% Section 6.1.2
% Boyd & Vandenberghe "Convex Optimization"
% Original by Lieven Vandenberghe
% Adapted for CVX Argyris Zymnis - 10/2005
%
% Comparison of the ell1, ell2, deadzone-linear and log-barrier
% penalty functions for the approximation problem:
%       minimize phi(A*x-b),
%
% where phi(x) is the penalty function
% Log-barrier will be implemented in the future version of CVX

% Generate input data
randn('state',0);
m=100; n=30;
A = randn(m,n);
b = randn(m,1);

% ell_1 approximation
% minimize   ||Ax+b||_1
disp('ell-one approximation');
cvx_begin
    variable x1(n)
    minimize(norm(A*x1+b,1))
cvx_end

% ell_2 approximation
% minimize ||Ax+b||_2
disp('ell-2');
x2=-A\b;

% deadzone penalty approximation
% minimize sum(deadzone(Ax+b,0.5))
% deadzone(y,z) = max(abs(y)-z,0)
dz = 0.5;
disp('deadzone penalty');
cvx_begin
    variable xdz(n)
    minimize(sum(max(abs(A*xdz+b)-dz,0)))
cvx_end


% log-barrier penalty approximation
%
% minimize -sum log(1-(ai'*x+bi)^2)

disp('log-barrier')

% parameters for Newton Method & line search
alpha=.01; beta=.5;

% minimize linfty norm to get starting point
cvx_begin
    variable xlb(n)
    minimize norm(A*xlb+b,Inf)
cvx_end
linf = cvx_optval;
A = A/(1.1*linf);
b = b/(1.1*linf);

for iters = 1:50

   yp = 1 - (A*xlb+b);  ym = (A*xlb+b) + 1;
   f = -sum(log(yp)) - sum(log(ym));
   g = A'*(1./yp) - A'*(1./ym);
   H = A'*diag(1./(yp.^2) + 1./(ym.^2))*A;
   v = -H\g;
   fprime = g'*v;
   ntdecr = sqrt(-fprime);
   if (ntdecr < 1e-5), break; end;

   t = 1;
   newx = xlb + t*v;
   while ((min(1-(A*newx +b)) < 0) | (min((A*newx +b)+1) < 0))
       t = beta*t;
       newx = xlb + t*v;
   end;
   newf = -sum(log(1 - (A*newx+b))) - sum(log(1+(A*newx+b)));
   while (newf > f + alpha*t*fprime)
       t = beta*t;
       newx = xlb + t*v;
       newf = -sum(log(1-(A*newx+b))) - sum(log(1+(A*newx+b)));
   end;
   xlb = xlb+t*v;
end


% Plot histogram of residuals

ss = max(abs([A*x1+b; A*x2+b; A*xdz+b;  A*xlb+b]));
tt = -ceil(ss):0.05:ceil(ss);  % sets center for each bin
[N1,hist1] = hist(A*x1+b,tt);
[N2,hist2] = hist(A*x2+b,tt);
[N3,hist3] = hist(A*xdz+b,tt);
[N4,hist4] = hist(A*xlb+b,tt);


range_max=2.0;  rr=-range_max:1e-2:range_max;

figure(1), clf, hold off
subplot(4,1,1),
bar(hist1,N1);
hold on
plot(rr, abs(rr)*40/3, '-');
ylabel('p=1')
axis([-range_max range_max 0 40]);
hold off

subplot(4,1,2),
bar(hist2,N2);
hold on;
plot(rr,2*rr.^2),
ylabel('p=2')
axis([-range_max range_max 0 11]);
hold off

subplot(4,1,3),
bar(hist3,N3);
hold on
plot(rr,30/3*max(0,abs(rr)-dz))
ylabel('Deadzone')
axis([-range_max range_max 0 25]);
hold off

subplot(4,1,4),
bar(hist4,N4);
rr_lb=linspace(-1+(1e-6),1-(1e-6),600);
hold on
plot(rr_lb, -3*log(1-rr_lb.^2),rr,2*rr.^2,'--')
axis([-range_max range_max 0 11]);
ylabel('Log barrier'),
xlabel('r')
hold off
ell-one approximation
 
