RELATIVE POLAR MULTIPLICITIES, THE REAL LINK, AND MORSE

INEQUALITIES FOR HYPERSURFACES

DAVID B. MASSEY
arXiv:2407.03313v1 [math.AG] 3 Jul 2024

Abstract. For a hypersurface defined by a complex analytic function, we derive Morse-type

inequalities between the Betti numbers of the real link of the hypersurface and the relative
polar multiplicities of the function.

1. Introduction
Let U be an open neighborhood of the origin in Cn+1 , where n ≥ 1, and let f : (U, 0) → (C, 0)
be a complex analytic function which is not locally constant at 0. We let Σf denote the critical
locus of f . To eliminate the trivial case, we assume that 0 ∈ Σf and let s := dim0 Σf . We shall
always use Z for our homology/cohomology coefficients.
We let X := V (f ) be the hypersurface defined by f . As in [11], we let KX denote the real
link of X at 0. We let LX denote the complex link of X at 0 (see [4]); this is the Milnor fiber of
the restriction of a generic linear form L to X. Lê proves in [4] that LX has the homotopy-type
of a bouquet of (n − 1)-spheres; the number
of (n − 1)-spheres in the bouquet is given by the
1
intersection number γf,L := Γ1f,L ·V (L) 0 , where Γ1f,L is the relative polar curve of f with respect


to L. In Corollaire IV.5.4.3 of [12], Teissier proves that for generic L, Γ1f,L ·V (L) 0 = mult0 Γ1f,L ,
which has a constant value for generic L; when L is this generic, we omit the L from the notation
and write simply Γ1f and γf1 .
From the Wang sequence of Lemma 8.4 of [11], the result of Kato and Matsumoto in [3] that
the Milnor fiber in (n − s − 1)-connected, and Alexander duality, we obtain the well-known fact
that H k (KX ) = 0 unless k = 0, k = 2n − 1, or n − 1 ≤ k ≤ n + s. In addition, it is also
well-known that the rank of H 2n−1 (KX ) is equal to the number of irreducible components of X
at the origin (as K is compact, this follows at once from Lemma 19.1.1 of [2]).
1
It is known, but perhaps not so well-known, that H e n−1 (KX ) injects into H e n−1 (LX ) ∼
= Zγf
(for s = 0, see Proposition 6.1.22 of [1]; for general s, see Corollary 4.10 of [8]).
In this paper, we recover this result in Corollary 3.3 and generalize it in Corollary 3.5; our
result is that there are Morse (or Morse-type) inequalities between the reduced Betti numbers,
b̃k (KX ), of KX and the relative polar multiplicities γf1 , . . . , γfn . The first of these inequalities
says precisely that b̃n−1 (KX ) ≤ γf1 .

2. Prior definitions and results

We continue with our notation from the introduction.
We need higher-dimensional relative polar varieties/cycles Γkf , where k is the dimension, as
developed by Lê, Hamm, and Teissier. We have written about these in many, many places; see,
for instance, [6] and Example 6.10 of [5]. We usually define them with respect to a linear choice

2010 Mathematics Subject Classification. 32S25, 32S15, 32S55.

1
2 DAVID B. MASSEY

of coordinates z := (z0 , . . . , zn ) and write Γkf,z for the k-dimensional relative polar cycle and γf,z
k
k k


(or γf,z (0)) for the intersection number Γf,z · V (z0 , . . . zk−1 ) 0 provided that the intersection is
k
proper. However, again, Corollaire IV.5.4.3 of [12] tells us that, for generic z, γf,z = mult0 Γkf,z
k
and is independent of the generic choice of z. Thus, we write simply γf for this common generic
value. Note that γf0 = 0, γfn+1 = 1, and γfn = −1 + mult0 f .

