compute_int(mu,j,delta,eta,LL,strategy,extra)

function int = compute_int(mu,j,delta,eta,LL,strategy,extra)
%COMPUTE_INT:  Determine which Lanczos vectors to reorthogonalize against.
%
%      int = compute_int(mu,eta,LL,strategy,extra))
%
%   Strategy 0: Orthogonalize vectors v_{i-r-extra},...,v_{i},...v_{i+s+extra} 
%               with nu>eta, where v_{i} are the vectors with  mu>delta.
%   Strategy 1: Orthogonalize all vectors v_{r-extra},...,v_{s+extra} where
%               v_{r} is the first and v_{s} the last Lanczos vector with
%               mu > eta.
%   Strategy 2: Orthogonalize all vectors with mu > eta.
%
%   Notice: The first LL vectors are excluded since the new Lanczos
%   vector is already orthogonalized against them in the main iteration.

%  Rasmus Munk Larsen, DAIMI, 1998.

if (delta<eta)
  error(' compute_int   DELTA >= ETA.')
end
switch strategy
  case 0
    I0 = find(abs(mu(1:j))>=delta);    
    if length(I0)==0
      [mm,I0] = max(abs(mu(1:j)));
    end    
    int = zeros(j,1);
    for i = 1:length(I0)
      for r=I0(i):-1:1
	if abs(mu(r))<eta | int(r)==1 
	  break;
	else
	  int(r) = 1;
	end
      end
      int(max(1,r-extra+1):r) = 1;
      for s=I0(i)+1:j
	if abs(mu(s))<eta | int(s)==1  
	  break;
	else
	  int(s) = 1;
	end
      end
      int(s:min(j,s+extra-1)) = 1;
    end
    if LL>0
      int(1:LL) = 0;
    end
    int = find(int);
  case 1
    int=find(abs(mu(1:j))>eta);
    int = max(LL+1,min(int)-extra):min(max(int)+extra,j);
  case 2
    int=find(abs(mu(1:j))>=eta);
end
int = int(:);