summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorMatt Argao <mcargao@email.wm.edu>2013-03-31 22:35:49 -0400
committerMatt Argao <mcargao@email.wm.edu>2013-03-31 22:35:49 -0400
commit4d08c162d5ce4ac8b7eaeb9d985bfef62c886e99 (patch)
tree45733367770ddc4585faa09976b403a858d5d8e5
parent2f3d161985ae2e57ed966981e58eb117d3a49c71 (diff)
downloadmode_match-3.0.tar.gz
mode_match-3.0.zip
Added descriptions to each function.v3.0
-rw-r--r--energy_vs_stability.m25
-rw-r--r--find_min.m4
-rw-r--r--fitness.m125
-rw-r--r--fitter_check.m54
-rw-r--r--gaussian_focus.m17
-rw-r--r--gbeam_propagation.m8
-rw-r--r--gbeam_propagation_froward_only.m9
-rw-r--r--mode_match.m29
-rw-r--r--optics_placer.m1
-rw-r--r--pick_visualization.m10
-rw-r--r--remove_similar_soln.m4
-rw-r--r--self_gbeam_propagation.m38
-rw-r--r--solution_stability.m9
-rw-r--r--solution_visualization.m64
-rw-r--r--stability_visualization.m7
15 files changed, 246 insertions, 158 deletions
diff --git a/energy_vs_stability.m b/energy_vs_stability.m
index 796c631..1915449 100644
--- a/energy_vs_stability.m
+++ b/energy_vs_stability.m
@@ -1,10 +1,22 @@
function [] = energy_vs_stability( possible_energy, stability, index, stability_max )
-%ENERGY_VS_STABILITY Summary of this function goes here
-% Detailed explanation goes here
+%Plots energy vs. stability
+% Using data sent from mode_match, this function plots energy vs.
+% stability of each solution for eyed comparisons.
n = size(stability,2);
- c = 1:n;
+ %Color code data points
+ if n == 3
+ c = [1, 0, 0; 0, 1, 0; 0, 0, 1];
+ elseif n == 2
+ c = [1, 0, 0; 0, 1, 0];
+ elseif n == 1
+ c = [1, 0, 0];
+ else
+ c = 1:n;
+ end
+
+ %Reorganize stability to coincide with energies
for i=1:n
fixed_stability(i,1) = stability(i);
end
@@ -13,6 +25,7 @@ function [] = energy_vs_stability( possible_energy, stability, index, stability_
energy(i,1) = possible_energy(index(i));
end
+ %Graph scatter plot
figure(n+1)
textCell = arrayfun(@(x,y) sprintf('(%3.2f, %3.2f)',x,y),energy,fixed_stability,'un',0);
scatter(energy,fixed_stability,25,c,'filled')
@@ -23,13 +36,11 @@ function [] = energy_vs_stability( possible_energy, stability, index, stability_
dx =0.0000000005;
dy =0.0000000005;
+ %Label data points relative to solution number
for i = 1:n
n_solution = num2str(i);
text(energy(i)+dx, fixed_stability(i)+dy, n_solution, 'FontSize',10);
end
-% for i = 1:n
-% text(energy(i), fixed_stability(i), textCell{i}, 'FontSize',8)
-% end
-
+
end
diff --git a/find_min.m b/find_min.m
index 2c3ae7d..a6962d2 100644
--- a/find_min.m
+++ b/find_min.m
@@ -1,7 +1,5 @@
function [ x_sol, energy ] = find_min( index, lens_placement, fitness_simplified, MaxFunEvals, MaxIter )
-%FIND_MIN Summary of this function goes here
-% Detailed explanation goes here
-
+%Minimizes energy for given solution
[x_sol, energy] = fminsearch(fitness_simplified, lens_placement(index,:), optimset('TolFun',1e-5,'MaxFunEvals',MaxFunEvals,'MaxIter', MaxIter));
end
diff --git a/fitness.m b/fitness.m
index 403d8d7..697d5fd 100644
--- a/fitness.m
+++ b/fitness.m
@@ -1,45 +1,16 @@
-function [Energy, Waist, Penalty] = fitness( q_0, q_final, x_final, optics_positions, optics_focal_length, lambda )
+function [Energy, Waist, Penalty] = fitness( q_0, q_final, x_final, optics_positions, optics_focal_length, lambda, self_flag )
%Outputs fitness of suggested solution
- x0 = 0;
- Energy = 0;
-
- x1=optics_positions(1);
- x2=optics_positions(2);
- x3=optics_positions(3);
