Coloring non-crossing strings

For a family of geometric objects in the plane $\mathcal{F}=\{S_1,\ldots,S_n\}$, define $\chi(\mathcal{F})$ as the least integer $\ell$ such that the elements of $\mathcal{F}$ can be colored with $\ell$ colors, in such a way that any two intersecting objects have distinct colors. When $\mathcal{F}$ is a set of pseudo-disks that may only intersect on their boundaries, and such that any point of the plane is contained in at most $k$ pseudo-disks, it can be proven that $\chi(\mathcal{F})\le 3k/2 + o(k)$ since the problem is equivalent to cyclic coloring of plane graphs. In this paper, we study the same problem when pseudo-disks are replaced by a family $\mathcal{F}$ of pseudo-segments (a.k.a. strings) that do not cross. In other words, any two strings of $\mathcal{F}$ are only allowed to"touch"each other. Such a family is said to be $k$-touching if no point of the plane is contained in more than $k$ elements of $\mathcal{F}$. We give bounds on $\chi(\mathcal{F})$ as a function of $k$, and in particular we show that $k$-touching segments can be colored with $k+5$ colors. This partially answers a question of Hlin\v{e}n\'y (1998) on the chromatic number of contact systems of strings.


Introduction
For a family F = {S 1 , . . . , S n } of subsets of a set Ω, the intersection graph G(F) of F is defined as the graph with vertex-set F, in which two vertices are adjacent if and only if the corresponding sets have non-empty intersection.
For a graph G, the chromatic number of G, denoted χ(G), is the least number of colors needed in a proper coloring of G (a coloring such that any two adjacent vertices have distinct colors). When talking about a proper coloring of a family F of subsets of a given set, we implicitly refer to a proper coloring of the intersection graph of F, thus the chromatic number χ(F) is defined in a natural way.
The chromatic number of families of geometric objects in the plane have been extensively studied since the sixties [2,13,18,19,20]. Since it is possible to construct sets of pairwise intersecting (straight-line) segments of any size, the chromatic number of sets of segments in the plane is unbounded in general. However, Erdős conjectured that trianglefree intersection graphs of segments in the plane have bounded chromatic number (see [12]). This was recently disproved [23]. The conjecture of Erdős initiated the study of the chromatic number of families of geometric objects in the plane as a function of their clique number, the maximum size of subsets of the family that pairwise intersect [8]. In this paper, we consider families of geometric objects in the plane for which the chromatic number only depends on local properties of the families, such as the maximum number of objects containing a given point of the plane.
Consider a set F = {R 1 , . . . , R n } of pseudo-disks (subsets of the plane which are homeomorphic to a closed disk) such that the intersection of the interiors of any two pseudo-disks is empty. Let H F be the planar hypergraph with vertex set F, in which the hyperedges are the maximal sets of pseudo-disks whose intersection is non-empty. A proper coloring of F is equivalent to a coloring of H F in which all the vertices of each hyperedge have distinct colors. If every point is contained in at most k pseudo-disks, Borodin conjectured that there exists such a coloring of H F with at most 3 2 k colors [3]. It was recently proved that this conjecture holds asymptotically [1] (not only in the plane, but also on any fixed surface). As a consequence, F can be properly colored with 3 2 k +o(k) colors.
It seems natural to investigate the same problem when pseudo-disks are replaced by pseudo-segments. These are continuous injective functions from [0, 1] to R 2 and are usually referred to as strings. Consider a set S = {C 1 , . . . , C n } of such strings. We will always assume that any two strings intersect in a finite number of points. We say that S is touching if no pair of strings of S cross, and that it is k-touching if furthermore at most k strings can "touch" in any point of the plane, i.e., any point of the plane is contained in at most k strings (see Figure 1(a) for an example).
Note that the family of all touching sets of strings contains all contact systems of strings, defined as sets of strings such that the interior of any two strings have empty intersection. In other words, if c is a contact point in the interior of a string s, all the strings containing c distinct from s end at c. In [15], Hliněný studied contact system of strings such that all the strings ending at c leave from the same side of s. Such a representation is said to be one-sided (see Figure 1(b) for an example). It was proved in [15] that if a contact system of strings is k-touching and every contact point is one-sided, then the strings can be colored with 2k colors.
