Tohoku Math. J.

53 (2001), 533-552

QUADRATIC VANISHING CYCLES, REDUCTION CURVES

AND REDUCTION OF THE MONODROMY GROUP OF
PLANE CURVE SINGULARITIES

NORBERT A'CAMPO

(Received January 5, 2000, revised December 21, 2000)

Abstract. The geometric local monodromy of a plane curve singularity is a diffeomor-

phism of a compact oriented surface with non empty boundary. The monodromy diffeomor-
phism is a product of right Dehn twists, where the number of factors is equal to the rank of
the first homology of the surface. The core curves of the Dehn twists are quadratic vanish-
ing cycles of the singularity. Moreover, the monodromy diffeomorphism decomposes along
reduction curves into pieces, which are invariant, such that the restriction of the monodromy
on each piece is isotopic to a diffeomorphism of finite order. In this paper we determine the
mutual positions of the core curves of the Dehn twists, which appear in the decomposition of
the monodromy, together with the positions of the reduction curves of the monodromy.

1. Introduction. Let f : C 2 C be a polynomial mapping with f(0) = 0 having

an isolated singularity at 0 C2. Let B C2 be a Milnor ball for the singularity. A disc
D C with center 0 C is called small relative to B if for all s D the intersection of B
with {p C2 | f(p) = s} is a transversal intersection of smooth differential manifolds.
A deformation of f is a polynomial mapping f : C2 × C C, (p, t) H>- ft(p), with
f0(p) = f(p). A deformation of f is small relative to the Milnor ball B and the disk D if
for all s D and all t C, |t| 1, the intersection of B with {p C2 | ft(p) = s} is a
transversal intersection of smooth differential manifolds and moreover, the map ft is regular
on ft 1( D) B.
Let d D be the intersection point of D with the positive real axis. The fibers
ft 1(d) B, |t | 1, are all diffeomorphic to the Milnor fiber F := f 1(d) B, and
the diffeomorphism can be chosen to be unique up to isotopy if one follows the family

A small deformation ft of f is a morsification of f if for all t ]0, 1], the map ft in

B ft 1(D) has only singularities of Morse type. A real morsification of a singularity with
a real defining equation f is a morsification ft with ft real, such that all critical points of the
maps ft:B Care real and the level set ft 1 (0) B, t ]0, 1], contains all the saddle
points of the restriction of ft to D2 = B R2. The level set ft 1 (0) D2 is then a divide for
the singularity of f, see [AC2,G-Z].
Let ft be a morsification, such that for some t e ]0, 1] the restriction to B has µ(f)
distinct critical values. A simply closed essential curve c on the surface ft¯ 1(d) B is a
quadratic vanishing cycle if there exists a continuously differentiable path γ : [0, 1] D
2000 Mathematics Subject Classification. Primary 14H20; Secondary 14D05, 14H50, 32S25.
534 N. A'CAMPO

such that γ(0) = d, γ(u),u [0, 1[, are regular values of the restriction of ft to B and in
t¯ 1 ([0, 1]) the curve c is null homotopic. A simply closed essential curve c on the Milnor
fiber F is a quadratic vanishing cycle if there exists a morsification ft and | t \ < 1, t ^ 0, such
that c is mapped to a quadratic vanishing cycle on ft¯ 1(d) B by the natural diffeomorphism
class.
The geometric monodromy group of the singularity of f is the image Γf of the geometric
monodromy representation in the mapping class group of the surface with boundary (F, F)
of the fundamental group of the complement (as germ) of the discriminant in the unfolding of
the singularity of f [T2]. It follows from the results and constructions of Egbert Brieskorn [B]
and a theorem of Helmut Hamm and Lê Dung Tráng [H-L], that the geometric monodromy
group is generated by the Dehn twists whose core curves are the curves of any distinguished
system of vanishing cycles. Distinguished systems of vanishing cycles for isolated plane curve
singularities have been constructed by [AC2, G-Z]. The set of quadratic vanishing cycles is
an orbit of the action of the geometric monodromy group of the singularity. It follows that in
principle the set of quadratic vanishing cycles is known.
A reduction curve for an isotopy class of diffeomorphisms T of a surface F is an essential
simply closed curve a on F, such that for some integer N > 0 the curves a and TN (a) are
homotopic and moreover, if N > 0 is chosen minimal, the curves Ti (a), 0 i N 1, can
be made pairwise disjoint by an isotopy of T. The set {Ti (a) | 0 i N 1} is then called
a reduction system for T.
The set of reduction curves for the geometric monodromy of a singularity with one local
branch is given by [AC1]. The purpose of the present paper is to describe the relative position
of a distinguished base of vanishing cycles and the set of reduction curves for the monodromy
diffeomorphism of isolated plane curve singularities.
Roughly speaking, we wish to give a picture of the Milnor fiber of a plane curve singu-
larity, that shows both a distinguished system of quadratic vanishing cycles and the reduction
curves of the geometric monodromy. The method uses the model of the Milnor fiber with
monodromy coming from a divide of the singularity [AC2-5,G-Z]. It turns out that the divides
constructed by Sabir Gusein-Zade are most suited for this purpose.
In Section 4 we will give an application of our constructions to the study of the geome-
tric monodromy group of an irreducible plane curve singularity with several essential Puiseux
pairs. It is proved that the geometric monodromy group of an irreducible plane curve singula-
rity in a natural way contains a product of many copies of the geometric monodromy groups
of its companion singularities.
I like to thank the referee for his valuable comments.