-
- % penalty calculation
- % do not put lenses too close to each other and end positions
- lens_size=0.03;
-
- d(1)=abs(x1-x2);
- d(2)=abs(x2-x3);
- d(3)=abs(x1-x3);
-
- d_from_start=abs(x0-optics_positions);
- d_from_end=abs(x_final-optics_positions);
-
-
- d=cat(2, d, d_from_start, d_from_end);
-
- coef = 1;
-
- penalty_lenses_too_closely_spaced =coef*sum( exp(-(d/(lens_size)).^12) );
-
- Energy = Energy + penalty_lenses_too_closely_spaced;
-
- % make sure that lenses are between ends
- d_from_start=(x0-optics_positions);
- d_from_end=(optics_positions-x_final);
-
- d = cat(2, d_from_start, d_from_end);
-
- coef = 1;
- distance_scaling=100;
- penalty_lenses_outside_optical_path = coef * sum(1 + tanh(distance_scaling*d));
-
- Energy = Energy + penalty_lenses_outside_optical_path;
-
- % make collimated region between 2nd and 3rd lens
+x0 = 0;
+Np=20; % # of pts between start position and second lens
+N_collimated = 10; % # of pts between second lens and third lens to be collimated region
+Energy = 0;
+
+x1=optics_positions(1);
+x2=optics_positions(2);
+x3=optics_positions(3);
+
+if self_flag == 1
+ % make collimated region between 2nd and 3rd lens
[w0, r0] = q2wr(q_0, lambda);
[ w, w_pos ] = self_gbeam_propagation( w0, optics_positions, optics_focal_length, x0, lambda );
@@ -51,20 +22,84 @@ function [Energy, Waist, Penalty] = fitness( q_0, q_final, x_final, optics_posit
Energy = Energy + penalty_not_collimated_beam;
-% % waist at end matches desired waist
+ % waist at end matches desired waist
coef = 10;
[waist_desired, r_desired] = q2wr(q_final, lambda);
penalty_waist = coef *((( waist_desired-w(end) )/waist_desired)^2);
Energy = Energy + penalty_waist;
-
+
coef=10;
dist_between_desired_and_final_waist_location =w_pos(end) - x_final;
penalty_final_waist_position = coef*dist_between_desired_and_final_waist_location^2;
-
+
Energy = Energy + penalty_final_waist_position;
+
+else
+ % check if waists match in forward and backward propagation
+ x_array1=linspace(x0,x2,Np);
+ x_array2=linspace(x2, x3, N_collimated);
+ x = cat(2,x_array1,x_array2);
+
+ q_f_trial_forward = gbeam_propagation(x,q_0,x0,optics_placer(optics_positions, optics_focal_length));
+ [Waist_trial_forward, Radius_trial_forward] = q2wr(q_f_trial_forward, lambda);
+ q_f_trial_backward = gbeam_propagation(x,q_final,x_final,optics_placer(optics_positions, optics_focal_length));
+ [Waist_trial_backward, Radius_trial_backward] = q2wr(q_f_trial_backward, lambda);
+
+ Penalty_waist_mismatch = sum(abs((Waist_trial_forward-Waist_trial_backward)./min(Waist_trial_forward, Waist_trial_backward)))/Np;
+
+ Energy = 1e-2*Penalty_waist_mismatch;
+
+ %intialize intermediate points between lenses
+ q_intermediate = q_f_trial_forward((x2<x) & (x<x3));
+ lambda_over_waist_sq = (-imag(1./q_intermediate)); %with numerical factor
+
+ coef = 1e-3;
+ penalty_not_collimated_beam = coef * sum(exp((std(lambda_over_waist_sq)/mean(lambda_over_waist_sq)/abs(optics_positions(2) - optics_positions(3)))));
+ Energy = Energy + penalty_not_collimated_beam;
+
+
+end
+
+
+
+% penalty calculation
+% do not put lenses too close to each other and end positions
+lens_size=0.03;
+
+
+
+d(1)=abs(x1-x2);
+d(2)=abs(x2-x3);
+d(3)=abs(x1-x3);
+
+d_from_start=abs(x0-optics_positions);
+d_from_end=abs(x_final-optics_positions);
+
+
+
+d=cat(2, d, d_from_start, d_from_end);
+
+coef = 1;
+
+penalty_lenses_too_closely_spaced =coef*sum( exp(-(d/(lens_size)).^12) );
+
