Skip to contents

Reference for the named parameter lists accepted by create.basin.complex(). Tables below are generated from the same defaults and resolvers as the constructor. They show the default for a graph with 100 vertices. The RTCB n.min default is max(20, ceiling(sqrt(n.vertices))); inspect object$parameters for the exact settings used in your analysis. Unknown names, invalid values, and method-inapplicable settings are errors.

Graph lengths and validation

edge.length.symmetry.tolerance is the nonnegative tolerance for matching lengths on the two orientations of an undirected edge. Neighbor lists and lengths are parallel; zero lengths are allowed. The inherited weight.list name in lcor() and lslope() also supplies edge lengths, not vertex masses or generic affinity weights. Derivative local correlation skips lengths at or below 1e-10. Vertex mass weights support, whereas vertex density can change the trajectory direction. Neither is inferred from the other.

Trajectory flow

modulation selects CLOSEST, NONE, DENSITY, EDGELEN or DENSITY_EDGELEN. CLOSEST uses exact connected plateaus and the nearest improving boundary edge; other modes require plateau.policy = "none". Density modes require vertex.density. edge.length.quantile.thld in (0,1] controls the preferred local edge-length threshold. long.edge.fallback is "allow_and_flag", "allow", or "forbid" when only longer improving edges are available. symmetric.seeding requests the backend's symmetric seed rule; max.trajectory.length bounds stored path length and store.trajectories controls path storage. tie.breaking = TRUE requires an explicit integer tie.seed and perturbs construction values; FALSE requires a NULL seed. primary.assignment.policy is "backend_primary", "largest_membership_then_extremum", or "none". It chooses a single convenience label without replacing overlapping membership.

Merge trees

plateau.tolerance = 0 is currently required: plateaus use exact equality. The only elder rule is "birth_then_representative_vertex": higher birth wins, with representative vertex breaking equal-birth ties. Assignment is "elder_at_merge"; root death uses "component_opposite_extreme". Persistence is in field units and is not a significance score.

Geodesic and relaxed trajectory basins

Geodesic reachability uses the edge-length quantile in (0,1], optional stored trajectories, and "none" or "largest_membership_then_extremum" assignment. RTCB exposes stopping and path-search controls: n.min limits the initial Dijkstra candidate region (clamped to graph size); m.min is the minimum monotonicity in (-1,1); optional q.min is the required favorable-change fraction in (0,1] and takes precedence over m.min; run.max limits consecutive adverse steps. tau0, kappa, k.max, h.max, d.path.max, and eta.step control the path search: local tolerance is tau0*(1-progress)^kappa, k.max bounds labels per vertex, h.max bounds hops, d.path.max bounds metric path length, and eta.step is the field-change deadband for classifying steps. epsilon.d and eps.num are numerical tolerances. sink.prune, sink.prune.max.iter, and max.paths.per.sink control sink pruning and retained path counts. See the details below for the path-budget interpretation. With cumulative favorable field change G and adverse change L, the path budget is lambda*G - L + tau >= -eps.num. Here lambda is (1-m.min)/(1+m.min), or (1-q.min)/q.min when q.min is supplied. These controls are method-specific, not interchangeable measures of scale.

Overlap cells

This family requires both directions. basin.merge.overlap.thld in [0,1] controls basin merging. The three min.*.cell.size.thld integers control minimum ascending/descending, ascending/ascending, and descending/descending cell sizes. cell.graph.params is forwarded to the cell graph backend. Primary assignment is "none"; overlapping cells are the authoritative result.

Refinement applicability and order

Every stage is disabled by default. Relative-value filtering, maxima and minima clustering, and geometric filtering apply to trajectory, geodesic and RTCB methods. Support filtering also applies to merge trees. Expansion applies only to geodesic and RTCB methods. No refinement stage applies to overlap cells. Nondefault settings in inapplicable stages are rejected even if disabled.

Stages run in the order above, followed by support filtering and expansion. Relative thresholds compare extrema with the mean construction field; clustering uses the overlap threshold in [0,1]. Geometric filters use the listed percentiles and positive integer hop radius. Support filters require minimum vertex count, trajectory count and normalized support mass; mass thresholds need supplied mass. Expansion assigns uncovered vertices to the nearest retained basin. Raw supports remain available independently of these retained results. See the canonical workflow for worked refinement examples.

Graph defaults

edge.length.symmetry.tolerance

1.49011611938477e-08

Resolved method defaults

trajectory_flow

modulation

"CLOSEST"

plateau.policy

"connected_exact"

edge.length.quantile.thld

0.9

long.edge.fallback

"allow_and_flag"

store.trajectories

TRUE

max.trajectory.length

10000L

symmetric.seeding

TRUE

tie.breaking

FALSE

tie.seed

NULL

primary.assignment.policy

"backend_primary"

superlevel_merge_tree

plateau.tolerance

0

elder.rule

"birth_then_representative_vertex"

primary.assignment.policy

"elder_at_merge"

root.death.policy

"component_opposite_extreme"

geodesic_reachability

edge.length.quantile.thld

0.9

store.trajectories

FALSE

primary.assignment.policy

"none"

rtcb

edge.length.quantile.thld

0.9

store.trajectories

FALSE

n.min

20L

m.min

0.7

q.min

NULL

run.max

2L

tau0

0.1

kappa

1.5

k.max

24L

h.max

1000L

d.path.max

Inf

eta.step

0

epsilon.d

1e-12

eps.num

1e-12

sink.prune

FALSE

sink.prune.max.iter

5L

max.paths.per.sink

64L

primary.assignment.policy

"backend_primary"

overlap_cell_complex

basin.merge.overlap.thld

0.1

min.asc.desc.cell.size.thld

1L

min.asc.asc.cell.size.thld

1L

min.desc.desc.cell.size.thld

1L

cell.graph.params

list()

primary.assignment.policy

"none"

Refinement defaults

relative.value

enabled

FALSE

min.relative.value.max

1.1

max.relative.value.min

0.9

maxima.clustering

enabled

FALSE

overlap.threshold

0.15

minima.clustering

enabled

FALSE

overlap.threshold

0.15

geometric.filter

enabled

FALSE

mean.neighbor.distance.percentile.max

0.9

mean.hop.distance.percentile.max

0.9

degree.percentile.max

0.9

hop.k

2L

support.filter

enabled

FALSE

min.basin.size

0L

min.trajectory.count

0L

min.basin.mass

0

expansion

enabled

FALSE

policy

"nearest_retained_basin"