Mathematics Department

Real Plane Algebraic Curves

Pedro González García
Master’s thesis in Mathematics. MAT-3900. 2020/2021
 Abstract

This master thesis studies several properties of real plane algebraic curves, focusing on the
case of even degree. The question of the relative positions of the connected components
of real plane algebraic curves originates in Hilbert’s sixteenth problem which, despite its
prominence, is still open in the case of higher degree curves. The goal of this thesis is an
exposition of fundamental contributions to this problem, which have been obtained within
the last century. The main aim of the thesis is to clarify these and to make them more
accessible.
Chapter 1 gives a brief introduction into the study of real plane algebraic curves. The
exposition of this chapter builds on the standard knowledge which are normally obtained
in an undergraduate course of algebraic curves, which usually focus only on complex plane
algebraic curves. In Chapter 2, several topological properties of real plane curves are
developed. The main statements here can be mostly established from Bezout’s theorem
and its consequences. The main result presented in this chapter is Harnack’s inequality
and the classification of the curves until degree five. The goal of Chapter 3 is to prove
Petrovski’s inequalities using Morse theoretic results along with the original arguments
which appeared in Petrovski’s manuscript. Chapter 4 presents results arising from the
complexification of a real plane curve. Finally, Chapter 5 mainly presents results from
Smith theory. In particular, this allows to see how Smith’s inequality generalizes Harnack’s
inequality which were presented in Chapter 2 to higher dimensions.

i
 CHAPTER 1

Introduction

Definition 1.1. Let K = C or K = R. Let n ∈ N. An affine algebraic set in the affine

space Kn is a set of the form

VK (f1 , . . . , fs ) = {(x1 , . . . , xn ) ∈ Kn : f1 (x1 , . . . , xn ) = 0, . . . , fs (x1 , . . . , xn ) = 0}

for one s ∈ N and f1 , . . . , fs ∈ K[X1 , . . . , Xn ] being polynomials.

Definition 1.2. In particular, an affine plane algebraic curve is the set VK (f ) for a
polynomial f ∈ K[X, Y ].

Here are some of the properties satisfied by the affine plane algebraic curves:

i) VK (f1 · f2 ) = VK (f1 ) ∪ VK (f2 ).

ii) VK (f ) = K2 if and only if f ≡ 0.

iii) If f |g then VK (f ) ⊆ VK (g).

Definition 1.3. An algebraic set Γ of Kn is called irreducible whenever Γ = Γ1 ∪ Γ2

implies that Γ = Γ1 or Γ = Γ2 , for Γ1 and Γ2 being algebraic sets in Kn .

Definition 1.4. A curve VK (f ) is irreducible if it is irreducible in the sense of an algebraic

set.

Definition 1.5. We say that the polynomial f is a minimal polynomial for the curve
VC (f ) = Γ if every other polynomial that vanish in the curve is a multiple of f .

We will stress differences between the real and the complex plane algebraic curves.
The first question is how big are these sets depending on the field we work with.

1
Chapter 1. Introduction

Proposition 1.6. Every plane affine complex algebraic curve is an infinite set.

Proof. This is taken from [3]. Let Γ = VC (f ). Suppose that n = degY (f ) where f ∈
K[X, Y ], then
fX (Y ) = a0 (X)Y n + a1 (X)Y n−1 + . . . + an (X)
with ai (X) ∈ C[X] for i = 1, . . . , n and a0 (X) 6= 0. By the fundamental theorem of
algebra we have that the number of distinct roots of a0 (X) is bounded by the degree of
a0 (X), then there are infinite values of x ∈ C that are not a root of a0 (X).
We pick one of this x ∈ C and let g(Y ) = f (x, Y ) ∈ C[Y ]. Then n = deg(g) and the
polynomial, by the fundamental theorem of algebra has 1 ≤ k ≤ n distinct roots in C, lets
call them α1 , α2 , . . . , αk . Then the points (x, α1 ), (x, α2 ) . . . (x, αn ) belong to Γ = VC (f ).
Since we have infinite possibilities for these x ∈ C, we are done.

Theorem 1.7. (Bezout’s Theorem). Let F, G ∈ C[X, Y, Z] be two homogeneous polinomi-

als i.e. with all its monomials being of the same degree, that are coprime and with degree
d and e respectively. Then, counting the points with multiplicity:

card(VC (F ) ∩ VC (G)) = d · e.

The proof of this result can be found in [3] and in [5]. Let us study now the real case.

Example.

Let K = R, f1 = X 2 + Y 2 − 1, f2 = X 2 + Y 2 and f3 = X 2 + Y 2 + 1. Then VR (f1 )

is marked in red in figure 1.1 and VR (f2 ) = {(0, 0)} in black. Since there is no real pair
(x, y) ∈ R2 that satisfying X 2 + Y 2 + 1 = 0, we have that VR (f3 ) = ∅.

Figure 1.1: VR (f1 ), VR (f2 ) and VR (f3 )

Thus, we have seen in this example that not only there could be only a finite number
of solutions, but also that it could be none at all in the real case. This also shows that,
unlike in the complex case, the curve Γ ⊂ R2 may have not an unique minimal polynomial
f , since ∅ = VR (X 2 + Y 2 + 1) = VR (X 2 + Y 2 + 2).

Proposition 1.8. Every algebraic set in R2 is a real algebraic curve. In particular, every
finite set in R2 is a real algebraic curve.

Proof. This is taken from [4]. We have VR (f1 , . . . , fs ) = VR (f1 ) ∩ . . . ∩ VR (fs ) = VR (f ) for
f = f12 + · · · + fs2 .

2
 Notice that f1 = X 2 + Y 2 − 1 has both negative and positive values (substituing
(x, y) = (0, 0) and (x, y) = (1, 1)), while f2 = X 2 + Y 2 ≥ 0 and f3 = X 2 + Y 2 + 1 > 0.
This leads us to consider the following definition, about the character of a polynomial:
Definition 1.9. The polynomial f ∈ R[X1 , . . . , Xn ] is indefinite if there exist a, b ∈ Rn
s.t. f (a) < 0 < f (b). It is called positive or negative definite polynomial if f (a) > 0 or
f (a) < 0 for every a ∈ Rn respectively, and positive or negative semidefinite polynomial
if f (a) ≥ 0 or f (a) ≤ 0 for every a ∈ Rn respectively.
The character of a polynomial does not change under an affine change of coordinates.
In the former example we see that only the behaviour of real curves defined by indefinite
polynomials is similar to the behaviour of complex curves, so we will start studing them.
Lemma 1.10. If f ∈ R[X, Y ] is indefinite, then VR (f ) is infinite.
Proof. This is taken from [4]. By the definiton of indefinite polynomial and perhaps after
a change of coordinates, we can assume that exist x, y1 , y2 ∈ R s.t. f (x, y1 ) < 0 < f (x, y2 ).
If we apply Bolzano’s theorem to the polynomial

fx (Y ) = a0 (x)Y n + a1 (x)Y n−1 + . . . + an (x)

with ai (x) ∈ R[X] for i = 1, . . . , n, we have that there exists (x, y) ∈ R2 s.t. f (x, y) = 0 i.e.
(x, y) ∈ VR (f ). By the continuity of f there exists one δ > 0 s.t. whenever x0 ∈ (x−δ, x+δ)
we still have the inequalities f (x0 , y1 ) < 0 < f (x0 , y2 ). Then applying Bolzano’s theorem,
for every fixed x0 ∈ (x − δ, x + δ) we get one y0 such that (x0 , y0 ) ∈ VR (f ), and thus VR (f )
is infinite.
Proposition 1.11. Let f ∈ R[X, Y ] be a polynomial of positive odd degree. Then f is
indefinite.
Proof. This is taken from [4]. Let f = f(0) +f(1) +· · ·+f(2k+1) with k ∈ N and f(p) being the
homogeneus part of degree p and being f(2k+1) 6= 0. The f(i) part (for i = 0, 1, . . . , 2k + 1)
looks like the following

f(i) (X, Y ) = ai0 X i + a1i−1 X i−1 Y + . . . + a0i Y i

so it is clear that f(i) (X, X) will be something of the form

f(i) (X, X) = λi X i

with λi = ai0 + ai−1

1 + . . . + a0i ∈ R. Then

f (X, X) = λ0 + λ1 X 1 + . . . + λ2k+1 X 2k+1

Suppose that λ2k+1 > 0 (respectively λ2k+1 < 0). Then we can always find one x ∈ R+
big enough (respectively x ∈ R− small enough) sattisfying f (x, x) > 0, and f (−x, −x) < 0.
Thus f is indefinite.

Lemma 1.12. Let f ∈ R[X1 , . . . , Xn ] be of positive degree d. If there exists P ∈ VR (f )

s.t. the gradient of f at the point P does not vanish, then f is indefinite.

3
Chapter 1. Introduction

Proof. This is taken from [4]. Let P = (p1 , . . . , pn ), it holds that (perhaps after a change of
∂f
coordinates) ∂X 1
(P ) 6= 0. Then the function fX2 ,...,Xn (X1 ) is strictly monotonous on some
neighborhood of p1 , and since fp2 ,...,pn (p1 ) = 0, then f changes sign in this neighborhood.

Remark. In the future it will be ocasionally used the following notation for the partial
∂f ∂f
derivatives fX = ∂X and fY = ∂Y .
Theorem 1.13. Let f ∈ R[X, Y ] be an irreducible polynomial. Then f is indefinite ⇐⇒
VR (f ) is infinite.
Proof. This is taken from [4]. ⇒ Is just lemma 1.10
⇐ Since f is irreducible, then the intersection of VR (f ) with any other VR (g) with g
not being a multiple of f , is finite, by Bezout’s theorem 1.7. In particular, since the degree
of fX and fY is smaller than the degree of f , then VR (f ) ∩ VR (fX ) ∩ VR (fY ) is finite. Since
VR (f ) is infinite, there must be a member of this set that does not vanish the gradient,
and by lemma 1.12, we have that f is indefinite.
Theorem 1.14. (Real Study’s lemma). Given f, g ∈ R[X, Y ] both of positive degree s.t.
f is irreducible in R[X, Y ], indefinite and VR (f ) ⊆ VR (g), then f divides g.
Proof. Suppose that f does not divide g. Then f and g have no common components
since f is irreducible. Also VR (f ) ∩ VR (g) = VR (f ) is a finite set by Bezout’s theorem 1.7,
but, since f is indefinite, we have by theorem 1.13 that VR (f ) is infinite, contradiction.
Notice that the inverse inclusion direction of the last theorem is already given by the
property iii) sattisfied by all the affine plane real algebraic curves.
Theorem 1.15. (Real Irreducibility Condition). If g ∈ R[X, Y ] is indefinite and irre-
ducible in R[X, Y ] then VR (g) is irreducible.
Proof. This is taken from [4]. Suppose that VR (g) = Γ1 ∪ Γ2 , with Γ1 = VR (h) and
Γ2 = VR (f ) by proposition 1.8. Then VR (g) = VR (h) ∪ VR (f ) = VR (hf ) and by the Real
Study’s lemma we get g|hf . Since g is irreducible, then either g|h or g|f . Thus, either
VR (g) ⊆ VR (h) or VR (g) ⊆ VR (f ) holds, then either VR (g) = Γ1 or VR (g) = Γ2 .
Remark. We can observe that the real affine curves VR (f ) with f indefinite have some
similar properties to complex affine curves.
Theorem 1.16. Let f ∈ R[X, Y ] be an indefinite polynomial of positive degree such that
it factorizes into irreducible and indefinite polynomial factors as follows f = f1m1 · . . . · fnmn .
Then, the following are equivalent:
i) VR (f ) ⊂ VR (g)

ii) f1 · . . . · fn divides g.

iii) There exist m ∈ N, s.t. f |g m .

In particular, V (f ) = V (g) ⇐⇒ f and g have the same irreducible and indefinite
components.

4
 Proof. It will be proven in a cyclical way.

i) ⇒ ii) Since each fi is an indefinite polynomial, then VR (fi ) is an infinite set. Then since
VR (f ) ⊂ VR (g), it is also true that VR (fi ) ⊂ VR (g), and by the Real Study’s Lemma
1.14, then fi divides to g. Since fi are coprime between each other, it follows that
f1 · . . . · fn divides g.

ii) ⇒ iii) Take m = max{m1 , . . . , mn }.

iii) ⇒ i) From f |g m we obtain that VR (f ) ⊂ VR (g m ), and since it holds that VR (g m ) = VR (g),

then VR (f ) ⊂ VR (g).

Definition 1.17. Let K = C or K = R and n ∈ N. A projective algebraic set in the

projective space KPn is a set of the form

VK (F1 , . . . , Fs ) = {(x0 : . . . : xn ) ∈ KPn : F1 (x0 : . . . : xn ) = 0, . . . , Fs (x0 : . . . : xn ) = 0}

for one s ∈ N and F1 , . . . , Fs ∈ K[X0 , . . . , Xn ] being homogeneous polynomials.

It is mandatory to work with homogeneus polynomials in the projective case because

both (x : y : z) ≡ (λx : λy : λz) are homogeneus coordinates of the same point in KP2 ,
then we need F (λx : λy : λz) = 0 if and only if F (x : y : z) = 0 for every (x : y : z) ∈ KP2 .

Definition 1.18. Let f ∈ K[X, Y ] be a polynomial of degree d, then its homogeneiza-

tion is defined as the polynomial
 
d X Y
F (X : Y : Z) = Z f , .
Z Z

Definition 1.19. Let F ∈ K[X, Y, Z] be a form of degree d and assume that Z (re-
spectively Y and X) does not divide F , then it is defined its dehomogeneization with
respect to Z (respectively with respect to Y and with respect to X) f ∈ K[X, Y ] as:
f (X, Y ) = F (X, Y, 1) (respectively f (X, Z) = F (X, 1, Z) and f (Y, Z) = F (1, Y, Z)).

Proposition 1.20. Let f ∈ R[X, Y ] be of positive degree d and F ∈ R[X, Y, Z] be its

homogeneization. Then:

i) f is indefinite ⇐⇒ F is indefinite.

ii) f is semi-definite ⇐⇒ F is semi-definite.

iii) If f is definite then F is semi-definite.

iv) If F is definite then f is definite.

i) ⇒ Recall that F (X : Y : Z) = Z d f X , Y . There exists (x1 , x2 ) ∈ R2 and

Proof. Z Z
(y1 , y2 ) ∈ R2 s.t. f (x1 , x2 ) < 0 and f (y1 , y2 ) > 0, and so F (x1 : x2 : 1) = f (x1 , x2 ) <
0 and F (y1 : y2 : 1) = f (y1 , y2 ) > 0.

5
Chapter 1. Introduction

⇐ There exists (x1 : x2 : x3 ) s.t. F (x1 : x2 : x3 ) > 0 and (y1 : y2 : y3 ) s.t.

F (y1 : y2 : y3 ) < 0. Suppose that x3 , y3 6= 0 (the other cases are analogous) then
   
x1 x2 x1 x2
0 < F (x1 : x2 : x3 ) = F : :1 =f ,
x3 x3 x3 x3
and    
y1 y2 y1 y2
0 > F (y1 : y2 : y3 ) = F : :1 =f , ,
y3 y3 y3 y3
then f is also indefinite.
ii) Is similar to i)
iii) Suppose f (X, Y ) > 0 for every (X, Y ) ∈ R2 , we can split f in f (X, Y ) = λ+g(X, Y )
with λ > 0 and g is positive semi-definite. Then
 
d d X Y
F (X : Y : Z) = Z λ + Z g ,
Z Z
is semidefinite, since
 
X Y
F (X : Y : 0) = 0 and F (X : Y : 1) = λ + g , > 0.
1 1

iv) Let F (X : Y : Z) > 0 for every (X : Y : Z) ∈ RP2 , then 0 < F (X : Y : 1) = f (X, Y )

for every (X, Y ) ∈ R2 and so f is definite.

Example.
f = X 2 + 1 is definite but F = X 2 + Z 2 is semi-definite non-definite.
Definition 1.21. Lets define Vd as the vector space of homogeneous polynomials of degree
d in K[X0 , X1 , X2 ].
Example. The following generic projective conic is inside V2
U00 X02 + U01 X0 X1 + U02 X0 X2 + U11 X12 + U12 X1 X2 + U22 X22 ∈ V2 .
Proposition 1.22. The vector space Vd has dimension d+2


2
.
Proof. This is based on [5]. A basis of Vd is a set with all the possible monomials of degree
d with the variables
X0 , X1 , X2 , then we need to see that there exists as many monomials
of this type as d+2 2
.
Any monomial of degree d can be represented as the product Xi1 Xi2 · · · Xid , where
i1 , i2 , . . . , id ∈ {0, 1, 2}. The order of this product is not important of course, then there
is as many monomials as combinations with repetition of the 3 elements {0, 1, 2} being
taken in groups of d, and this number is precisely
     
d+3−1 d+2 d+2
= = .
d d 2

6
Remark. Since Vd has dimension d+2


2
, if we dehomogeneizate a homogeneous polynomial
F ∈
K[X0 , X1 , X2 ] into f ∈ K[X, Y ] we lose a monomial (X0d concretely), then f has really
d+2
2
− 1 = d(d+3)
2
monomials.
Example. A projective conic looks like the following

F (X0 : X1 : X2 ) = a00 X02 + a01 X0 X1 + a02 X0 X2 + a11 X12 + a12 X1 X2 + a22 X22

and its dehomogeneization with respect to X0 is

f (X, Y ) = a00 + a01 X + a02 Y + a11 X 2 + a12 XY + a22 Y 2

2(2+3)
and as we noticed in the remark, we have 2
= 5 monomials.
We can also observe that since the conic can be represented by a polynomial up to a
multiplication by a constant, then we can differentiate 2 cases, when a00 6= 0 and a00 = 0:
         
a01 a02 a11 2 a12 a22
X+ Y + X + XY + Y 2 = −1
a00 a00 a00 a00 a00
a01 X + a02 Y + a11 X 2 + a12 XY + a22 Y 2 = 0

After this observation, we see that taking precisely 2(2+3)2

= 5 points in C2 , say (xi , yi )
for i = 1, 2, 3, 4, 5 then we have, substituing in the conic equation (lets stick now to the
equation with
  −1 on the rigth hand
 side, the
 other
 case
 is analogous), 5 equations for 5
variables ( a00 = a, a00 = b, a00 = c, a00 = d, aa22
a01 a02 a11 a12
00
= e):

axi + byi + cx2i + dxi yi + eyi2 = −1

that will give us the equation of a conic that pass by this (xi , yi ) points:

aX + bY + cX 2 + dXY + eY 2 = −1.

