diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h b/unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h
index 24a0df820..c7af02b11 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorBroadcasting.h
@@ -99,6 +99,10 @@ struct TensorEvaluator<const TensorBroadcastingOp<Broadcast, ArgType>, Device>
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
     : m_impl(op.expression(), device)
   {
+    // The broadcasting op doesn't change the rank of the tensor. One can't broadcast a scalar
+    // and store the result in a scalar. Instead one should reshape the scalar into a a N-D
+    // tensor with N >= 1 of 1 element first and then broadcast.
+    EIGEN_STATIC_ASSERT(NumDims > 0, YOU_MADE_A_PROGRAMMING_MISTAKE);
     const typename TensorEvaluator<ArgType, Device>::Dimensions& input_dims = m_impl.dimensions();
     const Broadcast& broadcast = op.broadcast();
     for (int i = 0; i < NumDims; ++i) {
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorConcatenation.h b/unsupported/Eigen/CXX11/src/Tensor/TensorConcatenation.h
index fa05cab30..3d153bb94 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorConcatenation.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorConcatenation.h
@@ -131,7 +131,9 @@ struct TensorEvaluator<const TensorConcatenationOp<Axis, LeftArgType, RightArgTy
     : m_leftImpl(op.lhsExpression(), device), m_rightImpl(op.rhsExpression(), device), m_axis(op.axis())
   {
     EIGEN_STATIC_ASSERT((static_cast<int>(TensorEvaluator<LeftArgType, Device>::Layout) == static_cast<int>(TensorEvaluator<RightArgType, Device>::Layout) || NumDims == 1), YOU_MADE_A_PROGRAMMING_MISTAKE);
-    EIGEN_STATIC_ASSERT(NumDims == RightNumDims, YOU_MADE_A_PROGRAMMING_MISTAKE)
+    EIGEN_STATIC_ASSERT(NumDims == RightNumDims, YOU_MADE_A_PROGRAMMING_MISTAKE);
+    EIGEN_STATIC_ASSERT(NumDims > 0, YOU_MADE_A_PROGRAMMING_MISTAKE);
+
     eigen_assert(0 <= m_axis && m_axis < NumDims);
     const Dimensions& lhs_dims = m_leftImpl.dimensions();
     const Dimensions& rhs_dims = m_rightImpl.dimensions();
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorPadding.h b/unsupported/Eigen/CXX11/src/Tensor/TensorPadding.h
index 07a6e8d4c..91e32d200 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorPadding.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorPadding.h
@@ -98,6 +98,11 @@ struct TensorEvaluator<const TensorPaddingOp<PaddingDimensions, ArgType>, Device
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device), m_padding(op.padding())
   {
+    // The padding op doesn't change the rank of the tensor. Directly padding a scalar would lead
+    // to a vector, which doesn't make sense. Instead one should reshape the scalar into a vector
+    // of 1 element first and then pad.
+    EIGEN_STATIC_ASSERT(NumDims > 0, YOU_MADE_A_PROGRAMMING_MISTAKE);
+
     // Compute dimensions
     m_dimensions = m_impl.dimensions();
     for (int i = 0; i < NumDims; ++i) {