2. Building blocks and cabling. It was observed by René Thom that the Chebyshev
polynomials T : C C up to affine equivalence of functions are precisely the polynomial
mappings from C to C with two or less critical values and with only quadratic singularities
[T1]. The standard Chebyshev polynomial T(p,z) of degree p has critical values + 1 , 1,
the symmetry T(p, z) = ( 1)pT(p, z), and the coefficient of zp is 2p 1. For p = 1, the
 VANISHING CYCLES AND REDUCTION CURVES 535

map T(p,z) has no critical values and for p = 2, the map T(p, z) has only the critical value
— 1. The Chebyshev polynomial T(p, z) satisfies the identity T(p, cos(x)) = cos(px), and
its restriction to [ 1, 1] is defined by T(p, t) = cos(parccos(t)).
Sabir Gusein-Zade has constructed real morsifications for real plane curve singularities
with Chebyshev polynomials. The building blocks for his construction are the real morsifica-
tion for the map f(x, y) = 2p 1xp 2q 1yq given by the family
fs(x, y) = spq(T(p, x/sq) T(q, y/s p)), s [0, 1].
For each s ]0, 1] the function fs:C2 C has µf = (p 1)(q 1) quadratic singularities
all at points with real coordinates, its critical values are contained in { 2spq, 0, 2spq} and
lims 0 ,s>0 fs = f.
If the exponents p and q are relatively prime to each other, the level set {f(x, y) = 0}
can be parametrized by the monomial map

It is a miracle that in this case the Chebyshev polynomials can be used to parametrize the level
sets {fs(x, y) = 0} as well. The map
t C i-> (sqT(q, t/s), s pT(p, t/s)) C2
parametrizes the level set {fs(x, y) = 0}. We recall that a divide is the image of a generic
relative immersion of a compact 1-manifold in the unit disk D2 in R2. For each s ]0, 1] the
intersection Pp,q;s := {fs(x, y) = 0} D2 with the unit disk D2 is a divide for the singularity
of {2p 1xp 2q 1yq = 0} at 0 C2.
The curve Pp,q := {f1(x, y) = 0} can be drawn in a rectangular box as in Figure 1. As a
first type of building block we will need the box B := [ 1, 1] × [ 1, 1] with the curve Pp,q.
If (p, q) = 1 holds, the curve Pp,q is the image of

Tp,q : [ 1, 1] B, Tp,q(t) := (T(p, t), T(q, t)),

FIGURE 1. Divide in box [ 1, 1] × [ 1,1] for 26x7 2 4 y 5 = 0.

536 N. A'CAMPO

which leaves the box through the corners. In general the immersed curve has several com-
ponents, which are immersions of the interval or of the circle. At most two components are
immersions of the interval, which leave the box through the corners.
Let P be any divide having one branch given by an immersion γ : [ 1, 1] D2. We
assume that the speed vector γ(t)˙ and the position vector γ(t) are proportional at t = ± 1 ,
i.e., the divide P meets D2 at right angles. Let Nγ : [ 1, 1] × [ 1, 1] D be the
corresponding immersion of a rectangular box, i.e., the restriction of Nγ to [ 1, 1] × {0} is
the immersion γ and the image of Nγ is in a small tubular neighborhood of P. For instance,
for a small value of the parameter η R>0 the following expression defines an immersion
Nγ : B R2 of the rectangular box B : = [ 1, 1] × [ 1, 1]:

(K(
Nγ(s, t) := γ(t) + sn ,
II )> (Oil
where J is the rotation ofR2 over π/2. The four corners Nγ(±1, ±1) are on the circle of
radius σ := 1 + η2. We finally define

that will be an immersion Nγ : B • D2 mapping the corners of the box B into D2.
We will denote by Pp,q * P the divide in D2, which is the image by Nγ : B D2 of
Pp,q B, see Figure 2. The number of double points δ(Pp,q * P) of Pp,q * P is computed
inductively from the number of double points δ(P) of P by:

2
*P) = (p 1)(q 1)/2 + δ(P)p .
Let Rηγ be the union of the image of Nηγ with the two chordal caps at the endpoints
of γ. The connected components of D2 \ Rηγ correspond via inclusion to the connected
components of D2 \ P. We declare a connected component of D2 \ PPA * P to be signed by
+, if the component contains a component of D2\Rηγ that corresponds to a + component of
D2 \ P. In this case we call the connected component of D2 \ Rηγ a P+-component. Observe
that there exists a chess board sign distribution for the components of D2 \ Rηγ that makes
P+ -components indeed to + components.
The field Φp,q of cones on the box B R2 is the subset in the tangent space of TB
given by:
0M :={(x,u)€TB\\(u,ei)R2\ >cos(a(x))||n||},
2
where e1 = (1, 0) R and α : B R is a function such that for every (x, u) TB with
x Pp,q and u Tx Pp,q we have the equality

\(ii,ei)Ri\ =cos(a(jc))||w||.

Moreover, α has the boundary values α ( ± 1 , t) = 0 and α(s, ±1) = π/2. We interpolate the
function α on B by upper and lower convexity, i.e., such that 2α/ t2 < 0 and 2α/ s 2 > 0.
The definition of α(x) seems to be conflicting at the double points of the curve Pp,q; at a
 VANISHING CYCLES AND REDUCTION CURVES 537

double point x = (x1,x2) of the curve Pp,q the two tangents lines to Pp,q have opposite
slopes tan(α(x)) and tan(α(x)), since the curve Pp,q is defined by the equation

T(q,x1) T(p,x2) = 0
that separates the variables. For example, a nice such function α is given by:

