diff --git a/contfrac.cc b/contfrac.cc
index 50d505a..7f49c9b 100644
--- a/contfrac.cc
+++ b/contfrac.cc
@@ -47,15 +47,20 @@ static void cf_helper(ContFrac<T> *me, T p, T q, int level)
 T div=p/q;
 {
 T rem=p%q;
-if(rem) cf_helper(me,q,rem,level+1);
+if(rem) 
+	{
+	if(rem<0) {--div; rem+=q;} //prevent negative a_i i>0
+	cf_helper(me,q,rem,level+1);
+	}
 else me->resize(level);
 }
 (*me)[level]=div;
 }
 
 template <typename T>
-ContFrac<T>::ContFrac(const T p, const T q) : NRVec<T>()
+ContFrac<T>::ContFrac(T p, T q) : NRVec<T>()
 {
+if(q<0) {p= -p; q= -q;}
 cf_helper<T>(this,p,q,0);
 }
 
@@ -117,11 +122,33 @@ return x;
 }
 
 
+//compare assuming they are canonical
+template <typename T>
+T ContFrac<T>::compare(const ContFrac<T> &rhs) const
+{
+int l=length();
+if(rhs.length()<l) l=rhs.length();
+for(int i=0; i<=l; ++i)
+	{
+	T d=(*this)[i]-rhs[i];
+	if(d) return (i&1)? -d :d;
+	}
+if(length()==rhs.length()) return 0;
+else if(length()<rhs.length()) return (length()&1) ? 1 : -1;
+else return (rhs.length()&1) ? -1 : 1;
+}
+
+
 template <typename T>
 void ContFrac<T>::canonicalize()
 {
 int n=this->length();
 if(n==0) return;
+if(n>0 && (*this)[1]<0) //handle negative a_i i>0
+	{
+	for(int i=0; i<=n; ++i) (*this[i]) = -(*this[i]);
+	*this = -(*this);
+	}
 this->copyonwrite();
 if((*this)[n]==1) {(*this)[n]=0; ++(*this)[n-1];} //avoid deepest 1/1
 for(int i=1; i<=n; ++i) //truncate if possible
@@ -167,13 +194,18 @@ return hnew;
 
 
 template <typename T>
-bool Homographic<T>::outputready(T &z) const
+bool Homographic<T>::outputready(T &z,bool first) const
 {
 bool inf=0;
 T q0,q1;
 if(v[1][0]==0) inf=1; else q0=v[0][0]/v[1][0];
 if(v[1][1]==0) inf=1; else q1=v[0][1]/v[1][1];
-if(!inf && q0==q1) {z=q0; return true;}
+if(!inf && q0==q1) 
+	{
+	z=q0; 
+	if(first && q0<0) --z; //prevent negative a1 etc.
+	return true;
+	}
 return false;
 }
 
@@ -190,6 +222,7 @@ ContFrac<T> Homographic<T>::value(const ContFrac<T>&x) const
 Homographic<T> h(*this);
 
 std::list<T> l;
+bool first=true;
 for(typename ContFrac<T>::iterator px=x.begin(); px!=x.beyondend(); ++px)
 	{
 	//digest next input term
@@ -197,10 +230,11 @@ for(typename ContFrac<T>::iterator px=x.begin(); px!=x.beyondend(); ++px)
 
 	//output as much as possible
 	T out;
-	while(h.outputready(out))
+	while(h.outputready(out,first))
 		{
 		l.push_back(out);
 		h=h.output(out);
+		first=false;
 		}
 
 	//terminate if exhausted
@@ -210,6 +244,11 @@ for(typename ContFrac<T>::iterator px=x.begin(); px!=x.beyondend(); ++px)
 		break;
 		}
 	}
+if(l.back()==1)  //simplify by removing a trailing 1
+	{
+	l.pop_back();
+	l.back()+=1;
+	}
 return ContFrac<T>(l);
 }
 
