Alexander and Thurston norms of graph links
Abstract.
We show that the Alexander and Thurston norms are the same for all irreducible EisenbudNeumann graph links in homology spheres. These are the links obtained by splicing Seifert links in homology spheres together along tori. By combining this result with previous results, we prove that the two norms coincide for all links in if either of the following two conditions are met; the link is a graph link, so that the JSJ decomposition of its complement in is made up of pieces which are all Seifertfibered, or the link is alternating and not a torus link, so that the JSJ decomposition of its complement in is made up of pieces which are all hyperbolic. We use the EN obstructions to fibrations for graph links together with the Thurston cone theorem on link fibrations to deduce that every facet of the reduced Thurston norm unit ball of a graph link is a fibered facet.
Key words and phrases:
Alexander norm, Thurston norm, EisenbudNeumann graph links, JSJ decomposition, link fibration2000 Mathematics Subject Classification:
Primary 57M27.1. Introduction
1.1. The Alexander norm (Anorm)
A seminorm defined on the first cohomology group of a connected, compact, orientable 3manifold , whose boundary (if any) is a union of tori, was introduced by McMullen [8] in the late 1990s. It is directly determined by the multivariable Alexander polynomial of the manifold and is called the Alexander norm of . We shall adopt the notation Anorm for the Alexander norm.
If we write a multivariable Alexander polynomial in multiindex notation, then with and denoting the support of the polynomial ; the set of all labeling nonzero constants . Let and be the exponents of any two arbitrarily chosen terms of . They are elements of . Let be a cohomology class in the space dual to the space containing the exponents so that where is the first Betti number of . The Anorm of is the supremum of taken over all the exponents in the support of .
1.2. The Thurston norm (Tnorm)
The Anorm is closely related to a seminorm for compact, oriented manifolds introduced in 1986 in [13], called the Thurston norm. We shall adopt the notation Tnorm for the Thurston norm. In the compact, oriented 3manifold with a boundary which may be empty, every homology class can be represented as where is a properly embedded, oriented surface. By Poincaré duality the relative second homology group is isomorphic to the first cohomology group, . The norm is determined by the Euler characteristic of the surface representing the cohomology class through this duality isomorphism.
1.3. EisenbudNeumann (EN) graph links
An EN graph link (see [3, 10]) is a link in a homology sphere , which can be built up by splicing together Seifert links in homology spheres. We use the notation to indicate not only the link but also the homology 3sphere it lies in. To each graph link there is a unique minimal graph. It is shown in [3] that the graph directly determines the Alexander polynomial and Thurston norm of .
1.4. Main theorem: The coincidence of Alexander and Thurston norms for graph links
Theorem 1.5.
The Alexander and Thurston norms coincide for all irreducible graph links with two or more components.
We prove this theorem by directly calculating the Anorm of graph link from the expression for its Alexander polynomial, and comparing it to the expression for its Tnorm. We derive three new results which are used in the proof. Each of these results involves a sum over the splice components of the graph of L;

a Tnorm decomposition formula,

the Newton polyhedron of the Alexander polynomial of as a Minkowski sum, and

an Anorm decomposition formula.
In the case of a graph knot, which is a graph link with one component, we show as a corollary that the two norms satisfy .
Without actually calculating the two norms themselves, McMullen proved in [8] that for a compact, connected, oriented manifold whose boundary if any is a union of tori, the two norms must coincide for all fibered cohomology classes which are primitive. A primitive cohomology class is one whose image is so that . Hence our result that the two norms coincide for graph links is new with regard to nonprimitive cohomology classes and the nonfibered cohomology classes.
1.6. Generalization of the main theorem
As an application of our main result, we combine it with previous results to obtain a more general theorem in the context of the JSJ (Jaco, Shalen and Johannson) decomposition of a link in . According to the JSJ decomposition, the link complement can be decomposed into irreducible pieces of two types, Seifertfibered and hyperbolic, by desplicing the pieces together along tori.
Theorem 1.7.
Let be a link in . Let the JSJ decomposition of the link complement consist of irreducible pieces so that
Then the Thurston and Alexander norms of coincide if either of the following two conditions are met:

is a graph link so that all the are Seifertfibered.

is an alternating link which is not a torus link, so that all the are hyperbolic.
1.8. Characteristic hyperplanes of the Thurston norm ball
Tnorm unit ball and fibrations
The Tnorm can be extended by convexity from integervalued classes to realvalued classes to determine a Tnorm unit ball. The cohomology classes of the Tnorm unit ball can be separated into two types; fibered and nonfibered. An integervalued cohomology class is fibered if the surface which represents it via Poincaré duality is also the fiber of a fibration of 3manifold over the circle . A realvalued cohomology class is fibered if it lies on a ray through the origin whose lattice points are integervalued fibered cohomology classes. Otherwise is a nonfibered cohomology class. We do not require the fiber to be connected.
Thurston cone theorem
The Thurston norm unit ball may not be a bounded set. However, we show in Long [6] that by introducing essential coordinates we can define a reduced Tnorm unit ball which is a polyhedron of the same dimension as the Newton polyhedron of the Alexander polynomial of 3manifold . The Thurston cone theorem (see [13]) states that the set of fibered cohomology classes is some union of the cones pointed at the origin (minus the origin) through the interiors of the facets (topdimensional faces) of the reduced Tnorm unit ball.
Fibration obstruction criterion
Using necessary and sufficient conditions for a cohomology class of a graph link to be a fibered class from [3], we show that the set of all nonfibered classes of is made up of a set of hyperplanes we call the characteristic hyperplanes. There is one characteristic hyperplane for each splice component of the graph of . The th characteristic hyperplane is made up of those classes whose Tnorm for the th splice component in the Thurston norm decomposition formula is zero.
The fibered facets of the reduced Thurston norm unit ball
We use the fibration obstruction criterion and the Thurston cone theorem to prove the following new theorem on graph link fibrations.
Theorem 1.9.
Every facet of the reduced Tnorm unit ball for the irreducible graph link L is a fibered facet.
This theorem implies that the boundary of each facet, which is made up of the lower dimensional faces, must be contained in the nonfibered set of cohomology classes.
1.10. Sample calculation
We conclude by applying our results to a sample graph link from Eisenbud and Neumann [3]. We find the Alexander polynomial, Thurston norm, reduced Tnorm unit ball and characteristic hyperplanes for this link and determine the intersection of the characteristic hyperplanes with the reduced Tnorm unit ball.
2. Alexander and Thurston norms
2.1. Alexander norm
Let be the Alexander polynomial of a compact, connected, orientable 3manifold with first Betti number , whose boundary (if any) is a union of tori. This polynomial can be expressed using multiindex notation as follows:
The Anorm is a seminorm defined on the first cohomology group of a 3manifold directly determined by the Alexander polynomial of . The Alexander polynomial is a Laurent polynomial; . Hence can be expressed as a finite sum as in Equation (2.1). Then . Further let be an element of the dual vector space to the space in which each lies in. Then there is a seminorm, which we shall call the Anorm, defined as follows.
Definition 2.2.
The Anorm of for the Alexander polynomial of is
(2)  
Remark 2.3.
The definition of the Anorm above can be extended to realvalued cohomology classes =.
Since the Anorm is completely determined by the Alexander polynomial we also use the notation . In Long [6], we note that the Alexander norm is the special case of a norm determined by Laurent polynomials in general for which the polynomial is an Alexander polynomial. We call this generalized norm the Laurent norm since each Laurent polynomial with integer coefficients determines a norm ; the Laurent norm for .
2.4. Thurston norm
The Tnorm for compact, oriented 3manifolds (with or without a boundary ) was first defined on the second relative homology group of in Thurston [13]. Any class of this relative second homology group can be represented by a compact, oriented twodimensional surface in . Each such surface has an integervalued Euler characteristic which can be used to define a norm on . By Poincaré duality each class of the group determines a class of the first cohomology group of ; . Due to this equivalence, the Tnorm can also be defined as a norm on . We use the formulation of the Tnorm as a norm on . In Dunfield [2], this version of the Tnorm is defined as follows:
Definition 2.5 (Dunfield, [2]).
For a compact, connected surface , let if and 0 otherwise. For a surface with multiple connected components let be the sum of the . Then the Tnorm of an integervalued class is
Remark 2.6.
