Tropical geometry
Grigory Mikhalkin
Contents
Chapter 1. Introduction 1
1. Overview 1
2. The tropical semiﬁeld T 2
3. The aﬃne space T
n
and the torus (T
×
)
n
≈ R
n
3
4. Integer aﬃne structures on smooth manifolds 4
5. Morphisms and isomorphisms of integer aﬃne manifolds 6
6. Examples of integer aﬃne surfaces 8
7. Integer aﬃne manifolds with corners 10
8. Tropical projective spaces 12
Chapter 2. Some (semi)algebraic notions 15
1. Tropical algebras 15
2. Examples 16
3. Spectra of tropical algebras 17
4. Quotient semiﬁelds 20
5. Aﬃne and convex functions in a tropical algebra 22
6. Aﬃne structure resulting from the semialgebraic data 23
7. Regular functions and tropical schemes 25
8. Regular maps 26
Chapter 3. Hypersurfaces and complete intersections in T
n
29
1. Integer aﬃne manifolds as tropical schemes 29
2. Hypersurfaces in T
n
29
3. Lines in the plane 32
4. Curves in the plane 38
5. Surfaces in TP
3
42
6. Complete Intersections 44
7. Balancing condition 44
Chapter 4. Tropical varieties 45
Chapter 5. Tropical equivalence 47
Bibliography 49
iii
CHAPTER 1
Introduction
1. Overview
Algebraic Geometry provides a uniform approach to some topologically
very distinct situations. As an example, let us consider a line in the aﬃne
2plane. Topologically this setup only makes sense if we ﬁx the ground ﬁeld,
i.e. the possible values for the coordinates in the 2plane. If the ground ﬁeld
is R we have the “most classical” situation: the plane is indeed a real plane
R
2
and the line is a real line R.
For the other choices of ground ﬁelds the topological picture is diﬀerent,
e.g. the complex plane C
2
is a 4manifold while over ﬁnite ﬁelds we do
not have any interesting topology at all. In the same time despite such
diﬀerences the behavior of lines remain the same. Namely, via any pair of
distinct points in the plane we can draw a unique line. Also, any pair of
lines intersect in a single point (unless they are parallel). This behavior is
dictated by the algebra of linear equations.
Figure 1. The three new intersection points are collinear
according to the Fano axiom
Some other properties of lines in the plane depend on the choice of the
ground ﬁeld. A famous example is the Fano axiom. Given any quadruple of
distinct points in the plane we may consider the triple of points obtained as
1
2 1. INTRODUCTION
the intersections of the pairs of lines corresponding to all possible choices of
two disjoint pairs among the four initial points. The Fano axiom states that
the resulting three points are collinear. This axiom clearly does not hold for
C or R, but it holds for the ﬁelds of characteristic 2.
When we pass to tropical geometry the ground ﬁeld gets replaced with
the tropical semiﬁeld T (which we introduce in the next section) with limited
arithmetics and algebra. E.g., it becomes no longer clear how to even deﬁne
the characteristic of T. Meanwhile, such geometric objects as points, lines,
etc. are perfectly welldeﬁned. In particular, the Fano axiom still holds in
tropical geometry. More general, in tropical geometry we may ﬁnd reﬂections
of properties from rather diﬀerent ﬁelds with diﬀerent algebraic origins.
Most algebraic constructions are obstructed by the absence of subtrac
tion in T. In the same time, geometry not only remains equally transparent,
but it gets more explicit and visual. The goal of this book is to justify this
statement.
2. The tropical semiﬁeld T
Definition 1.1. A commutative semiring is a set equipped with com
mutative and associative operations of addition and multiplication so that
the distribution law holds while the addition and multiplication operations
both have neutral elements. A commutative semiring R is called a semiﬁeld
if the nonzero elements of R form a group (denoted with R
×
) with respect
to multiplication.
Example 1.2. The nonnegative numbers R
≥0
equipped with the usual
addition and multiplication form a semiﬁeld. Its multiplicative group is the
group of positive numbers R
>0
.
The semiﬁeld introduced in the following deﬁnition is crucial for this
book.
Definition 1.3. The tropical semiﬁeld T is the set R∪{−∞} equipped
with the following two arithmetic operations called tropical addition and
tropical multiplication. If a, b ∈ T we set
“a +b” = max{a, b}
and
“ab” = a +b.
The quotation marks are used to signify that the arithmetic operations
we are referring to are tropical. It is easy to check that the usual commu
tativity, associativity and the distribution law hold in tropical arithmetics.
Namely, we have “a + b” = “b + a”, “(a + b) + c” = “a + (b + c)”, “ab” =
“ba”, “(ab)c” = “a(bc)” and “a(b + c)” = “ab + ac” for any a, b, c ∈ T. The
element −∞ = 0
T
is the additive zero while 0 = 1
T
is the multiplicative
3. THE AFFINE SPACE T
n
AND THE TORUS (T
×
)
n
≈ R
n
3
unit, “0
T
+ a” = max{−∞, a} = a, “1
T
b” = 0 + b = b, for any a ∈ T,
b ∈ T
×
= T {−∞}. In addition we have “ − ∞a” = −∞ for any a ∈ T.
However, in contrast with the classical addition the tropical addition is idem
potent:
“x +x” = x.
This property makes tropical subtraction impossible, T is only a semi
group with respect to addition. On the other hand, the nonzero elements
T
×
= T {−∞} form a group (isomorphic to R) with respect to multipli
cation and we have tropical division
“a/b” = a −b
as long as b = −∞.
Note that the semiﬁeld T has a natural (Euclidean) topology coming
from the identiﬁcation of T with the halfopen inﬁnite interval [−∞, +∞).
This topology is natural from the algebraic point of view. Indeed, the Eu
clidean topology on [−∞, +∞) is generated by the sets {x ∈ T  x > a} and
{x ∈ T  x < b} for a, b ∈ T
×
= (−∞, +∞).
Each inequality can be rephrased in agrebraic terms. Indeed the inequal
ity a ≤ b for a, b ∈ T is equivalent to the identity “a+b=b”.
3. The aﬃne space T
n
and the torus (T
×
)
n
≈ R
n
We deﬁne the tropical aﬃne nspace as a topological space by
T
n
= [−∞, +∞)
n
.
Accordingly, we deﬁne the ntorus there
(T
×
)
n
= (−∞, +∞) = R
n
⊂ T
n
.
This deﬁnition immediately gives the topology on T
n
. The algebro
geometric structure is given by regular functions on T
n
which come from
tropical polynomials.
Definition 1.4. A tropical polynomial f : T
n
→ T is a function given
by
f(x
1
, . . . , x
n
) = “
¸
j
1
,...,jn
a
j
1
...jn
x
j
1
1
. . . x
jn
n
”,
where a
j
1
...jn
∈ T, the indices j
k
are positive integers and the sum is ﬁnite.
Let us ﬁnd the geometric structure on T
n
that would enable us to dis
tinguish tropical polynomials from other continuous functions without a
reference to arithmetic operations in T. For that we restrict our attention
to the torus (T
×
)
n
.
4 1. INTRODUCTION
Note that if x ∈ T
×
then negative powers “x
−k
” = “
1
x
k
” = −kx also
make sense. Thus we also have the Laurent polynomials (T
×
)
n
→T deﬁned
by
“
¸
j
1
,...,jn
a
j
1
...jn
x
j
1
1
. . . x
jn
n
”,
where a
j
1
...jn
∈ T, j
k
∈ Z and the sum is still ﬁnite.
Each monomial
“a
j
1
...jn
x
j
1
1
. . . x
jn
n
” = j
1
x
1
+· · · +j
n
x
n
is an aﬃnelinear function in (T
×
)
n
= R
n
. Furthermore, the slope of this
function is (j
1
, . . . , j
n
) and thus it is integer. The geometric structure that
underlies such aﬃnelinear functions is the integer aﬃne structure.
4. Integer aﬃne structures on smooth manifolds
Definition 1.5. Let M be a smooth ndimensional manifold. An integer
aﬃne structure on M consists of an open covering U
α
and charts φ
α
: U
α
→
R
n
such that for each α, β the overlapping map φ
β
◦ φ
−1
α
can be obtained
as the restrictions of an integer aﬃnelinear transformation Φ
βα
: R
n
→R
n
.
Here a map f : R
n
→R
m
is called integer aﬃnelinear if it is a composition
of a Zlinear map R
n
→R
m
(i.e. a map given by m×n matrix with integer
values) and a translation by an arbitrary vector in R
m
. The map f is called
an integer aﬃnelinear transformation of R
n
if it is invertible in the class of
integer aﬃnelinear maps (note that the invertibility implies that m = n).
The manifold M equipped with such structure is called an integer aﬃne
manifold. As with all geometric structures of such kind we have the devel
oping map. Namely, if x ∈ U
α
⊂ M, y ∈ U
β
⊂ M and γ : [0, 1] → M is a
continuous path connecting x and y then we have the map Φ
γ
αβ
: R
n
→ R
n
deﬁned as follows.
The path γ([0, 1]) can be covered by a ﬁnite number of the charts U
α
j
,
j = 0, . . . k. We can make sure that U
α
j−1
∩ U
α
j
∩ γ([0, 1]) = ∅ for j > 0 so
that α
0
= α and α
k
= β. Then we deﬁne
Φ
γ
βα
= Φ
α
k
α
k−1
◦ · · · ◦ Φ
α
1
α
0
.
It is easy to see that Φ
γ
βα
depends only on α, β and γ but not on the choice
of U
α
j
. Furthermore, Φ
γ
βα
depends only on the relative homotopy class of
the path γ.
Recall that if we ﬁx x ∈ M then a point in the total space
˜
M of the
universal covering π :
˜
M → M corresponds to a pair (y, [γ]), where [γ] is
the relative homotopy class of a path from x to y. Thus if we ﬁx x and α
then we get a welldeﬁned map δ :
˜
M →R
n
by setting
δ(y, [γ]) = Φ
γ
βα
◦ Φ
α
,
where U
β
is chart containing y.
4. INTEGER AFFINE STRUCTURES ON SMOOTH MANIFOLDS 5
Definition 1.6. The map δ is called the developing map.
As it is easy to see the value δ(y, [γ]) ∈ R
n
does not depend on the
ambiguity in the choice of β. By construction, the developing map is always
an open embedding.
Definition 1.7. The integer aﬃne structure on a smooth manifold is
called complete if the developing map is proper.
Clearly, the product M × N of two integer aﬃne manifolds M and N
has a natural integer aﬃne structure.
Proposition 1.8. The product M × N is complete if and only if both
M and N are complete.
Proof. The proposition easily follows from the observation that the
universal covering of M × N can be obtained by taking the product of the
universal coverings for M and N.
For R
n
we have a notion of aﬃnelinear functions with integer slopes or
simply integer aﬃnelinear functions. These are the functions obtained from
linear maps R
n
→R deﬁned over Z (i.e. such that the image of the integer
lattice Z
n
⊂ R
n
is integer) after adding an arbitrary constant. Clearly, the
pullback of an integer aﬃnelinear function under an integer aﬃnelinear
map R
n
→R
n
is another integer aﬃnelinear function on R
n
.
Furthermore, for any open subset U of an integer aﬃne manifold M we
have a welldeﬁned notion of an integer aﬃnelinear function f : U →R. By
deﬁnition it is a function that corresponds to an aﬃnelinear function with
integer slope on R
n
in each chart. These functions correspond to tropical
monomials. While the choice of presentation as a tropical monomial depends
on the choice of chart, these functions always correspond to some tropical
monomials in any chart. Taking the maximal value of integer aﬃnelinear
function produces tropical (Laurent) polynomials. Thus geometrically, the
tropical structure on (T
×
)
n
may be rephrased as an integer aﬃnelinear
structure on R
n
.
Recall that the diﬀerential of the integer aﬃnelinear transformations
in R
n
is deﬁned over Z. Thus an integer tangent vector is mapped to an
integer tangent vector. Thus for any integer aﬃne (smooth) manifold M
and any point x ∈ M we have a welldeﬁned integer lattice in the tangent
space T
x
M. This lattice varies smoothly from point to point.
Conversely, if we have a smooth manifold with a coherent choice of
integer lattice in the tangent bundle then it does not necessarily come locally
from the tautological integer aﬃne structure on R
n
as this is a subject to
certain integrality condition. Locally such choice of lattice corresponds to
ﬁnding n linearly independent vector ﬁelds on R
n
. The integrality condition
is the (pairwise) commuting of these vector ﬁelds.
6 1. INTRODUCTION
Remark 1.9. We see that integer aﬃnelinear smooth manifolds locally
can be considered as examples of tropical varieties (as they locally coincide
with (T
×
)
n
. Similarly, smooth manifolds with a coherent (but not nec
essarily integrable) choice of integer lattice in the tangent bundle can be
considered as examples of almost tropical varieties.
5. Morphisms and isomorphisms of integer aﬃne manifolds
Let M and N be integer aﬃne varieties of dimensions m and n. A map
f : M → N is called an integer aﬃnelinear map (or just morphism of integer
aﬃnelinear varieties) if it is smooth and its diﬀerential maps any integer
vector tangent to M at any point x to an integer vector (tangent to N at
f(x)).
Consider a morphism f : M → N of integer aﬃnelinear varieties, a
point x ∈ M and any pair of charts U
α
∋ x, V
β
∋ f(x).
Proposition 1.10. The map ψ
−1
β
◦f ◦φ
α
is the restriction to the domain
where it is deﬁned (i.e. to U
α
∩ f
−1
(V
β
)) of an integer aﬃne linear map
R
m
→R
n
.
Proof. It suﬃces to show that if f : R
m
→ R
n
is a map whose diﬀer
ential takes integer vectors to integer vectors then f is integer aﬃne linear.
Applying a translation if needed we may assume that f takes the origin of
R
m
to the origin of R
n
.
We claim that the diﬀerential (df)
0
of such f at the origin coincides with
the map itself (after the natural identiﬁcation of R
m
with the tangent space
at its origin). The integrality assumption assures that (df)
0
is deﬁned over
Z. By the continuity argument the integrality assumption also implies that
(df)
x
= (df)
0
for every x ∈ R
m
.
Let v ∈ R
n
be any vector. It can be decomposed into a sum of integer
vectors v
j
, v =
¸
a
j
v
j
with a
j
∈ R. This allows to connect 0 and v with the
broken path such that each of its segment is parallel to one of the integer
vectors v
j
. Therefore, we have
f(v) =
¸
a
j
(df)
0
(v
j
) = (df)
0
(v).
A map f : M → N is called an isomorphism (or a symmetry) of integer
aﬃne manifolds if it is invertible and both f and f
−1
are morphism. Then
we say that M and N are isomorphic as integer aﬃne manifolds.
All isomorphisms of integer aﬃne manifolds M form a group. If the
quotient M/G by a subgroup G of this group is a manifold (which is the
case if this subgroup acts in a properly discontinous fashion, i.e. every point
x admits a neighborhood U ∋ x such that all translates by the elements of
G are disjoint) then it gets a natural integer aﬃne structure from M.
5. MORPHISMS AND ISOMORPHISMS OF INTEGER AFFINE MANIFOLDS 7
Clearly, R
n
is an aﬃne integer manifold tautologically. The group of
its symmetries is the group of all integer aﬃnelinear transformations of
R
n
. The action of the whole group is not properly discontinous, so we need
to restrict to a subgroup. The easiest properly discontinous subgroup is
the lattice generated by translation in linearly independent directions. But
there are other choices of subgroups, also using nontrivial linear parts (from
GL
n
(Z)).
Example 1.11. Consider the following examples of integer aﬃne man
ifolds obtained as the quotients of R
2
. Let M be the quotient of the plane
R
2
by the subgroup Λ generated by the vectors
a
b
,
c
d
.
For any choice of a, b, c, d ∈ R with ad−bc = 0 the resulting quotients are
integer aﬃne manifolds. All of them are diﬀeomorphic (and diﬀeomorphic
to S
1
×S
1
). However they are not all isomorphic as integer aﬃne manifolds.
E.g. if b = 0 then M is foliated by closed “horizontal” circles obtained
as the quotient (t, s), where for each circle s ∈ R is ﬁxed while t ∈ R varies.
The condition b = 0 ensures that the points (t, s) and (t + a, s) coincide so
that we get a closed circle.
closed “horizontal” circles
form a ﬁbration
no rational slope curve
is closed
Figure 2. Diﬀerent integer aﬃne structures n S
1
×S
1
.
Of course, being “horizontal” is not an intrinsic condition in M and
depends on the choice of chart to R
2
. But there is also an intrinsic property
that holds for these circles.
Definition 1.12. Let C ⊂ M be a curve. We say that it has rational
slope if it is tangent to an integer vector at its every point.
Alternatively we may deﬁne such curves as those which have rational
slope in each chart. This property does not depend on the choice of the
charts while being “horizontal” in one chart ensures rational slope in others.
Definition 1.13. Let C ⊂ M be a curve with rational slope and v ∈
T
x
C be a vector tangent to x ∈ C. We say that v is a primitive vector if it
is integer in T
x
M and cannot be presented as a nontrivial positive integer
multiple of another integer tangent vector.
8 1. INTRODUCTION
Proposition 1.14. For any x ∈ C the primitive tangent vector is unique
up to sign.
Proof. All vectors tangent to C form a 1dimensional real vector space
while the integer vectors form a lattice isomorphic to Z ⊂ R.
With the help of the primitive vectors we may deﬁne intrinsic length of
a curve C with rational slope. Indeed, a 1form α on C that takes value ±1
on primitive vectors is unique up to sign. Let γ ⊂ C be an arc on C.
Definition 1.15. The (intrinsic) length of γ is the integral
γ
α.
In charts the intrinsic length can be obtained by taking the Euclidean
length of γ and dividing it by the Euclidean length of a primitive vector
parallel to γ.
We return to Example 1.11. If b = 0 we can measure the length of the
“horizontal” circles. Clearly, all their lengths coincide and equal to a. If
they are the only closed curves with rational slope on M then a is the
isomorphism invariant.
We may also choose a, b, c, d ∈ R linearly independent over Q. Then no
circle in M can have rational slope. Indeed, suppose that on the contrary
we can ﬁnd such a circle and it is parallel at its every point (in a chart
obtained by reversing the quotient projection) to an integer vector
m
n
∈
R
2
. Then a multiple of
m
n
is proportional to an integer linear combination
j
a
b
+k
c
d
, j, k ∈ Z. But then n(ja+kc) = m(jb+kd) which contradicts
to the linear independence over Q.
6. Examples of integer aﬃne surfaces
Example 1.16. Let R : R
2
→ R
2
be the gliding reﬂection obtained by
the composition of the reﬂection at the xaxis with a translation by
a
0
,
a > 0. Let B be a translation by
0
d
, d > 0.
The quotient of R
2
by the properly discontinous subgroup G of symme
tries generated by R and B is a Klein bottle. Note that for this example we
have foliations both by horizontal and vertical circles. All of them, except
for two horizontal “core” circles have the same (intrinsic) length, equal to
2a and d respectively. The two horizontal “cores” have lengths equal to a,
see Figure 3 for one of the “cores”, the other is the result of identiﬁcation
of the horizontal sides of the rectangle.
6. EXAMPLES OF INTEGER AFFINE SURFACES 9
core horizontal cirle of length a
horizontal cirles of length 2a
Figure 3. A Klein bottle with horizontal circles.
Note also that the vertical circles are dual to the ﬁrst StiefelWhitney
class. Therefore d is an isomorphism invariant of M. The manifold M can
be obtained from the a × d rectangle by identifying the oriented opposite
sides, cf. Figure 3. The conjugation of G by an element from GL
2
(Z) results
in replacing the rectangle by a parallelogram, but the slopes of the sides of
this parallelogram will still have rational slope. Of course, such conjugation
does not change the isomorphism type of M.
So far our examples look very similar to examples of surfaces with Eu
clidean structure (cf. e.g. [49], [67]). Consider now a radically diﬀerent
example an integer aﬃne structure on a torus. First, we construct a non
trivial integer aﬃne annulus.
Example 1.17. Let A : R
2
→R
2
be a map obtained as the composition
of a translation by
0
d
, d > 0, and the linear transformation of R
2
deﬁned
by
1 1
0 1
. Let R be the quotient of R
2
by the group generated by A.
The surface R is an integer aﬃne annulus that is diﬀerent from the
quotient of R
2
by the group generated by any translation. Inside R we have
immersed curves with rational slope that have selfintersections as shown on
Figure 4. We identify the top side of the strip with the bottom so that the
corresponding bases match.
Note that the shear transformation is the only possible linear part for
an orientationpreserving deck transformation R
2
→R
2
corresponding to an
integer aﬃne linear transformation as shown in the following proposition.
Proposition 1.18. If an integer aﬃne linear transformation A : R
2
→
R
2
is ﬁxed point free and orientationpreserving then its linear part L has
both eigenvalues equal to 1.
Proof. Let λ and µ be the eigenvalues of L. Since L ∈ SL
2
(Z) and
preserves orientation we have λµ = 1. If λ, µ = R then λ = ¯ µ as L is real.
Therefore λ
2
= µ
2
 = 1 and L is an orthogonal matrix in some basis so
that A is a metric preserving transformation. From Euclidean planimetry
we know that A must be a translation.
10 1. INTRODUCTION
Figure 4. An integer aﬃne annulus and two curves with
rational slope there.
If λ, µ ∈ R with µ =
1
λ
= λ then we may choose the coordinates in R
2
so that L is given by (x, y) → (λx,
1
λ
y). Suppose that the translational part
of A is given by
a
b
. It suﬃces to ﬁnd (x, y) such that λx − x = a and
1
λ
y −y = b. But these linear equations clearly have solutions if λ = 0.
Example 1.19. Let T be the quotient of R
2
by the group generated by
the transformation A from Example 1.17 and a translation by
a
0
. This
is a compact surface diﬀeomorphic to the torus but not isomorphic to any
quotient of R
2
by a lattice of translations. Inside T we have immersed curves
with rational slope that have selfintersections as in the case of Example 1.17.
Example 1.20. Let K be the Klein bottle obtained as the quotient of
R
2
by the group generated by the transformation A from Example 1.17 and
the transformation R from Example 1.16. This group also acts in a properly
discontinuous manner so K is an integer aﬃne surface. Just like the torus
T from Example 1.19 the Klein bottle K has immersed curves with rational
slope that have selfintersections.
In fact, Examples 1.19 and 1.20 admit the same tiling by fundamental
domains in R
2
shown in Figure 5.
Remark 1.21. Any integer aﬃne manifold can also be considered as a
real aﬃne manifold as we have embedding GL
n
(Z) ⊂ GL
n
(R). See [5], [36],
[47] for a discussion of real aﬃne structures, particularly on a torus. See
also [16] for a discussion of aﬃne structures with singularities.
7. Integer aﬃne manifolds with corners
While integer aﬃne manifolds are modeled on open sets in R
n
the trop
ical aﬃne space T
n
has boundary and corners.
7. INTEGER AFFINE MANIFOLDS WITH CORNERS 11
Figure 5. Tiling of R
2
by fundamental domains for Exam
ples 1.19 and 1.20.
Definition 1.22. Let x = (x
1
, . . . , x
n
) be a point in T
n
= [−∞, +∞)
n
.
We call the sedentarity s(x) of x the number of coordinates x
j
equal to −∞.
The tropical aﬃne space T
n
is a manifold near its point x if and only if
x has sedentarity 0.
Let Φ : R
n
→ R
m
be an integer aﬃnelinear map. Let L be the linear
part of Φ which can be viewed as an (integer) m×n matrix. Let x ∈ T
n
R
n
be a point of positive sedentarity in T
n
. The image Φ(x) still makes sense
as a point in T
m
if whenever x
j
= −∞ the whole jth row of the matrix L
is nonnegative. Here we use the convention “a(−∞)” = −∞ if a > 0 and
“0(∞)” = 0.
This gives us partiallydeﬁned extensions
¯
Φ : T
n
T
m
of integer aﬃnelinear maps R
n
→ R
m
. The map
¯
Φ is continuous on the
domain of its deﬁnition. We treat such maps as integer aﬃne linear maps
between aﬃne tropical spaces. They allow us to extend the notion of integer
aﬃne structure to a larger class of spaces almost by repeating Deﬁnition 1.5
Definition 1.23. Let X be a topological space. We say that X is an
integer aﬃne manifold with corners if X is enhanced with an open covering
U
α
and charts φ
α
: U
α
→ T
n
such that for each α, β the overlapping map
φ
β
◦ φ
−1
α
can be obtained as the restrictions of a (partially deﬁned) integer
aﬃnelinear map
¯
Φ
βα
: T
n
T
n
that is deﬁned everywhere of φ
α
(U
α
).