α(x1,x2) : = arctan

The interest of the field < 5 M comes from the following lemma, that is immediate from
the definitions. The differential dNηγ : TB • TR2 maps the field of sectors Φp,q TB to
a subset in TR2, which will be denoted by Φη,p,qγ.
LEMMA. Let the image of γ : [ 1, 1] D be a divide P, that meets D2 at right
angles. For η > 0 small enough, the intersection S3 Φη,p,qγ of subsets in TR2 is a tubular
neighborhood of the knot L(P). The composition of Pp,q : [ 1, 1] • B : = [ 1, 1]×[ 1, 1]
and ofNγ : B • D2 is again a divide, whose knot is a torus cable knot of the knot L(P).
For small η the image Φη,p,qγ contains those vectors that have feet near P and form a
small angle with the tangent vectors of the divide P.
A sequence of pairs of integers (a i , bi)1 i k is a sequence of essential Puiseux pairs of
an irreducible plane curve singularity if the inequalities 2 ai < b i and bi/a 1a2 · · · ai <
bi+1/a 1a2··· aiai+1 are verified and if moreover, the integers b i and a1a2 · · · ai are relatively
prime. A sequence of essential Puiseux pairs defines a family of topologically equivalent sin-
gularities. A specific member fk(x, y) of this family is obtained from the Puiseux expansion
with fractional and strictly increasing exponents

\-x
by the rule, which takes into account the ramification of x1/a1a2···ak,

fk(x, y) =

FIGURE 2. The divide P2,9 *P2,3.

538 N. A'CAMPO

where θ runs over the a1a2 ···ak roots of £flifl2-«* — x = 0 in the algebraic closure of the
field C((x)). The coefficients of the polynomial fk(x, y) are integers.
For example, the Puiseux expansion y = x3/2+x7/4 leads to the polynomial f(2,3),(4,7) =
(y2 x3)2 4x5y x7 and the Puiseux expansion y = x3/2 + x 1 1 / 6 to the polynomial
(y2 x3)3 6x7y2 2x10 x11.
Let {fa,b(x, y) = 0} be a singularity having one branch and with essential Puiseux pairs
(ai, bi)1 i n. The theorem of Gusein-Zade [G-Z] very efficiently describes a divide for the
singularity {fa,b(x, y) = 0} in a closed form, namely the iteratively composed divide

where the numbers b'2,..., b'n can be computed recursively, as we will show here below. We
denote by Sk, 1 k n, the divide
P P P P
ak,b'k * ak_y,b'k_y *•••* a2,b'2 * a1,b1 ,

and let fk(x, y) be a specific equation for a singularity with essential Puiseux pairs
(ai, bi)1 i k.
Remember that the product a1a2 ... ak is the multiplicity of the singularity at 0 of the
curve {fk(x, y) = 0} and that the linking number λk of L(Sk) and L(Sk 1) in S3 can be
computed recursively by:

A.i = b1 , λk+1 = bk+1 bkak+1 + λkakak+1 .

The linking number λk is equal to the intersection multiplicity

dimC[[x, y]]/(fk(x, y), fk 1(x, y))

at 0 of the curves {fk(x, y) = 0} and {fk 1(x, y) = 0}. We have also for the linking number
λk an interpretation in terms of divides

λk = #(Sk Sk 1) = 2akδ(Sk 1) + b'k .

To see the first equality, observe that the fiber over 1 of the knot of a divide P, as con-
structed in [AC5], intersects the knot of a divide Q positively in #(P Q) points, provided
that the divides P and Q are mutually in general position. Remembering that we already have
computed recursively the numbers δ(Sk) and #(Sk Sk 1), we conclude that b'k too can be
computed recursively.
For example, for the Puiseux expansion y = x3/2 + x7/4 we have: λ1 = 3, λ2 =
1 3 ·2 + 3· 2· 2=13, b'2 = 13 2 ·2· 1 =9. Hence, the divide for the irreducible
singularity with Puiseux expansion y = x3/2 +x7/4 is the divide P2,9 * P2,3, see Figure 2. For
the Puiseux expansion y = x3/2+x11/6 we found: λ1 = 3, λ2 = 11 3·3 + 3·2·3 = 20, b'2 =
20 2 · 3 · 1 = 14. Hence, the divide for its singularity {(y2 x3)3 6x7y2 2x10 x11 = 0 }
is P3,14 * P2,3, see Figure 3.
An iterated *-composition of divides has to be evaluated from the right to the left.
Using [AC4-5], we can read off from this divide the Milnor fibration of the singularity
{fa,b(x, y) = 0}. In particular, we can describe the Milnor fiber with a distinguished base of
 VANISHING CYCLES AND REDUCTION CURVES 539

quadratic vanishing cycles, see the next section. Using the above iterated cabling construction,
in Section 4 we will also read off from the divide the reduction of the geometric monodromy
of an irreducible plane curve singularity, as described in [AC1]. For instance, intersection
numbers in the sense of Nielsen of quadratic vanishing cycles and reduction cycles can be
computed.
In general, for an isolated singularity of a real polynomial f(x, y) having several local
branches, the divide {f1(x, y) = 0} D of a real morsification ft(x, y) may have immersed
circles as componants. The above cabling construction Pp,q * P does not work if the divide
P consists of an immersed circle. Of course, if one is willing to change the equation of the
singularity to an equation, which defines a topologically equivalent singularity and which
has only real local branches, one will only have to deal with divides consisting of immersed
intervals. If we do not want to change the real equation, we will need a second type of building
blocks for a cabling construction, see Figure 4.
These building blocks are the divides Lp,q in the annular region A := {(x, y) D2
2 2
1/4 x +y 3/4}. If for the integers (p, q) = 1 holds, the divide Lp,q is the Lissajous

for
FIGURE 3. The divide P 3 , 1 4 * P2,3 (y2 x3)3 6x 7 y 2 2x10 x1

FIGURE 4. The building block L3,5 of Lissajous type.

