diff --git a/conjgrad.h b/conjgrad.h
index d16ca62..f181841 100644
--- a/conjgrad.h
+++ b/conjgrad.h
@@ -3,12 +3,69 @@
 #include "mat.h"
 #include "sparsemat.h"
 #include "nonclass.h"
+#include <iomanip>
 
 //conjugate gradient solution of a linear system
 
 //matrix can be any class which has nrows(), ncols(), diagonalof() and NRVec::gemv() available
 //does not even have to be explicitly stored
+//Conjugate gradient algorithm, cf. Bulirsch-Stoer book
 
 
 template<typename T, typename Matrix>
-extern void conjgrad(const Matrix &bigmat, const NRVec<T> &b, NRVec<T> &x, const bool doguess, const double tol, const int itmax, const bool verbose, bool issquare,const bool precondition);
+extern void conjgrad(const Matrix &bigmat, const NRVec<T> &b, NRVec<T> &x, const bool doguess, const double tol, const int itmax, const bool verbose, bool issquare,const bool precondition)
+{
+int m=bigmat.nrows();
+int n=bigmat.ncols();
+
+if(x.size()!=n || b.size() != m) laerror("incompatible vectors and matrix sizes in conjgrad");
+if(m!=n) issquare=0;
+
+double t,tt,bscal,ascal;
+
+NRVec<T> p,rr, *r;
+NRVec<T> q(m),s(m);
+if(issquare) r=&s; else r = new NRVec<T>(m);
+
+if(doguess)
+	{
+	x.gemv(0,bigmat,'t',-1.,b);
+        if(precondition) bigmat.diagonalof(x,true);
+        x.normalize();
+	}
+
+s.gemv(0,bigmat,'n',-1.,x);
+s+=b;
+if(!issquare) (*r).gemv(0,bigmat,'t',1,s);
+rr= *r;
+if(precondition) bigmat.diagonalof(rr,true);
+p=rr;
+
+for(int iter=0; iter<= itmax; iter++)
+	{
+	double err=p.norm();
+	if(verbose) cout << "conjgrad: iter= "<<iter<<" err= "<<
+		setiosflags(ios::scientific)<<setprecision(8) <<err<<
+		resetiosflags(ios::scientific)<<setprecision(12)<<"\n";
+	if(err <= tol) break;
+
+	q.gemv(0,bigmat,'n',1,p);
+	tt= (*r) * rr;
+	t=issquare?p*q:q*q;
+	if(!t) {if(!issquare) delete r; laerror("conjgrad: singular matrix 1");}
+	ascal=tt/t;
+	x.axpy(ascal,p);
+	s.axpy(-ascal,q);
+	if(!issquare) (*r).gemv(0,bigmat,'t',1,s);
+	rr= *r;
+	if(precondition) bigmat.diagonalof(rr,true);
+	if(!tt) {if(!issquare) delete r; laerror("conjgrad: singular matrix 2");}
+	bscal= ((*r)*rr)/tt;
+	rr.axpy(bscal,p);
+	p=rr;
+	}
+
+if(!issquare) delete r;
+
+}
+