\hideLIPIcs

Roma Tre University, Rome, [email protected]://orcid.org/0000-0001-5512-5298 Roma Tre University, Rome, [email protected]://orcid.org/0009-0001-4538-8198 Roma Tre University, Rome, [email protected]://orcid.org/0000-0003-2396-5174 Roma Tre University, Rome, [email protected]://orcid.org/0009-0008-6266-3324 Roma Tre University, Rome, [email protected]://orcid.org/0000-0003-4224-1550 Roma Tre University, Rome, [email protected]://orcid.org/0000-0001-5987-8713 Roma Tre University, Rome, [email protected]://orcid.org/0000-0002-5766-4567 Roma Tre University, Rome, [email protected]://orcid.org/0000-0001-9806-7411 \CopyrightCarlos Alegría, Susanna Caroppo, Giordano Da Lozzo, Marco D’Elia, Giuseppe Di Battista, Fabrizio Frati, Fabrizio Grosso, and Maurizio Patrignani \fundingThis research was supported, in part, by MUR of Italy (PRIN Project no. 2022ME9Z78 – NextGRAAL and PRIN Project no. 2022TS4Y3N – EXPAND). \ccsdesc[500]Theory of computation Computational geometry \ccsdesc[500]Theory of computation Design and analysis of algorithms \ccsdesc[500]Mathematics of computing Graph algorithms

Upward Pointset Embeddings of Planar $st$ -Graphs

Carlos Alegría Susanna Caroppo Giordano Da Lozzo Marco D’Elia Giuseppe Di Battista Fabrizio Frati Fabrizio Grosso Maurizio Patrignani

Abstract

We study upward pointset embeddings (UPSEs) of planar $st$ -graphs. Let $G$ be a planar $st$ -graph and let $S\subset\mathbb{R}^{2}$ be a pointset with $|S|=|V(G)|$ . An UPSE of $G$ on $S$ is an upward planar straight-line drawing of $G$ that maps the vertices of $G$ to the points of $S$ . We consider both the problem of testing the existence of an UPSE of $G$ on $S$ (UPSE Testing) and the problem of enumerating all UPSEs of $G$ on $S$ . We prove that UPSE Testing is \NP-complete even for $st$ -graphs that consist of a set of directed $st$ -paths sharing only $s$ and $t$ . On the other hand, for $n$ -vertex planar $st$ -graphs whose maximum $st$ -cutset has size $k$ , we prove that it is possible to solve UPSE Testing in $\mathcal{O}(n^{4k})$ time with $\mathcal{O}(n^{3k})$ space, and to enumerate all UPSEs of $G$ on $S$ with $\mathcal{O}(n)$ worst-case delay, using $\mathcal{O}(kn^{4k}\log n)$ space, after $\mathcal{O}(kn^{4k}\log n)$ set-up time. Moreover, for an $n$ -vertex $st$ -graph whose underlying graph is a cycle, we provide a necessary and sufficient condition for the existence of an UPSE on a given poinset, which can be tested in $\mathcal{O}(n\log n)$ time. Related to this result, we give an algorithm that, for a set $S$ of $n$ points, enumerates all the non-crossing monotone Hamiltonian cycles on $S$ with $\mathcal{O}(n)$ worst-case delay, using $\mathcal{O}(n^{2})$ space, after $\mathcal{O}(n^{2})$ set-up time.

keywords:

Upward pointset embeddings, planar st-graphs, st-cutset, non-crossing monotone Hamiltoninan cycles, graph drawing enumeration

1 Introduction

Given an $n$ -vertex upward planar graph $G$ and a set $S$ of $n$ points in the plane, an upward pointset embedding (UPSE) of $G$ on $S$ is an upward planar drawing of $G$ where the vertices are mapped to the points of $S$ and the edges are represented as straight-line segments. The Upward Pointset Embeddability Testing Problem (UPSE Testing) asks whether an upward planar graph $G$ has an UPSE on a given pointset $S$ .

Pointset embedding problems are classic challenges in Graph Drawing and have been considered for both undirected and directed graphs. For an undirected graph, a pointset embedding (PSE) has the same definition of an UPSE, except that the drawing must be planar, rather than upward planar. The Pointset Embeddability Testing Problem (PSE Testing) asks whether a planar graph has a PSE on a given pointset $S$ . Pointset embeddings have been studied by several authors. It is known that a graph admits a PSE on every pointset in general position if and only if it is outerplanar [10, 22]; such a PSE can be constructed efficiently [7, 8]. PSE Testing is, in general, \NP-complete [9], however it is polynomial-time solvable if the input graph is a planar $3$ -tree [29]. More in general, a polynomial-time algorithm for PSE Testing exists if the input graph has a fixed embedding, bounded treewidth, and bounded face size [5]. PSE becomes \NP-complete if one of the latter two conditions does not hold. PSEs have been studied also for dynamic graphs [4].

The literature on UPSEs is not any less rich than the one on PSEs. From a combinatorial perspective, the directed graphs with an UPSE on a one-sided convex pointset have been characterized [6, 23]; all directed trees are among them. Conversely, there exist directed trees that admit no UPSE on certain convex pointsets. Directed graphs that admit an UPSE on any convex pointset, but not on any pointset in general position, exist [2]. It is still unknown whether every digraph whose underlying graph is a path admits an UPSE on every pointset in general position, see, e.g., [27]. UPSEs where bends along the edges are allowed are studied in [6, 20, 21, 25]. From the computational complexity point of view [24], it is known that UPSE Testing is \NP-hard, even for planar $st$ -graphs and $2$ -convex pointsets, and that UPSE Testing can be solved in polynomial time if the directed graph is outerplanar and the given poinset is convex.

Our contributions. We tackle UPSE Testing for planar $st$ -graphs. Let $G$ be an $n$ -vertex planar $st$ -graph and $S$ be a set of $n$ points in the plane. We adopt the common assumption in the context of upward pointset embeddability, see e.g. [2, 6, 24], that no two points of $S$ lie on the same horizontal line. Our results are the following:

•

In Section 3, we show that UPSE Testing is \NP-hard even if $G$ consists of a set of internally-disjoint $st$ -paths (Theorem 3.1). A similar proof shows that UPSE Testing is \NP-hard for directed trees consisting of a set of directed root-to-leaf paths (Theorem 3.3). This answers an open question from [3] and strengthens a result therein, which shows \NP-hardness for directed trees with multiple sources and with a prescribed mapping for a vertex.
•

In Section 4, we show that UPSE Testing can be solved in $\mathcal{O}{(n^{4k})}$ time and $\mathcal{O}(n^{3k})$ space, where $k$ is the size of the largest $st$ -cutset of $G$ (Theorem 4.4). This parameter measures the “fatness” of the digraph and coincides with the length of the longest directed path in the dual [14]. By leveraging on the techniques developed for the testing algorithm, we also show how to enumerate all UPSEs of $G$ on $S$ with $\mathcal{O}(n)$ worst-case delay, using $\mathcal{O}(kn^{4k}\log n)$ space, after $\mathcal{O}(kn^{4k}\log n)$ set-up time (Theorem 4.5).
•

In Section 5, we provide a simple characterization of the pointsets in general position that allow for an UPSE of $G$ , if $G$ consists of two $st$ -paths. Based on that, we provide an $\mathcal{O}(n\log n)$ testing algorithm for this case (Theorem 5.1).
•

Finally, in Section 6, inspired by the fact that an UPSE of a planar $st$ -graph composed of two $st$ -paths defines a non-crossing monotone Hamiltonian cycle on $S$ , we provide an algorithm that enumerates all the non-crossing monotone Hamiltonian cycles on a given pointset with $\mathcal{O}(n)$ worst-case delay, and $\mathcal{O}(n^{2})$ space usage and set-up time (Theorem 6.1).

Concerning our last result, we remark that a large body of research has considered problems related to enumerating and counting non-crossing structures on a given pointset [1, 11, 18, 26, 30]. Despite this effort, the complexity of counting the non-crossing Hamiltonian cycles, often called polygonalizations, remains open [16, 26, 28]. However, it is possible to enumerate all polygonalizations of a given pointset in singly-exponential time [32, 33]. Recently, an algorithm has been shown [17] to enumerate all polygonalizations of a given pointset in time polynomial in the output size, i.e., bounded by a polynomial in the number of solutions. However, an enumeration algorithm with polynomial (in the input size) delay is not yet known, neither in the worst-case nor in the average-case acception. Our enumeration algorithm achieves this goal for the case of monotone polygonalizations.

2 Preliminaries

We use standard terminology in graph theory [15] and graph drawing [13]. For an integer $k>0$ , let $[k]$ denote the set $\{1,\dots,k\}$ . A permutation with repetitions of $k$ elements from $U$ is an arrangement of any $k$ elements of a set $U$ , where repetitions are allowed.

For a point $p\in\mathbb{R}^{2}$ , we denote by $x(p)$ and $y(p)$ the $x$ - and $y$ -coordinate of $p$ , respectively. The convex hull $\operatorname{\mathcal{CH}}(S)$ of a set $S$ of points in ${\mathbb{R}}^{2}$ is the union of all convex combinations of points in $S$ . The boundary $\mathcal{B}(S)$ of $\operatorname{\mathcal{CH}}(S)$ is the polygon with minimum perimeter enclosing $S$ . The points of $S$ with lowest and highest $y$ -coordinates are the south and north extreme of $S$ , respectively; we also refer to them as to the extremes of $S$ . The left envelope of $S$ is the subpath $\mathcal{E}_{L}(S)$ of $\mathcal{B}(S)$ that lies to the left of the line passing through the extremes of $S$ ; it includes the extremes of $S$ . The right envelope $\mathcal{E}_{R}(S)$ of $S$ is defined analogously. We denote the subset of $S$ in $\mathcal{E}_{L}(S)$ and in $\mathcal{E}_{R}(S)$ by $\operatorname{\mathcal{H}_{L}}(S)$ and $\operatorname{\mathcal{H}_{R}}(S)$ , respectively. A polyline $(p_{1},\dots,p_{k})$ , with $k\geq 2$ , is a chain of straight-line segments.

We call ray any of the two half-lines obtained by cutting a straight line at any of its points, which is the starting point of the ray. A ray is upward if it passes through points whose $y$ -coordinate is larger than the one of the starting point of the ray. We denote by $\rho(p,q)$ the ray starting at a point $p$ and passing through a point $q$ . For a set of points $S$ and a point $p$ whose $y$ -coordinate is smaller than the one of every point in $S$ , we denote by $\rho(p,S)$ (by $\ell(p,S)$ ) the rightmost (resp. leftmost) upward ray starting at $p$ and passing through a point of $S$ , that is, the clockwise (resp. counter-clockwise) rotation around $p$ which brings $\rho(p,S)$ (resp. $\ell(p,S)$ ) to coincide with any other upward ray starting at $p$ and passing through a point of $S$ is larger than $180^{\circ}$ .

A polyline $(p_{1},\dots,p_{k})$ is $y$ -monotone if $y(p_{i})<y(p_{i+1})$ , for $i=1,\dots,k-1$ . A monotone path on a pointset $S$ is a $y$ -monotone polyline $(p_{1},\dots,p_{k})$ such that the points $p_{1},\dots,p_{k}$ belong to $S$ . A monotone cycle on $S$ consists of two monotone paths on $S$ that share their endpoints. A monotone Hamiltonian cycle $(p_{1},\dots,p_{k},p_{1})$ on $S$ is a monotone cycle on $S$ such that each point of $S$ is a point $p_{i}$ (and vice versa).

A path $(v_{1},\dots,v_{k})$ is directed if, for $i=1,\dots,k-1$ , the edge $(v_{i},v_{i+1})$ is directed from $v_{i}$ to $v_{i+1}$ ; the vertices $v_{2},\dots,v_{k-1}$ are intern. A planar $st$ -graph is an acyclic digraph with one source $s$ and one sink $t$ , which admits a planar embedding in which $s$ and $t$ are on the boundary of the outer face. An $st$ -path in a planar $st$ -graph is a directed path from $s$ to $t$ . A drawing of a directed graph is straight-line if each edge is represented by a straight-line segment, it is planar if no two edges cross, and it is upward if the tail of every directed edge has a lower $y$ -coordinate than its head. A digraph that admits an upward planar drawing is an upward planar graph. Every upward planar graph admits an upward planar straight-line drawing [14]. An Upward Pointset Embedding (UPSE, for short) of an upward planar graph $G$ on a pointset $S$ is an upward planar straight-line drawing of $G$ that maps each vertex of $G$ to a point in $S$ . In this paper, we study the following problem.

In the remainder, we assume that not all points in $S$ lie on the same line, as otherwise there is an UPSE if and only if the input is a directed path. Recall that no two points in $S$ have the same $y$ -coordinate. Unless otherwise specified, we do not require points to be in general position, i.e., we allow three or more points to lie on the same line.

3 NP-Completeness of UPSE Testing

In this section we prove that UPSE Testing is \NP-complete. The membership in \NP is obvious, as one can non-deterministically assign the vertices of the input graph $G$ to the points of the input pointset $S$ and then test in polynomial time whether the assignment results in an upward planar straight-line drawing of $G$ . In the remainder of the section, we prove that UPSE Testing is \NP-hard even in very restricted cases.

We first show a reduction from 3-Partition to instances of UPSE in which the input is a planar $st$ -graph composed of a set of internally-disjoint $st$ -paths. An instance of 3-Partition consists of a set $A=\{a_{1},\dots,a_{3b}\}$ of $3b$ integers, where $\sum_{i=1}^{3b}a_{i}=bB$ and $B/4\leq a_{i}\leq B/2$ , for $i=1,\dots,3b$ . The 3-Partition problem asks whether $A$ can be partitioned into $b$ subsets $A_{1},\dots,A_{b}$ , each with three integers, so that the sum of the integers in each set $A_{i}$ is $B$ . Since 3-Partition is strongly \NP-hard [19], we may assume that $B$ is bounded by a polynomial function of $b$ . Given an instance $A$ of 3-Partition, we show how to construct in polynomial time, precisely $\mathcal{O}(b\cdot B)$ , an equivalent instance $(G,S)$ of UPSE Testing.

The $n$ -vertex planar $st$ -graph $G$ is composed of $4b+1$ internally-disjoint $st$ -paths. Namely, for $i=1,\dots,3b$ , we have that $G$ contains an $a_{i}$ -path, i.e., a path with $a_{i}$ internal vertices, and $b+1$ additional $k$ -paths, where $k=2B+1$ . Note that $n=2+(b+1)k+\sum_{i=1}^{3b}a_{i}=2+(b+1)k+bB$ .

The points of $S$ lie on the plane as follows (see Figure 1(a)):

•

$p_{1}$ is the origin, with coordinates $(0,0)$ .
•

Consider $b+1$ upward rays $\rho_{1},\dots,\rho_{b+1}$ , whose starting point is $p_{1}$ , such that the angles $\alpha_{1},\dots,\alpha_{b+1}$ that they respectively form with the $x$ -axis satisfy $3\pi/4>\alpha_{1}>\dots>\alpha_{b+1}>\pi/4$ . Let $\ell$ be a line intersecting all the rays, with a positive slope smaller than $\pi/4$ . For $j=1,\dots,b+1$ , place $k$ points $p_{j,1},\dots,p_{j,k}$ (in this order from bottom to top) along $\rho_{j}$ , so that $p_{j,k}$ is on $\ell$ and no two points share the same $y$ -coordinate. Observe that $p_{b+1,k}$ is the highest point placed so far.
•

Place $p_{n}$ at coordinates $(0,10\cdot y(p_{b+1,k}))$ .
•

Finally, for $j=1,\dots,b$ , place $B$ points along a non-horizontal segment $s_{j}$ in such a way that: (i) $s_{j}$ is entirely contained in the triangle with vertices $p_{j,k}$ , $p_{j+1,k}$ , and $p_{n}$ , (ii) for any point $p$ on $s_{j}$ , the polygonal line $\overline{p_{1}p}\cup\overline{pp_{n}}$ is contained in the region $R_{j}$ delimited by the polygon $\overline{p_{1}p_{j,k}}\cup\overline{p_{j,k}p_{n}}\cup\overline{p_{n}p_{j+1,k}% }\cup\overline{p_{j+1,k}p_{1}}$ , and (iii) no two distinct points on any two segments $s_{i}$ and $s_{j}$ share the same $y$ -coordinate.

Note that $S$ has $2+(b+1)k+bB=n$ points. The described reduction is the main ingredient for the proof of the following theorem.

Theorem 3.1.

UPSE Testing is \NP-hard even for planar $st$ -graphs consisting of a set of directed internally-disjoint $st$ -paths.

Proof 3.2.

First, the construction of $G$ and $S$ takes polynomial time. In particular, the coordinates of the points in $S$ can be encoded with a polylogarithmic number of bits. In order to prove the NP-hardness, it remains to show that the constructed instance $(G,S)$ of UPSE Testing is equivalent to the given instance $A$ of 3-Partition. Refer to Figure 1(b).

Suppose first that $A$ is a positive instance of 3-Partition, that is, there exist sets $A_{1},\dots,A_{b}$ , each with three integers, such that the sum of the integers in each set $A_{j}$ is $B$ . We construct an UPSE of $G$ on $S$ as follows. We map $s$ to $p_{1}$ and $t$ to $p_{n}$ . For $j=1,\dots,b+1$ , we map the $k$ internal vertices of a $k$ -path to the points $p_{j,1},\dots p_{j,k}$ , so that vertices that come first in the directed path have smaller $y$ -coordinates. Furthermore, for $j=1,\dots,b$ , let $A_{j}=\{a_{j_{1}},a_{j_{2}},a_{j_{3}}\}$ . Then we map the $a_{j_{1}}$ internal vertices of an $a_{j_{1}}$ -path, the $a_{j_{2}}$ internal vertices of an $a_{j_{2}}$ -path, and the $a_{j_{3}}$ internal vertices of an $a_{j_{3}}$ -path to the set of $B$ points in the triangle with vertices $p_{j,k}$ , $p_{j+1,k}$ , and $p_{n}$ , so that vertices that come first in the directed paths have smaller $y$ -coordinates and so that the internal vertices of the $a_{j_{1}}$ -path have smaller $y$ -coordinates than the internal vertices of the $a_{j_{2}}$ -path, which have smaller $y$ -coordinates than the internal vertices of the $a_{j_{3}}$ -path. This results in an UPSE of $G$ on $S$ .

Suppose next that $(G,S)$ is a positive instance of UPSE Testing. Trivially, in any UPSE of $G$ on $S$ , we have that $s$ is drawn on $p_{1}$ and $t$ on $p_{n}$ . Consider the points $p_{1,1},\dots p_{{b+1},1}$ . The paths using them use all the $(b+1)k$ points $p_{j,i}$ , with $j=1,\dots,b+1$ and $i=1,\dots,k$ . Indeed, if these paths left one of such points unused, no other path could reach it from $s$ without passing through $p_{1,1},\dots p_{{b+1},1}$ , because of the collinearity of the points along the rays $\rho_{1},\dots,\rho_{b+1}$ . Hence, there are at most $b+1$ paths that use $(b+1)k$ points. Since the $a_{i}$ -paths have less than $k$ internal vertices, these $b+1$ paths must all be $k$ -paths. Let $P_{1},\dots,P_{b+1}$ be the left-to-right order of the $k$ -paths around $p_{1}$ . For $j=1,\dots,b+1$ , path $P_{j}$ uses all points $p_{j,i}$ on $\rho_{j}$ , as if $P_{j}$ used a point $p_{h,i}$ with $h>j$ , then two among $P_{j},\dots,P_{b+1}$ would cross each other. Note that, after using $p_{j,k}$ , path $P_{j}$ ends with the segment $\overline{p_{j,k}p_{n}}$ . Hence, for $j=1,\dots,b$ , the region $R_{j}$ is bounded by $P_{j}$ and $P_{j+1}$ ; recall that $R_{j}$ contains the segment $s_{j}$ . The $a_{i}$ -paths must then use the points on $s_{1},\dots,s_{b}$ . Since $B/4<a_{i}<B/2$ , no two $a_{i}$ -paths can use all the $B$ points in one region and no four $a_{i}$ -paths can lie in the same region. Hence, three $a_{i}$ -paths use the $B$ points in each region, and this provides a solution to the given 3-Partition instance.

We next reduce the 3-Partition problem to the instances of UPSE testing in which the input is a directed tree consisting of a set of root-to-leaf paths. Consider an instance of 3-Partition consisting of a set $A=\{a_{1},\dots,a_{3b}\}$ of $3b$ integers, where $\sum_{i=1}^{3b}a_{i}=bB$ and $B/4\leq a_{i}\leq B/2$ , for $i=1,\dots,3b$ . We construct a directed tree $T$ as follows. The root $s$ of $T$ has $4b-1$ children. Among them, $b-1$ are leaves $v_{1},\dots,v_{b-1}$ , while each of the remaining $3b$ children is the first vertex of a directed path $P_{i}$ , for $i=1,\dots,3b$ , consisting of the $a_{i}+1$ vertices $v_{i,1},v_{i,2},\dots,v_{i,a_{i}+1}$ , where $v_{i,1}$ is the child of $s$ and $v_{i,a_{i}+1}$ is a leaf. All the edges of $T$ are directed from the root $s$ to the leaves. Note that the number of vertices of $T$ is $n=1+(b-1)+\sum_{i=1}^{3b}(a_{i}+1)=b(B+4)$ . The points of $S$ lie on the plane as follows (see Figure 2(a)):

•

$p_{1}$ is the origin, with coordinates $(0,0)$ .
•

Consider $b-1$ upward rays $\rho_{1},\dots,\rho_{b-1}$ , whose starting point is $p_{1}$ , such that the angles $\alpha_{1},\dots,\alpha_{b-1}$ formed by $\rho_{1},\dots,\rho_{b-1}$ with the $x$ -axis satisfy $3\pi/4>\alpha_{1}>\dots>\alpha_{b-1}>\pi/4$ . These rays split the half plane above the $x$ -axis into $b$ regions $R_{j}$ , with $j=1,2,\dots,b$ . In the interior of each region $R_{j}$ , place three points $p_{j,1},p_{j,2},$ and $p_{j,3}$ in such a way that $p_{j,1}$ is lower than $p_{j,2}$ , which is lower than $p_{j,3}$ , and so that they are all visible from $s$ . Along the line passing through $s$ and $p_{j,3}$ place $B$ points above $p_{j,3}$ .
•

Let $y_{m}$ be the highest $y$ -coordinate used so far. Let $\ell$ be a line with positive slope smaller than $\pi/4$ intersecting all the rays $\rho_{1},\dots,\rho_{b-1}$ at points that have $y$ -coordinates larger than $y_{m}$ . For $j=1,\dots,b-1$ , place a point $p_{\ell,j}$ at the intersection of $\rho_{j}$ with $\ell$ .

Note that $S$ has $1+3b+bB+(b-1)=b(B+4)=n$ points. The described reduction is the main ingredient for the proof of the following theorem.

Theorem 3.3.

UPSE Testing is \NP-hard even for directed trees consisting of a set of directed root-to-leaf paths.

Proof 3.4.

First, the construction of $T$ and $S$ takes polynomial time. In particular, the coordinates of the points in $S$ can be encoded with a polylogarithmic number of bits. In order to prove the NP-hardness, it remains to show that the constructed instance $(T,S)$ of UPSE Testing is equivalent to the given instance $A$ of 3-Partition. Refer to Figure 2(b).

Suppose first that $A$ is a positive instance of 3-Partition, that is, there exist sets $A_{1},\dots,A_{b}$ , each with three integers, such that the sum of the integers in each set $A_{j}$ is $B$ . We construct an UPSE of $G$ on $S$ as follows. We map $s$ to $p_{1}$ . For $j=1,\dots,b-1$ , we map the child $v_{j}$ of $s$ to $p_{\ell,j}$ . Furthermore, for $j=1,\dots,b$ , let $A_{j}=\{a_{j_{1}},a_{j_{2}},a_{j_{3}}\}$ . Then we map the $a_{j_{1}}$ internal vertices of an $a_{j_{1}}$ -path, the $a_{j_{2}}$ internal vertices of an $a_{j_{2}}$ -path, and the $a_{j_{3}}$ internal vertices of an $a_{j_{3}}$ -path to the set of $B$ points in the region $R_{j}$ , so that the neighbors of $s$ in the $a_{j_{1}}$ -path, in the $a_{j_{2}}$ -path, and in the $a_{j_{3}}$ -path lie on $p_{j,1}$ , $p_{j,2}$ , and $p_{j,3}$ , respectively, so that vertices that come first in the directed paths have smaller $y$ -coordinates, and so that the internal vertices of the $a_{j_{1}}$ -path have larger $y$ -coordinates than the internal vertices of the $a_{j_{2}}$ -path, which have larger $y$ -coordinates than the internal vertices of the $a_{j_{3}}$ -path. This results in an UPSE of $T$ on $S$ .

Suppose next that $(T,S)$ is a positive instance of UPSE Testing. It is obvious that the root $s$ of $T$ has to be placed on $p_{1}$ . From the root $s$ only $4b-1$ points are visible. These are the points $p_{\ell,j}$ , for $j=1,\dots,b-1$ , and the points $p_{h,1},p_{h,2},$ and $p_{h,3}$ , for $h=1,\dots,b$ (all these points are filled green in Figure 2(a)). Since $T$ has $4b-1$ children, each child must use one of the above points. Consider point $p_{\ell,b-1}$ . Since this is the highest point in the set $S$ , the child that uses it must be a leaf. This also holds for $p_{\ell,b-2}$ , which is the highest of the remaining points. Iterating this argument we have that the points $p_{\ell,j}$ , with $j=1,\dots,b-1$ , must be used by the $b-1$ children of $s$ which are leaves of $T$ . Since all other vertices have smaller $y$ -coordinates, each path $P_{i}$ , with $i=1,\dots,3m$ , is constrained to be into a region $R_{j}$ , with $j=1,\dots,b$ (see Figure 2(b)). Since each region $R_{j}$ contains exactly three points $p_{j,1},p_{j,2},$ and $p_{j,3}$ visible from $s$ , each region hosts exactly three such paths, which use the remaining $B$ points, and this provides a solution to the given 3-Partition instance.

4 Testing and Enumeration Algorithms for Planar st-Graphs with Maximum st-Cutset of Bounded Size

An $st$ -cutset of a planar $st$ -graph $G=(V,E)$ is a subset $W$ of $E$ such that:

•

removing $W$ from $E$ results in a graph consisting of exactly two connected components $C_{s}$ and $C_{t}$ ,
•

$s$ belongs to $C_{s}$ and $t$ belongs to $C_{t}$ , and
•

any edge in $W$ has its tail in $C_{s}$ and its head in $C_{t}$ .