540 N. A'CAMPO

FIGURE 5. The cabling of L3,5 * P2,4.

curve
s [0, 1] i-> (1/2+ 1/4sin(2πqs))(sin(2πps), cos(2πps))
2
in A D . The curve Lp, q has q-fold rotational symmetry. If (p, q) = r > 1, the divide
Lp, q will be defined as the union of r rotated copies of Lp/r,q/r with rotations of angles
2πk/p, k = 0 , . . . , r 1, of D 2 . Again, the system of curves Lp, q has a q-fold rotational
symmetry.
The star-product Lp, q * P can be defined as above if the divide P consists of one im-
mersed circle. We leave many details to the reader. The two types of building blocks Pp,q
and Lp,q together with the star-products Pp,q * P and Lp, q * P will allow one to describe the
iterated cablings of real plane curve singularities in general. See Figure 5.

3. Visualization of the vanishing cycles for a divide. Let P be a connected divide

and let πP : S3 \ L(P) S 1 be the fibration of the complement of the link L(P) over
S 1 as in [AC4-5]. The fibration map is given with the help of an auxiliary Morse function
fP : D2 R. The fiber FP := jtp l{\) above 1 S1 projects to the positive components of
the complement of P in D2. One has that the closure of
{(x, u) TD2 | fP(x) > 0, (dfP)x(u) = 0, | | x | | 2 + | | M | | 2 = 1}

in S3 \ L(P) is the fiber surface FP.

To each critical point of fP : D2 R corresponds a vanishing cycle on the surface FP.
In the case, where the divide P is a divide of a singularity, the surface FP is a model for the
Milnor fiber and the system of vanishing cycles on FP is a model for a distinguished system
of quadratic vanishing cycles of the singularity.
Let M be a maximum of fP. The vanishing cycle ΔM is the non-oriented simply closed
curve on the fiber FP, see Figure 6,
8M:={u€TMD2\\\M\\2+\\u\\2 = 1}.
Let c be a crossing point of P. The point c is a saddle type singularity of fP. The van-
ishing cycle is the non-oriented simply closed curve δc on FP that results from the following
construction. Put P+ : = {x D2 | fP(x) > 0}. Let gc P+ {c} be the singular gradient
 VANISHING CYCLES AND REDUCTION CURVES 541

FIGURE 6. The vanishing cycle ΔM for a maximum.

FIGURE 7. Tear for vanishing cycle δc for a saddle point.

line through c, for which the endpoints are a maximum of fP or a point in D 2 . We splice gc
and get a double tear tc P+ {c} as in Figure 7. The tear tc is a closed curve that has at c
a non-degenerate tangency with gc from both sides. Moreover, tc is perpendicular to gc at the
endpoints of gc, if the endpoint is a maximum of fP and else t c has a tangency with D2. The
vanishing cycle δc is the closure in FP of the set

{(x, U) FP | x tc, (dfP)x u points to the inside of the tear tc

Let M be a minimum of fP. The following is a description of the vanishing cycle δm on

FP. The projection of δm in D 2 is a simply closed curve tm in P + {c1, c 2 , . . . , ck}, where
{c1, c 2 , . . . , ck} is the list of the double points of P that lie in the closure of the region of
M, see Figure 8. The curve tm and the singular gradient line gci, 1 i k, coincide in a
neighborhood of c i . Moreover, if the endpoint of gc is a maximum M, the curve tm leaves
trans versally the tear tci at M and enters trans versally the next tear tci±1. If the endpoint of gc
is on the boundary of D, the curve which leaves the tear tci becomes tangent to the boundary
of D2 and enters the next tear, see Figure 8. The vanishing cycle δm is the non-oriented simply
542 N. A'CAMPO

FIGURE 8. The vanishing cycle δM for a minimum.

closed curve on FP, that is, the closure of

{(x, u) FP | x tm, u points inwards to the disk bounded by tm}.
The link of a divide P is naturally oriented by the following recipe. Let γ :] 0, 1 [
D2 be a local regular parametrization of P. The orientation of L(P) is such that the map
t ]0, 1[ H>- (y(f): M 0 x ( 0 ) e £ ( ^ ) is oriented. Here λ(t) is a positive scalar function,
which ensures that the map takes its values in L(P). For a connected divide P we orient
its fiber surface FP such that the oriented boundary of (FP L(P), L(P)) coincides with
the orientation of L(P). Vanishing cycles δc, where c is a critical point of fP, do not carry
a natural orientation, since the third power of the geometric monodromy of the singularity
{x3 y2 = 0} reverses the orientations of the vanishing cycles. The main use of a vanishing
cycle δc in this paper is through the associated right Dehn twist c of FP. This use does
not require an orientation for the cycles δc, but requires an orientation of the surface FP.
Moreover, we orient the tangent space TD = D×R2 and its unit sphere S3 as boundary of its
unit ball, such that the linking numbers of links L(P1) and L(P2) are positive for generic pairs
of divides P1 and P2. In fact, the orientation TD2 is opposite to the orientation as tangent
space. We have LkS3(L(P1), L(P2)) = #(P1 P2), a fact which was already used in the
previous section.

