Half-Graph Index in Sparse Graphs

• Half-Graph Index is defined as the maximum order of a half-graph—a bipartite structure encoding total orders—that appears in a graph or its power.
• It serves as a quantitative measure of structural complexity, with explicit bounds in sparse graph classes such as planar, bounded-degree, and minor-free graphs.
• The index plays a critical role in parameterized complexity, influencing tractability results and kernelization strategies for problems like Independent Set.

Half-graph index denotes several distinct notions in current graph-theoretic and algorithmic literature. In the graph-theoretic sense used for sparse graph powers and semi-induced patterns, it is the maximum order of a half-graph that appears in a graph, a graph power, or a graph class; because half-graphs encode total orders, the parameter serves as a quantitative measure of structural complexity and of the failure of stability-like behavior (Sokołowski, 2021). In parallel, unrelated usages attach the phrase or close variants to Knill’s symmetric graph index, to the HL-index for hypergraph max-reachability, and to index-assisted Half-Space Proximal classification (Knill, 2012). The most developed combinatorial theory concerns bipartite order patterns, semi-ladders, and their consequences for sparse graph structure and parameterized complexity (Dreier et al., 7 Feb 2026).

1. Half-graphs, ladders, and the formal parameter

A half-graph $H_n$ is a bipartite graph on parts $A=\{a_1,\dots,a_n\}$ and $B=\{b_1,\dots,b_n\}$ whose edges encode a total order. One standard convention is

$E(H_n)=\{a_i b_j: 1\le i\le j\le n\}.$

The sparse-graph-powers literature also uses the strict convention $(b_i,a_j)\in E$ iff $i<j$; the two conventions differ only by the diagonal and are equivalent up to minor tweaks in the resulting bounds (Sokołowski, 2021). Under the strict convention, the term ladder is used synonymously with half-graph.

The same literature places half-graphs alongside two related bipartite patterns. For $2\ell$ distinct vertices $a_1,\dots,a_\ell$ and $b_1,\dots,b_\ell$, a semi-ladder of order $\ell$ satisfies $A=\{a_1,\dots,a_n\}$0 for all $A=\{a_1,\dots,a_n\}$1 and $A=\{a_1,\dots,a_n\}$2 for all $A=\{a_1,\dots,a_n\}$3. A co-matching of order $A=\{a_1,\dots,a_n\}$4 satisfies $A=\{a_1,\dots,a_n\}$5 iff $A=\{a_1,\dots,a_n\}$6. Every ladder and every co-matching is a semi-ladder (Sokołowski, 2021).

For general graphs, the parameterized-complexity literature defines the half-graph index through semi-induced copies across a cut. If $A=\{a_1,\dots,a_n\}$7 denotes the bipartite subgraph induced by the cut $A=\{a_1,\dots,a_n\}$8, then

$A=\{a_1,\dots,a_n\}$9

Matching and co-matching indices are defined analogously by semi-induced copies of $B=\{b_1,\dots,b_n\}$0 and $B=\{b_1,\dots,b_n\}$1 (Dreier et al., 7 Feb 2026). The semi-induced formulation is essential when dense edges inside $B=\{b_1,\dots,b_n\}$2 or inside $B=\{b_1,\dots,b_n\}$3 are irrelevant to the bipartite order pattern.

In graph powers, the parameter is radius-dependent. For a graph $B=\{b_1,\dots,b_n\}$4 and integer $B=\{b_1,\dots,b_n\}$5,

$B=\{b_1,\dots,b_n\}$6

The half-graph index at radius $B=\{b_1,\dots,b_n\}$7 is the maximum $B=\{b_1,\dots,b_n\}$8 such that $B=\{b_1,\dots,b_n\}$9 appears in $E(H_n)=\{a_i b_j: 1\le i\le j\le n\}.$0. Equivalently, $E(H_n)=\{a_i b_j: 1\le i\le j\le n\}.$1 contains a distance-$E(H_n)=\{a_i b_j: 1\le i\le j\le n\}.$2 ladder of order $E(H_n)=\{a_i b_j: 1\le i\le j\le n\}.$3 (Sokołowski, 2021).

