<?xml-model href='http://www.tei-c.org/release/xml/tei/custom/schema/relaxng/tei_all.rng' schematypens='http://relaxng.org/ns/structure/1.0'?><TEI xmlns="http://www.tei-c.org/ns/1.0">
	<teiHeader>
		<fileDesc>
			<titleStmt><title level='a'>On Torsor Structures on Spanning Trees</title></titleStmt>
			<publicationStmt>
				<publisher>siam.org</publisher>
				<date>09/30/2023</date>
			</publicationStmt>
			<sourceDesc>
				<bibl> 
					<idno type="par_id">10517279</idno>
					<idno type="doi">10.1137/22M149154X</idno>
					<title level='j'>SIAM Journal on Discrete Mathematics</title>
<idno>0895-4801</idno>
<biblScope unit="volume">37</biblScope>
<biblScope unit="issue">3</biblScope>					

					<author>Farbod Shokrieh</author><author>Cameron Wright</author>
				</bibl>
			</sourceDesc>
		</fileDesc>
		<profileDesc>
			<abstract><ab><![CDATA[\mathrm{ \mathrm{ \mathrm{\mathrm{ \mathrm{ . \mathrm{\mathrm{ \mathrm{ \mathrm{\mathrm{\mathrm{\mathrm{\mathrm{ \mathrm{\mathrm{ \mathrm{\mathrm{.© 2023 \mathrm{ \mathrm{\mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{\mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{\mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{\mathrm{ \mathrm{ \mathrm{ \mathrm{ \mathrm{]]></ab></abstract>
		</profileDesc>
	</teiHeader>
	<text><body xmlns="http://www.tei-c.org/ns/1.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xlink="http://www.w3.org/1999/xlink">
<div xmlns="http://www.tei-c.org/ns/1.0"><p>Conjecture 1.1 (Baker and Wang <ref type="bibr">[3]</ref>). Given a nonplanar ribbon graph G with no multiple edges and no loops, there exists a vertex q for which \rho q \not = \beta q .</p><p>We resolve this conjecture and extend the result to a class of graphs having multiple edges but no loops. In particular, it is proven here that any nonplanar graph with or without multiple edges admits a vertex q for which \rho q \not = \beta q when endowed with any ribbon structure. The criterion for the presence of such a vertex q in a nonplanar ribbon graph with multiple edges is the presence of what we call a proper witness pair. In Theorem 6.2, we prove that the presence of a proper witness pair in a nonplanar ribbon graph is sufficient to ensure the presence of a vertex q for which \rho q \not = \beta q . Further, we show in Proposition 3.3 that any nonplanar graph, when endowed with any ribbon structure, must contain a proper witness pair. In particular, any nonplanar graph endowed with any ribbon structure contains a vertex at which the torsor structures disagree. We prove Theorem 7.1, the Baker--Wang conjecture for loopless nonplanar ribbon graphs with no multiple edges admitting no proper witness pair. Finally, we consider two examples and some associated computations; in particular, we exhibit a nonplanar ribbon graph having distinct vertices p and q such that \rho p = \beta p while \rho q \not = \beta q .</p><p>In an independent work, Changxin Ding <ref type="bibr">[7]</ref> also resolves this conjecture. As in the current paper, Ding considers multiple cases, but the approach taken there utilizes decompositions of graphs to facilitate study of the difference between the two torsors. In the present document, we work instead with classes of graphs depending on the presence or absence of certain subgraphs, referred to here as witness pairs.</p><p>2. Background and terminology. By a graph, we mean a finite connected multigraph with no loops; for a graph G, we denote by V (G) its vertex set and by E(G) its collection of edges, given as unordered pairs of vertices. Occasionally, we use V and E to denote the vertices and edges of a graph G when the graph is unambiguous. Unless otherwise specified, multiple parallel edges are allowed. A ribbon graph is a finite graph together with a cyclic ordering of the edges around each vertex, referred to as a ribbon structure. Note that a ribbon structure on a graph induces a natural ribbon structure on any graph minor. Given two edges e 0 , e 1 incident to a given vertex v, we will write e 0 \prec e 1 if e 1 immediately succeeds e 0 in the order around v. If a vertex v has two edges e 0 and e 1 , the interval [e 0 , e 1 ] between e 0 and e 1 is the collection of edges \{ e 0 , f 1 , . . . , f k , e 1 \} , where e 0 \prec f 1 \prec \cdot \cdot \cdot \prec f k \prec e 1 .</p><p>Analogously, we define the open interval (e 0 , e 1 ) = \{ f 2 , . . . , f k \} . The length of an interval is its cardinality as a set.</p><p>A finite ribbon graph is equivalent to an embedding of the underlying graph into a closed oriented surface such that the ribbon structure agrees with the orientation of the surface (see, e.g., <ref type="bibr">[15,</ref><ref type="bibr">Theorem 3.7]</ref>). In all figures in this paper, the ribbon structure is assumed to be given by the counterclockwise orientation of the underlying surface.</p><p>A path P in a ribbon graph G is the image of a mapping P k \rightar G of the length-k path into G that is injective on edges but not necessarily on vertices. A cycle in G is the image of a mapping C k \rightar G of the length-k cycle into G that is injective on both edges and vertices. We sometimes refer to paths and cycles as sets of edges rather than as subgraphs. Let C be a cycle, and endow C with an orientation. A vertex v of C is incident to precisely two edges of C, e in and e out , where the labeling of these edges agrees with the orientation of C. An edge e incident to v is said to be to the left of C if e \in (e out , e in ) and to the right of C if e \in (e in , e out ).</p><p>Given a group \Gamma , we consider a group action of \Gamma on a set X as a function \Gamma \times X -\rightar X .</p><p>We say that a group action is regular if for every \gamma \in \Gamma and every x \in X, the restricted mappings \{ \gamma \} \times X \rightar X and \Gamma \times \{ x\} \rightar X are isomorphisms of sets. Stated differently, an action of \Gamma on X is regular if \Gamma acts simply and transitively on X. If X = \{ 1, . . . , N \} , we can consider a group action as a homomorphism from \Gamma to the symmetric group on N symbols \Gamma \rightar S N . We therefore sometimes work with the image of such a mapping, representing group elements as permutations. All multiplication of permutations will be from right to left, so that, for example, we have (123)(34) = (1234).</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="3.">Cycles and witnesses.