POLONICI MATHEMATICI LXXI.2 (1999)
Topology of families of affine plane curves
by H` a Huy Vui (Hanoi) and Pham Tien Son (Dalat)
Abstract. We determine bifurcation sets of families of affine curves and study the topology of such families.
1. Introduction. Let f
α(x, y) be a family of polynomials of two complex variables (x, y) ∈ C
2whose coefficients are polynomial functions of α ∈ C
n. We consider the family of affine curves {(x, y) ∈ C
2| f
α(x, y) = 0}.
In this paper, we first determine the bifurcation set B
f, i.e., the smallest set of parameters α such that the family is equisingular outside this set.
Then applying this result, we introduce the notions of semi-cycles vanishing at infinity and study the topology of the family. Finally, we show that our results imply some well-known facts on the topology of polynomial functions ([2], [3], [6], [7]).
2. Bifurcation set of families of affine plane curves. Let f
α(x, y) :=
P (x, y, α), α ∈ C
n, be a family of polynomials of two variables whose coef- ficients are polynomials of α.
2.1. Definition. The family of affine curves {(x, y) ∈ C
2| f
α(x, y) = 0}, α ∈ C
n, is said to be equisingular outside a set B ⊂ C
nif for all α
06∈ B there exist a neighborhood U
α0of α
0and a diffeomorphism h such that the diagram
{(x, y) | f
α0(x, y) = 0} × U
α0{(x, y, α) | f
α(x, y) = 0} ∩ π
−1(U
α0)
U
α0U
α0h
//
π
πid//
is commutative, where π is the second projection.
1991 Mathematics Subject Classification: 32S55, 57M25, 57Q45.
Key words and phrases: affine curves, singularity, equisingularity, Milnor number.
Supported in part by the National Basic Research Program in Natural Sciences, Viet- nam. The second author would like to thank the AUBONNE for financial support.
[129]
Let B
fbe the smallest set of parameters α such that the family {f
α(x, y) = 0} is equisingular ouside B
f. We call B
fthe bifurcation set of the family.
2.2. The following assumptions will be needed throughout the paper:
• deg(f
α) = deg
y(f
α) = d = const;
• the curves {f
α(x, y) = 0} are all reduced.
Let l be the linear function defined by l(x, y) = x. By assumption, the restriction map
l
α: f
α−1(0) → C, (x, y) 7→ x, is proper for each α ∈ C
n.
Let δ(x, α) := disc
y(f
α(x, y)) be the discriminant of f
αwith respect to y.
Then we may write
δ(x, α) = q
k(α)x
k+ q
k−1(α)x
k−1+ . . . where q
i(α), i = 0, . . . , k, are polynomials of α. Put
B
∞:= {α | q
k(α) = 0}.
Denote by C
f(f
α) the set of critical values of f
α.
2.3. Theorem. Assume that 0 6∈ C
f(f
α) for generic α. Then the bi- furcation set of the family of affine curves {f
α(x, y) = 0} is precisely the set
B = {α | 0 ∈ C
f(f
α)} ∪ B
∞.
P r o o f. We first prove that the family of affine curves {f
α(x, y) = 0} is equisingular outside B. For each polynomial f
α, define
grad f
α:= (∂f
α/∂x, ∂f
α/∂y).
Assume that α
06∈ B. Let U
α0:= {α ∈ C
n| kα − α
0k < δ} so that C
n\ B contains the closure of U
α0.
By the definition of B, q
k(α) 6= 0 for all α ∈ U
α0; hence there exists c
0> 0 such that if α ∈ U
α0, then δ(x, α) 6= 0 on the set {x ∈ C | |x| ≥ c
0}.
By the properties of resultants, the system of equations
f
α(x, y) = 0,
∂f
α/∂y = 0,
has no solution on the set {(x, y) ∈ C
2| |x| ≥ c
0} for any fixed α ∈ U
α0. Thus
(1) ∂f
α/∂y 6= 0 for all (x, y, α) ∈ V
α0∪ Ω,
where
V
α0:= {(x, y, α) ∈ C
2× C
n| f
α(x, y) = 0, α ∈ U
α0, |x| ≥ c
0} and Ω is an open neighborhood of the set
{(x, y, α) ∈ C
2× C
n| f
α(x, y) = 0, α ∈ U
α0, |x| = c
0}.
