function demo10b
% *** RCIP: composed operators ***
  close all
  format long
  format compact
%
% *** user specified quantities **********
  theta=pi/2     % corner opening angle
  npan=11;       % number of coarse panels
  evec=1;        % a unit vector 
% ****************************************
  T16=Tinit16;
  W16=Winit16;
  [IP,IPW]=IPWinit(T16,W16);
%
  Pbc=zeros(96,64);
  Pbc( 1:16, 1:16)=eye(16);
  Pbc(17:48,17:32)=IP;
  Pbc(49:96,33:64)=flipud(fliplr(Pbc(1:48,1:32)));
%
  PWbc=zeros(96,64);
  PWbc( 1:16, 1:16)=eye(16);
  PWbc(17:48,17:32)=IPW;
  PWbc(49:96,33:64)=flipud(fliplr(PWbc(1:48,1:32)));
%
  PbcBig=zeros(192,128);
  PbcBig(1:96,1:64)=Pbc;
  PbcBig(97:192,65:128)=Pbc;
%
  PwbcBig=zeros(192,128);
  PWbcBig(1:96,1:64)=PWbc;
  PWbcBig(97:192,65:128)=PWbc;
%
% *** panel breakpoints and interval lengths in parameter ***
  sinter=linspace(0,1,npan+1);
  sinterdiff=ones(npan,1)/npan;
%
% *** discretization points and weights ***  
  npcoa=16*npan
  scoa=zeros(npcoa,1);
  wcoa=zeros(npcoa,1);
  for k=1:npan
    myind=k*16-15:k*16;
    sdif=sinterdiff(k)/2;
    scoa(myind)=(sinter(k)+sinter(k+1))/2+sdif*T16;
    wcoa(myind)=W16*sdif;
  end
  zcoa=zfunc(scoa,theta) ;
  zpcoa=zpfunc(scoa,theta);
  zppcoa=zppfunc(scoa,theta);
% *** some extra presicion gained from symmetry ***
  zcoa(npcoa/2+1:npcoa)=conj(flipud(zcoa(1:npcoa/2)));
  zpcoa(npcoa/2+1:npcoa)=-conj(flipud(zpcoa(1:npcoa/2)));
  zppcoa(npcoa/2+1:npcoa)=conj(flipud(zppcoa(1:npcoa/2)));
% *************************************************
  nzcoa=-1i*zpcoa./abs(zpcoa);
  wzpcoa=wcoa.*zpcoa;
%
% *** The M_coa^{\circ} and K_coa^{\circ} matrices are set up ***
  Mcirccoa=M1Rinit(zcoa,zpcoa,zppcoa,wcoa,wzpcoa,npcoa);
  Kcirccoa=M1Ainit(zcoa,zpcoa,zppcoa,nzcoa,wcoa,wzpcoa,npcoa);
  starind=[npcoa-31:npcoa 1:32];
  Mcirccoa(starind,starind)=zeros(64);
  Kcirccoa(starind,starind)=zeros(64);
%
  for nsub=1:100
% *** recursion for the R matrix ***
    for level=1:nsub
      sloc=[(T16+1)/4;T16/4+0.75;T16/2+1.5]/2^(nsub-level)/npan;
      wloc=[W16/4;W16/4;W16/2]/2^(nsub-level)/npan;
      zloc=zfunc(sloc,theta);
      zploc=zpfunc(sloc,theta);
      zpploc=zppfunc(sloc,theta);
      zloc=[conj(flipud(zloc));zloc];
      zploc=[-conj(flipud(zploc));zploc];
      zpploc=[conj(flipud(zpploc));zpploc];
      wloc=[flipud(wloc);wloc];
      nzloc=-1i*zploc./abs(zploc);
      wzploc=wloc.*zploc;
      Mloc=M1Rinit(zloc,zploc,zpploc,wloc,wzploc,96);
      Kloc=M1Ainit(zloc,zploc,zpploc,nzloc,wloc,wzploc,96);
      MKloc=zeros(192);
      MKloc(1:96,97:192)= Mloc;
      MKloc(97:192,1:96)=-Kloc;
      MAT=eye(192)+MKloc;
      if level==1
	R=inv(MAT([17:80 113:176],[17:80 113:176]));
      end
      R=SchurBana(PWbcBig,PbcBig,MAT,R);
    end
%
    R1=R(1:64,1:64);
    R2=R(65:128,1:64);
    R3=R(1:64,65:128);
    R4=R(65:128,65:128);
    R13=R1\R3;
    R4213=R4-R2*R13;
%
    R1G=speye(npcoa);
    R1G(starind,starind)=R1;
    R4213G=speye(npcoa);
    R4213G(starind,starind)=R4213;
    R2G=sparse(npcoa);
    R2G(starind,starind)=R2;
    R13G=sparse(npcoa);
    R13G(starind,starind)=R13;
%
% *** Solving main linear system ***
    g=2*real(conj(evec)*nzcoa);     
    [rhotildecoa,it]=myGMRESR(Mcirccoa,Kcirccoa,R1G,R2G,R13G, ...
			      R4213G,g,npcoa,100,eps);
    GMRES_iter_RCIP=it
%
% *** Post processing ***
    rhohatcoa=R1G*rhotildecoa;
    zeta=real(conj(evec)*zcoa.*abs(wzpcoa));
    qcoa=rhohatcoa'*zeta
    qref=1.95243329423584
    myerr=abs(qref-qcoa)/abs(qref);
    myerr(myerr<eps)=eps;
%
    set(0,'DefaultAxesFontSize',16)
    set(0,'DefaultAxesLineWidth',1) 
%
    figure(1)
    semilogy(nsub,myerr,'*')
    axis([0 100 1e-16 1e0])
    title('Convergence of {\itq} with mesh refinement','FontSize',16)
    xlabel('Number of subdivisions {\itn}_{sub}','FontSize',16)
    ylabel('Estimated relative error in {\itq}','FontSize',16)  
    axis square
    grid on  
    hold on
%  
    figure(2)
    semilogy(nsub,it,'*')
    axis([0 100 1 100])
    title('GMRES iterations for full convergence','FontSize',16)
    xlabel('Number of subdivisions {\itn}_{sub}','FontSize',16)
    ylabel('Number of iterations ','FontSize',16)  
    axis square
    grid on  
    hold on  
  end

