Matlab Users - Revision history

Tfour: /* Example: Creating a layered model */

2010-08-19T21:34:39Z

‎Example: Creating a layered model

137.229.24.182: /* simanneal_gs.m */

2008-10-02T01:32:58Z

‎simanneal_gs.m

137.229.24.182: /* simanneal_gs.m */

2008-10-02T01:14:09Z

‎simanneal_gs.m

137.229.24.182: /* Simulated Annealing */

2008-10-02T01:12:01Z

‎Simulated Annealing

137.229.24.182 at 00:34, 2 October 2008

2008-10-02T00:34:32Z

Tfour: /* Example: Creating a layered model */

2008-06-20T16:04:51Z

‎Example: Creating a layered model

Tfour: /* Example: Creating a layered model */

2008-06-20T15:41:49Z

‎Example: Creating a layered model

Tfour: /* Layered elastic Models */

2008-06-20T06:56:31Z

‎Layered elastic Models

Tfour: /* Example: Creating a layered model */

2008-06-20T06:55:08Z

‎Example: Creating a layered model

Tfour: /* Example: Creating a layered model */

2008-06-20T06:49:20Z

‎Example: Creating a layered model

@@ Line 70: / Line 70: @@
 There are a handful of Matlab functions that are designed for easy implementation of the simulated annealing algorithm (Cervelli et al., 2001). The functions are located in /usr/local/matlab/TomSimAnneal/
-The main workhorses are ''simanneal_gs.m'' and ''cool_simanneal.m'' while te other functions in the directory are useful for auxiliary tasks or play minor roles in process.
+The main workhorses are ''simanneal_gs.m'' and ''cool_simanneal.m'' while the other functions in the directory are useful for auxiliary tasks or play minor roles in process.
 ===''simanneal_gs.m''===

@@ Line 75: / Line 75: @@
 This is the actual simulated annealing algorithm, it is called ''_gs'' because of the geometric sampling method it uses (this isn't really important, but I thought you might ask).
+The help output (type simanneal_gs at the command line) looks like:
     % [allp MSE bestMOD lowMSE]=SIMANNEAL_GS(m,x,data,cov,CoSch,model)
@@ Line 106: / Line 106: @@
     % Tom Fournier; revised from sammeal_gs_sp Jan 2008.
 bootstrap.m

@@ Line 74: / Line 74: @@
 ===''simanneal_gs.m''===
 This is the actual simulated annealing algorithm, it is called ''_gs'' because of the geometric sampling method it uses (this isn't really important, but I thought you might ask).
     % [allp MSE bestMOD lowMSE]=SIMANNEAL_GS(m,x,data,cov,CoSch,model)
-−
+   %
-% Simulated Annealing Algorithm model parameter inversion. Uses a geometric
+   % Simulated Annealing Algorithm model parameter inversion. Uses a geometric
-% sampleing of the model space to improve performance.
+   % sampleing of the model space to improve performance.
-−
+   %
-−
+   %
-%INPUT
+   %INPUT
-% m       = model bondaries(4nx2),4 parameters for n isotropic sources,
+   % m       = model bondaries(4nx2),4 parameters for n isotropic sources,
-%           min-max for each (east, north, depth, volume)
+   %           min-max for each (east, north, depth, volume)
-silverthrone matlab/TomSimAnneal> head -35 simanneal_gs.m
+   % x       = data locations (3nx1)
-function [allp MSE bestMOD lowMSE]=simanneal_gs(m,x,data,cov,CoSch,model)
+   % data    = displacment/velocity (3nx1)
-% [allp MSE bestMOD lowMSE]=SIMANNEAL_GS(m,x,data,cov,CoSch,model)
+   % cov     = data covariance (3nx3n)
+   % CoSch   = cooling scheduele (from cool_simanneal)
-% Simulated Annealing Algorithm model parameter inversion. Uses a geometric
+   % model   = @function (w/ inputs: m,x)
-% sampleing of the model space to improve performance.
+   %               -m      = model parameters (vector)
+   %               -x      = model inputs (site locations)(vector)
-−
+   %
-%INPUT
+   %OUPUT
-% m       = model bondaries(4nx2),4 parameters for n isotropic sources,
+   % PARAM = final model parameters(10xn)
-%           min-max for each (east, north, depth, volume)
+   % MSE   = vector of MSE values.
-% x       = data locations (3nx1)
+   % bestMOD = best fit model parameters (10xn)
-% data    = displacment/velocity (3nx1)
+   % lowMSE = lowest value from MSE, corresponds to model, bestMOD.
 % cov     = data covariance (3nx3n)
-% CoSch   = cooling scheduele (from cool_simanneal)
+   % Optimizations come from Cerevelli et al. 2001, JGR.
-% model   = @function (w/ inputs: m,x)
+   % Extensive references regarding Simulated Annealing can be found in the
-%               -m      = model parameters (vector)
+   % above paper.
 %               -x      = model inputs (site locations)(vector)
+   % Tom Fournier; revised from sammeal_gs_sp Jan 2008.
 %OUPUT
 % PARAM = final model parameters(10xn)
 % MSE   = vector of MSE values.
 % bestMOD = best fit model parameters (10xn)
 % lowMSE = lowest value from MSE, corresponds to model, bestMOD.
-−
 % Optimizations come from Cerevelli et al. 2001, JGR.
 % Extensive references regarding Simulated Annealing can be found in the
 % above paper.
-−
 % Tom Fournier; revised from sammeal_gs_sp Jan 2008.

← Older revision		Revision as of 01:12, 2 October 2008
Line 68:		Line 68:

	==Simulated Annealing==		==Simulated Annealing==
−	There are a handful of Matlab functions that are designed for ~~the~~ easy implementation of the simulated annealing algorithm (Cervelli et al., 2001).	+	There are a handful of Matlab functions that are designed for easy implementation of the simulated annealing algorithm (Cervelli et al., 2001). The functions are located in /usr/local/matlab/TomSimAnneal/
		+
		+	The main workhorses are ''simanneal_gs.m'' and ''cool_simanneal.m'' while te other functions in the directory are useful for auxiliary tasks or play minor roles in process.
		+
		+	===''simanneal_gs.m''===
		+	This is the actual simulated annealing algorithm, it is called ''_gs'' because of the geometric sampling method it uses (this isn't really important, but I thought you might ask).
		+
		+	% [allp MSE bestMOD lowMSE]=SIMANNEAL_GS(m,x,data,cov,CoSch,model)
		+	%
		+	% Simulated Annealing Algorithm model parameter inversion. Uses a geometric
		+	% sampleing of the model space to improve performance.
		+	%
		+	%
		+	%INPUT
		+	% m = model bondaries(4nx2),4 parameters for n isotropic sources,
		+	% min-max for each (east, north, depth, volume)
		+	silverthrone matlab/TomSimAnneal> head -35 simanneal_gs.m
		+	function [allp MSE bestMOD lowMSE]=simanneal_gs(m,x,data,cov,CoSch,model)
		+	% [allp MSE bestMOD lowMSE]=SIMANNEAL_GS(m,x,data,cov,CoSch,model)
		+	%
		+	% Simulated Annealing Algorithm model parameter inversion. Uses a geometric
		+	% sampleing of the model space to improve performance.
		+	%
		+	%
		+	%INPUT
		+	% m = model bondaries(4nx2),4 parameters for n isotropic sources,
		+	% min-max for each (east, north, depth, volume)
		+	% x = data locations (3nx1)
		+	% data = displacment/velocity (3nx1)
		+	% cov = data covariance (3nx3n)
		+	% CoSch = cooling scheduele (from cool_simanneal)
		+	% model = @function (w/ inputs: m,x)
		+	% -m = model parameters (vector)
		+	% -x = model inputs (site locations)(vector)
		+	%
		+	%OUPUT
		+	% PARAM = final model parameters(10xn)
		+	% MSE = vector of MSE values.
		+	% bestMOD = best fit model parameters (10xn)
		+	% lowMSE = lowest value from MSE, corresponds to model, bestMOD.
		+	%
		+	% Optimizations come from Cerevelli et al. 2001, JGR.
		+	% Extensive references regarding Simulated Annealing can be found in the
		+	% above paper.
		+	%
		+	% Tom Fournier; revised from sammeal_gs_sp Jan 2008.
		+
		+
		+	bootstrap.m
		+	layermogiform.m
		+	nlinfit_cov.m
		+
		+	layerokadaform.m
		+	scatter_sample.m