+Energy = Energy + penalty_lenses_too_closely_spaced;
+
+% make sure that lenses are between ends
+d_from_start=(x0-optics_positions);
+d_from_end=(optics_positions-x_final);
+
+d = cat(2, d_from_start, d_from_end);
+
+coef = 1e-2;
+distance_scaling=100;
+penalty_lenses_outside_optical_path = coef * sum(1 + tanh(distance_scaling*d));
+
+Energy = Energy + penalty_lenses_outside_optical_path;
+
+Penalty = [ penalty_lenses_too_closely_spaced; penalty_lenses_too_closely_spaced; penalty_not_collimated_beam];
- Penalty = [ penalty_lenses_too_closely_spaced; penalty_lenses_outside_optical_path; penalty_not_collimated_beam; penalty_waist; penalty_final_waist_position];
end
diff --git a/fitter_check.m b/fitter_check.m
index 9491295..b7866ae 100644
--- a/fitter_check.m
+++ b/fitter_check.m
@@ -1,39 +1,39 @@
%Permute all possible lens combinations out of set of lenses
-lens_set = [.075, .203, .05, .03];
-lens_set = [.075,.203];
-lens_permutations = pick(lens_set,3,'or');
+lens_set = [.075, .203]; %Given lenses of unique focal lengths
+lens_permutations = pick(lens_set,3,'or'); %3 lens solutions
%Pre-defined Constants
-lambda= 1.064E-6 ;
-Ltot= 1.010675025828971 ;
-r0= 1.0E+100 ;
-w0= 2.563E-5 ;
-x0= 0 ;
-wf= 3.709E-5 ;
-rf= 1.0E+100 ;
-xf= Ltot;
-q0=wr2q(w0,r0,lambda);
-qf=wr2q(wf,rf,lambda);
+lambda= 1.064E-6 ; %Wavelength of beam
+Ltot= 1.010675025828971 ; %Length of optical system
+r0= 1.0E+100 ; %Initial radius of curvature
+w0= 2.563E-5 ; %Initial waist
+x0= 0 ; %Starting position of beam
+wf= 3.709E-5 ; %Desired final waist
+rf= 1.0E+100 ; %Desired final radius
+lens_width = .03; %Lens width
+show_lens_width = 1; %Set to 1 to enable display of lens width on solution propagation plot
+show_lens_position = 1; %Set to 1 to enable display of position of center of lens on solution propagation plot
+display_prop = [show_lens_width, show_lens_position];
+n_truncate = 3; %number of digits in truncated solution
+n_visualizations = 2; %number of best solutions to visualize
+n_hist = 1000; %number of sample points in histogram
+stability_max = 1; %max stability (y-axis) shown on energy vs. stability graph
+self_flag = 0; %Set to 1 to use Self's gaussian beam propagation, otherwise set to 0
%End list
+q0=wr2q(w0,r0,lambda); %Calculate intial q
+qf=wr2q(wf,rf,lambda); %Calculate final q
+
%Mode match
-[ possible_lens_placement, possible_lens_set, possible_sample_energy] = mode_match( q0, qf, Ltot, lambda, lens_permutations );
+[ possible_lens_placement, initial_lens_placement, possible_lens_set, possible_sample_energy] = mode_match( q0, qf, Ltot, lambda, lens_permutations, lens_width, self_flag );
%Remove similar solutions
-n_truncate = 3; %number of digits in truncated solution
[ possible_lens_placement_uniq, possible_lens_placement, possible_sample_energy, possible_lens_set, index ] = remove_similar_soln( possible_sample_energy, possible_lens_placement, possible_lens_set, n_truncate );
-n_visualizations = 5; %number of best solutions to visualize
-pick_visualization( possible_sample_energy, possible_lens_placement_uniq, possible_lens_placement, possible_lens_set, index, n_visualizations, q0, qf, Ltot, lambda );
-
-%Visualize fitness function for fixed f2 and f3
-lens_set = [.075, .075, .203];
-f2= 0.40361319425309 ;
-f3= 0.80361319425309 ;
-%fitness_simplified=@(x) fitness(q0, qf, Ltot, [x, f2, f3], lens_set, lambda );
-%figure(6)
-%ezplot(fitness_simplified, [0,Ltot])
+%Visualize solutions
