Persistent Transcendental Bézout Theorems
Persistent Transcendental Bézout Theorems
Persistent Transcendental Bézout Theorems
where B r stands for the closed ball of radius r. This quantity has at least
two degree-like features. First, assume that
log µ( f , r)
lim sup < k + 1.
r→∞ log r
Then, remarkably, f is a polynomial of the total degree ≤ k. This is
a minor generalization of Liouville’s classical theorem. Thus one can
distinguish polynomials in terms of the maximum modulus.
The second feature of the maximum modulus is given by a statement
which readily follows from Jensen’s formula. Let ζ( f , r) be the number of
zeros of an entire function f : C → C inside the ball B r . Then, provided
f (0) ̸= 0, one has for every a > 1
(1) ζ( f , r) ≤ C log µ( f , ar) ∀r > 0 ,
where C is a positive constant depending on a and f (0). For instance,
in Example 1.1 both ζ and log µ grow linearly in r.
These two features might have given a hope that log µ(r) is an ap-
propriate substitute of the degree for an entire map f : Cn → Cn (this
was known as the transcendental Bézout problem). However, this anal-
ogy was overturned by Cornalba and Shiffman [11] who famously con-
structed, for n = 2, an entire map f with log µ( f , r) ≤ Cε r ε for every
ε > 0 (and hence of growth order zero), with ζ( f , r) growing arbitrarily
fast. As Griffiths wrote in [15] “This is the first instance known to this
author when the analogue of a general result in algebraic geometry fails to
hold in analytic geometry."
1.2. Coarse zero count. One of the motivations for the present paper
is to further explore the Cornalba–Shiffman example using the notion of
coarse zero count introduced in [7], which is based on topological persis-
tence. The idea, roughly speaking, is to discard the zeros corresponding
to small oscillations of the map. It turns out that with such a count we
are able to get a Jensen-type estimate (1), albeit with a somewhat worse
power of log µ(r) in the right-hand side, see (2) below.
2
Given an analytic map f : Cn → Cn and positive numbers δ, r > 0, we
define the counting function ζ( f , r, δ) of δ-coarse zeros of f inside a ball
B r as the number of connected components of the set f −1 (Bδ )∩ B r which
contain zeros of f , see Figure 1.
Br
ζ(f, r) = 7
ζ(f, r, δ) = 3
Theorem 1.2. For any analytic map f : Cn → Cn and any a > 1, r > 0,
µ( f ,ar)
and δ ∈ (0, e ), we have
µ( f , ar) 2n−1
(2) ζ( f , r, δ) ≤ C log ,
δ
where the constant C depends only on a and n.
Note that by Liouville’s theorem, unless f is constant, µ( f , ar) is un-
bounded. Therefore, for any given δ > 0, the condition δ ∈ (0, µ( f , ar)/e)
holds for all r large enough.
Remark 1.3. Consider a higher-dimensional generalization of Example
1.1: take an analytic map f : Cn → Cn given by
f (z1 , . . . , zn ) = (ez1 + 1, . . . , ezn + 1) .
It is easy to see that log µ( f , r) grows linearly in r and ζ( f , r, δ) grows
as r n when r → ∞, for δ sufficiently small. It would be interesting to
3
understand whether the power of the logarithm in (2) is sharp or it can
be improved, possibly, to n.
It follows from Theorem 1.2 that for the Cornalba-Shiffman example
the coarse count of zeros grows slower than any positive power of r, see
Theorem 1.5 below for precise asymptotics.
Remark 1.4. Consider a function f of growth order ≤ ρ, that is, for all
ε > 0, there exist positive constants Aε , B, such that
ρ+ε
| f (z)| ≤ Aε e B|z|
everywhere. Then by (2), ζ( f , r, δ) grows slower than r (2n−1)ρ+ε for every
ε > 0. At the same time, it was shown in [8, equation (1.9)] that for any
α > 0, ζ( f +c, r) grows slower than r (2n−1)ρ+1+α for almost all c > 0 small
enough. While the growth rate in (2) is slightly sharper, it is interesting
to note that the power (2n − 1)ρ appears in both bounds.
1.3. Cornalba–Shiffman example: a coarse perspective. Let us re-
mind the Cornalba–Shiffman construction. Let g : C → C be given by
∞
Y z
g(z) = 1− i .
i=1
2
For k ≥ 1 an integer, let
g(z)
g k (z) =
1 − 2zk
be the function defined by the same product with k-th term excluded.
All the infinite products converge uniformly on compact subsets of C and
hence g and g k are holomorphic by Weierstrass’ theorem. For a positive
integer c we define a polynomial Pc : C → C as
c
1
Y
Pc (w) = w− .
j=1
j
Given a strictly increasing sequence of positive integers c = {ci }, c1 <
c2 < . . . define f : C2 → C as
∞
X 2
f (z, w) = 2−ci g i (z)Pci (w).
i=1
1
1
2
1
3
1
4
2 4 8 z
that the set {|F | ≤ δ}, while possibly being complicated for small ra-
dius r, stabilizes for large radii and can be described rather accurately.