In this section, we consider instances $(G,S)$ where $G$ is a planar $st$ -graph, whose maximum $st$ -cutset has bounded size $k$ . In Theorem 4.4, we show that UPSE Testing can be solved in polynomial time for such instances $(G,S)$ . Moreover, in Theorem 4.5, we show how to enumerate all UPSEs of $(G,S)$ with linear delay. The algorithm for Theorem 4.4 is based on a dynamic programming approach. It exploits the property that, for an $st$ -cutset $W$ defining the connected components $C_{s}$ and $C_{t}$ , the extensibility of an UPSE $\Gamma^{\prime}$ of $C_{s}\cup W$ on a subset $S^{\prime}$ of $S$ to an UPSE of $G$ on $S$ only depends on the drawing of the edges of $W$ , and not on the embedding of the remaining vertices of $C_{s}$ , provided that in $\Gamma^{\prime}$ there exists an horizontal line that crosses all the edges of $W$ . The algorithm for Theorem 4.5 leverages a variation of the dynamic programming table computed by the former algorithm to efficiently test the extensibility of an UPSE of $C_{s}\cup W$ (in which there exists a horizontal line that crosses all the edges of $W$ ) on a subset $S^{\prime}$ of $S$ to an UPSE of $G$ on $S$ .

The proofs of Theorems 4.4 and 4.5 exploit two dynamic programming tables $T$ and $Q$ defined as follows. Each entry of $T$ and $Q$ is indexed by a key that consists of a set of $h\leq k$ triplets $\langle e_{i},p_{i},q_{i}\rangle$ , where, for any $i=1,\dots,h$ , it holds that $e_{i}\in E(G)$ , $p_{i},q_{i}\in S$ , and $y(p_{i})<y(q_{i})$ . Moreover, each key $\chi=\bigcup^{h}_{i=1}\langle e_{i},p_{i},q_{i}\rangle$ satisfies the following constraints:

•

the set $E(\chi)=\bigcup^{h}_{i=1}e_{i}$ is an $st$ -cutset of $G$ and, for every $i,j$ , with $i\neq j$ , it holds true that $e_{i}\neq e_{j}$ (that is, $|E(\chi)|=h$ );
•

for every $i,j$ , with $i\neq j$ , it holds true that $p_{i}=p_{j}$ (resp. that $q_{i}=q_{j}$ ) if and only if $e_{i}$ and $e_{j}$ have the same tail (resp. the same head); and
•

let $\ell_{\chi}$ be the horizontal line passing through the tail with largest $y$ -coordinate among the edges in $E(\chi)$ , i.e., $\ell_{\chi}:=y=y(p_{i})$ s.t. $y(p_{j})\leq y(p_{i})$ for any $\langle e_{j},p_{j},q_{j}\rangle\in\chi$ ; then $\ell_{\chi}$ intersects all the segments $\overline{p_{j}q_{j}}$ , possibly at an endpoint.

For brevity, we sometimes say that the edge $e_{i}$ has its tail (resp. its head) mapped by $\chi$ on $p_{i}$ (resp. on $q_{i}$ ). We also say that $e_{i}$ is drawn as in $\chi$ if its drawing is the segment $\overline{p_{i}q_{i}}$ .

Let $\chi=\bigcup^{h}_{i=1}\langle e_{i},p_{i},q_{i}\rangle$ be a key of $T$ and of $Q$ ; see Figure 3(a). Let $G_{\chi}$ be the connected component containing $s$ of the graph obtained from $G$ by removing the edge set $E(\chi)$ .

The entry $T[\chi]$ contains a Boolean value such that $T[\chi]=\texttt{True}$ if and only if there exists an UPSE of $G_{\chi}^{+}=G_{\chi}\cup E(\chi)$ on some subset $S^{\prime}\subset S$ with $|S^{\prime}|=|V(G_{\chi}^{+})|$ such that:

•

the lowest point $p_{s}$ of $S$ belongs to $S^{\prime}$ and $s$ lies on it, and
•

for $i=1,\dots,h$ , the edge $e_{i}$ is drawn as in $\chi$ .

If $T[\chi]=\texttt{False}$ , the entry $Q[\chi]$ contains the empty set $\emptyset$ . If $T[\chi]=\texttt{True}$ and $E(\chi)$ coincides with the set of edges incident to $s$ , then $Q[\chi]$ stores the set $\{\bot\}$ . If $T[\chi]=\texttt{True}$ and $E(\chi)$ does not coincide with the set of edges incident to $s$ , $Q[\chi]$ stores the set $\Phi$ of keys with the following properties. Let $e_{\tau}$ be any edge whose tail $v_{\tau}$ has maximum $y$ -coordinate among the edges in $E(\chi)$ , i.e., $\langle e_{\tau},p_{\tau},q_{\tau}\rangle$ is such that $y(p_{\tau})\geq y(p_{j})$ for any $\langle e_{j},p_{j},q_{j}\rangle\in\chi$ . For each $\varphi\in\Phi$ , we have that:

•

$T[\varphi]=\texttt{True}$ ;
•

$E(\chi)\cap E(\varphi)$ contains all and only the edges in $E(\chi)$ whose tail is not $v_{\tau}$ , and each edge $e_{i}\in E(\chi)\cap E(\varphi)$ is drawn in $\varphi$ as it is drawn in $\chi$ ; and
•

all the edges in $E(\varphi)\setminus E(\chi)$ have $v_{\tau}$ as their head.

Additionally, we store a list $\Lambda$ of the keys $\sigma$ such that $T[\sigma]=\texttt{True}$ and $E(\sigma)$ is the set of edges incident to $t$ . Note that each edge in $E(\sigma)$ has its head mapped by $\sigma$ to the point $p_{t}\in S$ with largest $y$ -coordinate.

We use dynamic programming to compute the entries of $T$ and $Q$ in increasing order of $|V(G_{\chi})|$ . By the definition of $T$ , we have that $G$ admits an UPSE on $S$ if and only if $\Lambda\neq\emptyset$ .

First, we initialize all entries of $T$ to False and all entries of $Q$ to $\emptyset$ .

If $|V(G_{\chi})|=1$ , then $G_{\chi}$ only consists of $s$ . We set $T[\chi]=\texttt{True}$ and $Q[\chi]=\{\bot\}$ for every key $\chi=\bigcup^{h}_{i=1}\langle e_{i},p_{i},q_{i}\rangle$ such that:

•

$e_{1},\dots,e_{h}$ are the edges incident to $s$ ;
•

$p_{1}=\dots=p_{h}=p_{s}$ ; and
•

for every distinct $i$ and $j$ in $\{1,\dots,h\}$ , we have that $p_{s}$ , $q_{i}$ , and $q_{j}$ are not aligned.

If $|V(G_{\chi})|>1$ , we compute $T[\chi]$ and $Q[\chi]$ as follows, see Figure 3(b). If two segments $\overline{p_{i}q_{i}}$ and $\overline{p_{j}q_{j}}$ , with $i\neq j$ , cross (that is, they share a point that is internal for at least one of the segments), then we leave $T[\chi]$ and $Q[\chi]$ unchanged; in particular, $T[\chi]=\texttt{False}$ and $Q[\chi]=\emptyset$ . Otherwise, we proceed as follows. Let $e_{\tau}$ be any edge whose tail $v_{\tau}$ has maximum $y$ -coordinate among the edges in $E(\chi)$ . Let $H^{-}$ be the set of edges obtained from $E(\chi)$ by removing all the edges having $v_{\tau}$ as their tail, and let $H^{+}$ be the set of edges of $G$ having $v_{\tau}$ as their head. We define the set $H:=H^{-}\cup H^{+}$ . We have the following.

Claim 1.

$H$ is an $st$ -cutset of $G$ .

Proof 4.1.

Recall that, since $E(\chi)$ is an $st$ -cutset, removing the edges of $E(\chi)$ from $G$ yields two connected components $C_{s}$ and $C_{t}$ such that $s$ belongs to $C_{s}$ and $t$ belongs to $C_{t}$ ; see Figure 4(a). Let $C^{\prime}_{t}$ be the graph consisting of $C_{t}$ , the vertex $v_{\tau}$ , and the edges having $v_{\tau}$ as their tail (these are the edges in $E(\chi)\setminus H^{-}$ , which are not part of $H$ ). Also, let $C^{\prime}_{s}$ be the graph obtained by removing from $C_{s}$ the vertex $v_{\tau}$ and the edges in $H^{+}$ (i.e., these are the edges outgoing from $v_{\tau}$ ); see Figure 4(b). We have that $C^{\prime}_{s}$ and $C^{\prime}_{t}$ do not share any vertex, since $C_{s}$ and $C_{t}$ do not share any vertex, since $V(C^{\prime}_{s})\subset V(C_{s})$ and since the only vertex in $V(C^{\prime}_{t})\setminus V(C_{t})$ is $v_{\tau}$ , which does not belong to $C^{\prime}_{s}$ . Moreover, by construction $G=C^{\prime}_{t}\cup C^{\prime}_{s}\cup H$ , in particular the only edges connecting vertices in $C^{\prime}_{s}$ with vertices in $C^{\prime}_{s}$ are those in $H$ , which have their tails in $C^{\prime}_{s}$ and their heads in $C^{\prime}_{t}$ . Also, we have that $s$ belongs to $C^{\prime}_{s}$ and $t$ belongs to $C^{\prime}_{t}$ . To prove that $H$ is an $st$ -cutset of $G$ , it only remains to argue that each of $C^{\prime}_{s}$ and $C^{\prime}_{t}$ is connected. Since $C_{t}\subseteq C^{\prime}_{t}$ and since $C_{t}$ is connected, we have that every pair of vertices distinct from $v_{\tau}$ is connected by an undirected path in $C^{\prime}_{t}$ . Also, the heads of the edges outgoing from $v_{\tau}$ belong to $C_{t}$ and, by construction, such edges belong to $C^{\prime}_{t}$ . Hence, there exists an undirected path in $C^{\prime}_{t}$ between $v_{\tau}$ and every vertex of $C_{t}$ . Therefore, $C^{\prime}_{t}$ is connected. Suppose now, for a contradiction, that $C^{\prime}_{s}$ is not connected and thus there exists a vertex $v$ which is not in the same connected component as $s$ in $C^{\prime}_{s}$ . Since $G$ is a planar $st$ -graph, it contains a directed path from $s$ to $v$ . If such a path does not belong entirely to $C^{\prime}_{s}$ , it contains an edge which is directed from a vertex not in $C^{\prime}_{s}$ to a vertex in $C^{\prime}_{s}$ . Moreover, such an edge belongs to $H$ , however we already observed that all the edges in $H$ are outgoing from the vertices in $C^{\prime}_{s}$ , a contradiction.

Consider the set $S_{\downarrow}$ consisting of the points in $S$ whose $y$ -coordinates are smaller than $y(p_{\tau})$ . We have the following crucial observation.

{observation}

$T[\chi]=\texttt{True}$ if and only if there exists some key $\varphi$ , with $E(\varphi)=H$ , such that $T[\varphi]=\texttt{True}$ , the edges in $H^{-}$ are drawn in $\varphi$ as in $\chi$ , the edges in $H^{+}$ have their heads mapped by $\varphi$ on $p_{\tau}$ and their tails on a point in $S_{\downarrow}$ .

In view of Section 4, we can now define a procedure to compute $T[\chi]$ and $Q[\chi]$ . Assume that the edges $e_{1},\dots,e_{|H^{-}|},\dots,e_{|H|}\in H$ are ordered so that the edges of $H^{-}$ precede those of $H^{+}$ . By Section 4, if $|S_{\downarrow}|<|H^{+}|$ , then we leave $T[\chi]$ and $Q[\chi]$ unchanged, i.e., $T[\chi]=\texttt{False}$ and $Q[\chi]=\emptyset$ . In fact, in this case, there are not enough points in $S_{\downarrow}$ to map the tails of the edges in $H^{+}$ . Otherwise, let $D$ be the set of all permutations with repetitions of $|H^{+}|$ points from $S_{\downarrow}$ . We define a set $\Phi$ of keys that, for each $(d_{1},\dots,d_{|H^{+}|})\in D$ , contains a key $\varphi$ such that:

(i)

$E(\varphi)=H$ ;
(ii)

for any $i=1,\dots,|H^{-}|$ , the triple containing $e_{i}$ in $\varphi$ is the same as the triple containing $e_{i}$ in $\chi$ (note that $e_{i}\in H^{-}$ );
(iii)

for any $j=|H^{-}|+1,\dots,|H|$ , the triple containing $e_{j}$ in $\varphi$ has $q_{j}=p_{\tau}$ , and $p_{j}=d_{j-|H^{-}|}$ (note that $e_{j}\in H^{+}$ ); and
(iv)

for every $i=1,\dots,|H^{-}|$ and $j=|H^{-}|+1,\dots,|H|$ , it holds $p_{i}=p_{j}$ if and only if $e_{i}$ and $e_{j}$ have the same tail.

Let $\Phi^{\texttt{T}}=\{\varphi:\varphi\in\Phi\wedge\ T[\varphi]=\texttt{True}\}$ . By Section 4, we have $T[\chi]=\texttt{True}$ if and only if $|\Phi^{\texttt{T}}|\geq 1$ . Thus, we set $T[\chi]=\bigvee_{\varphi\in\Phi}T[\varphi]$ and $Q[\chi]=\Phi^{\texttt{T}}$ . We say that $\chi$ stems from any key $\varphi\in\Phi$ with $T[\varphi]=\texttt{True}$ .

We now upper bound the sizes of $T$ and $Q$ and the time needed to compute them.

Claim 2.

Tables $T$ and $Q$ have size in $\mathcal{O}(n^{3k})$ and $\mathcal{O}(kn^{4k}\log n)$ , respectively.

Proof 4.2.

