Algebraic geometry of the

ring of continuous functions

Nicolas Addington
October 2007

Abstract
Maximal ideals of the ring of continuous functions on a compact space
correspond to points of the space. For an algebraic geometer, this is
exciting. But if we try to go further than this, it doesn’t really lead
anywhere.

1. Let X be a compact Hausdorff space. We work throughout with the ring

C(X) of continuous, real-valued functions on X. If A ⊆ X, let

I(A) = {f ∈ C(X) : f (x) = 0 ∀x ∈ A}

be the ideal of functions that vanish on A. If a ⊆ C(X) is an ideal, let

V (a) = {x ∈ X : f (x) = 0 ∀f ∈ a}

be the common vanishing locus of the functions in a. As usual,

(1) (a) If a ⊆ b then V (b) ⊆ V (a).
(b) V (0) = X and V (1) = ∅.
P T
(c) V ( ai ) = V (ai ).
(d) V (a ∩ b) = V (a) ∪ V (b) = V (ab).
(2) (a) If A ⊆ B then I(B) ⊆ I(A).
(b) I(X) = 0 and I(∅) = (1).
S T
(c) I( Ai ) = I(Ai ).
(3) A ⊆ V (I(A)) and a ⊆ I(V (a)).
(4) V (I(V (a))) = V (a) and I(V (I(A))) = I(A).

1
2. If x ∈ X is a point then I(x) is a maximal ideal, since it is the kernel of the
surjective homomorphism C(X) → R sending f to f (x). Conversely,
Proposition. Every maximal ideal is the ideal of a point.*

Proof. Let m be maximal, and suppose that V (m) is empty. For each x ∈ X,
choose fx ∈ m such that fx (x) 6= 0 and let Ux be the set where fx does not
vanish. Since x ∈ Ux , the collection of all Ux cover X, so we can extract a finite
subcover Ux1 , . . . , Uxk . Now fx21 + · · · + fx2k never vanishes, hence is a unit, but
this is impossible since m is a proper ideal.
Thus there is a point x ∈ V (m), so m ⊆ I(V (m)) ⊆ I(x), and since m is
maximal, m = I(x).

3. Since maximal ideals of C(X) correspond to points just as in algebraic ge-

ometry, we would like prime ideals to correspond to subvarieties in some sense:
maybe subspaces, or closed subspaces, or submanifolds if X is a manifold—
really, we would like the prime ideals to tell us what we should mean by “sub-
variety.” We are sorely disappointed:
Proposition. If p is prime then V (p) consists of a single point.

Proof. Since p is proper, V (p) consists of at least one point as we saw above.
Suppose that x, y ∈ V (p) and x 6= y. Let U and V be neighborhoods of x
and y, respectively, with U ∩ V = ∅. Since X is compact and Hausdorff, it is
completely regular, so there are bump functions f and g supported in U and
V , respectively, with f (x) = g(y) = 1. Now f ∈ / p and g ∈
/ p, but f g = 0 ∈ p,
which is a contradiction.
Observe in particular that every prime ideal is contained in a unique maximal
ideal.

4. Thus disappointed, we might hope that every prime ideal is maximal, that
is, that there are no strange prime ideals properly contained in I(x). But we
are disappointed further: let X = [−1, 1] and
 
f (x)
r = f ∈ C(X) : lim k = 0 ∀k ≥ 0 .
x→0 x

This r is not prime, for if f (x) = max{x, 0} and g(x) = max{−x, 0} then f ∈
/r
and g ∈/ r, but f g = 0 ∈ r. But r is radical, for if f n ∈ r then
s
f (x) n f (x)n
lim = lim =0
x→0 xk x→0 xkn
* If X is not compact, there are more maximal ideals: the ideal a = C (X) of functions with
c
compact support is proper, hence is contained in a maximal ideal, but V (a) = ∅. In fact, the
maximal ideals of C(X) are in bijection with the points of the Stone-Čech compactification
of X.

2
so f ∈ r. Now r is a radical ideal properly contained in I(0), and since a radical
ideal is the intersection of all primes that contain it, this implies that there are
prime ideals properly contained in I(0).
We remark that while r is not prime, its contraction to C ∞ (X) ⊆ C(X) is
prime, as follows. If f ∈ r is C ∞ then by l’Hôpital’s rule,

f (x) f 0 (x)
0 = lim = lim
x→0 xk+1 x→0 (k + 1)xk

for all k ≥ 0, so f 0 ∈ r. Thus f (k) (0) = 0 for all k ≥ 0, so r ∩ C ∞ (X) is the

kernel of the map C ∞ (X) → R[[x]] sending a function to its Taylor series

f 00 (0) 2 f 000 (0) 3

f 7→ f (0) + f 0 (0)x + x + x + ···
2 3!
and R[[x]] is an integral domain.
The pathological example just described comes in an infinite family: given
any function g that vanishes only at 0, let
 
f (x)
rg = f ∈ C(X) : lim = 0 ∀k ≥ 0 .
x→0 g(x)k

2
As before, rg as radical. It is not zero since e−1/g ∈ rg . If h ∈ rg then rh is
properly contained in rg (since h ∈/ rh ). Thus there is a huge, messy lattice of
radical ideals, and hence of primes, living under I(x).

5. A discussion of the Nullstellensatz may shed light the preceding example.

Proposition.
(1) V (a) is closed.
(2) V (I(A)) = Ā.
(3) I(A) is radical.
Proof.
\
(1) Each f ∈ S is continuous and V (a) = f −1 (0).
f ∈a

(2) V (I(A)) is a closed set containing A, so Ā ⊆ V (I(A)). Since X is compact

and Hausdorff, it is completely regular, so if x ∈ / Ā there is a function f
with f (x) = 1 and f |Ā = 0. This f ∈ I(A), so x ∈ / V (I(A)).
(3) If f (x)n = 0 then f (x) = 0.
Parts 1 and 2 of the proposition say that the vanishing locus of an ideal is
closed and every closed set is the vanishing locus of some ideal. (Thus the
Zariski topology on X is the same as the original topology.) Part 3 says that

3
the ideal of a subset of X is√radical. If there were a fourth part, it would be the
Nullstellensatz: I(V (a)) = a, or equivalently, every radical ideal is the ideal of
some subset of X. But the radical ideals are nice if and only if the prime ideals
are nice.

Proposition. The following are equivalent:

(1) Every radical ideal is I(A) for some A ⊆ X.
(2) Every prime ideal is I(A) for some A ⊆ X.
Proof. Every prime ideal is radical, so (1) ⇒ (2). Every radical ideal is an
intersection of primes, so (2) ⇒ (1).
And we have seen that (2) fails, even for very nice spaces like [−1, 1].
The reader may object that we should not expect a Nullstellensatz since we
are working with real-valued functions and R is not algebraically closed. But this
is not the problem: everything we have said goes through for complex-valued
functions with only slight modifications.

6. As wild as the primes of C(X) are, at least we can study them locally. If
x ∈ X, let

J(x) = {f ∈ C(X) : f |U = 0 for some open U 3 x}

be the ideal of functions that vanish in a neighborhood of x, so C(X)/J(x) is

the ring of germs of continuous functions at x.
Proposition. The prime ideals contained in I(x) are in bijection with the prime
ideals of C(X)/J(x).
Proof. It suffices to show that J(x) ⊆ p for all primes p ⊆ I(x). If f ∈ J(x)
then f vanishes in a neighborhood U of x. Let g be a function supported in U
with g(x) = 1. Then g ∈ / I(x), so g ∈
/ p, but f g = 0 ∈ p, so f ∈ p.
We give a second proof, based on the following fact:
Proposition. The ring of germs C(X)/J(x) is isomorphic to the localization
C(X)I(x) .

Proof. First we show that any f ∈/ I(x) becomes a unit in C(X)/J(x). It suffices
to show that f 2 becomes a unit. Let y = f (x)2 and g = max{f 2 , y/2}. Then g
never vanishes, hence is a unit, and g − f 2 ∈ J(x).
Second we show that the natural map C(X) → C(X)I(x) sends any f ∈ J(x)
to 0. Since f ∈ J(x), f vanishes on a neighborhood U of x. Let g be a function
supported in U with g(x) = 1. Then g ∈ / I(x) and f g = 0, so f /1 = 0/g.

Now for any ring R and prime p, the primes of Rp are in bijection with the
primes contained in p.

4
7. We conclude with some remarks about finite generation.
Proposition. If a radical ideal r is finitely generated then V (r) is open.
2 2
pSuppose that r = (f1 , . . . , fk ). For each i, |fi | = fi ∈ r, so |fi | ∈ r. Let
Proof.
f = |f1 | + · · · + |fk |. Then f ∈ r, so there are g1 , . . . , gk ∈ C(X) such that
f = g1 f1 + · · · + gk fk , so

f = g1 f1 + · · · + gk fk
≤ |g1 ||f1 | + · · · + |gk ||fk |
≤ (|g1 | + · · · + |gk |)(|f1 | + · · · + |fk |)
= (|g1 | + · · · + |gk |)f 2

Outside of V (r) = V (f ), we can divide by f to get

1 ≤ (|g1 | + · · · + |gk |)f,

but if f (x) = 0 this inequality fails. Thus the complement of V (r) is the pullback
of the closed set [1, ∞) via the continuous function (|g1 | + · · · + |gk |)f .
Since V (r) is also closed, when X is connected the only finitely generated radical
ideals are 0 and the whole ring. In particular, if X = [−1, 1] then I(0) is not
finitely generated. We contrast this with the C ∞ situation:
Proposition. In C ∞ (Rn ) we have I(0) = (x1 , . . . , xn ).

Proof. We follow Lemma 2.1 of Milnor’s Morse Theory. Let f ∈ I(0). Then
Z 1
d
f (x1 , . . . , xn ) = f (tx1 , . . . , txn ) dt
0 dt
Z 1  
∂f ∂f
= (tx1 , . . . , txn ) x1 + · · · + (tx1 , . . . , txn ) xn dt.
0 ∂x1 ∂xn
R1 ∂f
Taking gi = 0 ∂xi
(tx1 , . . . , txn ) dt we get f = g1 x1 + . . . + gn xn .
Neither C(X), nor C ∞ (X) if X is a smooth manifold, is typically Noetherian:
to produce an ascending chain of ideals, we need only produce a descending chain
of closed sets, which is easy.

8. In conclusion, while it is well-known that algebraic geometry produces some

pretty terrible topological spaces, here we have seen that topology produces
some pretty terrible rings.
I thank Dan Turetsky, Matt Davis, and Sam Eckles for helpful discussions.