The Complexity of (List) EdgeColoring Reconfiguration Problem^{†}^{†}thanks: This work is partially supported by JSPS KAKENHI Grant Numbers JP26730001 (A. Suzuki), JP15H00849 and JP16K00004 (T. Ito), and JP16K00003 (X. Zhou).
Abstract
Let be a graph such that each edge has its list of available colors, and assume that each list is a subset of the common set consisting of colors. Suppose that we are given two list edgecolorings and of , and asked whether there exists a sequence of list edgecolorings of between and such that each list edgecoloring can be obtained from the previous one by changing a color assignment of exactly one edge. This problem is known to be PSPACEcomplete for every integer and planar graphs of maximum degree three, but any complexity hardness was unknown for the nonlist variant. In this paper, we first improve the known result by proving that, for every integer , the problem remains PSPACEcomplete even if an input graph is planar, bounded bandwidth, and of maximum degree three. We then give the first complexity hardness result for the nonlist variant: for every integer , we prove that the nonlist variant is PSPACEcomplete even if an input graph is planar, of bandwidth linear in , and of maximum degree .
1 Introduction
Recently, reconfiguration problems [10] have been intensively studied in the field of theoretical computer science. The problem arises when we wish to find a stepbystep transformation between two feasible solutions of a combinatorial (search) problem such that all intermediate results are also feasible and each step conforms to a fixed reconfiguration rule, that is, an adjacency relation defined on feasible solutions of the original search problem. (See, e.g., the survey [9] and references in [6, 13].)
1.1 Our problem
In this paper, we study the reconfiguration problem for (list) edgecolorings of a graph [11, 12]. Let be the set of colors. A (proper) edgecoloring of a graph is a mapping such that holds for every two adjacent edges . In list edgecoloring, each edge has a set of colors, called the list of . Then, an edgecoloring of is called a list edgecoloring of if holds for every edge . Figure 1 illustrates three list edgecolorings of the same graph with the same list ; the list of each edge is attached to the edge, and the color assigned to each edge is written in bold with an underline. Clearly, an (ordinary) edgecoloring of is a list edgecoloring of for which holds for every edge of , and hence list edgecoloring is a generalization of edgecoloring.
Ito et al. [12] introduced an adjacency relation defined on list edgecolorings of a graph, and define the list edgecoloring reconfiguration problem, as follows: Suppose that we are given two list edgecolorings of a graph (e.g., the leftmost and rightmost ones in Figure 1), and we are asked whether we can transform one into the other via list edgecolorings of such that each differs from the previous one in only one edge color assignment; such a sequence of list edgecolorings is called a reconfiguration sequence. We call this problem the list edgecoloring reconfiguration problem. For the particular instance of Figure 1, the answer is “yes” as illustrated in the figure, where the edge whose color assignment was changed from the previous one is depicted by a thick line.
For convenience, we call the problem simply the nonlist variant (formally, edgecoloring reconfiguration) if holds for every edge of a given graph.
1.2 Known and related results
Despite recent intensive studies on reconfiguration problems (in particular, for graph colorings [1, 2, 3, 4, 5, 7, 11, 12, 14, 16]), as far as we know, only one complexity result is known for list edgecoloring reconfiguration. Ito et al. [11] proved that list edgecoloring reconfiguration is PSPACEcomplete even when restricted to and planar graphs of maximum degree three. (Since the list of each edge is given as an input, this result implies that the problem is PSPACEcomplete for every integer .) They also gave a sufficient condition for which any two list edgecolorings of a tree can be transformed into each other, which was improved by [12]; but these sufficient conditions do not clarify the complexity status for trees, and indeed it remains open.
As a related problem, the (list) vertexcoloring reconfiguration problem has been studied intensively. (list vertexcoloring reconfiguration and its nonlist variant are defined analogously.) Bonsma and Cereceda [3] proved that vertexcoloring reconfiguration is PSPACEcomplete for every integer . On the other hand, Cereceda et al. [5] proved that both list vertexcoloring reconfiguration and its nonlist variant are solvable in polynomial time for any graph if . Thus, the complexity status of vertexcoloring reconfiguration is analyzed sharply with respect to the number of colors.