First, we give an upper bound on the number of entries of $T$ (and, thus, of $Q$ ), which we denote by $\rho$ . Each entry of $T$ is associated with a key $\chi$ defined by an $st$ -cutset $E(\chi)$ of size at most $h\leq k$ , a permutation (possibly with repetitions) of $h$ points in $S$ describing a mapping of the tails of the edges in $E(\chi)$ with points in $S$ , and a permutation (possibly with repetitions) of $h$ points in $S$ describing a mapping of the heads of the edges in $E(\chi)$ with points in $S$ . Recall that $|S|=n$ , that $\binom{a}{b}\leq a^{b}$ , and that the number of permutations with repetition of $h$ elements from a set $U$ is $|U|^{h}$ . Therefore, we have that $\rho\leq\binom{m}{k}\cdot n^{k}\cdot n^{k}\leq(mn^{2})^{k}$ . Since $m\in\mathcal{O}(n)$ , we thus have $\rho\in\mathcal{O}(n^{3k})$ .

We can now upper bound the size of $T$ and $Q$ . Since each entry of $T$ stores a single bit, we immediately have that $T$ has $\mathcal{O}(n^{3k})$ size. Instead, each entry of table $Q$ stores $\mathcal{O}(n^{k})$ keys of size $\mathcal{O}(k\log n)$ ; thus, $Q$ has $\mathcal{O}(kn^{4k}\log n)$ size. The upper bound on the number of keys comes from the number of ways to map the tails of the at most $k$ edges incoming into $v_{\tau}$ on the points of $S$ , which has size $n$ ; this number is $\binom{n}{k}\in\mathcal{O}(n^{k})$ . The upper bound on the size of each key comes from the fact that it consists of at most $k$ triplets each containing an identifier of $\mathcal{O}(n)$ edges and two identifiers of $\mathcal{O}(n)$ points.

Claim 3.

Table $T$ and $Q$ can be computed in $\mathcal{O}(n^{4k})$ and $\mathcal{O}(kn^{4k}\log n)$ time, respectively.

Proof 4.3.

We determine the time needed to compute, for each key $\chi$ , the value $T[\chi]$ and $Q[\chi]$ . For each key $\chi$ , we need to verify whether the $h$ segments $\overline{p_{i}q_{i}}$ intersect at a point different from a common endpoint, which can be tested in $\mathcal{O}(k\log k)$ time [31]. Moreover, if $|V(G_{\chi})|>1$ , computing $T[\chi]$ requires accessing the value of up to $|S_{\downarrow}|^{|H^{+}|}<n^{k}$ entries of $T$ , and verifying whether at least one of them contains the value True. Since $n>k$ , the $\mathcal{O}(k\log k)$ term in the running time is dominated by the $\mathcal{O}(n^{k})$ term, hence the time needed to compute each entry $T[\chi]$ is thus $\mathcal{O}(n^{k})$ . Since, by Claim 2, there are $\mathcal{O}(n^{3k})$ keys $\chi$ , it follows that $T$ can be computed in overall $\mathcal{O}(n^{4k})$ time. On the other hand, the time needed to compute each entry $Q[\chi]$ is upper bounded by the time needed to write the $\mathcal{O}(n^{k})$ keys contained in $Q[\chi]$ , each of which has $\mathcal{O}(k\log n)$ size, i.e., $\mathcal{O}(kn^{k}\log{n})$ time per entry. It follows that $Q$ can be computed in overall $\mathcal{O}(kn^{4k}\log{n})$ time.

Finally, recall that in order to verify whether $G$ admits an UPSE on $S$ , we need to check whether $\Lambda\neq\emptyset$ . Computing the maximum size of an $st$ -cutset of a planar $st$ -graph $G$ can be done in linear time, as it reduces to the problem of computing the length of a shortest path in the dual of any embedding of $G$ (between the vertices representing the left and right outer faces of this embedding) [12, 14]. Therefore, the overall running time to test whether $G$ admits an UPSE on $S$ is dominated by the time needed to compute $T$ , that is, $\mathcal{O}(n^{4k})$ time.

To obtain an UPSE $\Gamma$ of $G$ on $S$ , if any, we proceed as follows. Suppose that the algorithm terminates with a positive answer and let $\sigma$ be any key in $\Lambda$ . We initialize $\Gamma$ to a drawing of the edges in $E(\sigma)$ , where each edge $e_{i}\in E(\sigma)$ is drawn as in $\sigma$ . Then, in $\mathcal{O}(n^{k})$ time, we can search in $T$ a key $\chi$ with $T[\chi]=\texttt{True}$ such that $\sigma$ stems from $\chi$ , and update $\Gamma$ to include a drawing of the edges in $E(\chi)\setminus E(\sigma)$ , where each edge $e_{i}\in E(\chi)\setminus E(\sigma)$ is drawn as in $\chi$ ; note that the edges in $E(\chi)\cap E(\sigma)$ are drawn in $\chi$ as they are drawn in $\sigma$ . Iteratively apply such a procedure until a key $\alpha$ is reached such that $T[\alpha]=\texttt{True}$ and $E(\alpha)$ is the set of edges incident to $s$ . This yields the desired UPSE of $G$ on $S$ . As the depth of the recursion is linear in the size of $G$ , we can therefore compute $\Gamma$ in $\mathcal{O}(n^{k+1})$ time.

From the above discussion, we have the following theorem.

Theorem 4.4.

Let $G$ be an $n$ -vertex planar $st$ -graph whose maximum $st$ -cutset has size $k$ and let $S$ be a set of $n$ points. UPSE Testing can be solved for $(G,S)$ in $\mathcal{O}{(n^{4k})}$ time and $\mathcal{O}{(n^{3k})}$ space; if an UPSE of $G$ on $S$ exists, it can be constructed within the same bounds.

We now turn our attention to the design of an algorithm for the enumeration of the UPSEs of $G$ on $S$ . The algorithm exploits the table $Q$ and the set $\Lambda$ . By Claims 2 and 3, these can be computed in $\mathcal{O}(kn^{4k}\log{n})$ time and space. Our enumeration algorithm defines and explores an acyclic digraph $\mathcal{D}$ . The nodes of the digraph correspond to the keys $\chi$ of the dynamic programming table $Q$ such that $Q[\chi]\neq\emptyset$ , plus a source $n_{\cal S}$ and a sink $n_{\cal T}$ . Let $\chi_{i}$ and $\chi_{j}$ be two keys of $Q$ such that $Q[\chi_{i}]\neq\emptyset$ and $Q[\chi_{j}]\neq\emptyset$ , and let $n(\chi_{i})$ and $n(\chi_{j})$ be the nodes corresponding to $\chi_{i}$ and $\chi_{j}$ in $\mathcal{D}$ , respectively. There exists an edge directed from $n(\chi_{i})$ to $n(\chi_{j})$ in $\mathcal{D}$ if $\chi_{j}\in Q[\chi_{i}]$ . Also, there exists an edge directed from $n_{\cal S}$ to each node $n(\sigma)$ such that $\sigma\in\Lambda$ . Finally, there exists an edge directed to $n_{\cal T}$ from each node $n(\chi)$ such that $Q[\chi]=\{\bot\}$ . Note that $n_{\cal S}$ is the unique source of $\mathcal{D}$ , $n_{\cal T}$ is the unique sink of $\mathcal{D}$ , and $\mathcal{D}$ has no directed cycle. Hence, $\mathcal{D}$ is an $n_{\cal S}n_{\cal T}$ -graph.

The exploration of $\mathcal{D}$ performed by our enumeration algorithm is a depth-first traversal, in which every distinct path from $n_{\cal S}$ to $n_{\cal T}$ corresponds to an UPSE of $G$ on $S$ . We initialize a current UPSE $\Gamma$ of $G$ on $S$ as $\Gamma=S$ (where no edge of $G$ is drawn). When the visit traverses an edge of $\mathcal{D}$ directed from a node $n(\chi_{i})$ to a node $n(\chi_{j})$ , it adds to $\Gamma$ the edges in $E(\chi_{j})\setminus E(\chi_{i})$ , drawn as in $\chi_{j}$ . Note that these are all the edges in $E(\chi_{j})$ if $n(\chi_{i})=n_{\cal S}$ and it is an empty set if $n(\chi_{j})=n_{\cal T}$ . Whenever the traversal reaches $n_{\cal T}$ , it outputs the constructed UPSE $\Gamma$ of $G$ on $S$ . When the visit backtracks on a node $n(\chi_{i})$ coming from an edge $(n(\chi_{i}),n(\chi_{j}))$ , it removes from $\Gamma$ the edges in $E(\chi_{j})\setminus E(\chi_{i})$ .

To prove the correctness of the algorithm, we show what follows:

(i)

Distinct $n_{\cal S}n_{\cal T}$ -paths in $\mathcal{D}$ correspond to different UPSEs of $G$ on $S$ .
(ii)

For each UPSE of $G$ on $S$ , there exists in $\mathcal{D}$ an $n_{\cal S}n_{\cal T}$ -path corresponding to it.

For a directed path $\mathcal{P}$ in $\mathcal{D}$ , let $E(\mathcal{P})$ be the set that contains all the edges in the sets $E(\chi)$ , where $\chi$ is any key corresponding to a node in $\mathcal{P}$ .

•

To prove Item i, we proceed by contradiction. Let $\Gamma$ be an UPSE of $G$ on $S$ that is generated twice by the algorithm, when traversing distinct $n_{\cal S}n_{\cal T}$ -paths $\mathcal{P}_{1}$ and $\mathcal{P}_{2}$ . Let $n(\chi_{x})$ be the closest node to $n_{\mathcal{S}}$ in $\mathcal{P}_{1}$ and $\mathcal{P}_{2}$ such that $(n(\chi_{x}),n(\chi_{1}))$ is an edge in $\mathcal{P}_{1}$ and $(n(\chi_{x}),n(\chi_{2}))$ is an edge in $\mathcal{P}_{2}$ , with $n(\chi_{1})\neq n(\chi_{2})$ , where $\chi_{x}$ , $\chi_{1}$ , and $\chi_{2}$ are keys of $Q$ . Note that, since the path $\mathcal{P}_{x}$ from $n_{\mathcal{S}}$ to $n(\chi_{x})$ (possibly such a path is a single node if $n_{\mathcal{S}}=n(\chi_{x})$ ) is the same in $\mathcal{P}_{1}$ and $\mathcal{P}_{2}$ , the restriction $\Gamma_{x}$ of $\Gamma$ to the edge set $E(\mathcal{P}_{x})$ is also the same in $\mathcal{P}_{1}$ and $\mathcal{P}_{2}$ . Hence, the tail $v_{\chi_{x}}$ with largest $y$ -coordinate of an edge in $E(\chi_{x})$ is uniquely defined by $\Gamma_{x}$ . This implies that the edge sets $E(\chi_{1})$ and $E(\chi_{2})$ coincide, as they are both obtained from $E(\chi_{x})$ by replacing the edges outgoing from $v_{\chi_{x}}$ with the edges incoming into $v_{\chi_{x}}$ in $G$ . Since $E(\chi_{1})=E(\chi_{2})$ and $\chi_{1}\neq\chi_{2}$ , it follows that $\chi_{1}$ and $\chi_{2}$ must differ in the way such keys map the tails of the edges incoming into $v_{\chi_{x}}$ to the points of $S$ . Then the UPSEs yielded by $\mathcal{P}_{1}$ and $\mathcal{P}_{2}$ are different, a contradiction.
•

To prove Item ii, we show that, if there exists an UPSE $\Gamma$ of $G$ on $S$ , then there exists a path in $\mathcal{D}$ from $n_{\mathcal{S}}$ to $n_{\mathcal{T}}$ that yields $\Gamma$ . For $i=1,\dots,n$ , let $S_{i}$ be the set that consists of the lowest $i$ points of $S$ . Also, for $i=1,\dots,n-1$ , let $\Gamma_{i}$ be the restriction of $\Gamma$ to the vertices of $G$ mapped to $S_{i}$ and to all their incident edges, including those whose other end-vertex is not in $S_{i}$ . We claim that there exists a path $\mathcal{P}_{i}$ in $\mathcal{D}$ that starts from a node $n_{i}$ and ends at $n_{\mathcal{T}}$ such that: (1) the set $E(\mathcal{P}_{i})$ coincides with the set of edges that are embedded in $\Gamma_{i}$ ; (2) the embedding of the edges in $E(\mathcal{P}_{i})$ defined by the keys $\chi$ corresponding to nodes in $\mathcal{P}_{i}$ is the same as in $\Gamma_{i}$ ; and (3) let $\chi_{i}$ be the key corresponding to $n_{i}$ , then $E(\chi_{i})$ contains all and only the edges $e$ of $G$ such that an end-vertex of $e$ is mapped by $\Gamma$ to a point in $S_{i}$ and the other end-vertex of $e$ is mapped by $\Gamma$ to a point not in $S_{i}$ . The claim implies Item ii, as when $i=n-1$ , we have that $E(\mathcal{P}_{n-1})$ is the edge set of $G$ , by (1), and that the embedding of the edges in $E(\mathcal{P}_{n-1})$ defined by the keys $\chi$ corresponding to nodes in $\mathcal{P}_{n-1}$ is $\Gamma$ , by (2), hence $(n_{\mathcal{S}},\chi_{n-1})\cup\mathcal{P}_{n-1}$ is the desired path from $n_{\mathcal{S}}$ to $n_{\mathcal{T}}$ that yields $\Gamma$ .