Corollary 1.23. For d+2 − 1 = d(d+3)

2 2
points in C2 , we can always find a curve of degree
d passing through them.
Proof. We just need to use last proposition 1.22 and generalize the procedure described
for a conic above, instead of a system of 2(2+3)
2
= 5 equations for 2(2+3)
2
= 5, variables we
d(d+3) d(d+3)
will get 2 equations for 2 variables, with the variables being the coefficients of a
curve of degree d that pass by d(d+3)
2
chosen points.
Theorem 1.24. Let Γ = VC (F ) for an irreducible homogeneous polynomial F ∈ C[X, Y, Z]
of degree d, then the maximum number of singular points that Γ can have is (d−1)(d−2)
2
.
Proof. This is taken from [3] and [5]. For d = 1, suppose there would be one singular
point in the line Γ, then any other line intersecting Γ in that point will intersect with
multiplicity bigger than one, contradicting Bezout’s theorem.
For d = 2, suppose there would be a singular point. Then choosing any line going from
this point to any other of the conic, the line will be intersecting the conic with multiplicity
bigger than 2, contradicting again Bezout’s theorem.

7
Chapter 1. Introduction

Now that we have it for d = 1, 2, we will suppose d ≥ 3. Suppose that Γ has 1 +

(d−1)(d−2) 2
2
= d −3d+4
2
singular points. Taking (d−2)(d+1)
2
points, by the last corollary 1.23,
2
we can always find a curve C of degree d − 2 passing by the d −3d+4 2
singular points and
another
(d − 2)(d + 1) d2 − 3d + 4
− =d−3
2 2
points of Γ, that by hypothesis they are regular points.
Since the multiplicity of intersection between C and Γ in the singular points should be
at least 2, counting the multiplicities, we have at least
 2 
d − 3d + 4
2· + d − 3 = d2 − 2d + 1 = d(d − 2) + 1
2
intersection points (counted with multiplicity) between C and Γ, since Γ was irreducible,
this contradicts Bezout’s theorem.
Theorem 1.25. (Sign change criterion). Let F ∈ R[X0 , X1 , X2 ] be a homogeneous poly-
nomial of positive degree without multiple irreducible factors. We have that F is indefinite
⇐⇒ there exists a regular point P ∈ VR (F ).
Proof. Based on [4]. ⇒ If F is indefinite, we have that at least one irreducible factor of
F , say Fi , is indefinite. Thus VR (Fi ) is an infinite set by lemma 1.10 and proposition 1.20.
Since the set of singular points of a homogeneous polynomial is finite by theorem 1.24,
then there must be a regular point inside VR (Fi ), and thus inside VR (F ).
⇐ Suppose now that there exists a regular point P ∈ VR (F ). This point has at least
one coordinate that is not zero, suppose that is the first one, thus P = (1 : a : b). Taking
now f (X, Y ) = F (1, X, Y ), since fX (a, b) and fY (a, b) are not both null, then there exists
a regular point ((a, b) concretely) in the set VR (f ). Using lemma 1.12 we obtain that f is
indefinite, and by proposition 1.20 we have that F is indefinite also.
Lemma 1.26. Suppose that f ∈ R[X1 , . . . , Xn ] is irreducible in R[X1 , . . . , Xn ]. Then f is
reducible in C[X1 , . . . , Xn ] ⇐⇒ either f or −f is a sum of two squares in R[X1 , . . . , Xn ].
√
Proof. This is taken from [4]. ⇐ If f = r12 + r22 with r1 , r2 ∈ R[X1 , . . . , Xn ] and i = −1
then f = (r1 + ir2 )(r1 − ir2 ) is reducible in C[X1 , . . . , Xn ].
⇒ Suppose that f factors non-trivially f = (r1 + ir2 )(s1 + is2 ) with r1 , r2 , s1 , s2 ∈
R[X1 , . . . , Xn ]. Taking complex conjugates we obtain f ·f = (r12 +r22 )(s21 +s22 ). Since f was
irreducible in R[X1 , . . . , Xn ] then either (r12 +r22 ) or (s21 +s22 ) divides f , then f = a2 (r12 +r22 )
for some a ∈ C−{0} (analogously for (s21 +s22 )) and so f or −f is a sum of two squares.
Definition 1.27. Define the following function for d ∈ N:
 2 
d (d − 1)(d − 2)
α(d) := max , .
4 2
2
For d ≤ 5, α(d) = d4 and for d ≥ 6, (d−1)(d−2)
2
. Taking derivatives at d in both d
4
and
(d−1)(d−2) α(d)
2d
it is obtained that d is monotically increasing in N. Thus
α(dj ) α(d1 + d2 )
≤ ,
dj d1 + d2

8
for j = 1, 2
 
d1 d2
α(d1 ) + α(d2 ) ≤ + α(d1 + d2 ) = α(d1 + d2 ).
d1 + d2 d1 + d2

Theorem 1.28. Let F ∈ R[X, Y, Z] be a semi-definite irreducible form of positive degree

d. Then |VR (F )| ≤ α(d).

Proof. This is taken from [4]. First, assume that F is reducible in C[X, Y, Z]. Then by
lemma 1.26 we have that F = R12 +R22 for some forms R1 , R2 ∈ R[X, Y, Z] both necessarily
of degree d2 . Moreover, R1 and R2 must be coprime, since F is irreducible in in R[X, Y, Z].

2
Now by Bezout’s theorem, the set VR (F ) = VR (R1 ) ∩ VR (R2 ) contains at most d2 points
counted with multiplicity.
Suppose now that F is irreducible in C[X, Y, Z]. By 1.25, since F is a semi-definite poly-
nomial, each point of VR (F ) is singular. Therefore by theorem 1.24, |VR (F )| ≤ (d−1)(d−2)
2
.

Proposition 1.29. For any semi-definite homogeneous polynomial F ∈ R[X, Y, Z] of

degree d ≥ 2, then F is divisible by the square of some indefinite (homogeneous) polynomial
⇐⇒ VR (F ) is infinite.

Proof. The first inclusion is taken from [4]. ⇒ Since F is divisible by the square of a
indefinite form G, then VR (G) ⊂ VR (F ). By lemma 1.10 and proposition 1.20, VR (F ) is
infinite.
⇐ Let VR (F ) be infinite, by theorem 1.28, F must be reducible. Suppose that F =
F1 · . . . · Fn , then by theorem 1.13 and proposition 1.20 at least we must have one Fi
indefinite irreducible polynomial, for 1 ≤ i ≤ n. Since F is semi-definite, ignoring all
the Fm semi-definite polynomials of F = F1 · . . . · Fn , there must be one Fj irreducible
indefinite polynomial for j 6= i, s.t. it have the same negative and positive interval that
Fi , making Fi · Fj ≥ 0. In oher words, they must change the sign on the same points,
i.e. V (Fi ) = V (Fj ). Therefore by the Real Study’s lemma Fi |Fj and Fj |Fi , since they are
irreducible, up to a constant multiplication, they must be the same polynomial Fi = Fj .
Thus F is divisible by the square of the indefinite homogeneous polynomial Fi .
The following result is taken from [4]:

Corollary 1.30. If F ∈ R[X, Y, Z] is a homogeneous polynomial of positive degree without

multiple irreducible factors (in particular, if F is irreducible), then:

1. VR (F ) is infinite ⇐⇒ F is indefinite.

2. VR (F ) is finite non-empty ⇐⇒ F is semi-definite nondefinite.

3. VR (F ) is empty ⇐⇒ F is definite.

9
 CHAPTER 2

Topology on real algebraic curves

First it is important to notice that there are topological differences between the ambient
spaces, R2 and C2 , affine and projective. First R2 and RP2 are connected R-topological
manifolds of dimension 2. Moreover, R2 is orientable while RP2 is both non-orientable and
compact. C2 and CP2 are both connected, orientable R-topological manifolds of dimension
4, and CP2 it is also compact.
As an R-topological manifold, every non-singular curve in C2 has always dimension
2. In CP2 the projective algebraic curves are connected since if there were 2 different
connected components in a complex projective curve, by Bezout’s theorem they have to
intersect with multiplicity equal to the multiplication of their degrees. In RP2 things are
very different.
Let F ∈ R[X, Y, Z] be a homogeneous polynomial such that its decomposition into
coprime irreducible factors is
k k
F = cF1k1 · . . . · Fsks · Fs+1
s+1
· . . . · Ftkt · Ft+1
t+1
· . . . · Frkr

with c ∈ R − {0}, 0 ≤ s ≤ t ≤ r and s, t, r, kj ∈ N for all 1 ≤ j ≤ r. Suppose that Fj is

indefinite, for all j ≤ s, Fj is semi-definite nondefinite for all j with s + 1 ≤ j ≤ t and Fj
is definite, for all j with t + 1 ≤ j ≤ r. Then using last corollary 1.30 it is obtained that

VR (F ) = VR (F1 ) ∪ . . . ∪ VR (Fs ) ∪ VR (Fs+1 ) ∪ . . . ∪ VR (Ft )

where VR (Fs+1 ) ∪ . . . ∪ VR (Ft ) is a finite set. (This is taken from [4]).

Definition 2.1. Using the same notation as above, we define the irreducible components
of the curve VR (F ). There are two different types:

1. Each VR (Fj ), with j < s.

t
S s
S
2. Each point in the set VR (Fj ) \ VR (Fj ).
j=s+1 j=1

10
Note. Each irreducible component of type 2 is an isolated point in VR (F ), and each irre-
ducible component of type 1 is an algebraic irreducible curve.
Then, it can be concluded that every non-singular real curve Γ is of type 1 and has
dimension 1 as an R-topological manifold. A real singular curve Γ that has only irreducible
components of type 2 is of dimension 0. In a singular curve Γ that have both types of
irreducible components then the subset of regular points of the curve is a manifold of
dimension 1, not necessarily dense in the curve Γ. In this last case there are different
dimensions at a local level of the real plane curve, lets see an example.
Example. The cubic VR (X 2 + Y 2 − Y 3 ) has dimension 1, but the local dimension at the
isolated point (0, 0) is 0. Nevertheless, the local dimension of this curve is 1 at any other
point of this cubic. This can be seen in figure 2.1.

Figure 2.1: VR (X 2 + Y 2 − Y 3 )

Definition 2.2. For every point x in a real plane curve Γ and every sufficient small open
disk U with center at x, the set U \ {x} is homeomorphic to the union of an even number
of open segments, which are denoted as half branches of Γ at the point x.
Complex projective curves are unbounded, compact sets and furthermore, by Bezout’s
theorem, they are connected. Nevertheless as we have seen in the last example, in the
projective case, real curves are compact sets but they can fail to be connected. It will be
seen in the following results that there is a bound on the connected components that a
real non-singular curve may have, and that this bound it is always the best possible.
Definition 2.3. An oval of a real projective curve Γ is a connected component whose
complement is not connected, equivalently, an oval is a connected component bounding a
disk in RP2 .
Definition 2.4. A pseudoline is a curve that it is topologically equivalent to a line.
Lemma 2.5. Let Γ = VR (F ) be a non-singular algebraic curve of degree d in RP2 with
F ∈ R[X, Y, Z], then:
i) If d is even, every connected component of Γ is an oval.
i) If d is odd, one of the connected components of Γ is a pseudoline, while all the others
are ovals.

11
Chapter 2. Topology on real algebraic curves

Proof. This is taken from [1]. The real projective plane can be thought as the sphere S2
where all the antipodal points are being identified:
2
π : S2 −→ RP2 = S x ∼ (−x).

Since Γ is a non-singular curve, it can not have self-intersections i.e. Γ can only have
at maximum 1 pseudoline because the pseudolines in RP2 are seen in S2 topologically
equivalent to maximum circles, and is clear that every two maximum circles intersects
exactly at two (antipodal) points in the sphere.
Now let γ be a path in S2 that crosses transversaly to π −1 (Γ) and joins two antipodal
points which are not in π −1 (Γ). If C1 and C2 are connected components of Γ, then γ
intersects π −1 (C2 ) in an even number of points if C2 is an oval and γ intersects π −1 (C1 )
in an odd number of points if C1 is a pseudoline. This can be visualized in figure 2.2:

γ
C1

C2

Figure 2.2: S2 with all its antipodals points being identified.

Counting the sign changes of F along the path γ, we see that Γ have a connected
component that is a pseudoline iff F takes opposite signs near antipodal points i.e. if d is
odd. If F takes tha same sign near antipodal points, then d is even, and there can not be
any pseudoline, so all the connected components must be ovals.

Now we can easily think in the restrictions for the number of connected components for
the case of degree 2. Suppose that there would be more than 1 connected component in
this case, due to the last lemma 2.5, this is the same as saying more than 1 oval. Suppose
there would be 2 ovals, then joining one interior point of each oval, we would have a line
that it is intersecting with multiplicity 4 to a curve of degree 2, which by Bezout’s theorem,
this is impossible. Of course for more than 2 ovals this is also impossible. Then for degree
d = 2 we have 1 connected component as maximum. This can be seen in figure 2.3:

Figure 2.3: Counting geometric intersections between two ovals and a line.

12
Definition 2.6. Let Γ be a curve of degree d ∈ N, then we define its genus as

(d − 1)(d − 2)
g(d) = .
2

Theorem 2.7. (Harnack’s theorem).

The number of connected components c of a non-singular projective curve of degree
d ≥ 2 is bounded:
(d − 1)(d − 2)
c≤ + 1 = g(d) + 1.
2
Proof. This is taken from [1]. By the last commentary on the previous page, it is assumed
that d > 2 since for d = 2 it has already been shown that c ≤ g(2) + 1 = 1. It suffices
also to consider irreducible curves, since for a reducible curve f = g · h, with deg(g) = d1 ,
deg(h) = d2 , the following inequality holds

g(d1 ) + 1 + g(d2 ) + 1 ≤ g(d1 + d2 ) + 1

whenever d1 > 1 or d2 > 1.

Suppose now that Γ is an non-singular irreducible curve of degree d with more than
g(d) + 1 components. Then by the last lemma 2.5, Γ contains p = g(d) + 1 ovals
Ω1 , Ω2 , . . . , Ωp and at least one other connected component (which it will be an oval if
d is even or it will be a pseudoline if d is odd).
Lets pick 12 d(d − 1) − 1 points on Γ. Since 12 d(d − 1) − 1 ≥ g(d) + 1 = p for d ≥ 2, it
can be chosen a point on each of the ovals Ω1 , Ω2 , . . . , Ωp and the rest of the points on the
other connected component left that has Γ. By corollary 1.23, since

(d − 2)(d − 2 + 3) (d + 1)(d − 2) 1
= = d(d − 1) − 1,
2 2 2

there is a curve of degree d − 2, say ∆, passing through all this 12 d(d − 1) − 1 points. The
curves Γ and ∆ have no common irreducible component since Γ is irreducible of degree d
and ∆ is of degree d − 2. By Bezout’s theorem 1.7, the number of intersection points of Γ
and ∆, counted with multiplicity can not be greater than d(d − 2).
If ∆ intersects transversely (i.e. with multiplicity one, see [5]) an oval Ωi in one point,
then ∆ necessarily intersects Ωi in another point. Hence the number of intersection points
between ∆ and Γ counted with multiplicity is at least

d(d − 1) d2 − d + d2 − 3d + 2
− 1 + g(d) + 1 = = (d − 1)2
2 2

since there are at least p = g(d) + 1 ovals. Is clear that (d − 1)2 > d(d − 2) for d > 2, then
by Bezout’s theorem it is concluded that Γ can not have more than g(d) + 1 connected
components.

Note. Notice that since by hypothesis the curve is non-singular, it can not have any self
intersections between the different connected components.

13
Chapter 2. Topology on real algebraic curves

Then the number

(d − 1)(d − 2)
+ 1 = g(d) + 1
2
bounds the number of connected components that a non-singular real curve can have, now
it is natural to ask if this is the best bound possible. The answer will be in fact affirmative
and it is seen in the following proposition.

Proposition 2.8. (Harnack’s construction). The bound g(d) + 1 for a non-singular real
curve of degree d ≥ 2 is the best possible i.e. for each d, there exist a non-singular curve
of degree d in RP2 which reaches this bound.

Proof. This is taken from [1]. Choose a line L in RP2 . Starting with d = 2, it will be
constructed by induction a non-singular curve Γd of degree d in RP2 with exactly g(d) + 1
connected components and which, furthermore, sattisfies the following properties:

i) Γd has a connected component Cd intersecting L in d distinct points.

ii) There is an orientation of L and an orientation of Cd , such that these d distinct

points are arranged in the same order on L as on Cd . Taking this order into account,
we denote this points as a1 , . . . , ad .

iii) For each i = 1, 2, . . . , d − 1 the union of the interval [ai , ai+1 ] on L and the segment
of Cd bounded by ai and ai+1 form an oval in RP2 (which is not smooth).

For d = 2 it is clear that any conic has exactly one connected component, and it can
always be chosen Γ2 to be a non-singular conic intersecting L in 2 distinct points, then i)
holds. We can easily see in the following drawing that it is satisfied also ii). The oval that
is described in iii) is drawed in blue in figure 2.4:

a1 a2 L

Γ2 = C2

Figure 2.4: Drawing of the oval (in blue) sattisfying property iii) for d = 2.