More precisely, we show that there exists k0 , which depends only on δ,
such that {|F | ≤ δ} contains intervals {2k }×[0, 1] for all k ≥ k0 . Thus, for
k ≥ k0 the zeros on each of the intervals {2k }×[0, 1] are counted coarsely
as one zero and the coarse count increases at the rate log r. This implies
ζ(F, r, δ) = O(log r). Furthermore, for k ≥ k0 , {|F | ≤ δ} will never in-
tersect hyperplanes H k = {(w, z) | Re(z) = 2k + 2k−1 }. In other words,
{|F | ≤ δ} consists of parts contained between those hyperplanes and
which contain intervals {2k } × [0, 1], as shown on Figure 3 (shaded re-
gions represent the set {|F | ≤ δ}). This implies that ζ(F, r, δ) = Θ(log r),
which we can improve to ζ(F, r, δ) ∼ log r as claimed by Theorem 1.5.
Putting together Theorems 1.2 and 1.5, we come to the following con-
clusion. It follows from Theorem 1.2 for n = 2 that
On the logarithmic scale, this inequality tells us that for fixed a and δ,
n
(5) τ( f , r, δ) ≤ C1 log(µ( f , ar)/δ) ,
where C1 depends only on a and n.
Note that in view of Remark 1.3, estimate (5) is sharp.
Remark 1.7. Estimates analogous to (5) for the usual count of zeros have
been proven under positive lower bounds on the Jacobian of f in [18,
19, 20]. Upper bounds in Theorems 1.2 and 1.6 apply to all holomorphic
mappings f : Cn → Cn . A detailed comparison of our results with those
in [18, 19, 20] are carried out in Section 7. In particular, we give a
different proof of a result from [20] using Theorem 1.6.
The proof of Theorem 1.6 does not use persistence techniques and
does not extend to an estimate on ζ( f , r, δ), as it does not control penin-
sulas. In general, ζ( f , r, δ) and ζ0 ( f , r, δ) can behave rather differently.
Indeed, this is the case for the Cornalba–Shiffman example. as we dis-
cuss below.
Let F be a Cornalba-Shiffman map defined previously. It is natural to
ask what is the possible growth of the coarse count of islands ζ0 (F, r, δ).
We show that, as opposed to ζ(F, r, δ), ζ0 (F, r, δ) can grow arbitrarily
7
slow, with an upper bound depending on c. More precisely, we prove the
following theorem.
Theorem 1.8. Let λ, δ > 0, l ≥ 1 an integer and denote by exp2 (x) = 2 x .
If ci = ⌊exp2 . . . exp2 (λi)⌋ then there exists a constant ml,λ,δ such for all
| {z }
l times
r ≥ exp2 . . . exp2 (1) it holds
| {z }
l+1 times
1
ζ0 (F, r, δ) ≤ log . . . log r + ml,λ,δ .
λ | 2 {z }2
l+1 times
2i
In particular, for ci = 2 as in [11], we have that ζ0 (F, r, δ) = O(log2 log2 log2 r).
From the geometric perspective, the slow growth of ζ0 (F, r, δ) is due to
elongation of {|F | ≤ δ} in the w-direction. Namely, as r increases, new
groups of zeros of roughly the same modulus r appear, while the com-
ponents of {|F | ≤ δ} which contain these zeros grow in the w-direction
faster than r (the diameter of their w-projection grows faster than r).
Hence, it takes larger r for a component of {|F | ≤ δ} to be fully con-
tained in B r i.e. to contribute to ζ0 (F, r, δ).
1.5. Discussion. Below we discuss some extensions of our results as
well as directions for further research.
1.5.1. Analytic mappings from Cn to Ck . We expect that a bound analo-
gous to Theorem 1.2 should hold for entire mappings f : Cn → Ck . How-
ever, to have geometric meaning in this case, the definition of ζ( f , r, δ)
should be generalized. There are two essentially dual directions of do-
ing so. First, we can look at coarse homology groups of the zero set: for
0 ≤ d ≤ 2n − 1, set
ζd ( f , r, δ) = dim Im H d ({ f = 0} ∩ B r ) → H d ({| f | ≤ δ} ∩ B r ) .
Considering generic algebraic maps f , we expect only 0 ≤ d ≤ n − k to
have geometric significance. Of particular interest is d = n − k, since this
is the dimension where vanishing cycles appear. We expect the upper
bound 2n
ζd ( f , r, δ) ≤ C log(µar ( f )/δ) .
Second, we may set
md ( f , r, δ) = dim Im H d ({| f | > δ} ∩ B r ) → H d ({| f | > 0} ∩ B r ) .
We expect the upper bound
2n
md ( f , r, δ) ≤ C log(µar ( f )/δ)
for 0 ≤ d ≤ 2n − 1.
8
1.5.2. Affine varieties. It would be interesting to generalize our main re-
sults to more general affine algebraic varieties Y ⊂ CN . The starting
case would be varieties which compactify to smooth projective varieties
X ⊂ CP N by a normal crossings divisor D = X \ Y. We expect that the
methods of [10] combined with those of [7], in particular the subadditiv-
ity theorem for persistence barcodes, should be useful for this purpose.