In order to prove the claim, we proceed by induction. In the base case, we have $i=1$ , hence $S_{1}$ consists only of the point $p_{s}$ and $\Gamma_{1}$ is the restriction of $\Gamma$ to all the edges incident to $s$ . Since $\Gamma$ is an UPSE, $\Gamma_{1}$ is an embedding of such edges in which $s$ lies on $p_{s}$ and any two heads of such edges are not aligned with $p_{s}$ . Hence, by construction, there is a key $\chi$ such that $E(\chi)$ consists of the set of edges incident to $s$ , such that $Q[\chi]=\{\bot\}$ , and such that the embedding of the edges in $E(\chi)$ on $S$ defined by $\chi$ is $\Gamma_{1}$ . It follows that $\mathcal{D}$ contains a node $n(\chi)$ corresponding to $\chi$ , and thus a path $\mathcal{P}_{1}=(n(\chi),n_{\mathcal{T}})$ with the properties required by the claim.

For the inductive case, we have $i>1$ . Let $p_{i}$ be the point of $S_{i}$ with highest $y$ -coordinate and let $v_{i}$ be the vertex of $G$ mapped to $p_{i}$ by $\Gamma$ . By induction, there exists a path $\mathcal{P}_{i-1}$ in $\mathcal{D}$ that starts from a node $n_{i-1}$ and ends at $n_{\mathcal{T}}$ such that: (1) the set $E(\mathcal{P}_{i-1})$ is the set of edges embedded in $\Gamma_{i-1}$ ; (2) the embedding of the edges in $E(\mathcal{P}_{i-1})$ defined by the keys corresponding to nodes in $\mathcal{P}_{i-1}$ defines $\Gamma_{i-1}$ ; and (3) let $\chi_{i-1}$ be the key corresponding to $n_{i-1}$ , then $E(\chi_{i-1})$ contains all and only the edges whose end-vertices are mapped by $\Gamma$ one to a point in $S_{i-1}$ and the other to a point not in $S_{i-1}$ . Note that (3) ensures that all the edges incoming into $v_{i}$ are in $E(\chi_{i-1})$ .

Consider the edge set $H_{i}$ composed of the edges outgoing from $v_{i}$ and of the edges in $E(\chi_{i-1})$ , except for those incoming into $v_{i}$ . We prove that $H_{i}$ is an $st$ -cutset. Indeed, by (3), every edge of $G$ that in $\Gamma$ starts from a point below $p_{i}$ and ends at a point above $p_{i}$ is in $E(\chi_{i-1})$ . Then $H_{i}$ comprises all the edges that start from $p_{i}$ or from a point below $p_{i}$ and end at a point above $p_{i}$ . Hence, the removal of the edges of $H_{i}$ splits $G$ into two connected subgraphs, one induced by the vertices (including $s$ ) mapped by $\Gamma$ to $S_{i}$ , and one induced by the vertices (including $t$ ) mapped by $\Gamma$ to the points above $p_{i}$ .

Since $H_{i}$ is an $st$ -cutset, there exists a key $\chi_{i}$ such that $E(\chi_{i})=H_{i}$ and the edges of $E(\chi_{i})$ are embedded in $\chi_{i}$ as in $\Gamma_{i}$ . Note that $p_{i}$ is the tail of an edge in $E(\chi_{i})$ with largest $y$ -coordinate, hence our algorithm, starting from the $st$ -cutset $E(\chi_{i})$ , removes the edges outgoing from $v_{i}$ , and adds the edges incoming into $v_{i}$ , thus it constructs the $st$ -cutset $E(\chi_{i-1})$ and, from there, the key $\chi_{i-1}$ in which the edges of $E(\chi_{i-1})$ are mapped as in $\Gamma_{i-1}$ . The algorithm then inserts $\chi_{i-1}$ in $Q[\chi_{i}]$ , and hence the digraph $\mathcal{D}$ contains the edge $(n_{i},n_{i-1})$ , where $n_{i}$ is the node of $\mathcal{D}$ corresponding to $\chi_{i}$ . This completes the induction, hence the proof of the claim and the one of Item ii.

It remains to discuss the running time of our enumeration algorithm. Concerning the set-up time, the table $Q$ can be constructed in $\mathcal{O}(kn^{4k}\log n)$ time, by Claim 3. Also, the digraph $\mathcal{D}$ can be constructed in linear time in the size of $Q$ , which is $\mathcal{O}(kn^{4k}\log n)$ by Claim 2; indeed, the edges outgoing from a node $n(\chi)$ in $\mathcal{D}$ are those toward the nodes whose corresponding keys are in $Q[\chi]$ . Concerning the space usage, again by Claim 2, we have that $Q$ and $\mathcal{D}$ have $\mathcal{O}(kn^{4k}\log n)$ size. Finally, we discuss the delay of our algorithm. The paths from $n_{\mathcal{S}}$ to $n_{\mathcal{T}}$ have $\mathcal{O}(n)$ size. Between an UPSE and the next one, at most two paths are traversed (one to backtrack and one to again reach $n_{\mathcal{T}}$ ). The total number of edges which are deleted from or added to the current embedding when traversing such paths is in $\mathcal{O}(n)$ , hence the delay of our algorithm is $\mathcal{O}(n)$ . We get the following.

Theorem 4.5.

Let $G$ be a $n$ -vertex planar $st$ -graph whose maximum $st$ -cut has size $k$ and let $S$ be a set of $n$ points. It is possible to enumerate all UPSEs of $G$ on $S$ with $\mathcal{O}(n)$ delay, using $\mathcal{O}(kn^{4k}\log n)$ space, after $\mathcal{O}(kn^{4k}\log n)$ set-up time.

5 Planar st-Graphs Composed of Two Monotone st-Paths

In this section, we discuss a special, and in our opinion interesting, case of Theorem 4.4, namely the one in which the underlying graph of the given planar $st$ -graph is a cycle. Applying Theorem 4.4 to this setting would yield an $\mathcal{O}(n^{8})$ -time testing algorithm. Now, based on a characterization of the positive instances, we give a much faster algorithm for this case, provided that the points of $S$ are in general position.

Theorem 5.1.

Let $G$ be an $n$ -vertex planar $st$ -graph consisting of two $st$ -paths $P_{L}$ and $P_{R}$ , and let $S$ be a pointset with $n$ points in general position. We have that $G$ admits an UPSE on $S$ with $P_{L}$ to the left of $P_{R}$ if and only if $|P_{L}|\geq|\operatorname{\mathcal{H}_{L}}(S)|$ and $|P_{R}|\geq|\operatorname{\mathcal{H}_{R}}(S)|$ . Also, it can be tested in $\mathcal{O}(n\log n)$ time whether $G$ admits an UPSE on $S$ .

Proof 5.2.

Provided the characterization in the statement holds, we can easily test whether $G$ admits an UPSE on $S$ by computing the convex hull $\operatorname{\mathcal{CH}}(S)$ of $S$ (which can be done in $\mathcal{O}(n\log n)$ time), deriving the sets $\operatorname{\mathcal{H}_{L}}(S)$ and $\operatorname{\mathcal{H}_{R}}(S)$ (which can be done in $O(n)$ time, by scanning $\operatorname{\mathcal{CH}}(S)$ ), and finally comparing their sizes with the ones of $P_{L}$ and $P_{R}$ (which can be done in $O(1)$ time). Therefore, in the following we focus on proving the characterization.

For the necessity, suppose for a contradiction that there exists an UPSE on $S$ with $P_{L}$ to the left of $P_{R}$ and that $|P_{L}|<|\operatorname{\mathcal{H}_{L}}(S)|$ ; the case in which $|P_{R}|<|\operatorname{\mathcal{H}_{R}}(S)|$ is analogous. Since $|P_{L}\cup P_{R}|=n$ , a vertex $d$ of $P_{R}$ must be drawn on a point in $\operatorname{\mathcal{H}_{L}}(S)$ . Consider the subpath $P_{d}$ of $P_{R}$ between $s$ and $d$ . The drawing of $P_{d}$ splits $\operatorname{\mathcal{CH}}(S)$ into two closed regions, to the left and to the right of $P_{d}$ . In any UPSE of $G$ on $S$ with $P_{L}$ to the left of $P_{R}$ , we have that $P_{L}$ lies in both regions, namely it lies in the region to the left of $P_{d}$ with the edge incident to $s$ and it lies in the region to the right of $P_{d}$ at $t$ . Hence, the drawing of $P_{L}$ crosses the drawing of $P_{d}$ , and thus the one of $P_{R}$ , a contradiction.

In the following, we prove the sufficiency by induction on the size of $S$ (and, thus, of $V(G)$ ). We give some preliminary definitions; see Figures 5, 6 and 7. Let $p_{s}$ and $p_{t}$ be the south and north extreme of $S$ , respectively. Consider the line $\ell_{st}$ through $p_{s}$ and $p_{t}$ . Let $S_{L}$ ( $S_{R}$ ) be the set consisting of the points of $S$ lying in the closed half-plane delimited by $\ell_{st}$ that includes all points that lie to the left (resp. right) of $\ell_{st}$ , including $p_{s}$ and $p_{t}$ . Note that $\operatorname{\mathcal{H}_{L}}(S)\subseteq S_{L}$ and $\operatorname{\mathcal{H}_{R}}(S)\subseteq S_{R}$ . Moreover, since $S$ is in general position, it holds that $S_{L}\cap S_{R}=\{p_{s},p_{t}\}$ .

In the base case, it either holds that (a) $S_{L}=\{p_{s},p_{t}\}$ and $|\operatorname{\mathcal{H}_{R}}(S)|=|P_{R}|$ , or (b) $S_{R}=\{p_{s},p_{t}\}$ and $|\operatorname{\mathcal{H}_{L}}(S)|=|P_{L}|$ . We discuss the former case (see Figure 5), as the latter case is symmetric. In this case, an UPSE $\Gamma$ of $G$ on $S$ clearly exists and is, in fact, unique. In particular, the drawing of $P_{R}$ in $\Gamma$ coincides with the right envelope $\mathcal{E}_{R}(S)$ of $S$ , while the drawing of $P_{L}$ in $\Gamma$ is the $y$ -monotone polyline that assigns to the $j$ -th internal vertex of $P_{L}$ (when traversing $P_{L}$ from $s$ to $t$ ) the point of $S_{R}\setminus\operatorname{\mathcal{H}_{R}}(S)$ with the $j$ -th smallest $y$ -coordinate. Since each of such paths is $y$ -monotone, it is not self-crossing. Also, no edge of $P_{L}$ crosses an edge of $P_{R}$ , as the drawing of $P_{R}$ in $\Gamma$ coincides with $\mathcal{E}_{R}(S)$ .

If the base case does not hold, then we distinguish two cases based on whether both $S_{L}$ and $S_{R}$ contain a vertex different from $p_{s}$ and $p_{t}$ (Case A), or only one of them does (Case B). In the following, we assume that in Case B the set $S_{R}$ contains a vertex different from $p_{s}$ and $p_{t}$ , the case in which only $S_{L}$ contains a vertex different from $p_{s}$ and $p_{t}$ can be treated symmetrically. More formally, in Case A we have that $\{p_{s},p_{t}\}\subset S_{L}$ and $\{p_{s},p_{t}\}\subset S_{R}$ , whereas in Case B we have that $S_{L}=\{p_{s},p_{t}\}$ and $\{p_{s},p_{t}\}\subset S_{R}$ . Note that, in Case B, since the conditions of the base case do not apply and by the hypothesis of the statement, we have that $|P_{R}|>|\operatorname{\mathcal{H}_{R}}(S)|$ holds.

If Case A holds, we distinguish two subcases. In Case A1, it holds $|P_{L}|\geq|S_{L}|$ , whereas in Case A2, it holds $|P_{L}|<|S_{L}|$ . We discuss Case A1 (see Figure 6); Case A2 can be treated symmetrically, given that in this case it holds that $|P_{R}|\geq|S_{R}|$ .

Suppose that Case A1 holds true. Then $\operatorname{\mathcal{H}_{L}}(S)$ contains a point $p$ different from $p_{s}$ and $p_{t}$ ; see Figure 6(a). Since by the hypotheses of this case $|P_{L}|\geq|S_{L}|\geq|\operatorname{\mathcal{H}_{L}}(S)|$ and $|\operatorname{\mathcal{H}_{L}}(S)|\geq 3$ , we have that $P_{L}$ contains at least one internal vertex. Let $S^{\prime}=S\setminus\{p\}$ , let $P^{\prime}_{L}$ be an $st$ -path with $|P^{\prime}_{L}|=|P_{L}|-1$ , and let $G^{\prime}$ be the planar $st$ -graph $P^{\prime}_{L}\cup P_{R}$ . Since $|\operatorname{\mathcal{H}_{L}}(S^{\prime})|\leq|S_{L}|-1$ and since $|S_{L}|\leq|P_{L}|$ , we have that $|\operatorname{\mathcal{H}_{L}}(S^{\prime})|\leq|P_{L}|-1=|P^{\prime}_{L}|$ . Thus, the graph $G^{\prime}$ and the pointset $S^{\prime}$ satisfy the conditions of the statement. By induction, we have that $G^{\prime}$ admits an UPSE $\Gamma^{\prime}$ on $S^{\prime}$ , see Figure 6(b).