+pick_visualization( possible_sample_energy, possible_lens_placement_uniq, possible_lens_placement, possible_lens_set, index, n_visualizations, q0, qf, Ltot, lambda, lens_width, display_prop );
-[ w, w_pos ] = self_gbeam_propagation( w0, possible_lens_placement, possible_lens_set, x0, lambda )
-%solution_visualization(q0,x0,qf,xf, optics_placer(possible_lens_placement, possible_lens_set), lambda)
+%Plot energy vs. stability for each solution
+[stability] = stability_visualization( possible_lens_placement_uniq, q0, qf, Ltot, possible_lens_placement, possible_lens_set, lambda, n_visualizations, n_hist, index, self_flag );
+energy_vs_stability( possible_sample_energy, stability, index, stability_max)
diff --git a/gaussian_focus.m b/gaussian_focus.m
index ff444ee..cceeda0 100644
--- a/gaussian_focus.m
+++ b/gaussian_focus.m
@@ -1,9 +1,14 @@
function [ w, s ] = gaussian_focus( w0, s0, f, lambda )
-%GAUSSIAN_FOCUS Summary of this function goes here
-% Detailed explanation goes here
+%Focusing of Gaussian beams based on Sidney A. Self's "Focusing of
+%spherical Gaussian beams" Applied Optics, Vol. 22, Issue 5, pp. 658-661 (1983)
+% Input: lambda = wavelength;
+% w0 = waist before the lens;
+% s0 = distance of waist before lens (s>0 before lens)
+% f = focal length of lens (f>0 for converging lens)
+% Output: w = new waist;
+% s = position of waist from lens
-zR = pi*w0^2/lambda;
-s = f*(1+(s0/f-1)/((s0/f-1)^2+(zR/f)^2));
+zR = pi*w0^2/lambda; %Rayleigh Range
+s = f*(1+(s0/f-1)/((s0/f-1)^2+(zR/f)^2)); %Eq. (9b)
w = w0/sqrt((1-s0/f)^2+(zR/f)^2);
-end
-
+end \ No newline at end of file
diff --git a/gbeam_propagation.m b/gbeam_propagation.m
index d81f70e..585048d 100644
--- a/gbeam_propagation.m
+++ b/gbeam_propagation.m
@@ -1,4 +1,4 @@
-function q = gbeam_propagation(x_pos, q_in, x_in, optics_elements)
+function [q] = gbeam_propagation(x_pos, q_in, x_in, optics_elements)
% calculate the 'q' parameter of the Gaussian beam propagating through optical
% 'optics_elements' array along 'x' axis at points 'x_pos'
% takes the gaussian beam with initial q_in parameter at x_in
@@ -8,7 +8,7 @@ function q = gbeam_propagation(x_pos, q_in, x_in, optics_elements)
if any(x_pos >= x_in)
% Forward propagation to the right of x_in
- q(x_pos >= x_in) = gbeam_propagation_froward_only(x_pos(x_pos>=x_in), q_in, x_in, optics_elements);
+ [q(x_pos >= x_in)] = gbeam_propagation_froward_only(x_pos(x_pos>=x_in), q_in, x_in, optics_elements);
end
if any(x_pos < x_in)
@@ -45,8 +45,10 @@ function q = gbeam_propagation(x_pos, q_in, x_in, optics_elements)
% final assignment of the backwards propagating beam
% which we need to flip back
q(x_pos<x_in) = fliplr(q_backw);
- end
+ end
+
+
end
%!test
diff --git a/gbeam_propagation_froward_only.m b/gbeam_propagation_froward_only.m
index afc1316..2cdc1d8 100644
--- a/gbeam_propagation_froward_only.m
+++ b/gbeam_propagation_froward_only.m
@@ -1,4 +1,4 @@
-function q = gbeam_propagation_froward_only(x_pos, q_in, x_in, optics_elements)
+function [q] = gbeam_propagation_froward_only(x_pos, q_in, x_in, optics_elements)
% calculate the 'q' parameter of the Gaussian beam propagating through optical
% 'optics_elements' only in the positive direction along 'x' axis at points 'x_pos'
% takes the gaussian beam with initial q_in parameter at x_in
@@ -8,6 +8,10 @@ function q = gbeam_propagation_froward_only(x_pos, q_in, x_in, optics_elements)
%
% all x_pos must be to the right of x_in
% x_pos must be monotonic!
