Difference between revisions of "Phase space mode"

From SourceWiki
Jump to navigation Jump to search
Line 1: Line 1:
In this mode the model explores the phase space of the various parameters that are being optimised. The parameter values set in '''[[Readoptimfile]]''' have boundary values within which parameter values may vary. The subroutine Space_survey runs the model over the range of possible parameter values. For each combination of parameter values a cost function is determined,
+
In this mode the model explores the phase space of the various parameters that are being optimised. The parameter values set in '''[[Readoptimfile]]''' have boundary values within which parameter values may vary. The subroutine '''Space_survey''' runs the model over the range of possible parameter values. For each combination of parameter values a cost function is determined,
  
 
<math>C = \frac{1}{n_{i}}\sum_{i} \left|{M_{i}-m_{i}}\right|+\frac{1}{n_{j}}\sum_{j} \left|{M_{j}-m_{j}}\right|+\frac{1}{n_{k}}\sum_{k} \left|{M_{k}-m_{k}}\right|+...</math><br>
 
<math>C = \frac{1}{n_{i}}\sum_{i} \left|{M_{i}-m_{i}}\right|+\frac{1}{n_{j}}\sum_{j} \left|{M_{j}-m_{j}}\right|+\frac{1}{n_{k}}\sum_{k} \left|{M_{k}-m_{k}}\right|+...</math><br>
  
 
where, C is the cost function value, M is the measured value, m is the model predicted value, n is the normalisation constant, and the suffix i, j, k, refers to the different optimisation datasets.
 
where, C is the cost function value, M is the measured value, m is the model predicted value, n is the normalisation constant, and the suffix i, j, k, refers to the different optimisation datasets.

Revision as of 18:32, 16 November 2007

In this mode the model explores the phase space of the various parameters that are being optimised. The parameter values set in Readoptimfile have boundary values within which parameter values may vary. The subroutine Space_survey runs the model over the range of possible parameter values. For each combination of parameter values a cost function is determined,

[math]\displaystyle{ C = \frac{1}{n_{i}}\sum_{i} \left|{M_{i}-m_{i}}\right|+\frac{1}{n_{j}}\sum_{j} \left|{M_{j}-m_{j}}\right|+\frac{1}{n_{k}}\sum_{k} \left|{M_{k}-m_{k}}\right|+... }[/math]

where, C is the cost function value, M is the measured value, m is the model predicted value, n is the normalisation constant, and the suffix i, j, k, refers to the different optimisation datasets.