1: \begin{abstract} 

2: 

3: This paper proves the global convergence of a triangularized

4: orthogonalization-free method (\TOM{}). \TOM{}, in general, applies

5: a triangularization idea to the gradient of an objective function and

6: removes the rotation invariance in minimizers. More precisely, in this

7: paper, the \TOM{} works as an eigensolver for sizeable sparse matrices

8: and obtains eigenvectors without any orthogonalization step. Due to the

9: triangularization, the iteration is a discrete-time flow in a non-conservative

10: vector field. The global convergence relies on the stable manifold theorem,

11: whereas the convergence to stationary points is proved in detail in this

12: paper. We provide two proofs inspired by the noisy power method and the

13: noisy optimization method, respectively.

14: 

15: \end{abstract}

16: