HP MLIB User's Guide Vol. 2 7th Ed.

Chapter 14 METIS Routines 993
Computes fill reducing orderings of sparse matrices mlib_METIS_NodeND
options[2] Determines the algorithm used during
initial partioning. Possible values are:
1—Edge-based region growing (Default)
2—Node-based region growing
options[3] Determines the algorithm used for
refinement. Possible values are:
1—Two-sided node FM refinement
2—One-sided node FM refinement (Default)
One-sided FM refinement is faster than two-sided, but
in some cases two-sided refinement may produce better
orderings.
options[4] Used for debugging purposes. Always set it
to 0 (Default)
options[5] Used to select whether or not to compress
the graph and to order connected components
separately. The possible values and their meanings are:
0—Do not try to compress the graph and do not order
each connected component separately.
1—Try to compress the graph. (A compressed graph is
actually formed if the size of the graph can be reduced
by at least 15%.) (Default)
2—Order each connected component of the graph
separately. This option is particularly useful when
after a few levels of nested dissection, the graph breaks
up in many smaller disconnected subgraphs. This is
true for certain types of LP matrices.
3—Try to compress the graph and also order each
connected component separately.
options[6] Used to determine how many separators to
find at each step of nested dissection. The larger the
number of separators found at each step, the higher the
runtime and better the quality (in general). The default
value is 1, unless the graph has been compressed by
more than a factor of 2, in which case it becomes 2.
Reasonable values are in the range of 1 to 5. For most
problems, a value of 5 increases the runtime by a
factor of 3.