Calling sedumi: 230 variables, 100 equality constraints
------------------------------------------------------------
SeDuMi 1.21 by AdvOL, 2005-2008 and Jos F. Sturm, 1998-2003.
Alg = 2: xz-corrector, Adaptive Step-Differentiation, theta = 0.250, beta = 0.500
Split 30 free variables
eqs m = 100, order n = 261, dim = 261, blocks = 1
nnz(A) = 6200 + 0, nnz(ADA) = 10000, nnz(L) = 5050
 it :     b*y       gap    delta  rate   t/tP*  t/tD*   feas cg cg  prec
  0 :            2.34E+02 0.000
  1 :   3.05E+01 5.69E+01 0.000 0.2433 0.9000 0.9000   2.72  1  1  9.7E-01
  2 :   4.57E+01 2.12E+01 0.000 0.3727 0.9000 0.9000   1.14  1  1  4.3E-01
  3 :   5.16E+01 7.84E+00 0.000 0.3699 0.9000 0.9000   1.04  1  1  1.8E-01
  4 :   5.40E+01 2.49E+00 0.000 0.3174 0.9000 0.9000   1.01  1  1  5.9E-02
  5 :   5.48E+01 6.63E-01 0.000 0.2665 0.9000 0.9000   1.00  1  1  1.6E-02
  6 :   5.50E+01 1.94E-01 0.000 0.2918 0.9031 0.9000   1.00  1  1  4.8E-03
  7 :   5.51E+01 4.71E-02 0.000 0.2433 0.9149 0.9000   1.00  1  1  1.3E-03
  8 :   5.51E+01 7.72E-03 0.000 0.1639 0.9000 0.9089   1.00  1  1  1.8E-04
  9 :   5.51E+01 5.53E-05 0.000 0.0072 0.9990 0.9990   1.00  1  1  
iter seconds digits       c*x               b*y
  9      0.0  14.1  5.5128921594e+01  5.5128921594e+01
|Ax-b| =   5.2e-13, [Ay-c]_+ =   9.7E-15, |x|=  1.6e+01, |y|=  8.8e+00

Detailed timing (sec)
   Pre          IPM          Post
1.000E-02    4.000E-02    0.000E+00    
Max-norms: ||b||=1.957607e+00, ||c|| = 1,
Cholesky |add|=0, |skip| = 0, ||L.L|| = 2.90116.
------------------------------------------------------------
Status: Solved
Optimal value (cvx_optval): +55.1289
ell-2
deadzone penalty
 
Calling sedumi: 430 variables, 200 equality constraints
------------------------------------------------------------
SeDuMi 1.21 by AdvOL, 2005-2008 and Jos F. Sturm, 1998-2003.
Alg = 2: xz-corrector, Adaptive Step-Differentiation, theta = 0.250, beta = 0.500
Split 30 free variables
eqs m = 200, order n = 461, dim = 461, blocks = 1
nnz(A) = 600 + 6000, nnz(ADA) = 400, nnz(L) = 300
Handling 60 + 0 dense columns.
 it :     b*y       gap    delta  rate   t/tP*  t/tD*   feas cg cg  prec
  0 :            3.46E+01 0.000
  1 :   5.46E+00 1.32E+01 0.000 0.3811 0.9000 0.9000   6.12  1  1  1.2E+00
  2 :   1.30E+01 7.01E+00 0.000 0.5309 0.9000 0.9000   1.33  1  1  6.5E-01
  3 :   1.70E+01 3.25E+00 0.000 0.4643 0.9000 0.9000   1.20  1  1  3.1E-01
  4 :   1.96E+01 1.30E+00 0.000 0.3994 0.9000 0.9000   1.10  1  1  1.2E-01
  5 :   2.07E+01 4.75E-01 0.000 0.3656 0.9000 0.9000   1.04  1  1  4.5E-02
  6 :   2.12E+01 1.56E-01 0.000 0.3290 0.9054 0.9000   1.02  1  1  1.4E-02
  7 :   2.14E+01 3.94E-02 0.000 0.2520 0.9000 0.9047   1.01  1  1  3.8E-03
  8 :   2.15E+01 6.65E-03 0.000 0.1687 0.9000 0.9095   1.01  1  1  7.5E-04
  9 :   2.15E+01 2.58E-04 0.000 0.0389 0.9900 0.9906   1.00  1  1  4.5E-05
 10 :   2.15E+01 8.56E-06 0.000 0.0331 0.9904 0.9900   1.00  1  1  
iter seconds digits       c*x               b*y
 10      0.1  15.0  2.1468211792e+01  2.1468211792e+01
|Ax-b| =   3.9e-13, [Ay-c]_+ =   5.2E-15, |x|=  1.2e+01, |y|=  9.5e+00

