ON THE GEOMETRY OF THE ZEROS OF A POLYNOMIAL WITH

COMPLEX COEFFICIENTS

A THESIS

SUBMITTED TO THE FACULTY OF ATLANTA UNIVERSITY

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR

THE DEGREE OF MASTER OF SCIENCE

BY

GENEVIEVE MADELINE KNIGHT

DEPARTMENT OF MATHEMATICS

ATLANTA UNIVERSITY

AUGUST 1962

/ / / S!
 3L
cl
PREFACE

The subject treated in this thesis is the study of polynomials from

a non-algebraic standpoint. This point of view is largely that of the

geometric theory of function of a complex variable, hence, the title On the

Geometry of the Zeros of a Polynomial with Complex Coefficients.

An attempt is made to show the connection of polynomials with the

geometric theory of functions of a complex variable. We will concern our¬

selves with problems centered around the study of zeros of a polynomial

f(z) as functions of various parameters. The parameters being coefficients

of f(z), or the zeros of the coefficients of some related polynomial g(z).

Our methods for investigation will involve the geometric operations

with complex numbers and certain principles which are based upon these opera¬

tions. Among these is the principle that a sum of vectors cannot vanish if

the vectors are all drawn from a point 0 on a line L to points all on the

same side of L. Our basic principle is the so-called Principle of Argument

and its corrollaries. Among these are Roue he* s Theorem and Hurwitz's Theorem.

Hence, due to the nature of not only its problems but also its methods, our

subject may be considered as belonging to the geometric theory of functions.

The writer wishes to express her deepest gratitude to Or. Lonnie Cross

for having encouraged the development of the desire to seek the truth and

for having made many helpful criticisms and suggestions. The writer wishes

to acknowledge her indebtedness to The Atlanta University Fellowship Committee

which made this year's study possible} to Dr. Saxena for his informative

lectures} to my sister Loretta J. Knight for the drawing of accompanying

figures} and to Mrs. A. C. Batey for typing this thesis.

ii
 TABLE OF CONTENTS

Page

PREFACE ii
ABBREVIATIONS iv

Chapter

I. INTRODUCTION 1
Some Basic Theorems 1
The Zeros of the Derivative 6
Physical Interpretation 7
Geometric Interpretation 9
Function - Theoretic Interpretation 11

II. THE CRITICAL POINTS OF A POLYNOMIAL AND SOME OF THEIR

GENERALIZATION 15
Their Convex Enclosure 15
The Critical Points of a Real Polynomial 17
Some Generalizations 20

III. INVARIANTIVE FORMULATION 23

The Derivative Under Linear Transformations 23
Circular Regions 26
Zeros of Polar Derivatives 27
Successive Polar Derivatives 29

IV. COMPOSITE POLYNOMIAL 32

Apolar Polynomials ...... . . ...... 32
Combinations of a Polynomial and its Derivatives 35

V. THE CRITICAL POINTS OF POLYNOMIALS IN PRESCRIBED REGIONS . . . 1*0

The Critical Points of a Rational Function ...... . . 1*0
Polynomials With Two Given Zeros 1*3
Mean - Value Theorems 1*5
Polynomials With n-known Zeros 1*8

BIBLIOGRAPHY 50

iii
 ABBREVIATIONS

Eq.(m,n) The equality given in the nth formula of section ra.

Ineq.(m,n) The inequality given in the nth formula of section m.

Ex.(m,n) The nth exercise given at the close of section m.

Fig.(m,n) The nth figure accompanying section m.

Th. (m,n) The nth theorem in section m.

Cor.(m,n) A corollary to Th.(m,n).

Lera.(m,n) A lemma used in proving Th.(m,n).

C (m,n) The binomial coefficient n j j (n-m) Î

Sg x Signura of real number x. If x ? 0. sg x

= lj if x < 0, sg x * -1 and if x * 0, sg x
= 0.
R(z) Real part of the complex number: z = x + iy.

!(■) Imaginary part of the complex number

z = x + iy.

Arg(z) Argument (amplitude, phase angle) of z.

x - iy, imaginary conjugate of z.

f(k)U) Kth derivative of f(z)j ex. f’(z),

f'«(z) for k = 1,2.

Iff If and only if.

Implies.

i.e. That is.

Q.E.D. Quod erat demonstrandum (what was to be shown).

Greater than or equal to.

Less than or equal to.

Such that.

iv
 CHAPTER I

INTRODUCTION

Some Basic Theorems «--Before we go into the study of the different

specific problems connected with the zeros of polynomials, let us consider

certain general theorems which we may use as references.

The first of these theorems concerns the sufficient condition for the

non-vanishing of a sum of complex numbers. It requires that each term in

the sum be a vector drawn from the origin to a point on the same side of

seme line through the origin.

Theorem (1,1). If each complex number w., jsl,2,...n, has the pro-

perties that w. 0 and

J

(1,1) y £rarg w. ^ T+ "IT , j = l,2,...n, where )fis a real con-

J
n
stant, then their sum w = 2 w. cannot vanish.
’ » J

Proof:

Let us consider two cases: T - 0 and Ï ^0.

Case (i). If *
when the w.
0
0are vectors drawn from the origin

to points on the positive real axis or in the upper half-plane» If arg =

0 for all j, then R(w..) 0 for all j, and hence R(w) ^ 0. If arg w^ 0

for some value of j, then I(wj) "7 0 for that j and hence I(w) 70. There¬

fore, if Y * 0, w \ 0.
- Y*i
Case (ii). When T\ 0, we consider the quantities w. * e w.,
J J
These satisfy (1,1) with If* 0, and their sum w* does not vanish. As w1 »
- fi
e w , it follows that w k 0.
v

The point w lies inside the convex sector consisting of the origin

1
 2

and all the points z for which Y — arg z Y* ~t & ^ 7r) if all the

;
1
points w lie in the same sector. Q.E.D.
3
2
We now investigate the so-called Principle of argument.

Theorem (1,2). Let f(z) be analytic interior to a simple closed Jordan

Curve C, continuous, and different from zero on C. Let K be the curve de¬

scribed in the w- plane by w = f(z) and let A^arg f(z) denote the net change

in arg f(z) as the point z traverses C once over in the counterclockwise

direction. Then the number, N, of zeros of f(z) interior to C, counted with

their multiplicities, is N = 1/2JT arg f(z), i.e., it is the net number

of times that k winds about the point w = 0.

Proofs

Let us consider the case only when f (z) is a polynomial. Let *2>

Morris Marden, The Geometry of The Zeros of a Polynomial in A Complex

Variable. New York, 19^9., p. 1.
2
Dr. Lonnie Cross, "Principle of Argument1'. (Lecture, delivered to
class in Complex Variables, Atlanta University, Atlanta, Georgia, March 22,
1962).
 3

...z be the zeros of f(z) inside C and z _,z ^ z be the

n n + 1 n + 2' n + p
zeros outside G. Then

f(z) « A (z - z) • • • ( z - z ) (z - zn -| ) • » » ( z - z ), arg f(z) -

n 1 n n + J. n + p

arg A +
n
£ arg (z - z.) +
/*'
3
1
arg (z - z.).
As the point z traces C counterclockwise, arg (z - z^) increases by

27T when 1 f j < n, and has a zero net change when n< j £ n + p.

Hence we have

N - 1/(2 7T i)

for any single closed curve C.

/ f
HFCÏJ
dz.

Q.E.D.
3
Using Th. (1,2), we are able now to derive another important theorem.

Rouché’s Theorem (1,3). If f and g are analytic interior to a simple

closed Curve C, and if they are continuous on C, and if

(1,3) * /g(z) / on C,
then the function F(z) - f(z) + g(z) has the same number of zeros interior

to C as does g(z).

w - plane

(Fig. 1,2)

Phillip Franklin, Function of Complex Variables. Ehglewood Cliffs,

New Jersey, 1958 , pp. 200-201.
 1*

Proof:

(1,14.) Let F(z) = w g(z), w » 1 +

Let be the number of zeros of g(z) inside C, then according to Th. (1,2)

(1,5) A6arg g(z) = 2irNf .

Since / g(z) 1~ 1 011 the w

, defined in eq. (1,1*), describes (see

figure (1,2) a closed curve /"* which lies interior to the circle with center

at w = 1 and radius 1. Thus, the point w always remains in the right half¬

plane. The net change in arg w as w varies on T is therefore zero. Hence,

Acarg F(z) = A^arg w + Acarg g(z) - 2TTN^

and according to Th. (1,2), F(z) also has zeros in C. Q.E.D.

We shall now apply Rouchl's Theorem to prove a proposition, which we

1
shall make frequent reference to later. That is, we shall prove that the

zeros of a polynomial are continuous functions of the coefficients of the

polynomial.

Theorem (1,1*). Let

n n mj
f(z) = aQ + a1z + ... a^z - an Q (z - Zj) , an / 0.
n 1 n
F(z) - (aQ + ) + (a^ + €^)z ...(an - 1 + - l)z + aRz ,

and let 0 < f ,

k
min
I °k - I j j * 1*2, •••! * lj k ^ 1^ ••• p«

Then given any positive number 6 i, for / ^ / - £ , i ■ 0, 1, ...,

n - 1, then F(z) has precisely m, zeros in the circle C with center z and
■K K K

radius fk.

1
Morris Marden, op. cit., p. 1*.
 5

Proof;
On C^ the polynomial
3(z) = 6o+61z + .... +6n - 1 z1 ^ has the property
n-i
/J(z)/ ± »k = t (■^*+1**1) ;
whereas on C
J *K
If we choose £ ^ , we have the relation I J. ( z)/ Z. /f (z)| on
/
Ck. According to Rouche's Theorem, F(z) has the same number of zeros in¬
side C as does f(z). Since ineq. (1,6) insures that the only zeros of
K

f(z) in is the one of multiplicity m^ at z^, therefore, F(z) has pre¬

cisely m^ zeros in C . Q.E.D.
Let us also note that Th. (1,U) is a special case of Hurwitz's Theorem,
which is stated and proved below.
Hurwitz's Theorem. Th. (1,5). Let ff (z) J , (n = 1,2,....) be a
sequence of functions which are analytic in a region R and which converge
uniformly to a function f(z) ^ 0 in every closed subregion of R. Let
be an interior point of R. If 2 is a limit point of the sequence j f^(z) j,
then ^ is a zero of f(z). Conversely, if J is an m- fold zero of f(z),
every sufficiently small neighborhood K of ^contains exactly m zeros,
counted with their multiplicities of each f (z) n 21 N(K).
n

Proof;
Assume ) 0. Since f(z) is analytic in R, it can have only a
finite number of zeros in R. Choose P (> 0)^ f (z) 0 within and on
the circle K; / z ~ JJ =
P» Let £ = min f f (z) j for z on K. Since the
 6

sequence , converge to f (a) uniformly within and on K, we can

I *(•>/
find a N > 0, N * N(K) * /fn(z)-f^(t€for all n > N. Hence,

I f^(z) - f(z) |< M on K and, by Rouche's Theorem, the sum function

f (z) « I f (z) - f (z) JI + f(z) has as many zeros in K as does f(z).

n L n

Since f^(z) / 0 inside K for all n 2 N, a point at which f( ^ ) / 0

can not have a limit point of the zeros of the sequence f (z).
n
Conversely, if we assume that^, is a m - fold zeros of f(z), we may

again choose a positive number /» , } f(z) / 0 on K. Reasoning as in

the previous paragraph, we now conclude from Rouche's Theorem that each

f (z) n 2 N, has precisely m - zeros in K. Q.E.D.

n
Hurwita's Theorem provides us with the principle means of passing from

certain theorems on the *eros of polynomials to the corresponding theorems

on the zeros of entire functions.

The Zeros of the Derivative .--Considering Rolle's Theorem in the theory

of functions of a real variable, let us determine the zeros of polynomials

in a complex variable by studying the location of the zeros of the deriva¬

tive f'(z) of the polynomial

m
1 2 P
(2,1) f(z) (z - z/^z - Zg) ” ... (z — zp)

n ■
f
ZL m., in relation to distinct zeros z^, of f( ? ).
J
I J
Since f ' (z) may be written as

f'(z) ■ f(z) £ f»(z)/f(z)J

* f (z) d £ log f (z) J dz,

the zeros fall into two classes. First, there are the points z^ for which

> 1. As zeros of f'(z), they have the individual multiplicities of

nij - 1, and the total multiplicity of N - P. Secondly, there are P - 1

 7

1
zeros of the logarithmic derivative.

(2,2) F(z) = dz.

For convenience, we shall prescribe the location of the zeros of f(z)

and hence we shall know "a priori" the location of the first class of zeros

of f1 (z). It will remain for us to determine the location of the second

class of zeros f'(z), namely, those of F(z). From eqs. (2,1) and (2,2),

we see that these are the zeros of the function

n

in which the m. are the positive integers.

J
In order to obtain seme knowledge about the problems under considera¬

tion, we shall interpret the zeros of f'(z) from the stand point of physics,

geometry, and function theory. Observe that we will not use the fact that

mi are positive integers in our physical and geometrical interpretations

« o
as theorems concerning the zeros of a rational function F(Z) = ■■ - ,
f (z)
whose decomposition into partial fractions has the form (2,3), withm^ as

arbitrary real constants. But in our function - theoretic interpretation,

we shall restrict m. to the positive integers.

J
Physical Interpretations.—In place of F(z), let us introduce its con-

iueate imaginary:

T
Ruel V. Churchill, Complex Variables and Applications, New York,
i960, p. 162.
 8
It may be represented by a vector having direction from Zj to z and

having magnitude of m. times the reciprocal of the distance fan z. to z.

* J
In other words, the jth term may be regarded as the force with which a

fixed mass (or electric charge) m. at z. repels (attracts if m <0) a movable

J J J
unit mass (or charge) at z, the law of repulsion being the inverse distance

law*

An equivalent interpretation may be made in terms of masses repelling

according to the inverse square law. The derivative of this result can be
1
found in books on Newtonian Potential Theory.

If an infinite, thin rod L. of linear mass (or charge) density m./2

« J
passes through the point at right angles to the z-plane;the resultant

force upon a unit particle at z due to the particles of L repelling ac-

J
cording to the inverse - square law is a force directed along the line z.
J
to z and inversely proportional to the distance z. to z. This means that
v

we may interpret the imaginary conjugate of F(z) as the resultant force at

z, in the Newtonian field, due to a system of n infinitely thin rod3 at

z , J * 1,2, ...n.
J
Corresponding to each of these physical interpretations of the func¬

tion F, we have a physical interpretation of the zeros of F(z). These zeros

are the positive of equilibrium.

Theorem (3,1). The zeros of the function defined by

?
F(z) = £
jal
mi/(z - z.), with all m
J J
real, are the points

of equilibrium in the field of force due to the system of p masses (point

charges) m at the fixed points z. repelling a movable unit mass at z

J J

O. D. Kellogg, Potential Theory. Berlin, 1929, p. 10. ex.

 9

according to the inverse distance law.

Geometric Interpretation.—Let us begin with the case p ■ 3 when eq.

(2,3) becomes
m_ i&2

(M) F(z)
z-i,
7
z-z,
" “3
Observe that F(z) has two zeros z'^ and z’^» unless n ■ m-^ + “ 0

when it has only one finite zero z'^. We wish to know the location of the

points z'^ and z%^ relative to the triangle with vertices z^, z^, and zy
Let us investigate this condition by the following:

^ Theorem (U,l). The zeros z1^ and z'^ of the defined function F(z) ■

^ m.(z - z )-1 are the foci of the conic which touches the line segments
Js> 3 5 .
T_, T . and T . which divides these segments into the ratios m : m_, in :
3 12 1 2’ 2
vxy and m^: m^, respectively. If n - + m^ + =0, the conic is an

ellipse or hyperbola according as > 0

or Oj where as, if n ■ 0

but if V « m z +mz +mz t 0, the conic is a parabola whose axis is

112233
parallel to the line joining the origin to the point V.
 10

Proof:

We want to prove this theorem in the case leading to the ellipse.

Let us make use of a characteristic property of ellipses. Considering
that at any point P of an ellipse the focal radii should lie to the same
side of the tangent and make equal angles with this tangent. Let us con¬
sider the line segment z^z2 and the point on this line. By definition,

(4,2) T
3 = (ra z
l 2 +
Vl^ / (m
i +
V 5 i,e*»

(4,3) “i /(T3 “ + m
2 " z
2)
x
°*
We may on combining the fractions in eq. (4,1) write

(4,4) F(z)
Cl-l.) Ci-2Ü C2-Zj)
Thus, we have
m
F(I ) 3 . hCTt-2.')CT3 — 2i)
7J-2, CT3-H,)(T3-^)lT3-23)
i«G« <
(it,5)
Ca.-T})Ciy-Ts)
n
In the case m^m^ 0, we observe that nm^ ”7 0 and that is an in¬
ternal division point of the line segment a^a . From eq. (4,5), we have the
relation
(4,6) arg - 7* ) - awf ~^r -0
^ -r3) ** at=ry~
As shown in figure above z1^ and z‘2 must both lie on the same side of the
z1z2, and angle o( = angle P.
In the case m.nu0, we observe that nm Z. 0, and T is an external
± 3 3
division point of the line segment z^Zg. Fran eq. (4,5) we have

(4,7) arg . Ù.I r.Ti) _ JL. ^

0.
C2,-V & trÿ ■“* (^r)
 11

Therefore if z^ is taken at point z’^, then zlf and z’^ must be on

the same side of the line 35^2^ 311(1

angle o4 =
angle -ft . In all cases

in which > 0, the points z^ and z ^ are the foci of the ellipse

described in Th. (i*,l).