We show how to modify $\Gamma^{\prime}$ to obtain an UPSE $\Gamma$ of $G$ on $S$ as follows; see Figures 6(b) and 6(c). The drawing of $P_{R}$ is the same in $\Gamma$ as in $\Gamma^{\prime}$ . Let $h_{p}$ be the horizontal line passing through $p$ . Since $\Gamma^{\prime}$ is an UPSE of $G^{\prime}$ on $S^{\prime}$ and since $y(p_{s})<y(p)<y(p_{t})$ , we have that $h_{p}$ intersects the drawing of $P^{\prime}_{L}$ in a single point. Such a point belongs to a segment that is the image of an edge $e_{p}$ of $P^{\prime}_{L}$ . Let $d$ and $q$ be the extremes of such a segment that are the images of the tail and of the head of $e_{p}$ in $\Gamma^{\prime}$ , respectively. We show how to modify the drawing of $P^{\prime}_{L}$ to obtain a $y$ -monotone drawing of $P_{L}$ that does not intersect $P_{R}$ . To this aim, we replace the drawing of $e_{p}$ with the $y$ -monotone polyline composed of the segments $\overline{dp}$ and $\overline{pq}$ . Note that such a polyline lies in the interior of the region delimited by the segment $\overline{dq}$ (representing $e_{p}$ ) and by the horizontal rays originating at $d$ and $q$ and directed leftward. Due to the fact that $P^{\prime}_{L}$ is represented as a $y$ -monotone polyline in $\Gamma^{\prime}$ , such a region is not traversed by the drawing of any edge. Thus, $\Gamma$ is an UPSE of $G$ on $S$ . We refer to the described procedure as the $p$ -leftward-outer-extension of $\Gamma^{\prime}$ ; a $p$ -rightward-outer-extension of $\Gamma^{\prime}$ is defined symmetrically.

If Case B holds, recall that $S_{L}=\{p_{s},p_{t}\}\subset S_{R}$ , and since the base case does not apply, we have that $|P_{R}|>|\operatorname{\mathcal{H}_{R}}(S)|$ . Let $p$ be any point in $\operatorname{\mathcal{H}_{R}}(S)\setminus\{p_{s},p_{t}\}$ and $S^{\prime}=S\setminus\{p\}$ . By the conditions of Case B, the path $P_{R}$ contains at least one internal vertex. We let $P^{\prime}_{R}$ be an $st$ -path with $|P^{\prime}_{R}|=|P_{R}|-1$ , and we let $G^{\prime}$ be the $st$ -graph $P_{L}\cup P^{\prime}_{R}$ . We distinguish two cases based on the size of $\operatorname{\mathcal{H}_{R}}(S^{\prime})$ . In Case B1, it holds $|P^{\prime}_{R}|\geq|\operatorname{\mathcal{H}_{R}}(S^{\prime})|$ , whereas in Case B2, it holds $|P^{\prime}_{R}|<|\operatorname{\mathcal{H}_{R}}(S^{\prime})|$ .

In Case B1, the pair $(G^{\prime},S^{\prime})$ satisfies the conditions of the statement. In particular, it either matches the conditions of the base case or again those of Case B. Thus, since $|S^{\prime}|=|S|-1$ (and $|V(G^{\prime})|=|V(G)|-1$ ), we can inductively construct an UPSE $\Gamma^{\prime}$ of $G^{\prime}$ on $S^{\prime}$ , and obtain an UPSE of $G$ on $S$ via a $p$ -rightward-outer-extension of $\Gamma^{\prime}$ .

In Case B2, which is the most interesting, we proceed as follows; see Figure 7. Let $p^{+}$ be the point of $\operatorname{\mathcal{H}_{R}}(S)$ with the smallest $y$ -coordinate and above $p$ and let $p^{-}$ be the point of $\operatorname{\mathcal{H}_{R}}(S)$ with the largest $y$ -coordinate and below $p$ . Let $X$ be the set of points of $S$ that lie in the interior of the triangle $\Delta p^{+}pp^{-}$ , including $p^{+}$ and $p^{-}$ and excluding $p$ . Clearly, the right envelope of $\operatorname{\mathcal{CH}}(X)$ forms a subpath of the right envelope of $\operatorname{\mathcal{CH}}(S^{\prime})$ . The set $\operatorname{\mathcal{H}_{R}}(X)$ consists of $p^{-}$ , $p^{+}$ , and of $k$ vertices not belonging to $\operatorname{\mathcal{H}_{R}}(S)$ , depicted as squares in Figure 7(a). Denote by $k^{*}=|P_{R}|-|\operatorname{\mathcal{H}_{R}}(S)|$ the number of points in the interior of $\operatorname{\mathcal{CH}}(S)$ that need to be the image of a vertex of $P_{R}$ in an UPSE of $G$ on $S$ . Observe that $k>k^{*}>0$ holds true. Indeed, $k^{*}>0$ holds true since $(G,S)$ does not satisfy the conditions of the base case, and $k>k^{*}$ holds true since $(G,S)$ does not satisfy the conditions of Case B1. Let $p^{\wedge}$ be the point of $\operatorname{\mathcal{H}_{R}}(S^{\prime})$ with the smallest $y$ -coordinate and above $p$ , and let $p^{\vee}$ be the point with the largest $y$ -coordinate and below $p$ . Up to renaming, let $a_{0}=p^{+},a_{1},\dots,a_{\alpha}=p^{\wedge}$ be the subsequence of points of $\mathcal{E}_{R}(X)$ encountered when traversing $\mathcal{E}_{R}(X)$ from $p^{+}$ to $p^{\wedge}$ and observe that these points have decreasing $y$ -coordinates. Similarly, let $b_{0}=p^{-},b_{1},\dots,b_{\gamma}=p^{\vee}$ be the subsequence of points of $\mathcal{E}_{R}(X)$ encountered when traversing $\mathcal{E}_{R}(X)$ from $p^{-}$ to $p^{\vee}$ and observe that these points have increasing $y$ -coordinates. We define the set $X^{*}\subset\operatorname{\mathcal{H}_{R}}(X)$ as follows. If $k^{*}\leq\alpha$ , then let $X^{*}_{\wedge}=\{a_{i}|1\leq i\leq k^{*}\}$ and $X^{*}_{\vee}=\emptyset$ , otherwise let $X^{*}_{\wedge}=\{a_{i}|1\leq i\leq\alpha\}$ and $X^{*}_{\wedge}=\{b_{i}|1\leq i\leq k^{*}-\alpha\}$ . In both cases, let $X^{*}=X^{*}_{\wedge}\cup X^{*}_{\vee}$ . Observe that $|X^{*}|=k^{*}$ . Also, by the definition of $k^{*}$ , the path $P_{R}$ contains $\operatorname{\mathcal{H}_{R}}(S)-2+k^{*}$ internal vertices and since $\operatorname{\mathcal{H}_{R}}(S)\geq 3$ in Case B, we have that $P_{R}$ contains at least $k^{*}+1$ internal vertices.

Let $S^{*}=S\setminus X^{*}$ , let $P^{*}_{R}$ be an $st$ -path with $|P_{R}|-k^{*}$ vertices, and let $G^{*}$ be the $st$ -graph $P_{L}\cup P^{*}_{R}$ . We have that the pair $(G^{*},S^{*})$ satisfies the conditions of the statement, and in particular the base case. In fact, $|P^{*}_{R}|=|P_{R}|-k^{*}$ , and by the definition of $k^{*}$ , we have that $|P_{R}|-k^{*}=|\operatorname{\mathcal{H}_{R}}(S)|$ . Moreover, by construction, $\operatorname{\mathcal{H}_{R}}(S)=\operatorname{\mathcal{H}_{R}}(S^{*})$ , since the vertices of $X^{*}$ lie in the interior of $\operatorname{\mathcal{CH}}(S)$ . Thus, since $|S^{*}|=|S|-k^{*}$ , by induction $G^{*}$ admits an UPSE $\Gamma^{*}$ on $S^{*}$ ; see Figure 7(b).

We now show how to transform $\Gamma^{*}$ into an UPSE $\Gamma$ of $G$ on $S$ . Since the base case applies to $(G^{*},S^{*})$ , we have that the endpoints of the edges of $P^{*}_{R}$ are consecutive along $\mathcal{E}_{R}(S)$ . In particular, there exist two adjacent edges $e^{-}$ and $e^{+}$ of $P^{*}_{R}$ such that the tail of $e^{-}$ is mapped to $p^{-}$ , the head of $e^{-}$ , which is the tail of $e^{+}$ , is mapped to $p$ , and the head of $e^{+}$ is mapped to $p^{+}$ . Therefore, the UPSE $\Gamma$ of $G$ on $S$ can be obtained from $\Gamma^{*}$ as follows; see Figure 7(c). We initialize $\Gamma=\Gamma^{*}$ . The drawing of $P_{L}$ is the same in $\Gamma$ as in $\Gamma^{*}$ . Next, we show how to modify the drawing of $P^{*}_{R}$ to obtain a $y$ -monotone drawing $P_{R}$ that does intersect the drawing of $P_{L}$ and uses the same points as $P^{*}_{L}$ and the points in $X^{*}$ . To this aim, we replace the drawing of $e^{+}$ with the (unique) $y$ -monotone polyline connecting $p$ and $p^{+}$ that passes through all the points in $X^{*}_{\wedge}$ . Also, we replace the drawing of $e^{-}$ with the (unique) $y$ -monotone polyline connecting $p^{-}$ and $p$ that passes through all the points in $X^{*}_{\vee}$ ; note that $X^{*}_{\vee}$ might be empty, in which case the polyline still coincides with the drawing of $e^{-}$ . This concludes the construction of $\Gamma$ . To see that $\Gamma$ is an UPSE of $G$ on $S$ observe that the above polylines (i) are each non-self-crossing, as they are $y$ -monotone, (ii) do not cross with each other as they entirely lie either above or below $p$ (and only meet at $p$ ), and (iii) do not cross any edge of $\Gamma^{\prime}$ as they lie in the region $F$ (shaded gray in Figures 7(b) and 7(c)) obtained by subtracting from the triangle $\Delta p^{+}pp^{-}$ (interpreted as a closed region) all the points of $\operatorname{\mathcal{CH}}(X)$ . Indeed, observe that in $\Gamma^{*}$ , the region $F$ is not traversed by any edge and that the only points of $S^{*}$ that lie on the boundary of $F$ are $p$ and the points in $\operatorname{\mathcal{H}_{R}}(X)\setminus X^{*}$ .

6 Enumerating Non-crossing Monotone Hamiltonian Cycles

Theorem 5.1 allows us to test whether an $n$ -vertex planar $st$ -graph $G$ composed of two $st$ -paths can be embedded as a non-crossing monotone Hamiltonian cycle on a set $S$ of $n$ points. We now show an efficient algorithm for enumerating all the non-crossing monotone Hamiltonian cycles on $S$ . Figure 8 shows two non-crossing monotone Hamiltonian cycles on a pointset.

Theorem 6.1.

Let $S$ be a set of $n$ points. It is possible to enumerate all the non-crossing monotone Hamiltonian cycles on $S$ with $\mathcal{O}(n)$ delay, using $\mathcal{O}(n^{2})$ space, after $\mathcal{O}(n^{2})$ set-up time.

Let $p_{1},\dots,p_{n}$ be the points of $S$ , ordered by increasing $y$ -coordinates. This order can be computed in $\mathcal{O}(n\log n)$ time. For $i\in[n]$ , let $S_{i}=\{p_{1},\dots,p_{i}\}$ . A bipath $B$ on $S_{i}$ consists of two non-crossing monotone paths $L$ and $R$ on $S_{i}$ , each of which might be a single point, such that (see Figure 9):

(i)

$L$ and $R$ start at $p_{1}$ ;
(ii)

each point of $S_{i}$ is the image of an endpoint of a segment of $B$ ; and
(iii)

if $L$ and $R$ both have at least one segment, then $L$ is to the left of $R$ .

We say that a bipath $B$ is extensible if there exists a non-crossing monotone Hamiltonian cycle on $S$ whose restriction to $S_{i}$ is $B$ . Consider a bipath $B$ on $S_{i}$ with $1<i<n$ . Let $p_{\ell(B)}$ and $p_{r(B)}$ be the endpoints of $L$ and $R$ with the highest $y$ -coordinate, respectively. Suppose first that $\ell(B)>r(B)$ , that is, $p_{\ell(B)}$ is higher than $p_{r(B)}$ . Then note that $\ell(B)=i$ ; also, it might be that $r(B)=1$ . Consider the ray $\rho(p_{r(B)},S_{\ell(B)}\setminus S_{r(B)})$ ; recall that this is the rightmost ray starting at $p_{r(B)}$ and passing through a point of $S_{\ell(B)}\setminus S_{r(B)}$ . We denote by $\mathcal{R}(B)$ the open region of the plane strictly to the right of $\rho(p_{r(B)},S_{\ell(B)}\setminus S_{r(B)})$ and strictly above the horizontal line through $p_{\ell(B)}$ ; see Figure 10(a). Similarly, if $p_{r(B)}$ is higher than $p_{\ell(B)}$ , then $\mathcal{L}(B)$ is the open region of the plane strictly to the left of the leftmost ray $\ell(p_{\ell(B)},S_{r(B)}\setminus S_{\ell(B)})$ from $p_{\ell(B)}$ through a point of $S_{r(B)}\setminus S_{\ell(B)}$ and strictly above the horizontal line through $p_{r(B)}$ ; see Figure 10(b).

For any $i\in[n-1]$ , we say that a bipath $B$ on $S_{i}$ is safe if:

(i)

$i=1$ ; or
(ii)

$i>1$ , $p_{\ell(B)}$ is higher than $p_{r(B)}$ , and $|\mathcal{R}(B)\cap S|\geq 1$ ; or
(iii)

$i>1$ , $p_{r(B)}$ is higher than $p_{\ell(B)}$ , and $|\mathcal{L}(B)\cap S|\geq 1$ .

We have the following lemma.

Lemma 6.2.

A bipath $B$ is extensible if and only it is safe.

Proof 6.3.