1.5.3. Harmonic mappings. A result analogous to Theorem 1.2 but in the
context of harmonic maps can be proven in a similar way. Namely, let
h = (h1 , . . . , hd ) : Rd → Rd be a harmonic map in the sense that h j : Rd →
R is harmonic for all j. In this case, the coarse counts ζ(h, r, δ), ζ0 (h, r, δ)
being defined analogously to the above, should satisfy the upper bound
d
(6) ζ0 (h, r, δ) ≤ ζ(h, r, δ) ≤ C2 log(µar (h)/δ) ,
for all a > C1 , r > 0, and δ ∈ (0, µar (h)/e), where C1 depends on d
only, and C2 depends on a and d, but not on r or δ. (By Liouville’s
theorem the condition on δ holds for all r large enough, if our mapping
is not constant.) This bound is sharp asymptotically in r for all fixed
δ > 0, as can be easily seen from the example h = (h1 , . . . , hd ) where
hi (x 1 , . . . , x d ) = (e x i+1 sin(x i )) for 1 ≤ i ≤ d − 1 and hd (x 1 , . . . , x d ) =
(e x 1 sin(x d )). Note that log(µar (h)) is closely related to the notion of the
doubling index of the harmonic function (see for example [21, Equation
(12)]).
SInce our proof of inequality (6) is based on Cauchy estimates, as in
Proposition 4.1, and bounds on barcodes of polynomials as in Proposition
3.1 (see also [7, Proposition 4.12]), it is not hard to extend (6) to a
certain quasi-analytic class of functions.
1.5.4. Near-holomorphic mappings. Let us also note that Theorem 1.2
implies the following result about the count of islands for continuous
functions that are close to holomorphic ones.
Corollary 1.9. Fix b < 1 and δ > 0, and let h : Cn → Cn be a continuous
function such that there exists a holomorphic function f : Cn → Cn with
dC 0 (h, f ) < 2b δ. Then
2n−1
ζ(h, r, (1 + b)δ) ≤ C log(µ(h, ar)/δ) ,
where C depends on a, b, n only.
1.5.5. A dynamical interlude. A dynamical counterpart of the transcen-
dental Bézout problem is the count of periodic orbits of entire maps
f : Cn → Cn . Here by a k-periodic orbit we mean a fixed point of the
iteration f ◦k = f ◦ · · · ◦ f (k times). There exists a vast literature on the
orbit growth of algebraic maps f (see e.g. [2]). For instance, it follows
9
from the Bézout theorem that if the components of f are generic poly-
nomials of degree ≤ d, the number of k-periodic orbits does not exceed
d kn . Can one expect a bound on the number of k-periodic orbits in the
ball of radius r in terms of the maximum modulus function µ( f , r)? The
naive answer is “no" due to the the Cornalba-Shiffman example. Nev-
ertheless, Theorem 1.2 above readily yields such a bound on the coarse
count ζ( f k , r, δ), where
f k (z) := f ◦k (z) − z .
One can check that the maximum modulus function behaves nicely under
the composition and the sum:
µ( f ◦ g, r) ≤ µ( f , µ(g, r)), µ( f + g, r) ≤ µ( f , r) + µ(g, r) .
Fix a > 1 and δ > 0, set µ̃(r) := µ( f , ar), and put
µk (r) = µ̃◦k (r) + r .
By Theorem 1.2 we have the desired estimate
µk (r) 2n−1
(7) ζ( f k , r, δ) ≤ C max log ,1 .
δ
A few questions are in order.
Question 1.10. Can one find a transcendental entire map f for which
estimate (7) is sharp?
A natural playground for testing this question are transcendental Hénon
maps whose entropy as restricted to a family of concentric discs grows
arbitrarily fast [1].
Further, recall that a k-periodic orbit of an entire map f : Cn → Cn is
called primitive if it is not m-periodic with m < k. Denote by νk ( f , r) the
number of primitive periodic orbits lying in the ball of radius r.
Question 1.11. Does there exist a transcendental entire map f of order 0
ε
(i.e., the modulus µ( f , r) grows slower than e r for every ε > 0) such that
νk ( f , r) grows arbitrarily fast in k and r?
For instance, taking f (z) = h(z) + z, where h is the Cornalba-Shiffman
map, we see that ν1 ( f , r) can grow arbitrarily fast. Can one generalize
this construction to k ≥ 2?
Finally, let us mention that the failure of the transcendental Bézout
theorem appears as one of the substantial difficulties in the work [16]
dealing with a dynamical problem of a completely different nature, namely
with embeddings of Zk -actions into the shift action on the infinite dimen-
sional cube (see (2) on p. 1450 in [16]). In particular, the authors ana-
lyze the structure of zeroes of so-called tiling-like band-limited maps (see
10
p.1477 and Lemma 5.9). It would be interesting to perform our coarse
count of zeroes (i.e., to calculate ζ) for this class of examples.
ACKNOWLEDGEMENTS
We thank Misha Sodin and Steve Oudot for illuminating discussions.
Therefore
r k ρ
|rk (u)| ≤ µρ (g)
ρ ρ−r
and picking ρ = ar, we get
|rk (u)| ≤ Ca a−k µar (g) ≤ Ca a−k µar ( f ).