2. Half-graph index in powers of sparse graphs

The modern quantitative theory asks how large half-graphs or semi-ladders can be in $E(H_n)=\{a_i b_j: 1\le i\le j\le n\}.$4 when $E(H_n)=\{a_i b_j: 1\le i\le j\le n\}.$5 belongs to a sparse class. Earlier results established boundedness non-constructively for nowhere dense classes, but recent work gives explicit and nearly tight asymptotic bounds for planar graphs, bounded-degree graphs, bounded-pathwidth graphs, bounded-treewidth graphs, and $E(H_n)=\{a_i b_j: 1\le i\le j\le n\}.$6-minor-free graphs (Sokołowski, 2021).

The principal asymptotic bounds stated for radius $E(H_n)=\{a_i b_j: 1\le i\le j\le n\}.$7 are as follows.

Graph class	Lower bound	Upper bound
Bounded maximum degree $E(H_n)=\{a_i b_j: 1\le i\le j\le n\}.$8	$E(H_n)=\{a_i b_j: 1\le i\le j\le n\}.$9	$(b_i,a_j)\in E$0
Planar graphs	$(b_i,a_j)\in E$1	$(b_i,a_j)\in E$2
Bounded pathwidth $(b_i,a_j)\in E$3	$(b_i,a_j)\in E$4	$(b_i,a_j)\in E$5
Bounded treewidth $(b_i,a_j)\in E$6	$(b_i,a_j)\in E$7	$(b_i,a_j)\in E$8
$(b_i,a_j)\in E$9-minor-free graphs	$i<j$0	$i<j$1

These bounds show that the half-graph index can grow exponentially in the power radius even in sparse settings, but that the growth remains quantitatively constrained by the ambient sparsity notion (Sokołowski, 2021).

For planar graphs, the central theorem gives a constructive upper bound $i<j$2 for distance-$i<j$3 semi-ladders, improving the previously available $i<j$4 bound. The proof yields an explicit bound

$i<j$5

where $i<j$6 is a polynomial in $i<j$7 extracted from the structural analysis (Sokołowski, 2021).

The same work also proves a fully polynomial neighborhood-complexity bound for planar graphs. If $i<j$8 has size $i<j$9, and $2\ell$0 is the truncated distance-$2\ell$1 profile of $2\ell$2 on $2\ell$3, then

$2\ell$4

This theorem is a key ingredient in the planar semi-ladder bound and refines earlier bounded-expansion estimates in the planar case (Sokołowski, 2021).

3. Structural role in parameterized complexity

The half-graph index enters parameterized complexity through a trichotomy with the matching index and co-matching index. A theorem of Ding, Oporowski, Oxley, and Vertigan implies that sufficiently large twin-free bipartite graphs contain a large matching, co-matching, or half-graph. Using twin classes and semi-induced cuts, this yields the following graph-level consequence: there exists a function $2\ell$5 such that every graph with neighborhood diversity at least $2\ell$6 has matching index, co-matching index, or half-graph index at least $2\ell$7 (Dreier et al., 7 Feb 2026).

This organizes graph classes into eight regimes according to whether the three indices are bounded or unbounded. Several of the regimes inherit known tractability through bounded neighborhood diversity or bounded mim-width, but the role of the half-graph index becomes particularly sharp when combined with the co-matching index. If both $2\ell$8 and $2\ell$9 are bounded by a constant $a_1,\dots,a_\ell$0, then Independent Set is fixed-parameter tractable: one can decide whether $a_1,\dots,a_\ell$1 in time $a_1,\dots,a_\ell$2 (Dreier et al., 7 Feb 2026). Dominating Set is also fixed-parameter tractable in this semi-ladder-free regime.