  function A=SchurBana(PW,P,K,A)
  starL=[17:80 113:176];
  circL=[1:16 81:112 177:192];
  starS=[17:48 81:112];
  circS=[1:16 49:80 113:128];
  VA=K(circL,starL)*A;
  PTA=PW(starL,starS)'*A;
  PTAU=PTA*K(starL,circL);
  DVAUI=inv(K(circL,circL)-VA*K(starL,circL));
  DVAUIVAP=DVAUI*(VA*P(starL,starS));
  A(starS,starS)=PTA*P(starL,starS)+PTAU*DVAUIVAP;
  A(circS,circS)=DVAUI;
  A(circS,starS)=-DVAUIVAP;
  A(starS,circS)=-PTAU*DVAUI;
  
  function [x,it]=myGMRESR(A,B,R1,R2,R13,R4213,b,n,m,tol)
% *** GMRES with low-threshold stagnation control ***
  V=zeros(n,m+1);
  H=zeros(m);
  cs=zeros(m,1);
  sn=zeros(m,1);
  bnrm2=norm(b);
  V(:,1)=b/bnrm2;
  s=bnrm2*eye(m+1,1);
  for it = 1:m                                  
    it1=it+1;                                   
    w=A*((R4213*(B*(R1*V(:,it))))+R2*V(:,it))-R13*(B*(R1*V(:,it)));
    for k = 1:it
      H(k,it)=V(:,k)'*w;
      w=w-H(k,it)*V(:,k);
    end
    H(it,it)=H(it,it)+1;
    wnrm2=norm(w);
    V(:,it1)=w/wnrm2;
    for k=1:it-1                                
      temp     = cs(k)*H(k,it)+sn(k)*H(k+1,it);
      H(k+1,it)=-sn(k)*H(k,it)+cs(k)*H(k+1,it);
      H(k,it)  = temp;
    end
    [cs(it),sn(it)]=rotmat(H(it,it),wnrm2);     
    H(it,it)= cs(it)*H(it,it)+sn(it)*wnrm2;
    s(it1) =-sn(it)*s(it);                      
    s(it)  = cs(it)*s(it);                         
    myerr=abs(s(it1))/bnrm2;
    if (myerr<=tol)|(it==m)                     
      predicted_residual=myerr
      y=triu(H(1:it,1:it))\s(1:it);             
      x=fliplr(V(:,1:it))*flipud(y);
      true_residual=norm(x+A*((R4213*(B*(R1*x)))+R2*x)-R13*(B*(R1*x))-b)/bnrm2
      break;
    end
  end

