Graphical Interface for Global Optimization Algorithms

To know where to begin, make sure you have read the tutorials first !

algorithm.jpg

Algorithm Options

Algorithm

You can choose between Parallel Tempering (highly recommended) and Simulated Annealing (use this only if you know how... you'd have to carefully choose the number of trials and temperature schedule). Normally, the default choice (Parallel Tempering) should not be changed.

Temperature

These are the minimum and maximum temperatures to ensure that the algorithm will search all possible configurations, while insisting on the best configurations. It is highly recommended not to change the default choices, with a "smart" temperature schedule which will be tuned by the algorithm.

If you choose a different schedule than 'smart', then the the meaning of the max, min and gamma parameters is the following (nb is the number of parallel worlds (30) (for parallel tempering), or the number of trials (for simulated annealing) ; i is the index of the world (for parallel tempering) or the trial number (for simulated annealing).

Displacement amplitude

These are used to define the amplitude of displacements of all parameters during random moves. Again, you should not change the default parameters which should be fine for any optimization, with an exponential schedule.

If you choose a different schedule than 'smart', then the the meaning of the max, min and gamma parameters is the following (nb is the number of parallel worlds (30) (for parallel tempering), or the number of trials (for simulated annealing) ; i is the index of the world (for parallel tempering) or the trial number (for simulated annealing).

Options

There are several options available:

Running the optimization

You can choose between two types of runs (from the "Run/Stop" menu):

Normally you should always use the multiple runs mode, in order to obtain several solutions. The main objective is not to ensure that the global minimum will be reached, since the parallel tempering algorithm already garanties that. But it will provide several solutions so that you can have a better feeling about whether you have found the actual global minimum: if you find repeatedly the same solution,then that should be it.

The only case when you should not be using a multiple run is if you want to optimize your structure in several stages (increasing progressively the number of degrees of freedom).

Before running the optimization, put in the number of trials per run the desired number of trials. For a single run, you can use any large number (as long as you use Parallel Tempering), i.e. 100 million trials or more. For a multiple runs, you should choose a number depending on the complexity of the structure, e.g.:

During the optimization, a Tracked Data window will open and display the evolution of the various cost functions (the Chi2 of the different data sets, etc...). To stop the optimization, just use again the "Run/Stop" menu.

If you have used a multiple run, you can use the "Solutions" menu to browse the obtained solutions. Just clicking on any of the solutions listed will restore that configuration and update the different displays (crystal 3D structure, powder pattern,...).

Least-squares minimization

(new in Fox >= 1.8.2) From the "Run/Stop" menu you can select "Least Squares fit" - this will refine the structure, the diffraction data scale as well as the preferred orientation fraction and amplitude parameters (if applicable) for 20 cycles. After the refinement is finished you can browse the results from the intermediate cycles using the "Solutions" menu.

None: Manual/Algorithm (last edited 2011-06-19 14:48:33 by VincentFavreNicolin)