The mechanism is a structural lemma producing a large clique or independent set $a_1,\dots,a_\ell$3 with a strong neighborhood dichotomy: for every outside vertex $a_1,\dots,a_\ell$4, either $a_1,\dots,a_\ell$5 or $a_1,\dots,a_\ell$6. This supports kernelization rules such as the deletion rule for a large clique $a_1,\dots,a_\ell$7 when $a_1,\dots,a_\ell$8 (Dreier et al., 7 Feb 2026).

By contrast, bounded half-graph index alone is insufficient for tractability. There exists a graph class with half-graph index at most $a_1,\dots,a_\ell$9 on which Independent Set is $b_1,\dots,b_\ell$0-hard, and a graph class with half-graph index at most $b_1,\dots,b_\ell$1 on which Dominating Set is $b_1,\dots,b_\ell$2-hard (Dreier et al., 7 Feb 2026). The negative side is therefore as important as the positive one: half-graph-freeness controls one order-type obstruction, but not all of the complexity generated by co-matchings or other dense cut patterns.

The same work gives an approximation algorithm tied directly to the half-graph index. If $b_1,\dots,b_\ell$3 has half-graph index $b_1,\dots,b_\ell$4 and maximum independent set size $b_1,\dots,b_\ell$5, then in time $b_1,\dots,b_\ell$6 one can compute an independent set of size at least $b_1,\dots,b_\ell$7 (Dreier et al., 7 Feb 2026). The recursion is based on the fact that a semi-induced half-graph in a reduced subgraph would extend to a larger half-graph in the original graph, so the index decreases along the branching process.

4. Planar cage machinery and the geometry of semi-ladders

The most elaborate structural analysis of the half-graph index appears in the planar upper bound for semi-ladders in graph powers. Starting from a large distance-$b_1,\dots,b_\ell$8 semi-ladder, the proof repeatedly extracts more rigid subconfigurations: quasi-cages, cages, ordered cages, identity ordered cages, neighbor cages, and separating cages (Sokołowski, 2021).

A quasi-cage of order $b_1,\dots,b_\ell$9 consists of a distance-$\ell$0 semi-ladder together with two poles $\ell$1 and two geodesic trees $\ell$2 rooted at $\ell$3 and $\ell$4. The trees expose how shortest paths to the $\ell$5-vertices interleave in the planar embedding. Lemma 5.1 states that if $\ell$6 contains a distance-$\ell$7 semi-ladder of order $\ell$8, then $\ell$9 contains a quasi-cage of order $A=\{a_1,\dots,a_n\}$00 (Sokołowski, 2021).

Successive extraction lemmas improve the topological regularity. Every quasi-cage of order $A=\{a_1,\dots,a_n\}$01 contains a cage of order $A=\{a_1,\dots,a_n\}$02; every cage has an order; every ordered cage of order $A=\{a_1,\dots,a_n\}$03 can be re-embedded to yield an identity ordered cage of order $A=\{a_1,\dots,a_n\}$04; every sufficiently large identity ordered cage contains a neighbor cage; and every sufficiently large neighbor cage contains a separating cage (Sokołowski, 2021).

The final bottleneck is quantitative. Any separating cage has order $A=\{a_1,\dots,a_n\}$05. Chaining the extraction lemmas yields the explicit planar upper bound $A=\{a_1,\dots,a_n\}$06 and hence the asymptotic $A=\{a_1,\dots,a_n\}$07 bound on the semi-ladder index (Sokołowski, 2021).

This analysis is significant because it turns an abstract exclusion-of-orders statement into a constructive topological theorem. The proof relies on nooses, cyclic orders, rightmost shortest paths, and a neighborhood-complexity theorem, producing a geometric account of why large semi-ladders cannot persist in planar graph powers (Sokołowski, 2021).

5. Other meanings of “half-graph index”

Outside the combinatorics of half-graphs and semi-ladders, the same phrase or close variants denote different constructions.