If x ∈ U
α
⊂ X then we deﬁne its sedentarity as the sedentarity of its
image φ
α
(x) ∈ T
n
.
Proposition 1.24. The sedentarity s(x) of a point x ∈ X does not
depend on the choice of the chart U
α
.
Proof. Suppose, on the contrary, that s(
¯
Φ
βα
(x)) < s(x) = k for x ∈
T
n
. Without the loss of generality we may assume that x
1
= · · · = x
k
= −∞.
12 1. INTRODUCTION
Then we know that the top k rows of the matrix giving the linear part L of
¯
Φ
βα
must consist of nonnegative (integer) numbers. Since L is invertible
we may also assume without the loss of generality that the top k ×k minor
is not degenerate. But then the ﬁrst k coordinates of
¯
Φ
βα
(x) are all equal
to −∞ and this supplies a contradiction.
Let X
s
be the locus of points of sedentarity s in an ndimensional integer
aﬃne manifold with corners X.
Proposition 1.25. The space X
s
is a disjoint union of integer aﬃne
manifolds of dimension n −s (without boundary or corners).
Proof. Restrictions of the overlapping maps
¯
Φ
βα
to the coordinate (n−
s)planes in T
n
(those deﬁned by x
j
1
= · · · = x
js
= −∞) provides the
required integer aﬃne structure.
Definition 1.26. The integer aﬃne structure on a manifold with corners
is called complete if every component of X
s
is a complete integer manifold
for each s = 0, . . . , n.
Let X and Y be two integer aﬃne manifolds with corners.
Definition 1.27. A map f : X → Y is called a morphism if for every
x ∈ X there exists charts U
X
α
∋ x, U
Y
β
∋ f(x) and a map Φ : T
n
→ T
m
,
T
n
⊃ U
X
α
, T
m
⊃ U
Y
β
, such that f(t) = (φ
Y
β
)
−1
◦ Φ ◦ φ
X
α
.
Note that this is a straightforward extension of the deﬁnition of mor
phisms of manifolds without corners.
Clearly, any open subset U of a manifold with corner X is itself a man
ifold with corners (though not necessarily complete even in the case when
the ambient manifold with corners X is complete).
Definition 1.28. The (tropical) monomial on U is any morphism U →
T.
Proposition 1.29. If U is complete then for any monomial κ : U →T
we have κ(U) ⊃ R.
Proof. The map κ can be lifted to a morphism from the universal
covering
˜
U →T. Its image has to contain R as
˜
U contains R
n
.
Note that T
n
has the tautological structure of an integer aﬃne manifold
with corners. Furthermore, we can glue several copies of T
n
together to get
compact integer aﬃne manifolds with corners. For us the most important
example is that of the tropical projective space.
8. Tropical projective spaces
Consider the set
TP
n
= T
n+1
{0
T
n+1 }/ ∼
8. TROPICAL PROJECTIVE SPACES 13
where 0
T
n+1 = (−∞, . . . , −∞) is the origin in T
n+1
and we set (x
0
, . . . , x
n
) ∼
(y
0
, . . . , y
n
) if there exists λ ∈ T
×
such that x
j
= “λy
j
” = λ + y
j
for any
j = 0, . . . , n. Clearly the set TP
n
gets a natural topology of the quotient.
Furthermore, it admits a natural structure of an integer aﬃne manifold with
corners. As usual, we use the homogeneous coordinate notations x = [x
0
:
· · · : x
n
] ∈ TP
n
to denote the equivalence class of (x
0
, . . . , x
n
).
To see that we cover TP
n
with n + 1 open charts
U
j
= {x ∈ TP
n
 x
j
= 0
T
= −∞},
j = 0, . . . , n, (φ
j
(x))
k
= “
x
k
x
j
” = x
k
− x
l
, k = j.Here (φ
j
(x))
k
denotes the
kth coordinate of the image φ
j
(x) and the target of φ
j
is the hyperplane
T
n
⊂ T
n+1
given by {x ∈ T
n+1
 x
j
= 1
T
= 0}.
The overlapping maps
¯
Φ
jk
: T
n
T
n
, j = k are given by
(
¯
Φ
jk
)
l
= “
x
l
x
k
x
j
” = x
l
+x
k
−x
j
.
Clearly
¯
Φ
jk
is an integer aﬃne map deﬁned on {x
j
= −∞} ⊂ T
n
.
Proposition 1.30. The space TP
n
is homeomorphic to the nsimplex
Σ
n
so that a point inside a kface of Σ
n
corresponds to a point of sedentarity
n − k. Furthermore, the integer aﬃne structure induced in the interior of
each kface is isomorphic to the tautological integer aﬃne structure on R
k
.
Proof. The map
x → (
x
1
x
0
 +· · · +x
n

, . . . ,
x
n
x
0
 +· · · +x
n

)
provides the required homeomorphism to the standard simplex in R
n
≥0
(cut
by the halfspace x
1
+· · · +x
n
≤ 1).
Similarly to Proposition 1.8 we get the following statement.
Proposition 1.31. If X and Y are integer aﬃne manifolds with corners
then X × Y is also an integer aﬃne manifold with corners. Furthermore,
X ×Y is complete if and only if both X and Y are complete.
Remark 1.32. In a similar way we may construct tropical counterparts
of more general toric varieties. A complex smooth toric variety is obtained
by gluing several copies of aﬃne spaces C
n
(or, more generally, products of
aﬃne spaces C
k
with tori (C
×
)
n−k
) by maps such that each coordinate is
given by a monomial.
The tropical counterparts are obtained by gluing copies of T
k
× T
n−k
by the maps given by the corresponding tropical monomials. As in the
case with projective space there is a sedentaritypreserving homeomorphism
with the corresponding polyhedron (see e.g. [15]). E.g. Figure 6 shows the
tropical plane blown up at 6 points which is diﬀeomorphic (as a manifold
with corners) to a hexagon.
14 1. INTRODUCTION
Figure 6. The tropical projective plane and the tropical
projective plane blown up at three points. Interior of both
polygons are isomorphic to the complete aﬃne space R
2
with
the tautological integer aﬃne structure.
CHAPTER 2
Some (semi)algebraic notions
1. Tropical algebras
Definition 2.1. A Tcone is a set V with a choice of an element O ∈ V
called the origin equipped with a product operation
T ×V → V, (a, v) → “av”,
a ∈ T, v ∈ V , such that “(ab)v” = “a(bv)” for any a, b ∈ T, v ∈ V ,
“av” = “bv” if a = b and “v0
T
” = O.
Definition 2.2. A tropical algebra A is a semiring (recall according
to Deﬁnition 1.1 A has an additive zero 0
A
∈ A and a multiplicative unit
1
A
∈ A) equipped with a Tcone structure compatible with the semiring
operations, i.e. such that “a(fg)” = “(af)g” and O = 0
A
, subject to the
following additional property. For any f, g, h ∈ A if “fg” = “fh” for f, g, h ∈
A then either we have equality g = h or the element f is a zero divisor, i.e.
there exists
˜
f ∈ A such that “f
˜
f” = 0
A
.
Proposition 2.3. There is a natural embedding
ι
A
: T ⊂ A
which respects the semiring addition and multiplication: ι
A
(“a + b”) =
“ι
A
(a) +ι
A
(b)”, ι
A
(“ab”) = “ι
A
(a)ι
A
(b)”, ι
A
(−∞) = 0
A
and ι
A
(0) = 1
B
.
Conversely if A is semiring and ι
A
: T ⊂ A is such an embedding then
A is a tropical algebra as long as “0
A
f” = 0
A
and “ι
A
(a)f” = ι
A
(b)f for
any f ∈ A and a = b ∈ T.
Implicitly using this proposition we identify T with its image in A. In
particular, we have 0
A
= −∞ ∈ A and 1
A
= 0 ∈ A.
Proof. Deﬁne ι
A
(a) = “a1
A
”. Note that ι
A
is an embedding since A
is a cone. We have
ι
A
(“a +b”) = “(a +b)1
A
” = “a1
A
+b1
A
” = “ι
A
(a) +ι
A
(b)”
and
ι
A
(“ab”) = “(ab)1
A
” = “a(b1
A
)” = “(a1
A
)(b1
A
)” = “ι
A
(a)ι
A
(b)”.
To check the converse statement we note that ι
A
gives a Tcone structure
on A by “af” = “ι
A
(a)f”.
15
16 2. SOME (SEMI)ALGEBRAIC NOTIONS
Definition 2.4. Let A and B be two tropical algebras. A map φ : A →
B is called a homomorphism of tropical algebras (or just a Thomomorphism)
if for any a, b ∈ A we have φ(“a+b”) = “φ(a) +φ(b)”, φ(“ab”) = “φ(a)φ(b)”
and, in addition, φ is identity on T, i.e. the diagram
T
ι
A
//
ι
B
?
?
?
?
?
?
?
A
φ
B
is commutative. As usual, an isomorphism is an invertible homomorphism;
an epimorphism is a surjective homomorphism and a monomorphism is an
injective homomorphism.
Definition 2.5. A tropical algebra B is called an integral domain if it
does not have zero divisors, i.e. for any f, g ∈ B such that “fg” = 0
B
we
have either f = 0
B
or g = 0
B
.
2. Examples
Example 2.6. Consider the semiring
T[x] = {“
k
¸
j=0
a
j
x
j
”  a
j
∈ T, k ∈ N ∪ {0}}
of formal tropical polynomials in one variable x. These polynomials can be
added and multiplied according to formal polynomial laws (recall that −∞
is our additive zero) and form The embedding ι : T ⊂ T[x] is tautological
a → a.
Similarly, the semiring T[x
1
, . . . , x
n
] of formal tropical polynomials in n
variables
“
¸
(j
1
,...,jn)∈J
a
j
1
...jn
x
j
1
1
. . . x
jn
n
”,
where a
j
∈ T and J is a ﬁnite subset of (N ∪ {0})
n
, is another example
of tropical algebra. For convenience we will use multiindex notations for
multivariable monomials: if x = (x
1
, . . . , x
n
) ∈ T
n
and j = (j
1
, . . . , j
n
) ∈ Z
n
then
x
j
= x
j
1
1
. . . x
jn
n
.
Example 2.7. Consider the tropical algebra O(T
n
) of functions
T
n
→T, x → f(x),
where f ∈ T[x
1
, . . . , x
n
] and x = (x
1
, . . . , x
n
) ∈ T
n
. The addition and
multiplication on O(T
n
) are pointwise tropical addition and multiplication,
while constant functions give the embedding T → O(T
n
). Elements of O(T
n
)
are called regular functions on T
n
.
3. SPECTRA OF TROPICAL ALGEBRAS 17
The tautological map
τ : T[x
1
, . . . , x
n
] → O(T
n
)
is an epimorphism. Note that τ
−1
(−∞) = {−∞}. Nevertheless, τ is not a
monomorphism (unless n = 0). E.g.
“0x
2
1
+ax
1
+ 0” = “0x
2
1
+ 0”
whenever a ≤ 0 ∈ T. Indeed, in this case we have (depending on x
1
∈ T)
either “ax
1
” = x
1
+a ≤ 2x
1
= “0x
2
1
” or “ax
1
” ≤ 0.
Definition 2.8. A tropical algebra A is called ﬁnitely generated if there
exist f
1
, . . . , f
n
∈ A such that any f ∈ A can be presented in the form
f = “
n
¸
j=1
a
j
f
j
”.
The elements f
1
, . . . , f
n
are called generators of A.
Equivalently, A is ﬁnitely generated if there exists an epimorphism
T[x
1
, . . . , x
n
] → A
for some n ∈ N.
Example 2.9. Consider the algebra T[x
1
, . . . , x
n
, x
−1
1
, . . . , x
−1
n
] of Lau
rent polynomials in n variables “
¸
j∈J
a
j
x
j
”, where a
j
∈ T and J is a ﬁ
nite subset of Z
n
. This algebra is ﬁnitely generated by 2n generators
x
1
, . . . , x
n
, x
−1
1
, . . . , x
−1
n
.
3. Spectra of tropical algebras
Let A be a tropical algebra. Let B
1
and B
2
be two other tropical algebras
and φ
j
: A → B
j
be two epimorphisms.
Definition 2.10. The maximal spectrum Spec
m
(A) is the set of all
Thomomorphisms A →T.
The spectrum Spec(A) is the set of all epimorphisms A → B up to the
equivalence above, where B is an integral domain.
Example 2.11. We have
Spec
m
(T) = Spec(T) = {pt},
the only tropical epimorphism of T to another tropical algebra is the identity
T →T.
Definition 2.12. If f ∈ A and x ∈ Spec
m
(A) then we deﬁne the value
f(x) ∈ T as the image of f under the epimorphism x : A →T.
If U ⊂ Spec
m
(A) we denote
Funct(U) = {g : U →T  ∃f ∈ A : ∀x ∈ U g(x) = f(x)}.
18 2. SOME (SEMI)ALGEBRAIC NOTIONS
Pointwise addition and multiplication turn Funct(U) to a tropical algebra.
Clearly we have the natural evaluation epimorphism A → Funct(U).
Proposition 2.13. For any a ∈ T and x ∈ Spec
m
(A) we have
ι
A
(a)(x) = a,
thus the image of ι
A
corresponds to the constant functions on Spec
m
(A).
Proof. Since x : A → T is a homomorphism of tropical algebras it is
identity on T.
Definition 2.14. The evaluation epimorphism A → Funct(Spec
m
(A))
is called the reduction epimorphism. We say that the tropical algebra A is
reduced if the reduction epimorphism is an isomorphism.
Let a : A → B be a homomorphism of tropical algebras.
Definition 2.15. The induced map
a
∗
: Spec
m
(B) → Spec
m
(A)
is the map which takes an epimorphism x : B →T to x ◦ a : A →T.
Since a is an epimorphism of tropical algebras, so is x◦a. In particular,it
implies that x ◦ a maps onto T.
Proposition 2.16. If a : A → B is an epimorphism of tropical algebras
then a
∗
is an injection.
Proof. If x = x
′
: B → T then there exists f ∈ B such that x(f) =
x
′
(f). But then x(a(g)) = x
′
(a(g)) for any g ∈ A such that a(g) = f.
Example 2.17. Any x ∈ Spec
m
(A) is an epimorphism A → T. It
induces an embedding Spec
m
(T) ⊂ Spec
m
(A) (cf. Example 2.11) that cor
responds to the point x.
More generally, we have the following inclusions corresponding to such
embeddings when we pass to considerations of the full spectrum Spec(A).
Definition 2.18. If x ∈ Spec
m
(A), x : A → T, and F ∈ Spec(A),
F : A → B, we say that x is contained in F if x is contained in the image
F
∗
: Spec
m
(B) → Spec
m
(A). In other words, x is contained in F if there
exists y ∈ Spec
m
(B), y : B →T, such that x = y ◦ F.
Thus F deﬁnes a subset of Spec
m
(A). Clearly, this subset can be natu
rally identiﬁed with Spec
m
(B).
Definition 2.19. A subset X ⊂ Spec
m
(A) is called a basic closed set
if every tropical epimorphism Funct(X) → T corresponds to a point of
X. In other words, if x : Funct(X) → T is a tropical epimorphism then
the composition of the evaluation epimorphism A → Funct(X) and x is
contained in Z.
3. SPECTRA OF TROPICAL ALGEBRAS 19
In other words X is closed if the evaluation epimorphism A → Funct(X)
deﬁnes X (and not a larger set).
Proposition 2.20. An intersection of basic closed sets in Spec
m
(A) is
a basic closed set.
Proof. Suppose that X =
¸
j
X
j
and all X
j
⊂ Spec
m
(A) are basic
closed sets. Any tropical epimorphism x : Funct(X) → T can be composed
with the restriction epimorphism Funct(X
j
) → Funct(X). Therefore, the
composition of the evaluation epimorphism A → Funct(X) and x belongs
to X
j
for every j.
Recall that any collection of subsets deﬁne a topology as a prebasis. We
apply this construction in the following deﬁnition.
Definition 2.21. A set X ⊂ Spec
m
(A) is called closed if it can be
presented in the form
X =
¸
α∈J
X
α
,
where J ∋ α is any parameterizing set and each X
α
is the union of a ﬁnite
number of basic closed sets.
It follows immediately from this deﬁnition that the intersection of any
number of closed sets is closed and that the union of a ﬁnite number of
closed sets is open as well. Furthermore, an empty set is closed as the
parameterizing set J can be empty. The whole set Spec
m
(A) is an example
of a basic open set as it is presented by the identity epimorphism A → A.
Thus Deﬁnition 2.21 gives a topology on Spec
m
(A).
A set U ∈ Spec
m
(A) is called open if Spec
m
(A) U is a closed set. We
refer to this topology as the spectrum topology on Spec
m
(A) to distinguish
it from a diﬀerent topology (the Zariski topology) which we introduce later
on.
Proposition 2.22. The spectrum topology on Spec
m
(T[x
1
, . . . , x
n
]) co
incides with the Euclidean topology on [−∞, +∞)
n
. Furthermore any closed
set in the Euclidean topology is a basic closed set in the spectrum topology.
Proof. If F ⊂ Spec
m
(T[x
1
, . . . , x
n
]) is a basic closed set then it corre
sponds to an epimorphism T[x
1
, . . . , x
n
] → A. Consider Spec
m
(A). As each
Thomomorphism A → T also gives a Thomomorphism T[x
1
, . . . , x
n
] → T
by composition we have the identiﬁcation of Spec
m
(A) and F. Since all
tropical polynomials are continous functions any accumulation point of F
also deﬁnes a Thomomorphism A →T. Thus F must be closed.
Conversely, if F ⊂ [−∞, +∞)
n
is closed then we may consider the re
striction homomorphism T[x
1
, . . . , x
n
] → Funct(F). If y / ∈ F then we may
have two tropical polynomials f, g such that f(y) = g(y) but such that
20 2. SOME (SEMI)ALGEBRAIC NOTIONS
f(x) = g(x) for any x ∈ F. Thus a point y ∈ [−∞, +∞)
n
does not give a
homomorphism from Funct(F) unless y ∈ F.
Similarly we get the following proposition.
Proposition 2.23. The spectrum topology on Spec
m
(T[x
1
, . . . , x
n
, x
−1
1
, . . . , x
−1
n
])
coincides with the Euclidean topology on R
n
. Furthermore any closed set in
the Euclidean topology is a basic closed set in the spectrum topology.
4. Quotient semiﬁelds
As in classical Commutative Algebra if A is a tropical algebra A which
is an integral domain then we can make a semiﬁeld Q ⊃ T out of T by
allowing fractions.
Lemma 2.24. If A is a tropical integral domain and f
j
, g
j
∈ A, g
j
=
0
A
, j = 1, 2, 3, are such that “f
1
g
2
” = “f
2
g
1
” and “f
2
g
3
” = “f
3
g
”
2 then
“f
1
g
3
” = “f
3
g
1
”.
Proof. Take a product of the lefthand and the righthand sides of our
hypotheses “f
1
g
2
” = “f
2
g
1
” and “f
2
g
3
” = “f
3
g
2
”. We get
“f
1
g
2
f
2
g
3
” = “f
2
g
1
f
3
g
2
”.
Since A is a tropical algebra either the statement of the lemma holds or
“f
2
g
2
” is a zero divisor (cf. Deﬁnition 2.2). Since A is an integral domain
and g
2
= 0
A
we have f
2
= 0
A
. Then, in turn, f
1
= 0
A
and f
3
= 0
A
which
also veriﬁes the statement of the lemma.
Definition 2.25. The quotient semiﬁeld Q = Rat(A) of a tropical inte
gral domain A is the set of pairs (f, g), f, g ∈ A, g = 0
A
up to the following
equivalence relation (cf. Lemma 2.24 (f
1
, g
1
) ∼ (f
2
, g
2
) if
f
1
g
2
= f
2
g
1
∈ A.
We equip Q with operations of addition
“(f
1
, g
1
) + (f
2
, g
2
)” = (“f
1
g
2
+f
2
g
1
”, “g
1
g
2
”)
and multiplication
“(f
1
, g
1
)(f
2
, g
2
)” = (“f
1
g
1
”, “f
2
g
2
”).
It is easy to see that the equivalence class of the results of these operations
does not change if we replace (f
j
, g
j
), j = 1, 2, with an equivalent pair.
In accordance with the classical case we denote (f, g) ∈ Q with “
f
g
”.
Elements of the semiﬁeld Q are called rational functions associated with A.
From now on we suppose that a tropical algebra A is an integral domain
and Q = Rat(A) is its quotient semiﬁeld.
4. QUOTIENT SEMIFIELDS 21
Proposition 2.26. Q is a semiﬁeld that contains A as a subsemiring.
The embedding T ⊂ A ⊂ Q makes Q into a tropical algebra. The map
q : A → Q, q(f) = “
f
1
A
” is a monomorphism of tropical algebras.
Proof. Clearly, Q is a semiring since A is a semiring. Since we have
the inversion operation
“
1
f/g
” = “
g
f
”
Q is a semiﬁeld. If “
a
1
A
” is equivalent to “
b
1
A
”, a, b ∈ A then, by deﬁnition,
a = b. In particular, this gives an embedding T ⊂ Q which makes Q a
tropical algebra and q a tropical algebra monomorphism.
Proposition 2.27. Any homomorphism h : A → B of tropical algebras
naturally extends to a homomorphism H : Rat(A) → Rat(B).
Proof. We set H(“
f
g
”) = “
h(f)
h(g)
”.
The homomorphism q from Proposition 2.26 deﬁnes a map
q
∗
: Spec
m
(Q) → Spec
m
(A)
by taking x : Q →T to x ◦ q : A →T.
Definition 2.28. A point x ∈ Spec
m
(A) is called ﬁnite if x ∈ q
∗
(Spec
m
(A)).
We denote the set of all ﬁnite points in Spec
m
(A) with (Spec
m
(A))
◦
.
Proposition 2.29. Nonzero elements of A have ﬁnite values at ﬁnite
points of the spectrum. I.e. if x ∈ Spec
m
(A) is ﬁnite and f = −∞∈ A then
f(x) = −∞ ∈ T.
Proof. Since a homomorphism x : A → T can be factorized through
q : A → Q it can be extended to “
1
A
f
”. We have “
1
A
f
”(x) = “
1
T
f(x)
” ∈ T,
therefore f(x) ∈ T
×
.
Example 2.30. Consider the tropical algebra T[x
1
, . . . , x
n
] from Exam
ple 2.7. Its quotient semiﬁeld coincides with the quotient semiﬁeld of the
algebra T[x
1
, . . . , x
n
, x
−1
1
, . . . , x
−1
n
] as “
x
−1
j
1
T
” ∼ “
1
T
x
j
” (recall that 1
T
= 0). We
denote the resulting semiﬁeld in these cases with T(x
1
, . . . , x
n
) and call its
elements tropical rational functions in n variables.
Proposition 2.31. If Φ : A → B is a homomorphism then
Φ
∗
((Spec
m
(B))
◦
) ⊂ (Spec
m
(A))
◦
.
Proof. By Proposition 2.27 we have the induced map of the spectra
of Q
A
and Q
A
Spec
m
(B)
◦
→ Spec
m
(A)
◦
that agrees with Φ
∗
since H is an
extension of h. The required map is induced by the composition A → Q
A
→
Q
B
.
22 2. SOME (SEMI)ALGEBRAIC NOTIONS
5. Aﬃne and convex functions in a tropical algebra
Definition 2.32. An element f in a tropical algebra A ⊃ T is called
a primitive aﬃne function if f = −∞ and whenever we have f = a + b,
a, b ∈ A, and we have either f = a or f = b.
Recall that Q
∗
= Q {0} is an abelian group with respect to tropical
multiplication. Denote with Aﬀ(A) the subgroup of Q
∗
generated by all
primitive aﬃne functions in A ⊂ Q.
Definition 2.33. Elements of Aﬀ(A) are called aﬃne functions associ
ated with A.
An element of Q is called convex if it is a tropical sum of elements from
Aﬀ(A) ⊂ Q. All convex functions form a semiring Conv(A) ⊂ Q.
Proposition 2.34. If f ∈ Aﬀ(A) and a, b ∈ Conv(A) are such that
f = a +b then either f = a or f = b.
Proof. There exists a primitive aﬃne functions h ∈ A such that “f +
h” ∈ A is a primitive aﬃne function while “a + h”, “b + h” ∈ A. We have
“a+h+b +h” = “a+b +h” = “f +h” which contradicts to the primitivity
of “f +h”.
Definition 2.35. We say that a tropical algebra A is tame if the fol
lowing conditions hold:
• for every c ∈ T
∗
the image ι
A
(c) ∈ A is a primitive aﬃne function
(we call such functions constant) so that T
∗
⊂ Aﬀ(A) is a subgroup;
• the quotient group Aﬀ(A)/T
∗
is a free abelian group of ﬁnite rank;
• the subset Aﬀ(A) generates Q
A
in the semiﬁeld sense.