</head><p>Here we are concerned in particular with nonplanar graphs. Following <ref type="bibr">[5]</ref>, we discuss planarity and nonplanarity of ribbon graphs in terms of the absence or presence of nonseparating cycles. Definition 3.1. A cycle C is nonseparating if for any orientation of C, there exists a path P that intersects C only in the endpoints of P such that the first edge of P is on the left of C and the last edge of P is on the right of C. We will call P a witness for the nonseparating cycle C and refer to the ordered pair (C, P ) as a witness pair. Definition 3.2. Given a nonseparating cycle C, we will say that a witness P for C is a proper witness if the endpoints of P are distinct vertices of C. In this case, we call the pair (C, P ) a proper witness pair (see Figure <ref type="figure">3</ref>.1).</p><p>We say that a connected ribbon graph G is nonplanar if and only if it contains a nonseparating cycle. Note that this definition of nonplanarity is not equivalent to the underlying graph G being nonplanar in the typical sense, but all nonplanar graphs yield nonplanar ribbon graphs when endowed with any ribbon structure. A planar graph G can be endowed with a ribbon structure such that the resulting ribbon graph is nonplanar.</p><p>We have the following structural property of nonplanar graphs endowed with a ribbon structure. Proposition 3.3. A connected nonplanar graph G endowed with any ribbon structure admits a proper witness pair.</p><p>Proof. Starting from an arbitrary ribbon graph G, first note that the operations of edge deletion and edge contraction cannot create a proper witness in G. Thus, by   Kuratowski's theorem, it is enough to illustrate that K 5 and K 3,3 necessarily admit a proper witness pair when endowed with any ribbon structure.</p><p>Given any ribbon structure on K 5 , there necessarily exists a nonseparating cycle C; if some witness path P is of length 1, then the fact that K 5 is loopless ensures that P must be proper since the endpoints of P in C are distinct. Further, if P is of length 2, then the fact that K 5 has no multiple edges ensures that P must be proper. Now, C either includes all vertices or excludes at least one vertex. In the case that C includes all vertices, any witness path must be of length 1 and so must be proper.</p><p>Suppose then that C excludes at least one vertex and that all witness paths have length greater than 1. Then, since C is necessarily of length 3 or 4, any witness path has length either 2 or 3. Let P be some such witness path. If P is proper, we are done. If P is improper, then P has length 3, and we can write P = \{ \{ a, b\} , \{ b, c\} , \{ c, a\} \} and C = \{ \{ a, d\} , \{ d, e\} , \{ e, a\} \} . Suppose without loss of generality that \{ a, b\} is on the left of C. Then either the edges \{ d, b\} , \{ e, b\} , \{ d, c\} , and \{ e, c\} are on the left of C or one is on the right. In the former case, we have a proper witness P \prime = \{ \{ d, b\} , \{ b, c\} , \{ c, a\} \} , and in the latter, we have some proper witness P \prime by replacing either \{ c, a\} or \{ b, c\} , \{ c, a\} with the edge on the right of C (see Figure <ref type="figure">3</ref>.2).</p><p>Consider now K 3,3 . Since K 3,3 is nonplanar, any ribbon structure admits a nonseparating cycle C. In the case that C contains all vertices, the fact that K 3,3 contains no loops ensures that any witness path P is a proper witness. Suppose C excludes at least one vertex. Since all cycles in bipartite graphs are of even length, any witness path P must have length 2. But since there are no multiple edges in K 3,3 , the endpoints of P in C must be distinct; hence, P is proper (see Figure <ref type="figure">3</ref>.3).</p><p>We next prove a similar result describing ribbon graphs admitting K 4 as a minor in such a way that the inherited ribbon structure is nonplanar. Proposition 3.4. Suppose that G is a graph containing the complete graph K 4 as a minor. If G is endowed with a ribbon structure making the K 4 minor with the induced ribbon structure nonplanar, then G admits a proper witness pair.</p><p>Proof. As in the proof of Proposition 3.3, we begin with the observation that the presence of a nonseparating cycle with proper witness in the graph K 4 is sufficient to guarantee the presence of such a cycle with proper witness in any graph G containing K 4 as a minor. It is therefore sufficient to show that each nonplanar ribbon structure on K 4 admits a proper witness pair.</p><p>Since each vertex in K 4 is of degree 3, there are only two choices of cyclic ordering of the edges around each vertex (see Figure <ref type="figure">3</ref>.4). In particular, we can fix a planar embedding of K 4 and enumerate all ribbon structures by choosing either a clockwise or a counterclockwise orientation of edges about each vertex.</p><p>Moreover, by symmetry, we only need to consider the number of vertices oriented in either direction rather than their relative positions. We refer to the ribbon structure in which i vertices are oriented clockwise and j are oriented counterclockwise as being of type (i, j). By reversing the orientation of the plane, it is clear that the presence of a proper witness pair in the ribbon structure of type (i, j) ensures the presence of such a pair in the structure of type (j, i). Now, since the ribbon structures of type (4, 0) and (0, 4) give rise to planar ribbon graph structures, we have only two choices of ribbon structures yielding a nonplanar ribbon graph K 4 , arising, respectively, from the pairs in types (3, 1) and (2, 2). For each of these two ribbon structures, one can directly find the associated surface of minimal genus and verify that there exists a proper witness (see Figure <ref type="figure">3</ref>.5).</p><p>In the case of a nonplanar ribbon graph which does not admit a proper witness pair, there is another kind of witness pair which will be of use to us. Definition 3.5. Let G be a nonplanar ribbon graph that does not admit a proper witness pair. Suppose that G admits a witness pair (C, P ) in G such that C and P share a unique vertex z. We say that (C, P ) is a tight witness pair if the edges e and f incident to z in C are such that the interval [e, f ] in the ribbon structure about z has minimal length among all cycles C passing through z that have P as a witness path.</p><p>The tightness of a witness pair (C, P ) ensures that there are relatively few edges incident to z on one side of C. More precisely, this property means that for an oriented cycle C, the number of edges containing z on one side of C is as small possible. This definition is used in particular in the proof of Theorem 7.</p><p>Every finite nonplanar ribbon graph having no proper witness pair has at least one tight witness pair. This follows from the fact that G is a finite graph, and so we are taking a minimum over a finite collection of witness pairs. 4. Divisors on graphs. In this section, we review the basic notions of the divisor theory of graphs, providing a setting in which to discuss the Picard group of a graph as the group of chip-firing equivalence classes. Alternative presentations of this material can be found in Chapter 4 of <ref type="bibr">[12]</ref> or in Chapter 2 of <ref type="bibr">[6]</ref>. Given a graph G, let Div(G) denote the free abelian group with formal generators corresponding to the vertices of G. Elements of Div(G) are called divisors on G; a given divisor D has the form</p><p>where a v \in Z for all v \in V and (v) is the formal generator corresponding to v. It is often convenient to think of a divisor as an arrangement of chips placed at the vertices of the graph G. The degree of a divisor D is the sum \sum v\in V a v . The collection of degree d divisors is denoted by Div d (G). Note that, given any vertex q of G, the group Div 0 (G) is generated by the divisors \{ (v) -(q) : v \in V \} .</p><p>Denote by \scrM (G) the collection of functions f : V \rightar Z. The combinatorial Laplacian on G is the mapping \Delta : \scrM (G) \rightar Div 0 (G) given by</p><p>The image of \Delta is the set of principal divisors on G; such divisors form a subgroup Prin(G) of Div 0 (G). We say that two divisors D and D \prime are linearly equivalent if they differ by a principal divisor, so that D \prime = D + \Delta f for some f \in \scrM (G). The (degree 0) Picard group of G is the quotient group</p><p>In general, we write Pic d (G) = Div d (G)/ Prin(G); this is the collection of linear equivalence classes of degree-d divisors. Pic 0 (G) admits a regular action on Pic d (G) for any d given by addition of divisor classes. It is known that the cardinality of the Picard group of G is equal to the number of spanning trees of the graph G (see <ref type="bibr">[2]</ref> and references therein).</p><p>For a graph G, the genus of G is the first Betti number</p><p>any spanning tree of G has exactly | E(G)| -g edges. Given a spanning tree T \in \scrT (G) excluding edges e 1 , . . . e g , a divisor B \in Div g (G) is referred to as a break divisor for T if B is of the form B = \sum g i=1 s(e i ), where s(e i ) is one vertex of the edge e i . Note that there are many possible distinct break divisors associated to a given tree. We have the following property of break divisors.</p><p>Proposition 4.1 (Theorem 1.1 of <ref type="bibr">[1]</ref>). Each divisor class of Pic g (G) contains a unique break divisor.</p><p>Consequently, we can use break divisors as canonical representatives of the equivalence classes in Pic g (G).</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="5.">Bernardi and rotor-routing torsors.</head><p>In addition to having the same cardinality as \scrT (G), the Picard group Pic 0 (G) also admits at least two regular group actions, or torsor structures, on \scrT (G). One is derived from a process due to Bernardi, and the other arises from a process referred to as rotor-routing <ref type="bibr">[4]</ref>, <ref type="bibr">[10]</ref>. We deal with torsor structures that depend on a choice of base vertex in the graph G. We remark that K\' alm\' an, Lee, and T\' othm\' er\' esz have recently shown in <ref type="bibr">[11]</ref> that, using the machinery of planar trinities, it is possible to give a canonical definition of the Bernardi action on planar graphs without an initial choice of a base vertex. Moreover, this work extends these results to the situation of balanced plane digraphs.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="5.1.">Bernardi torsor.</head><p>The first torsor structure considered in this work arises from a family of combinatorial maps, collectively referred to as the Bernardi process. As mentioned above, Pic 0 (G) is in bijective correspondence with the set of spanning trees \scrT (G). The works of Bernardi and Baker and Wang establish a family of bijections \beta (q,e) : \scrT (G) \rightar Pic g (G) witnessing this correspondence by mapping spanning trees to break divisors <ref type="bibr">[4]</ref>, <ref type="bibr">[3]</ref>. These maps are parametrized by pairs (q, e) consisting of a vertex q and an edge e incident to q.</p><p>The Bernardi process uses the data (q, e) and a spanning tree T to perform a tour \tau (q,e) (T ) of the graph G. This tour can be thought of as a special walk in G beginning and ending at q, traversing each edge of T twice, and excluding each edge outside T . More specifically, the tour \tau (q,e) (T ) takes the form \tau (q,e) (T ) = (v 0 , e 1 , v 1 , . . . , e k , v k ), where v 0 = v k = q and e = e 1 = \{ v 0 , v 1 \} . The process of deriving the break divisor \beta (q,e) (T ) from T is illustrated graphically in Figure <ref type="figure">5</ref>.1. Since, by Proposition 4.1, each divisor class of Pic g (G) contains a unique break divisor, it follows that the maps \beta (q,e) induce explicit combinatorial bijections between spanning trees and divisor classes of degree g.</p><p>From these bijections, Baker and Wang define a group action \beta q : Pic 0 (G) \times \scrT (G) \rightar \scrT (G) using the natural action of Pic 0 (G) on Pic g (G) <ref type="bibr">[3]</ref>. Fix initial data (q, e). Given any divisor class [D] \in Pic 0 (G), define</p><p>) . A central result of the paper of Baker and Wang states that the action (5.1) is in fact independent of the chosen edge e used in defining the bijection \beta (q,e) (see <ref type="bibr">[3,</ref><ref type="bibr">Theorem 4.1]</ref>). As a result of this, one can define a group action of Pic 0 (G) on \scrT (G) after fixing only a vertex of G, the base vertex of the action.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="5.2.">Rotor-routing torsor.</head><p>Another regular action is derived from the rotorrouting process on G, introduced in <ref type="bibr">[14]</ref> and studied further in <ref type="bibr">[10]</ref> and <ref type="bibr">[9]</ref>. The rotor-routing process is a discrete-time dynamical system derived from G. The state space consists of rotor configurations on G: pairs (\sigma , v), where v is a vertex of G, interpreted as the location of a chip, and \sigma : V (G) \rightar E(G), a rotor function assigning to each vertex w of G an edge \sigma (w) containing w. The edge \sigma (w) is referred to as the rotor at w.</p><p>A single step of the rotor-routing process takes the rotor configuration (\sigma , v) to (\sigma \prime , w), where \sigma \prime is the rotor configuration for which \sigma \prime (u) = \sigma (u) for all u \not = v, \sigma \prime (v) is the successor of \sigma (v) in the ribbon structure about v, and w is the other vertex of \sigma \prime (v). Informally, we move the rotor at v once according to the ribbon structure about v and move the chip along the new edge to its other vertex w.</p><p>For the purposes of defining the action of Pic 0 (G) on \scrT (G), we choose a vertex q of G as a sink of the system and ignore the rotor at q. Given a spanning tree T of G and a vertex q of G, we can obtain a rotor function \sigma T such that \sigma T (v) is the first edge on the unique path in T from v to q. Conversely, given a rotor function \sigma , we can obtain a subgraph H \sigma of G whose only edges are the rotors of \sigma .</p><p>Given any vertex q of G and any spanning tree T of G, iterating the rotor-routing process from any configuration (\sigma T , v) will eventually yield a configuration (\tau , q); the chip will necessarily reach the vertex q [9, <ref type="bibr">Lemma 3.6</ref>]. Moreover, it can be shown that the graph H \tau derived from \tau will be a spanning tree of G [9, <ref type="bibr">Lemma 3.10]</ref>.</p><p>Using this observation, we can construct a group action \rho q of Div 0 (G) on \scrT (G) by defining the action of generators \{ (v) -(q) : v \in V (G)\} and extending linearly. Rotor-routing process applied to a spanning tree of K 4 . The original spanning tree T , indicated at the left by thickened edges, is mapped to the rightmost spanning tree \rho q([(v) -(q)])(T ). At each stage, the location of the chip is indicated by a filled vertex.</p><p>Beginning from the rotor configuration (\sigma T , v), iterate the process until reaching some (\tau , q), and let T \prime be the spanning tree given by the final rotor function. This action is trivial on any divisor in Prin(G) and so descends to an action by Pic 0 (G). The rotor-routing action \rho q : Pic 0 (G) \times \scrT (G) \rightar \scrT (G) is then given by setting</p><p>See Figure <ref type="figure">5</ref>.2 for an illustration of the rotor-routing action. In the investigation of <ref type="bibr">[5]</ref>, the following definition is of crucial use to establish the main findings.</p><p>Definition 5.1. A rotor configuration (\sigma , v) is called a unicycle if the image of the rotor function \sigma contains a unique directed cycle and v is a vertex of this cycle or if v has some neighbor w such that \sigma (v) = \sigma (w). This definition will be of particular use to us due to the interaction between the rotor-routing action and unicycles. Lemma 5.2 (Lemma 4.9 of <ref type="bibr">[10]</ref>). Let (\sigma , v) be a unicycle on a graph with m edges. Then, in iterating the rotor-routing process 2m times from (\sigma , v), the chip traverses each edge of G exactly once in each direction, each rotor makes exactly one full rotation, and the final state is again (\sigma , v).</p><p>The following lemma is an easy consequence of Lemma 5.2. Lemma 5.3. Suppose that G is a ribbon graph, (\sigma , w) a unicycle on G, and z a vertex satisfying \sigma (z) = \{ z, w\} . Then, for all neighbors v \not = w of z satisfying \sigma (v) = \{ v, z\} , the rotor at z reaches \{ z, v\} before the rotor at v completes a full rotation.</p><p>Proof. Let v be a neighbor of z such that \sigma (v) = \{ v, z\} . Suppose for the sake of contradiction that in iterating the rotor-routing process from (\sigma , z), the rotor at v completes a full rotation before the rotor at z reaches \{ z, v\} . Then, in order for the rotor at z to complete a full rotation, we pass through some rotor configuration (\rho , v) such that \rho (z) = \{ z, v\} . Then, iterating the rotor-routing process one additional time, the rotor at v has completed more than one full turn before the rotor at z returns to its starting position, contradicting Lemma 5.2.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="5.3.">Comparison of torsor structures.</head><p>The rotor-routing and Bernardi processes described in the previous subsections give rise to regular group actions of Pic 0 (G) on \scrT (G), yielding two torsor structures on \scrT (G) via (5.1) and (5.2). Each of the two structures is dependent on a choice of some vertex of G as base vertex for the action. These torsor structures have been the subject of much study in recent years <ref type="bibr">[3,</ref><ref type="bibr">5,</ref><ref type="bibr">13,</ref><ref type="bibr">11]</ref>.</p><p>Fix some vertex q \in V (G), and consider \rho q and \beta q . To see that the two homomorphisms \rho q and \beta q agree, we must have that</p><p>for each T \in \scrT (G) for each divisor class [D] \in Pic 0 (G). In particular, since divisors of the form (v) -(q) generate Div 0 (G), by (5.1) and (5.2), it is sufficient to verify that for each v \in V (G), we have the equality</p><p>In contrast, to show that \rho q \not = \beta q we need only find a single tree T \in \scrT for which the above equality does not hold; doing so immediately yields that, as permutations of \scrT (G),</p><p>and therefore \rho q and \beta q cannot be equal.</p><p>We have the following theorem about the rotor-routing torsor structure on \scrT (G).</p><p>Theorem 5.4 (Theorem 2 of <ref type="bibr">[5]</ref>). Let G be a connected ribbon graph. The action \rho q of Pic 0 (G) on \scrT (G) is independent of the base vertex q if and only if G is a planar ribbon graph.</p><p>The work of Baker and Wang established an analogous result for the torsor structure derived from the Bernardi process.</p><p>Theorem 5.5 (Theorems 5.1 and 5.4 of <ref type="bibr">[3]</ref>). Let G be a connected ribbon graph. The action \beta q of Pic 0 (G) on \scrT (G) is independent of the base vertex q if and only if G is a planar ribbon graph.</p><p>In the case of a planar ribbon graph G, neither the rotor-routing action nor the Bernardi action is dependent on the choice of base vertex; both actions are in this sense canonical.</p><p>Theorem 5.6 (Theorem 7.1 of <ref type="bibr">[3]</ref>). For a planar ribbon graph G, the Bernardi and rotor-routing processes define the same Pic 0 (G)-torsor structure on \scrT (G).</p><p>However, in the case of a nonplanar G, the situation was not as clear. The work <ref type="bibr">[3]</ref> concludes with the following conjecture.</p><p>Conjecture 5.7 (Conjecture 7.2 of <ref type="bibr">[3]</ref>). Let G be a nonplanar ribbon graph with no multiple edges and no loops. Then there exists a vertex q of G such that \rho q \not = \beta q .</p><p>6. Baker--Wang conjecture and proper witnesses. We now prove the Baker--Wang conjecture in the case that the graph G admits a proper witness pair. This will be done by first proving the conjecture in a more restricted situation. Lemma 6.1. Suppose that G is a ribbon graph with proper witness pair (C, P ) such that P has endpoints x and z. If the two edges e 0 and e 1 of C incident to z satisfy e 0 \prec e 1 , then there exists a vertex q of G such that \rho q \not = \beta q .</p><p>Proof. Suppose that e 1 = \{ z, q\} , so that q is the other vertex of e 1 in C. Now let e \prime = \{ u, v\} be an edge of P , and extend the collection of edges (C \setminu e 1 ) \cup (P \setminu e \prime ) to a spanning tree T of G. We will show that \rho q ([(z) -(q)], T ) \not = \beta q ([(z) -(q)], T ) . Rotor-routing process applied to the spanning tree T , thickened at the left, as in the proof of Lemma 6.1. After one step, the chip is moved from z to q, and the process is terminated, yielding the thickened tree T \prime at the right.</p><p>In particular, denoting by T \prime the spanning tree \rho q ([(z) -(q)], T ), we show that the divisor classes [(z) -(q)] + [\beta q,e1 (T )] and [\beta q,e1 (T \prime )] have distinct break divisor representatives. By the definition of the Bernardi action \beta q , we immediately obtain \rho q \not = \beta q .</p><p>Acting on T with \rho q ([(z) -(q)]) entails only a single step of the rotor-routing process: changing the rotor at z from e 0 to e 1 and moving the chip to q, terminating the process. As a result, the spanning trees T and T \prime differ only by a single edge. In particular, T \prime = (T \setminu e 0 ) \cup e 1 . This situation is depicted in Figure <ref type="figure">6</ref>.1.</p><p>Consider the divisors (z) - (q)+\beta (q,e1) (T ) and \beta (q,e1) (T \prime ). By the definition of the Bernardi bijection, we know that \beta (q,e1) (T \prime ) is a break divisor for T \prime . On the other hand, the divisor (z) -(q) + \beta (q,e1) (T ) also a break divisor for T \prime , distinct from \beta (q,e1) (T \prime ). Since a break divisor for a spanning tree is a formal sum of endpoints of edges outside that tree, we can see the following: (z) -(q) + \beta (q,e1) (T ) = (z) -(q) + (q) + \sum e / \in T e\not =e1 s(e) = (z) + \sum e / \in T \prime e\not =e0 s(e) .</p><p>Recall from section 4 that s(e) denotes either endpoint of the edge e. Since z is an endpoint of e 0 , this shows that (z) -(q) + \beta (q,e1) (T ) is a break divisor for T \prime .</p><p>We are reduced to showing that (z) -(q) + \beta (q,e1) (T ) and \beta (q,e1) (T \prime ) are distinct break divisors. By the construction of T , the collection of edges P \cap T consists of two connected components P z and P x , containing z and x, respectively. In the Bernardi tour of T , P x is traversed before P z , while the opposite is true for the tour of T \prime . This results in different placement of the chip associated to the edge e \prime in the two tours: Without loss of generality, the T tour places a chip at v, while the T \prime tour places a chip at u; this difference is depicted in Figure <ref type="figure">6</ref>.2.</p><p>It now remains to show that the difference in placement of the chip associated to e \prime yields a difference between (z) - (q)+\beta (q,e1) (T ) and \beta (q,e1) (T \prime ) as divisors. Suppose for the sake of contradiction that they are the same despite their difference in choice of s(e \prime ). Since e \prime is cut at v in the tour of T , the equality of (z) -(q) + \beta (q,e1) (T ) and \beta (q,e1) (T \prime ) at u implies that there must be some other edge f 0 = \{ u, w 1 \} , excluded from both T and T \prime , which is cut at u in the tour of T and at its other endpoint w 1 in the tour of T \prime . Similarly, for (z) -(q) + \beta (q,e1) (T ) and \beta (q,e1) (T \prime ) to agree at w 1 , Fig. <ref type="figure">6</ref>.2. Bernardi process with initial data (q, e 1 ) applied to the spanning trees T and T \prime as in the proof. For simplicity, only the relevant parts of the tours are included in the figure. At the left, the tour cuts e \prime at u, while at the right, e \prime is cut at v. Here, the arcs surrounding vertices indicate that for this cycle, all edges are cut at their heads, in accordance with the described tour of the tree T \prime . The depicted edges outside the cycle are the edges of T \prime incident to u and v.</p><p>we see that there must be some other edge f 1 = \{ w 1 , w 2 \} , excluded from both T and T \prime , which is cut at w 1 in the tour of T and is cut at its other endpoint w 2 in the tour of T \prime .</p><p>Continuing in this fashion, we see there must exist a cycle D in G having vertices u = w 0 , w 1 , . . . , w m = v and edges f 0 , . . . , f m - 1 , e \prime , where f i = \{ w i , w i+1 \} , such that all edges f i are excluded from both trees T and T \prime . Moreover, orienting D such that e \prime is oriented from v to u, we see that in the tour of T , each edge of D must be cut at its tail, while in the tour of T \prime , each edge of D must be cut at its head. This situation is depicted in Figure <ref type="figure">6</ref>.3.</p><p>We show that the existence of such a cycle D with edges being cut in this way is impossible. To see this, we orient the cycle C in a fixed direction and partition the vertices w i of D into two classes L and R according to whether the first edge outside C on the unique path in T from z to w i falls on the left or right of the cycle C. Since T and T \prime are trees, if some vertex w i \in L is visited in the tour of T before some other vertex w j \in R (or vice versa), then all edges outside T incident to w i will be cut before the tour visits w j .</p><p>Consider now the tour of T . Since v is visited before u in this tour, e \prime is cut at its tail, and the edge f m - 1 must be cut before u is visited. Thus, to avoid contradiction, it must be that w m - 1 is in the same partition class as v and that f m - 1 is cut at its tail w m - 1 . Now, applying the same reasoning, it must be that w m - 2 is in the same partition class as w m - 1 and that f m - 2 is cut at its tail w m - 2 . Following this pattern around the cycle D, it must be that w 0 = u is in the same class as w 1 and that the edge f 0 is cut at its tail w 0 = u. But this is impossible since, by construction, u and v cannot be in the same partition class.</p><p>Thus, it cannot be that (z) -(q) + \beta (q,e1) (T ) = \beta (q,e1) (T \prime ). Since the divisors (z) - (q)+\beta (q,e1) (T ) and \beta (q,e1) (T \prime ) are necessarily distinct break divisors, they cannot be representative of the same element of Pic g (G), and so \rho q \not = \beta q .</p><p>For a ribbon graph G admitting a proper witness pair, either we have z, e 0 , and e 1 , as in Lemma 6.1 above, or there are some edges f 1 , . . . , f k incident to z such that e 0 \prec f 1 \prec \cdot \cdot \cdot \prec f k \prec e 1 .</p><p>We now prove that in the latter case, we can still find vertices q \prime and z \prime and an edge e for which the break divisors \beta (q \prime ,e) (T \prime ) and (z \prime ) -(q \prime ) + \beta (q \prime ,e) (T ) still differ as described.</p><p>Theorem 6.2. For any nonplanar ribbon graph G admitting a proper witness pair (C, P ), there exists a vertex q for which \rho q \not = \beta q .</p><p>Proof. Suppose that the proper witness pair (C, P ) is such that the path P intersects C in vertices x and z. If this witness pair satisfies e 0 \prec e 1 in the ribbon structure about z, then we are finished. Suppose also that the pair (C, P ) fails to meet the conditions of Lemma 6.1. In particular, there exists an edge f = \{ z, v\} such that f \in [e 0 , e 1 ] and f \not = e 0 , e 1 . Let G \prime be the subgraph of G obtained by deleting the vertex z and all edges to which it is incident. Denote by F the connected component of the subgraph G \prime containing v. See Figure <ref type="figure">6</ref>.4. Note in particular that F may intersect the cycle C or the path P .</p><p>If the component F meets neither C nor P , then one can ignore the edge f , applying the same reasoning as in Lemma 6.1: The rotor-routing process carried out  in the subgraph F will not influence the rotors on the path P , so the break divisors (z) - (q)+\beta q,e1 (T ) and \beta q,e1 (T \prime ) will still differ as described in the proof. We therefore assume that there is no such edge f with F \cap C and F \cap P empty.</p><p>Suppose that the interval (e 0 , e 1 ) about z contains some positive number of edges f . We illustrate a method of finding a new witness pair (C \prime , P \prime ) and reassigning the roles of e 0 , e 1 , z, and q such that the new interval I serving as the replacement of (e 0 , e 1 ) has strictly fewer edges than before. This means that we can eventually find a witness pair (C \prime \prime , P \prime \prime ) such that the interval between the two edges of C \prime \prime incident to z is empty, and therefore Lemma 6.1 can be applied.</p><p>Suppose that there is some edge in (e 0 , e 1 ) that can be extended to a path from z to a vertex of P . Let f be the maximal such edge, and suppose it is extended to a path from Q z,x \prime through F connecting z to some vertex x \prime \in V (P ). Then, necessarily, there is a path C z,x from z to x in C which does not contain q, and there is a path P x,x \prime from x to x \prime in P . In this case, we replace the nonseparating cycle C by the new cycle C \prime = (C \setminu C z,x \prime ) \cup Q z,x \prime \cup P x,x \prime and the path P by the new path P \prime = P \setminu P x,x \prime . Following this, we repeat this process, replacing e 0 with f and replacing x with x \prime . Doing so, we see that the new interval (f, e 1 ) has cardinality strictly less than that of the original interval (e 0 , e 1 ). This process can be seen in Figure <ref type="figure">6</ref>.5.</p><p>Suppose that there is some edge in (e 0 , e 1 ) that can be extended to a path from z to a vertex of C. Let f be the minimal such edge, and suppose it is extended to a path Q z,c through F connecting z to some vertex c \in V (C). We have the following two cases:</p><p>\bullet If c = x or if c lies on the path from z to x in C, denote by C z,c the path from z to c in C which avoids q. Replace the nonseparating cycle C by the new cycle</p><p>, and leave the path P unchanged. We now let f fill the role of e 0 and see that the interval (f, e 1 ) has cardinality strictly less than that of (e 0 , e 1 ). \bullet If the vertex c lies on the path from z to x in C which passes through q or if c = q, let C z,c denote the path from z to c in C passing through (or stopping at) q. Replace the nonseparating cycle C by the new cycle</p><p>, and leave the path P unchanged. We now let f fill the role of e 1 and see that the interval (e 0 , f ) has cardinality strictly less than that of (e 0 , e 1 ). This process can be seen in Figure <ref type="figure">6</ref>.6. Using the above approaches, it follows that in the presence of a proper witness pair (C, P ) in G, we can always recover a witness pair (C \prime \prime , P \prime \prime ) which either satisfies the conditions of Lemma 6.1 or is such that no edges incident to z inhibit the proof of Lemma 6.1. Thus, we conclude that if G is a nonplanar ribbon graph admitting a proper witness pair, then there is some vertex q of G such that \rho q \not = \beta q .</p><p>Considering Proposition 3.3 alongside Theorem 6.2, we immediately obtain the following corollary. Corollary 6.3. For a nonplanar graph G endowed with any ribbon structure, there exists some vertex q of G such that \rho q \not = \beta q .</p><p>7. Simple graphs without proper witnesses. We next show that for nonplanar ribbon graphs G without multiple edges or loops that do not contain a proper witness pair, there is some vertex q for which \rho q \not = \beta q . This completes the proof of the Baker--Wang conjecture.</p><p>Theorem 7.1. Let G denote a nonplanar ribbon graph with no multiple edges and no loops such that G does not admit a proper witness pair. Then there is some vertex q of G such that \rho q \not = \beta q .</p><p>Proof. As in the proof of Lemma 6.1, we will construct a spanning tree T such that the Bernardi and rotor-routing torsors disagree on T .</p><p>In order to do so, we must first fix some terminology regarding our construction. It is recommended that the reader reference Figure <ref type="figure">7</ref>.1 while we introduce the required terminology.</p><p>1. Since G is a nonplanar ribbon graph with no proper witness pair, there exists at least one tight witness pair (Definition 3.5). Suppose that (C, P ) is one such pair and that z is the unique vertex in the intersection of P and C. 2. Suppose that q is a neighbor of z in C. Orient \{ z, q\} from q to z, and extend this orientation to all of C. We may assume that the interval on the left of C is the minimal interval described in the definition of a tight witness pair. Let e be the edge \{ z, q\} , and let e \prime be the edge immediately succeeding e in the ribbon structure about q. 3. Partition all neighbors y of z outside the cycle C into two classes L and R so that y \in L if and only if \{ y, z\} is on the left of C and y \in R if and only if \{ y, z\} is on the right of C.  In this case, we may form a new cycle C \prime which agrees with C on the (oriented) path from z to c in C but which contains the edges \{ w, z\} an \{ w \prime , c\} . Doing so then yields a witness pair (C \prime , P ) such that the length of the ribbon structure interval about z on the left of C \prime is strictly less than the length of the ribbon structure interval about z on the left of C. This is precisely what the tightness of the witness pair (C, P ) prohibits, and so such a situation is impossible. Consider the rotor-routing process taking T to T \prime := \rho q ([(z) -(q)], T ). At some point, the rotor at z shifts to the edge \{ z, w 0 \} , forming a unique cycle in the subgraph of H defined by the rotors at vertices of H. Restricting the rotor configuration to the graph H, we obtain a unicycle U = (\sigma , z) in H. Further, by the assumption that G has no proper witness, we see that all paths from any vertex of H other than z to any vertex on the path C must pass through z. In particular, combining this with the fact that (C, P ) is a tight witness pair ensures that no rotors outside of H or the subgraphs W will move until the rotor at z shifts to e.</p><p>Consider now the rotor-routing process in the graph H beginning from starting configuration U . By Lemma 5.3, for all neighbors v of z in R H , the rotor at z shifts to \{ z, v\} before the rotor at any v \in R H completes a full rotation. Considering the implication of this statement in the graph G, we see that the rotor at z must shift to e before the rotors at any v \in R H complete a full rotation, and so the entire rotor-routing process taking T to T \prime terminates before the rotor of any vertex in R H completes a full rotation. Now, either the rotor-routing process associated to \rho q ([(z) -(q)], T ) never deposited a chip at any vertex in R H , and so all edges from z to vertices in R H lie in the tree T \prime , or this rotor-routing process deposited a chip at some vertex of R H , and so the edge joining z to this vertex lies outside the tree T \prime .</p><p>In the former case, since no v \in R H received a chip, all vertices of R H have degree 1 in the subgraph T \prime \cap H. This means in particular that on the Bernardi tour associated to T \prime , all vertices of R H receive at least one chip from their neighbors in H in addition to whatever chips they receive from edges to neighbors outside H. However, on the Bernardi tour associated to T , all edges of H incident to vertices in R H will be cut at their other endpoint. Therefore, no vertex of R H will receive any chips from neighbors in H on this tour, while each such vertex receives the same number of chips as in the tour of T \prime from neighbors outside H. This means in particular that the break divisors \beta (q,e) (T \prime ) and (z) -(q) + \beta (q,e) (T ) differ at all vertices of R H and so cannot be representative of the same element of Pic g (G).</p><p>In the latter case, suppose that the rotor-routing process on T deposited a chip at the vertex v, and so \{ z, v\} lies outside T \prime . Then, in this case, the chip must have returned to z across an edge connecting z to some vertex \ell in L H . Indeed, all edges incident to z on the right of C are included in the tree T , so in order for the chip to return to z across an edge connecting z to a vertex in R H , the rotor at that vertex must complete a full rotation, which is impossible by Lemma 5.3. Further, by appealing to this same lemma, we also see that the edge \{ \ell , z\} across which the chip returned to z must be contained in T \prime . This follows from the fact that \{ \ell , z\} is in the tree T by construction, and so 5.3 ensures that the rotor at z reaches \{ \ell , z\} before the chip returns across this edge. In particular, the rotor at z continues to the next edge in the ribbon structure at z, and so the rotor at \ell remains \{ \ell , z\} at the termination of the process.</p><p>Thus, for all v \in R H which received a chip during the rotor-routing process, the unique path from v to z in T \prime passes through a vertex of L H . Note in particular that this need not be the case if G has multiple edges. Indeed, in the event that there is some v \in R H such that there are parallel edges between z and v, v may receive a chip in the rotor-routing process and eventually terminate in a configuration such that one of the edges \{ v, z\} is still in the tree T \prime . This means that in the Bernardi tour associated to T \prime , the edge \{ z, v\} will be cut at z, and so z will receive a chip from an edge to at least one neighbor in R H . Considering also that z receives a chip from an edge in the cycle C, we see that z must have a coefficient of at least 2 in the break divisor \beta (q,e) (T \prime ). On the other hand, all edges incident to z except for e are included in T , so that in the divisor \beta (q,e) (T ), the vertex z has coefficient 0. Thus, the break divisors (z) -(q) + \beta (q,e) (T ) and \beta (q,e) (T \prime ) necessarily differ at z and cannot be representative of the same element of Pic g (G).</p><p>Combining the above, we conclude that in either case, the rotor-routing and Bernardi torsors disagree on the tree T , so that \rho q \not = \beta q .</p><p>In view of Theorems 6.2 and 7.1 we obtain the following corollary, verifying the Baker--Wang conjecture.</p><p>Corollary 7.2 (Baker--Wang conjecture). Given a nonplanar ribbon graph with no multiple edges and no loops, there exists a vertex q for which \rho q \not = \beta q .</p><p>8. Examples. We conclude with two examples to address the following questions:</p><p>1. Is the assumption of no multiple edges necessary for the Baker--Wang conjecture? 2. For nonplanar ribbon graphs, is the difference between the two torsor structures independent of the base vertex? Our examples show that the answer to question (1 is yes, while the answer to question 2 is no.</p><p>Fix a labeling of the N spanning trees of the underlying graph and consider the actions \rho q and \beta q as homomorphisms:</p><p>The difference between the two torsors on a given divisor class [D] can be computed as \rho q ([D]) - 1 \beta q ([D]) \in S N . The difference between the two torsor structures is dependent on the base vertex if there exist two vertices p and q and a divisor class [D] such that</p><p>Example 8.1. Consider the graph depicted in Figure <ref type="figure">8</ref>.1, which we call the rounded bowtie graph. For the rounded bowtie graph, we have that \rho q = \beta q for all vertices q.</p><p>We compute all Bernardi and rotor-routing torsor structures on the graph. This graph has only four spanning trees, yielding a particularly straightforward collection of computations.</p><p>For a given spanning tree T and a given choice of vertex q and incident edge e = \{ v, q\} , we first compute the rotor-routing action, yielding a permutation \rho q ([(v) -(q)]) \in S n . Following this, we calculate the permutation associated to the Bernardi action by computing break divisors \beta (q,e) (T ) for all spanning trees T . Break divisors will be written as row vectors with coordinates in the order (a, b, c). With  these in hand, we translate the break divisors by the divisor (v) -(q), yielding new divisors, (8.1) (v) -(q) + \beta (q,e) (T )</p><p>for each spanning tree T . Finally, we can check for linear equivalence between these divisors and the other break divisors \beta (q,e) (S) for other spanning trees S. In particular, if some pair of spanning trees S and T satisfy (v) -(q) + \beta (q,e) (T ) \sim \beta (q,e) (S) , then the permutation \beta q ([(v) - (q)]) sends the spanning tree T to S. Calculating these divisors (8.1) and performing these comparisons for all spanning trees then yields a permutation representation for \beta q ([(v) -(q)]).</p><p>In order to verify that \rho q = \beta q for all vertices q of the graph in 8.1, we perform this process for all choices of vertices q and v of the graph in order to understand the action of divisor classes of Div 0 (G) generators on the trees. There are three nonequivalent choices for q and v: one in which the target vertex q has degree 4 and the vertex v has degree 2, one in which q and v both have degree 2, and one in which q has degree 2 and v has degree 4. In Figure <ref type="figure">8</ref>.2, we have included a depiction of the four spanning trees of the rounded bowtie graph, together with their associated Bernardi tours and break divisors for this choice of q = c and v = a.</p><p>In calculating the Bernardi action, we make use of the Bernardi bijection \beta (c,e) , where e = \{ c, a\} . Consider the divisor (a) -(c) + \beta (c,e) (T 1 ) = (1, 1, 0). This is the break divisor \beta (c,e) (T 3 ), so we conclude that \beta c ([(a) -(c)]) maps T 1 to T 3 . Similarly, we see  We now turn to the issue of the discrepancy between base vertices, illustrating in particular a graph, depicted in Figure <ref type="figure">8</ref>.3, having vertices p and q such that \rho p = \beta p while \rho q \not = \beta q .</p><p>Example 8.2. Consider the graph depicted in Figure <ref type="figure">8</ref>.3, which we call the pointed bowtie graph. For the pointed bowtie graph and the two indicated vertices p and q, we have \rho p = \beta p while \rho q \not = \beta q .</p><p>We deal with divisor classes of the form [(z) -(q)] for the sake of their simplicity. In Figure <ref type="figure">8</ref>.3, all vertices are labeled, and in addition, we have included all spanning trees of the graph, together with the labeling to be used in the cycle representation of our actions. We will show that Before proceeding, note that by Theorem 7.1, we are guaranteed that \rho q \not = \beta q for this choice of q. We compute now the action of the divisor [(a) - (p)] on each spanning tree using the actions \rho q and \beta q . Since the base vertex of these actions is q and not p, we must first rewrite [(a) -(p)] = [(a) -(q)] -[(p) -(q)], after which one can compute the actions of [(a) -(q)] and [(p) -(q)] under both \rho q and \beta q . Finally, \rho q ([(a) -(p)]) = \rho q ([(a) -(q)] \bigl[ \rho q ([(p) -(q)]) \bigr] - 1 , \beta q ([(a) -(p)]) = \beta q ([(a) -(q)] \bigl[ \beta q ([(p) -(q)]) \bigr] - 1 .</p><p>With these permutations in hand, we are able to form the product \rho q ([(a) -(p)] - 1 \beta q ([(a) - (p)]). To start this process, we have produced in Figure <ref type="figure">8</ref>.4 the Bernardi tours and break divisors associated to each of the spanning trees. For this example, break divisors will be written as row vectors with coordinates in the order (a, b, p, c, q). With these in hand, we can now compute the cycle representation of the permutation \beta q ([(p) -(q)]) using the Bernardi bijection \beta (q,e) , where e is the edge \{ p, q\} .</p><p>Beginning from the spanning tree T 1 , computation of the translates [(p) -(q)] + [\beta (q,e) (T i )] and their break divisor representatives yields that the permutation \beta q ([(p) -(q)]) has cycle representation (193)(278)(456). On the other hand, it can be verified from performing rotor-routing on the trees T 1 , . . . , T 9 that \rho q ([(p) -(q)]) = (123)(459)(678). We have \beta q ([(p) -(q)]) = (193)(278)(456) , \rho q ([(p) -(q)]) = (123)(459)(678) . Following this same reasoning, we can obtain cycle representations \beta q ([(a) -(q)]) = (139)(287)(465) , \rho q ([(a) -(q)]) = (132)(495)(687) .</p><p>Combining the above expressions, it follows that \rho q ([(a) -(p)]) - 1 \beta q ([(a) -(p)]) = (158)(269) .</p><p>It remains to see that the torsor actions of \beta p ([(a) -(p)]) and \rho p ([(a) -(p)]) coincide. This is most easily done by taking the same approach as above, utilizing the Bernardi bijection \beta (p,f ) . These computations are more straightforward than those above, so we leave the work as an exercise. One obtains ). Now note that by the symmetry of the pointed bowtie graph, the coincidence of \rho p and \beta p on [(a) -(p)] ensures their coincidence on all generators of Div 0 (G). We now have \rho q \not = \beta q while \rho p = \beta p .</p></div><note xmlns="http://www.tei-c.org/ns/1.0" place="foot" xml:id="foot_0"><p>Given e i and v i , the edge e i+1 is chosen to be the first edge of T succeeding e i in the ribbon structure about v i ; the process can be thought of as originating with the edge e 0 immediately preceding e in the ribbon structure about q. We think of this process as starting with the initial data, traversing each edge of T to the other endpoint, and cycling through the ribbon structure until finding another edge of T , ignoring each edge f not included in T . In such a tour, we ignore g many edges f 1 , . . . , f g / \in T , each being passed over two distinct times, once from each endpoint. In performing such a tour of G, we construct a divisor on G by placing a chip at our feet the first time we ignore a given edge f / \in T . In this case, we say that f is cut at the first endpoint, from which we ignore it. The divisor obtained in this fashion is necessarily a break divisor for the tree T . The mapping \beta (q,e) : \scrT (G) \rightar B(G) is given by mapping a spanning tree T to the resulting break divisor.Copyright &#169; by SIAM. Unauthorized reproduction of this article is prohibited. Downloaded 06/25/24 to 128.95.104.109 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy</p></note>
			<note xmlns="http://www.tei-c.org/ns/1.0" place="foot" xml:id="foot_1"><p>Copyright &#169; by SIAM. Unauthorized reproduction of this article is prohibited. Downloaded 06/25/24 to 128.95.104.109 . Redistribution subject to SIAM license or copyright; see https://epubs.siam.org/terms-privacy</p></note>
		</body>
		</text>
</TEI>
