- Fixed bug in MD routines where multiple roughnesses and normalization fluctuation was specified
- Fixed normalization constant was not respected by the Stochastic MD output window
- Fixed small graphing bug
- Updated random number generator for MD routines to Mersenne twister
- Should now work on XP/Vista 64
Version 1.6.3
- Fixed stochastic fit not working properly
Version 1.6.1
- Graphing fixes
- SLD only fixes
- Slight speed increase in MD routines
- Fixed occasional hang (hopefully)
- Internal Changes to improve stability
- Initial parameter step value (parameter temperature) can now be selected
- SLD can now be used instead of normalized electron density
- Added data patching
Version 1.5
- versions 1.3 and 1.4 were internal
- Updated levmar to version 2.2 - several very small bugs fixed
- Updated itextsharp to version 4.0.8
- Added ED graph for model dependent fitting
- Fixed pdf report and added more information for model dependent fits
- Mastergraph now has publication quality graphs with transparency for overlaying
- default points per Angstrom is now 3
- default amount of subphase length is 40
- default fitness function is log error
- Fixed incorrect absorption calculation
- added option for debugging
- added adaptive ability to STUN to the GUI
- separated STUN and SA dialogs
- choice of tunneling algorithms
- numerous small bug fixes
- Can resume between runs
- Improved error handling
- Fitting from previous versions cannot be resumed.
- calculation ~ 21% faster for absorbing films and ~ 12% faster for non absorbing under standard parameters
- larger super and subphases for very rough films affect the calculation less
- Fixed SLD calculator error with using bulk densities (thanks Luka)
- Added XR only option to reject fits with negative electron density
- Negative points in the data file are now ignored, and those files now load
- Fixed problem with loading the data files when there are extra lines at the end
- fitness scores are now divided by the number of data points
- fixed localization issues. Now our European, South American, and Indian friends can fit data without problems
- fixed graph issues
- cleaned up statistical error issues
- cleaned up file names and locations. Instead of deleting old fits, they are moved to a new folder
Version 1.2
- IMPORTANT - You will not be able to resume runs created with previous versions.
- Renamed the project to StochFit. This better captures the nature of the algorithms.
- Implemented consistent Chi square values throughout StochFit. Previously the Chi square reported in the model independent screen was the fit function, which was not necessarily a Chi square . The reflectivity Chi square values are reduced Chi squares of R/Rf. The Chi square of the electron density profile fit is just a least squares difference as there are no errors in that case.
- Improved graph zoom behavior when errors are very large.
- More options for the objective function in the model independent fit.
- Updated pdf and graphing library libraries.
- Added Simulated Annealing and Stochastic Tunneling optimization algorithms for model independent fits. These are slower but more effective for finding alternate solutions
- Small optimization algorithm improvements including the ability to easily resume from annealing
- Increased speed of the model independent algorithm (depending on options and hardware) between 15 and 300 percent.
- Better random number generation and behavior.
- Fixed stochastic model dependent fit crash on multi processor systems.
- Improved cutting and pasting of graphs
Version 1.1
- Fixed pdf report parameters to be consistent
- Added Levenberg-Marquardt fit convergence details
- Fixed error estimation problem
- Model dependent box fit with constraints is now available
Version 1.0 - Initial Release