In Knill’s graph-theoretic topology, the symmetric or “half” index of an injective function $A=\{a_1,\dots,a_n\}$08 at a vertex $A=\{a_1,\dots,a_n\}$09 is

$A=\{a_1,\dots,a_n\}$10

It satisfies the integral geometric index formula

$A=\{a_1,\dots,a_n\}$11

where $A=\{a_1,\dots,a_n\}$12 is the unit sphere of $A=\{a_1,\dots,a_n\}$13 and $A=\{a_1,\dots,a_n\}$14 is a discrete level-surface graph inside $A=\{a_1,\dots,a_n\}$15. For geometric graphs of odd dimension, $A=\{a_1,\dots,a_n\}$16, and therefore curvature vanishes pointwise (Knill, 2012). This is a local topological invariant, not a bipartite order parameter.

In hypergraph indexing, the phrase “Half-Graph Index” is used in the supplied terminology for HL-index, meaning Hypergraph Labeling Index. HL-index is a vertex-to-hyperedge labeling scheme for max-reachability in hypergraphs. For each vertex $A=\{a_1,\dots,a_n\}$17, it stores labels $A=\{a_1,\dots,a_n\}$18 consisting of pairs $A=\{a_1,\dots,a_n\}$19, and queries are answered by

$A=\{a_1,\dots,a_n\}$20

Its construction uses dominant tuples, transitive covering, the scalar summary $A=\{a_1,\dots,a_n\}$21, and a dynamically pruned neighbor-index $A=\{a_1,\dots,a_n\}$22 (Xie et al., 29 Dec 2025). This usage is algorithmically unrelated to the half-graph order $A=\{a_1,\dots,a_n\}$23.

In instance-based learning, the supplied terminology also associates a “Half-Graph Index” with an ANN-backed query procedure for the Half-Space Proximal graph. The method retrieves a candidate set via HNSW and then applies the HSP predicate $A=\{a_1,\dots,a_n\}$24 to remove candidates closer to the retained neighbor $A=\{a_1,\dots,a_n\}$25 than to the query $A=\{a_1,\dots,a_n\}$26. The resulting Probabilistic Asymptotic HSP classifier has the same complexity as indexed probabilistic kNN, plus $A=\{a_1,\dots,a_n\}$27 local filtering, and in the reported experiments often outperforms both exact and indexed kNN (Talamantes et al., 2021).

These usages share the word index but not a common mathematical object. In one case the index measures the largest embedded order pattern; in the others it is a local topological quantity or an indexing data structure.

6. Conceptual significance, dualities, and open directions

Half-graphs matter because they encode total orders inside graph relations. Their presence is therefore a combinatorial witness to instability-like behavior, while their exclusion forces regularity in neighborhoods, cuts, or graph powers (Sokołowski, 2021). This explains why half-graph bounds connect model-theoretic themes, sparse graph structure, and algorithm design.

Several dualities sharpen this picture. Matching-free classes are dual to co-matching-free classes under complementation, and half-graph-freeness is self-dual: a class is half-graph-free iff its complement class is half-graph-free (Dreier et al., 7 Feb 2026). These symmetries are reflected in the parallel complexity classifications of Independent Set and Clique.

The half-graph index also interacts with broader structural measures. Bounded matching index implies bounded mim-width across branch decompositions, yielding polynomial-time algorithms in several regimes. Semi-ladder bounds control the complexity of fixed-parameter algorithms for distance-$A=\{a_1,\dots,a_n\}$28 dominating set on sparse graphs. Recent work points toward further links with neighborhood complexity, VC-dimension, and cham-width (Sokołowski, 2021).

Several quantitative gaps remain open. For planar graphs, the lower bound $A=\{a_1,\dots,a_n\}$29 and the upper bound $A=\{a_1,\dots,a_n\}$30 leave an exponential-versus-quasi-exponential gap in the power radius (Sokołowski, 2021). In parameterized complexity, extending the fixed-parameter tractability of Independent Set from semi-ladder-free classes to graph classes of bounded cham-width is identified as a natural next step (Dreier et al., 7 Feb 2026). More broadly, the coexistence of strong tractability under bounded half-graph and co-matching indices with hardness under bounded half-graph index alone shows that half-graph exclusion is powerful but not, by itself, a complete algorithmic regularity principle (Dreier et al., 7 Feb 2026).