@@ -273,7 +312,7 @@ return 1;
 
 
 template <typename T>
-bool BiHomographic<T>::outputready(T &z) const
+bool BiHomographic<T>::outputready(T &z,bool first) const
 {
 T q[2][2];
 for(int i=0; i<2; ++i) for(int j=0; j<2; ++j)
@@ -283,6 +322,7 @@ for(int i=0; i<2; ++i) for(int j=0; j<2; ++j)
 	if(q[i][j]!=q[0][0])  return false;
 	}
 z=q[0][0]; 
+if(first && z<0) --z;
 return true;
 }
 
@@ -303,6 +343,7 @@ std::list<T> l;
 typename ContFrac<T>::iterator px=x.begin();
 typename ContFrac<T>::iterator py=y.begin();
 
+bool first=true;
 do
 	{
 	//select next input term
@@ -316,10 +357,11 @@ do
 
 	//output as much as possible
 	T out;
-	while(h.outputready(out))
+	while(h.outputready(out,first))
 		{
 		l.push_back(out);
 		h=h.output(out);
+		first=false;
 		}
 
 	//terminate if exhausted
@@ -330,6 +372,13 @@ do
 		}
 	}
 	while(px!=x.beyondend() || py!=y.beyondend());
+
+if(l.back()==1)  //simplify by removing a trailing 1
+        {
+        l.pop_back();
+        l.back()+=1;
+        }
+
 return ContFrac<T>(l);
 }
 
@@ -344,7 +393,7 @@ if(den<0)
 	den= -den;
 	}
 T g=gcd(num,den); 
-if(MYABS(g)>1) 
+if(g>1) 
 	{
 	num/=g; 
 	den/=g;
@@ -381,6 +430,7 @@ return *this;
 }
 
 
+//try avoiding overflows at the cost of speed
 template <typename T>
 Rational<T>  Rational<T>::operator+(const Rational &rhs) const
 {
@@ -426,6 +476,45 @@ return *this;
 }
 
 
+//unary -
+template <typename T>
+ContFrac<T> ContFrac<T>::operator-() const
+{
+int l=length();
+if(l==0)
+	{
+	ContFrac<T> r(0);
+	r[0]= -(*this)[0];
+	return r;
+	}
+if((*this)[1]!=1)
+	{
+	ContFrac<T> r(l+1);
+	r[0]= -(*this)[0]-1;
+	r[1]= 1;
+	r[2]= (*this)[1]-1;
+	for(int i=2; i<=l; ++i) r[i+1] = (*this)[i];
+	return r;
+	}
+else //a_1-1 == 0
+	{
+	if(l==1) //we have trailing 0, actually the input was not canonical
+		{
+		ContFrac<T> r(0);
+        	r[0]= -(*this)[0]-1;
+		return r;
+		}
+	else
+		{
+		ContFrac<T> r(l-1);
+		r[0]= -(*this)[0]-1;
+		r[1]= 1+(*this)[2];
+		for(int i=3; i<=l; ++i) r[i-1] = (*this)[i];
+		return r;
+		}
+	}
+}
+
 
 
 /***************************************************************************//**
diff --git a/contfrac.h b/contfrac.h
index 736ebd7..e463121 100644
--- a/contfrac.h
+++ b/contfrac.h
@@ -25,17 +25,16 @@
 
 namespace LA {
 
-//simple finite continued fraction class
+//Support for rationals and a simple finite continued fraction class
 //NOTE: 0 on any position >0 means actually infinity; simplify()  shortens the vector
-//presently implements just conversion to/from rationals and floats
-//maybe implement arithmetic by Gosper's method cf. https://perl.plover.com/classes/cftalk/TALK
-//
+//includes Gosper's arithmetics - cf. https://perl.plover.com/classes/cftalk/TALK
+//maybe implement the self-feeding  Gosper's algorithm for sqrt(int)
+//maybe do not interpret a_i=0 i>0 as termination???
 
 template <typename T>
 class ContFrac;
 