Proposition 2.36. If A is tame then for any f ∈ Q
A
there exist func
tions g, h ∈ Conv(A) such that f = “
g
h
”.
Proof. Since Aﬀ(A) provides a set of generators for the semiﬁeld Q
A
any element in Q
A
can be written as a ratio of two polynomial functions
from the elements of Aﬀ(A).
Corollary 2.37. If A is tame then Spec
m
(Conv(A)) = Spec
m
(Q).
Proof. Since we have the inclusion Conv(A) ⊂ Q any epimorphism
Q → T determines an epimorphism Conv(A) → T by taking restriction.
Since Conv(A) generates the semiﬁeld Qthis gives an embedding Spec
m
(Q) ⊂
Spec
m
(Conv(A)).
To ﬁnish the proof we need to show that any epimorphismx : Conv(A) →
T can be extended to Q. This follows from Proposition 2.29 and Corollary
2.36.
Example 2.38. The free tropical algebra A = T[x
1
, . . . , x
n
] is tame.
The group Aﬀ(A) corresponds to the group of all aﬃnelinear functions
6. AFFINE STRUCTURE RESULTING FROM THE SEMIALGEBRAIC DATA 23
f : R
n
→R whose slope is integer:
f(x) =< s, x > +t,
s = (s
1
, . . . , s
n
) ∈ Z
n
, t ∈ R. The function f is primitive aﬃne for A if
s
j
≥ 0, j = 1, . . . , n. Convex functions are ﬁnite tropical sums of elements
of Aﬀ(A).
The tropical algebra A
′
= T[x
1
, . . . , x
n
, x
−1
1
, . . . , x
−1
n
] is also tame. We
have A
′
⊃ A and Aﬀ(A
′
) = Aﬀ(A) ⊂ A
′
. All elements of Aﬀ(A
′
) are
primitive aﬃne for A
′
.
The tropical semiﬁeld Rat(A) = Rat(A
′
) is itself a tropical algebra.
However, it is not tame as Aﬀ(Rat(A)) is empty. E.g. both “
1
T
1
T
+x
1
” and
“
1
T
1
T
+x
−1
1
” are elements of Rat(A). However, we have the following expression
for the tropical sum of these elements
“
1
T
1
T
+x
1
+
1
T
1
T
+x
−1
1
” = “
1
T
+x
−1
1
+ 1
T
+x
1
1
T
+x
1
+x
−1
1
+ 1
T
” = “
1
T
+x
−1
1
+x
1
1
T
+x
1
+x
−1
1
” = 1
T
.
Thus 1
T
is not a primitiveaﬃne function in Q.
6. Aﬃne structure resulting from the semialgebraic data
If A is tame then Aﬀ(A)/T
∗
is a free ﬁnitely generated Abelian group.
Consider
T = Hom(Aﬀ(A)/T
∗
, R) ≈ R
n
.
This is an aﬃne space with the tautological integer aﬃne structure.
Proposition 2.39. If A is tame then we have a natural embedding
(Spec
m
)
◦
֒→ T.
Proof. The embedding Aﬀ(A) ⊂ A generates a homomorphism
(1) T[x
1
, . . . , x
n
, x
−1
1
, . . . , x
−1
n
] → Q
A
.
This gives a map (Spec
m
)
◦
→ T. We need to show injectivity of this map.
Suppose that s
1
, s
2
∈ (Spec
m
)
◦
, s
1
, s
2
: Q
A
→ T are distinct, but they
produce the same homomorphism after the composition with (1). But any el
ement of Q
A
can be expressed in terms of the elements fromT[x
1
, . . . , x
n
, x
−1
1
, . . . , x
−1
n
]
(using addition, multiplication and division) since A is tame. As the values
of the functions fromT[x
1
, . . . , x
n
, x
−1
1
, . . . , x
−1
n
] at s
1
and s
2
are all the same
we get that the values of all the functions from Q
A
at s
1
and s
2
are also the
same which leads us to a contradiction.
Thus we may treat the ﬁnite part of the maximal spectrum of a tame
tropical algebra A as certain (sedentarity 0) points in the aﬃne space asso
ciated to Aﬀ(A). This gives us a way to consider topological spaces much
more general than integer aﬃne manifolds with corners. Unfortunately, most
of them won’t be useful for us as they’ll be rather far from being a manifold.
24 2. SOME (SEMI)ALGEBRAIC NOTIONS
Example 2.40. Let K ⊂ [0, 1] ⊂ R be the Cantor set and let A be
the space of all functions A → T that can be obtained as a restriction of a
tropical polynomial f ∈ T[x], f : T →T, to K.
Then Spec
m
(A) = K as we can evaluate any f ∈ A on any point
y ∈ K. Conversely, if y : A → T is a Thomomorphism then it gives a T
homomorphismT[x] →T (as the restriction to T produces a Thomomorphism
T[x] → A) and thus corresponds to a point y ∈ T. If y / ∈ K then the ho
momorphism y : T[x] →T cannot factor through A as the value of f at y is
not determined by the values at K.
Note that Aﬀ(A) = Aﬀ(T[x]), so the tropical algebra A is still tame, so
in a sense we are considering the Cantor set enhanced with an integer aﬃne
structure.
In the following chapters we introduce tropical ndimensional varieties.
Locally they may look like either T
n
or some more general polyhedral n
dimensional complexes in T
N
, N > n. They will never look like the Cantor
set from Example 2.40. The next example provides a tropical algebra whose
spectrum is a tropical variety (as we’ll see later).
Example 2.41. Let A be the algebra obtained by restriction of tropical
polynomials in two variables to the tripod Y ⊂ T
2
deﬁned by
Y = [(−∞, 0), (0, 0)] ∪ [(0, −∞), (0, 0)] ∪ [(0, 0), (+∞, +∞)],
see Figure 1. The projection (x, y) → x gives a map π : Y → T that
induces a homomorphism π
∗
: T[x] → A. Furthermore, the map σ : T → Y ,
x → (x, “x + 0”) also induces a homomorphism σ
∗
: A → T[x] that is right
inverse to π
∗
, i.e. π
∗
◦ σ
∗
= Id. The map σ
∗
◦ π
∗
gives a retraction of A to
the subalgebra of functions constant on the ray [(0, −∞), (0, 0)]
Figure 1. A planar tropical line and its retractions.
7. REGULAR FUNCTIONS AND TROPICAL SCHEMES 25
Note that Y is symmetric with respect to permutation of x and y. Thus
we also have a right inverse to the projection homomorphism T[y] → A. We
have Spec
m
(A) = Y , the space Y is called the planar tropical line.
7. Regular functions and tropical schemes
Let f ∈ Q and x ∈ Spec
m
(A). We say that the value of f at x is f(x) if
the epimorphism x : A → T extends to an epimorphism ¯ x :
¯
A → T, where
¯
A ⊂ Q is a subalgebra such that
¯
A ⊃ A ∪ {f} and ¯ x(f) = f(x). Since A
generates Q as a semiﬁeld the value f(x) ∈ T is unique (if it exists). Note
that for any f ∈ Q and x ∈ Spec
m
(A)
◦
the value f(x) exists (and not equal
to −∞ ∈ T).
A point x ∈ Spec
m
(A) is called regular for f ∈ Q if there exists an open
neighborhood U ∋ x, U ⊂ Spec
m
(A), and an element g ∈ Conv(A) such that
the values g(y) and f(y) exist and g(y) = f(y) for any y ∈ U.
Let U ⊂ Spec
m
(A) be any subset.
Definition 2.42. The tropical algebra
˜
O(U) associated to a subset U
consists of all elements of Q that are regular at every point of U.
The tropical algebra O(U) consists of functions f : U → T such that
there exists an element
˜
f ∈
˜
O(U) ⊂ Q such that
˜
f(x) = f(x) for any x ∈ U.
An element of O(U) is called a regular function on U.
Note that O(U) is a quotient of
˜
O(U) as we have the evaluation epimor
phism
ev
U
O
:
˜
O(U) → O(U),
see Deﬁnition 2.14.
Definition 2.43. A point x ∈ Spec
m
(A) is called a pole for f ∈ Q if f
is not regular at x. A point x ∈ Spec
m
(A) is called a zero of f ∈ Q if f is
regular at x, but “
1
f
” has a pole at x.
Definition 2.44. Each element f ∈ A deﬁnes a set V
f
⊂ Spec
m
(A) of
its zeroes. This set is called a hypersurface deﬁned by f.
Proposition 2.45. The union of ﬁnite number of hypersurfaces is a
hypersurface.
Proof. We claim that
n
¸
j=1
V
f
j
is a hypersurface deﬁned by
n
¸
j=1
f
j
. Clearly
all points of Spec
m
(A) are regular for any f ∈ A. Suppose that x ∈
Spec
m
(A) is regular for “
1
n
¸
j=1
f
j
”. Then x is also regular for “
1
f
j
” as it can be
obtained from “
1
n
¸
j=1
f
j
” by taking a product with all f
j
′ , j
′
= j.
26 2. SOME (SEMI)ALGEBRAIC NOTIONS
Definition 2.46. A tropical scheme is a pair consisting of a topological
space X and a sheaf
˜
O
X
of tropical algebras on X such that for every point
x ∈ X there is an open neighborhood U ∋ x, a tropical integral domain A
and an open set U
A
⊂ Spec
m
(A) such that the pair (U,
˜
O
X

U
) is isomorphic
to the pair (U
A
,
˜
O
Spec
m
(A)

U
A
).
The scheme is called reduced if for any U the tropical algebra
˜
O
X
(U) is
reduced. In such case we set O
X
(U) =
˜
O
X
(U).
Since the restriction of a sheaf to an open set is a sheaf the tropical
integral domain A has to be such that
˜
O
A

U
A
form a sheaf. Note that clearly
we always have the required restriction homomorphisms ρ
V
U
:
˜
O
A
(V ) →
˜
O
A
(U) for V ⊃ U that are also always monomorphisms as we just take an
embedding of the elements of Q
A
that are regular on V in the larger set
of those elements which are regular on U. From now on we restrict our
attention to reduced schemes X. The sheaf O
X
is called the structure sheaf
of X.
Let f ∈ O
X
(V ) and x ∈ V for an open V ⊂ X Choose an open neighbor
hood U ∋ x, U ⊂ Spec
m
(A). Thus x corresponds to a tropical epimorphism
x
A
: A →T. The value of f(x) is x
A
(ρ
V
U∩V
(f)) ∈ T.
Proposition 2.47. The value f(x) does not depend on the choice of the
aﬃne neighborhood U.
Proof. Suppose that x corresponds to a tropical epimorphism x
A
′ :
A
′
→ T another aﬃne neighborhood U
′
∋ x with U
′
= Spec
m
(A
′
). Since
x ∈ U∩U
′
∩V both epimorphisms have to factor through the tropical algebra
O(U ∩U
′
∩V ) where both x
A
(ρ
V
U∩V
(f)) and x
A
′ (ρ
V
U
′
∩V
(f)) have a common
lift ρ
V
U∩U
′
∩V
(f).
Definition 2.48. Let Z ⊂ X be any subset and f : Z → T be a
function. The function f is called regular if for any x ∈ Z there exists an
open neighborhood U ∋ x and g ∈ O
X
(U) such that f(y) = g(y) for any
y ∈ Z ∩ U.
Once again, all regular functions on Z together with pointwise addition
and multiplication form a tropical algebra which we denote Funct(Z).
8. Regular maps
Definition 2.49. A regular map between tropical schemes
Φ : (X, O
X
) → (Y, O
Y
)
is a pair consisting of a continuous map
f : X → Y
and a collection of tropical algebra homomorphisms
Φ
∗
: O
Y
(U) → O
X
(Φ
−1
(U))
8. REGULAR MAPS 27
for any open set U ⊂ Y that is consistent with the restriction homomor
phisms of the sheaves O
X
and O
Y
, i.e. such that for any pair of open sets
V ⊂ U ⊂ Y the diagram
O
Y
(U)
ρ
U
V
//
Φ
∗
O
Y
(V )
Φ
∗
O
X
(Φ
−1
(U))
ρ
Φ
−1
(U)
Φ
−1
(V )
//
O
Y
(Φ
−1
(V ))
is commutative. Here ρ
U
V
are the corresponding restriction homomorphisms
for regular functions.
For simplicity of notations we will often suppress the symbols O
X
and
O
Y
and write a regular map just as Φ : X → Y .
Definition 2.50. A regular map Φ : X → Y is called a scheme embed
ding if Φ is a settheoretical embedding and for all open U ⊂ Y the homo
morphisms Φ
∗
: O
Y
(U) → O
X
(Φ
−1
(U)) is an epimorphism. In this case X
is called a closed subscheme of Y , once we identify X with Φ(X) ⊂ Y .
Let V ⊂ Y be any set and W = Φ
−1
(V ). Suppose that f ∈ Funct(V )
and Φ : X → Y is a regular map. As usual, we have a settheoretical
pullback of the function f, namely Φ
∗
(f) : W →T, x → f(Φ(x)).
Proposition 2.51. The function Φ
∗
(f) is regular in U, i.e. Φ
∗
(f) ∈
Funct(W).
Proof. Since f ∈ Funct(V ) for every x ∈ V there exists an open neigh
borhood U ∋ x and g ∈ O
Y
(U) such that g(y) = f(y) for every y ∈ U ∩ V .
We have Φ
∗
(g) ∈ O
X
(Φ
−1
(U)) by deﬁnition of the tropical map and, clearly,
Φ
∗
(g)(z) = f(Φ(z)) for every z ∈ Φ
−1
(U) ∩ W.
CHAPTER 3
Hypersurfaces and complete intersections in T
n
1. Integer aﬃne manifolds as tropical schemes
After a bit of algebraic formalism we return to our geometric objects:
integer aﬃne manifolds.
Theorem 3.1. Any integer aﬃne manifold X with corners can be nat
urally considered as a reduced tropical scheme.
Proof. Locally X is modeled on an open set in T
n
= Spec
m
(T[x
1
, . . . , x
n
])
so that regular functions correspond to monomials, cf. Deﬁnition 1.28.
In particular, we may characterize the regular functions in terms of
integeraﬃne structure. Recall that a monomial is just a aﬃnelinear mor
phism to T.
Let U ⊂ X be an open set and f : U →T be a continuous function.
Proposition 3.2. A function f : U →T is regular at x ∈ U if and only
if there exist an open subset W ⊂ U and a ﬁnite collection of monomials
κ
1
, . . . , κ
l
: W →T such that f
W
= max{κ
1
, . . . , κ
l
}.
Proof. We may choose W so that it is contained in a single chart φ
α
:
U
α
→T
n
. Then the second characterization coincides with the deﬁnition of
a tropical polynomial.
Also we may speak about tropical hypersurfaces in integer aﬃne man
ifolds with corners. A subspace V ⊂ X is called a hypersurface if for
any x ∈ V there exists a chart φ
α
: U
α
→ T
n
and a tropical polynomial
f
α
: T
n
→ T. such that V ∩ U
α
= φ
−1
α
(V
fα
), where V
fα
is the hypersurface
associated to f
α
. Thus to see the structure of hypersurfaces in X it suﬃces
to look carefully at the structure of hypersurfaces in T
n
.
2. Hypersurfaces in T
n
Let f : T
n
→T be a tropical polynomial
(2) f(x) = “
¸
j∈Z
n
a
j
κ
j
(x)” = max
j
a
j
+κ(x),
x = (x
1
, . . . , x
n
) ∈ T
n
. Here the some is taken over the ﬁnite number
of multiindexes j, a
j
∈ T and κ
j
(x) = “x
j
1
1
. . . x
jn
n
”, so that “a
j
κ
j
” are
29
30 3. HYPERSURFACES AND COMPLETE INTERSECTIONS IN T
n
monomials. Recall that the hypersurface V
f
is the locus of all points x ∈ T
n
such that “
1
T
f
” is not regular at x.
Proposition 3.3. The hypersurface V
f
is the locus of points x ∈ T
n
where the maximal value in (2) is attained by more than one monomial
a
j
κ
j
.
Proof. If more than one monomial assumes the maximum at x then f
is strictly convex at x and thus “
1
T
f
” = −f cannot be convex. If only one
monomial is maximal at x then f is locally linear at x and thus −f is also
regular at x.
The monomials a
j
κ
j
naturally deﬁne a stratiﬁcation of V
f
. Let J = {j ∈
Z
n
 a
j
= 0
T
} be the indices parameterizing the monomials that appear in
f. The set J is ﬁnite since f is a polynomial. For each x ∈ T
n
we deﬁne
K
f
(x) = {j ∈ J  f(x) = “a
j
κ
j
”,
in other words K
f
(x) is the set of the indices of the monomials where f(x)
assumes its maximum. Vice versa, for a subset K ⊂ J of cardinality greater
than one we may deﬁne the stratum V
K
f
⊂ V
f
by
V
K
f
= {x ∈ T
n
 K
f
(X) = K.
Note that V
K
f
is deﬁned by a system of linear inequalities in T
n
⊃ R
n
and thus is a convex polyhedron (possibly unbounded) in T
n
. This means
that it is the closure in T
n
of a convex polyhedral domain in R
n
.
Proposition 3.4. We have T
n
=
¸
K⊂J
V
K
f
and
V
f
=
¸
K>1
V
K
f
.
Each component of T
n
V
f
naturally corresponds to a point j ∈ J, such
that “a
j
κ
j
” is maximal in this component.
Proof. this proposition is the direct corollary of Proposition 3.3.
For many subsets of J we have V
K
f
= ∅. If V
K
f
= ∅ we say that K ∈
Subdiv
f
and denote with ∆
K
the convex hull of K in R
n
⊃ K. Denote with
∆
f
the Newton polyhedron of f, i.e. the convex hull of J in R
n
. Each ∆
K
is contained in a minimal aﬃnelinear subspace in R
n
. Denote with ∆
◦
K
the
relative interior of ∆
K
, i.e. the interior in the corresponding aﬃnelinear
space.
Theorem 3.5. The polyhedra ∆
K
form a subdivision of the polyhedron
∆
f
which is dual to the corresponding strata V
K
f
. Namely, we have the
following properties.
• If K
1
, K
2
∈ Subdiv
f
and K
1
∩K
2
= ∅ then K
1
∩K
2
∈ Subdiv
f
and
∆
K
1
∩ ∆
K
2
= ∆
K
1
∩K
2
.
2. HYPERSURFACES IN T
n
31
• The (relatively) open polyhedra ∆
◦
K
are disjoint: for any K
1
, K
2
∈
Subdiv
f
, K
1
= K
2
we have ∆
◦
K
1
∩ ∆
◦
K
2
= ∅.
• ∆
f
=
¸
K∈Subdiv
f
∆
◦
K
.
• For any K ∈ Subdiv
f
we have dimV
K
f
+dim∆
K
= n. Furthermore
the aﬃnelinear subspaces in R
n
generated by V
K
f
∩ R
n
and ∆
K
are orthogonal. (More rigorously, the Newton polygon ∆
f
and the
hypersurface V
f
∩ R
n
belong to dual vector spaces R
n
, but we may
identify them by introducing a scalar product to R
n
.)
• If ∆
K
1
⊂ ∆
K
2
then V
K
1
f
⊃ V
K
2
f
.
In particular, to each facet (i.e. an (n − 1)dimensional face of V
f
) we
may associate a positive integer number equal to the integer length of the
corresponding interval in Subdiv
f
. Here the integer length of an interval
I ⊂ R
n
with ∂I ∈ Z
n
is the total number of integer subintervals in it (i.e.
#(I ∩ Z
n
) −1).
Proof. The last two properties come as straightforward applications of
Linear Algebra.
Note that for every j ∈ J the locus “a
j
κ
j
(x)” = f(x) is deﬁned with
a system of linear inequalities and therefore is convex. Suppose that K ∈
Subdiv
f
and k ∈ ∆
◦
K
. Then, by convexity, “a
k
κ
j
(k)” = f(x) exactly on
V
K
f
. Thus without loss of generality we may assume that K coincides with
∆
K
∩ Z
n
.
Thus ∆
◦
K
are disjoint and form a subdivision of ∆
f
. Suppose that K
1
∩
K
2
= ∅. Then a generic point x of the convex hull of V
K
1
f
∪ V
K
2
f
must
correspond to K
1
∪ K
2
.
Remark 3.6. Subdivisions that appear in Theorem 3.5 are called con
vex, regular or, sometimes, coherent lattice subdivisions of the polyhedron
∆
f
, cf. e.g. [15]. The function j → a
j
is called the height function of the
subdivision. In real algebraic geometry such subdivisions appeared after the
discovery of the patchworking technique by Viro [68]. These subdivisions
come as projections of the top faces of the polyhedral domain in R
n
× R
obtained as the convex hull of the undergraph of j → a
j
, see [15].
Not all subdivisions are convex. Figure 1 depicts a classical example of
a nonconvex lattice subdivision (see e.g. [68], [15]). To see nonexistence
of the height function it suﬃces to look at the attachments of the wouldbe
faces around the inner square.
Remark 3.7. Theorem 3.5 gave a description of hypersurfaces in T
n
.
However, the same construction works also for hypersurfaces V in (T
×
)
n
,
TP
n
and other toric varieties as long as every component of V (its subset that
constitute a hypersurface itself) has nonempty intersection with the torus
(T
×
)
n
. Then the hypersurface V is still given by a tropical polynomial f in
32 3. HYPERSURFACES AND COMPLETE INTERSECTIONS IN T
n
Figure 1. A nonconvex lattice subdivision.
n variables and can be obtained by taking the closure in the corresponding
toric variety of the toric part V
f
∩(T
×
)
n
of the aﬃne hypersurface V
f
⊂ T
n
.
We often will use the same notation V
f
for a hypersurface in other toric
varieties.
3. Lines in the plane
The easiest examples to visualize are planar curves, i.e. hypersurfaces
in T
2
. Note that Y from Example 2.41 is an example of a tropical line in
the plane. Indeed, it is the hypersurface of f(x, y) = “x + y + 1
T
”. All
three monomials are equal at the origin while everywhere on the three rays
two of the three monomials are equal, but greater then the third remaining
monomial.
A general polynomial of degree 1 in two variables is of the form
f(x, y) = “ax +by +c”.
Thus a line in T
2
is the hypersurface associated to this tropical polynomial.
Note that as long as a, b, c = 0
T
any tropical line can be obtained from Y
by a translation in R
2
. More precisely, we have to take Y ∩ R
2
, apply the
translation and take the closure in T
2
again.
Indeed, the hypersurface, associated to “
f(x,y)
c
′′
= f(x, y)−c = max{(x+
a−c, y+b−c, 0} coincides with V
f
. But max{(x+a−c, y+b−c, 0} corresponds
to max{x, y, 0} under the translational change of coordinates x → x+a−c,
y → y +b −c. Note that the horizontal and vertical rays of Y end with an
inﬁnite point (as the axes {y = −∞} and {x = −∞} are included in T
2
),
but the diagonal ray is open.
3. LINES IN THE PLANE 33
If one of the coeﬃcients of f assumes the value 0
T
= −∞ then the
corresponding monomial is never maximal in f. Thus the corresponding hy
persurface is the closure of the straight line which maybe horizontal, vertical
or diagonal, depending on which monomial disappears, see Figure 2.
Figure 2. Five lines in T
2
.
Consider now the case when two of the coeﬃcients of f assume the value
−∞. If f(x, y) = c, c ∈ T
×
, then “
1
T
f
” = −c is regular everywhere on T
2
, so
V
f
= ∅. If f(x, y) = “ax” = x +a, a ∈ T
×
, then “
1
T
f
” = −x−a is regular as
long as x = infty, but not deﬁned at the coordinate yaxis {x = −∞} of T
2
.
Thus in this case V
f
coincides with the yaxis. Similarly the hypersurface
of f(x, y) = “ax” = x +b, b ∈ T
×
, is the xaxis of T
2
.
The projective space TP
2
provides a compactiﬁcation of T
2
by attaching
an extra line (called the inﬁnite line). When we consider, e.g. a family
(3) f
t
= “ty +c”, t → −∞
the corresponding horizontal line moves to inﬁnity and coincides with that
inﬁnite line in the limit.
We may draw the corresponding deformation on the (ﬁnite) triangle.
For that we need to reparameterize R
2
to the interior of a ﬁnite triangle.
