Benchmark report

Tree-decomposition solver comparison

Validated decompositions of primal and incidence graphs derived from preprocessed model-counting formulas. Every solver receives the same ten-second wall-clock limit.

Dataset
Loading…
Budget
Graph views
Generated

Analysis scope

Tree components are omitted. Filters apply to the comparison and graph records.

Quality coverage at 10 seconds

How often each solver returns a decomposition close to the best width observed.

stronger weaker

Quality–coverage curve

Number of selected graphs with a validated width no more than Δ above the best observed for that graph.

Width quality profile Solver coverage at reported width excess thresholds.
Higher is better. A missing or invalid result does not satisfy any threshold. Hover or focus a point for its exact value.

Instance results

All selected graphs, with every displayed solver visible without horizontal scrolling.

best observed widest valid result

Methods and definitions

Valid
A decomposition accepted by the common validator. Coverage is valid results divided by selected graphs.
Best observed
The minimum width among validated results from the solvers displayed here. It is an empirical reference, not a proven optimum.
Deadline
The runner stops search at the stated budget. Solvers with an internal deadline receive an earlier warning so they can finish and write a decomposition before that cutoff.
Quality coverage
The number of selected graphs on which a solver returns a validated width within Δ of the best observed width. Missing results count against coverage.
Component graph
One connected graph passed independently to every solver. This is the default aggregate unit.
Tree component
A connected component with |E| = |V| − 1, hence exact width one. These components are omitted by default.
Width
The largest bag size minus one.
Total bag size
The sum of bag sizes, counting a vertex once for each bag containing it.
Bag count
The number of bags in the decomposition.
Elapsed
Search time before normal exit or the fixed budget signal.

The default minimum width of 30 keeps the headline comparison focused on graphs where small absolute width differences are less informative. Set it to “No minimum” to inspect the complete non-tree dataset.