So taking maxima over u, v and v ′ , we obtain
µ r (R k ) ≤ Ca a−k µar ( f ).
□
Proof of Theorem 1.2. Let us now suppose that f = ( f1 , . . . , f n ) : Cn → Cn
is a complex analytic self-mapping. Let h = | f |. Proposition 4.1 implies
that on the ball B r of radius r, there exists a polynomial mapping pk :
Cn → Cn of degree < k such that
dC 0 ( f , pk ) < Ca a−k µar ( f ).
In particular for k = ⌈loga (µar ( f )/δ)⌉ = ⌈log(µar ( f )/δ)/ log(a)⌉, which
makes sense as a positive integer since a > 1,
dC 0 ( f , pk ) < Ca δ,
and therefore
dC 0 (h, |pk |) < Ca δ.
Hence by Theorem 2.1 and Proposition 3.1 from Section 3
2n−1
N0,Ca′ δ (h) ≤ N0,ε (|pk |) ≤ Cn k2n−1 ≤ Cn,a log(µar ( f )/δ) ,
for Ca′ = 2Ca , Cn,a = Cn ((log(a))−1 + 1)2n−1 and suitable small ε > 0. In
turn by a change of variable
2n−1
N0,δ (h) ≤ Cn,a
′
log(µar ( f )/δ) ,
14
′
for Cn,a = (log(Ca′ ) + 1)2n−1 Cn,a . We used the condition µar ( f )/δ > e in
order to absorb additive terms into suitable multiplicative terms in both
inequalities.
Now, ζ(r, f , δ) ≤ N0,δ (h) by definition of the persistence module of h.
□
The goal of this section is to prove Theorem 1.5. We will break down
its proof into Propositions 5.1, 5.4 and 5.6. Since there are no zeros of
F when r < 2 we always assume r ≥ 2. Firstly, we estimate µ r (F ) as
needed for the first part of Theorem 1.5.
Proposition 5.1. For all c and all r ≥ 2 it holds
1 3 3 7
(log2 r)2 − log2 r + 1 ≤ log2 µ r (F ) ≤ (log2 r)2 + log2 r + C,
2 2 2 2
Q∞
where C = 4 + log2 i=1
(1 + 2 ) .
−i
Proof. Let k ≥ 1 be an integer such that 2k ≤ r < 2k+1 . To prove the first
inequality it is enough to take (z, w) = (−r, 0). Now
∞
Y
|F (−r, 0)| ≥ |g(−r)| ≥ |g(−2 )| = k
(1 + 2k−i ),
i=1
and
∞
Y Yk k−1
Y r k−1
k(k−1) 2
(1 + 2 ) >
k−i
(1 + 2 ) =
k−i
(1 + 2 ) > 2
j 2 > .
i=1 i=1 j=0
2
log r−2
Since log2 r < k + 1 we have that k−1 2 >
2
2 and hence |F (−r, 0)| ≥
log 2 r
r
2 −1
2 . Applying logarithm to both sides proves the first inequality.
To prove the second inequality we firstly notice that if |z| ≤ r then
∞
Y ∞
Y k
Y ∞
Y
|g(z)| ≤ (1 + 2−i |z|) < (1 + 2k+1−i ) = (1 + 2 j ) · (1 + 2−i ).
i=1 i=1 j=0 i=1
Q∞
Denoting C1 = i=1
(1 + 2−i ) and estimating
k k+1
(k+1)(k+2) log2 r
Y Y k+2
(1 + 2 ) ≤
j
2j = 2 2 ≤ (2r) 2 ≤ (2r) 2 +1 ,
j=0 j=1
yields
log2 r
(8) |g(z)| < C1 (2r) 2 +1 .
15
Similarly,
log2 r
|g i (z)| < C1 (2r) 2 +1 ,
for all i ≥ 1. Using this inequality we further estimate that if |(z, w)| ≤ r
then
∞
X |g i (z)| · |Pci (w)| log2 r
+1
∞
X |Pci (w)|
| f (z, w)| ≤ 2
≤ C1 (2r) 2
2
.
i=1 2 ci i=1 2 ci
On the other hand
Q ci
X∞
|Pci (w)| X∞
j=1
|(w − 1/ j)| X∞ ∞
(r + 1)ci X (r + 1)i
2
= 2
< 2
≤ .
i=1 2 ci i=1 2 ci i=1 2 ci i=1
2i 2
To bound the last term we proceed as follows
∞
(r + 1)i r +1 i X r + 1 i
X X
i2
= i
+ i
<
i=1
2 1≤i≤log (r+1)
2 i>log (r+1)
2
2 2
∞
X X 1
< (r + 1)i + < (r + 2)log2 (r+1) + 2 < (2r)log2 2r + 2.
1≤i≤log2 (r+1) j=0
2j
Putting all the inequalities together, we obtain
log2 r
(9) | f (z, w)| < C1 (2r) 2 +1 ((2r)log2 2r + 2).
Since |F (z, w)| = |g(z)|2 + | f (z, w)|2 , combining (8) and (9) proves
p
(i−1)i 2 (i−1)i 1
c−
Lemma 5.2. For each δ ≥ 2 2 −ci the whole interval {2i }×[0, 2 i 2ci
δ ci ]
is contained in {|F | ≤ δ}.
Proof. Firstly we notice that
i−1 ∞ i−1
(i−1)i
Y Y Y
|g i (2i )| = (2i− j − 1) · (1 − 2i− j ) < 2i− j = 2 2 .
j=1 j=i+1 j=1
Secondly
2 (i−1)i 2
|F (2i , w)| = | f (2i , w)| = 2−ci |g i (2i )||Pci (w)| < 2 2 −ci |Pci (w)|.
Now, if w ∈ [0, 1], |Pci (w)| < 1 and the claim follows by the assump-
(i−1)i 1
c−
tion on δ. If w ∈ (1, 2 i 2ci
δ ci ], |Pci (w)| < wci and thus |F (2i , w)| <
(i−1)i 2
2 2 −ci wci ≤ δ and the claim follows. □
−i(i+1)
Corollary 5.3. If δ ≥ 2 2 then the whole interval {2i } × [0, 1] is con-
tained in {|F | ≤ δ}.
−i(i+1) (i−1)i 2
Proof. Since ci2 ≥ i 2 , δ ≥ 2 2 implies that δ ≥ 2 2 −ci and thus
(i−1)i 1
c−
{2i } × [0, 1] ⊂ {2i } × [0, 2 i 2ci
δ ci ] ⊂ {|F | ≤ δ}
by Lemma 5.2. □
17
Proposition 5.4. The following estimates hold for r > 2 :
P − log2 r(log2 r+1)
ci , if 0 < δ < 2 2
1≤i≤log2 r
− log2 r(log2 r+1)
1
ζ(F, r, δ) ≤ log2 r + 2 − −2 log2 δ + ≤δ<
p P
ci , if 2 2 .
p 2
δ
1≤i< −2 log2
log2 r, if δ ≥ 12
k−1 k ∞
2 + 2k−1 2k + 2k−1
Y 1 Y (k−1)k
= −1 · · 1− > C0 2 2 .
i=1
2i 2 i=k+1 2i
□
Proposition 5.6. For all r ≥ 2 it holds
¨
⌊log2 r⌋ − 1, if δ ≤ C0
ζ(F, r, δ) ≥ log2 r−1 .
⌊log2 r⌋ − 2 log2 δ − 2 log2 C0 − 2, if C0 < δ ≤ C0 r 2
p
Proof. We first prove the case δ ≤ C0 . By Lemma 5.5, we see that on each
hyperplane {Re(z) = 2i + 2i−1 }, i ≥ 1 it holds |F (z, w)| > C0 ≥ δ. Hence
{|F | ≤ δ} does not intersect any of these hyperplanes and in particular
zeros (2i , 1), i ≥ 1 all belong to different connected components of {|F | ≤
δ}. In other words
ζ(F, r, δ) ≥ the number of points (2i , 1), i ≥ 1 in B r ≥ ⌊log2 r⌋ − 1,
which finishes the proof of the first case.
To prove the second case, we firstly denote by r0 > 2 the unique real
log2 r0 −1
6. COUNTING ISLANDS
Let us start with the proof of Theorem 1.6 which, as was mentioned
in the introduction, is a an easy corollary of Rouché’s theorem.
20
Proof of Theorem 1.6. By Rouché’s theorem for analytic mappings from
Cn to Cn (see e.g. [12, Section 2.1.3]) if g : Cn → Cn is a polynomial
mapping of degree at most k such that dC 0 ( f |Br , g|Br ) < δ/2, then
τ( f , r, δ) ≤ τ(g, r, δ/2).
By Bézout’s theorem, however,
τ(g, r, δ/2) ≤ k n .
By Proposition 4.1, it is enough to take k such that Ca a−k µar ( f ) < δ/2.
It is easy to see that the optimal such k satisfies
k ≤ Ca′ log(µar ( f )/δ).
Combining the three displayed inequalities finishes the proof. □
where
I = the number of islands with zeros from {2i } × [0, 1], i > log2 bi0 .
21
w
1
z
FIGURE 6. Components count for ζ(F, r, δ), but not for ζ0 (F, r, δ)
Since the first term of the right-hand side of this inequality depends only
on δ and c, we wish to estimate I. If r ≤ bi0 then 2i > r and I = 0. Thus,
we assume that r > bi0 . Firstly, from 2) it follows that
(12) I ≤ the number of islands with zeros from {2i } × [0, 1], i > i0 .
The goal of this section is to compare the results of this paper to the
results of [20]. More precisely, we will deduce Theorem 5.1 in [20] from
Theorem 1.6, as well as show that Theorem 1.2 does not follow from
Theorem 5.1 in [20]. We start by recaling this result.
For an entire map f : Cn → Cn let J f denote the complex Jacobian
matrix. Given a sequence of zeros ξ = {ξi } ⊂ f −1 (0) we define ζξ ( f , r)
to be the number of elements of ξ inside a ball B r .
Theorem 7.1 (Theorem 5.1 in [20]). Let f : Cn → Cn be an entire map
and ξ = {ξi } a sequence of zeros of f . If there exist real numbers c > 0 and
b such that
(∀i) | det J f (ξi )| ≥ c(µ( f , |ξi |))−b ,
then for any a > 1 it holds
ζξ ( f , r) = O((log µ( f , ar))n ),
when r → ∞.
Firstly, we give a proof of Theorem 7.1 using Theorem 1.6. The strat-
egy of the proof follows [18]. Namely, the main results of [18], Theo-
rems 1.1 and 1.2, are proven using a lemma which was referred to as
“Weak Bézout estimate", see [18, Lemma 3.1]. This lemma establishes
an inequality
µ( f , r + 1) 2n
(14) τ( f , r, δ) ≤ Cn (r + 1) ,
δ
with Cn which depends only on n. The proof of (14) relies on a global
version of the Chern-Levine-Nirenberg inequality, see [18, Theorem 2.1]
and references therein. Substituting (14) with Theorem 1.6 and using
23
the same general arguments as in [18] proves Theorem 7.1. To imple-
ment this strategy, we will need the following lemma.
Lemma 7.2. Let f : Cn → Cn be an entire map and ξ a zero of f such that
J f (ξ) ̸= 0. Then, for all z ∈ Cn such that
1
|z| ≤
(µ( f ,|ξ|+1))n
2 n! | det J f (ξ)| +1
it holds
| det J f (ξ)| · |z|
| f (ξ + z)| ≥ .
2n!(µ( f , |ξ| + 1))n−1
The proof of Lemma 7.2 can be extracted from the proof of Theorem
1.1 in [18]. We present it here for the sake of completeness. We will
use the following auxiliary statement, which is a direct consequence of
Schwarz lemma, see [18, Lemma 3.4] for details.
Lemma 7.3. Let f : Cn → Cn be an entire map such that f (0) = 0, J f (0) =
0. Then for all z such that |z| ≤ 2µ(1f ,1) it holds | f (z)| ≤ 12 |z|.
Proof of Lemma 7.2. We start by proving the following auxiliary inequal-
ity
(µ( f , |ξ| + 1))n−1
(15) ∥J f (ξ)−1 ∥op ≤ n! ,
| det J f (ξ)|
where ∥ · ∥op denotes the operator norm. Indeed,
1
∥J f (ξ)−1 ∥op = ∥ adj(J f (ξ))∥op ,
| det J f (ξ)|
and thus we need to prove that
∥ adj(J f (ξ))∥op ≤ n!(µ( f , |ξ| + 1))n−1 .
By Cauchy-Schwarz inequality
∥ adj(J f (ξ))∥op ≤ n · max | adj(J f (ξ))i, j |,
i, j
For each i and j, adj(J f (ξ))i, j is a sum of (n − 1)! terms, each of which is
a product of n − 1 partial derivatives of f . Thus
max | adj(J f (ξ))i, j | ≤ (n − 1)! · (max |∂i f (ξ))|)n−1 .
i, j i
Finally, Cauchy’s inequality yields that maxi |∂i f (ξ))| ≤ µ( f , |ξ|+1) which
completes the proof of (15).
24
Now, let g : Cn → Cn be an entire map given by g(z) = (J f (ξ))−1 f (ξ+
z). Since g(0) = 0 and J g (0) = idCn we may apply Lemma 7.3 to the map
g(z) − z, which gives us
1
(16) |g(z) − z| ≤ |z|,
2
1
for all z, such that |z| ≤ 2µ(g(z)−z,1) . Moreover, µ(g(z) − z, 1) ≤ µ(g, 1) + 1
1
implies that (16) holds for all z with |z| ≤ 2(µ(g,1)+1) and triangle inequal-
ity further implies that
1
(17) |g(z)| ≥ |z|
2
1
as long as |z| ≤ 2(µ(g,1)+1) . From the definition of g and (17) it follows
that
1
(18) | f (ξ + z)| · ∥(J f (ξ))−1 ∥op ≥ |z|,
2
1
for all z such that |z| ≤ 2(µ(g,1)+1) . Applying (15) to (18) yields
(µ( f , |ξ| + 1))n−1 1
(19) | f (ξ + z)| · n! ≥ |z|,
| det J f (ξ)| 2
1
for all z such that |z| ≤ 2(µ(g,1)+1) . Using (15) again gives us
(µ( f , |ξ| + 1))n−1
|g(z)| ≤ | f (ξ + z)| · ∥(J f (ξ)) ∥op ≤ | f (ξ + z)| · n!
−1
,
| det J f (ξ)|
and thus
(µ( f , |ξ| + 1))n
µ(g, 1) ≤ n! .
| det J f (ξ)|
Combining the last inequality and (19) finishes the proof. □
Proof of Theorem 7.1. For simplicity we assume that b ≥ 0 and f (0) ̸= 0.
The general case readily reduces to this one.
Let r ≥ 0 and ξi ∈ B r . By the assumption
| det J f (ξi )| ≥ c(µ( f , |ξi |))−b ≥ c(µ( f , r))−b ≥ c(µ( f , r + 1))−b ,
and since µ( f , |ξi | + 1) ≤ µ( f , r + 1), Lemma 7.2 gives us that
c|z|
(20) | f (ξi + z)| ≥ δ := ,
2n!(µ( f , r + 1))n−1+b
for all z such that
1
|z| ≤ ϵ := .