4. Reduction cycles and reduction tori. We consider a divide of the form Q :=

Pp,q * P, where the divide P is given by an immersion γ : [ 1, 1] • D2. So, the divide Q is
the image of Nγ•Tp,q. If we change the immersion by a reparametrization γ1 : = γ•φ, where
0 is an oriented diffeomorphism of [ 1, 1], the divide Q1 : = N γ 1 •φ([ 1, 1]) is isotopic
to Q by an transversal isotopy, which does not change the type of its knot. By choosing
4> appropriately and η small, one can achieve that each double point of P corresponds to a
 VANISHING CYCLES AND REDUCTION CURVES 543

FIGURE 9. Manhattan: crossing of the box with 4 by 4 strands.

system of p2 double points of the divide Q1, which look like the intersection points of a
system of p almost parallel lines with an other system of p almost parallel lines, see Figure
2, where p = 2, Figure 3, where p = 3 and Figure 9.
We may assume that the divide Q for each double point of P already has a grid of p2
intersections.
We will construct reduction curves for the monodromy of the knot L(Q) by the method
of [AC1]. The reduction curves of the cabling Pp,q *n P are the intersection of the fiber FQ
over 1 S1 of the fibration on the complement of the knot L(Q) with the boundary of a
regular tubular neighborhood U of the closed tubular neighborhood V of the knot L(P) for
which L(Q) V holds. The intersection (FQ U) FQ is indeed a system of reduction
curves provided that the torus U is transversal to the fibration of the knot L(Q).
Assume that the tubular neighborhood V was constructed with the field Φp,q and a par-
ticular value of the parameter η. The same field of sectors, but a slightly bigger parameter
value rj' yields a tubular neighborhood U of V in S3. The construction of the fibration will
be done as in [AC5]. The main choice for the construction of the fibration for the knot L(Q)
is a Morse function fQ : D2 R with fQ 1(0) = Q. For our purpose here, where we
must achieve the above transversality, we will choose fQ as follows. First, after applying a
regular transversal small isotopy, we may assume that the divide P has perpendicular recti-
linear crossings. Next, we consider a Morse function fP : D2 R for the divide P that
is Euclidian near its crossings. Let the fibration on the complement of the knot L(P) be
3 1
np^r, : S \L(P) S , where Π (X,U) := θP(x, u)/|θP(x, u)| and
P

1
θP,λ(x, u) := fP(x) + iλ dfP(x)(u) 1λ 2χ(x)HfP(x)(u, u).

The function χ : D2 R is a bump function at the crossing points of P and λ is a big real
2
parameter. We now choose a small positive real number v, such that {x D | |fP(x)| v}
is a regular tubular neighborhood of P, that meets each component of {x D2 | χ (x) = 1}.
Next we choose rj' > 0 such the corners of Nrt,y(B) are in {|fP(x)| = v}, i.e., r]/2 = v.
We construct the torus knot L(Q) with Q := Pp,q *n P, where 0 < η < rj'. Since Q
544 N. A'CAMPO

{|fP(x)| < v} holds, we can construct a Morse function fQ : D2 R for the divide Q, such
that on {|fP(x)| v} the function fQ is constant along the level sets of fP.
The following theorem follows directly from Lemme 2, [AC1, page 153] and the above
construction.
THEOREM 1. The torus d<t>n\p,qy is transversal to thefibration deduced from fQ on
the complement of the knot L(Q). The intersection

is a system of p closed curves on the fiber FQ, which is a reduction of the monodromy of
L(Q).
With a few examples, we now explain how to depict in the Milnor fiber a distinguished
system of vanishing cycles and the reduction curves for the monodromy of a singularity, for
which a divide of the form Q = Pp,q * P is given.
The fiber FQ with a distinguished system of vanishing cycles is already constructed in
Section 3.
For a (p, q) cabling the reduction system consists of p simply closed curves on the fiber
FQ . Each of them cuts out from FQ a surface diffeomorphic to the fiber FP of the divide P.
The p copies of FP in FQ are cyclicly permuted by the monodromy TQ .
One of those copies can be visualized more easily as follows. Let {x D2 | fP(x) v}.
For each double point c of P we connect the two components of {x D2 | fP(x) v}, that
are incident with the double point c, by a special bridge which projects diagonally through
the Manhattan part of the divide Q, that corresponds to c. The projection of the bridge is a
twisted strip Sc in D2, that realizes a boundary connected sum of the P+ -components. The
twist points of the strip Sc are precisely the critical points of fQ, that lie on the diagonal. The
boundary of Sc consists of two smooth curves, that intersect each other transversally and that
also intersect the divide Q transversally.

•component
(x) = v

fix, -
P-l--componej

FIGURE 10. Bridge through Manhattan.

 VANISHING CYCLES AND REDUCTION CURVES 545

FIGURE 11. Bridge through a block of Manhattan.

Let C be the union of the projections Sc of the bridges with {x D2 | fP(x) v}, see
Figure 10. In Figure 11, we have zoomed out one block to show more details. The copy FP, Q
of the fiber of the knot L(P) is the closure in the fiber FQ of the knot L(Q) of the set

{(x,u) TD2 x C, (dfQ)x ± 0, (dfQ)x(u) = 0} S3

The first reduction curve is the boundary of the surface R := FP,Q of the surface
FP,Q FQ.
The reduction system is the orbit {R, TQ(R), TQ2(R), ...} under the monodromy TQ of
the singularity with divide Q.
Our first example is the singularity with two essential Puiseux pairs (x3 y2)2 4x5y
7
x , whose link is a two stage iterated torus knot. Its divide Q = P2,9 * P2,3, (see Figure 2), has
two P+ -components, where P = P2,3, (see Figure12, where the projection of the reduction
curve R is drawn). In this case Manhattan consists of one block. The reduction curve R is the
pre-image in the fiber FQ of its projection proj(R) D2 under the map (x, u) H>- X a drawn
in Figure 12. That means R is the closure in FQ of the set

{(x,u) TD2|x proj(R), (dfQ)x £ 0, (dfQ)x(u) = 0, | | x | | 2 + | | U | | 2 = 1 } .

The reduction system is {R, TQ(R)}.
The curve R is homologically trivial in FQ . It turns out that the power TQ156 of the
monodromy is the composition of the right Dehn twists, whose core curves are {R, TQ(R)}.
The power TQ156 is a product of 2496 = 16 × 156 Dehn twists, since TQ is the product of those
Dehn twists whose core curves are the system of distinguished quadratic vanishing cycles of
the real morsification with divide Q. It turns out that the expression as product of Dehn twists
is far from being as short as possible. In fact, the right Dehn twist R with core curve R
can be written as a product of 36 right Dehn twists that have core curves coming from the
morsification with divide Q. More precisely, the Dehn twist R factors as
1 1 6
R =( M • b • m• a• • ) .
546 N. A'CAMPO

FIGURE 12. Divide Q = P2,9 * P2,3 with reduction curves R and T(R) (dotted).

The factors are right Dehn twists whose core curves are among the quadratic vanishing cycles
δm, δa, ΔM, δb of the divide Q as indicated in Figure 11, ΔM is the vanishing cycle of a P+-
region, δm of the maximum of Manhattan, andδa, δb of street corners of Manhattan. It follows
that TQ156 can also be written as a composition of 72 Dehn twists with core curves among the
vanishing cycles of the divide Q. The composition b • m • a • m 1 o A^1 is the Dehn
twist with core curve a : = b( m(a)).
The reduction curve R cuts off from FQ a piece FP, Q of genus one (see also Figure 1 on
page 159 of [AC1]), and the Dehn twists M and Aa act only on this piece, since the curves
ΔM and a lie entirely in this piece; in this piece, that is a copy of the fiber FP, they generate
the geometric monodromy group of the accompanying singularity x3 y2 = 0 with divide
P2,3 .
Our second example is the singularity with two branches (x 3 y2)(y3 x2). Its homo-
logical monodromy is of infinite order [AC1]. Each branch is a torus knot. Again Manhattan
consists of one block. In Figure 13 we have drawn the projections of the curves R, R' and
S, S', that together are the boundary components of the two diagonals through Manhattan. In
this case the curves R and R' are isotopic to each other, as are the curves S and S'. A com-
plete reduction system for the geometric monodromy is the system {R, S}. Each component
of this system carries a non-trivial homology class. The isotopy classes of the curves R and
S are permuted by the monodromy TP, and hence the system {R, S} is invariant under the
monodromy.
Let h be the action of TP on the homology H1(FP, Z) of the the fiber FP. Let δa,δb,δc,δd
be the vanishing cycles of the double points, that are the corners of Manhattan of P, and let
δm be the vanishing cycle of the maximum in the center of Manhattan.
 VANISHING CYCLES AND REDUCTION CURVES 547

FIGURE 13. Divide P for (x3 y 2 )(y 3 x2) with reduction system R S.

FIGURE 14. Divide P for (x3 y 2 )(y 3 x2) with reduction system A B.

If one chooses the orientations appropriately, one has

[R] = [δa] + [δm] + [δc], [S] = [δb] + [δm] + [Sal H[R]) = [S], h([S]) = [R],

and hence also h([R] [S]) = [R] [S]. Let [k] be any cycle on FP, that is, carried by
a simple oriented curve k and intersects the curves R and S each transversally in one point.
One has h10([k]) = [k] ± ([R] + [S]), which shows that the homological monodromy h is
not of finite order. We have drawn in Figure 13 the oriented projection of such a cycle k, that
intersects the curves A and B. The curve A is halfway in between the curves R and R' on the
cylinder they cut out. Let B be the curve halfway in between S and S'. The curves A and B
are the reduction curves of Figure 14 on page 167 of [AC1]. The reduction system A, B is
much easier to draw, see Figure 14, where are drawn the projections in D2. The projections
548 N. A'CAMPO

meet transversally at the maximum in Manhattan of fP. The curve δm intersects transversally
1
in two points each curve R and S. One has h 0([δm]) = [δm] ± 2([R] + [S]).
The power TP1° of the geometric monodromy, that is, a word of length 110 in the Dehn
twists of the divide P is equal to the composition of those right Dehn twists, whose core
curves are R and S. So, the power TP1° also can be written as the much shorter word
1 1 1 X
AcoAmoAao A' o A' o Ad o Am o Ab o A' o A~d .

REMARK. The curves A and m( c(δa)) are isotopic, where m and c are the right
Dehn twists with core curves δm and δc. It follows that the reduction system A, B consists of
3 2 3 2
quadratic vanishing cycles of the singularity of {(x y )(y x ) = 0} with two branches.
In contrast, a reduction curve of a singularity with only one branch can not be a quadratic
vanishing cycle, since all reduction curves are zero in the homology, see [L,AC1]. It follows
that the action of the monodromy of an irreducible plane curve singularity on homology is of
finite order [L].

5. Geometric monodromy group and reduction system. Let the polynomial f(a,b)
be an equation for an irreducible plane curve singularity with n essential Puiseux pairs
(ai, bi)1 i n. The number of simply closed curves contained in a complete reduction sys-
tem R for the monodromy of f is

anan 1 · . . . · a2 + an 1an 2 · . . . · a2 + · · · + a3a2 + a2 .

f
Let Γ ,red be the subgroup of the geometric monodromy group of Γf of f of those elements
y Γf that up to isotopy fix each component of R. Let Γf,0red be the subgroup of Γ which
is generated by the Dehn twist, whose core curves are quadratic vanishing cycles and do not
intersect any component of R. Obviously, one has Γf,0red Γf,red, but we do not know if
this inclusion is strict. A component of F \ R is called a top-component if its closure in
F meets only one component of R. Let Γf,top be the subgroup of Γf of those monodromy
transformations, which induce the identity in each component of F \ R that is not a top-
component. Let F9 be the intersection Γf,top Γf,0red. We have
THEOREM 2. Let f = f(a,b) be an irreducible singularity with n 2 essential
Puiseux pairs (ai, bi)1 i n. Let g = f(a',b') be a singularity with the n 1 essential Puiseux
pairs (a', b') = (ai, bi) 1 i n 1. The group Γf contains the product of an copies of the group

THEOREM 3. Let f(a,b) be an irreducible singularity with n 2 essential Puiseux

pairs. The group Γ0 is isomorphic to the product of anan 1·... ·a2 copies of the geometric
monodromy group of the singularity ya1 xb1 = 0.
PROOF OF THEOREM 2. Let P be the divide Pcln_ub' *•• •*pa1,bl *pal,bl for the sin-
gularity of g and let Q = Pan,b'n * P be the divide for the singularity of f. A copy FP, Q of the
fiber FP is constructed as a subset of the fiber FQ . Remember, that FP is obtained by connect-
ing with strips the sets {(x, u) TD2 | fP(x) > 0, (dfP)x(u) = 0}, where fP : D2 R is a
 VANISHING CYCLES AND REDUCTION CURVES 549

Morse function for the divide P. For each double point of P there are two connecting strips.
To each +-component of P corresponds a P+-component of Q with the same topology and
to each double point of P corresponds a Manhattan grid of Q, in which we have drawn dia-
gonally the projection of the strips that connect {(x, u) TD2 | x QP,+,(dfQ)x(u) = 0}.
Here, QP,+ denotes the union of the P+ -components of the complement of the divide Q.
From the divide P is deduced a distinguished base of quadratic vanishing cycles for the sin-
gularity of f. Let BP be the union of the curves of this base. This base can be drawn on the
fiber FP, see Section 3.
In order to prove the theorem, we will construct inside FP, Q a system of simply closed
curves with union BP, Q , each of them being a quadratic vanishing cycle for the singularity
g, such that the pairs (FP, BP) and (FP,Q, BP,Q) are diffeomorphic. This finishes the proof,
since the Dehn twist, whose cores are the quadratic vanishing cycles of BP,Q, generate a copy
of Γg in Γf. By acting with the geometric monodromy T of the singularity f, one obtains an
commuting copies of Γg in Γf.
To each +-region of P corresponds one P+ -region of Q. The maximum of fP, say at
M in the region, is also a maximum of fQ. The quadratic vanishing cycle ΔM : = {(M, u)
TD2 | | | M | | 2 + | | U | | 2 = 1} of FQ lies in FP and also in FP,Q. For each double point c of
P the quadratic vanishing cycle δc FP projects in D2 to a tear splicing the gradient line
gc of fP through c. The endpoints of gc are maxima of fP or points on D2. The function
fQ has exactly one gradient line gg c that has the same endpoints as gc and coincides with
gc in a neighborhood of the common endpoints. The gradient line gg c runs along a diagonal
through the Manhattan grid corresponding to c. Let gg c be the simply closed curve on FQ,
that projects to a tear f g c equal to gg c , except above a neighborhood of its endpoints where
tQ,c equals tc. We remark that 5 g c is a cycle in FP,Q. Let c1,..., cp be the p := an double
points of Q that occur along the gg c and let M 2 , . . . , Mp along gQ c be the maxima. Let
δQ,c1 be the quadratic vanishing cycle of the singularity f that corresponds to c1. One verifies
that the cycles 5 g c and

c p • MP O • · • c 2 • M2(ΔQ,C1)

are isotopic. Here c i or MI stands for the right Dehn twist of FQ whose core curve is the
quadratic vanishing cycle δci or ΔMI of the singularity f. Hence 5 g c FP,Q is a quadratic
vanishing cycle for the singularity f. So far, we have constructed for each maximum and for
each saddle point of fP a simply closed curve on FP, Q that is a quadratic vanishing cycle of
the singularity f. These cycles intersect on FP, Q as do the corresponding quadratic vanishing
cycles of the singularity g on FP.
We now wish to construct for each minimum of fP a vanishing cycle on Fp g. We have
to handle two cases: p odd, see Figure 15, and p even, see Figure 16.
If p is odd, a minimum M of fP will also be a minimum of fQ. Let SQm be the van-
ishing cycle on FQ corresponding to M, see Figure 15. The projection of 5 g m into D2 is a
smooth simply closed curve s' transversal to Q, that surrounds the region of M through its
neighboring + regions of Q. One needs to take care that in each neighboring + component
550 N. A'CAMPO

the projection runs through the maximum of fQ in that region. The points of s correspond to
pairs (x, u) with x s' and u pointing inwards to M. Let r be a simply closed cycle on FP,Q
that projects into D2 upon the curve r', which now surrounds the region of M through the
P+ -components of Q, see Figure 15. In the Manhattan grids r' is just a diagonal, again r' runs
through the maxima of the regions or touches D. On r we only allow pairs (x,u), where u
points inwards to M. It is clear that the cycle r on FP,Q intersects the cycles of the previous
construction as the vanishing cycle to the minimum of fP intersects the vanishing cycles of
the critical points of fP. It remains, however, to check that the cycle r is a quadratic vanish-
ing cycle of the singularity of g. By applying to r FQ the Dehn twist corresponding to the
critical points of fQ that are in between the curves r' and s', one can transform the isotopy
class of the curve 5 g m to the class of the curve r. This proves that r is indeed a quadratic
vanishing cycle of the singularity of g.