In this paper, our aim is to study k-touching sets of strings in their full generality. Observe that if S is k-touching, k might be much smaller than the maximum degree of G(S). However, based on the cases of pseudo-disks and contact system of strings, we conjectured the following in the conference version of this paper [5]: For some constant c > 0, any k-touching set of strings can be colored with ck colors.
This conjecture was subsequently proved by Fox and Pach [9], who showed that any ktouching set of strings can be colored with 6ek+1 colors (where e is the base of the natural logarithm). In Section 2, we show how to slightly improve their bound for small values of k. We also show that for any odd k, the clique on 9 2 (k − 1) vertices can be represented as a set of k-touching strings, so the best possible constant c in Conjecture 1.1 is between 4.5 and 6e ≈ 16.3.
In Section 3, we give improved bounds when any two strings can intersect a bounded number of times. In Section 4, we restrict ourselves to contact systems of strings where any two strings intersect at most once (called 1-intersecting), which were previouly studied by Hliněný [15]. He asked whether there is a constant c such that every one-sided 1intersecting k-touching contact system of strings is (k + c)-colorable. We prove that they are ( 4k 3 + 6)-colorable, and that every k-touching contact system of straight-line segments is (k + 5)-colorable. Note that we do not need our contact systems to be one-sided.
Before giving general bounds, let us first mention two classical families of touching strings for which coloring problems are well understood.
If a k-touching set of strings has the property that the interior of each string is disjoint from all the other strings, then each string can be thought of as an edge of some (planar) graph with maximum degree k. By a classical theorem of Shannon, the strings can then be colored with 3k/2 colors. If moreover, any two strings intersect at most once, then the electronic journal of combinatorics 23(4) (2016), #P4.4 they can be colored with k + 1 colors by a theorem of Vizing (even with k colors whenever k 7, using a more recent result of Sanders and Zhao [24]). In this sense, all the problems considered in this article can be seen as a extension of edge-coloring of planar graphs. An x-monotone string is a string such that every vertical line intersects it in at most one point. Alternatively, it can be defined as the curve of a continuous function from an interval of R to R. Sets of k-touching x-monotone strings are closely related to bar kvisibility graphs. A bar k-visibility graph is a graph whose vertex-set consists of horizontal segments in the plane (bars), and two vertices are adjacent if and only if there is a vertical segment connecting the two corresponding bars, and intersecting no more than k other bars. It is not difficult to see that the graph of any set of k-touching x-monotone strings is a spanning subgraph of some bar (k − 2)-visibility graph, while any bar (k − 2)visibility graph can be represented as a set of k-touching x-monotone strings. Using this correspondence, it directly follows from [4] that k-touching x-monotone strings are (6k − 6)-colorable, and that the complete graph on 4k − 4 vertices can be represented as a set of k-touching x-monotone strings. If the left-most point of each x-monotone string intersects the vertical line x = 0, then it directly follows from [7] that the strings can be colored with 2k − 1 colors (and the complete graph on 2k − 1 vertices can be represented by k-touching x-monotone strings in this specific way).

General bounds
Before proving our first results on the structure of sets of k-touching strings in general, we make two important observations: Observation 2.1. The family of intersection graphs of 2-touching strings is exactly the class of planar graphs.
The class of planar graphs being exactly the class of intersection graphs of 2-touching pseudo-disks (see [17]) it is clear that planar graphs are intersection graphs of 2-touching strings (by taking a connected subset of the boundaries of each pseudo-disk). Furthermore, every intersection graph of 2-touching strings is contained in an intersection graph of 2touching pseudo-disks, and is thus planar. Indeed, it is easy given a set of 2-touching strings S = {C 1 , . . . , C n } to draw a set of 2-touching pseudo-disks F = {R 1 , . . . , R n } such that C i ⊂ R i for every i ∈ [1, n].
Observation 2.2. We can assume without loss of generality that the strings in any set of k-touching strings are polygonal lines (i.e. each string is the union of finitely many straight-line segments) and that no endpoint of a string of S coincides with an intersection between strings of S.
To see this, take a set S of k-touching strings and consider the following graph G: the vertices are the contact points and the endpoints of the strings of S, and the edges connect two points if they are consecutive in some string of S. The resulting graph is planar, but might contain multiple edges. Subdivide each edge once, and observe that the electronic journal of combinatorics 23(4) (2016), #P4.4 the resulting graph H is a simple planar (finite) graph, and each string of S is the union of some edges of H. Since H is a simple planar graph, by Fáry's theorem [6] it has an equivalent drawing in which all the edges are (straight-line) segments. If some endpoint of a string of S coincides with an intersection x between strings of S, one can prolong each string ending at x with a small enough segment, and Observation 2.2 follows.
The following result was proved by Fox and Pach [9] (this shows Conjecture 1.1). In the following, e is the base of the natural logarithm.
Their theorem is a direct consequence of the following bound on the number of edges in a graph represented by a set of k-touching strings. Their proof is inspired by the probabilistic proof of the Crossing Lemma.
). Any graph represented by a set of n k-touching strings has less than 3ekn edges.
When k = 3, their proof can easily be optimized to show that the number of edges is less than 12n. Hence, every such graph has a vertex with degree less than 24. These graphs are thus 23-degenerate and have chromatic number at most 24. We now show how to modify their proof to slightly improve this bound.
This is a direct consequence of the following lemma.
Proof. Let S be a set of n 3-touching strings, and let m be the number of edges in the corresponding intersection graph G. By Observation 2.2, we can assume that each string of S is a polygonal line (a union of finitely many segments) and that no endpoint of a string of S coincides with an intersection between strings of S. Consider a point p contained in three different strings s 0 , s 1 , s 2 . Then there is a small disk D centered in p that only intersects the strings s 0 , s 1 , s 2 , and is such that the boundary C of D intersects each s i (for i = 0, 1, 2) in exactly two points, say p i , p i . Assume that walking around C in clockwise order, we see p 0 , p 0 , p 1 , p 1 , p 2 , p 2 . Then for i = 0, 1, 2, we replace s i ∩ D by a new string s i between p i and p i as follows. For each i = 0, 1, 2, let q i be a point of D that is after p i and before p i+1 in clockwise order (with indices modulo 3). For two points x, y, let S(x, y) denote the (straight-line) segment between x and y. Then we replace s i ∩ D (the portion of s i between p i and p i ) by the concatenation of S(p i , q i ), S(q i , q i−1 ), and S(q i−1 , p i ) (see Figure 2). Note that the resulting set of strings is still 3-touching, and the intersection graph remains unchanged. Repeating this operation if necessary, we can assume without loss of generality that for any three strings s 0 , s 1 , s 2 as above, we see p 0 , p 0 , p 1 , p 2 , p 2 , p 1 when walking around C in clockwise order. In this case we say that s 1 is "sandwiched" between s 0 and s 2 at p (see Figure 3, left).  Figure 2: A local modification turning one 3-touching point into three 2-touching points, without changing the intersection graph.
We now define two spanning subgraphs G 1 and G 0 of G as follows. Two strings a and c are adjacent in G 1 if they intersect and for every intersection point p of a and c, there exists a third string b that is sandwiched between a and c at p. For every pair of strings a and c adjacent in G 1 , let P 1 (a, c) be an arbitrarily chosen intersection point between a and c. By definition, there is a unique string that is sandwiched between a and c at P 1 (a, c).
Two strings a and c are adjacent in G 0 if they are adjacent in G but not in G 1 , i.e., if there exists an intersection point P 0 (a, c) of a and c such that either P 0 (a, c) is not contained in another string or P 0 (a, c) is contained in a third string b that is not sandwiched between a and c. For i = 0, 1, the edge-set of G i is denoted by E i , and the cardinality of E i is denoted by m i .
For each edge ab ∈ E 1 , assume that ab gives a charge of 1 to the unique string c sandwiched between a and b at P 1 (a, b). Let N 0 (c) be the set of neighbors of c in G 0 . The total charge ρ(c) received by c is at most the number of pairs of vertices x, y ∈ N 0 (c) such that in S \ {c}, P 0 (x, y) is a 2-contact point. If we modify the set of strings intersecting c so as to only preserve those 2-touching points (all the other pairs of strings are made disjoint), we obtain a planar graph with vertex-set N 0 (c) and with at least ρ(c) edges. It follows that ρ(c) 3|N 0 (c)| − 6. Summing for all strings c, we obtain that the total charge c∈S ρ(c) = m 1 6m 0 − 6n. Since m 0 + m 1 = m, we have 7m 0 m + 6n, as claimed. We are now ready to prove the lemma. We select each string of S uniformly at random with probability p (to be chosen later). In the subset of chosen strings, we slightly modify the electronic journal of combinatorics 23(4) (2016), #P4.4 each 3-touching point x as follows. There is a small disk D with boundary C centered in x that only intersects the three strings s 0 , s 1 , s 2 containing x, and (with the same notation as before), we can assume without loss of generality that we see p 0 , p 0 , p 1 , p 2 , p 2 , p 1 when walking around C in clockwise order. Then for some point q on s 1 , close to x, we replace s 0 ∩D by the concatenation of the segments S(p 0 , q) and S(q, p 0 ) (see Figure 3). Note that this preserves all edges in G 0 . Let S be the new set of strings. For each edge ab ∈ E 0 , the probability that ab appears as an edge in S is p 2 , and for each edge ab ∈ E 1 , the probability that ab appears as an edge in S is at least the probability that a and b were both selected and the string c sandwiched between a and b at P 1 (a, b) was not selected, which is p 2 (1 − p). The expected number of vertices of the graph represented by S is then pn and its expected number of edges is at least By Observation 2.1, the graph is planar and therefore, p 2 7 (m(7−6p)+6pn) 3pn. This can be rewritten as m n 21−6p 2 p(7−6p) . Taking p = 3− 11/2, we obtain m 6 7 (6+ The ideas of Lemma 2.6 can be used to slightly improve the multiplicative constant in Theorem 2.3 for all k. Since our improvement is minor (we obtain a bound of 6k × 2.686 instead of 6k × 2.718), we omit the details. We now show that the constant c in Conjecture 1.1 is at least 9 2 .
the electronic journal of combinatorics 23(4) (2016), #P4.4 7 Theorem 2.8. For every odd k 1, k = 2 + 1, there exists a set of k-touching strings S k such that the strings of S k pairwise touch and such that |S k | = 9 = 9(k − 1)/2. Thus χ(S k ) = 9 = 9k 2 − 5. Proof. Consider n touching strings s 1 , . . . , s n that all intersect n points c 1 , . . . , c n in the same order (see the set of bold strings in Figure 4(a) for an example when n = 4), and call this set of strings an n-braid. For some > 0, take three 2 -braids S 1 , S 2 , S 3 , and for i = 1, 2, 3, connect each of the strings of S i to a different intersection point of S i+1 (with indices taken modulo 3), while keeping the set of strings touching (see Figure 4(a)). We call this set of touching strings a 2 -sun. Observe that a 2 -sun contains 6 strings that pairwise intersect, and that each intersection point contains at most 2 + 1 strings. Moreover, each of the 6 strings has an end that is incident to the infinite face.
We now consider three 2 -suns R 1 , R 2 , R 3 . Each of them has 6 strings with an end incident to the outerface. For each i = 1, 2, 3, we arbitrarily divide the strings leaving R i into two sets of 3 consecutive strings, say R i,i+1 and R i,i−1 . For each i = 1, 2, 3, we now take the strings of R i,i+1 and R i+1,i by pairs (one string in R i,i+1 , one string in R i+1,i ), and identify these two strings into a single string. This can be made in such a way that the resulting set of (6 × 3 )/2 = 9 strings is still (2 + 1)-touching (see Figure 4(b), where the three 2 -suns are represented by dashed circles, and only the portion of the strings leaving the suns is displayed for the sake of clarity). Hence we obtain a k-touching set of 9k 2 − 5 strings that pairwise intersect, as desired.

µ-intersecting strings
Let S be a k-touching set of strings. The set S is said to be µ-intersecting if any two strings intersect in at most µ points. We denote by H(S) the multigraph associated to S: the vertices of H(S) are the strings of S, and two strings with t common points correspond to two vertices connected by t edges in H(S). Note that the intersection graph G(S) of S is the simple graph underlying H(S).
We prove the following result (which only supersedes Theorem 2.3 for µ 5).
Theorem 3.1. Any k-touching set S of µ-intersecting strings can be properly colored with 3µk colors.
Again, the proof is based on an upper bound on the number of edges of such graphs. between strings of S. Let us slightly modify S in order to obtain a set S of 2-touching and µ-intersecting strings. For that, repeat the following operation while there exists an intersection point c with d(c) > 2. There is a small disk D centered in c such that D only intersects the strings containing c, and for any such string s, s ∩ D is the union of two segments. Pick a string s 1 containing c such that the angle between its two segments at c is minimal, and let s 2 be a string containing c, distinct from s 1 , such that a face f of D − S is bounded by s 1 , s 2 and the boundary of C. Let p 1 , p 1 be the intersection of s 1 and the boundary of D, and let q be a point of s 2 ∩ f distinct from c (and close to c). Then we replace D ∩ s 1 by the concatenation of the straight-line segments S(p 1 , q) and S(q, p 1 ) (see Figure 5). This is similar to the second modification used in Theorem 2.5 and illustrated in Figure 3, which was restricted to the case where c is contained in exactly three strings. Finally, since the number of edges of H(S) is precisely 1 2 c d(c)(d(c) − 1), we have that H(S) has less than 3 2 µkn edges, as desired. In particular, if a k-touching set S of strings is such that any two strings intersect in at most one point, Theorem 3.1 yields a bound of 3k for the chromatic number of S. We suspect that it is far from tight: There is a constant c > 0, such that every k-touching set of 1-intersecting strings can be colored with k + c colors.
In the next section, we show that this conjecture holds for k-touching (straight-line) segments, a special case of 1-intersecting strings. It is interesting to note that even though the bound for k-touching µ-intersecting graphs in Conjecture 1. not depend on µ, the chromatic number of these graphs has some connection with µ: sets of strings with µ = 1 have chromatic number at most 3k, whereas there exists sets of strings with large µ and chromatic number at least 9 2 (k − 1). Note that the constant c in Conjecture 3.3 is at least 4. Figure 6(a) depicts a 3touching set of seven 1-intersecting strings, in which any two strings intersect. Hence, this set requires seven colors. However this construction does not extend to k-touching sets with k 4, it might be that the constant is smaller for higher k. In Figure 6(b), the k-touching set S k contains k + 2 non-crossing strings, and is such that any two strings intersect. Hence, k + 2 colors are required in any proper coloring.

1-Intersecting contact system of strings
In this section, all the sets of strings we consider are 1-intersecting (any two strings intersect in at most one point). An interesting example of 1-intersecting set of strings is any family of non-crossing (straight-line) segments in the plane. Such a family is also known as a contact system of segments, and has been studied in [10], where the authors proved that any bipartite planar graph has a contact representation with horizontal and vertical segments.
More generally, a contact system of strings is a family of strings such that the interiors of the strings are pairwise non-intersecting. In other words, if c is a contact point in the interior of a string s, all the strings containing c and distinct from s end at c. A contact point p is a peak if every string containing p has an end at p. Otherwise, that is if p is an interior point of a string s and an end for all the other strings containing p, p is flat. A flat contact point p is one-sided if all the strings ending at p are on the same side of the unique string whose interior contains p. A contact system of strings in which every flat contact point is one-sided is also said to be one-sided.
It was proved by Hliněný [16] that the intersection graph of any one-sided 2-or 3touching set of segments is planar. Note that as a 2-touching set of segments is always the electronic journal of combinatorics 23(4) (2016), #P4.4 one-sided, it is also 4-colorable. In [22], Ossona de Mendez proved that it is NP-complete to determine whether a 2-touching set of segments is 3-colorable.
In [15], Hliněný studied the clique and chromatic numbers of one-sided k-touching contact systems of strings. He proved that the maximal clique in this class is K k+1 and that the graphs in this class are 2k-colorable. He also asked the following: is there a constant c such that if a contact system of strings is k-touching, 1-intersecting, and onesided, then it is (k +c)-colorable? Note that Conjecture 3.3 would imply a positive answer to this question.
In the first part of this section, we prove that 1-intersecting and k-touching contact systems of strings are 4 3 k + 6 -colorable. In the second part, we show that any ktouching contact system of segments is (k + 5)-colorable. Note that we do not assume the contact systems to be one-sided (but we also show that adding this assumption slightly improves the additive constants in our results).
Theorem 4.1. For any k 3, any 1-intersecting k-touching contact system of strings can be colored with 4 3 k + 6 colors. As in Theorem 2.3, the result is a consequence of a bound on the degeneracy of these graphs: Lemma 4.2. For any k 3, if S is a 1-intersecting k-touching contact system of strings, then G(S) contains a vertex of degree at most 4 3 k + 5. Proof. Assume that there is a counterexample, i.e. a 1-intersecting k-touching contact system S of n strings such that G(S) has minimum degree at least 4 3 k +6. In particular, G(S) has m n( 2 3 k + 3) edges. We take a counterexample for which n is minimal, and with respect to this, m is maximal. Observe that G(S) is connected, since otherwise by minimality of n, some connected component would have a vertex of degree at most 4 3 k + 5, a contradiction. Observe also that if some string of S has at most one contact point, then the corresponding vertex of G(S) has degree at most k − 1 4 3 k + 5, a contradiction. This implies that every string of S has at least two contact points. As a consequence, we can also assume that the two ends of each string of S are contact points (if not, delete the portion of a string between a free end and its first contact point).
Let H(S) be the plane graph whose vertices are the contact points of S, whose edges link two contact points if and only if they are consecutive on a string of S, and whose faces are the connected regions of R 2 \ S.
Let p i and f i be the number of contact points of exactly i strings of S that are respectively peaks and flat. Let us denote by c the total number of contact points, and note that c = k i=2 (p i + f i ). By counting the number of ends of a string of S in two different ways, we obtain that: the electronic journal of combinatorics 23(4) (2016), #P4.4 Consider a one-sided flat contact point p and let s be the unique string such that p is an interior point of s. If we draw a small open disk D containing p, a unique face f of H(S) has the property that f ∩ D is incident to s, and to no other string containing p.
We denote this face f of H(S) by F (p). Remark that since S is 1-intersecting, any face f of H(S) contains at least |F −1 (f )| + 3 vertices, thus at least |F −1 (f )| edges can be added to H(S) (inside f ) with the property that H(S) remains planar. Hence in total, one can add as many edges to H(S) as the number of one-sided contact points, while keeping H(S) planar. Since every flat 2-contact point is one-sided, and every planar graph on c vertices has at most 3c − 6 edges, it follows that H(S) has at most 3c − 6 − f 2 edges. As a consequence, the sum of the degrees of the vertices of H(S) is: By the definition of c, this is equivalent to: Since any pair of strings in S intersects at most once, the number of edges in G(S) satisfies the following equation.
Our goal is to use (1), (2), and (3) to prove that m is less than n( 2 3 k + 3), which is a contradiction. To prove this, we will see that in order to maximize m, we have to set all the values of f i and p i to zero, except for p 2 , f 3 , p k , f k (in other words, the weight has to be concentrated on the extremal variables). Once this is proved, bounding m will be significantly simpler.
Let us consider the linear program (LP1) defined on variables p i and f i with values in R + such that the equation (1) and the inequality (2) are satisfied, and such that the value m defined by (3) is maximized. Here n is considered to be a constant (it is not a variable of the linear program). Note that the solution m * of this problem is clearly an upper bound of the number of edges of G(S).

value of m in (3) is increased by
is a (concave) parabola with g(4) = k 2 −3k +6 > 0 (recall that 4 i k − 1, so k 5) and g(k) = 0, so it is positive in the interval [4, k − 1]. This concludes the proof of the claim.
Claim 4.5. The optimal solutions of (LP1) are such that p i = 0, for every 3 i k − 1.
If for some 3 i k − 1, p i = 0, choose a small > 0 and replace (i) p 2 by p 2 + (k−i) k−2 ; (ii) p i by p i − ; and (iii) p k by p k + i−2 k−2 . Then (1) remains valid, inequality (2) still holds (the left-hand side and the right-hand side remain unchanged), while the value of m in (3) is a (concave) parabola with g(3) = k 2 − 2k + 3 > 0 (recall that 3 i k − 1, so k 4) and g(k) = 0, so it is positive in the interval [3, k − 1]. This concludes the proof of the claim.
It follows from the previous claims that c = By equation (1) and the previous claims, we have 2n = 2p 2 + kp k + 2f 3 + (k − 1)f k . Hence, . This can be rewritten as Now, by equation (3), . We obtain that the graph G(S) has less than n( 2 3 k +3) edges, which is a contradiction. If the contact system we consider is one-sided, the argument we used for flat 2intersection points while establishing (2) in the previous proof works for all flat points, and it follows that H(S) has at most 3c − 6 − k i=2 f i edges. Consequently, inequality (2) can be replaced by the following stronger inequality: the electronic journal of combinatorics 23(4) (2016), #P4.4 Let (LP2) be the new linear program. It is not difficult to check that Claims 4.3 and 4.4 remain satisfied. Moreover, it can be proved that in some optimal solution of (LP2), we also have f 3 = 0. Similar computations give that in this case the graph G(S) contains a vertex of degree at most 4 3 k + 1. As a consequence: Theorem 4.6. For any k 3, any 1-intersecting one-sided k-touching contact system of strings can be colored with 4 3 k + 2 colors. We now show how to modify the proof of Theorem 4.1 to prove that k-touching contact systems of segments are (k + 5)-colorable. For technical reasons, we will consider instead the case of extendible contact system of strings, which we define next.
A pseudo-line is the homeomorphic image of a straight line in the plane. An arrangement of pseudo-lines is a set of pseudo-lines such that any two of them intersect at most once, and when they do they cross each other. We say that a contact system of strings is extendible if there is an arrangement of pseudo-lines, such that each string s of the contact system is contained in a distinct pseudo-line of the arrangement (this pseudo-line is called the support of s). Observe that a contact system of segments is clearly extendible. On the other hand, it was proved by de Fraysseix and Ossona de Mendez [11] that any extendible contact system of strings can be "stretched", i.e. continuously changed to a contact system of segments, while keeping the same underlying intersection graph.
We start with two observations about extendible contact systems of strings (the first one is similar to the first part of Observation 2.2).
Observation 4.7. Let S be an extendible contact system of strings, and let L be a corresponding arrangement of pseudo-lines. We can assume without loss of generality that each string of S is the union of finitely many segments, and each pseudo-line of L is the union of two rays (affine images of {0} × [0, +∞) in R 2 ) and finitely many segments.
The proof is similar to that of Observation 2.2. We consider the plane graph G whose vertices are the endpoints of the strings of S and the intersection points of S∪L, and whose edges connect two points if they are consecutive in some string of S or some pseudo-line of L. Since any two pseudo-lines intersect at most once, the resulting graph H is a simple planar (finite) graph, and by Fáry's theorem [6] it has an equivalent drawing in which all the edges are (straight-line) segments. We can then replace each end of a pseudo-line by finitely many segments and a ray, without changing G(S), and Observation 4.7 follows.