On the other hand, by the definition of U
α0, one has
(2) grad f
α6= 0 for all (x, y, α) ∈ C
2× U
α0with f
α(x, y) = 0, |x| ≤ c
0. From (1) and (2) we conclude that there exist smooth vector fields
ξ
j(x, y, α) = (ξ
1j(x, y, α), ξ
j2(x, y, α)),
η
j(x, y, α) = (η
j1(x, y, α), η
j2(x, y, α)), j = 1, . . . , n, such that
(3)
hξ
j(x, y, α), grad f
α(x, y)i + ∂f
α∂α
j(x, y) = 0, hη
j(x, y, α), grad f
α(x, y)i + √
−1 ∂f
α∂α
j(x, y) = 0, on the set X := {(x, y, α) ∈ C
2× C
n| f
α(x, y) = 0, α ∈ U
α0}, and (4)
( ξ
1j(x, y, α) = 0, η
1j(x, y, α) = 0,
for all (x, y, α) ∈ V
α0∪ Ω. (We can construct such vector fields locally and then extend them over X by a smooth partition of unity.)
To shorten notation, we write e
j= (0, . . . , 0, 1, 0, . . . , 0) with 1 in the jth place.
Let ϕ
j(x
j, y
j, α
j, τ ) := (ϕ
j1(x
j, y
j, α
j, τ ), ϕ
j2(x
j, y
j, α
j, τ ), α(x
j, y
j, α
j, τ )), j = 1, . . . , n, be solutions of the system
(5)
dx(τ )
dτ = ξ
1j(x(τ ), y(τ ), α(τ )), dy(τ )
dτ = ξ
2j(x(τ ), y(τ ), α(τ )), dα(τ )
dτ = e
j,
x(0) = x
j, y(0) = y
j, α(0) = α
j,
and ψ
j(u
j, v
j, β
j, τ ) := (ψ
j1(u
j, v
j, β
j, τ ), ψ
2j(u
j, v
j, β
j, τ ), β(u
j, v
j, β
j, τ )),
j = 1, . . . , n, be solutions of the system
(6)
du(τ )
dτ = η
j1(u(τ ), v(τ ), β(τ )), dv(τ )
dτ = η
j2(u(τ ), v(τ ), β(τ )), dβ(τ )
dτ = √
−1e
j,
u(0) = u
j, v(0) = v
j, β(0) = β
j,
where (x
j, y
j, α
j), (u
j, v
j, β
j) ∈ X. We conclude from (3) and (5) that d
dτ P (ϕ
j(x
j, y
j, α
j, τ )) = ∂P
∂x
∂ϕ
j1∂τ + ∂P
∂y
∂ϕ
j2∂τ +
n
X
k=1
∂P
∂α
k∂α
k∂τ
= ∂P
∂x
∂ϕ
j1∂τ + ∂P
∂y
∂ϕ
j2∂τ + ∂P
∂α
j∂α
j∂τ = 0, hence that
P (ϕ
j1(x
j, y
j, α
j, τ ), ϕ
j2(x
j, y
j, α
j, τ ), e
jτ + α
j) = 0.
Further, it follows from (4) and (5) that d
dτ |ϕ
j1(x
j, y
j, α
j, τ )|
2= d
dτ hϕ
j1(x
j, y
j, α
j, τ ), ϕ
j1(x
j, y
j, α
j, τ )i
= dϕ
j1(x
j, y
j, α
j, τ )
dτ , ϕ
j1(x
j, y
j, α
j, τ )
+
ϕ
j1(x
j, y
j, α
j, τ ), dϕ
j1(x
j, y
j, α
j, τ ) dτ
= 0, hence
(7) |ϕ
j1(x
j, y
j, α
j, τ )| = const for each (x
j, y
j, α
j) ∈ V
α0. Analogously,
P (ψ
1j(u
j, v
j, β
j, τ ), ψ
j2(u
j, v
j, β
j, τ ), √
−1 e
jτ + β
j) = 0;
moreover,
(8) |ψ
j1(u
j, v
j, β
j, τ )| = const for each (u
j, v
j, β
j) ∈ V
α0. Since deg(f
α) = deg
y(f
α) = d = const, the restriction map
l
Uα0: {(x, y) | f
α(x, y) = 0, α ∈ U
α0} → C, (x, y) 7→ x,
is proper. In addition, it is clear from (7) and (8) that the solutions
ϕ
j(x
j, y
j, α
j, τ ) and ψ
j(u
j, v
j, β
j, τ ), j = 1, . . . , n, can be extended over
their maximal intervals.