Detailed timing (sec)
   Pre          IPM          Post
4.000E-02    9.000E-02    0.000E+00    
Max-norms: ||b||=1.957607e+00, ||c|| = 1,
Cholesky |add|=0, |skip| = 0, ||L.L|| = 1.
------------------------------------------------------------
Status: Solved
Optimal value (cvx_optval): +21.4682
log-barrier
 
Calling sedumi: 300 variables, 131 equality constraints
   For improved efficiency, sedumi is solving the dual problem.
------------------------------------------------------------
SeDuMi 1.21 by AdvOL, 2005-2008 and Jos F. Sturm, 1998-2003.
Alg = 2: xz-corrector, Adaptive Step-Differentiation, theta = 0.250, beta = 0.500
eqs m = 131, order n = 301, dim = 301, blocks = 1
nnz(A) = 6500 + 0, nnz(ADA) = 7261, nnz(L) = 3696
 it :     b*y       gap    delta  rate   t/tP*  t/tD*   feas cg cg  prec
  0 :            9.36E+01 0.000
  1 :  -1.67E+01 7.44E+00 0.000 0.0795 0.9900 0.9900  -0.71  1  1  9.8E+01
  2 :  -7.32E+00 4.02E+00 0.000 0.5402 0.9000 0.9000   3.04  1  1  2.3E+01
  3 :  -2.19E+00 2.13E+00 0.000 0.5291 0.9000 0.9000   5.25  1  1  3.8E+00
  4 :  -1.52E+00 9.49E-01 0.000 0.4467 0.9000 0.9000   2.08  1  1  1.2E+00
  5 :  -1.33E+00 4.23E-01 0.000 0.4460 0.9000 0.9000   1.37  1  1  4.9E-01
  6 :  -1.25E+00 1.65E-01 0.000 0.3888 0.9000 0.9000   1.15  1  1  1.8E-01
  7 :  -1.22E+00 7.05E-02 0.000 0.4281 0.9000 0.9000   1.05  1  1  7.7E-02
  8 :  -1.21E+00 1.71E-02 0.000 0.2433 0.9327 0.9000   1.01  1  1  1.3E-02
  9 :  -1.20E+00 4.77E-03 0.000 0.2781 0.9000 0.9152   1.01  1  1  4.1E-03
 10 :  -1.20E+00 9.64E-04 0.000 0.2022 0.9244 0.9000   1.00  1  1  5.7E-04
 11 :  -1.20E+00 2.98E-05 0.000 0.0309 0.9903 0.9900   1.00  1  1  1.1E-05
 12 :  -1.20E+00 6.10E-06 0.000 0.2046 0.9156 0.9000   1.00  1  2  1.8E-06
 13 :  -1.20E+00 1.01E-06 0.000 0.1661 0.9000 0.9034   1.00  1  4  3.1E-07
 14 :  -1.20E+00 3.26E-09 0.000 0.0032 0.9990 0.9990   1.00  1  4  
iter seconds digits       c*x               b*y
 14      0.1  15.7 -1.2012704646e+00 -1.2012704646e+00
|Ax-b| =   2.4e-16, [Ay-c]_+ =   5.6E-16, |x|=  3.1e-01, |y|=  3.1e+00

Detailed timing (sec)
   Pre          IPM          Post
1.000E-02    7.000E-02    0.000E+00    
Max-norms: ||b||=1, ||c|| = 1.957607e+00,
Cholesky |add|=3, |skip| = 0, ||L.L|| = 1.86494.
------------------------------------------------------------
Status: Solved
Optimal value (cvx_optval): +1.20127