+
+ %Initialize q_lens (q at position of lens)
+ q_lens = zeros(1,size(optics_elements,2));
+
if (any(x_pos < x_in))
error('all beam positions must be to the right of the x_in');
end
@@ -45,8 +49,7 @@ function q = gbeam_propagation_froward_only(x_pos, q_in, x_in, optics_elements)
index = (x_last_calc <= x_pos);
x=x_pos(index);
q(index)= q_after_free_space(q_last_calc,x-x_last_calc);
-
-
+
end
diff --git a/mode_match.m b/mode_match.m
index 0a36332..f3f7408 100644
--- a/mode_match.m
+++ b/mode_match.m
@@ -1,4 +1,4 @@
-function [ possible_lens_placement, possible_lens_set, possible_sample_energy] = mode_match( q0, qf, Ltot, lambda, lens_permutations )
+function [ final_possible_lens_placement, initial_possible_lens_placement, possible_lens_set, possible_sample_energy] = mode_match( q0, qf, Ltot, lambda, lens_permutations, lens_width, self_flag )
%Shuffles lenses into random positions and stores possible solutions
% Shuffles over entire lens permutation array for n_shuffles times.
% Afterwards, solutions are sorted by Energy and truncated to n_truncate
@@ -7,6 +7,11 @@ function [ possible_lens_placement, possible_lens_set, possible_sample_energy] =
n_perms = size(lens_permutations,1);
n_shuffles=20; %number of random placements of lenses
+initial_MaxFunEvals = 1e8;
+initial_MaxIter = 10;
+final_MaxFunEvals = 1e8;
+final_MaxIter = 100;
+
%Initialize sample arrays
N = n_perms * n_shuffles;
possible_lens_placement = zeros(N,3);
@@ -14,25 +19,33 @@ possible_lens_set = zeros(N,3);
possible_sample_energy = zeros(N,1);
initial_rand_lens_placement=zeros(N,3);
-lens_size = .03; % physical size of the lens
-
for ip = 1:n_perms
f3=lens_permutations(ip,3);
x3=Ltot-f3; % last lense transfer collimated region to focused spot
for is = 1:n_shuffles
possible_lens_set((ip-1)*n_shuffles + is,:) = lens_permutations(ip,:);
- initial_rand_lens_placement_tmp = sort(lens_size+(x3-2*lens_size)*rand(1,2));
+ initial_rand_lens_placement_tmp = sort(lens_width+(x3-2*lens_width)*rand(1,2));
initial_rand_lens_placement((ip-1)*n_shuffles + is,:) = [initial_rand_lens_placement_tmp, x3];
end
end
+
+
+
parfor i = 1:N
-
- fitness_simplified=@(x) fitness(q0, qf, Ltot, x, possible_lens_set(i,:), lambda );
- [x_sol, energy]=fminsearch(fitness_simplified, initial_rand_lens_placement(i,:), optimset('TolFun',1e-5,'MaxFunEvals',1e8,'MaxIter',200));
+ fitness_simplified=@(x) fitness(q0, qf, Ltot, x, possible_lens_set(i,:), lambda, self_flag );
+ [ x_sol, energy ] = find_min(i, initial_rand_lens_placement, fitness_simplified, initial_MaxFunEvals, initial_MaxIter )
+
+ initial_possible_lens_placement(i,:) =x_sol;
- possible_lens_placement(i,:) =x_sol;
+end
+
+parfor i = 1:N
+ fitness_simplified=@(x) fitness(q0, qf, Ltot, x, possible_lens_set(i,:), lambda, self_flag );
+ [ x_sol, energy ] = find_min(i, initial_possible_lens_placement, fitness_simplified, final_MaxFunEvals, final_MaxIter )
+
+ final_possible_lens_placement(i,:) =x_sol;
possible_sample_energy(i) = energy;