34 3. HYPERSURFACES AND COMPLETE INTERSECTIONS IN T
n
One of the most natural ways (along with the map provided by Proposition
1.30 to do that is via the combination of the logarithmic moment map
Log : (C
×
)
2
→R
2
, Log(z, w) = (log z, log w),
which is the moment map for the (C
×
)
2
invariant form
dz
z
∧
d¯ z
¯ z
+
dw
w
∧
d ¯ w
¯ w
and the FubiniStudy moment map for CP
2
µ : (C
×
)
2
→R
2
, µ(z, w) = (
z
2
1 +z
2
+w
2
,
w
2
1 +z
2
+w
2
).
Note that the image µ(R
2
) is the interior of the triangle T = {(x, y) ∈
R
2
 x ≥ 0, y ≥ 0, x +y ≤ 1}. Both maps Log and µ have the same ﬁbers,
so we have a welldeﬁned map µ ◦ Log
−1
: R
2
→ Int(T), which is a diﬀeo
morphism. Furthermore, this diﬀeomorphism extends to a diﬀeomorphism
TP
2
→ T. When we need to speak about the inﬁnite points of varieties in
TP
2
it is more convenient to draw their images under this reparameteriza
tion. Note though that the image of a straight line in R
2
is (in general) no
longer straight in T.
Figure 3. The image of a complex projective lines under
µ ◦ Log
−1
is an inscribed ellipse in T.
Remark 3.8. one of the advantages of the parameterization µ ◦ Log
−1
with respect to the parameterization provided by Proposition 1.30 is that
the image of a line in RP
2
is an ellipse that is tangent to the three sides of the
triangle T, see Figure 3. The points of tangency with the sides corresponds
to the points of intersection with the three coordinate axes (the xaxis, the
yaxis and the inﬁnite line). These tangencies divide the circle into three
arcs, each corresponding to the real points of a line in a quadrant of (R
×
)
2
.
Note that a generic line in R
2
intersects three out of four quadrants.
The imaginary points of a line L ⊂ CP
2
that is real (i.e. invariant with
respect to the complex conjugation) are mapped inside this ellipse in the
3. LINES IN THE PLANE 35
21 fashion so that the the inverse image of a point inside the ellipse under
Log 
L
consists of a pair of complex conjugate points.
Furthermore, the image of any (not necessarily real) line in CP
2
is the
region in T
2
that is encompassed by an ellipse tangent to the sides of T.
Indeed, any line in CP
2
can be made real after the multiplication in (C
×
)
2
by a suitable pair (a, b), a, b ∈ C
×
. Note that the family of ellipses in R
2
is
5dimensional and each tangency gives a condition of codimension 1. Thus
we have a 2dimensional family of suitable ellipses and this corresponds to
the dimension of the space of lines in the projective plane.
The lines given by a binomial equation pass through an intersection point
of the coordinate axes (recall that we treat the inﬁnite line as one of the
coordinate axes!) and correspond to the degeneration of ellipses to intervals
passing through a vertex of the triangle. The lines given by a monomial
coincide with one of the coordinate axes and correspond to a side of the
triangle.
The same parameterization works well for images of tropical lines. In
deed, a generic line is made of three segment, where each segment is a
subinterval of a line passing through a vertex of T, see the ﬁrst part of Fig
ure 4. The second part of this ﬁgure shows how generic lines degenerate to a
binomial line. The last part of Figure 4 depicts the family (3) and its limit.
Figure 4. Images of tropical lines in T and their degenerations.
Thus we see that any line in TP
2
is either an R
2
translate of the tripod Y
from Example 1 or a degeneration of such translates. Note that two generic
lines in TP
2
intersect in a unique point: e.g. any pair of lines in Figure 2
has such “transverse” intersection. In the same time we may ﬁnd two lines
that have a whole ray in common, see Figure 5.
Later in this book we develop the tropical intersection theory which
allows to associate the cycle of the right dimension even for nontransverse
cycles A, B. This intersection cycle will be supported on the skeleton of the
settheoretical intersection of the expected dimension. Each facet F of this
skeleton will be included to the cycle with an integer (possibly negative)
coeﬃcient that depends only on the local structure of A and B near E.
36 3. HYPERSURFACES AND COMPLETE INTERSECTIONS IN T
n
stable intersection point
Figure 5. Nontransverse intersection of two lines in T
2
.
In particular, even though the lines from Figure 5 intersect along a ray,
their intersection cycle is the (sedentarity 0) endpoint of this ray. This
agrees with the notion of the stable intersection from [56] in the case when
the ambient space is an (integer aﬃne) smooth variety.
Proposition 3.9. Any pair of points p
1
, p
2
∈ TP
2
can be joined with a
line. Furthermore, this line is unique unless this pair of line and one of the
intersection points of the coordinate axes (the points of sedentarity 2) are
collinear.
Proof. Applying a translation in R
2
to the tripod Y from Example 1
we may ﬁnd a line L ∈ TP
2
such that its 3valent vertices coincide with p
1
.
If the sedentarity of p
1
is positive then we may ﬁnd a line L ∋ p
1
and a
family of nondegenerate lines L
t
so that the the trivalent point of L
t
tends
to p
1
. Generically, the line L separates TP
2
into three sectors, see Figure
6. If p
2
/ ∈ L then it is inside one of these sector. We can move L into this
sector so that p
2
is remained on L by a translation antiparallel to the ray
opposite to the sector of p
2
.
Theorem 3.10. Lines in TP
2
form themselves an integer aﬃne manifold
with corners isomorphic to TP
2
.
This manifold is called the dual projective plane and denoted with
(TP
2
)
∗
.
Proof. Note that from the algebraic point of view the statement is
trivial. Indeed, any line is given by a polynomial “ax+by+c”, a, b, c ∈ T
up to the simultaneous multiplication of the coeﬃcients a, b, c by the same
3. LINES IN THE PLANE 37
move into
this sector
Figure 6. Finding a line passing via two points in TP
2
.
scalar λ ∈ T
×
. These triples of coeﬃcients up to such rescaling form TP
2
by the very deﬁnition. Nevertheless, it is useful to look at the space of lines
from a geometric point of view. A chart near a line with the 3valent vertex
in R
2
is given by that 3valent vertex itself.
Consider now those degenerate lines that do not coincide with a coor
dinate axes (those given by a binomial). These lines pass through a vertex
of the triangle T and a point on its side. We still have such a distinguished
point by tracing the limit of the 3valent vertex under its approximation
by nondegenerate lines, but this point is the vertex of T, so it does no
longer determine the position of the line. Nevertheless, in the complement
of the three points corresponding to the coordinate lines we may identify the
space of all lines in TP
2
with the space of lines together with a distinguished
point (a 3valent vertex in the case of nondegenerate line and a vertex of T
otherwise). Furthermore, via this distinguished point we may identify the
nondegenerate lines with the points of R
2
.
Consider the inversion σ : R
2
→ R
2
, (x, y) → (“
1
x
”, “
1
y
”) = (−x, −y).
This inversion does not extend to the vertices of TP
2
, but does extend to the
vertices of TP
2
enhanced with lines passing through them. This extension
gives a chart to T × T
×
in a neighborhood of noncoordinate lines passing
via the vertex of TP
2
. Note that we may easily describe the same chart in
coordinates. Finally, a coordinate line L is mapped to the opposite vertex
of T by the inversion while choosing a nearby point in the image completely
determines the line nearby to L. This gives a chart to T
2
.
Remark 3.11. Because of the inversion σ from the proof of Theorem
3.10 it is convenient to depict the dual plane with the inverted triangle, see
Figure 7. As a map (TP
2
)
∗
TP
2
the inversion σ is only partially deﬁned.
However, replacing of the vertex of TP
2
with all lines passing through this
vertex is the tropical counterpart of the blowing up of this vertex. It allows
one to deﬁne a new manifold X (that is the result of blowing up of TP
2
in
38 3. HYPERSURFACES AND COMPLETE INTERSECTIONS IN T
n
all three vertices) and everywhere deﬁned maps X →TP
2
and X → (TP
2
)
∗
,
see Figure 7.
blow down three sides
blow down 3 other sides
Figure 7. Passing from the projective plane to the dual
projective plane.
Note that X is an integer aﬃne manifold with corners as tropical blowups
come with natural charts to T
2
. Furthermore, it is one of the toric varieties
from Remark 1.32, the one depicted on Figure 6.
4. Curves in the plane
Let us look at the conics in TP
2
. These are the hypersurfaces given by
quadratic polynomials
f(x, y) = “ax
2
+bxy +cy
2
+dx +ey +f”,
a, b, c, d, e, f ∈ T. We have six monomials and each can dominate the
polynomial f in a certain region in the plane (possibly empty).
The Newton polygon of f is the triangle ∆
f
with vertices (0, 0), (2, 0) and
(0, 2) or its subpolygon (in the case when some of the coeﬃcients vanish, i.e.
assume the value 0
T
= −∞). By Theorem 3.5 there is a lattice subdivision
of ∆
f
for each conic C ⊂ TP
2
and, conversely, each coherent subdivision of
∆
f
corresponds to a conic in TP
2
.
The smallest possible convex polygon with vertices in Z
2
is a triangle of
area
1
2
. Such triangles are called em the primitive triangles.
Definition 3.12. Curves dual to subdivision into primitive triangles are
called smooth planar tropical curves.
Primitive triangles do not contain lattice point other than their vertices.
Therefore, primitive triangulations (i.e. lattice decompositions of a Newton
polygon into primitive triangles) contain all lattice points of the polygon
among their vertices.
Consider a smooth conic V
f
⊂ TP
2
, see e.g. Figure 8. Because of the
smoothness condition each monomial m ∈ ∆
f
∩ Z
2
corresponds to a non
empty region in TP
2
. Furthermore, all the edges of V
f
has weight 1. Let us
deform just one of the coeﬃcients of f. It is easy to see that the resulting
deformation will leave the strata of V
f
disjoint from m invariant. In the
same time the edges of V
f
corresponding to the edges of Subdiv
f
adjacent
4. CURVES IN THE PLANE 39
the resulting region
Figure 8. Deforming one coeﬃcient.
to m will move enlarging or diminishing the corresponding region depending
on whether we increase or decrease the coeﬃcient of the monomial m.
Figure 10 shows some smooth conics together with the corresponding
subdivisions. It is easy to see that the ﬁgure exhaust all possible combina
torial types of smooth conics.
Figure 9. Smooth planar conics.
It is instructive to look at the possible degenerations of smooth conics.
The simplest degeneration correspond to a coarser subdivision of ∆
f
when
we take into Subdiv
f
the union of two nearby primitive triangles instead of
taking each one individually. We have two combinatorially diﬀerent cases:
the union of two could be a parallegram or it could be a triangle of area
1, see Figure ??. Note that the ﬁrst case corresponds to a reducible conic
that decomposes to the union of two lines. The second degeneration can be
interpreted as a smooth conic that is tangent to a coordinate axis in TP
2
as
we shall see later.
The higher is the degree the more possibilities we have for the com
binatorial type of the curve. List all combinatorial types would take too
40 3. HYPERSURFACES AND COMPLETE INTERSECTIONS IN T
n
weight 2 edge
Figure 10. Singular planar conics.
long already for the case of planar cubic. Figure 11 depicts a smooth and a
singular cubic.
Figure 11. Planar cubics.
The following two examples list two particularly simple combinatorial
types of smooth tropical curves of arbitrary degree. Note to specify a com
binatorial type of a planar tropical curve of degree d we need to produce a
lattice subdivision of the triangle ∆
d
⊂ R
2
with vertices (0, 0), (d, 0) and
(0, d) (or a subpolygon of this triangle).
Example 3.13. Consider the square lattice in Z
2
. If we subdivide each
square into two triangles by the diagonal parallel to the line x + y = 0 we
get a subdivision of R
2
that is compatible with ∆
d
for any d. The resulting
4. CURVES IN THE PLANE 41
subdivision and the tropical curve in the corresponding combinatorial type
are pictured on Figure 12. The tropical curves in this combinatorial type
(as well as all their degenerations) are called honeycombs. They proved to
be useful for a range of problems related to the Horn problem, see [32].
Figure 12. Honeycombs.
Note that the honeycomb triangulation of ∆
d
is symmetric with respect
to the exchange of the x and y coordinates. Furthermore, it is symmetric
with respect to the action of the symmetric group S
3
that interchanges these
two axes and the inﬁnite axis.
Our next example is not as symmetric.
Example 3.14. Let us subdivide ∆
d
⊂ R
2
into “ﬂoors” by the lines
y = 1, . . . , d−1. Each ﬂoor is a trapezoid that can be further subdivided into
the primitive triangles as shown on Figure 13. These subdivisions appeared
in [23] as coherent subdivisions of higherdimensional simplices.
Figure 13. The ItenbergViro subdivision in dimension 2.
42 3. HYPERSURFACES AND COMPLETE INTERSECTIONS IN T
n
Remark 3.15. The coherence of the subdivisions in Examples 3.13 and
3.14 are veriﬁed by existence of the corresponding tropical curves. To check
the latter we may note that because the lines y = 1, . . . , d−1 are compatible
with both types of subdivision our tropical curves are glued from the curves
dual to trapezoids of height 1 as shown on Figure 14. These curves are
called ﬂoors. The kth ﬂoor has d − k vertical rays pointing up, d − k + 1
rays pointing down and no other vertical edges.
Furthermore, we may ﬁx any positions (i.e. the xcoordinates) for the
vertical rays pointing down and ﬁnd a smooth tropical curve in the needed
combinatorial type with such rays. Because of that we may inductively stack
a k+1th ﬂoor on top of the kth ﬂoor. In particular we may combine the ﬂoors
of diﬀerent combinatorial types. Note also that any lattice subdivision of a
the Newton polygon of a ﬂoor (i.e. a subpolygon of the strip k −1 ≤ y ≤ k)
is coherent.
Figure 14. Floors and stacking them on top of each other.
Example 3.16. As our last example of a planar tropical curve we con
sider a rather involved example of a curve of degree 10. It appeared in the
work of Itenberg [21] disproving the Ragsdale conjecture (a conjecture on
topology of plane real curves that appeared in 1905 in [55] and was ﬁnally
disproved only in 1992 [21]). The counterexample is provided by this very
curve once we equip it with the suitable real phases, see Figure 15.
5. Surfaces in TP
3
We start by looking at the hyperplane in TP
3
, i.e. the hypersurface given
by the tropical polynomial “ax + by + cz + d”. Similarly to the case with
the lines in TP
2
it is easy to show that any hyperplane with a, b, c, d = 0
T
is the result of translation of the (standard) hyperplane of “x +y +z +1
T
”
by a vector in R
3
. Again if some (but not all) of the coeﬃcients a, b, c, d
assume the value 0
T
then we can interpret the corresponding hyperplane as
the limiting set of a family of translations of V
“x+y+z+1
T
”
in R
3
.
5. SURFACES IN TP
3
43
Figure 15. The ItenbergRagsdale curve of degree 10.
Figure 16 depicts a generic hyperplane H ⊂ TP
2
. It consists of 6 sectors,
all of them have a common vertex v ∈ R
3
. There are 4 outgoing rays from
v, in the direction (−1, 0, 0), (0, −1, 0), (0, 0, −1) and (1, 1, 1). Any pair of
these rays span a sector in R
3
diﬀeomorphic to the positive quadrant R
2
≥0
.
To get H we take the closure in TP
3
⊃ R
3
of the union of the 6 sectors.
vertex v
Figure 16. A tropical plane in the 3space.
44 3. HYPERSURFACES AND COMPLETE INTERSECTIONS IN T
n
The position of the vertex v ∈ R
3
completely determines a tropical
hyperplane V
“ax+by+cz+d”
with a, b, c, d = 0
T
. Similarly to the case with lines
in TP
2
all hyperplanes are parameterized by the dual space (TP
3
)
∗
⊃ R
3
.
More generally we have the following statement generalizing Theorem 3.10.
Theorem 3.17. The space of all hyperplanes (hypersurfaces given by
tropical polynomials of degree 1) in TP
n
forms an integer aﬃne manifold
with corners isomorphic to TP
n
.
Proof. Let us ﬁrst note that the Theorem is trivial if n = 1. Indeed a
hyperplane in TP
1
is given by a polynomial “ax + b” = max a +x, b in one
variable x, a, b ∈ T, “ab” = 0
T
. The corresponding hypersurface always just
the single point x = b −a ∈ TP
1
. Thus the set of such hyperplanes coincides
with the set of points in TP
1
.
To prove the theorem in general it suﬃces to prove show that if f is
a polynomial of degree 1 in n variables then the set V
f
determines the
coeﬃcients of f up to their simultaneous tropical multiplication by a non
zero constant. Indeed, once we prove this we can identify the space of
hyperplanes with the space of all coeﬃcients up to the simultaneous rescaling
which is the tropical projective nspace by deﬁnition.
Recall that TP
n
is topologically a simplex. Each edge of this simplex
corresponds to a tropical line TP
1
obtained as the intersection of (n − 1)
coordinate planes. The hyperplane V
f
cuts a point on each such TP
1
unless
this TP
1
is contained in V
f
. Each such point is a hyperplane in TP
1
and
determines two coeﬃcients of f up to scaling. If the line TP
1
is contained
in V
f
then both corresponding coeﬃcients must be equal to 0
T
.
We call this space of hyperplanes the dual projective space and denote
with (TP
n
)
∗
.
To understand the geometry of higherdegree surfaces in TP
3
(and more
generally the geometry of higherdimensional tropical varieties) it is conve
nient to introduce the notion of ﬂoor decomposition.
...
(TO BE CONTINUED)
...
6. Complete Intersections
7. Balancing condition
CHAPTER 4
Tropical varieties
45
CHAPTER 5
Tropical equivalence
47
Bibliography
[1] O. Aharony, A. Hanany, Branes, superpotentials and superconformal ﬁxed points,
http://arxiv.org hepth/9704170.
[2] O. Aharony, A. Hanany, B. Kol, Webs of (p, q) 5branes, ﬁve dimensional ﬁeld theories
and grid diagrams, http://arxiv.org hepth/9710116.
[3] V. I. Arnold, The situation of ovals of real plane algebraic curves, the involutions of
fourdimensional smooth manifolds, and the arithmetic of integral quadratic forms,
Funkcional. Anal. i Priloˇzen. 5 (1971) no. 3, 19.
[4] M. F. Atiyah, Angular momentum, convex polyhedra and algebraic geometry, Proc.
Edinburgh Math. Soc., 26 (1983) 121133.
[5] J. P. Benzecri, Vari´et´es localment aﬃnes, S´eminaire Ehresmann, May 1959.
[6] G. M. Bergman, The logarithmic limit set of an algebraic variety, Trans. AMS, 157
(1971), 459469.
[7] D. Bernstein, The number of roots of a system of equations, Functional Anal. Appl.
9 (1975), 183185.
[8] T. Bogart, A. Jensen, D. Speyer, B. Sturmfels, R. Thomas, Computing Tropical
Varieties, http://arxiv.org/abs/math.AG/0507563.
[9] F. Bourgeois, A MorseBott approach to contact homology, Dissertation, Stanford
University 2002.
[10] L. Brusotti, Curve generatrici e curve aggregate nella costruzione di curve piane
d’ordine assegnato dotate del massimo numero di circuiti, Rend. Circ. Mat. Palermo
42 (1917), 138144.
[11] M. Einsiedler, M. Kapranov, D. Lind, Nonarchimedean amoebas and tropical vari
eties, http://arxiv.org math.AG/0408311.
[12] Y. Eliashberg, A. Givental, H. Hofer, Introduction to symplectic ﬁeld theory, GAFA
2000, Special Volume, Part II, 560–673.
[13] M. Forsberg, M. Passare, A. Tsikh, Laurent determinants and arangements of hyper
plane amoebas, Advances in Math. 151 (2000), 45–70.
[14] K. Fukaya, Multivalued Morse theory, asymptotic analysis and mirror symmetry,
Graphs and patterns in mathematics and theoretical physics, 205–278, Proc. Sympos.
Pure Math., 73, Amer. Math. Soc., Providence, RI, 2005.
[15] I. M. Gelfand, M. M. Kapranov, A. V. Zelevinsky, Discriminants, resultants, and
multidimensional determinants. Mathematics: Theory & Applications. Birkh¨auser
Boston, Inc., Boston, MA, 1994.
[16] , M. Gross, B. Siebert, Aﬃne structures...
[17] M. Gross, P. M. H. Wilson, Large complex structure limits of K3 surfaces, J. Diﬀer
ential Geom. 55 (2000), no. 3, 475–546.
[18] A. Harnack,
¨
Uber Vieltheiligkeit der ebenen algebraischen Curven, Math. Ann. 10
(1876), 189199.
[19] A. Henriques, An analogue of convexity for complements of amoebas of varieties of
higher codimensions, Preprint, Berkeley, May 2001.
[20] D. Hilbert, Mathematische Probleme, Arch. Math. Phys. (3) 1 (1901), 213237.
[21] I. Itenberg,...
49
50 BIBLIOGRAPHY
[22] I. Itenberg, V. Kharlamov, E. Shustin, Welschinger invariant and enumeration of
real plane rational curves, http://arxiv.org/abs/math.AG/0303378.
[23] I. Itenberg, O. Viro, Patchworking real algebraic curves disproves the Ragsdale con
jecture, Math. Intelligencer, 18 (1996), 1928.
[24] I. Itenberg, O. Viro, Maximal real algebraic hypersurfaces of projective space, to ap
pear.
[25] M. M. Kapranov, A characterization of Adiscriminantal hypersurfaces in terms of
the logarithmic Gauss map, Math. Ann. 290 (1991), 277285
[26] M. M. Kapranov, Amoebas over nonArchimedean ﬁelds, Preprint, 2000.
[27] R. Kenyon, A. Okounkov, Planar dimers and Harnack curves, http://arxiv.org
math.AG/0311062.
[28] R. Kenyon, A. Okounkov, Limit shapes and the complex burgers equation,
http://arxiv.org mathph/0507007.
[29] R. Kenyon, A. Okounkov, S. Sheﬃeld, Dimers and amoebae, http://arxiv.org math
ph/0311005.
[30] V. M. Kharlamov, On the classiﬁcation of nonsingular surfaces of degree 4 in RP
3
with respect to rigid isotopies, Funktsional. Anal. i Prilozhen. 18 (1984), no. 1, 4956.
[31] A. G. Khovanskii, Newton polyhedra and toric varieties, Funkcional. Anal. i Priloˇzen.
11 (1977), no. 4, 56  64.
[32] , A. Knutson, T. Tao,...
[33] M. Kontsevich, Yu. Manin, GromovWitten classes, quantum cohomology and enu
merative geometry, Comm. Math. Phys. 164 (1994), 525562.
[34] M. Kontsevich, Y. Soibelman, Homological mirror symmetry and torus ﬁbrations,
Symplectic geometry and mirror symmetry (Seoul, 2000), 203–263, World Sci. Pub
lishing, River Edge, NJ, 2001.
[35] M. Kontsevich, Y. Soibelman, Aﬃne structures and nonarchimedean analytic spaces,
http://arxiv.org math.AG/0406564.
[36] N. H. Kuiper, Sur les surfaces localement aﬃnes, G´eom´etrie Diﬀ´erentielle, Colloq.
ICNRS, CNRS, Paris 1953, 79–87.
[37] V. P. Maslov, New superposition principle for optimization problems, in Seminaire
sur les Equations auc D´eriv´ees Partielles 1985/6, Centre Math´ematiques de l’
´
Ecole
Polytechnique, Palaiseau, 1986, expos´e 24.
[38] G. Mikhalkin, Real algebraic curves, moment map and amoebas, Ann. of Math. 151
(2000), 309  326.
[39] G. Mikhalkin, H. Rullg˚ard, Amoebas of maximal area, Intern. Math. Res. Notices 9
(2001), 441451.
[40] G. Mikhalkin, Amoebas of algebraic varieties, a report for the Real Algebraic
and Analytic Geometry congress, June 2001, Rennes, France, http://arxiv.org
math.AG/0108225.
[41] G. Mikhalkin, Decomposition into pairsofpants for complex algebraic hypersurfaces,
http://arxiv.org math.GT/0205011 Preprint 2002, to appear in Topology.
[42] G. Mikhalkin, Counting curves via lattice paths in polygons, C. R. Math. Acad. Sci.
Paris 336 (2003), no. 8, 629634.
[43] G. Mikhalkin, Enumerative tropical geometry in R
2
, J. Amer. Math. Soc. 18 (2005),
no. 2, 313–377.
[44] G. Mikhalkin, Maximal real algebraic hypersurfaces (in preparation).
[45] G. Mikhalkin, A. Okounkov, Geometry of shock curves in toric surfaces (in prepara
tion).
[46] G. Mikhalkin, I. Zharkov, Tropical curves, their Jacobicans and Θfunctions (in prepa
ration).
[47] T. Nagano, K. Yagi, The aﬃne structures on the real twotorus I, Bull. AMS, 79, no.
6 (1973), 1251–1253.
[48] V. Nikulin, I. Shafarevich, book
BIBLIOGRAPHY 51
[49] T. Nishinou, B. Siebert, Toric degenerations of toric varieties and tropical curves,
http://arxiv.org math.AG/0409060.
[50] L. Pachter, B. Sturmfels, Tropical geometry of statistical models, Proc. Natl. Acad.
Sci. USA 101 (2004), no. 46, 16132–16137.
[51] M. Passare, H. Rullg˚ard, Amoebas, MongeAmp`ere measures, and triangulations of
the Newton polytope. Preprint, Stockholm University, 2000.
[52] M. Passsare, A. Tsikh, Passare, Mikael; Tsikh, August Amoebas: their spines and
their contours, Idempotent mathematics and mathematical physics, 275–288, Con
temp. Math., 377, Amer. Math. Soc., Providence, RI, 2005.
[53] I. G. Petrovsky, On the topology of real plane algebraic curves, Ann. Math. 39 (1938),
187209.
[54] J.E. Pin, Tropical semirings, Idempotency (Bristol, 1994), 50–69, Publ. Newton
Inst., 11, Cambridge Univ. Press, Cambridge, 1998.
[55] Ragsdale,...
[56] J. RichterGebert, B. Sturmfels, Th. Theobald, First steps in tropical geometry,
Idempotent mathematics and mathematical physics, 289–317, Contemp. Math., 377,
Amer. Math. Soc., Providence, RI, 2005.
[57] V. A. Rohlin, Congruences modulo 16 in Hilbert’s sixteenth problem, Funkcional.
Anal. i Priloˇzen. 6 (1972), no. 4, 5864.
[58] L. Ronkin, On zeroes of almost periodic functions generated by holomorphic functions
in a multicircular domain, to appear in ”Complex Analysis in Modern Mathematics”,
Fazis, Moscow, 2000, 243256.
[59] H. Rullg˚ard, Stratiﬁcation des espaces de polynˆomes de Laurent et la structure de
leurs amibes, C. R. Acad. Sci. Paris, S´erie I, 331 (2000), 355358.
[60] H. Rullg˚ard, Polynomial amoebas and convexity, Preprint, Stockholm University,
2001.
[61] E. Shustin, A tropical calculation of the Welschinger invariants of real toric Del Pezzo
surfaces, http://arxiv.org math.AG/0406099.
[62] E. Shustin, Z. Izhakian, A tropical Nullstellensatz, http://arxiv.org
math.AC/0508413.
[63] D. Speyer, Horn’s problem, Vinnikov curves, and the hive cone, Duke Math. J. 127
(2005), no. 3, 395–427.
[64] D. Speyer, B. Sturmfels, The tropical Grassmanian, Adv. Geom. 4 (2004), no. 3,
389–411.
[65] D. Speyer, B. Sturmfels, Tropical Mathematics, http://arxiv.org math.CO/0408099.
[66] B. Sturmfels, Solving systems of polynomial equations, CBMS Regional Conference
Series in Mathematics, 97 Providence, RI, 2002.
[67] W. Thurston, book.
[68] O. Ya. Viro, Patchworking, 1979.
[69] O. Ya. Viro, Real plane algebraic curves: constructions with controlled topology,
Leningrad Math. J. 1 (1990), no. 5, 10591134.
[70] O. Ya. Viro, Dequantization of Real Algebraic Geometry on a Logarithmic Paper,
Proceedings of the European Congress of Mathematicians (2000).
[71] G. Wilson, Hilbert’s sixteenth problem, Topology 17 (1978), 5373.
Contents
Chapter 1. Introduction 1. Overview 2. The tropical semiﬁeld T 3. The aﬃne space Tn and the torus (T× )n ≈ Rn 4. Integer aﬃne structures on smooth manifolds 5. Morphisms and isomorphisms of integer aﬃne manifolds 6. Examples of integer aﬃne surfaces 7. Integer aﬃne manifolds with corners 8. Tropical projective spaces Chapter 2. Some (semi)algebraic notions 1. Tropical algebras 2. Examples 3. Spectra of tropical algebras 4. Quotient semiﬁelds 5. Aﬃne and convex functions in a tropical algebra 6. Aﬃne structure resulting from the semialgebraic data 7. Regular functions and tropical schemes 8. Regular maps Chapter 3. Hypersurfaces and complete intersections in Tn 1. Integer aﬃne manifolds as tropical schemes 2. Hypersurfaces in Tn 3. Lines in the plane 4. Curves in the plane 5. Surfaces in TP3 6. Complete Intersections 7. Balancing condition Chapter 4. Chapter 5. Bibliography Tropical varieties Tropical equivalence 1 1 2 3 4 6 8 10 12 15 15 16 17 20 22 23 25 26 29 29 29 32 38 42 44 44 45 47 49
iii
CHAPTER 1
Introduction
1. Overview Algebraic Geometry provides a uniform approach to some topologically very distinct situations. As an example, let us consider a line in the aﬃne 2plane. Topologically this setup only makes sense if we ﬁx the ground ﬁeld, i.e. the possible values for the coordinates in the 2plane. If the ground ﬁeld is R we have the “most classical” situation: the plane is indeed a real plane R2 and the line is a real line R. For the other choices of ground ﬁelds the topological picture is diﬀerent, e.g. the complex plane C2 is a 4manifold while over ﬁnite ﬁelds we do not have any interesting topology at all. In the same time despite such diﬀerences the behavior of lines remain the same. Namely, via any pair of distinct points in the plane we can draw a unique line. Also, any pair of lines intersect in a single point (unless they are parallel). This behavior is dictated by the algebra of linear equations.
Figure 1. The three new intersection points are collinear according to the Fano axiom Some other properties of lines in the plane depend on the choice of the ground ﬁeld. A famous example is the Fano axiom. Given any quadruple of distinct points in the plane we may consider the triple of points obtained as
1
This axiom clearly does not hold for C or R. Meanwhile. The tropical semiﬁeld T Definition 1. More general.. in tropical geometry we may ﬁnd reﬂections of properties from rather diﬀerent ﬁelds with diﬀerent algebraic origins. “(a + b) + c” = “a + (b + c)”. “ab” = “ba”. The Fano axiom states that the resulting three points are collinear. When we pass to tropical geometry the ground ﬁeld gets replaced with the tropical semiﬁeld T (which we introduce in the next section) with limited arithmetics and algebra. such geometric objects as points. The goal of this book is to justify this statement. In the same time. are perfectly welldeﬁned. Example 1.2. Its multiplicative group is the group of positive numbers R>0 . It is easy to check that the usual commutativity. the Fano axiom still holds in tropical geometry. associativity and the distribution law hold in tropical arithmetics. The quotation marks are used to signify that the arithmetic operations we are referring to are tropical. but it holds for the ﬁelds of characteristic 2. etc. Namely. it becomes no longer clear how to even deﬁne the characteristic of T.3. lines. E. geometry not only remains equally transparent. Definition 1. A commutative semiring is a set equipped with commutative and associative operations of addition and multiplication so that the distribution law holds while the addition and multiplication operations both have neutral elements.g.2
1. b} and “ab” = a + b. b. The semiﬁeld introduced in the following deﬁnition is crucial for this book. INTRODUCTION
the intersections of the pairs of lines corresponding to all possible choices of two disjoint pairs among the four initial points. c ∈ T. 2. The tropical semiﬁeld T is the set R ∪ {−∞} equipped with the following two arithmetic operations called tropical addition and tropical multiplication. In particular. If a. we have “a + b” = “b + a”. “(ab)c” = “a(bc)” and “a(b + c)” = “ab + ac” for any a. The nonnegative numbers R≥0 equipped with the usual addition and multiplication form a semiﬁeld. A commutative semiring R is called a semiﬁeld if the nonzero elements of R form a group (denoted with R× ) with respect to multiplication. b ∈ T we set “a + b” = max{a. The element −∞ = 0T is the additive zero while 0 = 1T is the multiplicative
. but it gets more explicit and visual. Most algebraic constructions are obstructed by the absence of subtraction in T.1.
Note that the semiﬁeld T has a natural (Euclidean) topology coming from the identiﬁcation of T with the halfopen inﬁnite interval [−∞. For that we restrict our attention to the torus (T× )n .. Let us ﬁnd the geometric structure on Tn that would enable us to distinguish tropical polynomials from other continuous functions without a reference to arithmetic operations in T. Definition 1.
where aj1 . +∞). the indices jk are positive integers and the sum is ﬁnite. Indeed. xnn ”. .4. b ∈ T× = T {−∞}. 3. +∞) is generated by the sets {x ∈ T  x > a} and {x ∈ T  x < b} for a.. However.jn j j aj1 .
. A tropical polynomial f : Tn → T is a function given by f (x1 .jn x11 . This deﬁnition immediately gives the topology on Tn . for any a ∈ T. a} = a. “1T b” = 0 + b = b. the nonzero elements T× = T {−∞} form a group (isomorphic to R) with respect to multiplication and we have tropical division “a/b” = a − b as long as b = −∞. The aﬃne space Tn and the torus (T× )n ≈ Rn We deﬁne the tropical aﬃne nspace as a topological space by Tn = [−∞. +∞). Indeed the inequality a ≤ b for a. b ∈ T× = (−∞. . +∞) = Rn ⊂ Tn . . T is only a semigroup with respect to addition. THE AFFINE SPACE Tn AND THE TORUS (T× )n ≈ Rn
3
unit. xn ) = “
j1 ..jn ∈ T.. . Accordingly. On the other hand.. b ∈ T is equivalent to the identity “a+b=b”. This topology is natural from the algebraic point of view. The algebrogeometric structure is given by regular functions on Tn which come from tropical polynomials. we deﬁne the ntorus there (T× )n = (−∞.. . the Euclidean topology on [−∞. +∞)n . This property makes tropical subtraction impossible.. . “0T + a” = max{−∞.. in contrast with the classical addition the tropical addition is idempotent: “x + x” = x. In addition we have “ − ∞a” = −∞ for any a ∈ T. Each inequality can be rephrased in agrebraic terms.3.
Here a map f : Rn → Rm is called integer aﬃnelinear if it is a composition of a Zlinear map Rn → Rm (i. [γ]). Furthermore.. The path γ([0. y ∈ Uβ ⊂ M and γ : [0. . Φγ depends only on the relative homotopy class of βα the path γ. j = 0. . The map f is called an integer aﬃnelinear transformation of Rn if it is invertible in the class of integer aﬃnelinear maps (note that the invertibility implies that m = n). xjn ”. An integer aﬃne structure on M consists of an open covering Uα and charts φα : Uα → Rn such that for each α.
. 1]) = ∅ for j > 0 so that α0 = α and αk = β.. Integer aﬃne structures on smooth manifolds Definition 1.. β and γ but not on the choice βα of Uαj . The manifold M equipped with such structure is called an integer aﬃne manifold.. . the slope of this function is (j1 . n 1
j1 . . INTRODUCTION
Note that if x ∈ T× then negative powers “x−k ” = “ x1k ” = −kx also make sense. Thus if we ﬁx x and α ˜ then we get a welldeﬁned map δ : M → Rn by setting δ(y. Thus we also have the Laurent polynomials (T× )n → T deﬁned by “ aj1 . jn ) and thus it is integer.. . 4.jn xj1 . jk ∈ Z and the sum is still ﬁnite.5. .4
1.. Each monomial “aj1 .jn xj1 . Furthermore. . . xjn ” = j1 x1 + · · · + jn xn n 1 is an aﬃnelinear function in (T× )n = Rn . . where [γ] is universal covering π : M the relative homotopy class of a path from x to y. 1]) can be covered by a ﬁnite number of the charts Uαj . Namely. βα where Uβ is chart containing y. if x ∈ Uα ⊂ M. ˜ Recall that if we ﬁx x ∈ M then a point in the total space M of the ˜ → M corresponds to a pair (y. We can make sure that Uαj−1 ∩ Uαj ∩ γ([0.jn
where aj1 .. 1] → M is a continuous path connecting x and y then we have the map Φγ : Rn → Rn αβ deﬁned as follows.e.. As with all geometric structures of such kind we have the developing map.. a map given by m × n matrix with integer values) and a translation by an arbitrary vector in Rm . . The geometric structure that underlies such aﬃnelinear functions is the integer aﬃne structure.jn ∈ T. βα It is easy to see that Φγ depends only on α. . β the overlapping map φβ ◦ φ−1 can be obtained α as the restrictions of an integer aﬃnelinear transformation Φβα : Rn → Rn . Let M be a smooth ndimensional manifold. k.. Then we deﬁne Φγ = Φαk αk−1 ◦ · · · ◦ Φα1 α0 . [γ]) = Φγ ◦ Φα .
Recall that the diﬀerential of the integer aﬃnelinear transformations in Rn is deﬁned over Z. The proposition easily follows from the observation that the universal covering of M × N can be obtained by taking the product of the universal coverings for M and N .6. Thus for any integer aﬃne (smooth) manifold M and any point x ∈ M we have a welldeﬁned integer lattice in the tangent space Tx M . Furthermore. these functions always correspond to some tropical monomials in any chart. The map δ is called the developing map. Locally such choice of lattice corresponds to ﬁnding n linearly independent vector ﬁelds on Rn . Clearly.8. The integrality condition is the (pairwise) commuting of these vector ﬁelds. These are the functions obtained from linear maps Rn → R deﬁned over Z (i. Proposition 1.e. The product M × N is complete if and only if both M and N are complete. For Rn we have a notion of aﬃnelinear functions with integer slopes or simply integer aﬃnelinear functions. Conversely. INTEGER AFFINE STRUCTURES ON SMOOTH MANIFOLDS
5
Definition 1. such that the image of the integer lattice Zn ⊂ Rn is integer) after adding an arbitrary constant. Proof. By deﬁnition it is a function that corresponds to an aﬃnelinear function with integer slope on Rn in each chart.4. Thus geometrically. the pullback of an integer aﬃnelinear function under an integer aﬃnelinear map Rn → Rn is another integer aﬃnelinear function on Rn . Taking the maximal value of integer aﬃnelinear function produces tropical (Laurent) polynomials. Clearly. if we have a smooth manifold with a coherent choice of integer lattice in the tangent bundle then it does not necessarily come locally from the tautological integer aﬃne structure on Rn as this is a subject to certain integrality condition. Definition 1. for any open subset U of an integer aﬃne manifold M we have a welldeﬁned notion of an integer aﬃnelinear function f : U → R. The integer aﬃne structure on a smooth manifold is called complete if the developing map is proper. By construction. the developing map is always an open embedding. the product M × N of two integer aﬃne manifolds M and N has a natural integer aﬃne structure. While the choice of presentation as a tropical monomial depends on the choice of chart.
. These functions correspond to tropical monomials.7. [γ]) ∈ Rn does not depend on the ambiguity in the choice of β. Thus an integer tangent vector is mapped to an integer tangent vector. the tropical structure on (T× )n may be rephrased as an integer aﬃnelinear structure on Rn . This lattice varies smoothly from point to point. As it is easy to see the value δ(y.
e. Consider a morphism f : M → N of integer aﬃnelinear varieties. a point x ∈ M and any pair of charts Uα ∋ x. A map f : M → N is called an integer aﬃnelinear map (or just morphism of integer aﬃnelinear varieties) if it is smooth and its diﬀerential maps any integer vector tangent to M at any point x to an integer vector (tangent to N at f (x)). It can be decomposed into a sum of integer vectors vj . All isomorphisms of integer aﬃne manifolds M form a group. We see that integer aﬃnelinear smooth manifolds locally can be considered as examples of tropical varieties (as they locally coincide with (T× )n . Let v ∈ Rn be any vector. This allows to connect 0 and v with the broken path such that each of its segment is parallel to one of the integer vectors vj .e.6
1. By the continuity argument the integrality assumption also implies that (df )x = (df )0 for every x ∈ Rm . Applying a translation if needed we may assume that f takes the origin of Rm to the origin of Rn .
Proof. every point x admits a neighborhood U ∋ x such that all translates by the elements of G are disjoint) then it gets a natural integer aﬃne structure from M . If the quotient M/G by a subgroup G of this group is a manifold (which is the case if this subgroup acts in a properly discontinous fashion. i. smooth manifolds with a coherent (but not necessarily integrable) choice of integer lattice in the tangent bundle can be considered as examples of almost tropical varieties. Then we say that M and N are isomorphic as integer aﬃne manifolds. Similarly. Vβ ∋ f (x).10.9. we have f (v) = aj (df )0 (vj ) = (df )0 (v).
. to Uα ∩ f −1 (Vβ )) of an integer aﬃne linear map Rm → Rn . We claim that the diﬀerential (df )0 of such f at the origin coincides with the map itself (after the natural identiﬁcation of Rm with the tangent space at its origin). Therefore. The map ψβ ◦f ◦φα is the restriction to the domain where it is deﬁned (i.
A map f : M → N is called an isomorphism (or a symmetry) of integer aﬃne manifolds if it is invertible and both f and f −1 are morphism. It suﬃces to show that if f : Rm → Rn is a map whose diﬀerential takes integer vectors to integer vectors then f is integer aﬃne linear. INTRODUCTION
Remark 1. 5. v = aj vj with aj ∈ R. The integrality assumption assures that (df )0 is deﬁned over Z.
−1 Proposition 1. Morphisms and isomorphisms of integer aﬃne manifolds Let M and N be integer aﬃne varieties of dimensions m and n.
11. Of course. .
closed “horizontal” circles form a ﬁbration
no rational slope curve is closed
Figure 2. b. The action of the whole group is not properly discontinous. s) coincide so that we get a closed circle. Definition 1. The easiest properly discontinous subgroup is the lattice generated by translation in linearly independent directions. Example 1. c.12. We say that it has rational slope if it is tangent to an integer vector at its every point. being “horizontal” is not an intrinsic condition in M and depends on the choice of chart to R2 . s). Diﬀerent integer aﬃne structures n S 1 × S 1 . Definition 1.g. We say that v is a primitive vector if it is integer in Tx M and cannot be presented as a nontrivial positive integer multiple of another integer tangent vector. d ∈ R with ad−bc = 0 the resulting quotients are integer aﬃne manifolds. E. Let C ⊂ M be a curve. All of them are diﬀeomorphic (and diﬀeomorphic to S 1 × S 1 ).13. so we need to restrict to a subgroup.
. But there is also an intrinsic property that holds for these circles. Let C ⊂ M be a curve with rational slope and v ∈ Tx C be a vector tangent to x ∈ C. This property does not depend on the choice of the charts while being “horizontal” in one chart ensures rational slope in others. The condition b = 0 ensures that the points (t. Alternatively we may deﬁne such curves as those which have rational slope in each chart. The group of its symmetries is the group of all integer aﬃnelinear transformations of Rn . s) and (t + a. But there are other choices of subgroups. Let M be the quotient of the plane a c R2 by the subgroup Λ generated by the vectors . b d For any choice of a. where for each circle s ∈ R is ﬁxed while t ∈ R varies. also using nontrivial linear parts (from GLn (Z)). if b = 0 then M is foliated by closed “horizontal” circles obtained as the quotient (t. Rn is an aﬃne integer manifold tautologically. Consider the following examples of integer aﬃne manifolds obtained as the quotients of R2 . However they are not all isomorphic as integer aﬃne manifolds.5. MORPHISMS AND ISOMORPHISMS OF INTEGER AFFINE MANIFOLDS
7
Clearly.
Indeed. With the help of the primitive vectors we may deﬁne intrinsic length of a curve C with rational slope. k ∈ Z.16. Examples of integer aﬃne surfaces Example 1. We return to Example 1.
. all their lengths coincide and equal to a. equal to 2a and d respectively. 6.
In charts the intrinsic length can be obtained by taking the Euclidean length of γ and dividing it by the Euclidean length of a primitive vector parallel to γ.15. a > 0. Clearly. b. the composition of the reﬂection at the xaxis with a translation by 0 0 . All vectors tangent to C form a 1dimensional real vector space while the integer vectors form a lattice isomorphic to Z ⊂ R. But then n(ja+kc) = m(jb+kd) which contradicts +k j d b to the linear independence over Q. d ∈ R linearly independent over Q. All of them. Let B be a translation by d The quotient of R2 by the properly discontinous subgroup G of symmetries generated by R and B is a Klein bottle.14.11. Definition 1. Then a multiple of n c a . c. Note that for this example we have foliations both by horizontal and vertical circles. see Figure 3 for one of the “cores”. suppose that on the contrary we can ﬁnd such a circle and it is parallel at its every point (in a chart m ∈ obtained by reversing the quotient projection) to an integer vector n m is proportional to an integer linear combination R2 . except for two horizontal “core” circles have the same (intrinsic) length. the other is the result of identiﬁcation of the horizontal sides of the rectangle. j. The two horizontal “cores” have lengths equal to a.8
1. For any x ∈ C the primitive tangent vector is unique up to sign. d > 0. If they are the only closed curves with rational slope on M then a is the isomorphism invariant. The (intrinsic) length of γ is the integral
γ
α. INTRODUCTION
Proposition 1. Let R : R2 → R2 be the gliding reﬂection obtained by a . Proof. If b = 0 we can measure the length of the “horizontal” circles. a 1form α on C that takes value ±1 on primitive vectors is unique up to sign. Let γ ⊂ C be an arc on C. Indeed. We may also choose a. Then no circle in M can have rational slope.
Let R be the quotient of R2 by the group generated by A. A Klein bottle with horizontal circles. We identify the top side of the strip with the bottom so that the corresponding bases match. e. such conjugation does not change the isomorphism type of M . cf. Figure 3.6. EXAMPLES OF INTEGER AFFINE SURFACES
9
horizontal cirles of length 2a
core horizontal cirle of length a
Figure 3. Proposition 1. [49]. Proof.g. Consider now a radically diﬀerent example an integer aﬃne structure on a torus. µ = R then λ = µ as L is real.
. Note also that the vertical circles are dual to the ﬁrst StiefelWhitney class. we construct a nontrivial integer aﬃne annulus. Note that the shear transformation is the only possible linear part for an orientationpreserving deck transformation R2 → R2 corresponding to an integer aﬃne linear transformation as shown in the following proposition.18. From Euclidean planimetry we know that A must be a translation. by 0 1 The surface R is an integer aﬃne annulus that is diﬀerent from the quotient of R2 by the group generated by any translation. Example 1. Let λ and µ be the eigenvalues of L. If λ. First. Of course. Therefore d is an isomorphism invariant of M .17. If an integer aﬃne linear transformation A : R2 → R2 is ﬁxed point free and orientationpreserving then its linear part L has both eigenvalues equal to 1. Let A : R2 → R2 be a map obtained as the composition 0 . ¯ Therefore λ2 = µ2  = 1 and L is an orthogonal matrix in some basis so that A is a metric preserving transformation. [67]). but the slopes of the sides of this parallelogram will still have rational slope. The manifold M can be obtained from the a × d rectangle by identifying the oriented opposite sides. Since L ∈ SL2 (Z) and preserves orientation we have λµ = 1. So far our examples look very similar to examples of surfaces with Euclidean structure (cf. Inside R we have immersed curves with rational slope that have selfintersections as shown on Figure 4. and the linear transformation of R2 deﬁned of a translation by d 1 1 . d > 0. The conjugation of G by an element from GL2 (Z) results in replacing the rectangle by a parallelogram.
Suppose that the translational part a of A is given by . Let T be the quotient of R2 by the group generated by a . [36]. See also [16] for a discussion of aﬃne structures with singularities.10
1. This the transformation A from Example 1. particularly on a torus. But these linear equations clearly have solutions if λ = 0. Remark 1. Just like the torus T from Example 1.
Example 1. Let K be the Klein bottle obtained as the quotient of R2 by the group generated by the transformation A from Example 1. y) such that λx − x = a and b 1 λ y − y = b.17 and a translation by 0 is a compact surface diﬀeomorphic to the torus but not isomorphic to any quotient of R2 by a lattice of translations. An integer aﬃne annulus and two curves with rational slope there. INTRODUCTION
Figure 4. µ ∈ R with µ = λ = λ then we may choose the coordinates in R2 1 so that L is given by (x.19.
.20 admit the same tiling by fundamental domains in R2 shown in Figure 5.21. Any integer aﬃne manifold can also be considered as a real aﬃne manifold as we have embedding GLn (Z) ⊂ GLn (R). Inside T we have immersed curves with rational slope that have selfintersections as in the case of Example 1. See [5]. Examples 1. In fact. 7. λ y).16. This group also acts in a properly discontinuous manner so K is an integer aﬃne surface.19 and 1.20. y) → (λx. [47] for a discussion of real aﬃne structures. Integer aﬃne manifolds with corners While integer aﬃne manifolds are modeled on open sets in Rn the tropical aﬃne space Tn has boundary and corners.17 and the transformation R from Example 1.17.19 the Klein bottle K has immersed curves with rational slope that have selfintersections. Example 1.
1 If λ. It suﬃces to ﬁnd (x.
¯ Proof. Tiling of R2 by fundamental domains for Examples 1. The map Φ is continuous on the domain of its deﬁnition. They allow us to extend the notion of integer aﬃne structure to a larger class of spaces almost by repeating Deﬁnition 1. β the overlapping map φβ ◦ φ−1 can be obtained as the restrictions of a (partially deﬁned) integer α ¯ aﬃnelinear map Φβα : Tn Tn that is deﬁned everywhere of φα (Uα ). If x ∈ Uα ⊂ X then we deﬁne its sedentarity as the sedentarity of its image φα (x) ∈ Tn . .22. Let X be a topological space. This gives us partiallydeﬁned extensions ¯ Φ : Tn Tm ¯ of integer aﬃnelinear maps Rn → Rm . We say that X is an integer aﬃne manifold with corners if X is enhanced with an open covering Uα and charts φα : Uα → Tn such that for each α.7. Let Φ : Rn → Rm be an integer aﬃnelinear map. The sedentarity s(x) of a point x ∈ X does not depend on the choice of the chart Uα . that s(Φβα (x)) < s(x) = k for x ∈ Without the loss of generality we may assume that x1 = · · · = xk = −∞. Proposition 1. . INTEGER AFFINE MANIFOLDS WITH CORNERS
11
Figure 5.
. The image Φ(x) still makes sense as a point in Tm if whenever xj = −∞ the whole jth row of the matrix L is nonnegative. Definition 1. on the contrary.5 Definition 1. . Let x = (x1 .19 and 1.24. xn ) be a point in Tn = [−∞. Here we use the convention “a(−∞)” = −∞ if a > 0 and “0(∞)” = 0. +∞)n . Let L be the linear part of Φ which can be viewed as an (integer) m×n matrix. Let x ∈ Tn Rn be a point of positive sedentarity in Tn .20. Suppose. The tropical aﬃne space Tn is a manifold near its point x if and only if x has sedentarity 0. We call the sedentarity s(x) of x the number of coordinates xj equal to −∞. . We treat such maps as integer aﬃne linear maps between aﬃne tropical spaces.23. Tn .
The (tropical) monomial on U is any morphism U → T. Its image has to contain R as U contains Rn .25.27. Definition 1. . Definition 1. any open subset U of a manifold with corner X is itself a manifold with corners (though not necessarily complete even in the case when the ambient manifold with corners X is complete). Y X Tn ⊃ Uα . 8. But then the ﬁrst k coordinates of Φβα (x) are all equal to −∞ and this supplies a contradiction. Since L is invertible we may also assume without the loss of generality that the top k × k minor ¯ is not degenerate. . Tm ⊃ Uβ . Uβ ∋ f (x) and a map Φ : Tn → Tm .29. Let X and Y be two integer aﬃne manifolds with corners. n. Proof. For us the most important example is that of the tropical projective space. Note that Tn has the tautological structure of an integer aﬃne manifold with corners.26. such that f (t) = (φY )−1 ◦ Φ ◦ φX . Proposition 1. INTRODUCTION
Then we know that the top k rows of the matrix giving the linear part L of ¯ Φβα must consist of nonnegative (integer) numbers. ¯ Proof. α β Note that this is a straightforward extension of the deﬁnition of morphisms of manifolds without corners. Furthermore. The map κ can be lifted to a morphism from the universal ˜ ˜ covering U → T. A map f : X → Y is called a morphism if for every Y X x ∈ X there exists charts Uα ∋ x. we can glue several copies of Tn together to get compact integer aﬃne manifolds with corners. Proposition 1. . The space Xs is a disjoint union of integer aﬃne manifolds of dimension n − s (without boundary or corners).28. Let Xs be the locus of points of sedentarity s in an ndimensional integer aﬃne manifold with corners X.12
1. Tropical projective spaces Consider the set TPn = Tn+1 {0Tn+1 }/ ∼
. Restrictions of the overlapping maps Φβα to the coordinate (n− n (those deﬁned by x s)planes in T j1 = · · · = xjs = −∞) provides the required integer aﬃne structure. . The integer aﬃne structure on a manifold with corners is called complete if every component of Xs is a complete integer manifold for each s = 0. If U is complete then for any monomial κ : U → T we have κ(U ) ⊃ R. Clearly. Definition 1.
g. X × Y is complete if and only if both X and Y are complete. As usual. . .8 we get the following statement. . The map x→( x1 xn . If X and Y are integer aﬃne manifolds with corners then X × Y is also an integer aﬃne manifold with corners. .
. j = 0. n. Furthermore.. (Φjk )l = “ xj ¯ Clearly Φjk is an integer aﬃne map deﬁned on {xj = −∞} ⊂ Tn . .32.8. . . . . . . .. we use the homogeneous coordinate notations x = [x0 : · · · : xn ] ∈ TPn to denote the equivalence class of (x0 . . the integer aﬃne structure induced in the interior of each kface is isomorphic to the tautological integer aﬃne structure on Rk . Clearly the set TPn gets a natural topology of the quotient. it admits a natural structure of an integer aﬃne manifold with corners. . n. −∞) is the origin in Tn+1 and we set (x0 . ) x0  + · · · + xn  x0  + · · · + xn 
provides the required homeomorphism to the standard simplex in Rn (cut ≥0 by the halfspace x1 + · · · + xn ≤ 1).. k = j. Furthermore. A complex smooth toric variety is obtained by gluing several copies of aﬃne spaces Cn (or. To see that we cover TPn with n + 1 open charts Uj = {x ∈ TPn  xj = 0T = −∞}. . . As in the case with projective space there is a sedentaritypreserving homeomorphism with the corresponding polyhedron (see e. The space TPn is homeomorphic to the nsimplex Σn so that a point inside a kface of Σn corresponds to a point of sedentarity n − k. Similarly to Proposition 1. . . . . .Here (φj (x))k denotes the xj kth coordinate of the image φj (x) and the target of φj is the hyperplane Tn ⊂ Tn+1 given by {x ∈ Tn+1  xj = 1T = 0}. yn ) if there exists λ ∈ T× such that xj = “λyj ” = λ + yj for any j = 0. (φj (x))k = “ xk ” = xk − xl . . Figure 6 shows the tropical plane blown up at 6 points which is diﬀeomorphic (as a manifold with corners) to a hexagon. Proposition 1. more generally. The tropical counterparts are obtained by gluing copies of Tk × Tn−k by the maps given by the corresponding tropical monomials. products of aﬃne spaces Ck with tori (C× )n−k ) by maps such that each coordinate is given by a monomial. In a similar way we may construct tropical counterparts of more general toric varieties. Furthermore. . xn ) ∼ (y0 .g. [15]).30. . xn ).. Remark 1.31. j = k are given by xl xk ¯ ” = xl + xk − xj . TROPICAL PROJECTIVE SPACES
13
where 0Tn+1 = (−∞. E. ¯ The overlapping maps Φjk : Tn Tn . Proposition 1. Proof.
. INTRODUCTION
Figure 6. Interior of both polygons are isomorphic to the complete aﬃne space R2 with the tautological integer aﬃne structure. The tropical projective plane and the tropical projective plane blown up at three points.14
1.
2. Implicitly using this proposition we identify T with its image in A. we have 0A = −∞ ∈ A and 1A = 0 ∈ A. “av” = “bv” if a = b and “v0T ” = O. For any f.e. There is a natural embedding ιA : T ⊂ A which respects the semiring addition and multiplication: ιA (“a + b”) = “ιA (a) + ιA (b)”. A tropical algebra A is a semiring (recall according to Deﬁnition 1. i. a ∈ T. Tropical algebras Definition 2. Deﬁne ιA (a) = “a1A ”. h ∈ A then either we have equality g = h or the element f is a zero divisor. ιA (“ab”) = “ιA (a)ιA (b)”. v) → “av”. b ∈ T.CHAPTER 2
Some (semi)algebraic notions
1. v ∈ V . Proof. Note that ιA is an embedding since A is a cone.1 A has an additive zero 0A ∈ A and a multiplicative unit 1A ∈ A) equipped with a Tcone structure compatible with the semiring operations. subject to the following additional property. (a. A Tcone is a set V with a choice of an element O ∈ V called the origin equipped with a product operation T × V → V. In particular. i. such that “(ab)v” = “a(bv)” for any a. To check the converse statement we note that ιA gives a Tcone structure on A by “af ” = “ιA (a)f ”. g.e. Proposition 2. g. We have ιA (“a + b”) = “(a + b)1A ” = “a1A + b1A ” = “ιA (a) + ιA (b)” and ιA (“ab”) = “(ab)1A ” = “a(b1A )” = “(a1A )(b1A )” = “ιA (a)ιA (b)”. ˜ ˜ there exists f ∈ A such that “f f ” = 0A .3.
15
. ιA (−∞) = 0A and ιA (0) = 1B . Conversely if A is semiring and ιA : T ⊂ A is such an embedding then A is a tropical algebra as long as “0A f ” = 0A and “ιA (a)f ” = ιA (b)f for any f ∈ A and a = b ∈ T.1. such that “a(f g)” = “(af )g” and O = 0A . h ∈ A if “f g” = “f h” for f. Definition 2. v ∈ V .
The addition and multiplication on O(Tn ) are pointwise tropical addition and multiplication. . Similarly.e.4. an epimorphism is a surjective homomorphism and a monomorphism is an injective homomorphism. .
.5.. . Let A and B be two tropical algebras. .. . Consider the semiring
k
T[x] = {“
j=0
aj xj ”  aj ∈ T. xn ] of formal tropical polynomials in n variables “ aj1 . .7. Definition 2. SOME (SEMI)ALGEBRAIC NOTIONS
Definition 2. in addition.jn )∈J
where aj ∈ T and J is a ﬁnite subset of (N ∪ {0})n . n 1
(j1 .jn xj1 . xn ) ∈ Tn . . Consider the tropical algebra O(Tn ) of functions Tn → T.. These polynomials can be added and multiplied according to formal polynomial laws (recall that −∞ is our additive zero) and form The embedding ι : T ⊂ T[x] is tautological a → a. x → f (x). . where f ∈ T[x1 . . . xn ) ∈ Tn and j = (j1 . . . g ∈ B such that “f g” = 0B we have either f = 0B or g = 0B .. As usual. while constant functions give the embedding T → O(Tn ). .e. A tropical algebra B is called an integral domain if it does not have zero divisors. . φ(“ab”) = “φ(a)φ(b)” and. . is another example of tropical algebra. . xn ] and x = (x1 . . 2. xjn . i. jn ) ∈ Zn then xj = xj1 . . b ∈ A we have φ(“a + b”) = “φ(a) + φ(b)”. . xjn ”.. A map φ : A → B is called a homomorphism of tropical algebras (or just a Thomomorphism) if for any a. For convenience we will use multiindex notations for multivariable monomials: if x = (x1 . . . Examples Example 2. Elements of O(Tn ) are called regular functions on Tn . φ is identity on T. the diagram T? ?
/A ??ιB ?? φ ?
ιA
B is commutative. the semiring T[x1 .6. . an isomorphism is an invertible homomorphism. . n 1 Example 2.16
2. i. . k ∈ N ∪ {0}}
of formal tropical polynomials in one variable x. for any f..
. . .11. Let B1 and B2 be two other tropical algebras and φj : A → Bj be two epimorphisms. Consider the algebra T[x1 . Definition 2. This algebra is ﬁnitely generated by 2n generators x1 . xn ] → A for some n ∈ N. Example 2. . . . . E. SPECTRA OF TROPICAL ALGEBRAS
17
The tautological map τ : T[x1 . fn ∈ A such that any f ∈ A can be presented in the form
n
f =“
j=1
aj fj ”.
j∈J
. .
The elements f1 . . . . n 1 3. 1 Definition 2. . Note that τ −1 (−∞) = {−∞}. . x−1 . x−1 . . fn are called generators of A. We have Specm (T) = Spec(T) = {pt}. Equivalently. the only tropical epimorphism of T to another tropical algebra is the identity T → T. . . where aj ∈ T and J is a ﬁrent polynomials in n variables “ nite subset of Zn . τ is not a monomorphism (unless n = 0). . . Spectra of tropical algebras Let A be a tropical algebra. Definition 2. . where B is an integral domain. The spectrum Spec(A) is the set of all epimorphisms A → B up to the equivalence above. xn . .10. xn . . The maximal spectrum Specm (A) is the set of all Thomomorphisms A → T. . . . . xn ] → O(Tn ) is an epimorphism. .8. Example 2. Nevertheless. x−1 .9. If U ⊂ Specm (A) we denote Funct(U ) = {g : U → T  ∃f ∈ A : ∀x ∈ U g(x) = f (x)}. in this case we have (depending on x1 ∈ T) either “ax1 ” = x1 + a ≤ 2x1 = “0x2 ” or “ax1 ” ≤ 0. A tropical algebra A is called ﬁnitely generated if there exist f1 . . x−1 ] of Laun 1 aj xj ”.3. .12.g. Indeed. “0x2 + ax1 + 0” = “0x2 + 0” 1 1 whenever a ≤ 0 ∈ T. . . A is ﬁnitely generated if there exists an epimorphism T[x1 . . If f ∈ A and x ∈ Specm (A) then we deﬁne the value f (x) ∈ T as the image of f under the epimorphism x : A → T. .
Since a is an epimorphism of tropical algebras.13.16. It induces an embedding Specm (T) ⊂ Specm (A) (cf. this subset can be naturally identiﬁed with Specm (B). Proposition 2.18. Clearly. Definition 2. We say that the tropical algebra A is reduced if the reduction epimorphism is an isomorphism. x is contained in F if there exists y ∈ Specm (B). In other words. Definition 2.15. Definition 2.19. if x : Funct(X) → T is a tropical epimorphism then the composition of the evaluation epimorphism A → Funct(X) and x is contained in Z. In other words. Proof. x : A → T.18
2. we say that x is contained in F if x is contained in the image F ∗ : Specm (B) → Specm (A). In particular. F : A → B. and F ∈ Spec(A). More generally. If a : A → B is an epimorphism of tropical algebras then a∗ is an injection. thus the image of ιA corresponds to the constant functions on Specm (A). SOME (SEMI)ALGEBRAIC NOTIONS
Pointwise addition and multiplication turn Funct(U ) to a tropical algebra. Let a : A → B be a homomorphism of tropical algebras. Proposition 2. Thus F deﬁnes a subset of Specm (A).14. so is x◦a. Example 2.11) that corresponds to the point x. Since x : A → T is a homomorphism of tropical algebras it is identity on T.17. The evaluation epimorphism A → Funct(Specm (A)) is called the reduction epimorphism. Clearly we have the natural evaluation epimorphism A → Funct(U ). For any a ∈ T and x ∈ Specm (A) we have ιA (a)(x) = a.it implies that x ◦ a maps onto T.
. such that x = y ◦ F . Definition 2.
Example 2. Any x ∈ Specm (A) is an epimorphism A → T. we have the following inclusions corresponding to such embeddings when we pass to considerations of the full spectrum Spec(A). y : B → T. If x = x′ : B → T then there exists f ∈ B such that x(f ) = But then x(a(g)) = x′ (a(g)) for any g ∈ A such that a(g) = f . If x ∈ Specm (A). x′ (f ). The induced map a∗ : Specm (B) → Specm (A) is the map which takes an epimorphism x : B → T to x ◦ a : A → T. Proof. A subset X ⊂ Specm (A) is called a basic closed set if every tropical epimorphism Funct(X) → T corresponds to a point of X.
the composition of the evaluation epimorphism A → Funct(X) and x belongs to Xj for every j. As each Thomomorphism A → T also gives a Thomomorphism T[x1 . . Furthermore any closed set in the Euclidean topology is a basic closed set in the spectrum topology. Thus Deﬁnition 2.22. . . Thus F must be closed. We apply this construction in the following deﬁnition. Conversely. Proposition 2. . Suppose that X =
j
Xj and all Xj ⊂ Specm (A) are basic
closed sets. A set X ⊂ Specm (A) is called closed if it can be presented in the form X= Xα . g such that f (y) = g(y) but such that
. SPECTRA OF TROPICAL ALGEBRAS
19
In other words X is closed if the evaluation epimorphism A → Funct(X) deﬁnes X (and not a larger set). A set U ∈ Specm (A) is called open if Specm (A) U is a closed set. If y ∈ F then we may / have two tropical polynomials f. .20. xn ] → T by composition we have the identiﬁcation of Specm (A) and F . . The whole set Specm (A) is an example of a basic open set as it is presented by the identity epimorphism A → A. . . Proposition 2. An intersection of basic closed sets in Specm (A) is a basic closed set. . +∞)n is closed then we may consider the restriction homomorphism T[x1 . . Any tropical epimorphism x : Funct(X) → T can be composed with the restriction epimorphism Funct(Xj ) → Funct(X). .3. if F ⊂ [−∞. Consider Specm (A).21 gives a topology on Specm (A). Recall that any collection of subsets deﬁne a topology as a prebasis. xn ] → Funct(F ). . The spectrum topology on Specm (T[x1 . . an empty set is closed as the parameterizing set J can be empty. Since all tropical polynomials are continous functions any accumulation point of F also deﬁnes a Thomomorphism A → T. It follows immediately from this deﬁnition that the intersection of any number of closed sets is closed and that the union of a ﬁnite number of closed sets is open as well. . If F ⊂ Specm (T[x1 . xn ]) is a basic closed set then it corresponds to an epimorphism T[x1 . .21. Therefore. . Proof. xn ] → A. . xn ]) coincides with the Euclidean topology on [−∞. Proof. . We refer to this topology as the spectrum topology on Specm (A) to distinguish it from a diﬀerent topology (the Zariski topology) which we introduce later on. +∞)n . . Definition 2.
α∈J
where J ∋ α is any parameterizing set and each Xα is the union of a ﬁnite number of basic closed sets. . Furthermore.
Take a product of the lefthand and the righthand sides of our hypotheses “f1 g2 ” = “f2 g1 ” and “f2 g3 ” = “f3 g2 ”.25. x−1 . g2 ) if f1 g2 = f2 g1 ∈ A. Lemma 2. Since A is a tropical algebra either the statement of the lemma holds or “f2 g2 ” is a zero divisor (cf. g1 )(f2 . .23.
. with an equivalent pair.20
2. . g1 ) ∼ (f2 . gj ). f1 = 0A and f3 = 0A which also veriﬁes the statement of the lemma. . “g1 g2 ”) and multiplication “(f1 . . SOME (SEMI)ALGEBRAIC NOTIONS
f (x) = g(x) for any x ∈ F . Furthermore any closed set in the Euclidean topology is a basic closed set in the spectrum topology. g Elements of the semiﬁeld Q are called rational functions associated with A. are such that “f1 g2 ” = “f2 g1 ” and “f2 g3 ” = “f3 g” 2 then “f1 g3 ” = “f3 g1 ”. The quotient semiﬁeld Q = Rat(A) of a tropical integral domain A is the set of pairs (f. j = 1. x−1 ]) n 1 coincides with the Euclidean topology on Rn . If A is a tropical integral domain and fj . “f2 g2 ”). Deﬁnition 2. We get “f1 g2 f2 g3 ” = “f2 g1 f3 g2 ”. Proof. j = 1. . . 2. Definition 2. Similarly we get the following proposition. f. 4. In accordance with the classical case we denote (f. g ∈ A. g) ∈ Q with “ f ”. . gj ∈ A. +∞)n does not give a homomorphism from Funct(F ) unless y ∈ F .2). Since A is an integral domain and g2 = 0A we have f2 = 0A . Lemma 2.24. in turn. Thus a point y ∈ [−∞. g2 )” = (“f1 g1 ”. 2. We equip Q with operations of addition “(f1 . xn . g1 ) + (f2 . g = 0A up to the following equivalence relation (cf. Then. The spectrum topology on Specm (T[x1 . 3. From now on we suppose that a tropical algebra A is an integral domain and Q = Rat(A) is its quotient semiﬁeld. Quotient semiﬁelds As in classical Commutative Algebra if A is a tropical algebra A which is an integral domain then we can make a semiﬁeld Q ⊃ T out of T by allowing fractions. . g2 )” = (“f1 g2 + f2 g1 ”. gj = 0A . Proposition 2. It is easy to see that the equivalence class of the results of these operations does not change if we replace (fj . g).24 (f1 .
Proposition 2. The required map is induced by the composition A → QA → QB . QUOTIENT SEMIFIELDS
21
Proposition 2. . . The map q : A → Q. Since we have the inversion operation g 1 ”=“ ” “ f /g f Q is a semiﬁeld. Proof. A A a = b. Proof. Q is a semiﬁeld that contains A as a subsemiring. In particular. I. By Proposition 2. Q is a semiring since A is a semiring. A Proof. .e. xn . Since a homomorphism x : A → T can be factorized through A A q : A → Q it can be extended to “ 1f ”. We set H(“ f ”) = “ h(f ) ”. . b ∈ A then. We denote the set of all ﬁnite points in Specm (A) with (Specm (A))◦ .31.27. We have “ 1f ”(x) = “ f1T ” ∈ T. Its quotient semiﬁeld coincides with the quotient semiﬁeld of the
j −1 algebra T[x1 . g h(g) The homomorphism q from Proposition 2.4.
. xn ) and call its elements tropical rational functions in n variables. If Φ : A → B is a homomorphism then Φ∗ ((Specm (B))◦ ) ⊂ (Specm (A))◦ .29.30.
x−1
Proposition 2. this gives an embedding T ⊂ Q which makes Q a tropical algebra and q a tropical algebra monomorphism. . . .26 deﬁnes a map q ∗ : Specm (Q) → Specm (A) by taking x : Q → T to x ◦ q : A → T. . a. Example 2. .26. The embedding T ⊂ A ⊂ Q makes Q into a tropical algebra. . Clearly. xn ] from Example 2. xn ] as “ 1T ” ∼ “ 1T ” (recall that 1T = 0). Proof. Proposition 2. .28. x−1 . if x ∈ Specm (A) is ﬁnite and f = −∞ ∈ A then f (x) = −∞ ∈ T.27 we have the induced map of the spectra of QA and QA Specm (B)◦ → Specm (A)◦ that agrees with Φ∗ since H is an extension of h.7. by deﬁnition. . Nonzero elements of A have ﬁnite values at ﬁnite points of the spectrum. . (x) therefore f (x) ∈ T× . Definition 2. Any homomorphism h : A → B of tropical algebras naturally extends to a homomorphism H : Rat(A) → Rat(B). Consider the tropical algebra T[x1 . q(f ) = “ 1f ” is a monomorphism of tropical algebras. If “ 1a ” is equivalent to “ 1b ”. A point x ∈ Specm (A) is called ﬁnite if x ∈ q ∗ (Specm (A)). . . We 1 xj denote the resulting semiﬁeld in these cases with T(x1 . .