2(c −1 · n!(µ( f , r + 1))n+b + 1)
25
Since ϵ < 12 < 1 a connected component of {| f | ≤ δ} which contains ξi
is itself fully contained in B r+1 . The same holds for each ξ j ∈ B r and thus
ζξ ( f , r) ≤ τ( f , r + 1, δ).
Finally, by (20), δ/e ≤ µ( f , a(r + 1))/e for all a > 1 and we may apply
Theorem 1.6 which yields
e · µ( f , a(r + 1)) n
ζξ ( f , r) ≤ τ( f , r + 1, δ) ≤ Cn,a log .
δ
Substituting the expression for δ into this inequality and using µ( f , r +
1) ≤ µ( f , a(r + 1)) gives us
n
2en! · (µ( f , a(r + 1)))n+b
ζξ ( f , r) ≤ Cn,a log
c|z|
for all z such that |z| ≤ ϵ. Taking |z| = ϵ yields
n
ζξ ( f , r) ≤ Cn,a An,b log µ( f , a(r + 1)) + Bn,c ,
where A, B, C depend on n, a, b, c as indicated. This completes the proof.
□
So far we proved that Theorem 7.1 follows from Theorem 1.6. Now,
we wish to show that Theorem 1.2 does not follow from Theorem 7.1.
Namely, one may imagine the following scenario. Let f : Cn → Cn be an
entire map. While the classical count of zeros of f might not satisfy the
transcendental Bézout bound, it may still happen that for a fixed δ > 0,
we can choose a sequence of zeros {ξi } from f −1 (Bδ ) such that Theorem
7.1 applies to this sequence and ζ( f , r, δ) = O(ζξ ( f , r)). In this case,
Theorem 7.1 would imply Theorem 1.2 (at least up to a constant which
depends on δ). However, our next result rules out this possibility.
Proposition 7.4. Let c = {ci } be an increasing sequence of positive integers
such that limi→+∞ cii = 0 and F the corresponding Cornalba-Shiffman map.
For all real numbers c > 0 and b the inequality
| det J f (ξ)| ≥ c(µ( f , |ξ|))−b ,
2
(21) | det J F (ξi, j )| ≤ 2−ci −i (g i (2i ))2 .
Moreover,
i−1
Y ∞
Y i−1
Y ∞
Y
g(2 ) =
i
(1 − 2 ) ·
i−l
(1 − 2 ) =
i−l
(1 − 2 ) ·
l
(1 − 2−l ),
l=1 l=i+1 l=1 l=1
for all j. Since |ξi, j | = |(2i , 1/ j)| < 2i+1 , we have that
2 2
3 2 7
−b
(24) 2−5b(i+1) < 2−5b(log2 |ξi, j |) < 2 2 (log2 |ξi, j |) + 2 log2 |ξi, j | .
Now, by Proposition 5.1
3 2 7
−b
(25) 2 2 (log2 |ξi, j |) + 2 log2 |ξi, j | ≤ 2 bC (µ(F, |ξi, j |))−b ,
27
where C > 0 is an absolute constant. Putting (23), (24) and (25) to-
gether gives us
for all but finitely many indices i, j. Since for every a > 0, 2−2i+bC < a for
all but finitely many i, the proof follows. □
Remark 7.5. Different results in the spirit of Theorem 7.1 were obtained
in [18] and [19]. In one way or another, these results rely on lower
bounds on det J f at zeros of f . Namely, in [18], | det J f (ξi )| is assumed to
be bounded from below by a constant, while in [19] the upper bound for
ζξ ( f , r) involves terms of the form log det J1f (ξi ) . From (22), it follows that
by taking {ci } which increases sufficiently fast, we can make | det J F (ξi, j )|
of Cornalba-Shiffman maps decrease arbitrarily fast. Since, by Theorem
1.5, ζ(F, r, δ) increases as log2 r independently of {ci }, we conclude that
Theorem 1.2 can not be deduced from the results of [18] and [19] using
the above-described strategy.
Remark 7.6. It is interesting to notice that Proposition 7.4 does not rule
out a possibility that Theorem 1.6, or at least the same bound for ζ0 , can
be deduced from Theorem 7.1. Indeed, we have not proven any lower
bound on the count of islands of Cornalba-Shiffman maps. As a matter
of fact, it is not even clear if for each δ > 0 and each sequence {ci },
ζ0 (F, r, δ) → +∞ as r → +∞. Namely, it may happen that starting from
certain finite r0 , all connected components of {|F | ≤ δ}, which contain
zeros of F , elongate all the way to infinity in w-direction and thus never
become islands, but rather remain peninsulas for all r > r0 .
Question 7.7. Is it true that for each δ > 0 and all sequences c, ζ0 (F, r, δ) →
+∞ as r → +∞? If so, what is the possible growth rate of ζ0 (F, r, δ) de-
pending on parameters δ and c?