The surface of this definition may or may not be connected. It is shown in [13] that if cohomology class satisfies for some integer , then because represents disjoint surfaces each with Tnorm . The Tnorm is additive for disjoint surfaces.
Also in [13], the Tnorm is extended using convexity from integervalued classes to realvalued classes to determine a convex set in called the Tnorm unit ball . Thus we consider the Tnorm as a seminorm on , the first cohomology group of with real coefficients, rather than .
3. EN graph links
3.1. Link splicing
We use the notation that denotes the link and its ambient space which is a homology sphere. Given two links, in and in , with and components respectively we may form the link in with components by selecting link components, of and of and splicing the two links together along and . To do this, first assume that and have tubular neighborhoods with meridian and longitude and respectively. To splice to along and we attach to and to .
The splice is a homeomorphism of a tubular neighborhood of which is a solid torus, , to a tubular neighborhood of , also a solid torus, , along a boundary torus . The action can be represented as . This union of two solid tori across a torus induces the appropriate map in homology connecting each meridian with a longitude.
3.2. Seifert links
Seifertfibered spaces
To construct a Seifertfibered space, we start with a solid torus and remove parallel tori to obtain the space . The fundamental group of this space is given by
This is an fibration of with base space a disc with points removed and circle fibers. and represent the longitude and meridian of the ambient solid torus . Each , represents a meridian of th torus determined by the th hole of the disc. The commutation relations say that each generator commutes with the longitude of which means the fibration has trivial monodromy. Next we attach solid tori, , to along the boundary tori in a special way using attaching maps that are homeomorphisms.
Each meridian generator, , of the solid torus, , is mapped to for . For the solid torus the meridian generator, is mapped to where is an integer:
The space we obtain by this procedure is called the Seifertfibered space
. It has exceptional fibers of type where and denote the number of times the th exceptional fiber wraps around the th torus longitudinally and meridianally respectively. The exceptional fibers are the core circles of the solid tori after we have attached them; if , then the core circle of which becomes the exceptional fiber is .
This space has fundamental group with presentation
It is equivalent to the fundamental group of with the one additional relation added for each of the solid tori attached. This construction of Seifertfibered spaces is taken directly from Zieschang [17].
Seifert link:
We obtain an component Siefert link from the Seifertfibered space by removing a tubular neighborhood from each of the first exceptional fibers. Each component , with , has a complement in the ambient Seifertfibered space which is equivalent to a torus knot, labeled by . We leave the remaining components alone so that the complement of link has singular fibers. This link is denoted . The ambient Seifertfibered space and together form the pair . Thus denotes the link and L denotes both the ambient space the link lies in and the link itself; . The following is an equivalent but more concise definition of a Seifert link.
Definition 3.3 (Eisenbud and Neumann, [3], pg. 24).
Let be a link in a manifold and let the interior of a closed tubular neighborhood of be denoted . Then is a Seifert link if the link exterior, which is , of in is a Seifertfibered space.
Remark 3.4.
If the Seifert link is in the 3sphere , we drop the boldfaced notation so that in this case.
3.5. Seifertfibered homology 3spheres
If we require that the Seifertfibered space has the homology of the sphere we must have the additional relations that
The notation means to remove the component from the equation. In this case it is not necessary to include the coefficients in uniquely labeling the Seifertfibered space so that if the space is in a homology 3sphere we use the notation and call it an unoriented Seifertfibered homology 3sphere of type . We use the notation with to indicate the two possible orientations of the Seifertfibered homology 3sphere. Thus we use oriented Seifertfibered homology 3spheres. The ambient oriented Seifertfibered homology 3sphere together with link form the pair denoted . This is the notation we shall use for all Seifert links which we assume are in oriented Seifertfibered homology 3spheres. The Seifert link as defined above is an component link with first homology group given by as would be the case if it was instead in the sphere. To a given unoriented Seifertfibered homology sphere, there exists a unique unordered tuple of coprime integers with .
3.6. Graphs of Seifert links and splice diagrams
We can represent a Seifert link in a Seifertfibered homology 3sphere as a graph with the following components.

Boundary vertices:
The boundary vertex corresponds to the solid torus which is a neighborhood of an exceptional fiber labeled by .

Arrowhead vertices:
The arrowhead vertex corresponds to one of the link components where a tubular neighborhood of an exceptional fiber has been removed. It also has the label .

Nodes: and
A node corresponds to a Seifert manifold embedded in a link exterior; the edges incident to a node correspond to the boundary components of the Seifert manifold, or for edges that lead to boundary vertices or arrowhead vertices to boundaries of tubular neighborhoods of fibers. Within each node we insert and corresponding to the two possible orientations of the Seifert manifold. Each node must have at least three edges incident on it.
We can now form a splice diagram (graph) by connecting two arrowhead vertices of the graphs of a pair of Seifertfibered links. An example of a splice diagram is shown in Figure 1. A splice diagram determines an EN graph link which is defined as follows:
Definition 3.7.
An EN graph link consists of either a Seifert link in a homology 3sphere or the splice of two or more Seifert links in homology 3spheres.
In the rest of this article, we assume that the graph of graph link L has arrowhead vertices labeled , nodes labeled and boundary vertices labeled . We also assume that the graph link is irreducible; it can not be expressed as a disjoint sum. In addition, by the term graph link we mean an EN graph link.
3.8. Alexander polynomial of a graph link
The Alexander polynomial of the graph link L is directly determined by its graph .
Theorem 3.9 (Eisenbud and Neumann [3], Theorem 12.1).
Assume (the graph of L) is connected. Then the Alexander polynomial of the graph link L is
(3) 
If , so that the link is a knot, the formula is
(4) 
Any terms of the form ( denotes an integer) which may occur on the righthand side of these two equations should be formally canceled against each other before being set equal to zero.
The Alexander polynomial has the same number of variables as arrowhead vertices and the product is over all vertices which are not arrowhead vertices; the nodes and boundary vertices. indicates the number of edges incident of each vertex. The linking numbers can be obtained directly from the graph. For any two distinct vertices and of a graph , let be the simple path in joining to , including and . Then we have that
(5) 
3.10. Tnorm of graph a link
The Tnorm of the graph link L is also directly determined by the graph of L.
Theorem 3.11 (EisenbudNeumann [3], Thm. 11.1).
The Tnorm of for the irreducible graph link L, which is not the unknot in , is
(6) 
Remark 3.12.
For a graph link with components, let be the greatest common divisor of the components of the vector . Up to homeomorphism, the surface that the cohomology class represents is the disjoint union of identical, connected, compact, oriented, twodimensional surfaces each having genus and holes. The genus of each of these surfaces can be determined using the wellknown formula for the Euler characteristic of such a surface,
and solving the equation for the Thurston norm,
for using the value of from Equation (6).
4. The coincidence of Thurston and Alexander norms for graph links
We now prove the main result of this article relating the Alexander and Thurston norms for graph links in homology 3spheres.
Theorem 4.1.
The Alexander and Thurston norms coincide for all irreducible graph links with two or more components.
Remark 4.2.
By Corollary 8.3 of Eisenbud and Neumann [3], there is a unique minimal splice diagram for each graph link such that every edge weight is nonnegative. Hence in this proof without loss of generality we can assume that the edge weights of all the arrowhead and boundary vertices are nonnegative; and . Even more, since a boundary vertex with an edge weight equal to one represents a nonsingular fiber, which is not an exceptional Seifert fiber, we can assume that for boundary vertices.
We prove the theorem by directly calculating the Anorm of the graph link L using Equation (3) for the Alexander polynomial and comparing the result to Equation (6) for the Tnorm as given in [3]. We derive three new results which are used in the proof. Each of these results involves a sum over the splice components of the graph of L;

a Tnorm decomposition formula,

the Newton polyhedron of the Alexander polynomial of as a Minkowski sum, and

an Anorm decomposition formula.
4.3. Tnorm decomposition formula
To show that the Tnorm is a sum of the Tnorms of each node of the graph of a graph link we need the following lemma.
Lemma 4.4.
The linking numbers of the arrowhead vertices into a node and into a boundary vertex attached to the same node of a graph link differ only by a factor of , where is the weight of the edge in the graph connecting the boundary vertex indexed by to the node.
Proof.
The proof follows directly from the formula for the linking numbers given by Equation (5). It says that to find the linking number between an arrowhead vertex and either a node or a boundary vertex we follow the path on the graph connecting the arrowhead vertex to the node or the boundary vertex. Along the way we multiply the product of all the signs of the nodes and also multiply all the edge weights of edges going into each node but not on the path. Since a boundary vertex is attached to the node by hypothesis, it is clear that the paths between any arrowhead vertex and either the boundary vertex or the node it is attached to are the same except the path to the boundary vertex contains the edge connecting that vertex to the node. Hence the weight of that edge, since it lies on the path, does not appear in the linking number between the arrowhead vertex and boundary vertex but it does in the linking number of that arrowhead vertex with the node. Hence the two linking numbers differ exactly by the factor which is the weight attached to the edge of boundary vertex indexed by . ∎
This lemma implies the following Corollary 4.5 to Theorem 3.11 which gives a Tnorm decomposition formula for the irreducible graph link L:
Corollary 4.5.
The Tnorm of graph link L, can be expressed as a sum of the Tnorms of the splice components of the graph of L.
where and is the contribution to the Tnorm of the link from the th splice component of the graph of the link.
Remark 4.6.
We use the notation that denotes number of boundary vertices attached to the th node, so that the boundary vertices attached to the th node can be indexed by . In addition, we denote by the edge weight of the th boundary vertex attached to the th node. In [3], the boundary vertices and the edge weights are ordered in a manner that does not specify which node the boundary vertex is attached to. Hence the notation for an edge weight is used in [3], which we have refined to .
Proof.
The expression for the Tnorm of this corollary can be obtained from the equation of the Tnorm given in [3], Equation (6), by a direct application of the Lemma 4.4 relating the linking numbers between arrowhead vertices to a node and the boundary vertices attached to the node. In effect, the terms of the original expression are replaced by the terms and the sum over both nodes and boundary vertices is replaced by a sum over nodes only. To be precise we shall present this proof in detail.
Without loss of generality, we can prove the formula for the Tnorm by showing it is true for the th vertex which is also the first node that has by hypothesis boundary vertices attached to it. The equation given in [3] for the Tnorm of an irreducible graph link, Equation (6), can be written
We’ve split the sum into contributions from the nodes first and then the boundary vertices. The contribution to the Thurston norm from the first node is
Applying the Lemma 4.4 that relates the linking numbers of the boundary vertices to the node we have that
We can combine the contribution from the node with the contributions from its boundary vertices to obtain
By repeating this procedure indicated for the th node on all of the nodes we obtain the Equation (4.5) as claimed.
In order to show that , we proceed by induction on the number of boundary vertices attached to the node. Without loss of generality we can show that this relation is true for the first node in order to prove it is true for all nodes. First, we use that in the definition of a node given in [3] and proceed by induction. Hence consider the first node indexed as and assume that it has only a single boundary vertex attached to it so that . We have that implies that
Next let us assume that for and we will show that this implies for . Adding the th boundary vertex to the th vertex, which is the first node, increases by one because of the additional edge into the node and also adds the term to . Adding this additional boundary vertex adds the strictly positive term to . By the inductive hypothesis without this additional boundary vertex. We find that we must have after the addition of the strictly positive contribution of the th boundary vertex. Hence by induction for all values of .∎
4.7. Newton polyhedron of the Alexander polynomial of a graph link
We also have to use another application of Lemma 4.4 in order to prove our fundamental theorem. It involves obtaining an expression for the Newton polyhedron of the graph link L.
The Newton polyhedron of a polynomial is the convex hull of the exponents of the Alexander polynomial .
The following proposition and corollary state two wellknown properties of the Minkowski sum. The second corollary, although somewhat trivial, is new and will be essential in our proof of Theorem 4.11.
Proposition 4.8.
(Gelfand, Kapranov and Zelevinsky [4], Prop. 6.1.2(b)) The Newton polyhedron of the product of two polynomials, and , is given by
Applying this proposition inductively to the product of with itself we obtain the following corollary.
Corollary 4.9.
The Newton polyhedron of a polynomial to a power , , is given by
Corollary 4.10.
Assume that divides for the rational polynomial . Then the Newton polyhedron of this rational polynomial satisfies the relation
(8) 
Proof.
We use that can be written as a product of polynomials so that and apply Proposition 4.8. ∎
Theorem 4.11.
The Newton polyhedron of the graph link L, is given by
Proof.
By Corollary 4.10, given polynomials and such that divides , their Newton polyhedra satisfy the equation . If we set and , then is the Alexander polynomial of L as given in Equation (3). We can prove this corollary by showing directly that is a solution of the equation .
Substituting and into Equation (8) we obtain
(10) 
In this equation, we’ve also used that
since , by Corollary 4.9. By Lemma 4.4, the linking numbers of the boundary vertices are the same as the node they are attached to up to a factor of . Using this result in , we obtain
The Newton polyhedra of this sum are all line segments with endpoints and . It is geometrically clear that is a scaling factor of each of these segments which reduces the length of the segment but does not change its direction. Hence we have the equality
We now substitute this expression for along with our candidate solution for into Equation (10) and obtain
In this equation we’ve introduced the constants and each of which multiplies the same line segment . The two terms on the right of this equation can be combined using Minkowski addition provided that and are nonnegative for all . By Corollary 4.5, . As mentioned in Remark 4.2, we may assume, without loss of generality, that edge weights for boundary vertices satisfy the inequality . This implies that . Hence since all the constants and which multiply the same line segment are positive, we can use Minkowski addition to combine the two expressions on the right of this equation and the sums involving cancel each other.
Thus we see that the equality of Equation (10) is satisfied.∎
A polyhedron which is the Minkowski sum of line segments is called a zonotope.
Corollary 4.12.
The Newton polyhedron of the graph link L is a zonotope consisting of line segments, one for each splice component ot the graph of L.
Proof.
By Equation (4.11), is a Minkowski sum of the Newton polyhedra, . The th component of this sum, which is the Newton polyhedron of the th splice component of the graph of L, has only two vertices, the endpoints and . The convex hull of these two vertices is a line segment. ∎
Remark 4.13.
If the graph link L has only one node, it is a Seifert link. By Corollary 4.12 for with , the Newton polyhedron of a Seifert link is a zonotope consisting of a single line segment. Dimca, Papadima and Suciu have found the same result in [1] for the Newton polyhedron of a Seifert link by finding a coordinate system for which the Alexander polynomial is a function of only one variable, which they call its essential variable.
4.14. Anorm decomposition formula
Since the Alexander norm is determined by the vertices of the Newton polyhedron , it can be viewed not only as a norm determined by an Alexander polynomial , but also as a norm determined by the Newton polyhedron of the Alexander polynomial . This means that the notations
can also be useful. In Long [6], we determine that the Anorm for the Newton polyhedron is equal to the width function of so that . In the theory of polyhedra, the width function is a wellknown function that can be proved to be Minkowski linear. This means that for arbitrarily chosen polyhedra and and all , it satisfies following:

