Intersection multiplicities as dimensions of local

rings

P. B. Kronheimer

February 6, 2011

Given a pair of algebraic curves, C and D in C2 and a point p in C2 , we are

going to define the intersection multiplicity, I p (C, D), and examine its proper-
ties. The definition we give here is rather different from the one in Kirwan’s
book (it requires more abstract algebra), but the result is the same.

1. The local ring at p

As usual, let C[x, y] denote the ring of polynomials with complex coefficients,
in two variables x and y. We shall write C(x, y) for the field of rational func-
tions in x and y. For us, a rational function is a formal expression

a(x , y)
f (x, y) =
b(x, y)

with a and b in C[x, y] and b non-zero (i.e. not the zero polynomial). Of course
we, understand that a/b = c/d if and only if ad = bc. Despite the name, a
“rational function” is not strictly speaking a function on C2 at all. For example,
there is no sensible way to evaluate the rational function x/ y at the point (0, 0).
More particularly, we cannot use a rational function on C2 to define a continu-
ous map from C2 to the Riemann sphere C ∪ {∞}, because discontinuity would
be unavoidable at points where both the numerator and denominator are zero.
The rational functions are a field, because if f = a/b is non-zero, then it has a
multiplicative inverse g = b/ a.
Given now a point p ∈ C2 , let us define the local ring at p to be the subset

Lp ⊂ C(x , y)

consisting of rational functions which can be represented as f = a/b with b(p)

non-zero. For such rational functions, the value of of f at p makes sense, and
is finite, for it is a(p)/b(p). We have therefore an evaluation map

ev : Lp → C
f 7→ f (p).
2 2. Intersection multiplicity

If ev( f ) is non-zero, then f is a unit in the ring Lp , because if f = a/b with a(p)
non-zero then b/ a is also an element of Lp .
A rational function can be differentiated with respect to x and y by follow-
ing the formal rules of calculus. If f belongs to Lp then so do its derivatives.
So, as well as evaluating f itself at p, we can evaluate also its derivatives,

∂ i+ j f
(p).
∂ xi ∂ y j

We can regard a polynomial P ∈ C[x, y] as defining a rational function

(which we should perhaps pedantically write as P/1). Viewed as a rational
function in this way, P belongs to the local ring Lp at every point p ∈ C2 .

2. Intersection multiplicity
Rather than define I p (C, D) as such, it is more convenient to work with the
defining polynomials P and Q for the curves C and D. We will define I p (P, Q)
for arbitrary polynomials P and Q in C[x, y] (including the case that P and Q
may have repeated factors). We do not even need to assume that P and Q are
non-zero at this point (though it will turn out that the intersection multiplicity
is infinite if one or both is zero).
We write
hP, Q i ⊂ C[x, y]
for the ideal generated by P and Q. That is,

hP, Q i = { RP + SQ | R, S ∈ C[x, y] }.

When a point p is given, we can regard P and Q as defining elements of the

local ring Lp , and we can then consider the ideal in Lp that they generate. We
write this as
hP, Q ip ⊂ Lp .
Concretely, this is
hP, Q ip = { f P + gQ | f, g ∈ Lp }.
Because f and g can always be put over a common denominator, an rational
function in hP, Q ip can always be expressed as (a1 P + a2 Q)/b, for polynomials
a1 , a2 and b, with b(p) non-zero. These leads to an alternative description of
this ideal as
hP, Q ip = { f = a/b | a ∈ hP, Q i, b(p) 6= 0 }. (1)
Consider now the quotient

Lp
Ap (P, Q) = .
hP, Q ip
 3

Here we have taken the quotient of a ring by an ideal; so Ap (P, Q) is itself

a ring. However, we are most interested in thinking of Ap (P, Q) simply as
a vector space over C. As such, it has a dimension. This dimension is our
definition of the intersection multiplicity:

Definition. We define I p (P, Q) to the dimension of the quotient Ap (P, Q), as a

complex vector space. So I p (P, Q) is either a non-negative integer or ∞.

3. The simplest cases

The simplest non-trivial case to look at is the case P = x and Q = y in C[x, y],
with p the point (0, 0). In the ring of polynomials, the ideal generated by x and
y consists of all polynomials which vanish at p. So hx, yip consists of rational
functions a/b with a(p) = 0 and b(p) 6= 0. Alternatively,

hP, Q ip = { f ∈ Lp | f (p) = 0 }.

Thus the ideal hP, Q ip is exactly the kernel of the evaluation map ev. By the
first isomorphism theorem, the quotient ring Ap is isomorphic to the image of
the evaluation map, which is C. Since this has dimension 1, our definition tells
us that
I p (x, y) = 1
in this case.
The next interesting case is when P and Q are powers of x and y:

P = xk
Q = yl .

In the polynomial ring C[x , y], these two generate the ideal consisting of all
polynomials S involving no terms containing x or y to powers lower than k
and l respectively. We can characterize these as

hx k , yl i = { S ∈ C[x, y] | (∂ i + j S/∂ xi ∂ y j )(p) = 0, for 0 ≤ i < k, 0 ≤ j < l }.

Using again the characterization (1) and the rule for differentiating a product,
we see that, at p = (0, 0),

hx k , yl ip = { f ∈ Lp | (∂ i + j f /∂ xi ∂ y j )(p) = 0, for 0 ≤ i < k, 0 ≤ j < l }.

This expresses the ideal hx k , yl ip ⊂ Lp as the kernel of a linear map

φ : Lp → Ckl

given by evaluating the derivatives of a function f ∈ Lp . The map φ is sur-

jective, because we can easily write down a rational function (or polynomial)
4 4. A case of infinite intersection multiplicity

whose various derivatives at (0, 0) take on any prescribed values. So the kernel
of φ (which is our ideal) has codimension kl in Lp . Thus the quotient Ap has
dimension kl, and we have shown

I p (x k , yl ) = kl .

4. A case of infinite intersection multiplicity

Let P be a non-zero polynomial with P(p) = 0. Let us show that

I p (P, P) = ∞

in this case. That is to say, let us show that the quotient space

Ap (P, P) = Lp /hPip

is infinite-dimensional. The proof will be by contradiction. It will not hurt to

assume that p = (0, 0).
So assume that Ap (P, P) is finite-dimensional. If S is a polynomial, let us
denote by S̄ the coset
S̄ = S + hPip ∈ Ap (P, P).
Because they belong to a finite-dimensional vector space, the elements

1̄, x̄, x¯2 , . . .

must be linearly dependent. This means that there is some non-trivial relation

g0 1̄ + · · · + gn x¯n = 0;

or more simply it means that there is a non-zero polynomial G(x) such that
G(x) + hPip = 0. This is the same as saying that G(x) belongs to hPip , or

a(x, y)
G(x) = P(x, y)
b(x, y)

for some b with b(0, 0) 6= 0. Because P(0, 0) is supposed to be zero, we have

G(0) = 0, and we can therefore write G(x) as xr G̃(x), for some r > 0 and some
polynomial G̃ with G̃(0) 6= 0. We can apply the same argument with y in place
of x, and we have
xr G̃(x) = (a/b)P
(2)
y s H̃(y) = (c/d)P

with s also positive and H̃(0) 6= 0. From these we obtain

xr c(x, y)b(x, y)G̃(x) = y s a(x, y)d(x, y) H̃(y),

 5

where everything in sight is a polynomial. Using the fact that C[x, y] has
unique factorization, we see that xr must divide a(x, y), for x certainly does
not divide y s and x cannot divide d(x, y) either because d(0, 0) is non-zero.
Thus
a(x, y) = xr ã(x, y)
for some polynomial ã(x, y). Now we can cancel xr in the first line of (2) to get

G̃(x) = ( ã/b)P.

This is a contradiction, because G̃(0) is non-zero, while P(0, 0) = 0. This verifies

the assertion that I p (P, P) = ∞ whenever P(p) = 0.

5. General properties of the intersection multiplicity

Having examined a few situations in the two sections above, we now turn to
verifying each property on the check-list (i)–(vi), essentially as they appear in
Kirwan’s book.
Property 1. For any P, Q and p, we have I p (P, Q) = I p (Q , P).
Proof. This is quite obvious from the definition.

Property 2. If P and Q have a common factor S with S(p) = 0, then I p (P, Q) = ∞,

and conversely.
Proof. If P and Q have a common factor S with S, then we both P and Q belong
to the ideal generated by S, so

hP, Q ip ⊂ hSip .

Therefore,
Lp Lp
dim ≥ dim .
hP, Q ip hSip
This means that
I p (P, Q) ≥ I p (S, S).
In the previous section, we saw that I p (S, S) is infinite if S(p) = 0. It follows
that I p (P, Q) is also infinite.
Now we show the converse: we suppose that P and Q have no common fac-
tor S with S(p) = 0, and we will show that I p (P, Q) is finite. In fact, we are free
to suppose that P and Q have no common factor at all in C[x, y], irrespective
of the condition that the factor vanish at p. This is because, if S is a common
factor not vanishing at p, then S is a unit in Lp , and the ideal in Lp generated by
P and Q is the same as the ideal in Lp generated by the polynomials P/S and
Q /S: we are thus free to throw out such common factors without changing the
question. We shall also suppose that p = (0, 0).
6 5. General properties of the intersection multiplicity

We will make use of two properties of resultants, from class. We write

RP,Q (y) for the resultant obtained from P and Q by eliminating the x variable.
The first property of resultants we need is that RP,Q (y) is not the zero polyno-
mial: this is true because P and Q have no common factor. We can therefore
write
RP,Q (y) = yl R̃(y)
for some l ≥ 0 and some R̃ with R̃(0) 6= 0. The second fact we need is that
RP,Q (y) belongs to the ideal in C[x, y] generated by P and Q. (Here we regard
RP,Q (y) as a polynomial in x and y which happens to have no dependence on
x.) This means that we can write

RP,Q (y) = a1 P + a2 Q (3)

for some a1 and a2 in C[x, y]. Because R̃ is non-zero at p = (0, 0), we can
consider the rational functions
ai
fi = ∈ Lp i = 1, 2
R̃

in the local ring Lp . Dividing (3) by R̃ we get

yl = f1 P + f2 Q .

In other words, yl belongs to the ideal hP, Q ip ⊂ Lp . We can apply the same
reasoning with x in place of y, to show that x k belongs to the ideal, for some
k ≥ 0. Putting these two together, we see

hP, Q ip ⊃ hx k , yl ip .

This gives us an inequality between the dimensions of the quotient spaces,

Lp Lp
dim ≤ dim ,
hP, Q ip hx k , yl ip

or in other words
I p (P, Q) ≤ I p (x k , yl ).
We saw earlier that the right-hand side is finite: indeed, it is exactly kl. It
follows that the left-hand side is also finite, which is what we wanted to show.

Property 3. The intersection multiplicity I p (P, Q) is zero if and only if at least one of
P and Q is non-zero at p.

Proof. If both P and Q are zero at p, then P and Q belong to the kernel of the
evaluation map, ev : Lp → C. The ideal which P and Q generate in Lp is also
contained in the kernel of the evaluation map, so the ideal hP, Q ip is a proper
 7

ideal, not the whole of Lp . The quotient Lp /hP, Q ip is therefore a non-zero

vector space, and its dimension is either a strictly positive integer or infinity.
This shows the “only if” direction.
If either P or Q is non-zero at p, then it is a unit in the local ring Lp . The
ideal generated by a unit is always the whole ring, so hP, Q ip = Lp . The quo-
tient Lp /hP, Q ip is the zero vector space, and it follows from the definition of
intersection multiplicity that I p (P, Q) = 0.

Property 4. If P and Q both have degree 1 and if the zero-sets C P and CQ are distinct
lines meeting in a single point p, then I p (P, Q) = 1.
Proof. By an affine change of coordinates, we come down to the case that P = x,
Q = y and p = (0, 0). This case was treated above, where we showed that the
intersection multiplicity was indeed 1.

Property 5. If Q = Q 1Q 2 , then I p (P, Q) = I p (P, Q 1 ) + I p (P, Q 2 ).

Proof. If P and Q have a common factor vanishing at p, then the same applies
to at least one of the pairs (P, Q 1 ) and (P, Q 2 ), because of unique factorization
in C[x, y]. It follows that one side of the equality to be proved is infinite only
if the other is. We may therefore assume that all three of the intersection multi-
plicities are finite.
Let L̄ denote the quotient ring

L̄ = Lp /hPip .

Let Q̄ denote the coset Q + hPip in L̄. The intersection multiplicity I p (P, Q) is
the dimension of the quotient
Lp
,
hP, Q ip
but we can also describe this as
L̄
hQ̄ i
by the third isomorphism theorem. (Here hQ̄ i is the ideal in L̄ generated by Q̄.)
The property to be proved is therefore

L̄ L̄ L̄
dim = dim + dim . (4)
hQ 1Q 2 i hQ̄ 1 i hQ̄ 2 i
Observe at this point that we have nested inclusions

hQ 1Q 2 i ⊂ hQ 2 i ⊂ L̄
from which we can already see that

L̄ hQ̄ 2 i L̄
dim = dim + dim .
hQ 1Q 2 i hQ 1Q 2 i hQ̄ 2 i
8 5. General properties of the intersection multiplicity

Comparing this with what we want to establish at (4), we see that we need only
show
hQ̄ 2 i L̄
dim = dim . (5)
hQ 1Q 2 i hQ̄ 1 i
At this point, let us recall that hQ̄ 2 i is simply the set of all multiples of Q̄ 2 in
L̄, so there is a surjective linear map
µ : L̄ → hQ̄ 2 i
r 7→ rQ̄ 2 .

Let us examine the inverse image of the subspace hQ 1 Q 2 i ⊂ hQ̄ 2 i under µ.

That is, let us examine those r for which µ(r) is a multiple of Q 1Q 2 . If we write
r as f + hPip for some f in Lp , then the corresponding condition on f is that
f Q 2 = gQ 1Q 2 + hP
for some g and h in Lp . Writing f = a/b and clearing all the denominators in
this equality, the corresponding condition on a is that
aFQ 2 = GQ 1 Q 2 + HP
for some polynomials F, G and H in C[x, y] with F(p) non-zero. Because P and
Q 2 have no common factor, Q 2 must divide H. We can therefore divide by Q 2 ,
which will give us
aF = GQ 1 + H̃P.
We can now divide by F and write this as

a = g̃Q 1 + h̃P

for some g̃ and h̃ in Lp . Reinterpreting this in terms of the original element r in

L̄, we conclude that
r ∈ ḡQ¯1
for some ḡ in L̄. That is, r belongs to the ideal hQ̄ 1 i. We have identified the
inverse image of hQ 1Q 2 i under µ as
µ−1 hQ 1Q 2 i = hQ̄ 1 i.
From the third isomorphism theorem, we now have
hQ̄ 2 i ∼ L̄
= .
hQ 1Q 2 i hQ̄ 1 i
The desired inequality of dimensions (5) follows.
Property 6. If P̃ = P + RQ for some R in C[x, y], then I p (P̃, Q) = I p (P, Q).
Proof. This is straightforward, because the pairs (P, Q) and (P̃, Q) generate the
same ideal, both in C[x, y] and the local ring Lp .
 9

6. Uniqueness
The intersection multiplicity I p (P, Q) for polynomials P, Q in C[x, y] is uniquely
characterized by the Properties 1 through 6 described above: that is,

Theorem 6.1. If I˜p (P, Q) is an “intersection multiplicity” which is defined for all p,
P and Q, and if it shares with the standard intersection multiplicity I p (P, Q) all 6 of
the properties described in the previous section, then in fact I˜p (P, Q) = I p (P, Q), for
all p, P and Q.

Proof. By Proposition 2, we know that if I p (P, Q) is infinite, then so is I˜p (P, Q).
So we need only consider the case that I p (P, Q) is finite. We prove by induction
on k that if I p (P, Q) = k then I˜p (P, Q) = k also. The base case is k = 0, which
is covered by Property 3.
For the induction step, we may as well take p = (0, 0). We write our poly-
nomials as P1 , P2 and we assume

I p (P1 , Q 2 ) = k > 0

and our induction hypothesis is that I˜p (P, Q) = I p (P, Q) whenever the latter is
k − 1 or less. Write
P1 = t1 (x) + ys1 (x, y)
P2 = t2 (x) + ys2 (x, y)
for some t1 , t2 in C[x]. Note that at least one of t1 , t2 must be non-zero, for oth-
erwise P1 and P2 have y as a common factor, and the intersection multiplicity
would be infinite, contrary to hypothesis. So let us assume t2 is non-zero. Both
t1 (0) and t2 (0) must be zero, because otherwise k = 0. There are now two cases,
according to whether or not t1 is non-zero.

Case 1: t1 = 0. In this case, let us write t2 (x) = x d2 r2 (x), where d2 > 0 and
r2 (0) 6= 0. We can use Property 5 and Property 6 etc. to obtain

I p (P1 , P2 ) = I p (y , P2 ) + I p (s1 , P2 )
= I p (y, t2 ) + I p (s1 , P2 )
= d2 + I p (s1 , P2 ).

The same applies to I. ˜ Since I p (s1 , P2 ) < k, we conclude that I p (P1 , P2 ) =

˜I p (P1 , P2 ) from our induction hypothesis.

Case 2: t1 6= 0. In this case, let us write t1 (x) = x d1 r1 (x) and t2 (x) = x d2 r2 (x),
where d1 , d2 > 0 and r1 (0) =6= 0 and r2 (0) 6= 0. We may suppose that d1 ≥ d2 .
Set
P3 = P1 − λx d1 −d2 P2 ,
10 7. Closing remarks

with λ chosen so that the coefficient of x d1 in P3 is zero. We have I p (P1 , P2 ) =

I p (P2 , P3 ), and ditto for I˜p . So let us replace (P1 , P2 ) with (P2 , P3 ) and start over.
After finitely many steps through this case, we will arrive eventually at Case 1.
So we are done.

7. Closing remarks
The construction of the field of rational functions C(x, y) from the ring C[x, y] is
an example of the general construction of a “field of fractions” from an integral
domain. The definition of Lp is an example of “localization”. The notation Lp
that we have used is not standard.