diff --git a/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h b/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h
index b4aa1ef2..538244c 100644
--- a/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h
+++ b/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h
@@ -751,15 +751,15 @@
 template<int StorageOrder,typename RealScalar, typename Scalar, typename Index>
 static void tridiagonal_qr_step(RealScalar* diag, RealScalar* subdiag, Index start, Index end, Scalar* matrixQ, Index n)
 {
-  // NOTE this version avoids over & underflow, however since the matrix is prescaled, overflow cannot occur,
-  // and underflows should be meaningless anyway. So I don't any reason to enable this version, but I keep
-  // it here for reference:
-//   RealScalar td = (diag[end-1] - diag[end])*RealScalar(0.5);
-//   RealScalar e = subdiag[end-1];
-//   RealScalar mu = diag[end] - (e / (td + (td>0 ? 1 : -1))) * (e / hypot(td,e));
   RealScalar td = (diag[end-1] - diag[end])*RealScalar(0.5);
-  RealScalar e2 = abs2(subdiag[end-1]);
-  RealScalar mu = diag[end] - e2 / (td + (td>0 ? 1 : -1) * sqrt(td*td + e2));
+  RealScalar e = subdiag[end-1];
+  // Note that thanks to scaling, e^2 or td^2 cannot overflow, however they can still
+  // underflow thus leading to inf/NaN values when using the following commented code:
+//   RealScalar e2 = abs2(subdiag[end-1]);
+//   RealScalar mu = diag[end] - e2 / (td + (td>0 ? 1 : -1) * sqrt(td*td + e2));
+  // This explain the following, somewhat more complicated, version:
+  RealScalar mu = diag[end] - (e / (td + (td>0 ? 1 : -1))) * (e / hypot(td,e));
+  
   RealScalar x = diag[start] - mu;
   RealScalar z = subdiag[start];
   for (Index k = start; k < end; ++k)