 
-//@@@operator== > >= etc.
 template <typename T>
 class Rational {
 public:
@@ -45,7 +44,7 @@ public:
 	Rational() {};
 	Rational(const T p, const T q) : num(p),den(q) {};
 	explicit Rational(const T (&a)[2]) :num(a[0]), den(a[1]) {};
-	Rational(const ContFrac<T> &cf) {cf.convergent(&num,&den);};
+	explicit Rational(const ContFrac<T> &cf) {cf.convergent(&num,&den);};
 	void simplify();
 
 	//basic rational arithmetics 
@@ -67,7 +66,21 @@ public:
 	Rational & operator+=(const Rational &rhs) {*this = *this+rhs; return *this;};
 	Rational & operator-=(const Rational &rhs) {*this = *this-rhs; return *this;};
 
-	
+	//combination with continued fractions
+	ContFrac<T> operator+(const ContFrac<T> &rhs) const {return rhs + *this;};
+	ContFrac<T> operator-(const ContFrac<T> &rhs) const {return -rhs + *this;};
+	ContFrac<T> operator*(const ContFrac<T> &rhs) const {return rhs * *this;};
+	ContFrac<T> operator/(const ContFrac<T> &rhs) const {return rhs.reciprocal() * *this;};
+
+
+	//relational operators, relying that operator- yields a form with a positive denominator
+	bool operator==(const Rational  &rhs) const {Rational t= *this-rhs; return t.num==0;};
+	bool operator!=(const Rational  &rhs) const {Rational t= *this-rhs; return t.num!=0;};
+	bool operator>=(const Rational  &rhs) const {Rational t= *this-rhs; return t.num>=0;};
+	bool operator<=(const Rational  &rhs) const {Rational t= *this-rhs; return t.num<=0;};
+	bool operator>(const Rational  &rhs) const {Rational t= *this-rhs; return t.num>0;};
+        bool operator<(const Rational  &rhs) const {Rational t= *this-rhs; return t.num<0;};
+
 };
 
 template <typename T>
@@ -96,7 +109,6 @@ class BiHomographic;
 
 
 
-//@@@operator== > >= etc.
 template <typename T>
 class ContFrac : public NRVec<T> {
 private:
@@ -106,16 +118,17 @@ public:
 	template<int SIZE>  ContFrac(const T (&a)[SIZE]) : NRVec<T>(a) {};
 	ContFrac(const NRVec<T> &v) : NRVec<T>(v) {}; //allow implicit conversion from NRVec
 	ContFrac(const int n) : NRVec<T>(n+1) {};
-	ContFrac(double x, const int n, const T thres=0); //might yield a non-canonical form 
+	explicit ContFrac(double x, const int n, const T thres=0); //WARNING: it might yield a non-canonical form 
 	//we could make a template for analogous conversion from an arbitrary-precision type
-	ContFrac(const T p, const T q); //should yield a canonical form
-	ContFrac(const Rational<T> &r) : ContFrac(r.num,r.den) {};
+	ContFrac(T p, T q); //should yield a canonical form
+	explicit ContFrac(const Rational<T> &r) : ContFrac(r.num,r.den) {};
 
 	void canonicalize();
 	void convergent(T *p, T*q, const int trunc= -1) const;
 	Rational<T> rational(const int trunc= -1) const {T p,q; convergent(&p,&q,trunc); return Rational<T>(p,q);};
 	double value(const int trunc= -1) const; //we could make also a template usable with an arbitrary-precision type
 	ContFrac reciprocal() const;
+	ContFrac operator-() const; //unary minus
 	int length() const {return NRVec<T>::size()-1;};
 	void resize(const int n, const bool preserve=true) 
 		{
@@ -123,6 +136,7 @@ public:
 		NRVec<T>::resize(n+1,preserve);
 		if(preserve) for(int i=nold+1; i<=n;++i) (*this)[i]=0;
 		} 
+
 	//arithmetics with a single ContFrac operand
 	ContFrac operator+(const Rational<T> &rhs) const {Homographic<T> h({{rhs.num,rhs.den},{rhs.den,0}}); return h.value(*this);};
 	ContFrac operator-(const Rational<T> &rhs) const {Homographic<T> h({{-rhs.num,rhs.den},{rhs.den,0}}); return h.value(*this);};
@@ -142,6 +156,16 @@ public:
 	ContFrac operator*(const ContFrac &rhs) const {BiHomographic<T> h({{{0,0},{0,1}},{{1,0},{0,0}}}); return h.value(*this,rhs);};
 	ContFrac operator/(const ContFrac &rhs) const {BiHomographic<T> h({{{0,1},{0,0}},{{0,0},{1,0}}}); return h.value(*this,rhs);};
 
