1: \begin{abstract}
2: For dense Hermitian matrices with small off-diagonal (numerical) ranks and in
3: a hierarchically semiseparable form, we give a stable divide-and-conquer
4: eigendecomposition method with nearly linear complexity (called SuperDC) that
5: significantly improves an earlier basic algorithm in [Vogel, Xia, et al., SIAM
6: J. Sci. Comput., 38 (2016)]. We incorporate a sequence of key stability
7: techniques and provide many improvements in the algorithm design. Various
8: stability risks in the original algorithm are analyzed, including potential
9: exponential norm growth, cancellations, loss of accuracy with clustered
10: eigenvalues or intermediate eigenvalues, etc. In the dividing stage, we give a
11: new structured low-rank update strategy with balancing that eliminates the
12: exponential norm growth and also minimizes the ranks of low-rank updates. In
13: the conquering stage with low-rank updated eigenvalue solution, the original
14: algorithm directly uses the regular fast multipole method (FMM) to accelerate
15: function evaluations, which has the risks of cancellation, division by zero,
16: and slow convergence. Here, we design a triangular FMM to avoid cancellation.
17: Furthermore, when there are clustered intermediate eigenvalues or when updates
18: to existing eigenvalues are very small, we design a novel local shifting
19: strategy to integrate FMM accelerations into the solution of shifted secular
20: equations so as to achieve both the efficiency and the reliability. We also
21: provide several improvements or clarifications on some structures and
22: techniques that are missing or unclear in the previous work. The resulting
23: SuperDC eigensolver has significantly better stability while keeping the
24: nearly linear complexity for finding the entire eigenvalue decomposition. In a
25: set of comprehensive tests, SuperDC shows dramatically lower runtime and
26: storage than the Matlab \textsf{eig} function. The stability benefits are also
27: confirmed with both analysis and numerical comparisons.
28:
29: \end{abstract}
30: