4 Moment maps and tropical charts [035I]
Original official author HTML, exact retained edition. Historical TeX conversion verdicts remain unchanged. Cited-edition alignment and mathematical self-containment are not assessed.
Complete original source context · Original author HTML
4 Moment maps and tropical charts
A complex manifold is locally defined using analytic charts . The charts help to transport the analysis from to . The idea in the non-archimedean setting is similar replacing the above charts by algebraic moment maps to multiplicative tori and the corresponding tropicalizations . The restriction of to the preimage of an open analytic subset will be called a tropical chart.
In this section, is an algebraically closed field endowed with a complete non-trivial non-archimedean absolute value . Note that the residue field is also algebraically closed. Let be the associated valuation and let be the value group. We will study analytifications, tropicalizations and moment maps of the algebraic variety over . This will be used in the next section to define -forms on .
4.1
We recall first the construction of the analytification of . Let be an open affine subset of , then is the set of multiplicative seminorms on extending the given absolute value on . This set is endowed with the topology generated by the functions with ranging over . By glueing, we get a topological space which is connected locally compact and Hausdorff. We can endow it with a sheaf of analytic functions leading to a Berkovich analytic space over which we call the analytification of . For a morphism of algebraic varieties over , we get an analytic morphism induced by composing the multiplicative semiorms with on suitable affine open subsets. We refer to [Be90] for details, or to [BPS11], §1.2, for a neat description of the analytification.
4.2
We will define some local invariants in . On an open affine neighbourhood , the point is given by a multiplicative seminorm on and we often write for . Dividing out the prime ideal , we get a multiplicative norm on the integral domain which extends to an absolute value on the quotient field of . The completion of this field is denoted by . It does not depend on the choice of and it may be also constructed analytically. The absolute value of is denoted by as it extends the given absolute value on . Note that the completed residue field of remains the same if we replace the ambient variety by the Zariski closure of in .
Let be the transcendence degree of the residue field of over . The quotient of the value group of by is a finitely generated abelian group and we denote its -rank by . Finally, we set . By [Be90], Proposition 9.1.3, we have for every open subset of .
Example 4.3
Let be the split multiplicative torus of rank with coordinates . Then a point of could be visualized by the coordinates and the multiplicative seminorm corresponding to is given by for every Laurent polynomial on . Conversely, every field extension with an absolute value extending the given absolute value on and every give rise to a point by . Note that and are not uniquely determined by .
In particular, we get an inclusion of into . For every , we have . However, there can be also other points with . If , then precisely the points of type 1 (i.e. the -rational points) and the points of type 4 satisfy (see [Be90], 1.4.4).
Returning to the case , there are some distinguished points of which behave completely different than -rational points. For positive real numbers , we define the associated weighted Gauss norm on by
for every Laurent polynomial . It follows from the Gauss Lemma that the weighted Gauss norm is a multiplicative seminorm giving rise to a point . The set is called the skeleton of . Every point satisfies (see [Du12], (0.12) and (0.13)).
4.4
Let be a split multiplicative torus over with coordinates . Then we have the tropicalization map
It is immediate from the definitions that the map is continuous and proper. To get a coordinate free approach, we could use the character group and its dual . Then is a map from to . We refer to [Gu12] for details about tropical geometry.
Remark 4.5
Note that we have a natural section of the tropicalization map. It is given by mapping the point to the weighted Gauss norm associated to . It follows from [Be90], Example 5.2.12, that this section is a homeomorphism of onto a closed subset of which is the skeleton introduced in 4.3. In this way, we may view the tropicalization map as a map from onto . Then it is shown in [Be90], §6.3, that the tropicalization map is a strong deformation retraction of onto the skeleton . This point of view is used very rarely in our paper.
4.6
For a closed subvariety of of dimension , the tropical variety associated to is defined by . The Bieri–Groves theorem says that is a finite union of -dimensional integral -affine polyhedra in . It is shown in tropical geometry that is an integral -affine polyhedral complex. The polyhedral structure is only determined up to subdivision which does not matter for our constructions. We will see below that the tropical variety is endowed with a positive canonical weight satisfying the balancing condition from 3.7. We get a tropical cycle of pure dimension which we also denote by forgetting the weight in the notation.
4.7
The tropical weight on an -dimensional polyhedron of is defined in the following way. By density of the value group in , there is . We choose with . Then the closure of in is a flat variety over whose special fibre is called the initial degeneration of at . Note that is a closed subscheme of . Let be the multiplicity of the irreducible component of . Then the tropical weight is defined by , where ranges over all irreducible components of . One can show that the definition is independent of the choices of and . It is a non-trivial fact from tropical geometry that is a tropical cycle (see [Gu12], §13, for details).
4.8
For an open subset of the algebraic variety , a moment map is a morphism to a split multiplicative torus over . The tropicalization of is
Obviously, this is a continuous map with respect to the topology on the analytification . Note that our moment maps are algebraic which differs from the moment maps in [CD12] which are defined analytically.
We say that the moment map of the open subset of refines the moment map if and if there is an affine homomorphism of the multiplicative tori such that on . Here, an affine homomorphism means a group homomorphism composed with a (multiplicative) translation on . This group homomorphism induces a homomorphism of character lattices. Its dual is the linear part of an integral affine map such that on .
If are finitely many moment maps of non-empty open subsets of with , then is a non-empty open subset of and is a moment map which refines every . Moreover, it follows easily from the universal property of the product that every moment map which refines every refines also .
Lemma 4.9
Let be a moment map on an open subset of and let be a non-empty open subset of . Then .
Proof: Let . We note that is a Laurent domain in and hence it has the same dimension as . We conclude that is not contained in the analytification of the lower dimensional Zariski-closed subset and hence .
4.10
If is a morphism of varieties over , then we define the push-forward of with respect to as the cycle , where the degree of is defined as if is generically finite and we set if . By restriction, the push-forward can be defined in the same way on prime cycles of and extends by linearity to all cycles of .
The following result is called the Sturmfels–Tevelev multiplicity formula. It was proved by Sturmfels and Tevelev [ST08] in the case of a trivial valuation and later generalized by Baker, Payne and Rabinoff [BPR11] for every valued field.
Proposition 4.11
Let be a moment map of the non-empty open subset of which refines the moment map , i.e. there is an affine homomorphism such that on . Then we have
in the sense of tropical cycles (see 3.9).
Proof: In fact, the Sturmfels–Tevelev multiplicity formula is the special case where is a closed subvariety of (see [Gu12], Theorem 13,17, for a proof in our setting deducing it from the original sources). In the general case, we conclude that
Since is dense in , the claim follows.
4.12
We will show that every open affine subset of has a canonical moment map. We note that the abelian group is free of finite rank (see [Sa66], Lemme 1). Here, we use that is algebraically closed (or at least that is geometrically reduced). We choose representatives in of a basis. This leads to a moment map . By construction, refines every other moment map on . Note that this moment map is canonical up to (multiplicative) translation by an element of .
Let be a morphism of algebraic varieties over and let is an open subset of with . Then induces a homomorphism of lattices. We get a canonical affine homomorphism of the canonical tori with . This will be applied very often in the case where is an open subset of in and . Then we get a canonical affine homomorphism .
4.13
Recall that an open subset of is called very affine if has a closed embedding into a multiplicative torus. Clearly, the following conditions are equivalent for an open affine subset of :
- (a)
is very affine;
- (b)
is generated as a -algebra by ;
- (c)
the canonical moment map from 4.12 is a closed embedding.
The intersection of two very affine open subsets is again very affine (see the proof of Proposition 4.16). Moreover, the very affine open subsets of form a basis for the Zariski topology. We conclude that all local considerations can be done using very affine open subsets.
On a very affine open subset, we will almost always use the canonical moment map which is a closed embedding by the above. To simplify the notation, we will set for the tropical variety of in . It is a tropical cycle in , where is the dual abelian group of . The tropicalization map will be denoted by . Recall that is only determined up to translation by an element of and hence and are only canonical up to an affine translation. This ambiguity is no problem as our constructions will be compatible with affine translations.
The following result of Ducros relates the local invariant from 4.2 with tropical dimensions.
Proposition 4.14
For , there is a very affine open neighbourhood of in such that for any open neighbourhood of in , there is a compact neighbourhood of in such that is a finite union of -dimensional integral -affine polytopes.
Proof: We choose rational functions on with such that the reductions form a transcendence basis of the residue field extension of . There are rational functions which are regular at such that form a basis of . By definition, we have . By (0.12) in [Du12], reduce to a transcendence basis of the graded residue field extensions of in the sense of Temkin. There is a very affine open neighbourhood of in such that are invertible on . Let be the coordinates of the canonical moment map . Then the graded reductions of generate a graded subfield of the graded residue field extension of . By construction, this graded subfield has transcendence degree over the graded residue field of . By [Du12], Theorem 3.2, is a finite union of integral -affine polytopes for every compact neighbourhood of in which is strict in the sense of [Be93]. For any open neighbourhood of in , Theorem 3.3 in [Du12] shows that there is a compact strict neighbourhood of in such that is a finite union of -dimensional polytopes.
4.15
A tropical chart on consists of an open subset of contained in for a very affine open subset of with for some open subset of . Here the canonical moment map from 4.12 plays the role of (tropical) coordinates for . By 4.13, is an embedding. The condition means that behaves well with respect to the tropical coordinates. In particular, is an open subset of .
We say that the tropical chart is a tropical subchart of if and . We note that the definition of tropical chart here is different from the tropical charts in [CD12], §3.1, which consist of an analytic morphism to a split torus and a finite union of polytopes containing the tropicalization.
Proposition 4.16
The tropical charts on have the following properties:
- (a)
They form a basis on , i.e. for every open subset of and for every , there is a tropical chart with . We may find such a such that the open subset of is relatively compact.
- (b)
The intersection of tropical charts and is a tropical subchart of both.
- (c)
If is a tropical chart and if is a very affine open subset of with , then is a tropical subchart of .
Proof: To prove (a), we may assume that is a very affine scheme. A basis of is formed by subsets of the form with all and real numbers . Using the ultrametric triangle inequality, it is easy to see that we may choose the basis in such a way that for all . Note that is contained in the analytification of the very affine open subset of . It is obvious that is a tropical chart proving (a).
To prove (b), let us consider the moment map
Since is separated, it is easy to see that is a closed embedding and hence is very affine. We conclude that is a very affine open subset of . By definition of a tropical chart, (resp. ) is an open subset of (resp. ). Note that
is an open subset of . An easy diagram chase yields . Since refines the moment map , we deduce that is a tropical chart. This proves (b).
Finally, we prove (c). Let be the canonical affine homomorphism from 4.12. Then we have on . Since is a tropical chart, is an open subset of and . Using , we get for the open subset of . We conclude that is a tropical chart proving (c).