1: \begin{abstract}

2:     In this paper, we study asymptotic behaviors of continuous-time and discrete-time gradient flows of 

3:     a ``lower-unbounded" convex function $f$ on a Hadamard manifold $M$, particularly,  

4:     their convergence properties to the boundary $M^{\infty}$ at infinity of $M$.

5:     %

6:     We establish a duality theorem that the infimum of the gradient-norm $\|\nabla f(x)\|$ of $f$ 

7:     over $M$ 

8:     is equal to the supremum of the negative of the recession function $f^{\infty}$ 

9:     of $f$ over the boundary $M^{\infty}$, provided the infimum is positive.

10:     Further, 

11:     the infimum and the supremum are

12:     obtained by the limits of the gradient flows of $f$, 

13:     %

14:      Our results feature convex-optimization ingredients of the moment-weight inequality 

15:      for reductive group actions by Georgoulas, Robbin, and Salamon, 

16:      and are applied to noncommutative optimization by B\"urgisser et al. FOCS 2019.

17:     We show that the gradient descent of 

18:     the Kempf-Ness function for an unstable orbit

19:     converges to a 1-parameter subgroup in the Hilbert-Mumford criterion, and 

20:     the associated moment-map sequence converges to the mimimum-norm point of the moment polytope.

21:     We show further refinements for operator scaling---the left-right action 

22:     on a matrix tuple $A= (A_1,A_2,\ldots,A_N)$.

23:     We characterize the gradient-flow limit of operator scaling

24:     via a vector-space generalization of 

25:     the classical Dulmage-Mendelsohn decomposition of a bipartite graph.

26:     %

27:     Also, for a special case of $N = 2$, 

28:     we reveal that this limit determines 

29:     the Kronecker canonical form of matrix pencils $s A_1+A_2$.

30: \end{abstract}

31: