You are on page 1of 8

Chow’s Theorem

Andreu Ferré Moragues

April 12, 2016

I: Preliminaries
We will begin by presenting the basic concepts that our proof of Chow’s theorem requires,
so first we introduce some notation:

Notation: Let B ⊆ Cn be an open set and U ⊆ B be an open subset of B and


suppose f1 , . . . , fm are holomorphic functions, fj : U → C. Then, their common zero set
is denoted by
N (f1 , . . . , fm ) = {z ∈ U : f1 (z) = · · · = fm (z) = 0.}
With this notation we can now give the

Definition 1: A subset A ⊆ B is called analytic if for each z ∈ A there exists an


open neighborhood of z, say U , and holomorphic functions f1 , . . . , fm on U such that
U ∩ A = N (f1 , . . . , fm ).

We have that if z0 is a point of B \ A, then we can choose an open neighborhood U


of z0 and holomorphic functions f1 , . . . , fm on U such that

z0 ∈ U \ N (f1 , . . . , fm ) ⊆ U ⊆ B,

so since the set N (f1 , . . . , fm ) is closed in U since the functions are continuous and we
are in a Hausdorff space, it follows that B \ A is open and A is closed in B. Therefore, as
an alternative definition for analytic set we could have taken a closed subset A ⊆ B such
that for each z ∈ A, there exists a neighborhood U and functions f1 , . . . , fm holomorphic
in U such that A ∩ U = N (f1 , . . . , fm ).

It can be shown that we cannot always find globally defining equations for analytic sets,
as in the following

Example 1: Suppose G = G1 ∪ G2 , where


1 1
G1 = {(z1 , z2 ) ∈ C2 : |z1 | < and |z2 | < 1}, G2 = {(z1 , z2 ) ∈ C2 : |z1 | < 1 and < |z2 | < 1}.
2 2
This G will be our open set. Now let A = {(z1 , z2 ) ∈ G2 : z1 = z2 }. By using G1
and G2 as open cover sets for A, and given the definition of A, it follows that A is an
analytic subset of G. So now suppose f is a holomorphic function in G that vanishes on A.

Then, as we saw in Andrew’s talk, f can be continued to the unit polydisk as it is a


Euclidean Hartogs figure. Let fˆ be its continuation. Given that g(z) = fˆ(z, z) vanishes
on 21 < |z| < 1, it also vanishes on 0 ≤ |z| ≤ 21 , by the symmetry of our open set G and

1
the definition of g exploiting this fact; so f vanishes on  = {(z1 , z2 ) ∈ G : z1 = z2 }. So
in particular, any finite set of globally defined holomorphic functions will contain  as
its null set, so A cannot be described by global holomorphic functions.

In particular, this establishes a connection between analytic sets and domains of ho-
lomorphy, that have been discussed previously. We need to introduce next the notion of
dimension for an analytic set. For this we will require the concept of regular and singular
points of an analytic set and rank at a point z of a finite family of holomorphic functions
in a neighbourhood of said point z.

Definition 2: Let G ⊆ Cn be a connected open set, and z ∈ G. Given f1 , . . . , fm


holomorphic functions in a neighborhood of z. Then the rank of f1 , . . . , fm at z is given
by
rkz (f1 , . . . , fm ) = rk J(f1 ,...,fm ) (z).
Definition 3: An analytic set A ⊆ G is called regular of codimension m at z ∈ A if
there exists a neighborhood U of z contained in G and holomorphic functions f1 , . . . , fm
on U such that A ∩ U = N (f1 , . . . , fm ) and rkz (f1 , . . . , fm ) = m. The number n − m is
called the dimension of A at z. Finally, we say that A is singular at z if it is not regular
at that point. We will use the notation Reg(A) for regular points of A and Sing(A) for
the singular points of A.

To be able to apply a theorem of Remmert and Stein, we will need to also introduce
the notion of irreducible analytic set and dimension.

Definition 4: As above, let G ⊆ Cn be a connected open subset and A ⊆ G an analytic


subset. We say that A is irreducible if Reg(A) is connected.

If A is irreducible, then A has the same dimension d at all regular points. We say
that this number, say d, is the dimension of A, and we denote it by dim(A).

Now, since Chow’s theorem will require us to work on projective space, which is a com-
plex manifold, we will need to introduce the following concepts:

Definition 5: Let π : Cn+1 \ {0} → CPn be the canonical projection and let x ∈ CPn .
Then `(x) := π −1 (x) ∪ {0}, the complex line through the origin in Cn+1 .

Definition 6: We say that a set X̂ ⊆ Cn+1 is a conical set if it is the union of a


family of complex lines through the origin. In other words:

z ∈ X̂ ⇒ λz ∈ X̂, for λ ∈ C.

We have that if X is an arbitrary subset of CPn , then π −1 (X) ∪ {0} is a conical set.

Finally, let us give the definition of analytic subset of a complex manifold X of pure
dimension n:

Definition 7: A subset A ⊆ X is called analytic if for each point p ∈ A there are


a connected open neighborhood U of p, and finitely many holomorphic functions on U ,

2
say f1 , . . . , fm such that

U ∩ A = {x ∈ U : f1 (x) = · · · = fm (x) = 0}.

From the definition, similarly to what we did for analytic sets, we have that A is in fact
a closed subset of X, and as one might expect, locally an analytic set in a manifold looks
like an analytic set in complex space, so most properties can get transferred.

3
II: Chow’s Theorem
Let us introduce a few results that will be needed in the proof of Chow’s theorem. The
first two, will be technical theorems that are needed for the proof to go through, and the
proof for which we shall omit. The interested reader can find it in [3]. It bears mentioning
however, that Remmert-Stein’s theorem is interesting in its own right, but we shall not
develop that here. The third lemma is proven here and is relatively straightforward.

Theorem 1: Let G ⊆ Cn be an open connected subset, and A ⊆ G an analytic set.


Then, for any z ∈ A, there is a fixed neighborhood of this z, say U ⊆ G and finitely
many holomorphic functions f1 , . . . , fN with N (f1 , . . . , fN ) = A ∩ U such that for all d
that makes sense, at every regular point w ∈ A ∩ U of dimension n − d, the rank of their
Jacobian at w is exactly equal to d.

Proof: As mentioned above, see [3].

Note that loosely speaking, this theorem is telling us that the dimension at a point
can be realized by a finite set of holomorphic functions on U . This will come handy
in the second part of Chow’s theorem, as we shall see. Now onto the Remmert-Stein
Theorem:

Theorem 2: Remmert-Stein. Suppose that G ⊆ Cn is an open connected subset. Let


K ⊆ G be an (n−d)-dimensional analytic subset and let A be an analytic subset of G\K
with all its components having dimension strictly greater than n − d. Then the closure
cl(A) of A in G is an analytic set in G.

Proof: Again, the proof can be found in [3].

But let us pause for a moment, and observe that while this theorem might feel ’natural’,
in the sense that it is an expected result, and one may even feel tempted to remove the
hypothesis on the dimension. So let us consider the following examples to show that this
inequality in dimensions cannot be omitted, and suggest that the result might be trickier
than it could appear at first sight:

Example 2: (Accumulation points cannot occur in analytic sets) Let n = 2, G = C2


with coordinates z1 , z2 , let K = {z1 = 0}, and let
∞  
[ 1
A= ,0 .
n=1
n

Then clearly, A is an analytic set in G\K because each of the points of our set is isolated,
so any polynomial of degree 1 will suffice to show this claim. However, its closure cl(A)
in G cannot be an analytic set, because (0, 0) ∈ cl(A), and this is a limit point, meaning
that if we take a neighborhood U of this point and suppose that a holomorphic function
has the zeros of U ∩ cl(A), then, by the identity theorem it is the 0 function, since the
set of zeros has a limit point in (0, 0). Therefore, cl(A) is not an analytic set as claimed.

4
Example 3: (Exploiting an essential singularity and our previous example) Again,
let n = 2, and take G = C2 with coordinates z1 , z2 . Let K = {z1 = 0}, and let
A = {z2 = e1/z1 , z1 6= 0}. Now A is closed in G \ K, and by construction, it is locally
given by a zero set of holomorphic functions, since we are away from z1 = 0.

However, cl(A) cannot be an analytic subvariety of G because if it were, it would ha-


ve to be one dimensional. Therefore, its intersection, with {z2 = 1} would have to be
zero-dimensional (here we should say that indeed, results on dimensionality of analytic
sets verify standard properties, so the claim is justified. See [3] for more details), so by
our previous example, it would have to consist only of isolated points. However, by Big
Picard, (0, 1) is a limit point in this intersection, not isolated.

While the previous example might have felt like something fixable given its nature, this
one should convince us that indeed, the hypothesis on dimension is not removable.

Lemma. P Let X̂ ⊆ Cn+1 be a conical set, f a holomorphic function near the origin,
and f = ν pν be its expansion into homogeneous polynomials. Suppose there exists
ε > 0 such that f |Bε (0)∩X̂ = 0 identically. Then pν |X̂ = 0 for every ν.

Proof: Let z 6= 0 be a point of Bε (0) ∩ X̂. Then, since each pν is homogeneous


X
λ 7→ f (λz) = pν (z)λν ,
ν

which vanishes for |λ| < 1 by the Weierstrass M -test and our initial hypothesis. Hence,
by the identity theorem applied to λ 7→ f (λz), pν (z) vanishes identically on lines through
the origin, for every ν, so since X̂ is conical and each pν is homogeneous it follows that
pν |X̂ = 0 identically for each ν, as we wanted. 

Remark: Suppose F1 , . . . , Fk ∈ C[x0 , . . . , xn ] are homogeneous polynomials. Then,

X̂ := {z ∈ Cn+1 : F1 (z) = · · · = Fk (z) = 0}

is a cone, so if we set X̂ 0 := X \ {0}, then its image under the canonical projection to
CPn is the set

X = {[z0 , . . . , zn ] ∈ CPn : F1 (z0 , . . . , zn ) = · · · = Fk (z0 , . . . , zn ) = 0}.

(Observe that homogeneity of the polynomials is required for this to make sense in projec-
tive space). Then, in each canonical open in projective space, Ui = {[z0 , . . . , zn ] : zi 6= 0}
we can define holomorphic functions fi,ν by
 
z0 zn
fi,ν ([z0 , . . . , zn ]) := Fν ,..., .
zi zi

So that then we have X ∩ Ui = {x ∈ Ui : fi,j (x) = 0, 1 ≤ j ≤ k}, which exhibits X as an


analytic set. We are now in position to give the promised proof of Chow’s theorem:

5
Chow’s theorem. Let X be an analytic set in projective space. Then X is a pro-
jective algebraic set (recall this means that X is a zero set of finitely many homogeneous
polynomials, say F1 , . . . , Fs ). Moreover, it also holds that if x ∈ X is a regular point of
codimension d, then rkz (F1 , . . . , Fs ) = d for every z ∈ π −1 (x).

Proof: Suppose that X is a nonempty analytic set in CPn , for otherwise the result
is trivial. Then, X̂ 0 = π −1 (X) is also analytic (notice that we only need to compose with
π, which is a holomorphic map between complex manifolds to see this). We also have
that dimz (X̂ 0 ) ≥ 1 for all z ∈ Cn+1 \ {0} because the Jacobian of π always has rank at
most n. (See the definition we gave for dimension in the first part). Therefore, it follows
that dim(X̂i0 ) ≥ 1, for any irreducible component of X̂ 0 .

Hence, we can apply Remmert-Stein with G = Cn+1 , K = {0} and A = X̂ 0 because


dim(X̂i0 ) = 1 > 0 = dim({0}), for any irreducible component X̂i0 of X̂ 0 . This allows us to
deduce that its closure, which we will denote by X̂ = X̂ 0 ∪ {0} is also analytic in Cn+1 .

By theorem 1 we can find an open neighborhood U of 0 in Cn+1 and a finite set of holo-
morphic functions, f1 , . . . , fm on U so that U ∩ X̂ = N (f1 , . . . , fm ) and rkz (f1 , . . . , fm ) = d
at any regular point z of dimension n + 1 − d in U ∩ X̂. Expand the fi ’s into homogeneous
polynomials pi,ν so that we have
X
fi = pi,ν .
ν∈Nn+1

Notice that by the previous lemma, since X̂ is a conical set and each of those functions
are holomorphic near 0 and U ∩ X̂ = N (f1 , . . . , fm ), then pi,ν |X̂ = 0 identically on X̂.