end
diff --git a/optics_placer.m b/optics_placer.m
index 247cc81..2407691 100644
--- a/optics_placer.m
+++ b/optics_placer.m
@@ -1,4 +1,5 @@
function optics = optics_placer( x ,f )
+%Places optical elements within array optics for use in other functions
x1=x(1);
x2=x(2);
x3=x(3);
diff --git a/pick_visualization.m b/pick_visualization.m
index 88f1989..5a11231 100644
--- a/pick_visualization.m
+++ b/pick_visualization.m
@@ -1,4 +1,4 @@
-function [ ] = pick_visualization( fitness_energy, possible_lens_placement_uniq, possible_lens_placement, possible_lens_set, index, n_visualizations, q0, qf, Ltot, lambda )
+function [ ] = pick_visualization( fitness_energy, possible_lens_placement_uniq, possible_lens_placement, possible_lens_set, index, n_visualizations, q0, qf, Ltot, lambda, lens_width, display_prop )
%Picks n_visualizations of sets of data and graphs each
x0 = 0;
@@ -6,11 +6,11 @@ n_possible_lens_placement = min(n_visualizations,size(possible_lens_placement_un
for n_graph = 1:n_possible_lens_placement
figure(n_graph)
- [w_final_trial, r_final_trial] = solution_visualization(q0, x0, qf, Ltot, optics_placer(possible_lens_placement(index(n_graph),:), possible_lens_set(index(n_graph),:)), lambda);
+ [w_final_trial, r_final_trial] = solution_visualization(q0, x0, qf, Ltot, optics_placer(possible_lens_placement(index(n_graph),:), possible_lens_set(index(n_graph),:)), lambda, lens_width, display_prop);
- str1=sprintf('\n f_1 = %0.4f, x_1 = %0.4f\n',possible_lens_set(index(n_graph),1),possible_lens_placement(index(n_graph),1));
- str2=sprintf('f_2 = %0.4f, x_2 = %0.4f\n',possible_lens_set(index(n_graph),2),possible_lens_placement(index(n_graph),2));
- str3=sprintf('f_3 = %0.4f, x_3 = %0.4f\n',possible_lens_set(index(n_graph),3),possible_lens_placement(index(n_graph),3));
+ str1=sprintf('\n (red) f_1 = %0.4f, x_1 = %0.4f\n',possible_lens_set(index(n_graph),1),possible_lens_placement(index(n_graph),1));
+ str2=sprintf(' (green) f_2 = %0.4f, x_2 = %0.4f\n',possible_lens_set(index(n_graph),2),possible_lens_placement(index(n_graph),2));
+ str3=sprintf(' (blue) f_3 = %0.4f, x_3 = %0.4f\n',possible_lens_set(index(n_graph),3),possible_lens_placement(index(n_graph),3));
tstr='Solution #';
str_w = ['w_{final}= ',num2str(w_final_trial)];
str_r = [', r_{final}= ',num2str(r_final_trial)];
diff --git a/remove_similar_soln.m b/remove_similar_soln.m
index 9bea732..a106aff 100644
--- a/remove_similar_soln.m
+++ b/remove_similar_soln.m
@@ -1,6 +1,6 @@
function [ possible_lens_placement_uniq, possible_lens_placement, possible_sample_energy, possible_lens_set, index ] = remove_similar_soln( possible_sample_energy, possible_lens_placement, possible_lens_set, n_truncate )
-%REMOVE_SIMILAR_SOLN Summary of this function goes here
-% Detailed explanation goes here
+%Removes similar solutions based on n_truncate (decimal tolerance to
+%determin uniqueness)
%Sorting possible solution according to energy
[possible_sample_energy, index] = sort(possible_sample_energy);