Function - Theoretic Interpretation.—We wish to undertake a classi¬

fication of polynomials, "purely function - theoretically", by entirely

ignoring the representation of polynomials and characterizing them (by their

critical points, distinct zeros, etc.).'*'

For a polynomial

(5,1) f(z) ■ (z - zx) (z - z2) ... (z - zn),

let us interpret the q distinct z'^. of its derivative

(5,2) f'(z) = n(z - z»^1*1 (z - ^P2

z'2)*
(z - ^ p. * n - 1,
q
/ J
from the point of view that f(z) is an analytic function of the complex

variable z. That is, w = f(z) maps any finite region S of the w - plane, the
2
mapping being conformal, except at the q points z' .. Specifically, if two
()

curves of the z-plane that intersect z’^. at an angle LjJ , they map into

two curves of the w - plane that intersect at an angle (p. + 1) 4>. For
J
this reason, the zeros of f1(z) are called the critical points of f(z).

Now let us consider, in particular, the families of curves:

R f(z) = a a
| f(z) | c

I f(z) = b arg f(z) = d

Konrad Knopp, Theory of Functions, New York, 19U5, p. 137.

2
Dr. Lonnie Cross, "Conformal Mapping". (Lecture delivered to class
in Complex Variables, Atlanta University, Atlanta, Georgia, April 17, 1962).
 12

where a,b, c(c >0), and d are real constants. These curves map into the

families of curves in the w- plane R (w) = a, I(w) « b, w * c and arg

1
w = d, respectively.

Curves R f(z) * a to arg f(z) = d, and their multiple points occur

at, and only at, the zeros z'. of f'(z).

•J -
Observe, if we set w * u + iv, and if we write jf (z) | * c as
2 2 2
F(x,y) = u + v - c =0,

then we discover that the multiple points of the curves jf (z) | * c are

the common solutions of the equations: . / .

We now conclude from these equations, since f(z) f 0 on the curve

j f(z)| = c, that every multiple point of this curve is a zero of f'(z).

Let us state some facts concerning the above families of curves. We

are interested in R f(z) » a and I f(z) = b because, when a * 0 and b * 0,

they are curves whose asymptotic behavior is the basis of Gauss' fourth

proof of the Fundamental Theorem of Algebra. Also, using these curves,

Liouville has given us a generalization of Rolle's Theorem, namely, that

the arc z"z of the curve I f (z) » 0 joining a pair of zeros, z and z ,
X 2 1 2
of f(z) passes through at least one zero z* of f'(z). By considering the
1
2 3
Riemann surface for w * f(z), De Boer was able to show that any k - branched

curve, R f(z) » a or I f(z) * b, not passing through a point z . divides the

J

1
Morris Marden, op. cit., p. 12.
2
E. G. Phillips, Function of a Complex Variable, New York, 1961, pp. 27»
3
De Boer, Extension of Rolle's Theory, Vol. 19 (188U), pp. 207-08.
 13

plane into k + 1 regions, each containing one less zero of f'(z) than f(z).

Also let us observe the curves j f(z)| *= c. From (5,1) and |f(z))= c,

we see that each is the locus of a point z, which moves so that the product

of its distance from the point z - is the constant c. As such, each is called
J
a lemniscate and the points z. are called its poles. When n » 1, it is a
J
Cassininan Curve, for zj_ = -1 and z^ * 1 with c = 1/2 (two small ovals),

c = 1 (Bernoulli lemniscate) and c » 2 (single large oval).

Assuming the points z. to be distinct, let us observe what happens in

J
the lemniscate as we allow c to increase from 0 to 0O. For c very small,

we infer from Th. (U,l) that the lemniscate consists of n small ovals each
containing a point z^> For any value of c, curve f(z) = c consists of N

closed Jordan Curves J^, J^, ...J^, Nin, the number N being a non-increas¬

ing function of c. Among these curves, no encloses J^, i / k. If c =

j f(z' ) I , j = 1,2, ..., or q, the curve passes through the zero z* . of

f*(z) and consequently has a multiple point at z' , p.+l tangent lines at
J J
z' being equally spaced. If, however, c Zj f(z’^)
J
I for j * 1,2,..., ki q
1
J
and c f(z‘J/ for j ** k + 1, k + 2, ..., q, i.e., if exactly k of the
J
points z1 lie exterior to the curve, then the lemniscate consists of

N * 1 + px + p2 + ... + pk

Jordan Curves J., each one of which contains one less zero zx . of f' (z) than
i* j
zeros z. of f(z), all counted with their multiplicities.
J
The use of the lemniscates If Ml ‘ c and the properties of the zeros

of f'(z) provides a method of determining the location of the critical points

1
of the Green's function of a region R. That is, if R is an infinite region

Tom M. Apostol, Mathematical Analysis, Reading, Massachusetts, 1957,

p. 28U.
which has a finite boundary B and for which the Green's function G(x,y)

with poles at infinity exists, then we may select a sequence LpI^L^ •••

of lemniscates which approaches B montonically. If the equation of is

f c
I k^z^a k* the Green's function for all the exterior of is

\(x>y) = ( l/n^.) log ( /fk(z)/)/ck,

when nk is the degree of the polynomial fk(z). The critical points of

Gk(x,y) are clearly the zeros z'j^ of f'k(z). If we now apply Hurwitz's
Theorem (Th. 1,5), we obtain the critical points of G(x,y) as the limit
points of z' .
 CHAPTER II

THE CRITICAL POINTS OF A POLYNOMIAL AND SOME

OF THEIR GENERALIZATIONS

Convex Enclosure.—In the previous chapter, we found that any criti¬

cal point (zero of the derivative) of the polynomial

f(z) = (z - z-^1 (z - z2)m2 ... (z - z/P ,

if not a multiple zero of f(z), is a zero of the function

The relative position of the real zeros and critical points of a real

differentiable function described in the well known Rolle's Theorem that be¬

tween any two zeros of the function lies at least one zero of its derivative.

But, Rolle’s Theorem is not generally true for analytic functions of a com¬

plex variable. For example, the function f(z) e2L - 1 vanishes for
27T zi
z = 0 and z = 1, but its derivative f'(z) * 27Tie never vanishes.

This ideal leads to the question of Rolle's Theorem being valid for the

polynomials with complex coefficients.

We shall attempt to answer this question, not with respect to Rolle's

Theorem, but with respect to a corollary of Rolle's Theorem which states

that any interval of the real axis which contains all the zeros of f(z) also

contains all the zeros of the derivative f'(z). A more general theorem

states that a line - segment L (not necessarily on the real axis) which con¬

tains all the zeros of a polynomial f(z) also contains all the zeros of its
1
derivative.

F. Irwin, Relation between the Zeros of a Rational Integral Function

and Its Derivatives, Amer. Math. Soc., XVI (191U-lf>), p. 138.

15
 16

Lucas' Theorem (6,1). Any convex polygon K which contains all the

zeros of a polynomial f(z) also contains all the zeros of the derivative

f (z).

From a physical point of view, this theorem is a consequence of Gauss'

Theorem with m. as positive integers. If the zeros of f'(z) are either mul-
w

tiple zeros of f(z) or the positions of equilibrium in the field of force

due to masses at the zeros of f(z), then in either case the zero of f'(z)

must lie in or on any convex polygon enclosing the zeros of f(z), (See

figure 6,1).

(Fig. 6,1)

Analytically, by Th. (1,1) if z', a zero of f'(z), were exterior to

K, it could not be a multiple zero of f(z). Moreover, the angle subte nded

at z' by K would be A(z'), where 0 ^A(z') ^ TT . Hence, if drawn from z,

each of the vectors (-w.) of formula (3,1) would lie in A(z') and therefore

+
each vector would equal -m jW^. Hence, by Th. (1, ), F(z') = - (w^+w^ •••

... +w)=0. This contradicts our assumption that z' is a zero of f'(z),
P
may be exterior to polygon K.
 17

Prom Lucas* Theorem, we may infer another theorem, Th. (6,2). Any

circle C which encloses all the zeros of a polynomial f(z) also encloses

all the zeros of its derivative f'(z).

If K is the smallest convex polygon enclosing the zeros of f(z), then

K lies in C and hence by Th. (6,1) all the zeros of f (z), being in K, also

lie in C.

Conversely, Th. (6,1) follows from Th. (6,2). If Th. (6,2) were valid,

through each pair of vertices of the polygon K of Th. (6,1) we could draw

a circle which contains K and hence the zeros of both f(z) and f*(z).

The region K* common to all these circles would also contain all the zeros

of f(z) and f’(z). Since this holds for all choices of circles passing

through K, all the zeros of f'(z) must lie in the region common to all possi¬

ble regions K* in polygons.

The Critical Points of a Real Polynomial.—In Lucas' Theorem we treated

the zeros z - of f(z) as independent parameters. If we impose some mutual

J
restaints upon z., the requirement that the z. be symmetrical in a line or
J u
point, we can expect the locus of the zeros of f ' (z) to be smaller region

than that given by Lucas' Theorem.