+	
+	//relational operators, guaranteed only to work correctly for canonicalized CF!
+	T compare(const ContFrac  &rhs) const;
+	bool operator==(const ContFrac  &rhs) const {return compare(rhs)==0;};
+	bool operator>(const ContFrac  &rhs) const {return compare(rhs)>0;};
+	bool operator<(const ContFrac  &rhs) const {return rhs.operator>(*this);};
+	bool operator!=(const ContFrac  &rhs) const {return ! this->operator==(rhs) ;}
+	bool operator<=(const ContFrac  &rhs) const {return ! this->operator>(rhs) ;}
+	bool operator>=(const ContFrac  &rhs) const {return ! this->operator<(rhs) ;}
+
 	//iterator
         class iterator {
         private:
@@ -176,7 +200,7 @@ T v[2][2]; //{{a,b},{c,d}} for (a+b.x)/(c+d.x) i.e. [denominator][power_x]
 
 	Homographic input(const T &x, const bool inf) const; 
 	Homographic output(const T &x) const;
-	bool outputready(T &x) const;
+	bool outputready(T &x, bool first) const;
 	bool terminate() const;
 	
 };
@@ -195,7 +219,7 @@ T v[2][2][2];  //{{{a,b},{c,d}},{{e,f},{g,h}}} i.e.[denominator][power_y][power_
 	BiHomographic inputy(const T &y, const bool inf) const;
         BiHomographic output(const T &z) const;
 	int inputselect() const;
-        bool outputready(T &x) const;
+        bool outputready(T &x,bool first) const;
         bool terminate() const;
 
 };
diff --git a/t.cc b/t.cc
index 1a21e57..d7ef878 100644
--- a/t.cc
+++ b/t.cc
@@ -2483,7 +2483,7 @@ ContFrac<int> z= x*Rational<int>({2,3});
 cout<<Rational<int>(z)<<endl;
 }
 
-if(1)
+if(0)
 {
 ContFrac<int> x(11,101);
 ContFrac<int> v(3,7);
@@ -2497,6 +2497,49 @@ cout<<Rational<int>(zz)<<endl;
 cout<<(Rational<int>(x)+3)*(Rational<int>(v)+4)/(Rational<int>(x)-Rational<int>(v))<<endl;
 }
 
+if(0)
+{
+double x;
+cin >>x;
+ContFrac<int> xx(x,15,100000);
+cout <<xx<<endl;
+ContFrac<int> xm= -xx;
+cout<< xm<<endl;
+cout<<xx+xm<<endl;
+}
+
+if(0)
+{
+Rational<int> x;
+cin >>x;
+ContFrac<int> xx(x);
+cout<<xx;
+ContFrac<int> xm= -xx;
+cout<< xm;
+ContFrac<int> yy(-x);
+cout<<yy;
+ContFrac<int> ym= -yy;
+cout<< ym;
+cout <<"ZEROs\n"<<xx+xm<<" "<<yy+ym<<" "<<xx+yy<<" "<<xm+ym<<endl;
+}
+
+if(1)
+{
+double x;
+cin >>x;
+ContFrac<int> xx(x,25,100000);
+ContFrac<int> xx1(x+1e-4,25,100000);
+ContFrac<int> xx2(x-1e-4,25,100000);
+xx.canonicalize();
+xx1.canonicalize();
+xx2.canonicalize();
+cout<<"small "<<xx2<<endl;
+cout<<"middle "<<xx<<endl;
+cout<<"big "<<xx1<<endl;
+cout << "TEST "<<(xx<xx1) <<" "<<(xx>xx2) <<endl;
+
+}
+
 
 
 }