SYMAMD (Symmetric AMD using COLAMD)
For symmetric A, apply COLAMD to A+A' structure to find fill-reducing order.
Mathematical Formulation
Approximate column degree for LU: deg(j) ≈ # rows in current column's nonzero pattern after elimination Aggressive absorption reduces degree overestimates from element merging.
Complexity
Time: O(nnz(A)·α(n)) average with aggressive absorption Space: O(nnz(A) + n) for quotient graph representation Slightly slower than AMD but better for unsymmetric patterns.
Implementations
SuiteSparse
- colamd.h - Column Approximate Minimum Degree ordering for sparse LU factorization
COLAMD computes a column permutation Q that reduces fill-in during LU factorization of an unsymmetric matrix A. The ordering minimizes the fill-in of A*Q when factored as LU.
SYMAMD computes a symmetric ordering for a symmetric matrix, using COLAMD on the matrix's structure. Both are related to the minimum degree family of algorithms.
References
- Davis, Gilbert, Larimore, Ng (2004). "A Column Approximate Minimum Degree Ordering Algorithm". ACM Trans. Math. Software 30(3):353-376.