diff --git a/self_gbeam_propagation.m b/self_gbeam_propagation.m
index 10bcf05..824be5a 100644
--- a/self_gbeam_propagation.m
+++ b/self_gbeam_propagation.m
@@ -1,11 +1,11 @@
function [ w, w_pos ] = self_gbeam_propagation( w0, x_lens, f, x0, lambda )
-%SELF_GBEAM_PROPAGATION Summary of this function goes here
-% Detailed explanation goes here
+%Beam propagation based on Sidney A. Self's "Focusing of spherical Gaussian
+%beams". Applied Optics, Vol. 22, Issue 5, pp. 658-661 (1983)
n_lens = 3;
-w = w0;
-w_pos = x0;
+w = w0; %Initial waist
+w_pos = x0; %Initial waist position
for i = 1:n_lens
x_from_next_lens = x_lens(i) - w_pos(i);
@@ -14,32 +14,4 @@ for i = 1:n_lens
w_pos(i+1) = x_lens(i) + s_new;
end
-% x_pos= x_lens;
-% x(1)=w_pos(1);
-% waist_prop(1)=w(1);
-%
-% x_temp=linspace(x0 ,x_pos(1));
-% waist_prop=[waist_prop, w(1)*sqrt(1+((x_temp-w_pos(1))/(pi*w(1)^2/lambda)).^2)];
-% x=[x x_temp];
-%
-% for i=1:n_lens
-% x_temp=linspace(x_pos(i),x_pos(i+1));
-% waist_prop=[waist_prop, w(i)*sqrt(1+((x_temp-w_pos(i))/(pi*w(i)^2/lambda)).^2)];
-% x=[x x_temp];
-% end
-%
-% x_temp=linspace(x_pos(end),x_pos(end)+abs(f(end)));
-% waist_prop=[waist_prop, w(end)*sqrt(1+((x_temp-w_pos(end))/(pi*w(end)^2/lambda)).^2)];
-% x=[x x_temp];
-%
-%
-% figure(1)
-% plot(x,waist_prop,'r',x,-waist_prop,'r');
-%
-
-
-
-
-
-end
-
+end \ No newline at end of file
diff --git a/solution_stability.m b/solution_stability.m
index 324621f..359335c 100644
--- a/solution_stability.m
+++ b/solution_stability.m
@@ -1,12 +1,13 @@
-function [ hist_h, hist_x ] = solution_stability( q_0, q_final, x_final, optics_positions, optics_focal_length, lambda, n_hist )
-%SOLUTION_STABILITY Summary of this function goes here
-% Detailed explanation goes here
+function [ hist_h, hist_x ] = solution_stability( q_0, q_final, x_final, optics_positions, optics_focal_length, lambda, n_hist, self_flag )
+%Plots histogram of n_hist number of sample points
+%Area under plot is equal to the stability of the solution.
+
lens_displacement = 10^-3;
x = lens_displacement*randn(n_hist,3);
for i=1:n_hist
xd= x(i,:);
- fitness_array(i) = fitness( q_0, q_final, x_final, optics_positions + xd, optics_focal_length, lambda );
+ fitness_array(i) = fitness( q_0, q_final, x_final, optics_positions + xd, optics_focal_length, lambda, self_flag );
end
[hist_h, hist_x] = hist(fitness_array,100);
diff --git a/solution_visualization.m b/solution_visualization.m
index 2cf8ce7..2b4158c 100644
--- a/solution_visualization.m
+++ b/solution_visualization.m
@@ -1,11 +1,19 @@
-function [waste_at_the_end, radius_at_the_end] = solution_visualization(q0,x0, qf, xf, optics, lambda)
+function [waste_at_the_end, radius_at_the_end, waist_at_lens_position] = solution_visualization(q0,x0, qf, xf, optics, lambda, lens_width, display_prop)
%Propagates beam with given input parameters
% Forward propagation and backward propagation are taken in order to
% visualize reasonable solutions.