FIGURE 15. Vanishing cycles on FQ from a minimum of fP and cycle r on

FIGURE 16. Vanishing cycle δQ,m on FQ from a minimum of fP and cycle r on FP, Q.
 VANISHING CYCLES AND REDUCTION CURVES 551

If p is even, then a minimum M of fP will be a maximum of fQ. Let δQ,M be the

vanishing cycle on FQ corresponding to maximum M := M, see Figure 16. Its projection into
D2 is the point M := M. Let r be a simply closed cycle on FP,Q that projects into D2 upon
the curve r' which now surrounds the region of M := M through the P+ -regions of Q, see
Figure 16. In the Manhattan grids r' is just a diagonal, again r' runs through the maxima of
the regions. On r we only allow pairs (x,u), where u points inwards to M. It is clear that the
cycle r on FP, Q intersects the cycles of the previous construction as the vanishing cycle to
the minimum of fP intersects the vanishing cycles of the critical points of fP. By applying to
ΔQ,M FQ the Dehn twist corresponding to the critical points of fQ that are in between the
curve r' and the point M : = M, one can transform the isotopy class of the curve SQ M to the
class of the curve r, and proves that r is indeed a quadratic vanishing cycle of the singularity
of g. As explained, this terminates the proof.
PROOF OF THEOREM 3. The proof of Theorem 2 constructs a copy ΓP,Q of the mono-
dromy group Γf of the singulatity f as subgroup in the monodromy group Γg of the singula-
rity g. This copy acts with support in a copy FP, Q of the fiber FP. The the first an 1 iterates
of the monodromy TQ of the singularity g constructs an copies of FP in FQ . By conjugation
with TQ one gets an copies from / > g . We end the proof by repeating this argument. One
gets anan 1 • • a2 commuting copies of the geometric monodromy group of the singularity

PROB LEM . We like to state the problem of presenting the geometric monodromy group
of plane curve singularities with generators and relations. It would be particulary nice to
express the presentation in terms of a divide of the singularity. The same problem can also
be stated for the homological monodromy group of plane curve singularities, but I think that
the problem for the geometric monodromy group is more tractable, since all reduction curves
can be taken into account. The theorems 2 and 3 are possibly first steps towards a solution
of this problem. However, an important missing piece in this program is a presentation with
generators and relations of the geometric monodromy group of the singularities yp xq = 0
for 3 p q, 7 p + q. The fundamental group of the complement of the discriminant
2 q
in the unfolding of the singularity y x = 0 is the braid group Bq 1. Bernard Perron
and Jean-Pierre Vannier have proved for the singularities y2 xq = 0 that the geometric
monodromy group is a faithful image of the braid group Bq 1 and that a similar result holds
for the singularities x(y2 xq) = 0 [P-V]. The fundamental group of the complement of
the discriminant in the unfolding of the singularity y3 x6 = 0 is the Artin A E6 group
of the Dynkin diagramm E6. Bronek Wajnryb has proved that the geometric monodromy
representation of A E6 into the mapping class group of the Milnor fiber of the singularity
3 6
y x = 0 is not faithful. Using the corresponding relation in the mapping class group,
Makoto Matsumoto has given a presentation of the mapping class group with local relations
[Ma].
552 N. A'CAMPO

REFERENCES
[AC1] N. A'CAMPO, Sur la monodromie des singularités isolées d'hypersurfaces complexes, Invent. Math. 20
(1973), 147-170.
[AC2] N. A'CAMPO, Le groupe de monodromie du déploiement des singularités isolées de courbes planes I, Math.
Ann. 213 (1975), 1-32.
[AC3] N. A'CAMPO, Le groupe de monodromie du déploiement des singularités isolées de courbes planes II, Actes
du Congrès International des Mathématiciens, tome 1, 395–404, Vancouver, B.C., 1974.
[AC4] N. A'CAMPO, Real deformations and complex topology of plane curve singularities, Ann. Fac. Sc. de
Toulouse Math. (6)8 (1999), 5-23.
[AC5] N. A'CAMPO, Generic immersions of curves, knots, monodromy and gordian number, Publ. Math. Inst.
Hautes Études Sci. 88 (1998), 151-169, (1999).
[B] E. BRIESKORN, Die Monodromie der isolierten Singularitäten von Hyperflächen, Manuscripta Math. 2
(1970), 103-170.
[G-Z] S. M. GUSEIN-ZADE, Matrices d' intersections pour certaines singularités de fonctions de 2 variables, Funk-
tsional. Anal. i Prilozen. 8 (1974), 11-15.
[H-L] H. H AMM AND LÊ DUNG˜ TRÁNG, Un théorème de Zariski du type Lefschetz, Ann. Sci. École Norm. Sup.
(4) 6 (1973), 317-366.
[L] LÊ DUNG˜ TRÁNG, Sur les noeuds algébriques, Compositio Math. 25 (1972), 281-321.
[Ma] M. MATSUMOTO, A presentation of mapping class groups in terms of Artin groups and geometric mono-
dromy of singularities, Math. Ann. 316 (2000), 401–418.
[M] J. MILNOR, Singular Points on Complex Hypersurfaces, Ann. of Math. Stud. 61, Princeton University Press,
Princeton, 1968.
[P-V] B. PERRON AND J.P. VANNIER, Groupe de monodromie géométrique des singularités simples, Math. Ann.
306 (1996), 231-245.
[T1] R. THOM, L'équivalence d'une fonction différentiable et d'un polynôme, Topology 3 (1965), Suppl. 2,
297-307.
[T2] R. THOM, Stabilité structurelle et morphogénèse, Benjamin et Édiscience, New York, 1972.
[W] B. WAJNRYB, Artin groups and geometric monodromy, Invent. Math. 138 (1999), 563-571.

MATHEMATISCHES INSTITUT
UNIVERSITÄT BASEL
RHEINSPRUNG 21, CH-4051 BASEL
SWITZERLAND