small improvements in tensor/tucker

t.cc

@@ -3683,7 +3683,7 @@ cin>>r>>n;
 INDEXGROUP shape;
         {
         shape.number=r;
-        shape.symmetry= -1;
+        shape.symmetry= 1;
         shape.range=n;
         shape.offset=0;
         }

tensor.cc

@@ -802,10 +802,30 @@ template<typename T>
 Tensor<T> Tensor<T>::flatten(int group) const
 {
 if(group>=shape.size()) laerror("too high group number in flatten");
-if(is_flat()) return *this;
+if(is_flat())
+	{
+	if(has_symmetry()) //get rid of formal symemtry
+		{
+		Tensor<T> r(*this);
+		r.shape.copyonwrite();
+                for(int g=0; g<r.shape.size(); ++g) r.shape[g].symmetry=0;
+                return r;
+		}
+	else
+		return *this;
+	}
 if(group>=0) //single group
 	{
-	if(shape[group].number==1) return *this;
+	if(shape[group].number==1) 
+		{
+		if(shape[group].symmetry==0) return *this;
+		else
+			{
+			Tensor<T> r(*this);
+			r.shape[group].symmetry=0;
+			return r;
+			}
+		}
 	if(shape[group].symmetry==0)
 		{
 		Tensor<T> r(*this);
@@ -816,7 +836,11 @@ if(group>=0) //single group
 if(group<0 && !is_compressed())
 	{
 	Tensor<T> r(*this);
-        for(int g=0; g<shape.size(); ++g) if(shape[g].number>1) r.split_index_group(g);
+        for(int g=0; g<shape.size(); ++g) 
+		{
+		if(shape[g].number>1) r.split_index_group(g);
+		}
+	for(int g=0; g<r.shape.size(); ++g) r.shape[g].symmetry=0;
         return r;
 	}
 
@@ -1285,7 +1309,7 @@ return r;
 
 
 //NOTE: Tucker of rank=2 is inherently inefficient - result is a diagonal tensor stored in full and 2 calls to SVD
-//we could avoid the second SVD, but the wasteful storage and erconstruction would remain
+//we could avoid the second SVD, but the wasteful storage and reconstruction would remain
 //
 template<typename T> 
 NRVec<NRMat<T> > Tensor<T>::Tucker(typename LA_traits<T>::normtype thr, bool inverseorder)
@@ -1314,9 +1338,9 @@ for(int i=0; i<r; ++i)
 	NRMat<T> um;
 	NRVec<indexgroup> ushape;
 	{
-	Tensor<T> u=unwind_index(I);
+	Tensor<T> uu=unwind_index(I);
-	ushape=u.shape; ushape.copyonwrite();
+	ushape=uu.shape; //ushape.copyonwrite(); should not be needed
-	um=u.matrix();
+	um=uu.matrix();
 	}
 	int mini=um.nrows(); if(um.ncols()<mini) mini=um.ncols(); //compact SVD, expect descendingly sorted values
 	NRMat<T> u(um.nrows(),mini),vt(mini,um.ncols());

tensor.h

@@ -47,6 +47,7 @@
 //@@@ will not be particularly efficient
 //
 //@@@conversions to/from fourindex, optional negarive range for beta spin handling
+//@@@use the fact that fourindex_dense is inherited from Mat/SMat and construct tensor from the (unsymmetrized) NRMat sharing data, just rewrite then the shape
 //
 //@@@?general permutation of individual indices - check the indices in sym groups remain adjacent, calculate result's shape, loopover the result and permute using unwind_callback
 //
@@ -173,6 +174,7 @@ public:
 
 	bool is_flat() const {for(int i=0; i<shape.size(); ++i) if(shape[i].number>1) return false; return true;};
 	bool is_compressed() const {for(int i=0; i<shape.size(); ++i) if(shape[i].number>1&&shape[i].symmetry!=0) return true; return false;};
+	bool has_symmetry() const {for(int i=0; i<shape.size(); ++i) if(shape[i].symmetry!=0) return true; return false;};
 	void clear() {data.clear();};
 	int rank() const {return myrank;};
 	int calcrank(); //is computed from shape