We combine a number of our previous results to obtain:

n−1
Proposition 2.1. Let λ0X := γf0 + γf1 = γf1 , λ1X := γf1 + γf2 , λ2X := γf2 + γf3 , . . . , λX :=
n−1 n n n n+1 n
γf + γf , and λX := γf + γf = γf + 1 = mult0 f . Then, there is a chain complex
n n−1 n−2 1 0
0 → ZλX → ZλX → ZλX → · · · → ZλX → ZλX → 0
such that the homology/cohomology at the λkX term is isomorphic to the stalk cohomology at the
origin in degree −k of Z•X [n]. Thus, the homology/cohomology at the λkX term is isomorphic to
Z when k = n and is zero when k 6= n.
Proof. The existence of the chain complex with the given stalk cohomology appears in Theorems
5.3 and 5.4 of [5], but with different shifts; for us now, we would use P• := Z•X [n] and include
shifts by −1 on the nearby and vanishing cycles. That the modules are actually free abelian
follows from the results in [7], but is more clear in [9]. Finally, the relation between the λkX and
the γfk is given in Example 8.4 of [5]. 

Remark 2.2. Whenever one has a chain complex where the terms in the complex have finite
rank, there are always Morse inequalities between the ranks of the terms in the complex and
the ranks of the cohomology of the complex. In a more-general context, this is presented in
Corollary 5.5 of [5]. However, the Morse inequalities which follow from Proposition 2.1 have
no real content; the reader should check that the Morse inequalities yield only that, for all k,
γfk ≥ 0.

3. Morse inequalities and the real link

The reader should be wondering why we bothered discussing Morse inequalities related to
Proposition 2.1 when we concluded that the related Morse inequalities told us essentially nothing.
The reason appears below, where we “dualize” the chain complex from Proposition 2.1 and look
at the associated Morse inequalities to yield non-trivial results about the Betti numbers of the
real link of X.
n−1
We continue to use λ0X := γf1 , λ1X := γf1 + γf2 , λ2X := γf2 + γf3 , . . . , λX := γfn−1 + γfn , and
λnX n
:= γf + 1 = mult0 f .
We let m0 : {0} ֒→ X be the inclusion, so that
H k (m! Z• [n]) ∼
0 X = H k+n (B ◦ ∩ X, B ◦ ∩ X\{0}) ∼
ǫ ǫ =H e n+k−1 (KX ),
where Bǫ◦ is a small open ball, centered at the origin.

Theorem 3.1. There is a chain complex

0 1 2 n−1 n
0 → ZλX → ZλX → ZλX → · · · → ZλX → ZλX → 0
such that the homology/cohomology at the λkX term is isomorphic to the costalk cohomology at
the origin in degree k of Z•X [n].
 RELATIVE POLAR MULTIPLICITIES AND THE REAL LINK 3

To be precise, the homology/cohomology at the λkX term is isomorphic to H k (m!0 Z•X [n]) ∼
=
e n+k−1
H (KX ).
Proof. As was shown just before Theorem 5.4 of [5], the complex in Proposition 2.1 is obtained
by starting with Z•X [n] and then applying iterated nearby and vanishing cycles, repeatedly using
the canonical map from ψzk [−1] to φzk [−1]. One obtains the “dual” complex above by instead
repeatedly using the variation map from φzk [−1] to ψzk [−1].

Alternatively one could begin with the Verdier dual D(Z•X [n] obtain the complex analogous
to Proposition 2.1 by repeatedly using the canonical map from ψzk [−1] to φzk [−1], then Verdier
dualizing, and finally using that Verdier dualizing commutes with ψzk [−1] and φzk [−1] (see
[10]). 

Remark 3.2. Suppose that n = 1; this special case is easy to analyze. Theorem 3.1 implies
that we have an exact sequence
0→H e 0 (KX ) → Z−1+mult0 f → Zmult0 f → H
e 1 (KX ) → 0.
From this we conclude two obvious things: the reduced Euler characteristic of KX is −1, i.e.,
the Euler characteristic is 0; and the number of connected components of KX is at most mult0 f .

The following corollary is immediate fromn the theorem.

1
Corollary 3.3. H e n−1 (LX ) ∼
e n−1 (KX ) injects into H = Zγf .

Now we want to look at the Morse inequalities derived from Theorem 3.1 and obtain highly
non-trivial results.
First, we need a trivial lemma, whose proof we leave to the reader:
Lemma 3.4. For all p such that 0 ≤ p ≤ n,
Pp k k p p+1
(1) k=0 (−1) λX = (−1) γf , and
Pp
(2) k n−k
k=0 (−1) λX = 1 + (−1)p γfn−p .

