Generalize TA::squared_norm to be applicable for tensor-of-tensor a…

…rrays as well.

src/TiledArray/dist_array.h

@@ -1213,6 +1213,17 @@ class DistArray : public madness::archive::ParallelSerializableObject {
     return TiledArray::expressions::TsrExpr<const DistArray>(*this, vars);
   }
 
+  ///
+  /// \brief This method creates a tensor expression but does not insist the
+  ///        annotation to be bipartite (outer and inner tensor annotations).
+  /// \param vars Annotation for the tensor expression.
+  /// \note Only use for unary evaluations when the indexing of the inner
+  ///       tensors is not significant, eg. norm computation.
+  ///
+  auto index_unchecked_tensor_expression(const std::string& vars) const {
+    return TiledArray::expressions::TsrExpr<const DistArray>(*this, vars);
+  }
+
   /// Create a tensor expression
 
   /// \param vars A string with a comma-separated list of variables
@@ -1917,7 +1928,8 @@ auto inner_product(const DistArray<Tile, Policy>& a,
 
 template <typename Tile, typename Policy>
 auto squared_norm(const DistArray<Tile, Policy>& a) {
-  return a(detail::dummy_annotation(rank(a))).squared_norm();
+  return a.index_unchecked_tensor_expression(detail::dummy_annotation(rank(a)))
+      .squared_norm();
 }
 
 template <typename Tile, typename Policy>