Edgecoloring in a graph can be reduced to vertexcoloring in the line graph of . By this reduction, we can solve list edgecoloring reconfiguration for any graph if . However, the reduction does not work the other way, and hence this reduction does not yield any computational hardness result. Indeed, the complexity of edgecoloring reconfiguration was an open question proposed by [11].
1.3 Our contribution
In this paper, we precisely analyze the complexity of (list) edgecoloring reconfiguration; in particular, we give the first complexity result for the nonlist variant.
We first improve the known result for list edgecoloring reconfiguration by proving that, for every integer , the problem remains PSPACEcomplete even if an input graph is planar, bounded bandwidth, and of maximum degree three. Recall that the problem is solvable in polynomial time if , and hence our result gives a sharp analysis of the complexity status with respect to the number of colors. We then give the first complexity result for the nonlist variant: edgecoloring reconfiguration is PSPACEcomplete for every integer and planar graphs whose maximum degrees are and bandwidths are linear in .
We here explain our main technical contribution roughly. (All the details will be given later.) Both our results can be obtained by constructing polynomialtime reductions from Nondeterministic Constraint Logic (NCL, for short), introduced by Hearn and Demaine [8]. This problem is often used to prove the computational hardness of puzzles and games, because a reduction from this problem requires to construct only two types of gadgets, called and and or gadgets. However, there is another difficulty for our problems, that is, how the gadgets communicate with each other. We handle this difficulty by introducing the “neutral orientation” to NCL. In addition, our and/or gadgets are very complicated, and hence for showing the correctness of our reductions, we clarify the sufficient conditions (which we call “internally connected” and “external adjacency”) so that the gadgets correctly work. We show that our gadgets indeed satisfy these conditions by a computer search.
2 Nondeterministic Constraint Logic
In this section, we define the nondeterministic constraint logic problem [8].
An NCL “machine” is an undirected graph together with an assignment of weights from to each edge of the graph. An (NCL) configuration of this machine is an orientation (direction) of the edges such that the sum of weights of incoming arcs at each vertex is at least two. Figure 2(a) illustrates a configuration of an NCL machine, where each weight2 edge is depicted by a thick (blue) line and each weight1 edge by a thin (orange) line. Then, two NCL configurations are adjacent if they differ in a single edge direction. Given an NCL machine and its two configurations, it is known to be PSPACEcomplete to determine whether there exists a sequence of adjacent NCL configurations which transforms one into the other [8].
An NCL machine is called an and/or constraint graph if it consists only of two types of vertices, called “NCL and vertices” and “NCL or vertices” defined as follows:

A vertex of degree three is called an NCL and vertex if its three incident edges have weights , and . (See Figure 2(b).) An NCL and vertex behaves as a logical and, in the following sense: the weight edge can be directed outward for if and only if both two weight edges are directed inward for . Note that, however, the weight edge is not necessarily directed outward even when both weight edges are directed inward.

A vertex of degree three is called an NCL or vertex if its three incident edges have weights , and . (See Figure 2(c).) An NCL or vertex behaves as a logical or: one of the three edges can be directed outward for if and only if at least one of the other two edges is directed inward for .
It should be noted that, although it is natural to think of NCL and/or vertices as having inputs and outputs, there is nothing enforcing this interpretation; especially for NCL or vertices, the choice of input and output is entirely arbitrary because an NCL or vertex is symmetric.
3 Our Results
In this section, we give our results. Observe that list edgecoloring reconfiguration can be solved in (most conveniently, nondeterministic [15]) polynomial space, and hence it is in PSPACE. Therefore, we show the PSPACEhardness. Our reductions from NCL take the same strategy, and hence we first give the common preparation in Section 3.1. Then, we prove the PSPACEhardness of list edgecoloring reconfiguration in Section 3.2 and the nonlist variant in Section 3.3.
3.1 Preparation for reductions
Suppose that we are given an instance of NCL, that is, an NCL machine and two orientations of the machine.
We first subdivide every NCL edge into a path of length three by adding two new vertices and . (See Figure 3(a) and (b).) We call the edge a link edge between two NCL vertices and , and call the edges and connector edges for and , respectively. Notice that every vertex in the resulting graph belongs to exactly one of stars such that the center of each corresponds to an NCL and/or vertex. Furthermore, these stars are all mutually disjoint, and joined together by link edges. (See Figure 3(c) as an example.)
Our reduction thus involves constructing three types of gadgets which correspond to link edges and stars of NCL and/or vertices; we will replace each of link edges and stars of NCL and/or vertices with its corresponding gadget. In our reduction, assigning the color to the connector edge always corresponds to directing the connector edge from to (i.e., the inward direction for ), while assigning the color to always corresponds to directing from to (i.e., the outward direction for ).
3.2 List edgecoloring reconfiguration
In this subsection, we prove the following theorem.
Theorem 3.1
For every integer , the list edgecoloring reconfiguration problem is PSPACEcomplete even if an input graph is planar, of maximum degree three, and has bounded bandwidth.
We prove the theorem in the remainder of this subsection. As we have mentioned in Section 3.1, it suffices to construct three types of gadgets which correspond to link edges and stars of NCL and/or vertices.
Link edge gadget.
Recall that, in a given NCL machine, two NCL vertices and are joined by a single NCL edge . Therefore, the link edge gadget between and should be consistent with the orientations of the NCL edge , as follows (see also Figure 4): If we assign the color to the connector edge (i.e., the inward direction for ), then must be colored with (i.e., the outward direction for ); conversely, must be colored with if we assign to . In particular, the gadget must forbid a list edgecoloring which assigns to both and (i.e., the inward directions for both and ), because such a coloring corresponds to the direction which illegally contributes to both and at the same time. On the other hand, assigning to both and (i.e., the outward directions for both and ) corresponds to the neutral orientation of the NCL edge which contributes to neither nor , and hence we simply do not care such an orientation.
Figure 5 illustrates our link edge gadget between two NCL vertices and . Figure 6(b) illustrates the “reconfiguration graph” of this link edge gadget together with two connector edges and : each rectangle represents a node of the reconfiguration graph, that is, a list edgecoloring of the gadget, where the underlined bold number represents the color assigned to the edge, and two rectangles are joined by an edge in the reconfiguration graph if their corresponding list edgecolorings are adjacent. Then, the reconfiguration graph is connected as illustrated in Figure 6(b), and the link edge gadget has no list edgecoloring which assigns to the two connector edges and at the same time, as required. Furthermore, the reversal of the NCL edge can be simulated by the path via the neutral orientation of , as illustrated in Figure 6(a). Thus, this link edge gadget works correctly.
And gadget.
Consider an NCL and vertex . Figure 7(a) illustrates all valid orientations of the three connector edges for ; each box represents a valid orientation of the three connector edges for , and two boxes are joined by an edge if their orientations are adjacent. We construct our and gadget so that it correctly simulates this reconfiguration graph in Figure 7(a).
Figure 8 illustrates our and gadget for each NCL and vertex , where , and correspond to the three connector edges for such that and come from the two weight NCL edges and comes from the weight NCL edge. Figure 7(b) illustrates the reconfiguration graph for all list edgecolorings of the and gadget, where each large box surrounds all colorings having the same color assignments to the three connector edges for . Then, we can see that these list edgecolorings are “internally connected,” that is, any two list edgecolorings in the same box are reconfigurable with each other without recoloring any connector edge. Furthermore, this gadget preserves the “external adjacency” in the following sense: if we contract the list edgecolorings in the same box in Figure 7(b) into a single vertex, then the resulting graph is exactly the graph depicted in Figure 7(a). Therefore, we can conclude that our and gadget correctly works as an NCL and vertex.
Or gadget.
Figure 9 illustrates our or gadget for each NCL or vertex , where , and correspond to the three connector edges for . To verify that this or gadget correctly simulates an NCL or vertex, it suffices to show that this gadget satisfies both the internal connectedness and the external adjacency. However, since this gadget has list edgecolorings, we have checked that our or gadget satisfies these sufficient conditions by a computer search of all list edgecolorings of the gadget.
Reduction.
As we have explained before, we replace each of link edges and stars of NCL and/or vertices with its corresponding gadget; let be the resulting graph. Since NCL remains PSPACEcomplete even if an input NCL machine is planar, bounded bandwidth and of maximum degree three, the resulting graph is also planar, bounded bandwidth and of maximum degree three; notice that, since each gadget consists of only a constant number of edges, the bandwidth of is also bounded.
In addition, we construct two list edgecolorings of which correspond to two given NCL configurations and of the NCL machine. Note that there are (in general, exponentially) many list edgecolorings which correspond to the same NCL configuration. However, by the construction of the three gadgets, no two distinct NCL configurations correspond to the same list edgecoloring of . We thus arbitrarily choose two list edgecolorings and of which correspond to and , respectively.
Clearly, the construction can be done in polynomial time.
Correctness.
We now prove that there exists a desired sequence of NCL configurations between and if and only if there exists a reconfiguration sequence between and .
We first prove the onlyif direction. Suppose that there exists a desired sequence of NCL configurations between and , and consider any two adjacent NCL configurations and in the sequence. Then, only one NCL edge changes its orientation between and . Notice that, since both and are valid NCL configurations, the NCL and/or vertices and have enough incoming arcs even without . Therefore, we can simulate this reversal by the reconfiguration sequence of list edgecolorings in Figure 6(b) which passes through the neutral orientation of as illustrated in Figure 6(a). Recall that both and and or gadgets are internally connected, and preserve the external adjacency. Therefore, any reversal of an NCL edge can be simulated by a reconfiguration sequence of list edgecolorings of , and hence there exists a reconfiguration sequence between and .
We now prove the if direction. Suppose that there exists a reconfiguration sequence from to . Notice that, by the construction of gadgets, any list edgecoloring of corresponds to a valid NCL configuration such that some NCL edges may take the neutral orientation. In addition, and correspond to valid NCL configurations without any neutral orientation. Pick the first index in the reconfiguration sequence which corresponds to changing the direction of an NCL edge to the neutral orientation. Then, since the neutral orientation contributes to neither nor , we can simply ignore the change of the NCL edge and keep the direction of as the same as the previous direction. By repeating this process and deleting redundant orientations if needed, we can obtain a sequence of valid adjacent orientations between and such that no NCL edge takes the neutral orientation.
This completes the proof of Theorem 3.1.
3.3 Edgecoloring reconfiguration
In this subsection, we prove the following theorem for the nonlist variant.
Theorem 3.2
For every integer , the edgecoloring reconfiguration problem is PSPACEcomplete even if an input graph is planar whose maximum degree is and bandwidth is linear in .
To prove the theorem, similarly as in the previous subsection, it suffices to construct three types of gadgets corresponding to a link edge and stars of NCL and/or vertices. However, since we deal with the nonlist variant, every edge has all colors as its available colors. Thus, we construct one more gadget, called a color gadget, which restricts the colors available for the edge. The gadget is simply a star having leaves, and we assign the colors to the edges of the star in both and . (See Figure 10(a) as an example for .) Note that, since the color set consists of only colors, these edges must stay the same colors in any reconfiguration sequence. Thus, if we do not want to assign a color to an edge , then we connect the leaf with the color to an endpoint of . (See Figure 10(b).) In this way, we can treat the edge as if it has the list of available colors. However, we need to pay attention to the fact that all edges sharing the endpoint cannot receive the color by connecting such gadgets to .
Figures 11 and 12 illustrate all gadgets for the nonlist variant, where the available colors for each edge is attached as the list of the edge. Notice that the gadget in Figure 11 forbids the colors , , and . We emphasize that all (red) endpoints of connector edges shared by other gadgets are attached with the same color gadgets which forbid two colors and , and hence we can connect the gadgets consistently. Then, the link edge gadget and and/or gadgets have , , edgecolorings, respectively. We have checked that all gadgets satisfy both the internal connectedness and the external adjacency by a computer search of all edgecolorings of the gadgets. Therefore, by the same arguments as the proof of Theorem 3.1, we can conclude that an instance of NCL is a instance if and only if the corresponding instance of edgecoloring reconfiguration is a instance.
Recall that NCL remains PSPACEcomplete even if an input NCL machine is planar, bounded bandwidth, and of maximum degree three. Thus, the resulting graph is also planar. Notice that only the size of the color gadget depends on , and the other (parts of) gadgets are of constant sizes. Since , the maximum degree of is , the degree of the center of each color gadget. In addition, since each of link edge and and/or gadgets contains only a constant number of color gadgets, the number of edges in each gadget can be bounded by a linear in . Since the bandwidth of the input NCL machine is a constant, that of can be bounded by a linear in .
This completes the proof of Theorem 3.2.
4 Conclusion
In this paper, we have shown the PSPACEcompleteness of list edgecoloring reconfiguration and its nonlist variant. We emphasize again that our result for list edgecoloring reconfiguration gives a sharp analysis of the complexity status with respect to the number of colors. In addition, our result is the first complexity hardness result for the nonlist variant.
References
 [1] Bonamy, M., Bousquet, N.: Recoloring bounded treewidth graphs. Electronic Notes in Discrete Mathematics 44, pp. 257–262 (2013)
 [2] Bonamy, M., Johnson, M., Lignos, I., Patel, V., Paulusma, D.: Reconfiguration graphs for vertex colourings of chordal and chordal bipartite graphs. J. Combinatorial Optimization 27, pp. 132–143 (2014)
 [3] Bonsma, P., Cereceda, L.: Finding paths between graph colourings: PSPACEcompleteness and superpolynomial distances. Theoretical Computer Science 410, pp. 5215–5226 (2009)
 [4] Bonsma, P., Mouawad, A.E., Nishimura, N., Raman, V.: The complexity of bounded length graph recoloring and CSP reconfiguration. Proc. of IPEC 2014, LNCS 8894, pp. 110–121 (2014)
 [5] Cereceda, L., van den Heuvel, J., Johnson, M.: Finding paths between colorings. J. Graph Theory 67, pp. 69–82 (2011)
 [6] Demaine, E.D., Demaine, M.L., FoxEpstein, E., Hoang, D.A., Ito, T., Ono, H., Otachi, Y., Uehara, R., Yamada, T.: Lineartime algorithm for sliding tokens on trees. Theoretical Computer Science 600, 132–142 (2015)
 [7] Hatanaka, T., Ito, T., Zhou, X.: The list coloring reconfiguration problem for bounded pathwidth graphs. IEICE Trans. on Fundamentals of Electronics, Communications and Computer Sciences E98A, 1168–1178 (2015)
 [8] Hearn, R.A., Demaine, E.D.: PSPACEcompleteness of slidingblock puzzles and other problems through the nondeterministic constraint logic model of computation. Theoretical Computer Science 343, pp. 72–96 (2005)
 [9] van den Heuvel, J.: The complexity of change. Surveys in Combinatorics 2013, London Mathematical Society Lecture Notes Series 409 (2013)
 [10] Ito, T., Demaine, E.D., Harvey, N.J.A., Papadimitriou, C.H., Sideri, M., Uehara, R., Uno, Y.: On the complexity of reconfiguration problems. Theoretical Computer Science 412, pp. 1054–1065 (2011)
 [11] Ito, T., Kamiński, M., Demaine, E.D.: Reconfiguration of list edgecolorings in a graph. Discrete Applied Mathematics 160(15), pp. 2199–2207 (2012)
 [12] Ito, T., Kawamura, K., Zhou, X.: An improved sufficient condition for reconfiguration of list edgecolorings in a tree. IEICE Trans. on Information and Systems E95D(3), pp. 737–745 (2012)
 [13] Ito, T., Ono, H., Otachi, Y.: Reconfiguration of cliques in a graph. Proc. of TAMC 2015, LNCS 9076, pp. 212–223 (2015)
 [14] Johnson, M., Kratsch, D., Kratsch, S., Patel, V., Paulusma, D.: Finding shortest paths between graph colourings. To appear in Algorithmica. DOI: 0.1007/s0045301500097
 [15] Savitch, W.J.: Relationships between nondeterministic and deterministic tape complexities. J. Computer and System Sciences 4, pp, 177–192 (1970)
 [16] Wrochna, M.: Reconfiguration in bounded bandwidth and treedepth. arXiv:1405.0847 (2014)
 [17] van der Zanden, T.C.: Parameterized complexity of graph constraint logic. Proc. of IPEC 2015, LIPIcs 9076, pp. 282–293 (2015)