1: \begin{abstract}

2: Algebraic multigrid (AMG) is an $\mathcal{O}(n)$ solution process

3: for many large sparse linear systems.  A hierarchy of progressively coarser

4: grids is constructed that utilize complementary relaxation and interpolation

5: operators.  High-energy error is reduced by relaxation, while low-energy error

6: is mapped to coarse-grids and reduced there.  However, large parallel

7: communication costs often limit parallel scalability.  As the multigrid

8: hierarchy is formed, each coarse matrix is formed through a triple matrix

9: product. The resulting coarse-grids often have significantly more nonzeros

10: per row than the original fine-grid operator, thereby generating

11: high parallel communication costs on coarse-levels. In this paper, we introduce

12: a method that systematically removes entries in coarse-grid matrices

13: after the hierarchy is formed, leading to an improved communication costs.

14: We sparsify by removing weakly connected or unimportant entries in the

15: matrix, leading to improved solve time.  The main trade-off is that if

16: the heuristic identifying unimportant entries is used too aggressively,

17: then AMG convergence can suffer.  To counteract this, the original

18: hierarchy is retained, allowing entries to be

19: reintroduced into the solver hierarchy if convergence is too slow.

20: This enables a balance between communication cost and convergence, as necessary.

21: In this paper we

22: present new algorithms for reducing communication and

23: present a number of computational experiments in support.

24: \end{abstract}

25: