, non-negative is automatically disabled)
Fitting Options
Here you can find tuning parameters for the various fitting algorithms. The defaults are carefully selected and should not be changed without getting my advice. (This means you, Lukasz!)
Model Free Fitting
Show progress: By default, a window appears during fitting to observe the progress. This window can be disabled if needed. (The computational cost of the window is negligible and it is useful to keep it visible).
Guess: Levenberg-Marquardt requires an initial guess for the distance distribution. By default ("Tikhonov Guess") the fitting will use a cleaned up digest of the distance distribution from the plain non-negative Tikhonov regularization as initial guess for the distance distribution. If disabled, a flat guess for the initial distance distribution is used. It typically does not make a difference which guess is used. If there are certain doubts, flat guess should be selected, because it makes no assumptions.
Deriv: Determines the smoothness constraint, basically the difference between zero and a derivative of the magnitude of the distance distribution scaled by the "smoothness" parameter. The second derivative (derive=2) is the default, but anything from 0 ..9 can be freely selected to allow experimentation.
Refine background (experimental feature*): Enabling this will allow co-fitting of the background during a model-free fit.
Model Based Fitting
(Algorithm options similar to model-free as described below)
Levenberg Marquardt options (separate settings for model free and model based).
Algorithm: Three settings for lambda (start, scale up, scale down). Changing these is not recommended but they can be used to modify the convergence behavior from pure steepest descent to pure conjugate gradient and to anything in between. By default, Levenberg-Marquard will scale the algorithm according to the up/down settings. There are six different implementations of the algorithm, "stock" is the National Instruments implementation that ships with LabVIEW. The default is "CA Hybrid Par", which uses parallelization and uses a hybrid lambda update that better protects from parameter correlations. Note that the model always calculates the partial derivatives in parallel (the main effort!) and the additional parallelization of the minimization algorithms only provides a small additional advantage.
Fitting Termination
Defines when the fitting should give up based on the number of iterations and tolerance. Typically a minimum is found quickly, so these settings are irrelevant. If the fitting terminates because this condition is reached, the best fit found so far is shown. Similarly, fitting can be terminated at any time by pressing the <esc> key.
Simulation Setup
To more closely simulate data that resembles your existing data, You can first load real data, find the background, and copy and use the real settings (e.g. number of DE points, point spacing, etc.) for the simulation.
Advanced
The program contains some advanced or experimental features that might be confusing or unstable. Enabling experimental features will expose additional controls that can be used at your own risk. Beware!
Currently available experimental features are:
- Denoising options (Select tab)
- Background fit criteria [L, H, L&H] (background tab)
- Rotation setting (L-curve tab)
LabVIEW™ Is a trademark of National Instruments Corporation.