It's OK to do this

cout << ((MatrixXf(1,2) << 0, 2).finished()) * ((MatrixXf(2,1) << 0, 2).finished()) << endl;

But when I try to do this

MatrixXf mtemp(2,1);
mtemp(0,0) = ((MatrixXf(1,2) << 0, 2).finished()) * ((MatrixXf(2,1) << 0, 2).finished());

It said 'cannot convert const Eigen::GeneralProduct to float in assignment'.

Because the result of the matrix product is a matrix. If you use vectors instead of dynamically sized matrices, then Eigen will predict that the result is a 1x1 matrix and allow you to do the conversion. Also using vectors is much faster.