Remark 7.8. The main technical ingredient in [20] and [19] is Theorem
3.6 from [20] (slightly modified in [19]). While the proof of this result
has certain similarities with the proof of Theorem 1.6, it seems that the
two approaches are fundamentally different. Namely, approximation of a
holomorphic map by a polynomial is the key idea in the proof of Theorem
1.6, while it is not directly used in the proof of Theorem 3.6 in [20].
It would be interesting to explore how the methods of [20] and [19]
relate to the coarse counts of zeros. In the opposite direction, it would
be interesting to deduce the results of [19] using the same strategy as
above, by proving a suitable analogue of Theorem 1.6.
28
REFERENCES
[1] L. Arosio, A. M. Benini, J. E. Fornæss, and H. Peters. Dynamics
of transcendental Hénon maps III: Infinite entropy. J. Mod. Dyn.,
17:465–479, 2021.
[2] M. Artin and B. Mazur. On periodic points. Ann. of Math. (2),
81:82–99, 1965.
[3] U. Bauer and M. Lesnick. Induced matchings and the algebraic
stability of persistence barcodes. J. Comput. Geom., 6(2):162–191,
2015.
[4] M. B. Botnan and M. Lesnick. An introduction to multiparameter
persistence. Preprint, arXiv:2203.14289, 2023.
[5] M. B. Botnan, S. Oppermann, and S. Oudot. Signed barcodes for
multi-parameter persistence via rank decompositions. In 38th In-
ternational Symposium on Computational Geometry, volume 224
of LIPIcs. Leibniz Int. Proc. Inform., pages Art. No. 19, 18. Schloss
Dagstuhl. Leibniz-Zent. Inform., Wadern, 2022.
[6] M. B. Botnan, S. Oppermann, S. Oudot, and L. Scoccola. On the
bottleneck stability of rank decompositions of multi-parameter per-
sistence modules. Preprint, arXiv:2208.00300, 2022.
[7] L. Buhovsky, J. Payette, I. Polterovich, L. Polterovich, E. Shelukhin,
and V. Stojisavljević. Coarse nodal count and topological persis-
tence. Preprint, arXiv:2206.06347, 2022.
[8] J. A. Carlson. A moving lemma for the transcendental Bezout prob-
lem. Ann. of Math. (2), 103(2):305–330, 1976.
[9] W.-L. Chow. On compact complex analytic varieties. American Jour-
nal of Mathematics, 71(4):893–914, 1949.
[10] M. Cornalba and P. Griffiths. Analytic cycles and vector bundles on
non-compact algebraic varieties. Invent. Math., 28:1–106, 1975.
[11] M. Cornalba and B. Shiffman. A counterexample to the “transcen-
dental Bézout problem”. Ann. of Math. (2), 96:402–406, 1972.
[12] J. P. D’Angelo. Several complex variables and the geometry of real
hypersurfaces. Studies in Advanced Mathematics. CRC Press, Boca
Raton, FL, 1993.
[13] W. Fulton. Intersection theory, volume 2 of Ergebnisse der Mathe-
matik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in
Mathematics. Springer-Verlag, Berlin, second edition, 1998.
[14] A. A. Gol’dberg, B. Y. Levin, and I. Ostrovskii. Entire and meromor-
phic functions. Springer, 1997.
[15] P. A. Griffiths. Function theory of finite order on algebraic varieties.
I(B). J. Differential Geometry, 7:45–66, 1972.
29
[16] Y. Gutman, Y. Qiao, and M. Tsukamoto. Application of signal anal-
ysis to the embedding problem of Zk -actions. Geom. Funct. Anal.,
29(5):1440–1502, 2019.
[17] D. G. C. Handron. Generalized billiard paths and Morse theory for
manifolds with corners. Topology Appl., 126(1-2):83–118, 2002.
[18] S. Ji. Bézout estimate for entire holomorphic maps and their Jaco-
bians. Amer. J. Math., 117(2):395–403, 1995.
[19] B. Q. Li. On the Bézout problem and area of interpolating varieties
in Cn . II. Amer. J. Math., 120(6):1191–1198, 1998.
[20] B. Q. Li and B. A. Taylor. On the Bézout problem and area of inter-
polating varieties in Cn . Amer. J. Math., 118(5):989–1010, 1996.
[21] A. Logunov. Nodal sets of Laplace eigenfunctions: proof of Nadi-
rashvili’s conjecture and of the lower bound in Yau’s conjecture.
Ann. of Math. (2), 187(1):241–262, 2018.
[22] J. Milnor. On the Betti numbers of real varieties. Proc. Amer. Math.
Soc., 15:275–280, 1964.
[23] S. Oudot and L. Scoccola. On the stability of multigraded Betti
numbers and Hilbert functions. Preprint, arXiv:2112.11901, 2023.
[24] L. Polterovich, D. Rosen, K. Samvelyan, and J. Zhang. Topological
Persistence in Geometry and Analysis, volume 74 of University Lecture
Series. American Mathematical Society, Providence, 2020.
[25] J.-P. Serre. Géométrie analytique et géométrie algébrique. Ann.
Inst. Fourier, VI (1955–56), pages 1–42, 1955.
30