36. Definition 2. Elements of Aﬀ(A) are called aﬃne functions associated with A. Recall that Q∗ = Q {0} is an abelian group with respect to tropical multiplication. There exists a primitive aﬃne functions h ∈ A such that “f + h” ∈ A is a primitive aﬃne function while “a + h”. Since Conv(A) generates the semiﬁeld Q this gives an embedding Specm (Q) ⊂ Specm (Conv(A)). Since we have the inclusion Conv(A) ⊂ Q any epimorphism Q → T determines an epimorphism Conv(A) → T by taking restriction. Aﬃne and convex functions in a tropical algebra Definition 2. The group Aﬀ(A) corresponds to the group of all aﬃnelinear functions
. . b ∈ A.29 and Corollary 2. b ∈ Conv(A) are such that f = a + b then either f = a or f = b. Definition 2. An element of Q is called convex if it is a tropical sum of elements from Aﬀ(A) ⊂ Q. • the quotient group Aﬀ(A)/T∗ is a free abelian group of ﬁnite rank. We say that a tropical algebra A is tame if the following conditions hold: • for every c ∈ T∗ the image ιA (c) ∈ A is a primitive aﬃne function (we call such functions constant) so that T∗ ⊂ Aﬀ(A) is a subgroup. Corollary 2. If A is tame then Specm (Conv(A)) = Specm (Q). Proof. Proposition 2. To ﬁnish the proof we need to show that any epimorphism x : Conv(A) → T can be extended to Q. and we have either f = a or f = b.32. . .34. a. h ∈ Conv(A) such that f = “ h ”. Proof. Example 2.35. Denote with Aﬀ(A) the subgroup of Q∗ generated by all primitive aﬃne functions in A ⊂ Q. Since Aﬀ(A) provides a set of generators for the semiﬁeld QA any element in QA can be written as a ratio of two polynomial functions from the elements of Aﬀ(A).22
2.38. • the subset Aﬀ(A) generates QA in the semiﬁeld sense. .37. The free tropical algebra A = T[x1 . An element f in a tropical algebra A ⊃ T is called a primitive aﬃne function if f = −∞ and whenever we have f = a + b. All convex functions form a semiring Conv(A) ⊂ Q. Proof. SOME (SEMI)ALGEBRAIC NOTIONS
5. If f ∈ Aﬀ(A) and a. If A is tame then for any f ∈ QA there exist funcg tions g. Proposition 2.33. This follows from Proposition 2. We have “a + h + b + h” = “a + b + h” = “f + h” which contradicts to the primitivity of “f + h”. xn ] is tame.36. “b + h” ∈ A.
s = (s1 . Convex functions are ﬁnite tropical sums of elements of Aﬀ(A). . j = 1. . . n. R) ≈ Rn . Suppose that s1 . . As the values of the functions from T[x1 . both “ 1T1T 1 ” and +x 1T “ 1 +x−1 ” are elements of Rat(A). x−1 . . If A is tame then we have a natural embedding (Specm )◦ ֒→ T . However. . s2 ∈ (Specm )◦ . . Proposition 2. The tropical algebra A′ = T[x1 . AFFINE STRUCTURE RESULTING FROM THE SEMIALGEBRAIC DATA
23
f : Rn → R whose slope is integer: f (x) =< s. n 1 This gives a map (Specm )◦ → T .g. x > +t. . . . The embedding Aﬀ(A) ⊂ A generates a homomorphism (1) T[x1 . xn . xn . most of them won’t be useful for us as they’ll be rather far from being a manifold.39. . The tropical semiﬁeld Rat(A) = Rat(A′ ) is itself a tropical algebra. Thus we may treat the ﬁnite part of the maximal spectrum of a tame tropical algebra A as certain (sedentarity 0) points in the aﬃne space associated to Aﬀ(A). . . it is not tame as Aﬀ(Rat(A)) is empty. 1T + x1 1T + x−1 1T + x1 + x−1 + 1T 1T + x1 + x−1 1 1 1
T 1
Thus 1T is not a primitiveaﬃne function in Q. x−1 . . E. . s2 : QA → T are distinct. . but they produce the same homomorphism after the composition with (1). This gives us a way to consider topological spaces much more general than integer aﬃne manifolds with corners. . . . x−1 ] n 1 (using addition. . . xn . x−1 . We n 1 have A′ ⊃ A and Aﬀ(A′ ) = Aﬀ(A) ⊂ A′ . x−1 ] is also tame. . . But any element of QA can be expressed in terms of the elements from T[x1 . . we have the following expression for the tropical sum of these elements “ 1T 1T 1T + x−1 + x1 1T + x−1 + 1T + x1 1 1 + ”=“ ”=“ ” = 1T . . . . sn ) ∈ Zn . Proof. . xn . . . Consider T = Hom(Aﬀ(A)/T∗ . x−1 . t ∈ R.6. All elements of Aﬀ(A′ ) are primitive aﬃne for A′ . Unfortunately. . . The function f is primitive aﬃne for A if sj ≥ 0. . 6.
. x−1 ] → QA . . However. . Aﬃne structure resulting from the semialgebraic data If A is tame then Aﬀ(A)/T∗ is a free ﬁnitely generated Abelian group. . s1 . We need to show injectivity of this map. This is an aﬃne space with the tautological integer aﬃne structure. . multiplication and division) since A is tame. . . . x−1 ] at s1 and s2 are all the same n 1 we get that the values of all the functions from QA at s1 and s2 are also the same which leads us to a contradiction.
see Figure 1. Furthermore.
. Conversely. If y ∈ K then the ho/ momorphism y : T[x] → T cannot factor through A as the value of f at y is not determined by the values at K. The next example provides a tropical algebra whose spectrum is a tropical variety (as we’ll see later). i. “x + 0”) also induces a homomorphism σ ∗ : A → T[x] that is right inverse to π ∗ . The projection (x. so in a sense we are considering the Cantor set enhanced with an integer aﬃne structure. 1] ⊂ R be the Cantor set and let A be the space of all functions A → T that can be obtained as a restriction of a tropical polynomial f ∈ T[x].40. Note that Aﬀ(A) = Aﬀ(T[x]). so the tropical algebra A is still tame. 0)] ∪ [(0. 0)] ∪ [(0. In the following chapters we introduce tropical ndimensional varieties. Then Specm (A) = K as we can evaluate any f ∈ A on any point y ∈ K.40.41. π ∗ ◦ σ ∗ = Id. A planar tropical line and its retractions. (0. f : T → T.e. +∞)]. They will never look like the Cantor set from Example 2. Let K ⊂ [0. −∞). SOME (SEMI)ALGEBRAIC NOTIONS
Example 2. (+∞. Let A be the algebra obtained by restriction of tropical polynomials in two variables to the tripod Y ⊂ T2 deﬁned by Y = [(−∞. −∞). x → (x. 0). the map σ : T → Y . 0)]
Figure 1. y) → x gives a map π : Y → T that induces a homomorphism π ∗ : T[x] → A. Locally they may look like either Tn or some more general polyhedral ndimensional complexes in TN . 0). N > n. (0. The map σ ∗ ◦ π ∗ gives a retraction of A to the subalgebra of functions constant on the ray [(0.24
2. to K. Example 2. (0. if y : A → T is a Thomomorphism then it gives a Thomomorphism T[x] → T (as the restriction to T produces a Thomomorphism T[x] → A) and thus corresponds to a point y ∈ T.
The tropical algebra O(U ) consists of functions f : U → T such that ˜ ˜ ˜ there exists an element f ∈ O(U ) ⊂ Q such that f (x) = f (x) for any x ∈ U . Suppose that x ∈ Specm (A) is regular for “ n1 ”. Note that for any f ∈ Q and x ∈ Specm (A)◦ the value f (x) exists (and not equal to −∞ ∈ T). Proposition 2. 7.
n n
Proof. Then x is also regular for “ f1j ” as it can be
fj
j=1
obtained from “
n
1 fj
” by taking a product with all fj ′ . Each element f ∈ A deﬁnes a set Vf ⊂ Specm (A) of its zeroes.
j=1
. Definition 2. This set is called a hypersurface deﬁned by f .42. A point x ∈ Specm (A) is called regular for f ∈ Q if there exists an open neighborhood U ∋ x. The union of ﬁnite number of hypersurfaces is a hypersurface. where ¯ ¯ ¯ ⊂ Q is a subalgebra such that A ⊃ A ∪ {f } and x(f ) = f (x). Thus we also have a right inverse to the projection homomorphism T[y] → A. Let U ⊂ Specm (A) be any subset. A point x ∈ Specm (A) is called a zero of f ∈ Q if f is 1 regular at x. Regular functions and tropical schemes Let f ∈ Q and x ∈ Specm (A). the space Y is called the planar tropical line. Since A ¯ A ¯ generates Q as a semiﬁeld the value f (x) ∈ T is unique (if it exists).45.44. REGULAR FUNCTIONS AND TROPICAL SCHEMES
25
Note that Y is symmetric with respect to permutation of x and y. U ⊂ Specm (A). Clearly
j=1
all points of Specm (A) are regular for any f ∈ A. We have Specm (A) = Y . We say that the value of f at x is f (x) if the epimorphism x : A → T extends to an epimorphism x : A → T. We claim that
j=1
Vfj is a hypersurface deﬁned by
fj .7. Definition 2. and an element g ∈ Conv(A) such that the values g(y) and f (y) exist and g(y) = f (y) for any y ∈ U . An element of O(U ) is called a regular function on U . j ′ = j. but “ f ” has a pole at x. ˜ Definition 2.43. The tropical algebra O(U ) associated to a subset U consists of all elements of Q that are regular at every point of U . A point x ∈ Specm (A) is called a pole for f ∈ Q if f is not regular at x.14. ˜ Note that O(U ) is a quotient of O(U ) as we have the evaluation epimorphism ˜ evU : O(U ) → O(U ). O see Deﬁnition 2.
OX U ) is isomorphic ˜ to the pair (UA . Let f ∈ OX (V ) and x ∈ V for an open V ⊂ X Choose an open neighborhood U ∋ x. OSpecm (A) UA ). The value of f (x) is xA (ρV ∩V (f )) ∈ T. The value f (x) does not depend on the choice of the aﬃne neighborhood U . Let Z ⊂ X be any subset and f : Z → T be a function. Note that clearly ˜ we always have the required restriction homomorphisms ρV : OA (V ) → U ˜A (U ) for V ⊃ U that are also always monomorphisms as we just take an O embedding of the elements of QA that are regular on V in the larger set of those elements which are regular on U . ˜ The scheme is called reduced if for any U the tropical algebra OX (U ) is ˜X (U ).49.46. 8. U ⊂ Specm (A). Regular maps Definition 2. In such case we set OX (U ) = O Since the restriction of a sheaf to an open set is a sheaf the tropical ˜ integral domain A has to be such that OA UA form a sheaf. Proof. reduced. OY ) is a pair consisting of a continuous map f :X→Y and a collection of tropical algebra homomorphisms Φ∗ : OY (U ) → OX (Φ−1 (U ))
. U Proposition 2.48. SOME (SEMI)ALGEBRAIC NOTIONS
Definition 2. From now on we restrict our attention to reduced schemes X. A tropical scheme is a pair consisting of a topological ˜ space X and a sheaf OX of tropical algebras on X such that for every point x ∈ X there is an open neighborhood U ∋ x. Since x ∈ U ∩U ′ ∩V both epimorphisms have to factor through the tropical algebra O(U ∩ U ′ ∩ V ) where both xA (ρV ∩V (f )) and xA′ (ρV ′ ∩V (f )) have a common U U lift ρV ∩U ′ ∩V (f ). OX ) → (Y. Once again. A regular map between tropical schemes Φ : (X. The sheaf OX is called the structure sheaf of X.47. a tropical integral domain A ˜ and an open set UA ⊂ Specm (A) such that the pair (U. Thus x corresponds to a tropical epimorphism xA : A → T.26
2. all regular functions on Z together with pointwise addition and multiplication form a tropical algebra which we denote Funct(Z). U Definition 2. The function f is called regular if for any x ∈ Z there exists an open neighborhood U ∋ x and g ∈ OX (U ) such that f (y) = g(y) for any y ∈ Z ∩ U. Suppose that x corresponds to a tropical epimorphism xA′ : A′ → T another aﬃne neighborhood U ′ ∋ x with U ′ = Specm (A′ ).
50. A regular map Φ : X → Y is called a scheme embedding if Φ is a settheoretical embedding and for all open U ⊂ Y the homomorphisms Φ∗ : OY (U ) → OX (Φ−1 (U )) is an epimorphism.e. i. clearly.8. we have a settheoretical pullback of the function f . namely Φ∗ (f ) : W → T. REGULAR MAPS
27
for any open set U ⊂ Y that is consistent with the restriction homomorphisms of the sheaves OX and OY . We have Φ∗ (g) ∈ OX (Φ−1 (U )) by deﬁnition of the tropical map and. Since f ∈ Funct(V ) for every x ∈ V there exists an open neighborhood U ∋ x and g ∈ OY (U ) such that g(y) = f (y) for every y ∈ U ∩ V . Let V ⊂ Y be any set and W = Φ−1 (V ). x → f (Φ(x)). Φ∗ (g)(z) = f (Φ(z)) for every z ∈ Φ−1 (U ) ∩ W . Suppose that f ∈ Funct(V ) and Φ : X → Y is a regular map. Here ρU are the corresponding restriction homomorphisms V for regular functions. As usual. i. Proof.e.51. such that for any pair of open sets V ⊂ U ⊂ Y the diagram OY (U )
Φ∗ ρU V
/ OY (V )
Φ∗
ρ
OX (Φ−1 (U ))
Φ−1 (U ) Φ−1 (V )
/ OY (Φ−1 (V ))
is commutative. once we identify X with Φ(X) ⊂ Y . For simplicity of notations we will often suppress the symbols OX and OY and write a regular map just as Φ : X → Y . Φ∗ (f ) ∈ Funct(W ). The function Φ∗ (f ) is regular in U . In this case X is called a closed subscheme of Y . Definition 2. Proposition 2.
.
.
28. xjn ”. Then the second characterization coincides with the deﬁnition of a tropical polynomial. . Proposition 3. xn ) ∈ Tn . . . . Proof. . . . 2. cf. In particular. . A subspace V ⊂ X is called a hypersurface if for any x ∈ V there exists a chart φα : Uα → Tn and a tropical polynomial fα : Tn → T. Let U ⊂ X be an open set and f : U → T be a continuous function. . Recall that a monomial is just a aﬃnelinear morphism to T. . xn ]) so that regular functions correspond to monomials. . such that V ∩ Uα = φ−1 (Vfα ).2.CHAPTER 3
Hypersurfaces and complete intersections in Tn
1. . κl }. Theorem 3. . . . aj ∈ T and κj (x) = “xj1 . Integer aﬃne manifolds as tropical schemes After a bit of algebraic formalism we return to our geometric objects: integer aﬃne manifolds. Also we may speak about tropical hypersurfaces in integer aﬃne manifolds with corners. Thus to see the structure of hypersurfaces in X it suﬃces to look carefully at the structure of hypersurfaces in Tn . Hypersurfaces in Tn Let f : Tn → T be a tropical polynomial (2) f (x) = “
j∈Zn
aj κj (x)” = max aj + κ(x). Here the some is taken over the ﬁnite number of multiindexes j. . κl : W → T such that f W = max{κ1 . Proof. Any integer aﬃne manifold X with corners can be naturally considered as a reduced tropical scheme. . A function f : U → T is regular at x ∈ U if and only if there exist an open subset W ⊂ U and a ﬁnite collection of monomials κ1 .1. . Deﬁnition 1. we may characterize the regular functions in terms of integeraﬃne structure. We may choose W so that it is contained in a single chart φα : Uα → Tn .
j
x = (x1 . so that “aj κj ” are n 1
29
. Locally X is modeled on an open set in Tn = Specm (T[x1 . where Vfα is the hypersurface α associated to fα .
If VfK = ∅ we say that K ∈ Subdivf and denote with ∆K the convex hull of K in Rn ⊃ K. Each ∆K is contained in a minimal aﬃnelinear subspace in Rn . Recall that the hypersurface Vf is the locus of all points x ∈ Tn such that “ 1T ” is not regular at x.e. K2 ∈ Subdivf and K1 ∩ K2 = ∅ then K1 ∩ K2 ∈ Subdivf and ∆K1 ∩ ∆K2 = ∆K1 ∩K2 . the interior in the corresponding aﬃnelinear space. the convex hull of J in Rn . f Proposition 3. The monomials aj κj naturally deﬁne a stratiﬁcation of Vf .5. for a subset K ⊂ J of cardinality greater than one we may deﬁne the stratum VfK ⊂ Vf by VfK = {x ∈ Tn  Kf (X) = K. Theorem 3. i. Proof. Denote with ∆f the Newton polyhedron of f .e. i.3. Namely. HYPERSURFACES AND COMPLETE INTERSECTIONS IN Tn
monomials.4. we have the following properties.30
3. Denote with ∆◦ the K relative interior of ∆K . Proof.
Each component of Vf naturally corresponds to a point j ∈ J. For each x ∈ Tn we deﬁne Zn Kf (x) = {j ∈ J  f (x) = “aj κj ”. If only one f monomial is maximal at x then f is locally linear at x and thus −f is also regular at x.3. The polyhedra ∆K form a subdivision of the polyhedron ∆f which is dual to the corresponding strata VfK . Vice versa. in other words Kf (x) is the set of the indices of the monomials where f (x) assumes its maximum. If more than one monomial assumes the maximum at x then f is strictly convex at x and thus “ 1T ” = −f cannot be convex. this proposition is the direct corollary of Proposition 3. Proposition 3. The set J is ﬁnite since f is a polynomial.
Tn
. For many subsets of J we have VfK = ∅. We have Tn =
K⊂J
VfK and
Vf =
K>1
VfK . Note that VfK is deﬁned by a system of linear inequalities in Tn ⊃ Rn and thus is a convex polyhedron (possibly unbounded) in Tn . • If K1 . The hypersurface Vf is the locus of points x ∈ Tn where the maximal value in (2) is attained by more than one monomial aj κj . Let J = {j ∈  aj = 0T } be the indices parameterizing the monomials that appear in f . such that “aj κj ” is maximal in this component. This means that it is the closure in Tn of a convex polyhedral domain in Rn .
Furthermore the aﬃnelinear subspaces in Rn generated by VfK ∩ Rn and ∆K are orthogonal. Here the integer length of an interval I ⊂ Rn with ∂I ∈ Zn is the total number of integer subintervals in it (i. To see nonexistence of the height function it suﬃces to look at the attachments of the wouldbe faces around the inner square. Remark 3. These subdivisions come as projections of the top faces of the polyhedral domain in Rn × R obtained as the convex hull of the undergraph of j → aj . The function j → aj is called the height function of the subdivision. Then. Suppose that K ∈ Subdivf and k ∈ ∆◦ . e. [68]. coherent lattice subdivisions of the polyhedron ∆f . [15]. Thus ∆◦ are disjoint and form a subdivision of ∆f .g. but we may identify them by introducing a scalar product to Rn .g. “ak κj (k)” = f (x) exactly on K VfK . The last two properties come as straightforward applications of Linear Algebra. #(I ∩ Zn ) − 1). Note that for every j ∈ J the locus “aj κj (x)” = f (x) is deﬁned with a system of linear inequalities and therefore is convex.e. cf.5 are called convex. Then a generic point x of the convex hull of VfK1 ∪ VfK2 must correspond to K1 ∪ K2 . Figure 1 depicts a classical example of a nonconvex lattice subdivision (see e. by convexity. the Newton polygon ∆f and the hypersurface Vf ∩ Rn belong to dual vector spaces Rn . Then the hypersurface V is still given by a tropical polynomial f in
.6. sometimes.5 gave a description of hypersurfaces in Tn . Remark 3.) • If ∆K1 ⊂ ∆K2 then VfK1 ⊃ VfK2 .e. Suppose that K1 ∩ K K2 = ∅. Not all subdivisions are convex. K1 = K2 we have ∆◦ 1 ∩ ∆◦ 2 = ∅.2. see [15].7. [15]). However. K2 ∈ K Subdivf . In particular. the same construction works also for hypersurfaces V in (T× )n . Subdivisions that appear in Theorem 3. TPn and other toric varieties as long as every component of V (its subset that constitute a hypersurface itself) has nonempty intersection with the torus (T× )n . (More rigorously. Theorem 3. In real algebraic geometry such subdivisions appeared after the discovery of the patchworking technique by Viro [68]. HYPERSURFACES IN Tn
31
• The (relatively) open polyhedra ∆◦ are disjoint: for any K1 . K K ∆◦ . Proof. Thus without loss of generality we may assume that K coincides with ∆K ∩ Z n . regular or. to each facet (i. an (n − 1)dimensional face of Vf ) we may associate a positive integer number equal to the integer length of the corresponding interval in Subdivf . • ∆f = K
K∈Subdivf
• For any K ∈ Subdivf we have dim VfK +dim ∆K = n.
Note that as long as a. the hypersurface. Note that Y from Example 2. n variables and can be obtained by taking the closure in the corresponding toric variety of the toric part Vf ∩ (T× )n of the aﬃne hypersurface Vf ⊂ Tn . More precisely. but greater then the third remaining monomial. b. HYPERSURFACES AND COMPLETE INTERSECTIONS IN Tn
Figure 1. Indeed. associated to “ f (x. All three monomials are equal at the origin while everywhere on the three rays two of the three monomials are equal. A general polynomial of degree 1 in two variables is of the form f (x. Lines in the plane The easiest examples to visualize are planar curves. y → y + b − c.
.y) = f (x. 0} under the translational change of coordinates x → x + a − c. it is the hypersurface of f (x. y+b−c.32
3. 3. but the diagonal ray is open. apply the translation and take the closure in T2 again.41 is an example of a tropical line in the plane. We often will use the same notation Vf for a hypersurface in other toric varieties. y) = “ax + by + c”. y) = “x + y + 1T ”.e. we have to take Y ∩ R2 . hypersurfaces in T2 . c = 0T any tropical line can be obtained from Y by a translation in R2 . ′′ Indeed. But max{(x+a−c. 0} corresponds to max{x. y. Thus a line in T2 is the hypersurface associated to this tropical polynomial. A nonconvex lattice subdivision. y+b−c. 0} coincides with Vf . i. y)−c = max{(x+ c a−c. Note that the horizontal and vertical rays of Y end with an inﬁnite point (as the axes {y = −∞} and {x = −∞} are included in T2 ).
When we consider. LINES IN THE PLANE
33
If one of the coeﬃcients of f assumes the value 0T = −∞ then the corresponding monomial is never maximal in f . a ∈ T× .3. y) = “ax” = x + a. y) = c. y) = “ax” = x + b. a family (3) ft = “ty + c”.
. then “ 1T ” = −x − a is regular as f long as x = inf ty. The projective space TP2 provides a compactiﬁcation of T2 by attaching an extra line (called the inﬁnite line). Thus the corresponding hypersurface is the closure of the straight line which maybe horizontal. Five lines in T2 . Similarly the hypersurface of f (x. but not deﬁned at the coordinate yaxis {x = −∞} of T2 . see Figure 2. vertical or diagonal. Thus in this case Vf coincides with the yaxis. If f (x. is the xaxis of T2 . depending on which monomial disappears.g. then “ 1T ” = −c is regular everywhere on T2 . c ∈ T× . t → −∞
the corresponding horizontal line moves to inﬁnity and coincides with that inﬁnite line in the limit. We may draw the corresponding deformation on the (ﬁnite) triangle. e. so f Vf = ∅. Consider now the case when two of the coeﬃcients of f assume the value −∞.
Figure 2. For that we need to reparameterize R2 to the interior of a ﬁnite triangle. If f (x. b ∈ T× .
8. each corresponding to the real points of a line in a quadrant of (R× )2 . which is a diﬀeomorphism.30 is that the image of a line in RP2 is an ellipse that is tangent to the three sides of the triangle T . Note that a generic line in R2 intersects three out of four quadrants. w) = (
dz z
∧
d¯ z z ¯
+
dw w
∧
dw ¯ w ¯
z2 w2 . HYPERSURFACES AND COMPLETE INTERSECTIONS IN Tn
One of the most natural ways (along with the map provided by Proposition 1. Log(z. x + y ≤ 1}. The points of tangency with the sides corresponds to the points of intersection with the three coordinate axes (the xaxis.e. this diﬀeomorphism extends to a diﬀeomorphism TP2 → T . The imaginary points of a line L ⊂ CP2 that is real (i.30 to do that is via the combination of the logarithmic moment map Log : (C× )2 → R2 . The image of a complex projective lines under µ ◦ Log−1 is an inscribed ellipse in T .34
3. the yaxis and the inﬁnite line). see Figure 3. When we need to speak about the inﬁnite points of varieties in TP2 it is more convenient to draw their images under this reparameterization. so we have a welldeﬁned map µ ◦ Log−1 : R2 → Int(T ). 1 + z2 + w2 1 + z2 + w2
Note that the image µ(R2 ) is the interior of the triangle T = {(x. w) = (log z. one of the advantages of the parameterization µ ◦ Log−1 with respect to the parameterization provided by Proposition 1. Note though that the image of a straight line in R2 is (in general) no longer straight in T . Remark 3. Furthermore. y) ∈  x ≥ 0. which is the moment map for the (C× )2 invariant form and the FubiniStudy moment map for CP2 µ : (C× )2 → R2 . log w). Both maps Log and µ have the same ﬁbers. µ(z. y ≥ 0. R2
Figure 3. These tangencies divide the circle into three arcs. ). invariant with respect to the complex conjugation) are mapped inside this ellipse in the
.
Furthermore. Each facet F of this skeleton will be included to the cycle with an integer (possibly negative) coeﬃcient that depends only on the local structure of A and B near E. see the ﬁrst part of Figure 4. The lines given by a monomial coincide with one of the coordinate axes and correspond to a side of the triangle.3. any pair of lines in Figure 2 has such “transverse” intersection. The lines given by a binomial equation pass through an intersection point of the coordinate axes (recall that we treat the inﬁnite line as one of the coordinate axes!) and correspond to the degeneration of ellipses to intervals passing through a vertex of the triangle. Indeed. Images of tropical lines in T and their degenerations. Thus we have a 2dimensional family of suitable ellipses and this corresponds to the dimension of the space of lines in the projective plane. Indeed. see Figure 5.g. any line in CP2 can be made real after the multiplication in (C× )2 by a suitable pair (a. The second part of this ﬁgure shows how generic lines degenerate to a binomial line. the image of any (not necessarily real) line in CP2 is the region in T2 that is encompassed by an ellipse tangent to the sides of T . a. LINES IN THE PLANE
35
21 fashion so that the the inverse image of a point inside the ellipse under Log L consists of a pair of complex conjugate points. a generic line is made of three segment.