  function [c,s]=rotmat(a,b)
  if  b==0
    c=1;
    s=0;
  elseif abs(b)>abs(a)
    temp=a/b;
    s=1/sqrt(1+temp^2);
    c=temp*s;
  else
    temp=b/a;
    c=1/sqrt(1+temp^2);
    s=temp*c;
  end

  function M1=M1Ainit(z,zp,zpp,nz,w,wzp,N)
% *** adjoint of double layer potential ***   
  M1=zeros(N);
  for m=1:N
    M1(:,m)=abs(wzp(m))*real(nz./(z(m)-z));
    M1(m,m)=-w(m)*imag(zpp(m)/zp(m))/2;      
  end
  M1=M1/pi;

  function M1=M1Rinit(z,zp,zpp,w,wzp,N)
% *** double layer potential ***   
  M1=zeros(N);
  for m=1:N
    M1(:,m)=imag(wzp(m)./(z(m)-z));
    M1(m,m)=w(m)*imag(zpp(m)/zp(m))/2;
  end
  M1=M1/pi;
  
  function zout=zfunc(s,theta)
  zout=sin(pi*s).*exp(1i*theta*(s-0.5));

  function zpout=zpfunc(s,theta)
  zpout=(pi*cos(pi*s)+1i*theta*sin(pi*s)).*exp(1i*theta*(s-0.5));
  
  function zppout=zppfunc(s,theta)
  zppout=(2i*pi*theta*cos(pi*s)-(theta^2+pi^2)*sin(pi*s)).* ...
	 exp(1i*theta*(s-0.5));  

  function T=Tinit16
% *** 16-point Gauss-Legendre nodes ***  
  T=zeros(16,1);
  T( 1)=-0.989400934991649932596154173450332627;
  T( 2)=-0.944575023073232576077988415534608345;
  T( 3)=-0.865631202387831743880467897712393132;
  T( 4)=-0.755404408355003033895101194847442268;
  T( 5)=-0.617876244402643748446671764048791019;
  T( 6)=-0.458016777657227386342419442983577574;
  T( 7)=-0.281603550779258913230460501460496106;
  T( 8)=-0.095012509837637440185319335424958063;
  T( 9)= 0.095012509837637440185319335424958063;
  T(10)= 0.281603550779258913230460501460496106;
  T(11)= 0.458016777657227386342419442983577574;
  T(12)= 0.617876244402643748446671764048791019;
  T(13)= 0.755404408355003033895101194847442268;
  T(14)= 0.865631202387831743880467897712393132;
  T(15)= 0.944575023073232576077988415534608345;
  T(16)= 0.989400934991649932596154173450332627;

  function W=Winit16
% *** 16-point Gauss-Legendre weights ***  
  W=zeros(16,1); 
  W( 1)= 0.027152459411754094851780572456018104;
  W( 2)= 0.062253523938647892862843836994377694;
  W( 3)= 0.095158511682492784809925107602246226;
  W( 4)= 0.124628971255533872052476282192016420;
  W( 5)= 0.149595988816576732081501730547478549;
  W( 6)= 0.169156519395002538189312079030359962;
  W( 7)= 0.182603415044923588866763667969219939;
  W( 8)= 0.189450610455068496285396723208283105;
  W( 9)= 0.189450610455068496285396723208283105;
  W(10)= 0.182603415044923588866763667969219939;
  W(11)= 0.169156519395002538189312079030359962;
  W(12)= 0.149595988816576732081501730547478549;
  W(13)= 0.124628971255533872052476282192016420;
  W(14)= 0.095158511682492784809925107602246226;
  W(15)= 0.062253523938647892862843836994377694;
  W(16)= 0.027152459411754094851780572456018104;
  
  function [IP,IPW]=IPWinit(T16,W16)
  A=ones(16);
  AA=ones(32,16);
  T32=[(T16-1)/2;(T16+1)/2];
  W32=[W16;W16]/2;
  for k=2:16
    A(:,k)=A(:,k-1).*T16;
    AA(:,k)=AA(:,k-1).*T32;         
  end
  IP=AA/A;
  IPW=IP.*(W32*(1./W16)');