First, we prove the necessity. Suppose that $B$ is extensible and let $C$ be any non-crossing monotone Hamiltonian cycle on $S$ whose restriction to $S_{i}$ is $B$ . Also suppose, for a contradiction, that $B$ is not safe, which implies that $i>1$ . Assume that $p_{\ell(B)}$ is higher than $p_{r(B)}$ , as the other case is symmetric. Then we have $\mathcal{R}(B)\cap S=\varnothing$ . Let $\overline{p_{r(B)}p^{\prime}_{r(B)}}$ be the segment of $C$ such that $y(p^{\prime}_{r(B)})>y(p_{r(B)})$ . Since all points in $S_{\ell(B)}\setminus S_{r(B)}$ belong to $L$ , we have $p^{\prime}_{r(B)}$ lies strictly above the horizontal line through $p_{\ell(B)}$ . This, together with the fact that $S$ contains no point strictly above the horizontal line through $p_{\ell(B)}$ and to the right of the ray $\rho(p_{r(B)},S_{\ell(B)}\setminus S_{r(B)})$ , implies that the ray $\rho(p_{r(B)}p^{\prime}_{r(B)})$ lies to the left of the ray $\rho(p_{r(B)},S_{\ell(B)}\setminus S_{r(B)})$ , which implies that the segment $(p_{r(B)},p^{\prime}_{r(B)})$ crosses the path $L$ , a contradiction to the fact that $C$ is non-crossing.

Second, we prove the sufficiency. Suppose that $B$ is safe. We show how to construct a non-crossing monotone Hamiltonian cycle $C$ on $S$ whose restriction to $S_{i}$ is $B$ . Assume first that $i=1$ . Then $C$ can be constructed as the union of two monotone paths. The first path is one of the two paths between $s$ and $t$ on the boundary of the convex hull of $S$ . The second path from $s$ to $t$ traverses all the points of $S$ that are not on the first path, in increasing order of $y$ -coordinate. Assume next that $i>1$ and refer to Figure 11. Assume also that $p_{\ell(B)}$ is higher than $p_{r(B)}$ , as the other case is symmetric. Then $\mathcal{R}(B)$ contains some points of $S$ . Consider the rightmost ray $\rho(p_{r(B)},S\setminus S_{i})$ starting from $p_{r(B)}$ and passing through a point $p$ in $S\setminus S_{i}$ . Observe that $\overline{p_{r(B)}p}$ is a segment on the boundary of the convex hull of $S\setminus S_{r(B)-1}$ . Hence, we can augment $R$ so that it becomes a monotone path from $s$ to $t$ , by adding to it the part of the boundary of the convex hull of $S\setminus S_{r(B)-1}$ from $p_{r(B)}$ to $t$ (by proceeding in counter-clockwise direction on this boundary from $p_{r(B)}$ to $t$ ). Also, we can augment $L$ so that it becomes a monotone path from $s$ to $t$ by making it pass through all the points in $S\setminus S_{i}$ that are not used by $R$ , and finishing at $t$ .

We now describe our algorithm. The algorithm implicitly defines and explores a search tree $T$ . The leaves of $T$ have level $n$ and correspond to non-crossing monotone Hamiltonian cycles on $S$ . The internal nodes at level $i$ correspond to extensible bipaths on $S_{i}$ and have at most two children each. The exploration of $T$ performed by our enumeration algorithm is a depth-first traversal. When a node $\mu$ is visited, the number of its children is established. If $\mu$ has at least one child, the visit proceeds with any child of $\mu$ . Otherwise, $\mu$ is a leaf; then the visit proceeds with any unvisited child of the ancestor of $\mu$ that has largest level, among the ancestors of $\mu$ that have unvisited children.

•

The algorithm starts at the root of $T$ , which corresponds to the (unique) safe bipath on $S_{1}$ .
•

At each node $\mu$ at level $i\in[n-2]$ of $T$ , corresponding to a bipath $B(\mu)$ , we construct either one or two bipaths on $S_{i+1}$ , associated with either one or two children of $\mu$ , respectively. Let $L(\mu)$ and $R(\mu)$ be the left and right non-crossing monotone paths composing $B(\mu)$ , respectively, and let $p_{\ell(B(\mu))}$ and $p_{r(B(\mu))}$ be the endpoints of $L(\mu)$ and $R(\mu)$ with the highest $y$ -coordinate, respectively. If $\overline{p_{\ell(B(\mu))}p_{i+1}}$ does not cross $R(\mu)$ , then let $B_{L}=B(\mu)\cup\overline{p_{\ell(B(\mu))}p_{i+1}}$ . We test whether $B_{L}$ is a safe bipath and, in the positive case, add to $\mu$ a child $\mu_{L}$ corresponding to $B_{L}$ . Analogously, if $\overline{p_{r(B(\mu))}p_{i+1}}$ does not cross $L(\mu)$ , then we test whether $B_{R}=B(\mu)\cup\overline{p_{r(B(\mu))}p_{i+1}}$ is a safe bipath and, in the positive case, add to $\mu$ a child $\mu_{R}$ corresponding to $B_{R}$ . Note that the algorithm guarantees that each node at a level smaller than or equal to $n-1$ of $T$ is safe, and thus, by Lemma 6.2, extensible.
•

Finally, at each node $\mu$ at level $n-1$ , we add a leaf $\lambda$ to $\mu$ corresponding to the non-crossing monotone Hamiltonian cycle $B(\mu)\cup\overline{p_{\ell(B(\mu))}p_{n}}\cup\overline{p_{r(B(\mu))}p_{n}}$ . Note that, since $\mu$ is extensible, such a cycle is indeed non-crossing.

In order to complete the proof of Theorem 6.1, we show what follows:

(i)

Each node of $T$ at level $i\neq n$ is internal.
(ii)

Each leaf corresponds to a non-crossing monotone Hamiltonian cycle on $S$ .
(iii)

Distinct leaves correspond to different non-crossing monotone Hamiltonian cycles on $S$ .
(iv)

For each non-crossing monotone Hamiltonian cycle on $S$ , there exists a leaf of $T$ corresponding to it.
(v)

Using $\mathcal{O}(n^{2})$ pre-processing time and $\mathcal{O}(n^{2})$ space, the algorithm enumerates each non-crossing monotone Hamiltonian cycle on $S$ with $\mathcal{O}(n)$ delay.

•

To prove Item i, we show that the leaves of $T$ have all level $n$ . Consider a node $\mu$ of $T$ with level $i<n-1$ , we prove that it has a child in $T$ . Recall that $B(\mu)$ is safe, otherwise it would not had been added to $T$ , and thus, by Lemma 6.2, it is extensible. Hence, there exists a non-crossing monotone Hamiltonian cycle $C$ on $S$ whose restriction to $S_{i}$ is $B(\mu)$ . Also, the restriction of $C$ to $S_{i+1}$ is a bipath $B^{\prime}(\mu)$ on $S_{i+1}$ which coincides with $B(\mu)$ , except that it contains either the segment $\overline{p_{\ell(B(\mu))}p_{i+1}}$ or the segment $\overline{p_{r(B(\mu))}p_{i+1}}$ . Since $B^{\prime}(\mu)$ is the restriction of $C$ to $S_{i+1}$ , it is extensible and thus, by Lemma 6.2, it is safe. It follows that $\mu$ has a child corresponding to $B^{\prime}(\mu)$ , which is inserted in $T$ when adding either the segment $\overline{p_{\ell(B(\mu))}p_{i+1}}$ or the segment $\overline{p_{r(B(\mu))}p_{i+1}}$ to $B(\mu)$ . The proof that a node with level $n-1$ is not a leaf is analogous.
•

To prove Item ii, consider a leaf $\lambda$ and its parent $\mu$ in $T$ . Note that $\mu$ is associated with a safe bipath $B(\mu)$ on $S_{n-1}$ ; by Lemma 6.2, we have that $B(\mu)$ is extensible. Since $B(\mu)$ is extensible, the (unique) monotone Hamiltonian cycle on $S$ whose restriction to $S_{n-1}$ is $B(\mu)$ is non-crossing. This cycle corresponds to $\lambda$ and is added to $T$ when visiting $\mu$ .
•

To prove Item iii, suppose for a contradiction that there exist two leaves $\lambda_{1}$ and $\lambda_{2}$ associated with two monotone Hamiltonian cycles $C_{1}$ and $C_{2}$ , respectively, with $C_{1}=C_{2}$ . Let $\mu$ be the lowest common ancestor of $\lambda_{1}$ and $\lambda_{2}$ in $T$ . Let $j$ be the level of $\mu$ . Denote by $\mu_{i}$ the child of $\mu$ leading to $\lambda_{i}$ , with $i\in\{1,2\}$ . By the construction of $T$ , we have that exactly one of the bipaths $B(\mu_{1})$ and $B(\mu_{2})$ contains the segment $\overline{p_{\ell(B(\mu))}p_{j+1}}$ , while the other one contains the segment $\overline{p_{r(B(\mu))}p_{j+1}}$ . This contradicts the fact that $C_{1}=C_{2}$ .
•

To prove Item iv, let $C$ be a non-crossing monotone Hamiltonian cycle on $S$ . Consider the safe bipath $B$ on $S_{n-1}$ obtained by removing from $C$ the point $p_{n}$ , together with its two incident segments. It suffices to show that $T$ contains a node $\mu$ such that $B=B(\mu)$ . In fact, in this case, $\mu$ is an extensible node of level $n-1$ whose unique child in $T$ is the leaf corresponding to $C$ . To prove that $T$ contains such a node $\mu$ , we prove by induction that, for every level $i=1,\dots,n-1$ , the tree $T$ contains a node corresponding to the restriction $B_{i}$ of $B$ to $S_{i}$ . The base case trivially holds. For the inductive case, suppose that $T$ contains a node $\nu$ whose associated bipath $B(\nu)$ is $B_{i-1}$ . Then $B_{i}$ is obtained by adding either the segment $\overline{p_{\ell(B(\nu))}p_{i}}$ or the segment $\overline{p_{r(B(\nu))}p_{i}}$ to $B_{i-1}$ . Since $B_{i}$ is extensible, by Lemma 6.2 it is safe, and hence $\nu$ has a child in $T$ corresponding to $B_{i}$ .
•

Finally, we prove Item v. To this aim, we compute in $\mathcal{O}(n^{2})$ time a table $D$ of $\mathcal{O}(n^{2})$ size that allows us to test in $\mathcal{O}(1)$ time whether a bipath $B$ on $S_{i}$ , with $i\in\{2,\dots,n-1\}$ , is safe. This, together with the fact that $T$ has $n$ levels, implies that the algorithm’s delay is in $\mathcal{O}(n)$ .

The table $D$ is indexed by triples $\langle p_{a},p_{b},X\rangle$ , where $p_{a},p_{b}\in S$ with $a<b$ and $X\in\{L,R\}$ . Each entry of $D$ contains a Boolean value $D[p_{a},p_{b},X]$ defined as follows.
- -
  
  Suppose that $X=R$ . Consider the rightmost ray $\rho(p_{a},S_{b}\setminus S_{a})$ starting from $p_{a}$ and passing through a point in $S_{b}\setminus S_{a}$ . We denote by $\mathcal{R}(p_{a},p_{b})$ the open region of the plane strictly to the right of $\rho(p_{a},S_{b}\setminus S_{a})$ and strictly above the horizontal line through $p_{b}$ . Then, $D[p_{a},p_{b},R]=\texttt{True}$ if and only if $\mathcal{R}(p_{a},p_{b})\cap S\neq\emptyset$ .
- -
  
  Suppose now that $X=L$ . Consider the leftmost ray $\ell(p_{a},S_{b}\setminus S_{a})$ starting from $p_{a}$ and passing through a point in $S_{b}\setminus S_{a}$ . We denote by $\mathcal{L}(p_{a},p_{b})$ the open region of the plane strictly to the left of the ray $\ell(p_{a},S_{b}\setminus S_{a})$ and strictly above the horizontal line passing through $p_{b}$ . Then, $D[p_{a},p_{b},L]=\texttt{True}$ if and only if $\mathcal{L}(p_{a},p_{b})\cap S\neq\emptyset$ .
For each fixed $a\in[n-1]$ , we show how to compute all the entries $D[p_{a},p_{b},R]$ with $b=a+1,a+2,\dots,n$ in overall $\mathcal{O}(n)$ time. This sums up to $\mathcal{O}(n^{2})$ time over all the entries $D[p_{a},p_{b},R]$ of $D$ with $a=1,2,\dots,n-1$ and $b=a+1,a+2,\dots,n$ . The computation of the entries $D[p_{a},p_{b},L]$ of $D$ is done symmetrically.

(a)

(b)

Figure 12: (a) For any $b\in\{a+1,a+2,\dots,c-1\}$ , we have that $p_{c}$ is in $\mathcal{R}(p_{a},p_{b})$ . (b) For any $b\in\{c,c+1,\dots,n\}$ , we have that $\mathcal{R}(p_{a},p_{b})$ is empty. Region $\mathcal{R}(p_{a},p_{b})$ is shaded gray.

We compute the point $p_{c}$ with $c>a$ such that the ray $\rho(p_{a},p_{c})=\rho(p_{a},S\setminus S_{a})$ is the rightmost among the rays starting from $p_{a}$ and passing through a point in $S\setminus S_{a}$ . This can be done in $\mathcal{O}(n)$ time by inspecting the points $p_{a+1},p_{a+2},\dots,p_{n}$ . Then, we set $D[p_{a},p_{b},R]=\texttt{True}$ for all the points $p_{b}$ with $b=a+1,a+2,\dots,c-1$ and $D[p_{a},p_{b},R]=\texttt{False}$ for all the points $p_{b}$ with $b=c,c+1,\dots,n$ . Indeed, for any $b\in\{a+1,a+2,\dots,c-1\}$ , we have that $p_{c}$ is in $\mathcal{R}(p_{a},p_{b})$ , since it is strictly above the horizontal line through $p_{b}$ (given that $b<c$ ) and strictly to the right of the ray $\rho(p_{a},p_{b})$ (given that $\rho(p_{a},p_{c})$ is the rightmost among the rays starting from $p_{a}$ and passing through a point in $S\setminus S_{a}$ ); see Figure 12(a). Also, for any $b\in\{c,c+1,\dots,n\}$ , we have that $p_{c}$ is in $S_{b}\setminus S_{a}$ , and, by definition of $p_{c}$ , no point is strictly to the right of the ray $\rho(p_{a},p_{c})$ , hence $\mathcal{R}(p_{a},p_{b})$ is empty; see Figure 12(b).