← Older revision		Revision as of 00:34, 2 October 2008
Line 66:		Line 66:

	The figure is created automatically, it is useful for double checking that the geometry of your model is what you think it is, and to be sure that the layered model solution matches the homogeneous solution. The best approach is to make scaleN as small as possible while still matching the analytical solution to within some tolerance, say 0.1mm or 0.01mm. Remember that the accuracy of the result, with a given scaleN value, may depend on the depth and dip of the dislocation, so if you are inverting for source geometry you should test several model geometries to ensure that you are picking a scale value that gives accurate results for all conditions.		The figure is created automatically, it is useful for double checking that the geometry of your model is what you think it is, and to be sure that the layered model solution matches the homogeneous solution. The best approach is to make scaleN as small as possible while still matching the analytical solution to within some tolerance, say 0.1mm or 0.01mm. Remember that the accuracy of the result, with a given scaleN value, may depend on the depth and dip of the dislocation, so if you are inverting for source geometry you should test several model geometries to ensure that you are picking a scale value that gives accurate results for all conditions.
		+
		+	==Simulated Annealing==
		+	There are a handful of Matlab functions that are designed for the easy implementation of the simulated annealing algorithm (Cervelli et al., 2001).

@@ Line 55: / Line 55: @@
 Choosing a proper scaleN value is important, particularly if you are going to make many function calls to the layer model scripts.  Use get_scaleN.m and get_scaleN_mogi.m to find the best scale value.  The scaleN scripts require a model vector, a maximum distance, and a scale value.
 <br>
->>m = [3 3 5 -45];   % For the Okada model only specify the length, width depth and dip
+    >>m = [3 3 5 -45];   % For the Okada model only specify the length, width depth and dip
-<br>
+    >>rmax = 50;
->>rmax = 50;
+    >>scaleN = 0.3;
-<br>
+    >> [Uprop,Uokada]=get_scaleN(m,rmax,scaleN);
 >>scaleN = 0.3;
-<br>
 >> [Uprop,Uokada]=get_scaleN(m,rmax,scaleN);
@@ Line 68: / Line 65: @@
 <br><br>
-The figure is created automatically, it is useful for double checking that the geometry of your model is what you think it is, and to be sure that the solution matches the homogeneous solution.  The best approach is to make scaleN as small as possible while still matching the analytical solution to within some tolerance.
+The figure is created automatically, it is useful for double checking that the geometry of your model is what you think it is, and to be sure that the layered model solution matches the homogeneous solution.  The best approach is to make scaleN as small as possible while still matching the analytical solution to within some tolerance, say 0.1mm or 0.01mm.  Remember that the accuracy of the result, with a given scaleN value, may depend on the depth and dip of the dislocation, so if you are inverting for source geometry you should test several model geometries to ensure that you are picking a scale value that gives accurate results for all conditions.

← Older revision		Revision as of 15:41, 20 June 2008
Line 49:		Line 49:


−	First make a vector with the depths to the bottom of each layer. Create a model vector, calculate the elastic parameters, and choose a scale term. Note the scaleN is not required the default is 1. ~~Also notice~~ that the program crashes, even though all of the inputs seem to be valid. You can get around this problem by adjusting the length scale, instead of using 1km as the length unit, use 10km. Now everything works just fine! Notice that the fault slip/opening does not have to be rescaled. Just remember that the units of displacement are the same as the units of slip/opening, whatever unit you choose.	+	First make a vector with the depths to the bottom of each layer. Create a model vector, calculate the elastic parameters, and choose a scale term. Note the scaleN is not required the default is 1. The elastic parameter inputs need to be relative to the first layer, so divide by the first value in each vector. Notice that the program crashes, even though all of the inputs seem to be valid. You can get around this problem by adjusting the length scale, instead of using 1km as the length unit, use 10km. Now everything works just fine! Notice that the fault slip/opening does not have to be rescaled. Just remember that the units of displacement are the same as the units of slip/opening, whatever unit you choose.

	<br><br>		<br><br>

@@ Line 22: / Line 22: @@
 A couple of caveats about get_scaleN functions.  For a given scaleN, the accuracy of the layered model is dependent on depth, dip and other parameters that affect the depth to any portion of the fault.  Keep this in mind when choosing a scale value.
-The other warning I should bring up is in regard to the LayeredGreens.m function.  This function will crash in cases where the depth of the plane is too deep.    I simply changed the length scales from 1km to 10km, it worked sometimes.  If you are interested in this problem, particularly if you want to try and fix it, let me know and I can tell you more about it.
+The other warning I should bring up is in regard to the LayeredGreens.m function.  This function will crash in cases where the layer depths are too deep.    I simply changed the length scales from 1km to 10km, it worked sometimes.  If you are interested in this problem, particularly if you want to try and fix it, let me know and I can tell you more about it.
 =Example: Creating a layered model=

← Older revision		Revision as of 06:49, 20 June 2008
Line 43:		Line 43:

	The scripts that calculate deformation in a layered half-space, LayeredGreens.m (Okada dislocation), and MogiLayers (isotropic point source), use the Lame constants and the depth to the bottom of each layer. The bottom layer extends to infinity, so the depth vector has a length ''n-1'', while the Lame parameter vectors have a length of ''n'', for a model with ''n'' layers. The layered model scripts require the length scale to be km.		The scripts that calculate deformation in a layered half-space, LayeredGreens.m (Okada dislocation), and MogiLayers (isotropic point source), use the Lame constants and the depth to the bottom of each layer. The bottom layer extends to infinity, so the depth vector has a length ''n-1'', while the Lame parameter vectors have a length of ''n'', for a model with ''n'' layers. The layered model scripts require the length scale to be km.
		+
		+	<br>
		+	[[Image:Layer okada example.JPG]]
		+	<br><br>
		+
		+
		+	First make a vector with the depths to the bottom of each layer. Create a model vector, calculate the elastic parameters, and choose a scale term. Note the scaleN is not required the default is 1. Also notice that the program crashes, even though all of the inputs seem to be valid. You can get around this problem by adjusting the length scale, instead of using 1km as the length unit, use 10km. Now everything works just fine! Notice that the fault slip/opening does not have to be rescaled. Just remember that the units of displacement are the same as the units of slip/opening, whatever unit you choose.
		+
		+	<br><br>
		+
		+	Choosing a proper scaleN value is important, particularly if you are going to make many function calls to the layer model scripts. Use get_scaleN.m and get_scaleN_mogi.m to find the best scale value. The scaleN scripts require a model vector, a maximum distance, and a scale value.
		+
		+	>>m = [3 3 5 -45];
		+	>>rmax = 50;
		+	>>scaleN = 0.3;
		+	>> [Uprop,Uokada]=get_scaleN(m,rmax,scaleN);
		+	<br><br>
		+	[[Image:ScaleN example.jpg]]
		+	<br><br>
		+
		+	The figure is created automatically, it is useful to double check that the geometry of your model is what you think it is, and to be sure that the solution matches the homogeneous solution. The best approach is to make scaleN as small as possible while still matching the analytical solution to within some tolerance.