Let us assume in particular that f(z) is a real polynomial and thus

its non-real zeros occur in conjugate imaginary pairs. Let us consider the

circles whose diameters are the line- segments joining the pairs of conju¬

gate imaginary zeros of f(z). These circles shall be called Jensen circles

of f(z)j (See figure 7,1)»

Ï
J. L. ¥. Jensen, Research on Theory of Equations, Acta Math. Soc.
Vol. XXXVI (1913), PP. 181-®.
 18

Jensen’s Theorem Th. (7,1). Every non-real zero of the derivative

of a real polynomial f(z) lies in or on at least one of the Jensen circles

of f(z).

(Fig. 7,1)

Proof:

Let us consider the equation

f'(z)/f(z) = ^
J /
the sum of the terms w_L = l/(x + iy - Xj. - iyx) and = l/(x+iy-xi+iy1 )

z
corresponding to the pair of zeros z^ * x^ + iy^ and 2 =
~ "^1 ^ias

imaginary part

XC^+^3 - -Hi(■*-*,)*>

whereas the term w^ = l/(x + iy - x^) corresponding to a real zero z^ = x^

of f (z) has the imaginary part

Kw^) - -y/ £*(x - x^)2 + y2^/.

Hence, sg I(w^ + Wg) = -sg y for every point z outside all the Jensen circles

and sg I(w^) = -sg y for every point z. Outside all the Jensen circles^
 19

(7,1) Sg I -sg y.

In particular, if z is a non- real point outside all the Jensen cir¬

cles f1 (z) = 0. Hence a result which proves the Jensen's Theorem.

Observe, that a non- real point on a Jensen circle can not be a zero

of f(z) if it lies outside all the other Jensen cirlces. Jensen's Theorem

supplements Rolle's Theorem in describing the location of the zeros of

f'(z) relative to those of f(z). A theorem which describes the number of

zeros of f'(z) is due to Walsh.^

Theorem (7,2). Let Ii£A xéfi be an interval of the real axisji

neither^ nor ft is a zero of the real polynomial f(z) or is a point in or

on any Jensen circle of f(z). Let K be the configuration of I and of the

closed interior of all the Jensen circles which intersect I. Then, if K

contains R zeros of f (z), it contains at least k - 1 and at most k + 1

zeros of f'(z).

Proof:

We shall prove this theorem by the theorem on Principle of Argument.

Let R be the boundary of the smallest rectangle which has parallel to the

coordinate axis and which encloses K. In view of eq. (7,1), R is mapped

by the function w = f'(z)/f(z) upon the w-plane into a curve which encir-

des the origin at most once. Hence,

and by eq. (1,2) the number of zeros of f'(z) within R differs by at most

one from the number of zeros of f (z) in R.

An immediate consequence of Th. (7,2) is the following corollary:

- - —

J. L. Walsh, On the Location of the Roots of the Derivative of a

Rational Function, Amer. Math. Soc. XXII (1920), pp. 12Ü-1U;.
 20

Any closed interval of the real axis contains at most one zero of

f'(z) if it contains no zero of f'(z) and if it is exterior to all the Jen¬

sen circles for f(z).

Some Generalizations.—From proofs given in the previous section, it

is clear that the Lucas' and Jensen's Theorems are essentially results re¬

garding the zeros of

F(z) = m ./ (z - z^), m ~p 0,
j-J
and that these results are valid when the positive numbers nu are not in¬

tegers. This expression is, however, only a special case of the linear

combination
n
(8,1) F(z) (z)
J J

where f.(z)
CB—
0 ft — ft * * » cz-£;r)
and where the nu are complex numbers
 21

(8,2) fl £ arg (7^) * ^ j-l,l,...,*.

Theorem (8,2). If all the zeros and poles of each rational function

f (z) entering in eq. (8,1) lie in a closed convex region K and if the m.
j w
(j ■ 1,2, ..,n) are constants satisfying in eq. (8,2), than all the zeros
*f1

of the zeros of the linear combination F(z) ■ £ m f. (z) lie in S(k, $),
.’si j J
a region which is star- shaped with respect to K and which consists of all

points from which K subtends an angle of at least <ÿ {TT- T)/(p + q).

Observe that the region S(K, /) may not be replaced by a smaller re¬

gion. If P:S is any point in S(K, /), two points Q^:t^ and Qg1 ^2

be found in K 3 • Denote by d^ and dg the distances of and

Q2 from P respectively and by w the angle formed by the ray PQ^ with the

positive real axis. Also let us define

/d p
h' [ - V (s
iJ * and

P + q ir
(S t )/d e
[ - 2 2J

Then, since Ihl ■ W ■lj arg = (p + q)w and arg ** (p + q)

(w + 0) + f = TT + (p + q)w. The vectors k^ and k^ are equal and opposite

and thus

k + k 0.
1 2
This means that the function

G(z) - fd2q(z - tl)P/d1P(z - t2)qJ +£e q z

^ * t
2^
P
/

has a zero at the point s, i.e., every point 3 of S(K, Ç&) is a zero of at
 22

least one function F(z) of type (8,1).

An application of Th. (8,2) can be made by introducing a polynomial

f(z) of degree P and N polynomials h..(z), j = 1,2, ...,n, each of degree at

most p - 1. Then

F(z). f(t) . w.t. fe) + ** • • : +>”»IA &

n fT)/ ~h * * * ' +

is a polynomial of type (8,1) with q = 0 and with

f(z) - h (z) = (z - et ) (z - B» ) ... (z - a ) .

J jl J2 jp
The a are clearly the points at which f(z) = h (z)
JP J
and the zeros of F(z) are the points where
n n
f(z) = 2. m h (z)/ «£-• m..
J^ J
Hence, we have established the following Mean - Value Theorem for

polynomials.

Let f(z) be a pth degree polynomial, let each h (z) (j l,2,...,n)

J
be a polynomial of degree at most p - 1 and let m.: ■ be complex constants
J
satisfying inequalities

— 1,1, *
arg. hnj é V + r*. JA+TT

If all the points Z at which f(z) = h (z) for at least one j ( j = 1,2,
J
...,n) lie in a convex region K, all the points at which

f(z) =
4
Z-.
£
m,h.(z)/ m.
3
j =.1 33

lie in the star - shaped region S(K, (7T — r)/p)."^

^Morris Marden, On the Zeros of Certain Rational Functions, Tran. Amer.

Math. Soc. XXXH (1930), pp. 650-68'.
 CHAPTER III

INVARIANTIVE FORMULATION

We have proven theorems concerning the zeros of the logarithmic deri¬

vative of the function o

* ,, , »;
1 n = £■ m
b
(?,1) f(z)-VC2-2j) , "

and obtained the results largely by use of Th. (1,1) and Th. (1,1;).

We wish now to investigate further generalizations, if any, which may

be derived by the use of conformal mapping.1 For example, we know by vir-

ture of the Lucas Theorem that any circle C containing all the zeros of a

polynomial f(z) also contain all the zeros of the derivative f'(z) off(z).

Since we may map the closed interior of C conformally upon the closed ex¬

terior of a circle C, can we then infer that, if all the zeros of f(z) lie

exterior to C, so do all the zeros of f'(z). Certainly not in general, as

we see from the example, f(z) = - 8 with C’ taken as the exterior of the

circle | z J = 1.

Let us consider now to generalize Th. (6,2) so we might obtain a re¬

sult which will be invariant relative to the non- singular linear transfor¬

mation

(9,2)
L- \if\--fO-
Specifically, let us denote by the points into which the zeros z. of
J 1
f(z) transformed by (9,2) and by Z'k the points into which the zeros zof

Marden, op. cit., p. 37-38.

23
 2h

f’(z)/f(z) are transformed, i.e.,

Z /6>J
(9,3) . -^ J+ Ic*

Clearly, the Z. are the zeros of F(Z), the transformation of f(z), name-
V

iy,

(9,ii) F(z).fr2+0

The Z’ are not in general the zeros of the logarithmic derivative

k
of F(Z). Let us investigate the choice of the m. necessary and sufficient

for a finite Z'k to be such a zero.

wmay}
The logarithmic derivative of F(Z) calculated from
era (9,U) is
is _

F'Q.) - hi , +
(9,5)
Faj ft (FIT? )cn*5
we thereby obtain

Jn
F'a'x) i
(9,6)
FU‘k) ri^-ts
Thus a necessary and sufficient condition for F’(Z'k) = if Z'k‘+0° is that

rn = 0. This condition will be satisfied if we choose )T - Oj that is, if

we select for (9,2) any non- singular linear integral transformation

(9,7) z = AZ + B, A^O •

Thus, if we restrict the transformation to translation, rotations, and those

of similitude, the zeros of f(z)/f(z) transform into those of F*(Z)/F(Z)

when the m. are chosen as arbitrary positive or negative numbers.

v

To satisfy the condition when Y* 0 and Z'k ^ QO , we must choose

n * 0. This implies that not all m. may be of the same sign. In particular,
J
 25

it excludes the case that all m. are positive integers. That is, under
J
the general transformation (9,2) the zeros of the logarithmic derivative

of a polynomial are not carried in to the zeros of the logarithmic deriva¬

tive of F(Z).