(Recall that γf0 = 0 and γfn+1 = 1.)

Now, from the theorem and the above lemma, we immediately conclude:
e k (KX ). Then, for all p
Corollary 3.5. (Morse Link Inequalities) For all k, let b̃k := rank H
such that 0 ≤ p ≤ n, we have the following inequalities:
(1)
p
X p
X
(−1)p (−1)k b̃n+k−1 ≤ (−1)p (−1)k λkX = γfp+1 and
k=0 k=0

(2)
p
X p
X
(−1)p (−1)k b̃2n−k−1 ≤ (−1)p n−k
(−1)k λX = (−1)p + γfn−p .
k=0 k=0

One can also conclude from Theorem 3.1 and Lemma 3.4 that the Euler characteristic of KX is
0, but this is well-known.
4 DAVID B. MASSEY

Example 3.6. Let c denote the number of irreducible components of X at 0; we remind the
reader that this equals the rank of H 2n−1 (KX ). Note that, if c 6= 1, then s = dim0 Σf = n − 1.
Now suppose that we take p = 0 in Corollary 3.5. Then we obtain two inequalities:
b̃n−1 ≤ γf1 and b̃2n−1 ≤ 1 + γfn .
This first inequality, which we have already discussed, is not obvious. However, the second
inequality simply says that c ≤ mult0 f , which is stupidly true.
What if p = 1? Then we obtain two inequalities:
−(b̃n−1 − b̃n ) ≤ γf2 and − (b̃2n−1 − b̃2n−2 ) ≤ −1 + γfn−1 .
Again the first inequality is not obvious and, as far as we know, is a new bound. Is the second
inequality obvious? That depends on the size of s. We remind the reader H k (KX ) = 0 unless
k = 0, k = 2n − 1, or n − 1 ≤ k ≤ n + s. So, if n + s + 1 ≤ 2n − 2, i.e., if s ≤ n − 3 (and hence
c = 1), then b̃2n−2 = 0 and the second inequality above becomes −b̃2n−1 ≤ −1 + γfn−1 , that
is, 1 = c = b̃2n−1 ≥ 1 − γfn−1 , which is trivially true. On the other hand, if b̃2n−2 6= 0, which
requires that s ≥ n − 2, then the second inequality above is new as far as we know.

References
[1] Dimca, A. Sheaves in Topology. Universitext. Springer-Verlag, 2004.
[2] Fulton, W. Intersection Theory, volume 2 of Ergeb. Math. Springer-Verlag, 1984.
[3] Kato, M., Matsumoto, Y. On the connectivity of the Milnor fibre of a holomorphic function at a critical
point. Proc. of 1973 Tokyo manifolds conf., pages 131–136, 1973.
[4] Lê, D. T. Sur les cycles évanouissants des espaces analytiques. C. R. Acad. Sci. Paris, Sér. A-B, 288:A283–
A285, 1979.
[5] Massey, D. Numerical Invariants of Perverse Sheaves. Duke Math. J., 73(2):307–370, 1994.
[6] Massey, D. Lê Cycles and Hypersurface Singularities, volume 1615 of Lecture Notes in Math. Springer-Verlag,
1995.
[7] Massey, D. Singularities and Enriched Cycles. Pacific J. Math., 215, no. 1:35–84, 2004.
[8] Massey, D. Milnor fibers and Links of Local Complete Intersections. International J. of Math., 25:18 pages,
2014.
[9] Massey, D. Calculations with Characteristic Cycles. https://doi.org/10.48550/arXiv.1101.4304, 2016.
[10] Massey, D. Natural Commuting of Vanishing Cycles and the Verdier Dual. Pacific J. of Math, 284-2:431–437,
2016.
[11] Milnor, J. Singular Points of Complex Hypersurfaces, volume 77 of Annals of Math. Studies. Princeton Univ.
Press, 1968.
[12] Teissier, B. Variétés polaires II: Multiplicités polaires, sections planes, et conditions de Whitney, Proc. of
the Conf. on Algebraic Geometry, La Rabida 1981. Springer Lect. Notes, 961:314–491, 1982.