Observation 4.8. Let S be an extendible contact system of strings, and let L be a corresponding arrangement of pseudo-lines. We can assume without loss of generality that for any strings s 1 , s 2 , s 3 ∈ S, if their supporting pseudo-lines intersect in a point p, then s 1 , s 2 , s 3 also intersect in p.
Consider a point p contained in 1 , 2 , 3 ∈ L, and such that the string s 1 ∈ S supported by 1 does not contain p. We now show how to modify 1 so that it avoids p, without changing the other elements of L and the elements of S. By Observation 4.7, we can assume that a small disk D centered in p only intersects the elements of S ∪ L containing the electronic journal of combinatorics 23(4) (2016), #P4.4 p, and the intersection of each of these elements with D consists of the union of at most two segments. By the definition of a pseudo-line arrangement, the pseudo-lines intersecting p pairwise cross. Note that 1 cuts the boundary of D in two arcs, say a, a . We then modify 1 by replacing D ∩ 1 by a. The set S is not modified, and the pseudo-lines intersecting p still pairwise cross, so we obtain a new pseudo-line arrangement. Repeating this operation if necessary, we finally obtain a pseudo-line arrangement satisfying Observation 4.8.
We can now prove the main result of this section. Theorem 4.9. For any k 3, any k-touching extendible contact system of strings can be properly colored with k + 5 colors.
As before, the result is a consequence of a bound on the degeneracy of the corresponding intersection graphs: Lemma 4.10. For any k 3, if S is a k-touching extendible contact system of strings, then G(S) contains a vertex of degree at most k + 4.
Proof. The proof is similar to that of Theorem 4.1. We consider a counterexample S consisting of n k-touching extendible strings. Since G(S) has minimum degree at least k + 5, G(S) has m 1 2 (k + 5) n edges. Again, we take n minimal, and with respect to this, m maximal. As before, we can assume that G(S) is connected and that the two ends of each string are contact points.
We again consider the plane graph H(S) whose vertices are the contact points of S, whose edges link two contact points if and only if they are consecutive on a string of S, and whose faces are the connected regions of R 2 \S. Let p i and f i be the number of contact points of exactly i strings of S that are respectively peaks and flat. Let p = k i=2 p i and c = k i=2 (p i + f i ). Recall that any face f of H(S) contains at least |F −1 (f )| + 3 vertices (where F −1 (f ), defined in the proof of Lemma 4.2, is the number of flat contact points "incident" to f ), thus at least |F −1 (f )| edges can be added to H(S) (inside f ) with the property that H(S) remains planar. We now show that, moreover, the vertices corresponding to the peaks of S all lie on the outerface of H(S). This directly implies that we can add f 2 + p − 3 edges to H(S), while keeping H(S) planar (as a consequence, H(S) has at most 3c − 6 − (f 2 + p − 3) = 3c − 3 − f 2 − p edges).
Indeed, if some peak x of S is not incident to the outerface, we choose a string s containing x and prolong s after x (following the pseudo-line supporting s) until it hits some other segment s (note that since S is extendible, the pseudo-lines supporting s and s intersect at most once, and thus s and s did not intersect previously). Let S be the new contact system of strings. It is still extendible, and by Observation 4.8 we can assume that s ∩ s is only contained in s and s , so the new contact point is 2-touching. Consequently, this new contact system is extendible and k-touching, which contradicts the maximality of m.
It follows that inequality (2) in the proof of Theorem 4.1 can be replaced by: We consider the linear program (LP3) defined on variables p i and f i with values in R + two edges incident to the same vertex, therefore each of the segments s 1 , . . . , s 60 hits an interior point of another segment).  Figure 7 depicts k-touching sets of segments requiring k + 2 colors, for k = 2, 3. However it does not appear to be trivial to extend this construction for any k 4. Note that Figure 7(b) also shows that there are intersection graphs of 3-contact representations of segments (with two-sided contact points) that are not planar (this remark also appears in [14]).