Eq. (9,6) does suggest that in place of the derivative f'(z) of a

given nth degree polynomial f(z), there be introduced the function

(9.8) f^z) = nf(z) - (z - Zq) f*(z).

Like f *(z), f]_(z) is a polynomial of degree n - 1. It is furthermore a gen¬

eralization of the derivative in the sense that

(9.9) lim ff (z)/(z - z)l - f’(z).

-, J- o —

n
The function f^(z) has been called by Laguerre the emanantn of f (z)

and by Polya - Szego "the derivative of f(z) with respect to the point ZQ,"

but we shall be conventional and call f-^(z) the polar derivative of f(z)

with respect to the pole ZQ or simply the polar derivative of f(z).

The zeros of the polar derivative are:

(a) the point Zq, if f(zQ) = 0;

(b) the multiple zeros of f(z), and

(c) the zeros of the function

(9,10)
JL£5L_ =-=^--2.
Since (9,10) is the logarithmic derivative of

n
-f(z) (z - zQ) ,

a function of type (9*1) with a total "degree" of zero, the zeros of (9,10)

and hence those of f^(z) are invariant under the general linear transformation
 26

z
rTTT'
Circular Regions .—We have associated with every nth degree polynomial

f(z) and (n-l) st degree polynomial called the polar derivative of f(z),

namely

f^z) = nf (z) + ("J -z) f * (z) ,

whose zeros remain invariant under the linear transformation (10,2). Since

f^(z) is a generalization of the ordinary derivative, its zeros may be ex¬

pected to satisfy some invariant form of the Lucas Theorem that any circle,

C, containing all the zeros of f'(z). In order to find the corresponding

theorem for the polar derivative, we need to consider the class of regions

which includes the interior of a circle as a special case and which remains

invariant under the transformation (9,2). This class is known as "circular

regions," consisting of closed interiors or exteriors of circles and the

closed half- plane.

We shall find the following lemma very useful in our future work in¬

volving circular regions.

Lemma (10,1). Let C(z) 25 /z - a» so that C(z) = 0

is the equation of the circle C with center at a point a and radius Y~«

Let z-,, Z and w be any three points connected by the relation w, =

1
— - -1
(Z - z-^) and let C’ be the circle with center at a' and radius Y"

where

(10,1) a’ = (Z - a) /0(Z) and f - Y/ / / •

Then the point w^ lies inside or outside the circle O’ according as the cir¬

cle C does or does not separate the two points Z and z^.
 27

Proof:

. 2 2
Calculate C ' (w^) | v1 - a« / - Y'
1 X 1
C'frj) JjZ - zx) -(Z - a)C(Z) J [(Z - Z;L)

2 2
- (z-a)c(zf1D -{► /c(z) Ü.
Using the identity AB + AB ./A/2 + IBÏ -|A - B I2 in the form (Z - a)

(Z - Z]L) + (Z - a) (Z - Z]L) = / Z - a /2 + /Z - | *
2 ,
-fz^-af
2
,

we obtain C (w^) = /w^/2 /C(z^) /C(Z)jl

If one of the points z^ and Z is inside and the other outside C, then

C(z^)/C(z) ^0 and hence C* (w^) 0, implying that w^ is inside circle C*.

If the points z^ and Z are both inside or outside circle C, then C(z1)/C

(Z) "> 0 and hence C’(w^) ^ 0, implying that is outside circle C. Hence

the result. Q.E.D.

Zeros of the Polar Derivative.--We are now ready to state the invar¬

iant form of the Lucas Theorem due to Lacjuerre.

La^uerre's Theorem (Th. 11,1). If all the zeros z_. of the nth de¬

gree polynomial f(z) lie in a circular region C and if Z is any zero of

(11,1) fx(z) = nf(z) + (J -z)f'(z),

the polar derivative of f(z), then not both points Z and J may lie outside

of C. Furthermore, if F(Z)% 0, any circle K through Z and either passes

through all the zeros of f(z) or separates these zeros.

Because of the importance of Lacjuerre ’ s Theorem, we shall give two

proofs of it.

The first proof will be concerned with spherical force fields. Let

us assume that Z and ~S. are both exterior to the region C. Since all the
 28

zeros of f(z) lie in C, it follows that F(Z) f 0 and hence, also Z f^

Through Z a circle P may be drawn which separates the region C from the

point . As a zero of f(z), Z must satisfy the equation

(11,2) f^Z)/ [ (^-Z) f(Z) jf

[f (Z)/f(Z)J - 0
and must be an equilibrium point in a plane force field due to particles

of total mass zeroi With this plane force field may be associated a spheri¬

cal force field in which P., P, and Q and circles C and f1 ' correspond re-

spectively to points z^, Z, and ^ and circles C and P and in which the

mass at P is m and the mass at Q is -n = -(m_ + m + ... + m ). The

j j 1 2 p
force at P due to the pair consisting of m at P and of (-m ) at Q
3 3 j Ù
acts in the direction of the circular arc P^PQ and toward the side of cir¬

cle P ' not containing C1. The vector Ë j's are all drawn from P to

points on the same side of the tangent line to P ' ' at P. According to Th.

(1,1), they cannot sum to zero. This means that P cannot be an equilibrium

point in the spherical field and that Z cannot be an equilibrium point in

the corresponding plane field. This contradiction to our assumption concern¬

ing Z proves the first part of Lequerre's Theorem.

To prove the second part, let us assume that a circle K through Z and

-7 has at least one z. in its interior, no z in its exterior, and the re-
0 3 3
maining z^ on its circumference. The corresponding circle K' through P and

Q on the sphere then has at least one P^. in its "exterior" and the remaining

P.'s on its circumference. The forces . are then directed from P along
J J

the tangent line to K* at P or to one side of this line and hence cannot

sum to zero. This contradicts the hypothesis that Z is a zero of f^(z) and
 29

so at least one z. must be exterior to K. Since a contradiction would also

J
follow if K were assumed to have at least one z, in the interior and no zJ
k j
in its interior, we conclude that K must separate the z. unless it passes
J
through all of them.
The proof just completed was based upon the properties of equilibrium
points, but the second proof of Laquerre's Theorem (11,1) will be based upon
the properties of the centroid of a system of masses. If Z is any zero of
(11,2), it satisfies the equation
n _ f»J3L
(u
>3) ïïi ~ 17,

On substituting into this equation

1 1
(11,U) T = (Z - J) , wj = (Z - 2^) ,
we derive the relation j>

w= fô »; i V;) h n- 2 > ,

w is the centroid of the system of masses m, at the point w .

0 j
Successive Polar Derivatives.—Corresponding to a given nth degree
polynomial f(z), let us construct the sequence of polar derivatives
(12,1) f (z) - (n - k + l)fk-+> x(z) + (J - z)f'k _ x(z),
k = 1,2 ...n, with f (z) * f(z). The poles fa may be equal or unequal.
(k)
Like the ordinary derivative f (z) off(z), the kth poles deriva¬
tive fk(z) is a polynomial of degree n - k. Just as the position of the
(k)
zeros of f (z) may be determined by repeated applications of Lucas
Theorem, the position of the zeros of fk(z) may be determined by repeated
applications of LZquerre’s Theorem. The result may be stated as
 30

Theorem (22,1). If all the zeros of an nth degree polynomial f(z)

lie in a circular region C and if none of the points J1, n - 1)

lies in region C, then each of the polar derivatives f^(z), f^Cz), ...

fk(z) in eq. (12,1) has all of its zeros in region C.

By Laguerre's Theorem, all the zeros of f^(z) lie in Cj hence, all

those of f (z) lie in C; hence, all those of f (z) lie in Cj etc.

Let us express the polar derivative f (z) directly in terms of f(z)

K

and From eq. (12,1) we successively eliminate f (z), f (z)

... *fk -
We find
£ (K-J ) ! cl*-), K-i) Sj fa)&)j

where (z) is the sum of all the products of the differences (^ - z)

taken j at a time, i = 1,2,...kj f^(z) is a generalization of the kth de

rivative of a polynomial is the sense that as ^^ oo , j * l,2,...kj

U y- — 75 ^— = lCK)&.
C?I * * " 0?K—^ '
Let us put f (z) still in another form which will clearly show the
K

relation of its coefficients to those of f(z). For this purpose, let us

write f(z) and fk(z) in the form

(k) i •
where we define A.j ■ 0 for and j -y —A .

