Segment Graph Representation

• Segment representations are mappings of graph vertices to line segments that intersect if and only if the vertices are adjacent, providing a clear geometric encoding of graphs.
• The construction uses an exponential coordinate transform on normalized L-representations to explicitly compute integer grid endpoints for each segment.
• This method achieves a bounded resolution of 4^n for planar graphs, significantly improving previous exponential bounds and aiding compact encodability.

A segment representation of a graph $G=(V,E)$ assigns to each vertex $v\in V$ a line segment $s(v)$ in the plane such that two segments intersect if and only if the corresponding vertices are adjacent. “Segment representations with small resolution” studies not the existence problem in general, but the integer-grid resolution needed when such a representation exists. Its main theorem states that any graph with an $L$-representation, and in particular any planar graph, has a segment representation on an integer grid of width and height $4^n$, where $n=|V|$ (Biedl, 2018).

1. Problem setting and complexity motivation

For a graph $G=(V,E)$, a segment representation is an assignment $v\mapsto s(v)$ of line segments in $\mathbb{R}^2$ such that

$$s(v)\cap s(w)\neq \emptyset \quad\text{if and only if}\quad (v,w)\in E.$$

Not all graphs admit such a representation, but planar graphs do (Biedl, 2018).

The central parameter is the resolution of the representation under the assumption that segment endpoints lie on integer grid points. Since endpoints can be perturbed to rational coordinates and then scaled, one may assume integer endpoints without loss of generality. The resolution is then the width and height of an integer grid containing all segment endpoints (Biedl, 2018).

This question is motivated by compact encodability. Low resolution means that a representation can be described with few bits, whereas very large coordinates may be unavoidable for some graph classes. The note also gives a complexity-theoretic reason that bounded resolution is meaningful: if every segment graph had a representation of exponential resolution, then the representation could be encoded with polynomially many bits, which would place the recognition problem in NP. Since segment graph recognition is $v\in V$0-hard, the general existence of such bounds is considered unlikely (Biedl, 2018).

2. $v\in V$1-representations and the covered graph classes

The theorem applies to graphs that admit an $v\in V$2-representation. In the form used here, an $v\in V$3-representation is a string representation in which every vertex is drawn as an axis-aligned $v\in V$4-shape consisting of one horizontal segment and one vertical segment sharing a common endpoint. After a $v\in V$5 rotation, each $v\in V$6-shape becomes the union of the top side and right side of its axis-aligned bounding box (Biedl, 2018).

A vertex $v\in V$7 is described by four integers

$v\in V$8

with bounding box

$v\in V$9

The corresponding rotated $s(v)$0-shape consists of:

• the top horizontal segment from $s(v)$1 to $s(v)$2,
• the right vertical segment from $s(v)$3 to $s(v)$4.

Before the segment construction, the representation is normalized so that:

1. no endpoint of an $s(v)$5 lies on another $s(v)$6,
2. every horizontal and vertical arm has positive length,
3. all endpoint coordinates are distinct,
4. after re-ranking coordinates, all $s(v)$7- and $s(v)$8-coordinates lie in $s(v)$9, equivalently $L$0 after indexing shift (Biedl, 2018).

This preserves the represented graph because only relative order matters. The relevance of $L$1-representations is that every planar graph has one, via earlier planar-graph representation machinery used in Gonçalvez–Isenburg–Pennarun’s proof of Scheinerman’s conjecture. Consequently, the main theorem immediately yields a bounded-resolution segment representation for every planar graph (Biedl, 2018).

3. Exponential coordinate transform and explicit construction

The construction replaces each rotated $L$2-shape by a diagonal segment whose slope depends exponentially on the top and right coordinates of the bounding box. For a vertex $L$3 with parameters $L$4, define a long diagonal

$L$5

as the segment from

$L$6

Equivalently, $L$7 is the downward diagonal of the rectangle

$L$8

and its slope is

$L$9

The actual segment $4^n$0 is obtained by pruning $4^n$1 to the two halfspaces

$4^n$2

Thus $4^n$3 is the portion of $4^n$4 that remains after clipping at the vertical line $4^n$5 and the horizontal line $4^n$6 (Biedl, 2018).

The resulting endpoints are

$4^n$7

and

$4^n$8

These are integer grid points because all exponents are integers. The construction is therefore explicit and constructive: once an $4^n$9-representation is given, the segment endpoints are directly computable (Biedl, 2018).

The geometric purpose of the exponential map $n=|V|$0, $n=|V|$1 is to separate coordinates exponentially. This produces controlled slope differences among the diagonals and is the mechanism that allows adjacency information in the $n=|V|$2-representation to be recovered as segment intersection.

4. Correctness: no false positives and no false negatives

The proof has two complementary lemmas. The first establishes that if two constructed segments intersect, then the corresponding vertices are adjacent. Let

$n=|V|$3

and assume $n=|V|$4. If also $n=|V|$5, then the diagonals $n=|V|$6 and $n=|V|$7 have the same order on both axes, so they cannot cross. Therefore an intersection can occur only when $n=|V|$8 and $n=|V|$9. If $G=(V,E)$0, then $G=(V,E)$1 lies entirely left of $G=(V,E)$2; if $G=(V,E)$3, then $G=(V,E)$4 lies entirely above $G=(V,E)$5. Hence intersection forces

$G=(V,E)$6

Then $G=(V,E)$7 lies on the vertical arm of $G=(V,E)$8 and on the horizontal arm of $G=(V,E)$9, so the original $v\mapsto s(v)$0-shapes intersect. This shows that the segment construction creates no false positives (Biedl, 2018).