Now define
h : {(x, y) | f
α0(x, y) = 0} × U
α0→ {(x, y, α) | f
α(x, y) = 0} ∩ π
−1(U
α0) by
h(x, y, α) = ψ
n(ϕ
n(. . . (ψ
1(ϕ
1(x, y, α
0, Re α
1− Re α
01), Im α
1− Im α
01), . . . , Re α
n− Re α
0n), Im α
n− Im α
0n) for (x, y) ∈ f
α−10(0) and α ∈ U
α0. We can easily check that the map h is a diffeomorphism, and that π ◦ h(x, y, α) = π(x, y, α) = α. This gives a trivialization of the fibration over the set U
α0.
We next prove that the set B is smallest. By contradiction, assume that for some α
0∈ B there exist a neighborhood U
α0of α
0and a diffeomorphism h : {(x, y) | f
α0(x, y) = 0} × U
α0→ {(x, y, α) | f
α(x, y) = 0} ∩ π
−1(U
α0).
Consider the following two cases:
Case 1: 0 ∈ C
f(f
α0). Since h is a diffeomorphism of {(x, y) | f
α0(x, y)=0}
× U
α0onto {(x, y, α) | f
α(x, y) = 0} ∩ π
−1(U
α0), it follows that 0 ∈ C
f(f
α) for all α ∈ U
α0, a contradiction.
Case 2: α
0∈ B
∞. In this case, the next lemma is needed.
2.4. Lemma. Let F be a polynomial of two complex variables such that the restriction map l|
V, V := F
−1(0), is proper where
l : C
2→ C, (x, y) 7→ x.
Suppose that the curve V is reduced. Then
χ(F
−1(0)) = d − deg disc
yF (x, y).
P r o o f. Let x
β, β = 1, . . . , p, be the critical values of l|
Vand (x
β, y
jβ), j = 1, . . . , i
β, be the corresponding critical points of l|
Vwith multiplicity l
jβ. We may write
disc
yF (x, y) = a(x − x
1)
γ1· · · (x − x
p)
γp, where a 6= 0 and γ
β= P
iβj=1
l
jβ, β = 1, . . . , p.
Take x
∗∈ C\{x
1, . . . , x
p}. Then l|
−1V(x
∗) consists of d := deg(F ) distinct points. In the x-plane, we consider a system of paths T
1, . . . , T
pconnecting x
1, . . . , x
pto x
∗such that
(i) no path T
jhas self-intersection points;
(ii) T
i∩ T
j= {x
∗} (i 6= j).
Put
S(V ) := l| b
−1V[
pi=1
T
i.
Then b S(V ) is a union of 1-dimensional curves. Let ˇ S(V ) be the set of all curves in b S(V ) \ l|
−1V(x
∗) which contain a point of Σ := {(x
β, y
βj) | j = 1, . . . , i
β, β = 1, . . . , p}.
Since S
pi=1
T
iis a deformation retract of C and the restriction map l|
V: F
−1(0) \ Σ → C \ {x
1, . . . , x
p}, (x, y) 7→ x,
is a locally trivial fibration, it follows that b S(V ) is a deformation retract of V .
It is not hard to see that
S(V ) := ˇ S(V ) ∪ l|
−1V(x
∗)
is a deformation retract of b S(V ) and also of V = F
−1(0). (The set S(V ) is called the skeleton of the curve V ([2]).) Hence,
χ(F
−1(0)) = χ(S(V )).
The set S(V ) can be identified with a 1-dimensional graph of d + P
p β=1i
βvertices and P
p β=1P
iβj=1
(l
jβ+ 1) edges. Thus, χ(S(V )) =
d +
p
X
β=1
i
β−
p
X
β=1 iβ
X
j=1
(l
jβ+ 1)
= d −
p
X
β=1 iβ
X
j=1
l
βj= d −
p
X
β=1
γ
β= d − deg disc
yF (x, y).
We now return to Case 2.
For each α ∈ C
n, let k(α) := max{j ∈ {0, . . . , k} | q
j(α) 6= 0}. By Lemma 2.4, we conclude that χ(f
α−1(0)) = d − k(α). Since α
0∈ B
∞= q
k−1(0) and α 6∈ B
∞for generic α, one gets
χ(f
α−1(0)) = d − k < χ(f
α−10(0)), a contradiction, which ends the proof of Theorem 2.3.
2.5. Remark. From the construction of B
∞, it is reasonable to call each α
0∈ B
∞a bifurcation value corresponding to the singularity at infinity of the family {f
α(x, y) = 0}.
3. Topology of families of affine plane curves. From now on, we assume that the curves {f
α= 0} are smooth for generic α.
Consider the family of affine curves {f
α= 0}. By Lemma 2.4, χ(f
α−1(0))
= d − k(α). On the other hand, by the definition of the set B
∞, we see that
α
0∈ B
∞iff k(α
0) < k = k(α) for generic α. Hence χ(f
α−1(0)) < χ(f
α−10(0))
for generic α.
Thus we can reformulate Theorem 2.3 as follows.
3.1. Theorem. α
0∈ C
nis a bifurcation value of the family of affine curves {f
α(x, y) = 0} if and only if either
(i) the curve {(x, y) | f
α0(x, y) = 0} is singular , or (ii) χ(f
α−1(0)) < χ(f
α−10(0)) for generic α.
3.2. Remark. In the general case, it is not sufficient to use the group H
1(f
α−1(0)) to distinguish the generic curve from the special curve. For example, consider the family of polynomials f
α(x, y) = y
3+ xy
2+ y − α, α ∈ C. Then B
f= {0} and rank H
1(f
α−1(0)) = rank H
1(f
0−1(0)) = 1.
There is no loss of generality in assuming that the map l
α= x is simple for each α near a given α
0(l
αis said to be simple iff l
−1α(x) consists of d − 1 distinguished points for every critical value x of l
α). Then the number of singular points of l
αis exactly k(α). Let
(x
1(α), y
1(α)), . . . , (x
k(α), y
k(α))
be the critical points of the map l
α, α 6∈ B
∞. Now we use the notations as in the proof of Lemma 2.4. Suppose that x
∗∈ C is a common regular value of l
α0and l
αfor all α near α
0and let e
j(α) := l
−1α(T
j) ∩ S(f
α−1(0)), j = 1, . . . , k, be the cycles of the group H
1(f
α−1(0), l
−1α(x
∗)) corresponding to the singular points (x
j(α), y
j(α)). These elements define a basis of the group H
1(f
α−1(0), l
−1α(x
∗)). By definition, α
0∈ B
∞iff there exist critical points (x
j(α), y
j(α)) of l
αsuch that k(x
j(α), y
j(α))k → ∞ as α → α
0. Moreover, since the map l
αis simple, the number of singular points of l
αtending to infinity as α → α
0is r := k − k(α
0). Therefore, we may assume without loss of generality that such critical points are
(x
1(α), y
1(α)), . . . , (x
r(α), y
r(α)).
3.3. Definition. We call e
j(α), j = 1, . . . , r, the semi-cycles vanishing at infinity as α → α
0.
From Theorem 3.1 and the above definition we easily obtain the follow- ing.
3.4. Theorem. α
0is a bifurcation value of the family of affine curves {f
α(x, y) = 0} if and only if either
(i) the curve {(x, y) | f
α0(x, y) = 0} is singular , or
(ii) there exist semi-cycles e
j(α) vanishing at infinity as α → α
0. The number of such semi-cycles is exactly k − k(α
0).
3.5. Remark. The number of semi-cycles e
j(α) vanishing at infinity can
be given in other ways as follows.
(i) Let (F, G) be the intersection number of two curves {F = 0} and {G = 0}. Then
χ(f
α−10(0)) − χ(f
α−1(0)) = k − k(α
0)
= (f
α, ∂f
α/∂y) − (f
α0, ∂f
α0/∂y)
= dim
CC[x, y]/(f
α, ∂f
α/∂y)
− dim
CC[x, y]/(f
α0, ∂f
α0/∂y).
(ii) In CP
2we consider the family of curves
Γ
α:= {(x : y : z) | z
df
αx/z, y/z = 0}.
Clearly, Γ
αis the compactification of Γ
α:= f
α−1(0). In addition, we assume that the homogeneous part of degree d of f
αdoes not depend on α. Then the curves Γ
αintersect the line z = 0 at the same points A
1, . . . , A
sfor any α. Let µ
Ai(Γ
α) be the Milnor number of the germ of the analytic curve Γ
αat A
i. By arguments of [3], we can show that
k − k(α
0) = χ(f
α−10(0)) − χ(f
α−1(0))
=
s
X
i=1
[µ
Ai(Γ
α0) − µ
Ai(Γ
α)].
Next, we describe the change in the homotopy type of the curve {f
α(x, y) = 0} as α → α
0, α
0∈ B
∞.
3.6. Definition. An operation of attaching a 1-dimensional cell to an affine curve V is a map
ϕ : I := [0, 1] → C
2such that
(i) ϕ(I) is diffeomorphic to I;
(ii) ϕ(I) ∩ V = {ϕ(0), ϕ(1)}.
The set V
0= V ∪ ϕ(I) is called V with a 1-cell attached .
3.7. Theorem. Let α
0be a bifurcation value corresponding to the singularity at infinity of the family {f
α(x, y) = 0} such that the curve {f
α0(x, y) = 0} is smooth. Then a generic curve f
α−1(0) may be obtained from f
α−10(0), up to homotopy type, by attaching exactly k − k(α
0) 1-dimen- sional cells.
P r o o f. Let S(f
α−1(0)) (resp. S(f
α−10(0))) be the skeleton of the affine
plane curve {f
α(x, y) = 0} (resp. {f
α0(x, y) = 0}) as in the proof of
Lemma 2.4. According to the construction of skeletons, the set S(f
α−1(0))
(resp. S(f
α−10(0))) is a graph with d + k (resp. d + k(α
0)) vertices and 2k
(resp. 2k(α
0)) edges.
Furthermore, S(f
α−10(0)) is obtained from S(f
α−1(0)) by deleting k−k(α
0) vertices (x
β(α), y
β(α)), β = 1, . . . , r, and k−k(α
0) pairs of edges. These pairs of edges connect a point (x
β(α), y
β(α)) tending to infinity to two distinct points in l
−1α(x
∗). In other words, the set S(f
α−1(0)) is S(f
α−10(0)) with k − k(α
0) 1-cells attached.
On the other hand, the graph S(f
α−1(0)) (resp. S(f
α−10(0))) is a deforma- tion retract of the curve f
α−1(0) (resp. f
α−10(0)). This proves the theorem.
3.8. Corollary. If α
0is a bifurcation value at infinity of the family of curves {f
α= 0}, then the number of connected components of the curve f
α−10(0) is greater than or equal to the one for f
α−1(0).
Moreover, we can describe a change mechanism of the number of con- nected components when passing from the general curves f
α−1(0) to the special curve f
α−10(0). For that, we need:
3.9. Definition. A subgraph B(α) of the graph S(f
α−1(0)) is said to be a block vanishing at infinity as α → α
0if the following four conditions are satisfied.
(i) B(α) is connected;
(ii) each vertex of B(α) either belongs to l
−1α(x
∗) or tends to infinity as α → α
0;
(iii) the number of connected components of S(f
α−1(0)) is different from that of S(f
α−1(0)) \ B(α);
(iv) B(α) is minimal in the sense that there exists no subgraph B
0(α) ! B(α) of S(f
α−1(0)) satisfying (i)–(iii).
Let v(α
0) be the number of blocks vanishing at infinity as α → α
0, and let b
0(α) and b
0(α
0) be the numbers of connected components of f
α−1(0) and f
α−10(0), respectively. By Theorem 3.7, we obtain the following.
3.10. Theorem. b
0(α
0) − b
0(α) = v(α
0).
4. Corollaries
4.1. We begin by recalling some facts on the topology of polynomials of two variables.
Let F : C
2→ C be a polynomial function. It is well known that there exists a finite set C(F ) ⊂ C, called the bifurcation set of F , such that the restriction
F : C
2\ F
−1(C(F )) → C \ C(F )
is a locally trivial C
∞-fibration (see, for example, [5], [6], [7], [3]).
We say that a value t
0∈ C is regular at infinity if there exist a small
δ > 0 and a compact K ⊂ C
2such that the restriction
F : F
−1(D
δ) \ K → D
δ, D
δ:= {t | |t − t
0| < δ}, is a trivial C
∞-fibration ([4]).
If t
0is not regular at infinity, it is called a critical value at infinity of F . Denote by C
∞(F ) the set of critical values at infinity of F . It is known ([3]) that C(F ) = C
f(F ) ∪ C
∞(F ).
Let F be a polynomial with isolated critical points only. Denote by µ
c(F ) the fibre Milnor number of F at c.
Let f
α(x, y) := F (x, y) − α, α ∈ C. Using the notation of Remark 3.5, we put
λ
α0(F ) :=
s
X
i=1