Substituting into (12,1) the expression for fk _ ^ (z) and f^ (z),

 31

equating the combined coefficient of zJ on the right side of (12,1) to that

on the left side and simplifying the resulting formula, we have

(k) , (k-1)
(k-1) (k-1)
(12,3) A. (n - k + 1) (A + A 2 v).
3 3 + 1

Let us show that by repeated application of (13,3) we may derive the formula

(k)
(12,U) A n(n - l)...(n - k + 1): £ cr (k,i) A
3 * ZTo
A= 0 1 +

where 0*(k,i) is the symmetric function consisting of the sum of all possi¬

ble products of j taken i at a time. First, we note that for k = 1

eq. (12 ,U) is the same as eq. (12,3)- We want to show that, if (12,1;) is
(k + 1)
valid, A. will be given by (12,U) with k replaced by k + 1, i.e.,
J
*

- nCn-1) ••• (n~«) ZjjrCXUj + 3

= n Cn-/) ••• i*-*) %
=o
trCx+i.iÏA' u
-
A

Thus eq. (12,1;) has been established by mathematical induction.

 CHAPTER IV

COMPOSITE POIZNQMIALS

Apolar Polynomials,—We have thus far been concerned with the relative

position of the zeros of certain pairs of polynomials. In chapter I and H,

the pair consisted of a polynomial and its ordinary derivative and in Chap¬

ter III, the pair consisted of a polynomial and its polar derivative. Hence,

we are now able to apply these results to the study of the comparative loca¬

tion of the zeros of other pairs or sets of related polynomials.

Let us begin with a pair of so-called apolar polynomials. Two poly¬

nomials

(13.1) f(z) - X C(n,k)

h-Q
V
n
g(z) - 2Z c(n,k) B Z
k 9 AnBn ^ °»
K=o
are said to be apolar if their coefficients satisfy the equation

(13.2) AQBn - C(n,l) J^Bn _ ± + C(n,2) A^ _ 2 + ...

+ ( -I)" A B =0.
n n
There are an infinite number of polynomials which are apolar to a given
3
polynomial. For example, the polynomial z + 1 is apolar to the polynomial
3 2
z + 3 o(z + 3 Pz + 1 for any choice of the constants o( and .

z
Let us denote by 4^,*. , 'n the zeros of f(z) and by^Tl, ^2,

n the zeros of g(z) such that

(13.3) f(z) - An(z-z1)(z-z2) (z-z^)

s

B
g(s) - n«-4> •••• <z
In terms of the elementary symmetric functions

32
 33

(13,U) S(n,p) =
£ Va* jp *

<7(n,p) =
’ SùP *
the sum of products of these zeros taken p at a time, we may substitute

(13.5) C(n,p) * (-1)P S(n,p) An

C(n,p) Bn_p = (-1)P **(n,p) Bn

into eq. (13,2) and so obtain the following criterion for apolarity.

Theorem (13,1). Two nth degree polynomials f(z) and g(z) are apolar

iff the elementary symmetric functions S(n,p) of the zeros of f(z) and the

elementary symmetric functions (f (n,p) of the zeros of g(z) satisfy the

relation

(13.6) ^ (-1)^ C(n,k) S(n,n-k) (T (n,k) - 0.

A simpler method for constructing a polynomial g(z) apolar to a given

polynomial f(z) is as follows.

Theorem (13,2). If the coefficients a,K - C (n,k) Ak of the polynomial

f(z) satisfy the linear relation

n
(13.7) L M ■ t = 0 ?

then the polynomial

(13.8) g(z) * L £(t-z) ^ is apolar to f(z), for as

(t-z)n = 2T (-l)n C(n,k) Zn’ktk and

g(z) = L f(t-z)nJ = £ (-l)n“ l^C (n,k),

of eq. (13,7) is seen to be of the form (13,2) with B ^ = (-l)11"1^ .

We want to consider now the relative location of the zeros of two apolar
 3U

polynomials, we will use the fundamental result of Grace.

Grace's Theorem (13*3)» If f(z) and g(z) are apolar polynomials and

if one of them has all its zeros in a circular region C, then the other will

have at least one zero in C.

Proof:

Let us assume that all the zeros z.,,z0,...z of f(z) lie in a circu-
1 z n
were a1 1 to
lar region C. If the zeros ^ -- -^-e ex
"

terior to C, «ill the zeros of each polar derivative f^Cz), k=£L,2,...n-1,

given by eq. (12,1) would according to Th. (13,1) also lie in C. Observe

that we may write

(n-1) (n-1)
(13,9) fn-1(z) = A
o + z>

where
(n-1)
nj + <T (n-1,1)^ + cT (n-l,2)A2 + Cr (n-l,n-l)Az_1^
U
(n-D . r
A + (T ( n-1, n-1) A^*^ .
1 = h; fei «*(n-l,l)A2 + <r (n-l,2)A^ +
Recall eqs. (13,it) and 0-3,9); hence the following

0*(n-l,k) + 0"(n-l,k-l) = <T{n,k).

Therefore^

f x (Jn) = n
/{A0 +
+ (T{ n-2)A2 + , + r (n,n)Anlf

(13,10)anf / Bn‘fA0Bn-G(n,l)A1Bn- 1 + + ... +

(-1)11 C(n,n)AnBQ^.

Since f(z) and g(z) are apolar, eq. (13,10) ■ ^ fR_1 (J n) = 0. The

point J n is therefore the zero of fn_^(z) and must lie in C.

must 1 e
Hence, at least one of the zeros ***"?n ^
z
in any circular region C containing the zeros z^, ^ 2> •••• n °f f(z) He
 35

in any circular region containing all the zeros of g(z). Q.E.D.

Frcsn Grace's Theorem, we may deduce the following corrollary due to
Takagi:
If f (z) and g(z) are apolar polynomials, any convex region A enclosing
all the zeros of f(z) must have at least one point in common with any con¬
vex region B enclosing all the zeros of g(z).

Proof:
Let us assume that A and B have no point in common, then we could
separate them by means of a circle C enclosing say A, but not containing
any zero of g(z). Therefore, a contradiction of Grace's Theorem. Hence
the result follows. Q.E.D.
Combinations of a Polynomial and its Derivative.—Let us consider an
1
important theorem due to Walsh.
Theorem (11;,1). Let
n
01;,1) f(z) = %
j=o J ■ a 7T (z-^/ ),
J

01;,2)
. n
g(z) " ± bV- b ir (*-£).}
J=o J J

(Z)
OU,37 h(z) = £ (n-j)/ b .

If all the zeros of f(z) lie in a circular region A, then all the zeros of
h(z) lie in the point set C, consisting of n circular regions obtained by
translating A in the amount and direction of the vector P..
J

Proof:
Let us assume Z to be any zero of h(z), i.e.,

J. L. Walsh, On the Location of the Roots of Certain Types of Polyno-

mials. Amer. Math. Soc. XXIV (1922), pp. I63-8O.
 36

(j)
04,4) h(z) - j^(n-J)' n-j f (Z)
b o

Since eq. (14,4) is a linear expression in the coefficients of f(z), we can

assume that f (z) is apolar to the polynomial obtained by replacing f (z) in

(14,4) by (Z-z)n, that is,

n.
(n-.i) ^ bn-J d ^ (Z-z)n / (dZ)*^ « n/ g(Z-z).
è (n-j)/
js.O -J

According to Grace's Theorem, at least one of the n zeros Z-0. of g(Z-z)

J
must lie in the circular region A containing the zeros of f(z). That is,

Z * + Ê. , where is a point of A. Q.E.D.

3
This result leads us to a corollary which enables us to determine

where all the zeros of a linear combination lie, i.e.;

If all the zeros of an nth degree polynomial f (z) lie in a circular

region A, all the zeros of the linear combination

(14,5) f (a) ■ f(z) -A,f'(z)

lie in the point- set comprised of both A and A' = A' being the

region obtained on translating A in the magnitude and directions of the

vector (n A1).

Proof:

If we consider the above corollary with conjunction to cor. (13,3),

then the apolarity of f(z) and g(Z-z) permits us to infer that any convex

region A containing all the zeros of f(z') must over lap every convex

region B’ containing the zeros of g(Z-z). Since B' may be considered as

locus of the point z = Z-ft when ft varies over a convex region B containing

the zeros of f(z), each zero Z of h(z) is expressible in the form z=&( +

are points of A and B respectively. In other words, the following result

 37

has been proved.

Theorem (lij.,2). Let the polynomials f(z), g(z) and h(z) be defined

by eqs. (1U,1), (1U,2), and (1U,3). Then if all the zeros of f(z) lie in

a convex region A and all the zeros of g(z) lie in a convex region B, then

all the zeros of h(z) lie in the convex region C7 which is the locus of the

points X - +'ft when the points o/ and $ vary independently over the regions

A and B respectively.

If g(z) is taken as the polynomial

g(z) - zn 1 (ü - n X,),

the convex region B may be taken as the line segment joining the points

z = 0 and 2= n This establishes the following result:

Let f(z) be an nth degree polynomial having all its zeros in a convex

region A and let A (z) be the polynomial

f\ (z) = (1 - XfZ) (1 - JLXZ) ... (1 - Ay»z).

Then all the zeros of the polynomial

F(z) = A (D)f(z) D = d/dz,

lie in the above defined convex region, kp.

Of special interest is the case that f (z) = z11 and that A (z) is an

nth degree polynomial for which points lie in a convex sector S with

vertex at the origin. Since each point f] also lies in S, each region

A^ will lie in S provided the preceding region A^_1 lies in S. The region

A may be taken as the point z = 0. Since the corresponding region A^ will

be the line segment joining the points z = 0 and z -nK the region A^

lies in S and hence all the subsequent regions k^,ky..,.k^1±e in S.

Since the 'X.j^ are the zeros of the polynomial

 38

g(z) - z:n A (l/z)

we have proved that .*

If all the zeros of the polynomial

n
g(z) » bQ + b^z + + bn z
lie in a convex sector S with the vertex at the origin, then so do also all

the zeros of the polynomial

G(z) = bo + b^ + (b2z2 / 2 / ) + ... + (bnzn /n /).

When the point set Pis the locus of the zeros of h(z), we can make the
following application

If all the zeros of an nth degree polynomial f (z) lie in the circle

C: lzl±. t and if all the zeros of the polynomial

(j,) (z) = ^.0+ C(m,l) z + ... CCm^njA^z21

lie in the circular region K: ^z/£s ( z - h / , s ? 0,

then all the zeros of the polynomial

y(z) = V(z)^| f'(z) £(tz)/llj+ ...+/A0

lie in the circle P : Jz/£kmax (l,s). The polynomial ^(z) is of the

form of h(z) with
3
TX
f^zjE f(z), f2(z)H (ta) , Bfc= t /nj tnC (n,k)
and hence the corresponding g(z) is

g(z) * (z/t)n (f) (t/z)/nj .

If E^jEg, *" E
n 32,6 zeros 0 (a), the zeros of g(z) are =

t/E^. Here circle is the same as circle C, but circle is namely the
point z = 0. According to the above theorem, the zeros of ^(z), not in C,
 39

lie in the circles /"7 centers at z = 0 and radii

k

rk - *•// 1-Bkl - rl /•
From the condition JzJ îsjz - t J on the zeros of it now

follows that IM ^.fs. The zeros of Ÿ (z), including those in C, there*

fore satisfy

|z |«£ K max(l,s).
 CHAPTER V

THE CRITICAL POINTS OF POLYNOMIALS IN PRESCRIBED REGIONS

The Critical Points of a Rational Function.—We want to investigate

the derivative of a rational function which has its zeros and poles dis-
1
tribu ted over any number of prescribed circular regions.

Theorem (15,1). For j » 0, 1, ..., p, let f. (z) denote a poly-

j
nomial of degree n. having all its zeros in the circular region <T.C (z) £
J i • 0
0 where (T = + 1 and
Ù 2 2
(15,1) c (z) * /z-c /
J ' 3 i
Then every finite zeros Z of the derivative of the rational function

(15.2) f(z) - 0 f q * p,
&(*)■■■ fp(*)
satisfied at least one of {he p + 2 inequalities

(15.3) j = V> • • f

-m— - tm-*e>
j" Y j-_, 4

where m = n or -n according as j £ q or J > q, n « Z rn.,

j j -o
ê

We shall interpret Th. (15,1) from a geometric standpoint. Accord¬

ing as CT “ 1 or -1, the region C . defined by the inequality

J J

Morris Marden, On the Zeros of the Derivative of a Rational Function.

Bull. Amer. Math. Soc., XLEI (1936), pp. U00-U05.
CT*(z) 0 is the closed interior or the closed exterior of the

1
circle with c. as center and Y as radius* As A* = or -1, the
J J
quantity t. when positive is the sequence of the common external tangent
JJC

of the circles C (z) * 0 and C (z) » 0. When n ^ 0, the equation -

J K
E(x + iy) » 0 may be written in the form

(x2 + y2)P ♦ <f> (yp) ■ 0

where (f> (^r) is a real polynomial with a combined degree of at most 2p - 1

in x and y. Being of this form, E(z) ■ 0 represents a so - called p-

circular 2p - ic curve, a curve of degree 2p which passes p times through

each circular point at infinity. Hence, the curve E(Z)“ 0 consists of at

the most p-branches, each of which is a finite Closed Jordan Curve. That

is, when n 0, Th. (15,1)~ y that each zero Z of f* (z) lies in at

least one of the given circular regions C^ or lies in one of the regions

For n -àp O, , the above is the equation of a bicircular quartic, a result

which for the mj positive integers and the Cj interior of circles coincides

the result due to Walsh.1 Fig.(l5,l) illustrates the case that

I
J. L. Walsh, op. cit.. p. 197.
m = m » m and the regions C,, C0 and C_ are interiors of circles of radii
1- «■2 3 i 123
r, » =* f3 = 102 with center at ^ » (-3 -i), C2 * 3 - i and * 2i. In

that case, curve E(z) consist of two nested ovals*

For the continuation of our proof, let us denote the numerator in eq.

For
(15,1) by F-^(z), and the denominator by Fg(z). a31?- zero, Z of f'(z),

the expression F(Z) = F2(Z)F^(Z) - F^Z) Fg(Z) = 0 is linear and symmetric

in the zero of each f^(z)» We may select a point ^^ in each region ï Z

will also satisfy the equation obtained from F(z) * 0 by letting f .(z) ■
• J
Th[
(z • us we find that either Z for seme value of j for which

n. > 1 or Z satisfies the equation.

(15,1) 4"
j-o
•fwi
L ,

In the first case, Z lies in a region C. and thus satisfies the jth
j
derivative of the inequalities (15,3). In the second case, Z lies in the

locus described by the roots of equation (15,5) when ^ ^ ^ ^

are allowed to vary independently over the circular region CQ, C-^,

Cp respectively.

We want to investigate Walsh's Cross-Ratio Theorem. (15,2). If the

points ^ ±> ^ 2> 3» var

ying independently have given circular

regions as their loci, then any point z forming a constant cross-ratio

with^, p ^ 2> and ^ also has a circular region as its locus.

Proof:

For each j - 1, 2, 3, let f (z) be a polynomial of degree n having

3 3
all its zeros in a circular region C . If IL + n. = n .and if no point is
j J- 2 3'
common to all the regions C., then each finite zero of the derivative of
J
 U3

the function

f(z) » fx (z) f^z)/ f(z)

lies in at least one of the circular regions C^, C^, or in a fourth

circular region C. This fourth region is the locus of a point z, whose

cross-ratio ~ with the

points 7 , 7, and 7 has the
C2-30CÏ3-£) <rl «*2 °*3
constant value (-d0/m ) as , and describe the region C , C and
t 1 1 <r 2 1 2
respectively*

Regarding the Th*(l5,2), we may say that the same conclusion holds

for the zeros of the derivative of the reciprocal f_(z)/f_(z)f (z) of the
3 12
above function. Since the total "degree" of f(z) Is n ■ n +n -n_ ■ 0,
12-3 j
111.(15,2) may be restated in terms of the Jacobian of the two binary forms.

Q.E.D.

Polynomials with Two Given Zeros.—We have developed several theorems

on the locations of all the critical points of a polynomial f (z) when the

location of all the zeros of f(z) is known. We shall attempt to investigate

the extent to which the prescription of only some of the zeros of f(z)

fixes the location of some of the critical points of f(z).

1
J. L. Walsh, The Jacobians of Binary Forms, Amer. Math. Soc.,
XXVII (1932), p. 112.
 (Fig.16,1)

We may derive immediately from Rolle’s Theorem Z*AZ+B, A. ^ 0 and

thereby transforming the real axis into an arbitrary line L* This result

states that, if the zeros of a polynomial are symmetric on a line L, then

between any pair of zeros lying on L may be found at least one zero of

the derivative.

We want to know whether or not, given two zeros of a polynomial, we

may determine the location of a zero of f ' (z) even when no additional

hypothesis is placed upon the remaining zeros. An affirmative answer is

1 2
given first by Grace and later by Heawood.

Grace-Heawood Theorem. (17,1)* If z^ and z^ are any two zeros of

an nth degree polynomial f(z), at least one zero of its derivative f‘(z)

and wifch a
will lie in the circle C with center at £”(z-j+z2)/2^ radius

[^(1/2) | Z£-z2j (cot ÏÏ/n)J.

Proof:

We may without loss of generality take z^ «* 1 and z^ «-1. (see fig.

Ï
J. H. Grace, op. cit., p, 136.
2
Heawood, Zeros of Polynomials, Bull. Amer. Hath. Soc., Voli. XI
(1902), pp. 101-109.
l6,l). By hypothesis, we have upon
- v 2 n-2 n-1
f'(z) « aQ +a1z + agz ♦ . . . a^z + z

the requirement that

r+i
(16*,1) 0-f(l)-f (-1) - J ^ fy (t)dt « 2ao + 2^/3 * 2*^/5

Since eq. (16,1) is a linear relation among the coefficients of f * (z), we

may apply Grace's Theorem. Thus at least one zero f (z) lies in every

circular region containing the zeros of the polynomial

g(z) * ^l(z-t)n“1dt » (l/n) £(z-l)n - (zfl)n j/ .