The converse lemma states that if $v\mapsto s(v)$1, then $v\mapsto s(v)$2 and $v\mapsto s(v)$3 intersect. After renaming, one may assume the intersection in the $v\mapsto s(v)$4-representation occurs between the vertical arm of $v\mapsto s(v)$5 and the horizontal arm of $v\mapsto s(v)$6, so

$v\mapsto s(v)$7

Because coordinates are integral and distinct, this sharpens to

$v\mapsto s(v)$8

The unpruned diagonals satisfy

$v\mapsto s(v)$9

so they cross at some point $\mathbb{R}^2$0. The proof derives bounds implying

$\mathbb{R}^2$1

and symmetrically

$\mathbb{R}^2$2

Thus the crossing survives both left-pruning and bottom-pruning, so it lies on both $\mathbb{R}^2$3 and $\mathbb{R}^2$4. This shows that the construction creates no false negatives (Biedl, 2018).

Together, these lemmas prove that $\mathbb{R}^2$5 is a correct segment representation.

5. Resolution bound, comparison with earlier work, and the $\mathbb{R}^2$6 extension

After normalization, all box parameters satisfy

$\mathbb{R}^2$7

Since every endpoint coordinate in the construction is built from numbers of the form $\mathbb{R}^2$8, all coordinates are on the order of $\mathbb{R}^2$9. The note states more precisely that the coordinates lie in

$$s(v)\cap s(w)\neq \emptyset \quad\text{if and only if}\quad (v,w)\in E.$$0

so the width and height are bounded by

$$s(v)\cap s(w)\neq \emptyset \quad\text{if and only if}\quad (v,w)\in E.$$1

up to a harmless factor of $$s(v)\cap s(w)\neq \emptyset \quad\text{if and only if}\quad (v,w)\in E.$$2. The theorem is therefore stated cleanly as a $$s(v)\cap s(w)\neq \emptyset \quad\text{if and only if}\quad (v,w)\in E.$$3-resolution bound (Biedl, 2018).

This is an upper bound, not a lower bound, and no claim of tightness is made. The contribution is not that graphs with $$s(v)\cap s(w)\neq \emptyset \quad\text{if and only if}\quad (v,w)\in E.$$4-representations are segment representable—that was already known from earlier work of Middendorf and Pfeiffer—but that one can guarantee a much smaller explicit resolution. The older approach yields only a much weaker bound of

$$s(v)\cap s(w)\neq \emptyset \quad\text{if and only if}\quad (v,w)\in E.$$5

whereas the new construction gives

$$s(v)\cap s(w)\neq \emptyset \quad\text{if and only if}\quad (v,w)\in E.$$6

The note also treats the more general class of graphs with an $$s(v)\cap s(w)\neq \emptyset \quad\text{if and only if}\quad (v,w)\in E.$$7-representation, where a vertex may be represented either by an $$s(v)\cap s(w)\neq \emptyset \quad\text{if and only if}\quad (v,w)\in E.$$8-shape or by its horizontal reflection. For this larger class, the $$s(v)\cap s(w)\neq \emptyset \quad\text{if and only if}\quad (v,w)\in E.$$9 construction does not seem to extend directly. Instead, the paper gives a different construction based on rectangles

$v\in V$00

using one diagonal of $v\in V$01, chosen according to whether the shape is an $v\in V$02 or a reflected $v\in V$03. This yields a segment representation of width $v\in V$04 and height $v\in V$05, hence overall resolution roughly $v\in V$06 (Biedl, 2018).

6. Constructive significance, upstream planar machinery, and open problems

The theorem is explicitly constructive. Once an $v\in V$07-representation is available, one can normalize coordinates by rank, compute $v\in V$08 for each vertex, and output the segment endpoints using the stated formulas. The note does not emphasize formal running-time analysis, but the coordinate construction itself is straightforwardly polynomial-time once the $v\in V$09-representation is known (Biedl, 2018).

Its relation to planar graph theory is upstream rather than internal. The note does not redevelop the machinery that yields $v\in V$10-representations for planar graphs; that role is played by earlier planar representation results, including the framework used in Gonçalvez–Isenburg–Pennarun’s proof of Scheinerman’s conjecture. The specific contribution here is the passage

$v\in V$11

Several open questions remain. The conclusion asks whether exponential resolution is actually necessary for planar graphs, or whether all planar graphs might admit polynomial-resolution segment representations. The same question is left open for the larger class of graphs with $v\in V$12-representations. It is also open whether every graph with an $v\in V$13-representation admits an exponential-resolution segment representation, whether one can explicitly construct a family of segment graphs with no exponential-resolution realization, and whether every segment graph has some bounded-resolution representation $v\in V$14 for a computable function $v\in V$15 (Biedl, 2018).

A plausible implication is that the note’s main importance lies in converting an existence theorem into a quantitatively controlled realization theorem. For planar graphs, segment representability was already known; what becomes explicit here is that one may realize such graphs by line segments on an integer grid of width and height at most $v\in V$16, via a direct exponential coordinate transform from a normalized $v\in V$17-representation.