Figure 4. Note that the family of ellipses in R2 is 5dimensional and each tangency gives a condition of codimension 1. where each segment is a subinterval of a line passing through a vertex of T . b ∈ C× . Thus we see that any line in TP2 is either an R2 translate of the tripod Y from Example 1 or a degeneration of such translates. b). The same parameterization works well for images of tropical lines. The last part of Figure 4 depicts the family (3) and its limit. Later in this book we develop the tropical intersection theory which allows to associate the cycle of the right dimension even for nontransverse cycles A.
. In the same time we may ﬁnd two lines that have a whole ray in common. B. This intersection cycle will be supported on the skeleton of the settheoretical intersection of the expected dimension. Note that two generic lines in TP2 intersect in a unique point: e.
Indeed. the line L separates TP2 into three sectors. This agrees with the notion of the stable intersection from [56] in the case when the ambient space is an (integer aﬃne) smooth variety. We can move L into this / sector so that p2 is remained on L by a translation antiparallel to the ray opposite to the sector of p2 . even though the lines from Figure 5 intersect along a ray. b. Proof. Applying a translation in R2 to the tripod Y from Example 1 we may ﬁnd a line L ∈ TP2 such that its 3valent vertices coincide with p1 . Generically. HYPERSURFACES AND COMPLETE INTERSECTIONS IN Tn
stable intersection point
Figure 5.9. c ∈ T up to the simultaneous multiplication of the coeﬃcients a. a. Lines in TP2 form themselves an integer aﬃne manifold with corners isomorphic to TP2 . Any pair of points p1 . Theorem 3. If the sedentarity of p1 is positive then we may ﬁnd a line L ∋ p1 and a family of nondegenerate lines Lt so that the the trivalent point of Lt tends to p1 . this line is unique unless this pair of line and one of the intersection points of the coordinate axes (the points of sedentarity 2) are collinear.36
3. Nontransverse intersection of two lines in T2 . Proposition 3. Furthermore. see Figure 6. Note that from the algebraic point of view the statement is trivial. Proof. This manifold is called the dual projective plane and denoted with (TP2 )∗ . If p2 ∈ L then it is inside one of these sector. In particular.10. any line is given by a polynomial “ax+by+c”. p2 ∈ TP2 can be joined with a line. c by the same
. b. their intersection cycle is the (sedentarity 0) endpoint of this ray.
10 it is convenient to depict the dual plane with the inverted triangle. Finding a line passing via two points in TP2 . We still have such a distinguished point by tracing the limit of the 3valent vertex under its approximation by nondegenerate lines. replacing of the vertex of TP2 with all lines passing through this vertex is the tropical counterpart of the blowing up of this vertex. so it does no longer determine the position of the line. Because of the inversion σ from the proof of Theorem 3. 1 1 Consider the inversion σ : R2 → R2 .3. a coordinate line L is mapped to the opposite vertex of T by the inversion while choosing a nearby point in the image completely determines the line nearby to L. A chart near a line with the 3valent vertex in R2 is given by that 3valent vertex itself. It allows one to deﬁne a new manifold X (that is the result of blowing up of TP2 in
. (x. Note that we may easily describe the same chart in coordinates. As a map (TP2 )∗ TP2 the inversion σ is only partially deﬁned. see Figure 7. scalar λ ∈ T× . However. Nevertheless. LINES IN THE PLANE
37
move into this sector
Figure 6.11. Nevertheless. These lines pass through a vertex of the triangle T and a point on its side. Consider now those degenerate lines that do not coincide with a coordinate axes (those given by a binomial). y) → (“ x ”. via this distinguished point we may identify the nondegenerate lines with the points of R2 . but this point is the vertex of T . This inversion does not extend to the vertices of TP2 . This extension gives a chart to T × T× in a neighborhood of noncoordinate lines passing via the vertex of TP2 . Finally. “ y ”) = (−x. −y). These triples of coeﬃcients up to such rescaling form TP2 by the very deﬁnition. Furthermore. it is useful to look at the space of lines from a geometric point of view. but does extend to the vertices of TP2 enhanced with lines passing through them. This gives a chart to T2 . in the complement of the three points corresponding to the coordinate lines we may identify the space of all lines in TP2 with the space of lines together with a distinguished point (a 3valent vertex in the case of nondegenerate line and a vertex of T otherwise). Remark 3.
4. Passing from the projective plane to the dual projective plane. Primitive triangles do not contain lattice point other than their vertices. each coherent subdivision of ∆f corresponds to a conic in TP2 . Furthermore.5 there is a lattice subdivision of ∆f for each conic C ⊂ TP2 and. see e. The smallest possible convex polygon with vertices in Z2 is a triangle of area 1 . Therefore. primitive triangulations (i. c. Consider a smooth conic Vf ⊂ TP2 .e. Curves dual to subdivision into primitive triangles are called smooth planar tropical curves. 0).e. i. conversely. Let us deform just one of the coeﬃcients of f . 2) or its subpolygon (in the case when some of the coeﬃcients vanish. Furthermore. Curves in the plane Let us look at the conics in TP2 .g. e. In the same time the edges of Vf corresponding to the edges of Subdivf adjacent
. 2 Definition 3. b. By Theorem 3. d. a.38
3.32. y) = “ax2 + bxy + cy 2 + dx + ey + f ”. assume the value 0T = −∞). lattice decompositions of a Newton polygon into primitive triangles) contain all lattice points of the polygon among their vertices. all the edges of Vf has weight 1. f ∈ T. It is easy to see that the resulting deformation will leave the strata of Vf disjoint from m invariant. These are the hypersurfaces given by quadratic polynomials f (x. HYPERSURFACES AND COMPLETE INTERSECTIONS IN Tn
all three vertices) and everywhere deﬁned maps X → TP2 and X → (TP2 )∗ . the one depicted on Figure 6. Note that X is an integer aﬃne manifold with corners as tropical blowups come with natural charts to T2 . (2.
blow down three sides blow down 3 other sides
Figure 7. We have six monomials and each can dominate the polynomial f in a certain region in the plane (possibly empty). The Newton polygon of f is the triangle ∆f with vertices (0. Because of the smoothness condition each monomial m ∈ ∆f ∩ Z2 corresponds to a nonempty region in TP2 . it is one of the toric varieties from Remark 1.12. Figure 8. 0) and (0. see Figure 7. Such triangles are called em the primitive triangles.
The second degeneration can be interpreted as a smooth conic that is tangent to a coordinate axis in TP2 as we shall see later. The simplest degeneration correspond to a coarser subdivision of ∆f when we take into Subdivf the union of two nearby primitive triangles instead of taking each one individually. List all combinatorial types would take too
. Deforming one coeﬃcient. We have two combinatorially diﬀerent cases: the union of two could be a parallegram or it could be a triangle of area 1. see Figure ??. CURVES IN THE PLANE
39
the resulting region
Figure 8. to m will move enlarging or diminishing the corresponding region depending on whether we increase or decrease the coeﬃcient of the monomial m.4.
Figure 9. Figure 10 shows some smooth conics together with the corresponding subdivisions. Note that the ﬁrst case corresponds to a reducible conic that decomposes to the union of two lines. The higher is the degree the more possibilities we have for the combinatorial type of the curve. It is easy to see that the ﬁgure exhaust all possible combinatorial types of smooth conics. Smooth planar conics. It is instructive to look at the possible degenerations of smooth conics.
HYPERSURFACES AND COMPLETE INTERSECTIONS IN Tn
weight 2 edge
Figure 10.13. Example 3. d) (or a subpolygon of this triangle). Consider the square lattice in Z2 . Note to specify a combinatorial type of a planar tropical curve of degree d we need to produce a lattice subdivision of the triangle ∆d ⊂ R2 with vertices (0. Planar cubics.
Figure 11. 0) and (0. 0). If we subdivide each square into two triangles by the diagonal parallel to the line x + y = 0 we get a subdivision of R2 that is compatible with ∆d for any d. The following two examples list two particularly simple combinatorial types of smooth tropical curves of arbitrary degree. Singular planar conics. (d. The resulting
. Figure 11 depicts a smooth and a singular cubic.40
3. long already for the case of planar cubic.
Let us subdivide ∆d ⊂ R2 into “ﬂoors” by the lines y = 1. . They proved to be useful for a range of problems related to the Horn problem. Note that the honeycomb triangulation of ∆d is symmetric with respect to the exchange of the x and y coordinates. . These subdivisions appeared in [23] as coherent subdivisions of higherdimensional simplices.4. Our next example is not as symmetric. . Example 3.
Figure 13.
. see [32]. Each ﬂoor is a trapezoid that can be further subdivided into the primitive triangles as shown on Figure 13. d−1. Honeycombs. The tropical curves in this combinatorial type (as well as all their degenerations) are called honeycombs. The ItenbergViro subdivision in dimension 2. . CURVES IN THE PLANE
41
subdivision and the tropical curve in the corresponding combinatorial type are pictured on Figure 12. Furthermore. it is symmetric with respect to the action of the symmetric group S3 that interchanges these two axes and the inﬁnite axis.
Figure 12.14.
Floors and stacking them on top of each other. Similarly to the case with the lines in TP2 it is easy to show that any hyperplane with a. Furthermore. d − k + 1 rays pointing down and no other vertical edges. b. we may ﬁx any positions (i. These curves are called ﬂoors. .14 are veriﬁed by existence of the corresponding tropical curves. To check the latter we may note that because the lines y = 1.13 and 3. HYPERSURFACES AND COMPLETE INTERSECTIONS IN Tn
Remark 3. 5.16.e. The coherence of the subdivisions in Examples 3.e. b.42
3. Note also that any lattice subdivision of a the Newton polygon of a ﬂoor (i.e. c. In particular we may combine the ﬂoors of diﬀerent combinatorial types. Surfaces in TP3 We start by looking at the hyperplane in TP3 . Because of that we may inductively stack a k+1th ﬂoor on top of the kth ﬂoor. Example 3. d = 0T is the result of translation of the (standard) hyperplane of “x + y + z + 1T ” by a vector in R3 . As our last example of a planar tropical curve we consider a rather involved example of a curve of degree 10. The counterexample is provided by this very curve once we equip it with the suitable real phases.
.
Figure 14. a subpolygon of the strip k − 1 ≤ y ≤ k) is coherent. i. The kth ﬂoor has d − k vertical rays pointing up. c. d assume the value 0T then we can interpret the corresponding hyperplane as the limiting set of a family of translations of V“x+y+z+1T ” in R3 . the xcoordinates) for the vertical rays pointing down and ﬁnd a smooth tropical curve in the needed combinatorial type with such rays. . It appeared in the work of Itenberg [21] disproving the Ragsdale conjecture (a conjecture on topology of plane real curves that appeared in 1905 in [55] and was ﬁnally disproved only in 1992 [21]).15. Again if some (but not all) of the coeﬃcients a. d− 1 are compatible with both types of subdivision our tropical curves are glued from the curves dual to trapezoids of height 1 as shown on Figure 14. . . the hypersurface given by the tropical polynomial “ax + by + cz + d”. see Figure 15.
Figure 16 depicts a generic hyperplane H ⊂ TP2 .
. 0. (0. (0. 1). There are 4 outgoing rays from v. all of them have a common vertex v ∈ R3 . It consists of 6 sectors. A tropical plane in the 3space. −1. 0. 1. Any pair of these rays span a sector in R3 diﬀeomorphic to the positive quadrant R2 .
vertex v
Figure 16. ≥0 To get H we take the closure in TP3 ⊃ R3 of the union of the 6 sectors. in the direction (−1. 0). SURFACES IN TP3
43
Figure 15. 0). −1) and (1. The ItenbergRagsdale curve of degree 10.5.
Proof. d = 0T . 6. More generally we have the following statement generalizing Theorem 3. Complete Intersections 7. b ∈ T. Indeed.. If the line TP1 is contained in Vf then both corresponding coeﬃcients must be equal to 0T . once we prove this we can identify the space of hyperplanes with the space of all coeﬃcients up to the simultaneous rescaling which is the tropical projective nspace by deﬁnition.. HYPERSURFACES AND COMPLETE INTERSECTIONS IN Tn
The position of the vertex v ∈ R3 completely determines a tropical hyperplane V“ax+by+cz+d” with a. . Let us ﬁrst note that the Theorem is trivial if n = 1. a. Balancing condition
. “ab” = 0T . The corresponding hypersurface always just the single point x = b − a ∈ TP1 .. (TO BE CONTINUED) . We call this space of hyperplanes the dual projective space and denote with (TPn )∗ . Recall that TPn is topologically a simplex. To prove the theorem in general it suﬃces to prove show that if f is a polynomial of degree 1 in n variables then the set Vf determines the coeﬃcients of f up to their simultaneous tropical multiplication by a nonzero constant. To understand the geometry of higherdegree surfaces in TP3 (and more generally the geometry of higherdimensional tropical varieties) it is convenient to introduce the notion of ﬂoor decomposition.10. Indeed a hyperplane in TP1 is given by a polynomial “ax + b” = max a + x. The space of all hyperplanes (hypersurfaces given by tropical polynomials of degree 1) in TPn forms an integer aﬃne manifold with corners isomorphic to TPn .. c. Theorem 3.17. Each such point is a hyperplane in TP1 and determines two coeﬃcients of f up to scaling. Thus the set of such hyperplanes coincides with the set of points in TP1 . Each edge of this simplex corresponds to a tropical line TP1 obtained as the intersection of (n − 1) coordinate planes. The hyperplane Vf cuts a point on each such TP1 unless this TP1 is contained in Vf . b. b in one variable x.44
3. Similarly to the case with lines in TP2 all hyperplanes are parameterized by the dual space (TP3 )∗ ⊃ R3 .
CHAPTER 4
Tropical varieties
45
.
.
CHAPTER 5
Tropical equivalence
47
.
.
Proc. Gross. Jensen. F. 26 (1983) 121133. q) 5branes. Trans. Math. B. Pure Math. and the arithmetic of integral quadratic forms. Amer. A. Soc. Dissertation. Bergman. resultants. Kapranov. Advances in Math. [11] M. 5 (1971) no. [17] M. [14] K. i Priloˇen. M. D. 189199. Harnack. Arnold. 45–70. 213237. S´minaire Ehresmann. Forsberg. [5] J. 205–278. Edinburgh Math. convex polyhedra and algebraic geometry. [9] F. M. A. (3) 1 (1901). Tsikh. The logarithmic limit set of an algebraic variety. May 2001. [7] D. An analogue of convexity for complements of amoebas of varieties of higher codimensions. Hanany. Brusotti. Preprint. B. 2005. ﬁve dimensional ﬁeld theories and grid diagrams. A. P. Inc. Eliashberg. Passare.AG/0507563. Aﬃne structures. 560–673.. Givental. Kapranov. ee e [6] G. 138144. 10 (1876). Introduction to symplectic ﬁeld theory. Multivalued Morse theory. Discriminants. Itenberg. Birkh¨user a Boston. M. http://arxiv. Berkeley. Mat. Hilbert. V. Siebert.. Hofer.org/abs/math. Arch. Ann. Nonarchimedean amoebas and tropical varieties. M.Bibliography
[1] O. R. Speyer. Gelfand. 3. Uber Vieltheiligkeit der ebenen algebraischen Curven. superpotentials and superconformal ﬁxed points.. D. M. M. Boston. Aharony. Math. Kol. Thomas. [3] V. 73. M. http://arxiv. Benzecri. Rend.. Soc. Graphs and patterns in mathematics and theoretical physics. no. Providence. Anal. AMS. H. Phys. Angular momentum. 151 (2000). asymptotic analysis and mirror symmetry. Fukaya. Mathematische Probleme.
49
. Funkcional. [2] O. Wilson. The situation of ovals of real plane algebraic curves. 3. 183185. M. Sympos. Part II. [10] L. A. the involutions of fourdimensional smooth manifolds. RI. Webs of (p. Bogart. 55 (2000). B. May 1959. The number of roots of a system of equations.org hepth/9710116. Mathematics: Theory & Applications. [12] Y. http://arxiv. [19] A. Atiyah. H. Hanany. Aharony. Circ. http://arxiv. 475–546. Functional Anal. Branes. Bourgeois. I. Special Volume. [21] I. Diﬀerential Geom. J. Sturmfels. Curve generatrici e curve aggregate nella costruzione di curve piane d’ordine assegnato dotate del massimo numero di circuiti.org math. [15] I. A MorseBott approach to contact homology.. z [4] M. Stanford University 2002. Zelevinsky. 9 (1975). Math. 19. Large complex structure limits of K3 surfaces. 157 (1971). ¨ [18] A.. [13] M.. and multidimensional determinants. Appl. [8] T. Henriques. Computing Tropical Varieties. Bernstein.. Palermo 42 (1917). [20] D. Lind. P.org hepth/9704170. Vari´t´s localment aﬃnes. MA. 459469. A. Gross.. [16] . Einsiedler.AG/0408311. Proc. A. Laurent determinants and arangements of hyperplane amoebas. 1994. GAFA 2000.
Manin. expos´ 24.AG/0311062. 6 (1973). Okounkov. 2000. NJ. [45] G. AMS. A. Shafarevich. book
. Planar dimers and Harnack curves. K.64. World Sci. 56 . Sur les surfaces localement aﬃnes. 18 (2005). 290 (1991). no. Math. Comm. M. Soibelman. 18 (1996). their Jacobicans and Θfunctions (in preparation). quantum cohomology and enumerative geometry. Enumerative tropical geometry in R2 . Mikhalkin. of Math. Khovanskii. [42] G. Kontsevich. [46] G. 4. A. Counting curves via lattice paths in polygons. 2001. I. Mikhalkin. Preprint.50
BIBLIOGRAPHY
[22] I. The aﬃne structures on the real twotorus I. Welschinger invariant and enumeration of real plane rational curves. Kharlamov. [44] G. [35] M. Amoebas of algebraic varieties. 629634. New superposition principle for optimization problems. Math. Rullg˚ ard. A characterization of Adiscriminantal hypersurfaces in terms of the logarithmic Gauss map. 2000). Mikhalkin. Anal. Kenyon. i Prilozhen. Bull. [24] I. in Seminaire ´ sur les Equations auc D´riv´es Partielles 1985/6. 164 (1994). On the classiﬁcation of nonsingular surfaces of degree 4 in RP3 with respect to rigid isotopies. 4956. Newton polyhedra and toric varieties. S. Patchworking real algebraic curves disproves the Ragsdale conjecture. z 11 (1977). France. Itenberg. Kapranov. M. Mikhalkin. [33] M. no. [36] N.org math. G. 309 . [37] V. Amoebas of maximal area. to appear in Topology. e [38] G. Aﬃne structures and nonarchimedean analytic spaces.AG/0108225. [39] G. 313–377. Okounkov. Soibelman. Kuiper.org/abs/math. Colloq. Ann. Symplectic geometry and mirror symmetry (Seoul. Rennes. H. Notices 9 (2001). 79. Y. Palaiseau. Real algebraic curves. http://arxiv. Kenyon. Mikhalkin. Tropical curves. [30] V. Nikulin. A.GT/0205011 Preprint 2002. Viro. Math. 8.. Ann. no.org math. http://arxiv. a report for the Real Algebraic and Analytic Geometry congress. 525562. http://arxiv. [43] G. CNRS. Tao. C. [28] R. Mikhalkin. I. A. P. Paris 336 (2003). 203–263. Centre Math´matiques de l’Ecole e e e Polytechnique. Intelligencer. Maximal real algebraic hypersurfaces (in preparation). Amer. O. Maslov. Mikhalkin. Limit shapes and the complex burgers equation. Res. [47] T. to appear... Kontsevich. Decomposition into pairsofpants for complex algebraic hypersurfaces.org math. [31] A. http://arxiv. 1928. [29] R. 1. Yu. Math. 1251–1253. [41] G. E. i Priloˇen. 277285 [26] M. 2. http://arxiv. T. Maximal real algebraic hypersurfaces of projective space. Knutson. Itenberg. Mikhalkin.org mathph/0507007. Homological mirror symmetry and torus ﬁbrations. GromovWitten classes. [34] M. Amoebas over nonArchimedean ﬁelds. Anal. Kapranov. June 2001. V. http://arxiv. R. Soc. G´om´trie Diﬀ´rentielle. O. 151 (2000). Sheﬃeld. Math.AG/0303378. Zharkov. [48] V. Kharlamov. H. no. 18 (1984). Acad. [23] I. Mikhalkin. e e e ICNRS. Itenberg. Yagi. J. 79–87. Shustin. Geometry of shock curves in toric surfaces (in preparation). Kontsevich. moment map and amoebas. Publishing. 1986. [32] . Sci. Okounkov.org math. 441451. Math.org mathph/0311005. [40] G. [25] M. Intern. [27] R. M.326. River Edge. Funkcional. Paris 1953. Okounkov. A. Phys. Dimers and amoebae. Y.AG/0406564. Viro. Kenyon. Nagano. http://arxiv. no. Funktsional.
[67] W. [63] D. Sturmfels.
. 2005. First steps in tropical geometry. Hilbert’s sixteenth problem. 3. Mikael. Idempotent mathematics and mathematical physics. [51] M. Tsikh. Ya. C. R. 1 (1990). Natl. [62] E. Math. Petrovsky. 395–427.org math. Acad. and triangulations of e the Newton polytope. On zeroes of almost periodic functions generated by holomorphic functions in a multicircular domain. CBMS Regional Conference Series in Mathematics. 1998. Ya. Moscow. Cambridge. Thurston. Solving systems of polynomial equations. Newton Inst. Patchworking. Sturmfels. Ronkin. Rullg˚ ard. 377. 50–69. Izhakian. 289–317. 39 (1938). Ya. no. 1979. Contemp.org math. B. Z. Siebert. http://arxiv. Leningrad Math. Providence. http://arxiv. Wilson. 4. RI. Sturmfels. Math. 2002. 6 (1972). Passare. Th. 389–411. Toric degenerations of toric varieties and tropical curves. Vinnikov curves. Adv. 4 (2004).. Anal. RichterGebert. i Priloˇen. 243256. Horn’s problem..org math. B.. Speyer. Speyer. to appear in ”Complex Analysis in Modern Mathematics”. Passsare. 187209. Sturmfels. Amoebas. Sturmfels. S´rie I. Shustin. Publ. 1994). Tropical semirings... Math. [59] H. Topology 17 (1978). [54] J. G. A tropical calculation of the Welschinger invariants of real toric Del Pezzo surfaces. and the hive cone. [56] J. Viro. [65] D. 46. 355358. no. Sci. book. Preprint.AC/0508413. [50] L. Acad. Duke Math. RI. Cambridge Univ. http://arxiv.. J. [64] D. 5373. Soc. 275–288. 2000.. 2000. Ann. Stockholm University. Nishinou. Paris. 10591134. [61] E. [55] Ragsdale. Pin. Rohlin. Tropical Mathematics. [57] V. Providence. Contemp. Geom. [69] O. Tsikh.org math. Press.AG/0409060. 11. Math. H. Soc. 3. Viro. 127 (2005). e [60] H. Real plane algebraic curves: constructions with controlled topology. Proc. USA 101 (2004). Rullg˚ ard. Stratiﬁcation des espaces de polynˆmes de Laurent et la structure de o leurs amibes. On the topology of real plane algebraic curves. Idempotency (Bristol. http://arxiv. B. J. 2005. A. B. [66] B. A. Amer. Speyer. Fazis. Sci.CO/0408099.E. Funkcional. 97 Providence. Amer. 5864. z [58] L. Rullg˚ ard. Passare. MongeAmp`re measures. Congruences modulo 16 in Hilbert’s sixteenth problem. [68] O. Pachter. Shustin. no. B. Proceedings of the European Congress of Mathematicians (2000). [71] G. 16132–16137. 2001. Theobald. Polynomial amoebas and convexity. RI. Stockholm University. no. Idempotent mathematics and mathematical physics. [70] O. no. 377. A tropical Nullstellensatz. Math. 5. [52] M. Preprint.BIBLIOGRAPHY
51
[49] T.AG/0406099. August Amoebas: their spines and their contours. Dequantization of Real Algebraic Geometry on a Logarithmic Paper.. The tropical Grassmanian. Tropical geometry of statistical models. 331 (2000). Viro. [53] I.