+%Array of lens positions
+n_lens = size(optics,2);
+
+for i = 1:n_lens
+ lens_position(i) = optics{i}.x;
+end
+
x=linspace(x0,xf,1000); % we will calculate beam profile between x0 and xf
+
%Forward propagation
-q_forward=gbeam_propagation(x,q0,x0,optics);
+q_forward =gbeam_propagation(x,q0,x0,optics);
[w_forward,r_forward]=q2wr(q_forward, lambda);
%Backward propagation
@@ -13,13 +21,53 @@ q_backward=gbeam_propagation(x,qf,xf,optics);
[w_backward,r_backward]=q2wr(q_backward, lambda);
%Plot beam profile
-plot ( ...
- x,w_forward, '-r', ...
- x,-w_forward, '-r', ...
- x, w_backward, '-.b', ...
- x, -w_backward, '-.b')
-legend({'forward propagation', '', 'backward propagation', ''})
+subplot(2,1,1); plot ( ...
+ x,w_forward, '-r', ...
+ x,-w_forward, '-r', ...
+ x, w_backward, '-.b', ...
+ x, -w_backward, '-.b')
+legend({'forward propagation', '', 'backward propagation', ''})
+
+%Find q and waist at lens positions
+waist_at_lens_position = zeros(1,n_lens);
+q_lens = zeros(1,n_lens);
+
+for i = 1:n_lens
+ q_lens(i) = gbeam_propagation(lens_position(i),q0,x0,optics);
+end
+
+
+for i = 1:n_lens
+ waist_at_lens_position(i) = q2wr(q_lens(i), lambda);
+end
+
+
+%Plot lenses
+color = ['m' 'g' 'c'];
+
+if display_prop(1) == 1
+ for i = 1:n_lens
+ half_width = lens_width/2;
+ corrected_lens_position = lens_position(i)-half_width;
+ corrected_waist = waist_at_lens_position(i)*2;
+ rectangle('Position', [corrected_lens_position,-waist_at_lens_position(i),lens_width,corrected_waist], 'EdgeColor', color(i), 'LineWidth',1);
+ end
+end
+
+if display_prop(2) == 1
+ for i = 1:n_lens
+ x1 = optics{i}.x;
+ y1 = waist_at_lens_position(i);
+ x2 = x1;
+ y2 = -waist_at_lens_position(i);
+
+ line([x1;x2], [y1;y2], 'LineWidth', 3, 'Color', color(i));
+ end
+end
+
[waste_at_the_end,radius_at_the_end] = q2wr(q_forward(end), lambda);
+
+
diff --git a/stability_visualization.m b/stability_visualization.m
index ba0a8de..37e9aad 100644
--- a/stability_visualization.m
+++ b/stability_visualization.m
@@ -1,12 +1,11 @@
-function [stability] = stability_visualization( possible_lens_placement_uniq, q0, qf, xf, optics_position, optics_set, lambda, n, n_hist, index )
-%STABILITY_VISUALIZATION Summary of this function goes here
-% Detailed explanation goes here
+function [stability] = stability_visualization( possible_lens_placement_uniq, q0, qf, xf, optics_position, optics_set, lambda, n, n_hist, index, self_flag )
+%Visualizes stability graph
n_possible_lens_placement = min(n, size(possible_lens_placement_uniq,1));
for i=1:n
figure(i)
- [hist_h, hist_x] = solution_stability( q0, qf, xf, optics_position(index(i),:), optics_set(index(i),:), lambda, n_hist );
+ [hist_h, hist_x] = solution_stability( q0, qf, xf, optics_position(index(i),:), optics_set(index(i),:), lambda, n_hist, self_flag );
subplot(2,1,2); plot(hist_x, hist_h);
area = trapz(hist_x, hist_h);
str_n = ['# of test points = ', num2str(n_hist)];