Minkowski additivity : .

Minkowski scaling: .
A proof of the Minkowski linearity of the width function can be found in Long [5].
The equivalence of the width function of and the Anorm for along with the Minkowski linearity of the width function imply that the Anorm has the following properties:
By inductively applying Minkowski linearity, we obtain the following Anorm decomposition formula.
Proposition 4.15.
Assume that the Newton polyhedron of the Alexander polynomial can be written as a Minkowski sum of component Newton polyhedra, , with , so that
Then the Anorm for is the sum of the Anorms of each component Newton polyhedron:
(11) 
Remark 4.16.
In Long [6], we derive a decomposition formula of the Alexander norm for the polynomial expressed as a sum involving the Alexander norms for each the irreducible factors of ; for and .
By substituting the expression for the Newton polyhedron of L, Equation (4.11), into the Anorm decomposition formula, Equation (11), we obtain as a corollary the Anorm decomposition formula for graph links.
Corollary 4.17.
The Anorm of graph link L, can be expressed as a sum of the Anorms of the splice components of the graph of L.
where denotes the contribution to the Alexander norm from the th splice component of the graph of L.
4.18. Proof of the main theorem
Proof.
We prove the theorem by showing that Equation (4.17), the Anorm decomposition formula for graph links, is equal to Equation (4.5), the Tnorm decomposition formula. Each term in the sum of the Anorm decomposition formula for graph links involves the Anorm of a Newton polyhedron which is a line segment with endpoints at the origin and at . The Anorm for the line segment can be obtained by direct application of the Definition 2.2 of the Anorm for the polynomial . This polynomial has two exponents and . The supremum which occurs in this definition must take its value on the difference of these two exponents since this polynomial has only two terms. Hence, we have that