Confluent Hasse diagrams

We show that a transitively reduced digraph has a confluent upward drawing if and only if its reachability relation has order dimension at most two. In this case, we construct a confluent upward drawing with $O(n^2)$ features, in an $O(n) \times O(n)$ grid in $O(n^2)$ time. For the digraphs representing series-parallel partial orders we show how to construct a drawing with $O(n)$ features in an $O(n) \times O(n)$ grid in $O(n)$ time from a series-parallel decomposition of the partial order. Our drawings are optimal in the number of confluent junctions they use.

💡 Research Summary

The paper establishes a precise characterization of when a transitively reduced directed acyclic graph (DAG) admits an upward confluent drawing: this is exactly when the reachability relation of the graph has order dimension at most two. The authors show that for any such graph one can construct an upward confluent Hasse diagram by first computing a two‑dimensional realizer (two linear extensions) of the underlying partial order, embedding each element at integer coordinates (2i, 2j) in a (2n+1) × (2n+1) grid, and then augmenting the point set with additional points that correspond to the Dedekind–MacNeille completion of the poset. These added points become the confluent junctions of the final drawing.

The construction proceeds in three phases. Phase 1 finds the two linear orders in O(n²) time (using the algorithm of Ma and Spinrad) and places the original elements on even rows and columns of the grid, guaranteeing that dominance in the plane mirrors the partial‑order relation. Phase 2 scans all odd‑indexed grid positions (i, j) and inserts a junction at (i, j) if four local monotonicity conditions hold; these conditions ensure that the new point represents a cut of the poset that is not already present, i.e., a new element of the lattice completion. Phase 3 generates all direct dominance pairs (edges of the transitive reduction) by a row‑wise stack algorithm reminiscent of Graham’s scan. For each column a “topmost” element seen so far is maintained, and as the algorithm sweeps rows it creates edges from the stack and from the column top to the current point, thereby enumerating all required track segments in O(n² + k) time, where k ≤ O(n²) is the number of edges. After the combinatorial structure is built, the whole picture is rotated 45° counter‑clockwise so that every track is monotone upward, satisfying the upward‑confluent requirement.

The authors prove that this construction yields a planar confluent diagram and that the number of junctions used (O(n²)) is optimal: any upward confluent drawing of the same DAG must contain at least as many junctions because each corresponds to a distinct element of the Dedekind–MacNeille completion, which is the smallest complete lattice containing the poset. Moreover, they show that if the poset’s dimension exceeds two, no upward confluent drawing exists, establishing the “if and only if” condition.

For the important subclass of series‑parallel partial orders, the paper presents a linear‑time algorithm that leverages the series‑parallel decomposition tree. By recursively embedding series and parallel compositions, only O(n) junctions are needed, and the entire drawing fits in an O(n) × O(n) grid. This improves both the feature count and the running time from quadratic to linear for this family.

Implementation details include rendering each track segment as a cubic Bézier curve with control points placed a small fixed distance above and below each junction; the authors argue that such curves cannot intersect because their convex hulls are disjoint and the curves lie within those hulls. Experiments on random two‑dimensional posets of size 100 demonstrate that the confluent Hasse diagrams are completely crossing‑free, whereas traditional Hasse diagrams exhibit many crossings. The series‑parallel case shows linear growth in both junction count and runtime, confirming the theoretical bounds.

In summary, the paper makes three major contributions: (1) a complete combinatorial characterization of upward‑confluent drawability via order dimension, (2) an optimal O(n²)‑feature algorithm for general two‑dimensional posets, and (3) a linear‑time, linear‑feature method for series‑parallel posets. These results bridge the gap between order theory, planar graph drawing, and confluent visualization, offering a practical way to produce clear, crossing‑free representations of dense partial orders that were previously difficult to visualize.