Since for d = 2 holds, using induction, suppose that we have already constructed Γd
with g(d) + 1 components and satisfying properties i), ii), iii). Then the curve Γd+1 is con-
structed as follows, we choose distinct points b1 , b2 , . . . , bd+1 on L s.t. a1 , . . . , ad , b1 . . . , bd+1
are ordered according to the orientation of L. Now choose lines L1 , L2 , . . . , Ld+1 distinct
from L and passing through b1 , b2 , . . . , bd+1 respectively like it is drawed in figure 2.5.

14
 L1 L2 Ld+1

L a1 a2 ad b1 b2 bd+1

Figure 2.5: Drawing lines through points b1 , . . . , bd+1 different than a1 , . . . , ad .

Then the curve Γd+1 is constructed as a small perturbation of the union Γd ∪ L. Iden-
tifying the curves with their respective homogeneous polynomials, we define

d+1
Y
Γd+1 = L · Γd +  · Li
i=1

where  ∈ R is small enough, and its sign is the opposite of the sign of L · Γd at the interior
of the ovals that lie between ai and ai+1 , this way the effect of the perturbation is to shrink
these ovals towards their interior. Hence this perturbation of Cd ∪ L has d − 1 ovals and
an additional connected component Cd+1 that intersects L in b1 , b2 , . . . , bd+1 and so we can
conclude that Γd+1 sattisfies i), ii) and iii) for b1 , b2 , . . . , bd+1 . We can visualize this for
the case d = 5 at figure 2.6 obtained from https://www.math.tamu.edu :

Figure 2.6: Harnack’s construction from the case of degree 5 to degree 6.

Since we are taking  small enough, the perturbation of the ovals of Γd which do not
intersect with L does not erase them, therefore the curve Γd+1 has precisely

(d − 1)(d − 2) d(d − 1)
g(d) + 1 + (d − 1) = +d= + 1 = g(d + 1) + 1
2 2
connected components.

15
Chapter 2. Topology on real algebraic curves

Definition 2.9. (M-curve).

A real plane curve that attains the maximum number of connected components is
called an M-curve.

Example. An elliptic curve with 2 components, such as Y 2 = X 3 − X is an M-curve.

Nevertheless, Y 2 = X 3 − X + 1 has a single component, thus is not an M-curve. See 2.7:

Figure 2.7: VR (Y 2 − X 3 + X) (M-curve) and VR (Y 2 − X 3 + X − 1) (not an M-curve).

Definition 2.10. (Simple Harnack’s curve).

A non-singular curve Γ ⊂ RP2 of degree d is called a non-singular simple Harnack’s
curve if it is an M-curve and:

1. All ovals of Γ are disjoint (this means, they have disjoint interiors) if d is odd.
(k−1)(k−2)
2. One oval of Γ contains 2
ovals in its interior while all other ovals are disjoint
if d = 2k is even.

Example. Let’s construct a simple Harnack’s curve of degree 6. One oval is going to
contain
(3 − 1) · (3 − 2)
=1
2
oval inside and since it is also an M-curve we have that it has (6−1)·(6−2)
2
+ 1 = 11 total
components. Then it will look like the following schematic drawing, and staring at the
image 2.6 from the proof of Harnack’s construction for the case d = 5 to d = 6, we see
that they are both the same curve as it is seen in the drawing below in figure 2.8.

Figure 2.8: Schematic drawing of a simple Harnack’s curve of degree 6.

16
Definition 2.11. The depth of an oval Ω in a non-singular real plane curve Γ is the
number of ovals of Γ containing Ω in their interiors. One oval is nested in another if it is
contained in its interior, we say that a nest has complexity k + 1 if there is at least one
oval of depth k.

Corollary 2.12. (Hilbert’s theorem) For a non-singular real plane curve of degree even
(d = 2k), the complexity of a nest is at most k.

Proof. This proof is based on [16].

Suppose it is true that there is a nest of complexity k + 1, then by the figure 2.9

Ωk+1
L Ω1

Figure 2.9: Hilbert’s theorem proof schematic drawing idea.

is seen that the line L is intersecting the curve with multiplicity at least 2 · (k + 1),
which is impossible by Bezout’s theorem since this multiplicity of intersection can be at
most 2k.
Notice that Harnack’s restriction 2.7 is only restricting how many connected compo-
nents can a non-singular real plane curve have as maximum, but it is not telling about
how this ovals or pseudolines are distributed, up to homeomorphism. Bezout’s theorem
can help on finding out how are distributed in RP2 in the first cases of lower degree.

Definition 2.13. The notation hP i denotes a pseudoline and hni denotes the number
of ovals of a curve. Also to denote m ovals inside one oval, it is denoted as h1hmii. To
express the union of this lasts different cases it is used q.

Theorem 2.14. (Classification of non-singular real plane curves of degree d ≤ 4).

Up to homeomorphism, the classification of non-singular real plane curves of degree
d ≤ 4 is:

i) d = 1 hP i.

ii) d = 2 h0i, h1i.

iii) d = 3 hP i, hP q 1i.

iv) d = 4 h0i, h1i, h2i, h1h1ii, h3i, h4i.

Proof. Based in key ideas from [8] and [15] using original own examples.

i) By lemma 2.5, a non-singular curve of degree d = 1 has one pseudoline and since
it can not have more connected components by Harnack’s theorem 2.7, hence its
clasification is hP i, as example take the non-singular curve defined by f (X, Y ) = Y .

17
Chapter 2. Topology on real algebraic curves

ii) By Harnack’s theorem 2.7 it can only have one connected component and if it is
non-empty, by lemma 2.5 it is an oval, like in VR (f ) for f (X, Y ) = X 2 + Y 2 − 1, then
one possibility is h1i. Nevertheless there are non-singular empty real conics like the
one defined by f (X, Y ) = X 2 + Y 2 + 1 then the other possibility is h0i.
iii) By Harnack’s theorem 2.7 a curve of degree 3 has at most 2 connected components.
By lemma 2.5 it has at least a pseudoline and the rest is ovals. Then clearly the
only two possibilities are like the ones that we study at example 2, respectively
f (X, Y ) = Y 2 − X 3 + X − 1 for hP i and f (X, Y ) = Y 2 − X 3 + X for hP q 1i which
are both non-singular curves.
iv) By Harnack’s theorem 2.7 a curve of degree 4 has at most 4 connected components.
By lemma 2.5 it is only formed by ovals. It can have a nest but of complexity at most
2 by corollary 2.12. It is possible to have at most 1 nest of this type without any
exterior ovals, i.e. h1h1ii, for example f (X, Y ) = (X 2 + Y 2 − 2)(2X 2 + Y 2 − 1) − 1.
Suppose that there would be an exterior oval to the nest, if it is traced a line between
the interior of the nest and the interior of a exterior oval, the line intersects with
multiplicity 6, wich by Bezout’s theorem 1.7 it is impossible.
Other possibility is h0i, thinking on the non-singular cuartic f (X, Y ) = X 4 + Y 4 + 1.
The case h1i is represented by f (X, Y ) = X 4 + Y 4 − 1 for example. To contruct 2
ovals that are not one nested inside the other it can be used the non-singular curve
defined by the polynomial f (X, Y ) = (2X 2 + Y 2 − 1)(X 2 + Y 2 − 1) + 0.11. Hence
it it can be added to the classification h2i.
Case h3i can be constructed out of a perturbation of the following picture which
represents in the left of figure 2.10 an union between an ellipse and another ellipse,
represented by the equation f (X, Y ) = (X 2 + 2Y 2 − 1) (2X 2 + Y 2 − 2). It is
represented also where f is positive and negative at figure 2.10:

Figure 2.10: Small perturbation of two ellipses.

The perturbation drawed in blue is made adding a perturbation in two extra odd
degree coefficients, in this case:
f (X, Y ) = (X 2 + 2Y 2 − 1)(2X 2 + Y 2 − 2) + 0.1X 3 + 0.1Y 3 .

18
 Looking at the partial derivatives and the homogeneization, it is seen that it is a
non-singular complex curve.
To construct the case h4i it will be needed to make again a slight perturbation in
the following union of 2 ellipses, f (X, Y ) = (X 2 + 2Y 2 − 1)(2X 2 + Y 2 − 1)= 0, where
it is represented also where f is positive and negative in figure 2.11:

Figure 2.11: Small perturbation of two transversal ellipses.

Looking at the left picture above, the optimal perturbation to obtain 4 ovals will
be shrinking slightly the negative parts towards their interior, and this can be done
adding one small  > 0 to f , i.e. (X 2 + 2Y 2 − 1)(2X 2 + Y 2 − 1) = −. In particular
for  = 0.1 we obtain the right above picture in blue. Notice that the partial
derivatives of X and Y are the same of f and f + , then the singular points that
there are in the intersection between the 2 ellipses in the left picture are no longer
in the right picture with the perturbation done. Doing the partial derivatives in the
homogeneous polynomial of f (X, Y ) = (X 2 + 2Y 2 − 1)(2X 2 + Y 2 − 1) + 0.1 it is seen
that VC (f ) is a non-singular curve of degree 4 with 4 ovals.

The degree 5 case relies on the clasification of degree 4, adding a pseudoline and
perturbing the curve. Two extra cases must be considered also.

Theorem 2.15. (Classification of non-singular real plane curves of degree 5).

Up to homeomorphism the classification of non-singular real curves of degree 5 is

hP i, hP q 1i, hP q 2i, hP q 1h1ii, hP q 3i, hP q 4i, hP q 5i, hP q 6i.

Proof. The idea of the proof can be found in [8] and [15].
By theorem 2.7 the number of connected components is bounded by (5−1)(5−2)
2
+ 1 = 7.
By lemma 2.5, the possible extra cases for degree 5 that can not be obtained from the
last theorem 2.14 adding a pseudoline to the curve and perturbing it (perturbing the
coefficients) in order to create a non-singular curve are hP q 5i and hP q 6i.

19
Chapter 2. Topology on real algebraic curves

For hP q 5i can be used:

 2 ! 
2 2 2 1 1 1 1
f (X, Y ) = (3X + Y − 1) X + 2 Y − − Y − −√
2 3 2 6

that it is represented with its sign in the following left picture below of figure 2.12.

Figure 2.12: Small perturbation of two ellipses and a tangent line to one of the ellipses.

The optimal way to proceed is to shrink towards itself the negative parts, using a small
 > 0 that makes the curve non-singular, it is obtained a real curve similar to the one at
the right in figure 2.12 for f +  = 0. Thus it is obtained a non-singular quintic with 5
ovals. To construct hP q 6i it can be obtained by the Harnack’s construction 2.8 or it may
be similarly used an optimal perturbation, this time in figure 2.13:

Figure 2.13: Base case to construct hP q 6i by making a perturbation.

20
 CHAPTER 3

Petrovski inequality for the even degree

In the last chapter it was seen the complete classification of non-singular real plane curves
until degree 5. Now for degree 6 it will be seek to find the M-curves using the same
methods from the last chapter. By Harnack’s theorem 2.7 the M-curve of degree 6 has
exactly
(6 − 1)(6 − 2)
+ 1 = 11
2
connected components. Since it is of even degree, by lemma 2.5 it will be only ovals,
so 11 ovals. Using Hilbert’s theorem 2.12 it is impossible to have a nest of complexity
greater than 3. Nevertheless, it is also impossible to have a nest of 3 ovals since joining the
center of one nest of 3 ovals with another oval by a line, this line will be intersecting with
intersection at least 8 to a curve of degree 6, which it is impossible by Bezout’s theorem
1.7. Also an M-curve of degree 6 can not have 2 different nest of complexity 2 each, since
a line between its centers will be intersecting with multiplicity 8, also impossible by the
Bezout’s theorem 1.7. Then as far as we easy get with Bezout’s theorem is to 2 possible
types of curves, one with 11 ovals all outside each other and the other with 11 ovals with
one nest of complexity 2. In fact the second type will be seen that it is in fact the only
possibility by the Petrovski inequality.
This far, it seems that Bezout’s theorem is not enough and that it will be needed
another topological results not mainly based on this theorem to go further.
Definition 3.1. Let f ∈ R[X, Y ], then the set Mt is defined as
Mt = {(x, y) ∈ R2 |f (x, y) ≤ t}
and the level curve set Lt it is defined as Lt = {(x, y) ∈ R2 |f (x, y) = t}.
Definition 3.2. (Critical values).
Let f ∈ R[X, Y ], it is said that a real number c0 is a critical value of f if this
polynomial takes the value c0 at some critical point (x0 , y0 ), i.e.
∂f ∂f
(x0 , y0 ) = (x0 , y0 ) = 0, f (x0 , y0 ) = c0 .
∂X ∂Y
21
Chapter 3. Petrovski inequality for the even degree

∂ ∂
Definition 3.3. It is said that a vector field X = X1 ∂X + X2 ∂Y in R2 (see appendix A) is
a gradient like vector field for a polynomial f ∈ R[X, Y ] if in every non-critical point
(x0 , y0 ) sattisfies

∂f ∂f
X · f (x0 , y0 ) = X1 (x0 , y0 ) (x0 , y0 ) + X2 (x0 , y0 ) (x0 , y0 ) > 0.
∂X ∂Y

Lemma 3.4. (Topological lemma).

If f has no critical value inside the interval [a, b], then M[a,b] is homeomorphic to the
product
f −1 (a) × [0, 1].

Proof. This is taken from [9]. Let X be a gradient like vector field for f . Since X(f ) > 0
at every non-critical point of f , it can be defined a new vector field Y on R2 outside the
critical points:
1
Y = X.
X ·f
Since by hypothesis M[a,b] = {(x, y) ∈ R2 |a ≤ f (x, y) ≤ b} contains no critical points
of f , this vector field Y is well-defined. Consider the integral curve cp (t) of Y which starts
at point p, i.e. cp (0) = p. To see how fast climbs this integral curve from f −1 (a) to f −1 (b),
it is the same as see how fast it goes inside the height function i.e. f :

d dcp 1
f (cp (t)) = (t) · f = Y (cp (t)) · f = X · f = 1.
dt dt X ·f

Thus the integral curve climbs up from f −1 (a) with a constant speed 1 with respect to
the height that is defined by f as shown in figure 3.1. Since it starts at the level f = a at
t = 0 it will reach the level b at the time t = b − a.

f −1 (b)

cp (t)

p f −1 (a)

Figure 3.1: The integral curve cp (t) climbing through the height of f .

There is no harm on consider [a, b] instead of [0, b − a] redefining the integral curve c
as c(t) := c(t + a). Then it can be defined the following continuous map

h : f −1 (a) × [a, b] −→ M[a,b]

(p, t) −→ cp (t).

22
 Then it can be considered the following continuous map

h−1 : M[a,b] −→ f −1 (a) × [a, b]

p −→ (cp (−f (p)), f (p)).

Thus M[a,b] is homeomorphic to f −1 (a) × [a, b]. Now since there can be constructed also
the following homeomorphism

H : [a, b] −→ [0, 1]
t−a
t −→
b−a
it is concluded that
M[a,b] ∼
= f −1 (a) × [0, 1].

Theorem 3.5. Let b < c be real numbers such that f ∈ R[X, Y ] has no critical values
in the interval [b, c]. Then Mb and Mc are homeomorphic. In particular, Lb and Lc are
homeomorphic.
Proof. This is taken from [9]. Denote again by M[b,c] the portion between Lb and Lc

M[b,c] = {(x, y) ∈ R2 |b ≤ f (x, y) ≤ c}.

By the previous definitions it is clear that Mb ∪ M[b,c] = Mc . Since the number of singular
points of f (x, y) is bounded, there can not be a convergent sequence of critical values
{bn }∞
n=1 that converges to b, which by hypothesis it is not a critical value. Thus it can be
assumed that f has no critical points in M[b−,c] for a small enough  > 0. By the topological
lemma above 3.4, M[b−,c] is homeomorphic to the product Lb− × [0, 1]. Also note that
M[b−,b] ⊂ M[b−,c] and f has no critical points either in M[b−,b] , thus M[b−,b] ∼
= Lb− × [0, 1]
too by lemma 3.4. Then we have a homeomorphism

h : M[b−,b] −→ M[b−,c]

where the restriction to the level curve Lb− must be the identity map. In addition, it can
be defined the following homeorphism using the identity map

H = id ∪ h : Mb− ∪ M[b−,b] −→ Mb− ∪ M[b−,c]

Since Mb− ∪ M[b−,b] = Mb and Mb− ∪ M[b−,c] = Mc , it can be obtained as consecuence

the homeomorphism
H : Mb −→ Mc .

Definition 3.6. (Positive and negative ovals).

An oval is said to be positive if when crossing the oval from the inside, the function
decreases and to be negative if when crossing the oval from the inside, the function
increases.

23
Chapter 3. Petrovski inequality for the even degree

Remark. Now, with this definition of positive and negative ovals it will be seeked to find
a restriction for the difference between the number of negative ovals and the number of
positive ovals that a curve of even degree has.
Lemma 3.7. (Lemma 1).
Let VR (F ) be a non-singular real plane curve of even order n with F ∈ K[X0 , X1 , X2 ]
being a homogeneous polynomial. Then it can be continuously slightly perturbed any of its
coefficients without changing its topological structure.
Proof. By Morse’s lemma 3.5 the coefficient a00 can be slightly variated in
F (X0 , X1 , X2 ) = an,0 X2n + an−1,1 X2n−1 X1 + . . . + a0,0 X0n ,
without changing the topological structure of the curve.
It can be made the following change of coordinates on the axis
H : RP2 −→ RP2
[X0 : X1 : X2 ] −→ [X2 : X1 : X0 ]
and F will still being a non-singular real curve under this kind of change of coordinates
(changing the axis), since H is an homeomorphism it can be deduced that the coefficient
an,0 can be also continuously variated without changing the topological structure of the
curve. Analogously it can be deduced that the coefficient a0,n can be also slightly perturbed
without changing the topological structure of the curve.
Every small perturbation in a coefficient ai,j with i + j ≤ n due to adding to ai,j a
small enough (in absolute value) epsilon  can be continuously controlled
|||X2i X1j X0n−i−j | < |0 |X0n + |1 |X1n + |2 |X2n (3.1)
by |0 |, |1 | and |2 | perturbations (respectively of a0,0 , a0,n and an,0 ) that do not change
the topological structure of the curve. This inequality is clear whenever a point takes
Xi = 0 for any i = 0, 1, 2. Now consider any point P s.t. Xi 6= 0 for every i = 0, 1, 2. Let
P = [1 : a : b]. It can always be considered that both |a| and |b| are bigger than 1 since
we can always divide all the coordinates by the smallest number (in absolute value) on
the coordinates of P and then do a change of coordinates if it would be necessary. Taking
µ = max{|a|, |b|} and supposing that µ = |a| (analogously for µ = |b|) is clear for || < |1 |
(|| < |2 | analogously) that
|||bi aj | < |1 |µn = |1 |an < |0 | + |1 |an + |2 |bn .
Thus taking || < min{|0 |, |1 |, |2 |} then the inequality (3.1) holds always true.
Since the perturbation of ai,j is continuously made and it is controlled by another
continuous perturbation that does not change the topology of the curve, it is deduced
that this perturbation of ai,j does not change either the topology of the curve.
Lemma 3.8. (Lemma 2).
Let VR (f ) be a non-singular real plane curve of even order n with f ∈ R[X, Y ]. Then it
can be slightly perturbed its equation without changing its order or its topological structure
s.t. the equations
∂f ∂f
= 0, = 0, (3.2)
∂X ∂Y
will have (n − 1)2 different finite solutions, real or imaginary.

24
Proof. Based on [6]. By the last lemma 3.7, the coefficients of f ∈ K[X, Y ] can be
continuously slightly variated such that the topology of the curve does not change, i.e.
such that no singularities will arise in the new curve. When the equations from (3.2)
have an infinite number of solutions is because they both have common components. This
situation can be solved by a continuous perturbation of the equation, lets see this. Suppose
that p1 , p2 , h ∈ R[X, Y ] are non-constant polynomials s.t.

∂f ∂f
= p1 · h, = p2 · h. (3.3)
∂X ∂Y
Then it will happen that the partials have a common component h(X, Y ) and due to this,
an infinite number of solutions. Since the set of irreducible polynomials in R[X, Y ] is dense,
whenever is made any continuous perturbation in f we are able to find one irreducible
polynomial in any of the partials, making (3.3) impossible to be, and thus having a finite
number of solutions.
If (3.2) had not all different solutions like it is being seeked, it is always possible to
get to a different polynomial with a small perturbation s.t. all its solutions in (3.2) are
distinct. Thus either real or imaginary, all the roots in (3.2) can be considered to be all
different, and by Bezout’s theorem they are going to be exactly (n − 1)2 solutions in the
projective plane.
The finite points set in the projective plane (points that are not in the line at infinity)
is dense, thus it is always possible to make all this (n − 1)2 solutions of (3.2) to be finite
ones due to a slight continuously perturbation in the coefficients of f .

Definition 3.9. (Plus and minus points).

Let (x0 , y0 ) be a real finite critical point of the function f (X, Y ). If at this point occurs
that !
∂2f ∂2f
∂X 2 (x 0 , y 0 ) ∂X∂Y
(x 0 , y0 )
D(x0 , y0 ) = det ∂2f ∂2f >0 (3.4)
∂Y ∂X
(x 0 , y0 ) ∂Y 2
(x0 , y0 )
then it is called a plus point. If it happens that D(x0 , y0 ) < 0, then it is called a minus
point.

Definition 3.10. (Degenerate points).

Let f ∈ R[X, Y ] be a real polynomial and (x0 , y0 ) be a critical point of f , then (x0 , y0 )
is called a degenerate point if D(x0 , y0 ) = 0. Analogously, (x0 , y0 ) is called a non-
degenerate point if D(x0 , y0 ) 6= 0.

Remark. By lemma 3.7, it is possible to make a slight perturbation in the coefficients of

the polynomial f ∈ R[X, Y ] without changing its topology. It is clear that this perturbation
will affect also to the second derivatives, so it can be always assumed that f ∈ R[X, Y ] is
a polynomial without any degenerate points.

Note. By theorem 3.5 when there is no critical value between [b, c], the topology of Lb
and Lc remains equal. Next lemma 3 will explain for a real polynomial of even degree
how the topology changes when one of this critical values are being actually crossed. By
the reasoning from the last remark it will be consider only plus and minus points, i.e.
non-degenerate points.

25
Chapter 3. Petrovski inequality for the even degree

Lemma 3.11. (Lemma 3).

Let (x0 , y0 ) be a real finite critical point of the polynomial f ∈ R[X, Y ] of even degree.
Suppose that (x0 , y0 ) is a plus point, then when C varies from f (x0 , y0 ) +  to f (x0 , y0 ) − 
for  > 0, then the difference p − m between the number p of positive ovals of the curve
f (X, Y ) = C and the number m of negative ovals of this curve increases by 1.
If (x0 , y0 ) is a minus point, then when C varies from f (x0 , y0 ) +  to f (x0 , y0 ) −  the
difference p − m of the curve f (X, Y ) = C decreases by 1, except once when decreases 2.

Proof. Taken from [6]. If D(x0 , y0 ) > 0, then at the critical point (x0 , y0 ) there is a
maximum or a minimum of f (X, Y ).
Suppose that (x0 , y0 ) is a maximum. Moving C from f (x0 , y0 ) +  to f (x0 , y0 ) − 
it is really lowering the level curve from f (x0 , y0 ) +  to f (x0 , y0 ) − . Since (x0 , y0 ) is a
maximum, in a small enough neighborhood around (x0 , y0 ) there is nothing above f (x0 , y0 ),
i.e. Lf (x0 ,y0 )+ = ∅. Nevertheless when the level decreases down the maximun suddenly
an oval around (x0 , y0 ) appears as is seen in figure 3.2, and since when crossing this oval
towards the outside, the function f decreases, thus a positive oval appears and so p − m
increases by 1.

Lf (x0 ,y0 )+ Lf (x0 ,y0 ) Lf (x0 ,y0 )−

Figure 3.2: Lowering the level curve around a maximum of f .

Suppose now that (x0 , y0 ) is a minimum. Then lowering the level curve from f (x0 , y0 )+
to f (x0 , y0 ) −  it is making one oval dissaper around (x0 , y0 ). Since it was a minimum
this oval is negative because going towards the outside is actually increasing the function
f . Then again p − m increases by 1.

Lf (x0 ,y0 )+ Lf (x0 ,y0 ) Lf (x0 ,y0 )−

Figure 3.3: Lowering the level curve around a minimun of f .

If D(x0 , y0 ) < 0 then there is a saddle point of the function f (X, Y ) in (x0 , y0 ). Then
there are 2 possibilities, that an oval touches another oval or that an oval touches itself.
Suppose first that a positive oval touches another positive oval in Lf (x0 ,y0 ) . Then
lowering the level curve from f (x0 , y0 ) +  to f (x0 , y0 ) −  passes from 2 separate positive
ovals to 1 positive oval created out of the union of the last 2, since the positive domain
expands as shown at figure 3.4.

26
 Lf (x0 ,y0 )−
Lf (x0 ,y0 )+ Lf (x0 ,y0 )

Figure 3.4: Lowering the level curve around two maximums of f .

So a positive oval disapears and p − m decreases 1. When a negative oval touches itself,
the drawing is the same but inverting the sign of the oval and inverting the direction of the
arrows, since lowering the level makes the negative domain smaller, it shrinks the negative
ovals towards themselves. Thus it is created 2 negative ovals from a negative oval, in
conclusion, p − m decreases 1 too.
Now consider the case when a positive oval touches a negative oval. Since all the most
outer ovals are positive (up to a change of sign in f ) it can only happen when a negative
oval is inside a positive oval. Lowering the level curve from f (x0 , y0 ) +  to f (x0 , y0 ) −  the
positive oval will get bigger since there would be more domain around it that is positive.

Lf (x0 ,y0 )+ Lf (x0 ,y0 ) Lf (x0 ,y0 )−

Figure 3.5: Lowering the level curve around a saddle point f .

Then it will be creating a new negative oval as in figure 3.5, so p − m decreases 1. If

we interchange the papers of the signs in the drawing, the arrows will invert the direction,
and so a positive oval dissapears, thus p − m decreases 1. A positive oval touching itself
is analogous.
Now as last case, an infinite oval touches itself, and since it is an infinite oval, this
case can only happens once. This occurs when a critical point looks locally as the saddle
point of a horse seat (see figure 3.6, taken from https://www.wikidata.org/wiki/Q357268).
Lowering the level curve, the positive oval (red one) becomes a negative oval (in black).
So a positive oval becomes a negative oval, and p − m decreases 2.

Figure 3.6: Saddle point.

27
Chapter 3. Petrovski inequality for the even degree

Lemma 3.12. (Lemma 4). Suppose that the real curve of even degree f (X, Y ) = C meets
the line at infinity in k different points (k does not depend on C since it relies on the
homogeneization) and s.t. all critical points of f (X, Y ) are finite and different; then the
difference between the number of minus points (−p) and the number of plus points (+p)
of the function f , is #(−p) − #(+p) = k − 1.

Proof. Taken from [6]. Denote by CM the maximal critical value that reaches f (X, Y ) and
Cm the minimal critical value. Consider the projective plane seen as the upper hemisphere
S2+ which its equator is the line of infinity. Let MC be the set of all points of the hemisphere
which corresponds to the points (x, y) of the plane for which f (x, y) > C. If k ≥ 1 then
when C > CM , thes set MC consist of k open regions G1 , G2 , . . . , Gk between two different
points where the curve f (X, Y ) = C meets the real line of infinity.

Figure 3.7: Projective real plane semisphere with regions.

When C < Cm the set MC consists already in all the semisphere region which has all
been formed around the point where f takes the minimal critical value Cm , because f is
a continuous function. Hence when C varies from C > CM to C < Cm all the k regions
G1 , . . . , Gk come together. By lemma 3.11, every 2 different regions Gi come together
when C passes through a minus point, thus this gives precisely k − 1 minus points.
Notice that every new positive oval will arise in a plus point and it will disappear in
a minus point, and every new negative oval will arise in a minus point and disappear in
a plus point by lemma 3.11. Perhaps, while C varies from CM +  to Cm −  new ovals
may arise, but also subsequently disappear in order to arrive to the final whole semisphere
region, thus all this does not affect to the difference #(−p) − #(+p) which remains k − 1.
If k = 0, then either the curve f (X, Y ) = C when C > CM consist of a single negative
oval (if there would be more than one, they would touch at some point creating another
higher critical value, and CM is the highest) and when C < Cm the curve is imaginary
containing no real points (as example take f (X, Y ) = X 2 + Y 2 ), or conversely, it can
happen also the opposite, for C > CM the curve is imaginary and when C < Cm the
curve consist of a single positive oval (as example take f (X, Y ) = −X 2 − Y 2 ). It can
not happen that both when C < Cm and C > CM the curve is imaginary or that when
C < Cm the curve f (X, Y ) = C consist of a single positive oval and when C > CM the
curve f (X, Y ) = C consist of a single negative oval, because f has only one height for
each (x, y) ∈ R2 .
Since in the first case the negative oval is disappearing and in the second case the
positive oval is appearing, in both cases there must exist a plus point where the negative
oval vanish and the positive oval arises respectively. Thus the difference in both cases is
−1 because all other ovals appearing and disappearing from CM +  to Cm − , give rise
to equal number of plus and minus points, so the difference remains −1.

28
Lemma 3.13. (Lemma 5).
Let f1 (X, Y ) and f2 (X, Y ) be 2 real polynomials of degree n vanishing simultaneously
at exactly n2 different finite points, and f (X, Y ) ∈ R[X, Y ] a polynomial of degree l < n
in R[X, Y ] not identically zero. Then f (X, Y ) cannot vanish in more than n · l points at
which f1 (X, Y ) and f2 (X, Y ) vanish simultaneously.
Proof. Taken from [6]. The polynomials f1 and f2 have no common factor because they
vanish simultaneosly at a discrete set of points. Denote by M1 (X, Y ) the greatest common
factor of f1 and f , and by M2 (X, Y ) the greatest common factor of f2 and f and let the
degrees of M1 (X, Y ) and M2 (X, Y ) respectively n1 and n2 (notice this number could be
both 0). Since f1 and f2 have no common factor, it can be written:

f1 (X, Y ) = M1 (X, Y ) · M 1 (X, Y )

f2 (X, Y ) = M2 (X, Y ) · M 2 (X, Y )
f (X, Y ) = M1 (X, Y ) · M2 (X, Y ) · M (X, Y )

where M 1 (X, Y ), M 2 (X, Y ), M (X, Y ) ∈ R[X, Y ]. The functions f1 ,f2 and f can only
vanish at a finite number of points which are the solutions of at least one of the following
systems

M1 (X, Y ) = 0, f2 (X, Y ) = 0
M2 (X, Y ) = 0, f1 (X, Y ) = 0
M (X, Y ) = 0, f1 (X, Y ) = 0 (or f2 (X, Y ) = 0).

All the left members of each of these systems of equations are relatively prime with their
right members. Therefore the first system has at most n1 · n solutions, the second at most
n2 · n and the third at most (l − n1 − n2 ) · n. The sum of these numbers is l · n < n2 .

Lemma 3.14. (Lemma 6).

Let A be a fixed complex number different from 0 and g(X, Y ) ∈ R[X, Y ]. Then the
condition that the real part of the product A · g(X, Y )2 = 0 at a given point (x0 , y0 ), real or
complex s.t. g(x0 , y0 ) non-zero real part, can always be expressed in the form of a linear
homogeneous equation, with real coefficients, precisely in the coefficients of g(x0 , y0 ).
Proof. Taken from [6]. Let A = a + bi and g(x0 , y0 ) = c + di where a, b, c, d ∈ R and c 6= 0.
Then A · g(x0 , y0 )2 = (a + bi)(c + di)2 and the real part of A · g(x0 , y0 )2 is precisely

ac2 − ad2 − 2bcd = c2 (a − 2bx − ax2 )

where x = dc .
The equation ax2 + 2bx − a = 0 has for any real numbers a, b ∈ R at least one real
finite root, since its discriminant is 4(b2 + a2 ) and therefore non-negative. Denote this root
by x0 . Then d = x0 c implies that the real part of A · g(x0 , y0 )2 is 0, and since the equation
d = x0 c is a linear homogeneous equation with real coefficients from g(x0 , y0 ), the proof is
finished.

29
Chapter 3. Petrovski inequality for the even degree

Theorem 3.15. (Petrovski inequalities).

Denoting by p the number of positive ovals of a non-singular real curve f (X, Y ) = 0
of even degree n and by m the number of negative ovals. Then

3n2 − 6n 3n2 − 6n
− −δ ≤p−m≤ +1−δ
8 8
where δ = 0 if the outer ovals are positive and δ = 1 if the outer ovals are negative.
Proof. This is taken from [6] and [13]. By lemma 3.8 it can always be assumed that the
following system
∂f ∂f
= 0, =0 (3.5)
∂X ∂Y
posseses (n − 1)2 different finite solutions (xi , yi ) for i = 1, . . . , (n − 1)2 . In this case the
following theorem of Euler-Jacobi holds (see lemma 3.16):
(n−1)2
X P (xi , yi )
= 0; (3.6)
i=1
D(x i , yi )

∂2f ∂2f
∂X 2
(x0 , y0 ) (x0 , y0 )
where D(x0 , y0 ) = ∂2f
∂X∂Y
∂2f and P (X, Y ) ∈ R[X, Y ] is an arbitrary
∂Y ∂X
(x0 , y0 ) ∂Y 2
(x0 , y0 )
polynomial of degree strictly lower than (n − 1) + (n − 1) − 2 = 2n − 4.
Again by lemma 3.8 the system (3.5) is not having multiple solutions, and thus the
denominator is never zero of any member of the sum, so is well-defined.
In particular, the equation (3.6) holds taking P (X, Y ) = F (X, Y ) · g(X, Y )2 where
g(X, Y ) is an arbitrary polynomial of degree 12 (n − 4) and

∂f ∂f
F (X, Y ) = nf (X, Y ) − X (X, Y ) − Y (X, Y )
∂X ∂Y
which is at most of degree n − 1 by its form. Notice that since f is non-singular, there is
no solution of the system (3.5) that satisfies f (X, Y ) = 0 also.
The polynomial g(X, Y ) can be chosen with real coefficients and s.t. vanishes in
1
(n − 4) + 3 12 (n − 4)



2 (n − 4 + 6)(n − 4) n2 − 2n − 8 n(n − 2)
= = = − 1 (3.7)
2 8 8 8
arbitrarly chosen critical points of f (X, Y ) using corollary 1.23.
Lemma 3.13 tells that g(X, Y ) can not vanish in all critical points of f (X, Y ), since it
is bounded by
1 n2 − 5n + 4
(n − 1) · (n − 4) = , n ∈ N, (3.8)
2 2
2
which is smaller than n2 − 2n + 2 and thus is smaller than n2 − 2n + 1 = (n − 1)2 , n ∈ N.
After these preliminary constructions of the auxiliary polynomials, it can be introduced
the kernel of the proof. Denote by k the number of points in which the curve f (X, Y ) = C
meets the line of infinity. If k > 0 and C > CM the curve consists of 21 k positive infinite
ovals. When C decreases from CM +  to 0 the curve obtains:

30
 k
p+α− new positive ovals,
2
m + β new negative ovals,
and loses α positive ovals and β negative ovals. While C passes through the critical values
of f (X, Y ) there are
k
p+α+β− plus-points
2
m+α+β−δ minus-points
since every positive oval arises in a plus point and every negative is erased in a plus point
too and every negative oval is created in a minus point and every positive oval is erased
in a minus point. Also δ = 0 if the outer ovals are positive and δ = 1 if the outer ovals
are negative, since it means that it have crossed a saddle point (see lemma 3.11) where an
infinite positive oval touches itself.
This reasoning also holds when the number of plus points and minus points are being
counted for k = 0. Recall the case k = 0 from lemma 3.12. When there was at C > CM
an imaginary curve, it is δ = 0, and lowering from CM it appears an outer oval which is
positive, thus all the outer ovals are positive. When there was at C > CM a negative oval,
all the outers ovals are negative, and δ = 1 since we are counting an extra minus point
taking into acount this negative oval at C > CM (without a starting minus point).
Then this far it has been seen that f (X, Y ) > 0 at
k
p+α+β− plus-points and at
2
m+α+β−δ minus-points of f.
Is clear since f (X, Y ) is non-singular, no plus points or minus points are counted while
f (X, Y ) = 0, so this counting holds true too for f (X, Y ) ≥ 0. By lemma 3.12 it is sattisfied
that
#(−p) − #(+p) = (#(−p)≥ + #(−p)< ) − (#(+p)≥ + #(+p)< ) = k − 1 (3.9)
where #(−p)≥ and #(−p)< denotes the number of minus points that sattisfies respectively
f (X, Y ) ≥ 0 and f (X, Y ) < 0. And analogously it is defined #(−p)≥ and #(−p)<
changing minus points for plus points in the last sentence. Recall that in 3.9 it was
defined plus and minus points as real critical points of the function f (X, Y ). Thus
#(−p) + #(+p) = (n − 1)2 − 2γ (3.10)
where γ denotes the number of complex and non real solutions of the system (3.5). Adding
(3.9) and (3.10) it is obtained
2#(−p) = 2#(−p)≥ + 2#(−p)< = (n − 1)2 − 2γ + k − 1
2#(−p)< = −2 (m + α + β − δ) + (n − 1)2 − 2γ + k − 1
(n − 1)2 + k − 1
#(−p)< = − γ − m − α − β + δ.
2

31
Chapter 3. Petrovski inequality for the even degree

Now substracting (3.9) to (3.10)

2#(+p) = 2#(+p)≥ + 2#(+p)< = (n − 1)2 − 2γ − k + 1

 
2 k
2#(+p)< = (n − 1) − 2γ − k + 1 − 2 p + α + β −
2
2
(n − 1) + 1
#(+p)< = − γ − p − α − β.
2
f (xi ,yi )
Notice that at (xi , yi ) ∈ (+p)> and (xi , yi ) ∈ (−p)< points it holds that D(x i ,yi )
>0
with D(xi , yi ) defined as at the beginning of the proof. Analogously for (xi , yi ) ∈ (−p)>
f (xi ,yi )
and (xi , yi ) ∈ (+p)< points it holds D(x i ,yi )
< 0. It will be denoted due to this as A-points
to the points of (+p)> ∪ (−p)< and B-points to the points of (−p)> ∪ (+p)< . There are

#(A − points) = #(+p)> + #(−p)< = (3.11)

k (n − 1)2 + k − 1
p+α+β− + −γ−m−α−β+δ = (3.12)
2 2
(n − 1)2 − 1
p−m+δ−γ+ (3.13)
2
A-points and

#(B − points) = #(−p)> + #(+p)< = (3.14)

(n − 1)2 + 1
m+α+β−δ+ −γ−p−α−β = (3.15)
2
(n − 1)2 + 1
m−p−δ−γ+ (3.16)
2
B-points.
Now in order to see that
n(n − 2) n(n − 2)
#(A − points) ≥ −γ and #(B − points) ≥ −γ
8 8
suppose the opposite, that

n(n − 2) n(n − 2)
#(A − points) < −γ and #(B − points) < − γ. (3.17)
8 8

If #(A − points) < n(n−2) 8

− γ choose 18 n(n − 2) coefficients of the polynomial g(X, Y )
s.t. it vanishes in all the A-points (this last thing can be done by (3.7)). Also choose
P (xi ,yi )
this 81 n(n − 2) coefficients s.t. the real parts of the components D(x i ,yi )
of the sum (3.6)
corresponding to imaginary solutions of the system (3.5) vanish also, this can be done by
lemma 3.14. The polynomial g(X, Y ) differs from 0 in at least one real critical point of
f (X, Y ) because using 3.17 it is deduced that

n(n − 2)
γ< ,
8
32
since this is the most extreme case that we could obtain from #(A − points) < n(n−2)
8
−γ
i.e. making #(A − points) = 0. Then the number of real critical points of f (X, Y ) which
is (n − 1)2 − 2γ exceeds the number
n(n − 2) 3 3
(n − 1)2 − 2 · + 1 = n2 − n + 2.
8 4 2
The number of critical points of f (X, Y ) where g(X, Y ) vanishes can not exceed by (3.8)
(n − 1)(n − 4) n2 5n
= − + 2,
2 2 2
thus f (X, Y ) has at least one real critical point where g(X, Y ) does not vanishes since
n2
2
− 5n
2
+ 2 < 34 n2 − 23 n + 2. Recall that
(n−1)2 2
X P (xi , yi ) (n−1)
X F (xi , yi ) · g(xi , yi )2
0= = ;
i=1
D(xi , yi ) i=1
D(xi , yi )
∂f ∂f
since F (X, Y ) = nf (X, Y ) − X ∂X (X, Y ) − Y ∂Y (X, Y ), in a real critical point (xj , yj ) this
last expression is not zero, since f is not singular and there is one real critical A-point,
lets say (xj , yj ), where g(X, Y ) does not vanishes and F (xj , yj ) = nf (xj , yj )
(n−1)2 2
X P (xi , yi ) (n−1)
X nf (xi , yi )
0= = · g(xi , yi )2 > 0;
i=1
D(xi , yi ) i=1
D(xi , yi )

thus it is a contradiction. Then

n(n − 2)
#(A − points) ≥ − γ. (3.18)
8
n(n−2)
Analogously, by a similar reasoning it can be seen that if #(B − points) < 8
− γ, then
(n−1)2 2
X P (xi , yi ) (n−1)
X nf (xi , yi )
0= = · g(xi , yi )2 < 0;
i=1
D(xi , yi ) i=1
D(xi , yi )

which is another contradiction, so

n(n − 2)
#(B − points) ≥ − γ. (3.19)
8
From (3.18) and (3.11), it is obtained that
(n − 1)2 − 1 n(n − 2)
p−m+δ−γ+ ≥ −γ
2 8
n(n − 2) (n − 1)2 − 1
p−m≥ − −δ
8 2
n2 − 2n n2 − 2n
p−m≥ − −δ
8 2
−3n2 + 6n
p−m≥ − δ.
8

33
Chapter 3. Petrovski inequality for the even degree

From (3.19) and (3.14), it is obtained that

(n − 1)2 + 1 n(n − 2)
m−p−δ−γ+ ≥ −γ
2 8
−(n − 1)2 − 1 −n(n − 2)
p−m+δ+ ≤
2 8
−n2 + 2n 4n2 − 8n
p−m≤ + +1−δ
8 8
3n2 − 6n
p−m≤ + 1 − δ.
8
Thus
3n2 − 6n 3n2 − 6n
− −δ ≤p−m≤ + 1 − δ. (3.20)
8 8

Lemma 3.16. (Jacobi-Euler Theorem).

Let f, g ∈ R[X, Y ] with n =deg(f ) ∈ N and m =deg(g) ∈ N. If Γ1 = VR (f ) and
Γ2 = VR (g). If Γ1 and Γ2 have no common component and no multiple intersection points,
then for any P ∈ R[X, Y ] with deg(P ) < n + m − 2 the following holds
X P (x, y)
= 0, (3.21)
J(x, y)
(x,y)∈Γ1 ∩Γ2

where J is the Jacobi determinant of the function F = (f, g) : R2 −→ R2 .

Corollary 3.17. (Restriction on M-curves of even degree 2k ≥ 6).

There can not be an M-curve of degree n = 2k ≥ 6 with all its ovals lying outside each
other.

Proof. Taken from [6]. Suppose that all the outer ovals are positive

12k 2 − 12k (2k − 1)(8k − 8) 12k 2 − 12k

 
(2k − 1)(2k − 2)
+1− +1 = −
2 8 8 8

where the right hand side is

16k 2 − 24k + 8 −12k 2 + 12k 4k 2 − 12k + 8 k 2 − 3k + 2 (k − 1)(k − 2)

+ = = = ≥ 0,
8 8 8 2 2
thus for k > 2, k ∈ N i.e. 2k ≥ 6 the last inequality is strict, which by the last theorem
3.15 is a contradiction.
In particular, for degree 6 it can not happen that the 11 ovals lie outside each other.
This shows that for higher degrees, M-curves do not have this particular simple topology
case like at the lower degree cases.

34
 CHAPTER 4

Real algebraic curves from a complex point of view

Now it will be seeked to find more information about the real plane algebraic curves
using the fact that it is a subset of a complex plane algebraic curve defined by the same
polynomial but taking also the complex roots into account. This complex curve defines
a subset of the complex plane C2 , and remember this last space has complex dimension
2 and real dimension 4. Notice that we have 2 real equations in a complex polynomial
equation:

f (x, y) = 0, x, y ∈ C,
Re(f (x, y)) = 0,
Im(f (x, y)) = 0.
for 4 real variables, since x = a + bi and y = c + di for a, b, c, d ∈ R.
Proposition 4.1. Let f ∈ R[X, Y ] be a polynomial and let α be a complex root of this
polynomial f . Then α is also a root of f .
Proof. Suppose that f (X, Y ) = an,0 Y n +an−1,1 Y n−1 X +an−1,0 Y n−1 +. . .+a0,0 , then letting
α = (α1 , α2 ) ∈ VC (f ) ⊂ C2 :
f (α1 , α2 ) = an,0 α2n + an−1,1 α2n−1 α1 + an−1,0 α2n−1 + . . . + a0,0 = 0.
Now for α = (α1 , α2 ) and using that ai,j ∈ R for every i, j = 1, 2, . . . , n and the conjugate
property (z + w) = z + w:
f (α1 , α2 ) = an,0 α2 n + an−1,1 α2 n−1 α1 + an−1,0 α2 n−1 + . . . + a0,0
= an,0 α2 n + an−1,1 α2 n−1 α1 + an−1,0 α2 n−1 + . . . + a0,0
= an,0 α2n + an−1,1 α2n−1 α1 + an−1,0 α2n−1 + . . . + a0,0
= f (α1 , α2 ) = 0.

35
Chapter 4. Real algebraic curves from a complex point of view

The real algebraic curve Γ lying inside the real projective plane RP2 remains unchanged
under the conjugation σ : CP2 −→ CP2 , σ(x : y : z) 7→ (x : y : z). Due to proposition
4.1, if the real projective plane curve divides the complex algebraic curve, then it divides
in equal halves. Figure 4.1 is just a fake sketch to understand better what is happening,
since CP2 has real dimension 4.

RP2
Γ

Figure 4.1: The real projective plane dividing a complex algebraic curve (visual sketch).

Since a complex projective curve is connected, the number of parts into which the
curves fixed by conjugation divides the non-singular complex curve is at most two.
Definition 4.2. It is said that a real curve VR (F ) is of type I if the curve splits the
complex curve VC (F ) into 2 halves. A real curve VR (F ) is of type II if the complement
VC (F ) \ VR (F ) is connected, i.e. the curve VR (F ) is not dividing VC (F ).
Example. A circle, VR (f ) for f (X, Y ) = X 2 + Y 2 − 1 is a real plane curve of type I.
Curves of type I have an additional structure that comes from the complex domain.
Since the real part of a curve of type I divides the complex curve in 2 halves, each of the
halves has the orientation defined by the complex structure and defines an orientation on
the real curve, which is the common boundary of each half. The set of real points get 2
exactly opposite orientations that is named complex orientation of the real curve, one
from the upper part and the other from the lower part. When an orientation is chosen for
one of the ovals, it determines the orientations of all the other ovals.
Definition 4.3. (Positive and negative injective pair).
Any pair of ovals s.t. one surrounds the other is called an injective pair. An injective
pair of oriented ovals is negative if the ovals are both oriented clockwise or counterclock-
wise. Otherwise is said to be positive. Please see figure 4.
The number of positive injective pairs of ovals is denoted by Π+ and the number of
negative injective pairs by Π− .

Negative pair Positive pair

36
 Since the real curve is precisely the set of fixed points under the complex conjugation, it
will be interesting to study more properties about the conjugation map σ. Thinking in this
map together with the map composition, it turns out to have the structure of an abelian
group, furthermore, there is a natural group isomorphism between {id, σ} = {σ 2 , σ} and
the finite additive group Z2 .
Definition 4.4. (Euler characteristic).
Let X be a topological space, then the Euler characteristic of X is defined as
∞
X X
X (X) = (−1)n dim(Hn (X)) = (−1)n card(CnX )
n=0 n=0

where Hn denotes the n-homology group (see appendix C) and card(CnX ) the number of
n-cells of X. The dimension of the n-homology group gives the number of Z that appear
at the direct sum of the n-homology group (the Z-rank of the n-homology group).
Definition 4.5. (Even and odd ovals).
An oval of a real curve Γ is said to be odd if its depth is even. Analogously an oval is
said to be even if its depth is odd. It is denoted by p the number of even ovals and by n
the number of odd ovals of Γ.
Notice that p is the number of positive ovals and n is the number of negative ovals
if all the outermost ovals are positive.
Definition 4.6. Let F ∈ R[X, Y, Z] be a homogeneous polynomial, then it is defined
B+ = (x : y : z) ∈ RP2 | F (x : y : z) ≥ 0 .


Proposition 4.7. Let p and n denote respectively the number of even and odd ovals of a
non-singular real curve of even degree Γ and X denote the Euler characteristic. Then
X (B+ ) = p − n.
Proof. Please see C. Using that
X X
X (B+ ) = (−1)n dim(Hn (B+ )) = (−1)n card(CnB+ ),
n=0 n=0

let D be the disk in R2 , that is one 2-cell with one 1-cell and one 0-cell, then is clear
that X (D) = 1 − 1 + 1 = 1 since dim(H0 (D)) = dim(H1 (D)) = dim(H2 (D)) = 1 because
H0 (D) ∼= H1 (D) ∼ = H2 (D) ∼= Z. With celullar homology is easy to compute it, there is
one 0-cell (a vertex v), one 1-cell (an edge e) and one 2-cell (U ) as shown in figure 4.2

v
e
U

Figure 4.2: Cellular decomposition of a disk.

37
Chapter 4. Real algebraic curves from a complex point of view

Now lets take a disk with a hole in it, D0 . Then H0 (D0 ) = Z since it is path connected.
By Hurewicz’s theorem C.20 (from C), we deduce that H1 (D0 ) ∼ = H1 (S1 ) ∼
= π1 (S1 ) ∼
= Z,
1
since a punctured disk has the same homotopy type as a circle S . There is no 2-cells on a
circle, so H2 (D0 ) ∼
= H2 (S1 ) ∼
= 0 again by the homotopy invariance of the homology groups.
0
Thus X (D ) = 1 − 1 + 0 = 0.
Define A∨B as the union by a point between A and B. A disk with n holes D0n is of the
same homotopy type as S1 ∨ . . . ∨ S1 which is path-connected, then H0 (S1 ∨ . . . ∨ S1 ) ∼ = Z.
For any family of topological spaces Xi for i = 1, . . . , p it holds:

Hn (X1 ∨ . . . ∨ Xp ) ∼
= Hn (X1 ) ⊕ . . . ⊕ Hn (Xp ),

then H1 (S1 ∨ . . . ∨ S1 ) ∼
= H1 (S1 ) ⊕ . . . ⊕ H1 (S1 ) ∼
= Z ⊕ . . . ⊕ Z = Zn and also it is obtained
that H2 (S1 ∨ . . . ∨ S1 ) ∼
= H2 (S1 ) ⊕ . . . ⊕ H2 (S1 ) ∼
= 0, concluding

X (D0n ) = 1 − n + 0 = 1 − n. (4.1)

Using that for every n ∈ N:

M
Hn (X) = Hn (Xα ) (4.2)
α

where Xα are the path-components of X i.e. path-connected components. Then for any
space X with path-components D1 , . . . , Dk we have that

k
X
X (X) = X (Dα ). (4.3)
α=1

It can always be assumed (up to a sign change in F ∈ R[X, Y, Z]) that F is negative on
the region outside all the ovals of VR (F ) i.e. the outermost ovals are positive and even.
Assume that there is only one even/positive oval, then by the reasoning above with D
as a disk in R2 , X (B+ ) = X (D) = 1. If there is an odd oval inside an even oval, then B+
is homeomorphic to a disk with a hole, then X (B+ ) = X (D0 ) = 0.
If there are 2k + 1 ovals in a nest (k ∈ Z), the deepest oval will be even and since
the rest would be unions of k pairs of one odd oval inside one even oval that have Euler
characteristic 0, by (4.3) it is deduced that it is like the case when there is only one even
oval case, i.e. X (B+ ) = 1 = k + 1 − k = p − n since B+ is made out of k − 1 annulus (of one
even oval and one odd oval inside) and one oval that are different connected components.
Now if there is a nest of order 2k, then the deeper oval is odd and the nest is made
out of k rings that are precisely pairs of an even oval with one odd oval inside, since this
rings have Euler characteristic 0, it is deduced by (4.3) that X (B+ ) = 0 = k − k = p − n.
Perhaps this can be better understood with the following picture with the cases for a
nest of order 4 (left) and a nest of order 3 (right) with the domain B+ colored in red.

38
 Figure 4.3: The domain B+ in the projective plane in red. In this figure p − n = 1.

The last case would be to study what happens with the characteristic when we have
more than one oval inside another oval. Suppose we have an even oval with 2 odd ovals
lying inside. This is equivalent to a disk with 2 holes on it, thus by the reasoning of the
beginning (4.1) X (B+ ) = 1 − 2 = p − n. In general, if there is an even oval with n odd
different ovals lying in its interior, then again by (4.1), X (B+ ) = 1 − n = p − n.
Any other even and odd ovals distribution that may arise are combinations of all the
cases it have been already studied with different connected components configurations
1 m
B+ , . . . , B+ , but since all our primarly cases sattisfied X = p − n, the sum (4.3) of the
different connected components would also sattisfy in general
1 m
X (B+ ) = X (B+ ∪ . . . ∪ B+ ) = p − n.

Using this homological definition of the Euler characteristic it can be seen due to
homological properties that for a g-torus Mg (see [7]) the Euler characteristic is
X (Mg ) = 2 − 2g.
Proposition 4.8. Any non-singular M-curve of even degree is of type I.
Proof. This proof is based on [15]. Let Γ be a complex non-singular M-curve of degree
d, thus of genus g = (d−1)(d−2)
2
(this result is explained in [5]). Then RΓ is the union of
(d−1)(d−2)
g+1 = 2
+ 1 disjoint ovals lying in RP2 . Since Γ is homeomorphic to a sphere
with g = (d−1)(d−2)
2
handles ([5]) then that many disjoint ovals of the M-curve g + 1, make
each oval necessarily divide each handle of the sphere. This can be seen in 4.4:

RΓ
1 1 2 2 3 3 g g+1
Γ

Figure 4.4: Counting ovals in a sphere with g handles.

39
Chapter 4. Real algebraic curves from a complex point of view

For a nested ovals case, using that the non-singular curve Γ is homeomorphic to a
g-torus, its Euler characteristic is 2 − 2g = 2 − (d − 1)(d − 2). Now cutting Γ along RΓ,
since the Euler characteristic behaves like a measure, it is clear that RΓ has measure zero
with respect to Γ, thus the Euler characteristic under this cutting remains equal.
Now cap every boundary circle with a disk. Each component of RΓ is giving now rise to
two boundary disks, thus the number of the boundary circles is 2(g +1) = (d−1)(d−2)+2.
The Euler characteristic is additive under the union, thus the Euler characteristic of
the surface is
2 − (d − 1)(d − 2) + (d − 1)(d − 2) + 2 = 4,
using X (D) = 1 for every D being a disk. Any closed connected surface is homeomorphic
to a sphere with g handles for some g ≥ 0 which has Euler characteristic 2 − 2g ≤ 2 < 4.
Since there is no connected closed surface with Euler characteristic 4, RΓ must divide Γ.
Thus RΓ is a dividing type curve and Γ is a curve of type I.

Let σ : CP2 −→ CP2 denote the complex conjugation. If Γ is a real curve of type I, its
complexification CΓ is divided by Γ into 2 components CΓ± that are interchanged by the
complex conjugation σ. The orientation of CΓ+ orients the ovals of its boundary and it
induces an orientation between 2 related ovals in a common nest. Let
 
±1 if the orientations of the ith and j th ovals agree or disagree
εij = .
0 if the orientations of the ith and j th ovals cannot be compared

It will be said that 2 different ovals can not be compared when they are not in the
same oval nest.

Lemma 4.9. Let Γ be a real curve of even degree 2k with l ovals of type I, then
X
2 εij = k 2 − l.
1≤i<j≤l

Proof. The idea of this proof is taken from [14]. See appendix C. Let D = li=1 Di be the
S
disjoint union on RP2 made up of the discs bounded by the l ovals of Γ. Visualizing it in
figure 4.5 for the upper case (the lower case is analogous) in the case where every oval lies
outside every other oval:

CΓ+

Γ
D1 D2 D3 Dl

Figure 4.5: Capping with l discs half of type I complex curve CΓ.

40
 Lets define
E+ = CΓ+ ∪ D+ and E− = CΓ− ∪ D−
where D+ is D together with the orientation of CΓ+ and D− is D together with the
orientation of CΓ− .
Let σ : CP2 −→ CP2 denote the complex conjugation and H2 (CP2 ; Z) ' Z the second
homology group with integer coefficients, and lets denote the homology conjugation map
as σ∗ : H2 (CP2 ; Z) −→ H2 (CP2 ; Z).
Denote now by [E+ ] and [E− ] respectively the homology clases in H2 (CP2 ; Z) of E+
and E− . The conjugation takes one to another with a minus sign due to their respectively
opposite orientation σ∗ ([E+ ]) = −[E− ], now

[E+ ] ⊕ [E− ] = [E+ ∪ E− ] = [CΓ] = 2k[L] (4.4)

where [L] is the homological class of a line and CΓ being the complexification Γ. Thus by
(4.4) it has to be that [E+ ] = k[L0 ] and [E− ] = k[L00 ] since the curve is of type I and the
real projective plane is dividing the upper and the lower part in 2 equal halves. Then

[E+ ] • [E− ] = k 2 (4.5)

where • denotes the number of intersections in the complex projective plane, which is
exactly k 2 by Bezout’s theorem 1.7.
Now, it will be counted the intersection [E+ ]•[E− ] using properties of the 2 homological
classes [E+ ] and [E− ] of the second homology group H2 (CP2 ; Z).
l
X l
X
[E+ ] • [E− ] = [E+ ] • (−σ∗ [E+ ]) = εij = l + 2 εij . (4.6)
i,j=1 i<j

This count comes out taking into account the order or the orientation of the disks
that are in [E+ ] and in −σ∗ [E+ ]. Every disk of [E+ ] intersects with its own projection in
−σ∗ [E+ ] counting +1. In the case of having a nest with one disk of [E+ ] having opposite
orientation with another disk of the nest, this opposite orientation will be the same in
−σ∗ [E+ ], and it can be changed at the cost of a minus sign making it the same homolgy
class, in order to count this intersection as a +1. Taking the sign into account −(+1) = −1.
When one disk of the nest in [E+ ] shares the same orientation with another disk of its
nest, the projection in −σ∗ [E+ ] shares this orientation and it is counted this intersection
as +1. This last 2 cases can be both visualised in 4.6.

Di+ Dk+
Dj− −
Dm

Figure 4.6: Two nests with 2 discs with opposite (left) and same (right) orientation.

41
Chapter 4. Real algebraic curves from a complex point of view

Then the count from (4.6) holds. Equaling now the sums of [E+ ] • [E− ] from (4.6) and
(4.5) the result is finally obtained.

Corollary 4.10. (Complexity of a nest). Result from [14]. P

The sum of the complexities of all the possible nests (i.e. i<j |εij |) in a real non-
singular algebraic curve of type I and even degree 2k with l ovals sattisfies the following
inequality
k2 − l X
≤ |εij |
2 i<j

Proof. Using lemma 4.9

k2 − l X X
= εij ≤ |εij |.
2 i<j i<j

Theorem 4.11. (Rokhlin complex orientation formula).

For any curve of type I and degree m = 2k with l ovals, then

2(Π+ − Π− ) = l − k 2 .

Proof. By the definition of Π+ , Π− and εij it holds that

l
X
Π− − Π+ = εij
i<j

then
l
X
+ −
2(Π − Π ) = −2 εij .
i<j

Using lemma 4.9, 2(Π+ − Π− ) = l − k 2 .

Corollary 4.12. (Gudkov congruence mod 4).

Let p and n denote respectively the number of positive and negative ovals of a non-
singular real algebraic curve Γ of type I and of even degree 2k, then

p − n ≡ k2 mod 4.

Proof. Let p and n denote respectively the number of positive ovals and negative ovals.
Since l is the number of total ovals l = p + n. By the definition of Π+ and Π− it holds
that
Π+ + Π− ≡ n mod 2, (4.7)
from where it can be deduced that

Π+ − Π− ≡ n mod 2. (4.8)

42
By the Rokhlin complex orientation formula from theorem 4.11 and the equation of (4.8),
it holds that
l − k2
≡ n mod 2
2
l − k 2 ≡ 2n mod 4
p + n ≡ k 2 + 2n mod 4
p − n ≡ k 2 mod 4.

Note. By the last results it was seen that an M-curve is of type I, so in particular it should
sattisfy for degree 2 · 3 = 6 that

X (B+ ) = p − n ≡ 9 mod 4 ≡ 1 mod 4.

Using all this information, the possible candidates that we have this far for an M-curve of
degree 6 are the following pairs of positive and negative ovals (p, n):

(10, 1) (2, 9) (6, 5) (8, 3) (4, 7). (4.9)

In fact, there is a stronger restriction for the M-curves of even degree 2k in particular,
named as the Gudkov-Rokhlin congruence stating that

X (B+ ) ≡ k 2 mod 8
which by proposition 4.7 is the same as p − n ≡ k 2 mod 8. Thus the latter list (4.9) is
reduced just to

(10, 1) (2, 9) (6, 5),

which are in fact the only M-curves for degree 6.

43
 CHAPTER 5

Smith’s inequality and Harnack’s formula generalization

Definition 5.1. It is said that K is a simplicial complex when K is a set whose

elements are called vertices, together with a collection of finite non-empty substets of the
set of vertices called simplices such that:
1. Every vertex is contained in some simplex.

2. Every non-empty subset of a simplex is a simplex.

Let v1 − v0 , . . . , vn − v0 be linearly independent vectors of a vector space V . The points
vi are vertices of the n-simplex [v0 , v1 , . . . , vn ].
A proper and non-empty subset of a simplex is called a face of that simplex. A
simplicial map from one simplicial complex to another is a map of the set of vertices
which carries simplices into simplices. We denote |K| the topological space of a simpicial
complex K together with the weak topology for K i.e. a subset of |K| it is closed if and
only if its intersection with each |s| is closed (using the usual topology at each |s|), being
|s| a simplex of K.
Definition 5.2. Let G be a finite group, then it is said that G acts simplicially on K
if each trasformation of K under the action of any member of G is a simplicial map. The
simplicial complex K, together with such an action is called a simplicial G-complex.
A simplicial action of G on K satisfying for g0 , g1 , . . . , gn ∈ G that [v0 , . . . , vn ] and
[g0 v0 , . . . , gn vn ] are both n-simplices s.t. there exists an element g ∈ G sattisfying g(vi ) =
gi (vi ) for every i it is said that this simplical action is regular. It can be seen that up to
homeomorphism at |K|, any simplical action G on K is regular, it will be assumed in the
sequel that the simplical action it is always regular (this result is in chapter 3 of [2]).
Definition 5.3. The vertices of a regular G-complex K, denoted by K/G, are orbits
v ∗ = Gv = { gv |g ∈ G} of the action of G on the vertices of K, and we take the simplices
of K/G to be
[v0∗ , . . . , vn∗ ].

44
 The simplex [v0 , . . . , vn ] of K is said to be over the simplex [v0∗ , . . . , vn∗ ]. Since K
is a regular G-complex it is deduced that K/G is a well defined simplicial complex.
Let [v0 , . . . , vn ] and [w0 , . . . , wn ] be simplices of K over the same simplex [v0∗ , . . . , vn∗ ] =
[w0∗ , . . . , wn∗ ] of K/G, then by regularity, there exists an element g ∈ G s.t. [w0 , . . . , wn ] =
g[v0 , . . . , vn ] = [gv0 , . . . , gvn ]. Thus all the simplices of K over a simplex of K/G form an
orbit of the action G on the simplices of K.
Definition 5.4. The complex K G is the subcomplex of K consisting of all simplices which
are pointwise fixed under G (i.e. every element of the group G).
An oriented n-simplex of K is a n-simplex s together with an equivalence class of
total orderings of the vertices of s. Two orderings are equivalent if they differ by an even
of permutations of the vertices.
If v0 , v1 , . . . , vn are the vertices of the simplex s, then [v0 , . . . , vn ] denotes the oriented
n-simplex of K consisting of the simplex s together with the equivalence class of the
ordering v0 < . . . < vn of its vertices represented by <.
Example. [v0 , v1 , v2 ] ≡ [v2 , v0 , v1 ] ≡ [v1 , v2 , v0 ].
Definition 5.5. Let Cn (K) be the free abelian group generated by the oriented
n-simplices sn sattisfying the relations

sn1 + sn2 = 0

whenever sn1 and sn2 are opposite oriented n-simplices corresponding to the same n-simplex
of K (for a definition of a free group, please see appendix C).
The rank of the free abelian group Cn (K) is equal to the number of n-simplices of K.
Define the homomorphisms ∂n : Cn (K) −→ Cn−1 (K) for every n ≥ 1 as follows:
n
X
∂n [v0 , v1 , . . . , vn ] = (−1)i [v0 , v1 , . . . , v̂i , . . . , vn ]
i=0

where [v0 , v1 , . . . , v̂i , . . . , vn ] denotes the oriented (n − 1)-simplex obtained by omitting vi .

If sn1 + sn2 = 0 in Cn (K) then it can be verified, since ∂n is a homomorphism:

∂n (sn1 ) + ∂n (sn2 ) = ∂n (sn1 + sn2 ) = ∂n (0) = 0.

In addition, by the definition of ∂n , it holds that ∂n ◦ ∂n+1 = 0 (see [7]).

Therefore there is a free non-negative chain complex C(K) = {Cn (K), ∂n } with
its correspondant oriented chain complex of K:
n ∂ 2 ∂n−1 1 ∂ ∂
· · · → Cn (K) −→ Cn−1 (K) −−−→ · · · −
→ C1 (K) −
→ C0 (K).
The n-th oriented homology group of K is

Hn (K) = Ker(∂n )/Im(∂n+1 ).

Since K is a simplicial G-complex, then C(K) inherits also the action of G putting

g[v0 , . . . , vn ] = [gv0 , . . . , gvn ].

45
Chapter 5. Smith’s inequality and Harnack’s formula generalization

Due to this, it is a module over the group ring ZG, whose elements are of the form:
X
ng · g,
g∈G

where all the ng are integers.

Definition 5.6. The norm σ ∈ ZG is defined to be the sum

X
σ= g
g∈G

that can act on any chain c ∈ C(K). The image σC(K) ⊂ C(K) of σ : C(K) −→ C(K)
is a subcomplex. As a map, σ is defined as
X
σ([v0 , v1 , · · · , vn ]) = [gv0 , gv1 , · · · , gvn ].
g∈G

Let L ⊂ K be a subcomplex which is invariant under the action of G, then G acts on

the chain C(K, L) = C(K)/C(L) and σC(K, L) is a subcomplex of C(K, L).
Restrict now the attention to a multiplicative group G of prime order p (G cyclic) and
to homology groups with coefficients in Z/pZ = Zp . Let g be a fixed generator of G, then
it is defined Zp G as the group ring which elements are of the form

x = a0 + a1 g + · · · + ap−1 g p−1 with a0 , a1 , · · · , ap−1 ∈ Zp . (5.1)

Notice that Zp G ∼
= Zp [g]/(g p − 1). For g ∈ G being a fixed generator of G lets define:

τ = 1 − g ∈ Zp G,
σ = 1 + g + g 2 + · · · + g p−1 ∈ Zp G.

There are several relations between this 2 elements τ, σ from Zp G.

Proposition 5.7. The kernel of the map

τ : Zp G −→ Zp G
x 7−→ τ x

is 1-dimensional and it is spanned by σ. In particular, σ · Zp G is 1-dimensional.

Proof. Proof taken from [12]. Let x be a general element from Zp G like in (5.1)

τ x = a0 + a1 g + · · · + ap−1 g p−1 − a0 g − a2 g 2 − · · · − ap−2 g p−1 − ap−1

= (a0 − ap−1 ) + (a1 − a0 )g + · · · + (ap−1 − ap−2 )g p−1 .

This is zero if and only if all the ai are equal, which is true whenever x is precisely a
multiple of σ, x = k · σ for any k ∈ Zp i.e. it is spanned by σ. Thus Ker(τ ) = σ · Zp G and
since it is spanned only by one element σ, the kernel is one dimensional.

46
Proposition 5.8. τ p−1 = σ
Proof. Taken from [12]. Using the binomial Newton’s theorem
p−1  
p−1 p−1
X p−1
τ = (1 − g) = (−1)i g i .
i=0
i

It must be proven that  

p−1
(−1)i ≡ 1 mod p
i
for 1 ≤ i ≤ p − 1. Thus, developing the combinatorial number:
 
p−1 (p − 1)(p − 2) · · · (p − i) 1 · 2···i
(−1)i = (−1)i ≡p (−1)i (−1)i = 1.
i (i)(i − 1) · · · (1) i · (i − 1) · · · 1
Then
p−1
X
p−1
τ = g i = σ.
i=0

Proposition 5.9. For every 0 ≤ i ≤ p − 1 it holds that σ ∈ τ i · Zp G.

Proof. Taken from [12]. By lemma 5.8 σ = τ p−1 = τ i · τ p−1−i .
Lemma 5.10. For every 0 ≤ i ≤ p − 1, the following exact sequence holds
f g
− τ i · Zp G →
0 −→ σ · Zp G → − τ i+1 · Zp G −→ 0.

Proof. Proof from [12]. Because of lemma 5.9, the first map is just an inclusion, which is
injective. Im(f ) =Ker(g) by lemma 5.7 and since g = τ , it is clear that g is surjective.
Lemma 5.11. Consider the following exact sequence
f g
D→A→
− B→
− C→E

with f and g being respectively injective and surjective homomorphisms, then the following
holds
dim(B) = dim(A) + dim(C).
In particular, it holds for the last sequence being an exact sequence.
Proof. Since g is a surjectice homomorphism, Im(g) = C, then:

dim(Ker(g)) + dim(C) = dim(Ker(g)) + dim(Im(g)) = dim(B).

Using the fact of being an exact sequence, Ker(g) = Im(f ) and because f is an injectice
homomorphism Ker(f ) = 0 thus:

dim(A) + dim(C) = dim(Im(f )) + dim(C) = dim(Ker(g)) + dim(C) = dim(B).

since dim(A) = dim(Ker(f )) + dim(Im(f )) = dim(Im(f )).

47
Chapter 5. Smith’s inequality and Harnack’s formula generalization

Lemma 5.12. For all 0 ≤ i ≤ p − 1, the subspace τ i · Zp G of Zp G is (p − i)-dimensional.

Proof. Result from [12]. The space τ 0 · Zp G = Zp G is p-dimensional (with respect to the
coefficient field Zp ), and the last lemmas 5.10, 5.11 and 5.7 imply for 0 ≤ i ≤ p − 1 that:

dim τ i · Zp G = dim (σ · Zp G) + dim τ i+1 · Zp G = 1 + dim τ i+1 · Zp G .

Now the lemma follows since

dim (τ · Zp G) = dim (Zp G) − 1 = p − 1

dim τ 2 · Zp G = dim (τ · Zp G) − 1 = p − 2

.. .. ..
. . .
p−2 p−3



dim τ · Zp G = dim τ · Zp G − 1 = 3 − 1 = 2
dim τ p−1 · Zp G = dim τ p−2 · Zp G − 1 = 2 − 1 = 1.

Thus for every 0 ≤ i ≤ p − 1

dim τ i · Zp G = p − i.

Lemma 5.13. Let ρ = τ i with 1 ≤ i ≤ p − 1. Set ρ = τ p−i . Then there is a short exact
sequence:
ρ=τ i
0 −→ ρ · Zp G −→ Zp G −−−→ ρ · Zp G −→ 0.

Proof. Taken from [12]. The third map ρ = τ i is surjective since Im(τ i : Zp G −→
τ i · Zp G) = τ i · Zp G. The second map is an inclusion, so it is injective. It is clear that
Ker(ρ) =Im(ρ).

By lemma 5.8 holds that τ = τ p−1 = σ and that σ = τ . Now, it will be considered
chain subcomplexes
ρC(K; Zp )
from the chain complex of C(K; Zp ) for ρ = τ i , 1 ≤ i ≤ p − 1 with coefficients in Zp . One
of the most important results is the following:

Theorem 5.14. For each ρ = τ j , 1 ≤ j ≤ p − 1, then

i ρ
0 −→ ρC(K; Zp ) ⊕ C(K G ; Zp ) →
− C(K; Zp ) →
− ρC(K; Zp ) −→ 0

is a short exact sequence of chain complexes, where i is the sum of the inclusions from
ρC(K; Zp ) to C(K; Zp ) and from C(K G ; Zp ) to C(K; Zp ) and ρ is just the map

ρ : C(K; Zp ) −→ C(K; Zp )
s 7−→ ρ · s.

48
Proof. Proof taken from [2]. There are two cases depending of the n-simplex s being in
K G or not.
If s ∈ K G , then τ s = 0, and as a consecuence ρs = 0 = ρs no matter what 1 ≤ j ≤ p−1,
and the sequence is exact:
id ρ
0 −→ C(K G ; Zp ) −
→ C(K G ; Zp ) →
− 0.
/ K G . Any n-chain in the orbit of s, i.e. G(s) = {gs|g ∈ G}, has the form:
Let now s ∈
p−1
X
c= ni g i s, ni ∈ Zp
i=0
Pp−1
with i=0 ni g i being an unique element in Zp G. By the last lemma 5.13 it holds that:
ρ=τ i
0 −→ ρ · Zp G ,−
→ Zp G −−−→ ρ · Zp G −→ 0
/ C(K G ; Zp ) that:
is an exact sequence. Thus it holds for any n-chain c ∈
i ρ
0 −→ ρC(K; Zp ) →
− C(K; Zp ) →
− ρC(K; Zp ) −→ 0.

Definition 5.15. (Smith’s homology groups).

For ρ = τ i , 1 ≤ i ≤ p − 1 being as always τ = 1 − g with g a generator of G, lets define:
Hnρ (K, L; Zp ) = Hn (ρC(K, L; Zp )).
Corollary 5.16. (Smith’s sequence).
Let G be a cyclic group of prime order p and let K be a regular simplicial complex.
Fixing ρ = τ i with 1 ≤ i ≤ p − 1 and letting ρ = τ p−i . Then by theorem 5.14, there exists
the following long exact sequence (see appendix C):
ρ
· · · → Hnρ (K; Zp )⊕Hn (K G ; Zp ) → Hn (K; Zp ) → Hnρ (K; Zp ) → Hn−1 (K; Zp )⊕Hn−1 (K G ; Zp ) → · · ·
Lemma 5.17. Consider the following exact sequence
f g h
D→A→
− B→
− C→
− E
with f ,g and h being homomorphisms, then the following holds
dim(B) ≤ dim(A) + dim(C).
Proof. Since g it is an homomorphism: dim(Ker(g)) + dim(Im(g)) = dim(B). Using the
fact of being an exact sequence, Ker(g) = Im(f ) and Im(g) = Ker(h):
dim(Im(f )) + dim(Ker(h)) = dim(Ker(g)) + dim(Im(g)) = dim(B).
Thus:
dim(Im(f )) + dim(Ker(f )) + dim(Ker(h)) + dim(Im(h)) ≥ dim(Im(f ))+dim(Ker(h)) = dim(B)
dim(A) + dim(C) = dim(Im(f )) + dim(Ker(f )) + dim(Ker(h))+dim(Im(h)) ≥ dim(B)
dim(A) + dim(C) ≥ dim(B).

49
Chapter 5. Smith’s inequality and Harnack’s formula generalization

Theorem 5.18. (Smith-Floyd I).

Let G be a cyclic group of order p prime and let X be a regular finite dimensional
simplicial G-complex such that all the homology groups with coefficients in Zp are finite
dimensional. For some 1 ≤ i ≤ p − 1, set ρ = τ i . Then all the Smith’s homology groups
Hkρ (X; Zp ) are finite-dimensional. In addition for n ≥ 0, we have:
∞ ∞
!
X X
G
dim(Hk (X ; Zp )) ≤ dim(Hk (X; Zp )) − dim(Hnρ (X; Zp )). (5.2)
k=n k=n

In particular, all the Hk (X G ; Zp ) homology groups are finite-dimensional.

Proof. Proof taken from [12]. Let ρ = τ p−i . For all k, by corollary 5.16, there is a long
exact sequence containing in particular the following segment
ρ
Hk+1 (X; Zp ) −→ Hkρ (X; Zp ) ⊕ Hk (X G ; Zp ) −→ Hk (X; Zp ). (5.3)
Set
ai = dim(Hiρ (X; Zp )) and ai = dim(Hiρ (X; Zp )).
Now it is deduced from the short exact sequence (5.3) and from lemma 5.17 that

ak + dim(Hk (X G ; Zp )) ≤ ak+1 + dim(Hk (X; Zp )). (5.4)

Interchanging the roles of ρ = τ i and ρ = τ p−i in corollary 5.16, it is obtained

ak + dim(Hk (X G ; Zp )) ≤ ak+1 + dim(Hk (X; Zp )). (5.5)

Let X be an N -dimensional simplicial complex, then for every k ≥ N + 1 it holds that

Hkρ (X; Zp ) = Hkρ (X; Zp ) = 0.

For the case k = N of (5.4) and (5.5)

aN + dim(HN (X G ; Zp )) ≤ dim(HN (X; Zp )) < ∞

aN + dim(HN (X G ; Zp )) ≤ dim(HN (X; Zp )) < ∞,

so aN , aN < ∞. Now, for the case k = N − 1 of (5.4) and (5.5)

aN −1 + dim(HN −1 (X G ; Zp )) ≤ aN + dim(HN −1 (X; Zp )) < ∞

aN −1 + dim(HN −1 (X G ; Zp )) ≤ aN + dim(HN −1 (X; Zp )) < ∞,

thus aN −1 , aN −1 < ∞. Going backwards it is seen that ak , ak < ∞ for every k as claimed
in the first conclusion of the proposition.
Now, for the second conclusion, rearranging (5.4) and (5.5)

dim(Hk (X G ; Zp )) ≤ ak+1 − ak + dim(Hk (X; Zp )), (5.6)

dim(Hk (X G ; Zp )) ≤ ak+1 − ak + dim(Hk (X; Zp )). (5.7)

50
Using (5.6) in an alternate way
N
X N
X
G
dim(Hk (X ; Zp )) ≤ dim(Hk (X; Zp )) + (an+1 − an ) + (an+2 − an+1 )+
k=n k=n
+(an+3 − an+2 ) + · · · + (aN +1 − aN )

it results a telescopic sum that eventually turns out into

N
X N
X
G
dim(Hk (X ; Zp )) ≤ dim(Hk (X; Zp )) + aN +1 − an ,
k=n k=n

or into
N
X N
X
dim(Hk (X G ; Zp )) ≤ dim(Hk (X; Zp )) + aN +1 − an .
k=n k=n

Anyway, since X is N -dimensional, both terms aN +1 = aN +1 = 0, thus

N
X N
X
G
dim(Hk (X ; Zp )) ≤ dim(Hk (X; Zp )) − an , (5.8)
k=n k=n

as the second conclusion claimed.

Corollary 5.19. (Smith-Floyd inequality mod p.). For any p ∈ N:
∞
X ∞
X
G
dim(Hk (X ; Zp )) ≤ dim(Hk (X; Zp )).
k=0 k=0

Corollary 5.20. (Harnack’s inequality generalization). Result from [15].

Consider a smooth complex algebraic manifold Γ and its fixed space under the conjugate
map i.e. RΓ obtained like in the last chapter, then the Smith’s inequality above 5.19
generalizes Harnack’s theorem 2.7.
∞
X ∞
X
dim(Hi (RΓ; Z2 )) ≤ dim(Hi (Γ; Z2 )).
i=0 i=0

Proof. Let’s consider Γ a complex non-singular projective curve and RΓ its real part. It
will be seek to arrive to Harnack’s inequality using Smith’s inequality.
The dimension of the group H0 (X; Z2 ) is the number of connected components, say L.
This holds since H0 (X; Z2 ) ∼ = Z2 for every X being a path-connected topological space
(see appendix B and C). By Hurewic’s theorem C.20, every oval in the real projective
plane Xi for i = 1, . . . , L sattisfies that

Z2 ∼
= π1 (Xi ; Z2 ) ∼
= H1 (Xi ; Z2 )

and by the equality (C.1), H1 (RΓ; Z2 ) = Lα=1 H1 (Xα ) = ZL2 . Thus

L

dim(H0 (RΓ; Z2 )) + dim(H1 (RΓ; Z2 )) = L + L = 2L.

51
Chapter 5. Smith’s inequality and Harnack’s formula generalization

Now, go back to the complex curve Γ of degree d, then its genus is g(d) = (d−1)(d−2) 2
.
Thus topologically is just like a g(d)-torus 5.1. Every n-torus is path connected, thus
H0 (X; Z2 ) ∼= Z2 , and using Hurewic’s C.20 together with the multiplicative property of
g(d)
π1 , it holds that π1 (Γ; Z2 ) ∼
= (Z2 × Z2 ) × · · · × (Z2 × Z2 ) ∼
= H1 (Γ; Z2 ). Since there is only
one 2-cell in its identification space of the g(d)-torus 5.2, it results that H2 (Γ; Z2 ) ∼ = Z2 .

1 2 g(d)

Figure 5.1: g(d)-torus.

Now computing the dimensions

dim(H0 (Γ; Z2 )) + dim(H1 (Γ; Z2 )) + dim(H2 (Γ; Z2 )) = 1 + 2g(d) + 1 = 2 + 2g(d)

and thus by corollary 5.19:

1
X ∞
X
dim(Hi (RΓ; Z2 )) ≤ dim(Hi (Γ; Z2 )) =⇒ 2L ≤ 2 + 2g(d) =⇒ L ≤ 1 + g(d).
i=0 i=0

In the last step it has been used again that the dimension of the identification space of a
g(d)-torus is up to dimension 2, thus

∞
X 2
X
dim(Hi (Γ; Z2 )) = dim(Hi (Γ; Z2 )).
i=0 i=0

It is drawed the identification space 5.2 of a g(d)-torus:

a2g(d)−1

a2g(d) a2

a2g(d)−1 a1

a2g(d) a2
a1 p

Figure 5.2: Identification space of a g(d)-torus.

52
Definition 5.21. (Euler characteristic with coefficients).
Let X be a complex, then it is defined the Euler characteristic with coefficients
to ∞
X
X (X; Zp ) = (−1)n dim(Hn (X; Zp )).
n=0

Theorem 5.22. (Smith-Floyd II).

Let G be a cyclic group of order p prime and let X be a regular finite dimensional
simplicial G-complex such that all the homology groups with coefficients in Zp are finite
dimensional (by the last theorem 5.18, it also holds that Hk (X G ; Zp ) homology groups are
finite-dimensional). Thus

X (X G ; Zp ) ≡ X (X; Zp )(mod p).

Proof. Taken from[2] and [12].

Let τ = 1 − g ∈ Zp G, with g ∈ G being a generator of the cyclic group G as usual. By
i
the last theorem 5.18 all the Smith special homology groups Hkτ (X) are finite-dimensional
and since X is finite-dimensional itself only finitely many of them are non-zero, making
also finite its Euler characteristic
∞
τi i
X
X (X; Zp ) = (−1)k dim(Hkτ (X; Zp )).
k=0

In particular, for ρ = τ in corollary 5.16 it is given the following long exact sequence
p−1 i ρ=τ p−1
· · · → Hkτ (X) ⊕ Hk (X G ) →
− Hk (X) −−→ Hkτ (X) → Hk−1
τ
(X) ⊕ Hk−1 (X G ) → · · ·

Using that i is a double inclusion and thus injective and that ρ = τ is surjective, by last
lemma 5.11 it is deduced that
p−1
X (X; Zp ) = X τ (X; Zp ) + X τ (X; Zp ) + X (X G ; Zp ).

Thus to see that X (X; Zp ) and X (X G ; Zp ) are equal modulo p, it is enough to prove that
p−1
X τ (X; Zp ) + X τ (X; Zp ) ≡ 0 (mod p).

In order to see that last statement it is necessary to prove for every 1 ≤ i ≤ p − 1 that
σ τ
0 → τ p−1 Ck (X) →
− τ i Ck (X) →
− τ i+1 Ck (X) → 0 (5.9)

is a short exact sequence. Notice that it is well defined because for every 1 ≤ i ≤ p − 1
it holds that σ ∈ τ i · Zp G. In addition τ is clearly surjective and τ p−1 = σ (lemma 5.8)
injective by the definition of σ. By lemma 5.7 also Im(σ)=Ker(τ ), thus (5.9) holds. In
conclusion, for 1 ≤ i ≤ p − 1 it follows that
τi+1 p−1 i i+1 p−1
· · · → Hk+1 (X; Zp ) → Hkτ (X; Zp ) → Hkτ (X; Zp ) → Hkτ τ
(X; Zp ) → Hk−1 (X; Zp ) → · · ·

and using lemma 5.11 it is obtained:

i p−1 i+1
X τ (X; Zp ) = X τ (X; Zp ) + X τ (X; Zp ).

53
Chapter 5. Smith’s inequality and Harnack’s formula generalization

Thus, since τ p = 0:
p−1 p−2
!
τi τ p−1 τ i+1 p−1
X X
X (X; Zp ) = (X (X; Zp ) + X (X; Zp )) + Xτ (X; Zp )
i=1 i=1
p−1
!
τi p−1
X
= X (X; Zp ) + (p − 1)X τ (X; Zp ).
i=2

Reordering the summands:

p−1
X τ (X; Zp ) = (p − 1)X τ (X; Zp )
τ p−1 τ p−1
X τ (X; Zp ) + X (X; Zp ) = p · X (X; Zp ).

Then
X (X G ; Zp ) ≡ X (X; Zp )(mod p).

Corollary 5.23. (Smith-Floyd II).

Consider a smooth complex algebraic manifold Γ and its fixed space under the conjugate
map i.e. RΓ, by theorem 5.22 it holds that X (RΓ; Z2 ) and X (Γ; Z2 ) have the same parity.
Definition 5.24. (Mod-p acyclic).
The simplicial complex X is said to be mod-p acyclic if
 
Zp if k = 0
Hk (X; Zp ) = .
0 else

Definition 5.25. (Mod-p homology n-sphere).

The simplicial complex X is said to be mod-p homology n-sphere if
 
 Zp if k = 0 
Hk (X; Zp ) = Zp if k = n .
0 else
 

Corollary 5.26. (Smith-Floyd).

Let p be a prime, let G be a finite ciclic group of order p and let X be a finite-
dimensional simplicial G-complex. Then
1. If X is mod-p acyclic, then so is X G . In particular, X G is non-empty.

2. If X is a mod-p homology n-sphere, then X G is either empty or a mod-p homology

m-sphere for some 0 ≤ m ≤ n.
Proof. Proof taken from [12].

1. Since X is mod-p acyclic then by theorem 5.18

∞
X ∞
X
G
dim(Hk (X ; Zp )) ≤ dim(Hk (X; Zp )) = 1.
k=0 k=0

54
 Since X is path connected, either X G is empty or X G is mod-p acyclic. But by
theorem 5.22
X (X G ; Zp ) ≡ X (X; Zp )(mod p) ≡ 1(mod p),
thus X G is mod-p acyclic.

2. Now let X be a mod-p homology n-sphere. By theorem 5.18, it holds that

∞
X ∞
X
G
dim(Hk (X ; Zp )) ≤ Hk (X; Zp ) = 2.
k=0 k=0

Then the left hand side is 0, 1 or 2. If at the left there is a 0, then it would be the
case when X G is empty. If in the left there is a 1 it necessarily would be due to the
0th homology group, so X (X G ) = 1, which contradicts theorem 5.22 since
 
G 2 if n = 2λ
X (X ) ≡ X (X)(mod p) =
0 if n = 2λ + 1

for any λ ∈ N, thus it can be discarded the case when is 1.

Then let the left hand side be 2, so X G is a mod-p homology m-sphere for one m ∈ N.
Now there is only left to check that 0 ≤ m < n, thus applying theorem 5.18
∞
X ∞
X
G
dim(Hk (X ; Zp )) ≤ dim(Hk (X; Zp )) = 0,
k=n+1 k=n+1

then Hk (X G ; Zp ) = 0 for all k ≥ n + 1, thus in this case X G is a mod-p homology

m-sphere for 0 ≤ m ≤ n.

55
 APPENDIX A

Vector fields

Definition A.1. (Manifold).

A set M ⊂ Rn is called a manifold if every point x ∈ M has an open set U in M s.t.
is homeomorphic to an open set V inside Rm . It will be denoted a parametrization to
℘ = (x1 , . . . , xn ) : V −→ U ⊂ M,
and a coordinate system of M to the following inverse map
ϕ = (ϕ1 , . . . , ϕm ) : U −→ V ⊂ Rm .
Definition A.2. A vector field on a manifold M is a continuous correspondence that
assgins to each point p ∈ M a tangent vector in Tp M i.e.
X : M −→ TM
p −→ X(p) = (p, X(p)) ∈ M × Tp M
where TM denotes the tangent bundle. It shall be refered to X : U −→ Rn as vector
field.
Definition A.3. (Coordinate vector fields).
Using the same notation above, given a coordinate system ϕ : U −→ Rm it can be
defined the following special vector fields
∂
: U −→ Rn
∂xi
∂ ∂℘
p 7−→ (p) := (ϕ(p)).
∂xi ∂xi
Every vector field in U can be defined then as X(p) = m ∂
P
i=1 Xi (p) ∂xi (p).

Definition A.4. A continuous curve c : R −→ M is called an integral curve of a vector

field X iff for every t ∈ R it sattisfies that
X(c(t)) = c0 (t).

56
 APPENDIX B

Algebraic topology

Definition B.1. (Homotopy).

Let X and Y be topological spaces and let f : X −→ Y and g : X −→ Y be
continuous applications. It is said that f is a homotopy to g if there exists a continuous
map H : X × [0, 1] −→ Y s.t.

H(x, 0) = f (x)
H(x, 1) = g(x).

Note. As notation it is used f ' g and this homotopy relation is in fact an equivalence
relation.

Definition B.2. (Homotopically equivalent).

Let X and Y be topological spaces. It is said that X and Y are homotopically
equivalent if there exist a continuous map f : X −→ Y and g : Y −→ X s.t. g ◦ f ' 1X
and f ◦ g ' 1Y where 1X and 1Y denotes respectively the identity maps in the topological
spaces X and Y .

Definition B.3. (Loop).

Let X be a topological space and let x0 ∈ X. It is called a loop in X with basepoint
on x0 to every f : [0, 1] −→ X continuous map that f (0) = f (1) = x0 . Two loops are said
to be homotopic if f '[0,1] g where '[0,1] denotes that f (0) = g(0) = f (1) = g(1) = x0
i.e. f and g are loops with the same basepoint x0 ∈ X and this relation, is an equivalence
relation.

Example. Let X = R2 , then f '[0,1] g can be visualised as:

g
f

x0

57
Appendix B. Algebraic topology

Note. The set π1 (X, x0 ) denotes the quotient set of the loops of basepoint x0 ∈ X with
respect to the homotopy relation between loops defined before i.e. '[0,1] . The quotient class
of any loop f is denoted as [f ]. This notions and some of the definitions above can be
found in [11].
Definition B.4. (Product of two loops).
Let X be a topological space and let f and g be two loops in the topological space X.
It is called product of the loops f and g to:

0 ≤ t ≤ 12
 
f (2t),
(f ∗ g)(t) = 1
g(2t − 1), 2
≤t≤1

Lemma B.5. Let [f ], [g] ∈ π1 (X, x0 ), then there is a well defined path class product

[f ]∗[g] = [f ∗ g].

Proof. Taking another representant of the equivalence class of f and lets call it f 0 i.e.
[f ] = [f 0 ], and the same for g and lets call it g 0 ([g] = [g 0 ]). All f, f 0 , g, g 0 are loops with
base in x0 .
Consider
Hf : [0, 1] × [0, 1] −→ X
the homotopy map relative to f to f 0 , and

Hg : [0, 1] × [0, 1] −→ X

the homotopy map relative to g to g 0 . Then applying the definition of the product of two
loops ∗ for the first coordinate of Hf and Hg :

Hf (2t, t0 ), 0 ≤ t ≤ 21
 
0
(Hf ∗ Hg )(t, t ) =
Hg (2t − 1, t0 ), 1
2
≤t≤1

and this construction is actually a homotopy between (f ∗ g) and (f 0 ∗ g 0 )

0 ≤ t ≤ 12
 
Hf (2t, 0) = f (2t),
(Hf ∗ Hg )(t, 0) = 1 = (f ∗ g)(t)
Hg (2t − 1, 0) = g(2t − 1), 2
≤t≤1
Hf (2t, 1) = f 0 (2t), 0 ≤ t ≤ 12
 
(Hf ∗ Hg )(t, 1) = = (f 0 ∗ g 0 )(t)
Hg (2t − 1, 1) = g 0 (2t − 1), 1
2
≤ t ≤ 1

and in conclusion [f ∗ g] = [f 0 ∗ g 0 ], thus the product

[f ]∗[g] = [f ∗ g]

is well-defined.
Theorem B.6. (First fundamental group).
Let X be a topological space and x0 ∈ X. Then π1 (X, x0 ) is a group with respect to the
operation ∗ defined in the last lemma B.5.
Proof. The proof of this theorem can be found at [10] in page 371 in theorem 51.2.

58
Definition B.7. Let γ : [0, 1] −→ X be a path in X from x0 to x1 i.e. γ(0) = x0 and
γ(1) = x1 with γ being a continuous map. It is defined

γ̂ : π1 (X, x0 ) −→ π1 (X, x1 )
[f ] −→ [γ] ∗ [f ] ∗ [γ]

where γ(t) = γ(1 − t) i.e. the reverse path going from x1 to x0 . It is well-defined since
[γ] ∗ [f ] ∗ [γ] is in π1 (X, x1 ).

Theorem B.8. The map γ̂ is a group isomorphism.

Proof. The proof of this theorem can be found at [10] in page 377 in theorem 52.1.

Corollary B.9. If X is a path connected i.e. there is always a path between every 2 points
and x0 and x1 are two points of X, then π1 (X, x0 ) is isomorphic to π1 (X, x1 ).

After this last corollary it makes sense to denote the first fundamental group of a
path-connected topological space X as π1 (X) instead of π1 (X, x0 ) for any x0 ∈ X.

Definition B.10. (Contractible).

If any loop in a path-connected topological space X can be continuously retracted
into its base point it is said that the space X is contractible. Thus π1 (X) has only one
element, so it must be the identity:

π1 (X) = e.

Definition B.11. Let h : (X, x0 ) −→ (Y, y0 ) be a continuous map. Then it is defined the
following map

h∗ : π1 (X, x0 ) −→ π1 (Y, y0 )
[f ] 7−→ h∗ ([f ]) = [h ◦ f ]

Theorem B.12. If X and Y are homeomorphic and path-connected, then h∗ : π1 (X) −→

π1 (Y ) it is an isomorphism.

Proof. The proof is in [10], results 52.4 and 52.5 in page 379.

Theorem B.13. (First fundamental group of a circumference S1 ).

π1 (S1 ) ∼
=Z

Proof. This proof is in [10], theorem 54.5 in page 392.

Theorem B.14. (Multiplicative property of the first fundamental group).

π1 (X × Y, x0 × y0 ) is isomorphic to π1 (X, x0 ) × π1 (Y, y0 ).

Proof. The proof is in [10], theorem 60.1 in page 421.

Corollary B.15. The first fundamental group of the torus T = S1 × S1 is π1 (T ) ∼

= Z × Z.

59
 APPENDIX C

Homology

Definition C.1. Let G be an abelian group. If {Gα }α∈J is a family of subgroups of G it

is said that these groups generate G if every element x ∈ G can be written as a finite
product of elements of the subgroups {Gα }α∈J . This subgroups will be known as the basis
of G.

This definition means that there is a finite sequence (x1 , x2 , . . . , xn ) of elements of the
groups Gα s.t. x = x1 ·x2 ·. . .·xn . Such a sequence is called a word of length n. Due to the
commutativity it can be rearranged the factors in the expression for x to group together
in (x1 , x2 , . . . , xn ) the factors that belong to a single subgroups Gα of the set {Gα }α∈J .
Furthermore, if any xi = 1 or it happens to be in a word xi together with xj = x−1 i it
can be always be both deleted from the original word making it shorter. Applying all this
reductions to the original word it is obtained what is called a reduced word.

Definition C.2. (Free product).

Let G be a group and {Gα }α∈J be a family of subgroups of G that generates G. Let
also that Gα ∪ Gβ consist only on the identity element alone whenever α 6= β. Then it
is said that G is the free product of the groups Gα if for every x ∈ G there is only one
reduced word in the groups Gα that represents x. In this case it is written

G = G1 ∗ . . . ∗ Gn .

Definition C.3. A standard n-simplex is defined as the following set

n+1
X
n n+1
∆ = {(t0 , . . . , tn ) ∈ R | ti = 1, ti ≤ 0}.
i=0

Definition C.4. A singular n-simplex in a space X is a continuous map

σ : ∆n −→ X.

60
Definition C.5. (N-chains).
Let Cn (X) denote the free abelian group with basis the set of singular n-simplices
Pin X.
The elements of Cn (X) called n-chains or singular n-chains, are finite formal sums i ni σi
where ni is a coefficient of a field (as for example Z or Z2 ) and where σi : ∆n −→ X.

Definition C.6. (Boundary map).

The boundary map is defined as

∂n : Cn (X) −→ Cn−1 (X)

n
X
σ 7−→ (−1)i · σ|[v0 ,...,vˆi ,...,vn ] .
i

Using this last definition it can be seen that ∂n ◦ ∂n−1 = 0, thus Im(∂k+1 ) ⊂Ker(∂k )
and it is obtained the following chain complex
∂n+2 ∂n+1 n ∂ ∂n−1
. . . −−−→ Cn+1 (X) −−−→ Cn (X) −→ Cn−1 (X) −−−→ . . .

Definition C.7. (Homology group).

It is defined the n-singular homology group as:
.
Hn (X) = Ker(∂n ) Im(∂n+1 ) .

Proposition C.8. Let Xα all the connected components which conform X, then
M
Hn (X) = Hn (Xα ). (C.1)
α

Proof. Since the n-simplices are path connected, then the singular simplex (σ : ∆n −→ X
being continuous) has a path connected image and in conclusion L the image always fall in
a single one of the path connected components, so Cn (X) = α Cn (Xα ). Thus for every
σα ∈ Ck (Xα ) for k ∈ N, using the boundary map of the complex (Cn (Xα ), ∂nα ) it holds for
every α that ∂kα (σα ) ∈ Ck−1 (Xα ). Defining the following map
M M M
∂k = ∂kα : Ck (Xα ) −→ Ck−1 (Xα )
α α α

it holds that Ker(∂k )= α Ker(∂kα ) and that Im(∂k )= α Im(∂kα ), thus

L L

. .
α
L M
Hk (X) = Ker(∂k ) Im(∂k+1 ) = α Ker(∂k ) α = Hk (Xα )
L
α Im(∂k+1 )
α

where the last equal holds since for every α and k ∈ N

α
Im(∂k+1 ) ⊂ Ker(∂kα ).

Proposition C.9. If X is path connected then H0 (X) ∼

= Z.

61
Appendix C. Homology

Proof. The proof is in [7], proposition 2.7 in page 109.

Corollary C.10. H0 (X) is a direct sum of Z, one for each path component of X.

Proof. Use the last 2 propositions C.8 and C.9.

Definition C.11. Let f : X −→ Y be a continuous map between 2 topological spaces,

then a map of chain complexes is defined as the following f]

f] (σ)=f ◦σ
f] : Cn (X) −→ Cn (Y ), ∆n /
>Y
σ
! f
X

defined for every σ ∈ Cn (X) i.e. σ : ∆n −→ X as f] (σ) = f ◦ σ : ∆n −→ Y .

Let (Cn (X), ∂ X ) and (Cn (Y ), ∂ Y ) denote the singular complexes with its boundaries
of X and Y . In page 111 in [7] is explained that this 2 boundary maps are related by
f] ◦ ∂ X = ∂ Y ◦ f] . It is deduced from here that f] (Ker∂ X ) ⊆ Ker∂ Y and f] (Im∂ X ) ⊆ Im∂ Y .
Then f] induces a homomorphism

Hn (f ) = f∗ : Hn (X) −→ Hn (Y ).

Theorem C.12. If f : X −→ Y and g : X −→ Y are homotopic maps, then for every

n ∈ N:
Hn (f ) = Hn (g) : Hn (X) −→ Hn (Y ).

Proof. The proof is in [7], theorem 2.10 in page 111.

Corollary C.13. The map f∗ : Hn (X) −→ Hn (Y ) induced by a homotopy equivalence

map f : X −→ Y is an isomorphism for all n ∈ N, i.e. Hn (X) ∼
= Hn (Y ).

Definition C.14. (Exact sequence).

A sequence of abelian groups {Ai }i∈I and homomorphisms {αi }i∈I such
αn+2 αn+1 n α αn−1
· · · −−−→ An+1 −−−→ An −→ An−1 −−−→ · · ·

is said to be exact at {Ai }i∈I if Ker(αn ) =Im(αn+1 ).

Definition C.15. (Short exact sequence).

A short exact sequence is an exact sequence of the form
f g
0 −→ A →
− B→
− C −→ 0.

Note. Exactness at A: implies that Ker(f ) =Im(0) and since f is a group homomor-
phism Ker(f ) = 0 ⇐⇒ f is injective.
Exactness at B: implies that Ker(g) =Im(f ), i.e. g ◦ f = 0.
Exactness at C: Im(g)=Ker(0) = C, then g is surjective.

62
Definition C.16. Given a topological space X and a subspace A ⊂ X, lets define
.
Cn (X, A) = C n (X) Cn (A) ,

thus any chain in A is trivial in Cn (X, A). The boundary map ∂ : Cn (X) −→ Cn−1 (X)
takes Cn (A) to Cn−1 (A), thus it induces a quotient boundary map

∂ : Cn (X, A) −→ Cn−1 (X, A)

and with it, a complex {Cn (X, A), ∂} which it’s homology groups Hn (X, A) are denoted
as relative homology groups.
Lemma C.17. Lets consider 3 chain complexes like these {An , ∂}, {Bn , ∂} and {Cn , ∂}
also having that for every n ∈ N it holds the following exact sequence
i j
0 −→ An →
− Bn →
− Cn −→ 0.

Then the map ∂ : Hn (C) −→ Hn−1 (A) is well-defined and it is an homomorphism.

Proof. The idea of this proof is in [7], page 116.
Theorem C.18. The following sequence of homology groups
∂ i
∗ j∗ ∂ ∗ i j∗
··· →
− Hn (A) −
→ Hn (B) −
→ Hn (C) →
− Hn−1 (A) −
→ Hn−1 (B) −
→ ···

is exact.
Proof. This proof is in theorem 2.16 in [7].
It is possible to compute the n-homology group Hn (X) when X is a cell complex
using what is called celullar homology. First are defined the n-chains. Let X be a cell
complex, then we define CnCW (X) as the free abelian group with basis the set of n-cells of
X, this is denoted as
CnCW (X) = hn-cells of Xi.
With this chains it is possible to find a boundary map ∂ as it is explained in section
’Cellular homology’ at page 137 on [7]. Thus there is a complex chain {C CW (X), ∂} and
its n-homology group is denoted as HnCW (X). In fact, it holds:
Theorem C.19.
HnCW (X) ∼
= Hn (X).
Proof. This proof is in theorem 2.35 at page 139 in [7].
Theorem C.20. (Hurewicz’s theorem).
If X is a path-connected topological space, then

H1 (X) ∼
= π1 (X)ab
where π1 (X)ab is the abelianization of the group π1 (X).
Proof. This result can be found in [7].

63
 Bibliography

[1] J. Bochnack, M. Coste, M.F. Roy, Real Algebraic Geometry, Ergebnisse der Mathe-
matik und ihrer Grenzgebiete, Springer, Berlin, 1998.

[2] G.E. Bredon. Introduction to compact transformation groups. Elsevier, Vol. 46, 1972.

[3] M.J. de la Puente. Curvas algebraicas y planas. Universidad de Cádiz. Servicio de

Publicaciones, 2007.

[4] M.J. de la Puente. Real Plane Algebraic Curves. Expositiones Mathematicae, Vol. 20,
pp 291-314, Madrid, 2002.

[5] G. Fischer. Plane Algebraic Curves. Bd. 15. American Mathematical Society, 2001.

[6] I.G. Petrovsky. On the topology of real plane algebraic curves. Annals of Mathematics,
Second Series, Vol. 39, No. 1, 1938.

[7] A. Hatcher. Algebraic topology. Cambridge university press, 2001.

[8] V. Kharlamov, O. Viro. Easy reading on topology of real plane algebraic curves. 2007.
URL: ’http://www.pdmi.ras.ru/ olegviro/introMSRI.pdf’.

[9] Y. Matsumoto. An introduction to Morse theory. Translation of mathematical mono-

graphs, vol. 208. American Mathematical Society, 2002.

[10] J. Munkres. Topology. Pearson, 2000.

[11] S.P. Novikov, V.A. Rokhlin. Topology II. Homotopy and homology. Springer, 2004.

[12] A. Putman. Smith theory and Bredon homology. University of Notre Dame.

[13] R. Schabert. Satz von Petrovski und maximale Anzahl reeller Nullstellen von ternären
psd Formen. Universität Konstanz. 2018.

[14] R. Sharpe. On the ovals of even-degree plane curves. Michigan Mathematical Jorunal
22(3): pp. 285-288, 1976.

64
 Bibliography

[15] O. Viro. Introduction to topology of real algebraic varieties. 2007. URL:

’http://archive.schools.cimpa.info/archivesecoles/20141218145605/es2007.pdf’.

[16] G. Wilson. Hilbert’s sixteenth problem. Topology. Vol. 17. pp 53-73. Pergamon press,
1978.

65