Next consider the chain of ideals generated by all pi,ν such that ν ≤ k, k ∈ N and
i = 1, . . . , m. In other words, Ik = hpi,ν : 1 ≤ i ≤ m, ν ≤ ki in C[x0 , . . . , xn ]. By cons-
truction Ik ⊆ Ik+1 for all k ∈ N, so since C is a field, by Hilbert’s basis theorem, we have
that C[x0 , . . . , xn ] is Noetherian. Hence, this chain stabilizes at some stage. Therefore,
we can find, as in the statement of the theorem, homogeneous polynomials F1 , . . . , Fs
such that every pi,ν is a finite linear combination of them.

It follows that each fi is also a linear combination of those Fj ’s, so we have N (f1 , . . . , fm ) =
N (F1 , . . . , Fs ) near the origin. But using the fact that X̂ is a cone, and that the poly-
nomials are homogeneous, it will also be true that X̂ = N (F1 , . . . , Fs ) (This also gets
typically denoted by V (F1 , . . . , Fs )).

Thus, applying the content of the Remark after the lemma we get that X is a projective
algebraic set. Finally, suppose we have that for each i = 1, . . . , m
s
X
fi = ai,j Fj .
j=1

Then, if we let A(z) = (ai,j (z)), by the chain rule we will have

J(f1 ,...,fm ) (z) = A(z)J(F1 ,...,Fs ) (z), for z ∈ X̂ near 0.

6
Hence, d = rkz (f1 , . . . , fm ) ≤ rkz (F1 , . . . , Fs ). Then, since π is a submersion, if X is
regular of codimension d at x, then so is X̂ at every z ∈ π −1 (x), so that rkz (F1 , . . . , Fs )
cannot be greater than d at these points. Since the rank is constant along π −1 (x), it must
be equal to d at every z ∈ π −1 (x) which finishes the proof. 

Remark: As was commented during the lectures, Chow’s theorem fails to hold over
the real numbers. One can find an explicit counterexample in [1]. Namely, one can show
(see Theorem 1.5 in [1]) that there exists a real-analytic singular Levi-flat hypersurfa-
ce H ⊆ P2 with all leaves complex hyperplanes, such that H is not contained in any
proper real-algebraic subvariety of P2 . We note that the construction relies on using
transcendental curves, as was also discussed. For more details, cf [1].

7
III: Consequences
We will now comment some remarkable consequences to Chow’s theorem. Their proofs
shall be omitted, since they involve delving much deeper into algebraic geometry and
complex manifolds, so we cannot do it here. However, they are still worth mentioning.

(1) Applying Chow’s theorem to the graph of a holomorphic map from a compact
projective variety into an algebraic variety, it follows that any such map is an
algebraic morphism.

(2) Every meromorphic function on CPn is rational [Here, in [3], divisors get used, for
instance].

(3) Chow’s theorem can be used to show that every compact Riemann surface is an
algebraic curve, by first showing that there is an analytic embedding into CPn . This
can also be done via Riemann-Roch or Hörmander’s L2 methods.

To end, some historical remarks on the theorem. It was originally proved by Chow in
1949 with a long and complicated proof that gets sketched in [2] in a simplified way. It
was then simplified in 1953 via Remmert-Stein, which is the proof presented here and
then it became a corollary in 1956 to Serre’s GAGA paper, (cf. [4]) which was much
further-reaching than Chow’s original theorem.

References
[1] Lebl, J.: Singular Levi-Flat Hypersurfaces in Complex Projective Space Induced
by Curves in the Grassmannian arxiv:1407.5913v3, 17 Dec 2014.

[2] Nash, E.: Three Approaches to Chow’s Theorem on-line reference


http://web.stanford.edu/∼ebwarner/AlgGeomSemTalk.pdf

[3] Fritzsche, K.; Grauert, H.: From Holomorphic Functions to Complex Manifolds
GTM Springer-Verlag, 2002.

[4] Serre, J.-P.: Géométrie algébrique et géométrie analytique, Annales de l’institut


Fourier, Vol. 6, pp. 1-42.

You might also like