Items i, ii, iii and iv prove the correctness of the enumeration algorithm, while Item v proves its efficiency. This concludes the proof of Theorem 6.1.

7 Conclusions and Open Problems

We addressed basic pointset embeddability problems for upward planar graphs. We proved that UPSE testing is \NP-hard even for planar $st$ -graphs composed of internally-disjoint $st$ -paths and for directed trees composed of directed root-to-leaf paths. For planar $st$ -graphs, we showed that UPSE Testing can be solved in $O(n^{4k})$ time, where $k$ is the maximum $st$ -cutset of $G$ , and we provided an algorithm to enumerate all UPSEs of $G$ on $S$ with $\mathcal{O}(n)$ worst-case delay. We also showed how to enumerate all monotone polygonalizations of a given pointset with $\mathcal{O}(n)$ worst-case delay. We point out the following open problems.

•

Our NP-hardness proofs for UPSE testing use the fact that the points are not in general position. Given a directed tree $T$ on $n$ vertices and a set $S$ of $n$ points in general position, is it \NP-hard to decide whether $T$ has an UPSE on $S$ ?
•

Can UPSE testing be solved in polynomial time or does it remain \NP-hard if the input is a maximal planar $st$ -graph?
•

We proved that UPSE testing for a planar $st$ -graph is in XP with respect to the size of the maximum $st$ -cutset of $G$ . Is the problem in FPT with respect to the same parameter? Are there other interesting parameterizations for the problem?
•

Let $S$ be a pointset and $\mathcal{P}$ be a non-crossing path on a subset of $S$ . Is it possible to decide in polynomial time whether $\mathcal{P}$ can be extended to a polygonalization of $S$ ? A positive answer would imply an algorithm with polynomial delay for enumerating the polygonalizations of a pointset, with the same approach as the one we adopted in this paper for monotone polygonalizations.

References

[1] Victor Alvarez, Karl Bringmann, Radu Curticapean, and Saurabh Ray. Counting triangulations and other crossing-free structures via onion layers. Discret. Comput. Geom., 53(4):675–690, 2015. URL: https://doi.org/10.1007/s00454-015-9672-3, doi:10.1007/S00454-015-9672-3.
[2] Patrizio Angelini, Fabrizio Frati, Markus Geyer, Michael Kaufmann, Tamara Mchedlidze, and Antonios Symvonis. Upward geometric graph embeddings into point sets. In Ulrik Brandes and Sabine Cornelsen, editors, Graph Drawing - 18th International Symposium, GD 2010, Konstanz, Germany, September 21-24, 2010. Revised Selected Papers, volume 6502 of Lecture Notes in Computer Science, pages 25–37. Springer, 2010. doi:10.1007/978-3-642-18469-7_3.
[3] Elena Arseneva, Pilar Cano, Linda Kleist, Tamara Mchedlidze, Saeed Mehrabi, Irene Parada, and Pavel Valtr. Upward point set embeddings of paths and trees. In Ryuhei Uehara, Seok-Hee Hong, and Subhas C. Nandy, editors, WALCOM: Algorithms and Computation - 15th International Conference and Workshops, WALCOM 2021, Yangon, Myanmar, February 28 - March 2, 2021, Proceedings, volume 12635 of Lecture Notes in Computer Science, pages 234–246. Springer, 2021. doi:10.1007/978-3-030-68211-8_19.
[4] Giuseppe Di Battista, Walter Didimo, Luca Grilli, Fabrizio Grosso, Giacomo Ortali, Maurizio Patrignani, and Alessandra Tappini. Small point-sets supporting graph stories. In Patrizio Angelini and Reinhard von Hanxleden, editors, Graph Drawing and Network Visualization - 30th International Symposium, GD 2022, Tokyo, Japan, September 13-16, 2022, Revised Selected Papers, volume 13764 of Lecture Notes in Computer Science, pages 289–303. Springer, 2022. doi:10.1007/978-3-031-22203-0_21.
[5] Therese Biedl and Martin Vatshelle. The point-set embeddability problem for plane graphs. In Tamal K. Dey and Sue Whitesides, editors, Proceedings of the 28th ACM Symposium on Computational Geometry, Chapel Hill, NC, USA, June 17-20, 2012, pages 41–50. ACM, 2012. doi:10.1145/2261250.2261257.
[6] Carla Binucci, Emilio Di Giacomo, Walter Didimo, Alejandro Estrella-Balderrama, Fabrizio Frati, Stephen G. Kobourov, and Giuseppe Liotta. Upward straight-line embeddings of directed graphs into point sets. Comput. Geom., 43(2):219–232, 2010. URL: https://doi.org/10.1016/j.comgeo.2009.07.002, doi:10.1016/J.COMGEO.2009.07.002.
[7] Prosenjit Bose. On embedding an outer-planar graph in a point set. Comput. Geom., 23(3):303–312, 2002. doi:10.1016/S0925-7721(01)00069-4.
[8] Prosenjit Bose, Michael McAllister, and Jack Snoeyink. Optimal algorithms to embed trees in a point set. J. Graph Algorithms Appl., 1(2):1–15, 1997. URL: https://doi.org/10.7155/jgaa.00002, doi:10.7155/JGAA.00002.
[9] Sergio Cabello. Planar embeddability of the vertices of a graph using a fixed point set is np-hard. J. Graph Algorithms Appl., 10(2):353–363, 2006. URL: https://doi.org/10.7155/jgaa.00132, doi:10.7155/JGAA.00132.
[10] Netzahualcoyotl Castañeda and Jorge Urrutia. Straight line embeddings of planar graphs on point sets. In Frank Fiala, Evangelos Kranakis, and Jörg-Rüdiger Sack, editors, Proceedings of the 8th Canadian Conference on Computational Geometry, Carleton University, Ottawa, Canada, August 12-15, 1996, pages 312–318. Carleton University Press, 1996. URL: http://www.cccg.ca/proceedings/1996/cccg1996_0052.pdf.
[11] Gi-Sang Cheon, Hong Joon Choi, Guillermo Esteban, and Minho Song. Enumeration of bipartite non-crossing geometric graphs. Discret. Appl. Math., 317:86–100, 2022. URL: https://doi.org/10.1016/j.dam.2022.04.008, doi:10.1016/J.DAM.2022.04.008.
[12] Thomas H. Cormen, Charles E. Leiserson, Ronald L. Rivest, and Clifford Stein. Introduction to Algorithms, 4th Edition. MIT Press, 2022. URL: https://mitpress.mit.edu/9780262046305/introduction-to-algorithms/.
[13] Giuseppe Di Battista, Peter Eades, Roberto Tamassia, and Ioannis G. Tollis. Graph Drawing: Algorithms for the Visualization of Graphs. Prentice-Hall, 1999.
[14] Giuseppe Di Battista and Roberto Tamassia. Algorithms for plane representations of acyclic digraphs. Theor. Comput. Sci., 61:175–198, 1988. doi:10.1016/0304-3975(88)90123-5.
[15] Reinhard Diestel. Graph Theory, 4th Edition, volume 173 of Graduate texts in mathematics. Springer, 2012.
[16] David Eppstein. Counting polygon triangulations is hard. Discret. Comput. Geom., 64(4):1210–1234, 2020. URL: https://doi.org/10.1007/s00454-020-00251-7, doi:10.1007/S00454-020-00251-7.
[17] David Eppstein. Non-crossing hamiltonian paths and cycles in output-polynomial time. In Erin W. Chambers and Joachim Gudmundsson, editors, 39th International Symposium on Computational Geometry, SoCG 2023, June 12-15, 2023, Dallas, Texas, USA, volume 258 of LIPIcs, pages 29:1–29:16. Schloss Dagstuhl - Leibniz-Zentrum für Informatik, 2023. URL: https://doi.org/10.4230/LIPIcs.SoCG.2023.29, doi:10.4230/LIPICS.SOCG.2023.29.
[18] Philippe Flajolet and Marc Noy. Analytic combinatorics of non-crossing configurations. Discret. Math., 204(1-3):203–229, 1999. doi:10.1016/S0012-365X(98)00372-0.
[19] M. R. Garey and D. S. Johnson. Computers and Intractability: A Guide to the Theory of NP-Completeness. W. H. Freeman, first edition edition, 1979.
[20] Emilio Di Giacomo, Henry Förster, Daria Kokhovich, Tamara Mchedlidze, Fabrizio Montecchiani, Antonios Symvonis, and Anaïs Villedieu. On 1-bend upward point-set embeddings of st-digraphs. In José A. Soto and Andreas Wiese, editors, LATIN 2024: Theoretical Informatics - 16th Latin American Symposium, Puerto Varas, Chile, March 18-22, 2024, Proceedings, Part I, volume 14578 of Lecture Notes in Computer Science, pages 3–18. Springer, 2024. doi:10.1007/978-3-031-55598-5_1.
[21] Francesco Giordano, Giuseppe Liotta, Tamara Mchedlidze, Antonios Symvonis, and Sue Whitesides. Computing upward topological book embeddings of upward planar digraphs. J. Discrete Algorithms, 30:45–69, 2015. URL: https://doi.org/10.1016/j.jda.2014.11.006, doi:10.1016/J.JDA.2014.11.006.
[22] Peter Gritzmann, Bojan Mohar, János Pach, and Richard Pollack. Embedding a planar triangulation with vertices at specified points. The American Mathematical Monthly, 98(2):165, 1991. doi:10.2307/2323956.
[23] Lenwood S. Heath, Sriram V. Pemmaraju, and Ann N. Trenk. Stack and queue layouts of directed acyclic graphs: Part I. SIAM J. Comput., 28(4):1510–1539, 1999.
[24] Michael Kaufmann, Tamara Mchedlidze, and Antonios Symvonis. On upward point set embeddability. Comput. Geom., 46(6):774–804, 2013. URL: https://doi.org/10.1016/j.comgeo.2012.11.008, doi:10.1016/J.COMGEO.2012.11.008.
[25] Michael Kaufmann and Roland Wiese. Embedding vertices at points: Few bends suffice for planar graphs. J. Graph Algorithms Appl., 6(1):115–129, 2002. URL: https://doi.org/10.7155/jgaa.00046, doi:10.7155/JGAA.00046.
[26] Dániel Marx and Tillmann Miltzow. Peeling and nibbling the cactus: Subexponential-time algorithms for counting triangulations and related problems. In Sándor P. Fekete and Anna Lubiw, editors, 32nd International Symposium on Computational Geometry, SoCG 2016, June 14-18, 2016, Boston, MA, USA, volume 51 of LIPIcs, pages 52:1–52:16. Schloss Dagstuhl - Leibniz-Zentrum für Informatik, 2016. URL: https://doi.org/10.4230/LIPIcs.SoCG.2016.52, doi:10.4230/LIPICS.SOCG.2016.52.
[27] Tamara Mchedlidze. Upward planar embedding of an n-vertex oriented path on O(n ${}^{\mbox{2}}$ ) points. Comput. Geom., 46(8):1003–1008, 2013.
[28] Joseph S. B. Mitchell and Joseph O’Rourke. Computational geometry column 42. Int. J. Comput. Geom. Appl., 11(5):573–582, 2001. doi:10.1142/S0218195901000651.
[29] Rahnuma Islam Nishat, Debajyoti Mondal, and Md. Saidur Rahman. Point-set embeddings of plane 3-trees. Comput. Geom., 45(3):88–98, 2012. URL: https://doi.org/10.1016/j.comgeo.2011.09.002, doi:10.1016/J.COMGEO.2011.09.002.
[30] Andreas Razen and Emo Welzl. Counting plane graphs with exponential speed-up. In Cristian S. Calude, Grzegorz Rozenberg, and Arto Salomaa, editors, Rainbow of Computer Science - Dedicated to Hermann Maurer on the Occasion of His 70th Birthday, volume 6570 of Lecture Notes in Computer Science, pages 36–46. Springer, 2011. doi:10.1007/978-3-642-19391-0_3.
[31] Michael Ian Shamos and Dan Hoey. Geometric intersection problems. In 17th Annual Symposium on Foundations of Computer Science, Houston, Texas, USA, 25-27 October 1976, pages 208–215. IEEE Computer Society, 1976. doi:10.1109/SFCS.1976.16.
[32] Manuel Wettstein. Counting and enumerating crossing-free geometric graphs. J. Comput. Geom., 8(1):47–77, 2017. URL: https://doi.org/10.20382/jocg.v8i1a4, doi:10.20382/JOCG.V8I1A4.
[33] Katsuhisa Yamanaka, David Avis, Takashi Horiyama, Yoshio Okamoto, Ryuhei Uehara, and Tanami Yamauchi. Algorithmic enumeration of surrounding polygons. Discret. Appl. Math., 303:305–313, 2021. URL: https://doi.org/10.1016/j.dam.2020.03.034, doi:10.1016/J.DAM.2020.03.034.

Upward Pointset Embeddings of Planar s⁢t𝑠𝑡stitalic_s italic_t-Graphs