But the zeros of g(z) are z= -i cot(k Tf/n), ksl, 2, ......n-1. That is,
K
not only at least one f'(z) lies in every circle C of Th.(l6,l) but also

58
that at least one aero of f(z) lies between the two points z 1 i cof*ftu

We can see that the Grace- Heawood may be used to prove the followings

If two zeros of an nth degree polynomial lie in or on a circle of

radius R, at least one zero of its derivative lies in or on the concentric

circle of radius R csc( 7T /n).

Proof:

Choose

f(z) 2i esc (2 7T /n)+- u exp (- 7fi /n) cos ( Tj/:njTtl

du* For this polynomial has on the unit circle the zeros z, ■ +1 and z
1 2
2/T i/n
■ e and its derivative has a zero on the circle / z| ■ esc ^/n at

the point z ■ e csc (TT/n). Q.E.D.

Mean-Value Theorems.—We want to derive resets similar to those of

%
section 16 above using the following Mean-Value Theorems. These theorems
* 1
were first proved by Marden and later treated as special cases. Both

Morris Marden, op. cit., p. 7.

 U6
theorems make use of the notation s(K, $ ) as in section 8 for the star¬

shaped region comprised of all points from which the convex region K sub¬

tends an angle at least <f) .

Theorem. (17,1). Let P(z) be an nth degree polynomial and let

>, ...
z be any m points of a convex region K.
m
Let <r
i
the mean-

value of P(z) in the points z., be defined by

Ml j

a r £ < ?(*: ) y
=l J
(17.1) J d=l ' J
where
**■ f Argfcj) *yu+r * /1+ir.
Then the star-shaped region S(K, (TT - f )n) contains at least one point
at which P(S) * (f .

If H denotes the smallest convex region of w- plane containing the

points W ■ P(z ), P(z ), ... P(z ), then according to Th. (17,2), applied
1, c m
to F(W) = ( U) - p(i )), (T is an arbitrary point of the region
* d
S(H, 7r- r ).
Theorem. (17,2). Let P(z) be an nth degree polynomial and let C:Z >*
Y (t) £ t real} a - t £ b J be a rectifiable curve1 drawn in a convex

region K. On curve C, let c< (t) be a continuous function whose argument

satisfies the inequality

/t - *rS * M.+As* ir.

Then, the star-shaped region S(K, (7T - f )/n) contains a point s i

(17.2) f f[vti)]* [rft)] a ; Pis)/ *[.

J-

1
Dr. Lonnie Cross, "Rectifiable Curves".. (Lecture delivered to the
class in Complex Variables, Atlanta University, Atlanta, Georgia, November
16, 1961).
 kl
Theorems (17,1) and (17,2) could be combined into a single theorem

if Stieltjes Integrals'1' were introduced into eq. (17,2). The proofs of

both theorems are essential the same. That is, to prove the first, let us

write eq. (17,1) In the form

(17,3)
J
i-'
Denoting by a ,a ...a the points at which P(z) assumes the value <T, we may
12 n
set up the equation

(17,k) ?(z)-<r - •••

If every a were to lie exterior to S(k, (IT, t )/n), the region K would
&
subtend to each a^, and angle Z(Tï -Y )/n. That is, a constant could

be found such that

0 é ar>
(17,5)
1
Hence, by Th. (1,1)

is in contradiction to eq. (17,3) • Q«£«D •

We shall now apply Th. (17,2) to the determination of^ the zeros of

P(z); that is, the points s for which P(z) = 0. Since J oS ^0

a,
in Th. (17,2), we deduce at once a result which for f » 0 and for Y arbi¬
trary.

Theorem (17,3). Let P(z) be an nth degree polynomial; let Cs/z/ ■

f(t) (t real; aéH b) be a rectifiable curve drawn in a convex region K

and let p<(t) be on G a continuous function whose argument satisfies the

inequality

I
Dr. Lonnie Cross, "Stieltjes Integrals” (Lecture delivered to the
class in Advanced Calculus, Atlanta University, Atlanta, Georgia, March
11, 1962).
 U8

M' - 4r(j J Ci) é XL -+ Y Z JUL + IT.

Then, if

<w.« L rOfÛtïU [flt)]JU = <?,

P (z) has at least one zero in the star-shaped region S(K, (71 - r)n).

Observe if we choose

Y-of4fr) = l} d-0A=!} =
Let us denote by Q(z) an nth degree polynomial which assumes the same values

at the £ and ^ • If we replace n by n-1 in Th« (17>3) and if we set

P(z) H Q»(z), then we find CQ (f) — Q (*1) 3 __ n

Z Q- CU-UÇ+’llM = Ci-*1)
That is, eq. (17,6) is satisfied. Hence, at least one zero of Q'(z) lies

in S(K, 7T /n-1). Since K may be taken as the line segment Joining the

points £ and ^ , we have established that if the nth degree polynomial

P(z) has two zeros z * Ç and z = *£ its derivative will have at least one

zero in the region comprised of all points from which the line-segment ft
subtends an angle of at least .

Polynomial With n-known Zeros.—Let us denote the radius of a circle

containing n zeros of an nth degree polynomial f(z) and by R' the radius

of the concentric circle containing at least p-1 zeros of f'(z). We shall

now obtain another upper on R', using induction.

We want to prove

Theorem (18,1). If an nth degree polynomial f(z) has n zeros in or

on a circle C of radius R and an (n-p) - fold zero at a point ? , then its

derivative has at least p-1 zeros in the concentric C of radius R’ *

R f(3n - 2p)/n J .
 Proof:

Without loss of generality in the proof, we assume that C is the

unit circle fel = 1.

If , then all the zeros of f(z) lie in the circle C and by

Th. (16,2), all n-1 zeros of f*(z) lie in C. In such a case, surely p-1

zeros of f’(z) lie in C.

If , the zeros of f*(z) lie in circle C and in a circle

P with center and radius "C=(n-p)/n. If C and P do not over¬

lap one another, exactly p-1 zeros of f'(z) lie in C and C. If C and P
do overlap, but if p does not enclose 7 , precisely p zeros of f'(z)

lie in the region comprised of C and P and hence in the circle O’ with

center at the origin and radius 1 + £ 2(n-p)/nJ =(3/) —- 3p)Jr\,

Finally, if C and P overlap^ and if P contains ^, then all the zeros

of f(z) lie in C’ and hence all n-1 zeros of f'(z) lie in C, In all cases,

therefore!, circle C contains at least p-1 zeros of f'(z). Q.E.D.

 BIBLIOGRAPHÏ

Apostol, Tan M. Mathematical Analysis, Reading, Massachusetts: Addiscn-

Wesley, 1957.

Churchill, Ruel V. Complex Variables and Applications. New York: Wiley

Sons, i960.

Copson, E. T. Function of Complex Variables. London: Oxford, 1935.

Cross, Lonnie. "Principle of Argument," Lecture delivered to class in Com¬

plex Variables, Atlanta University, Atlanta, Georgia, March 22, 1962.

. "Conformal Mapping," Lecture delivered to class in Complex

Variables, Atlanta University, Atlanta, Georgia, April 17, 1962.

. "Rectifiable Curves," Lecture delivered to class in Com¬

plex Variables, Atlanta University, Atlanta, Georgia, November 16,
1961.

. "Stieltjes Integrals," Lecture delivered to class in Ad¬

vanced Calculus, Atlanta University, Atlanta, Georgia, March 11,
1962.

De Doer, H. E. Extensions of Rolled Theory. Vol. XIX (188U), pp. 207-208.

Franklin, Phillip. Function of Complex Variables. Englewood Cliffs, New

Jersey: Prentice Hall, 1958.

Grace, J. H. "The Zeros of a Polynomial," American Mathematical Society,

XI (1902), 352-357.

Heawood, P. G. "Zeros of Polynomials," American Mathematical Society, XI

(1902), 101-109.

Irwin, F. "Relation Between the Zeros of a Rational Integral Function and

Its Derivative," American Mathematical Society, XVI (1920), I38.

Jensen, J. L. W. "On the Location of the Roots of the Derivative of a

Rational Function," American Mathematical Society, XXII (1920),
128-DiU.

Kellogg, O. D. Potential Theory. Berlin Germany, 1929.

Knopp, Konrad. Theory of Functions. New York: Dover, 19^5•

Marden, Morris. The Geometry of The Zeros of a Polynomial in A Complex

Variable. New York: American Mathematical Society, 19U9.

50
 51

Phillips, E. G* Function of a Complex Variable. New York: Oliver and

Boyd, 1961.

Titcbmarch. E. C. Theory of a Complex Variable» London: Oxford, 1935»

Walsh, J. L. "On the Location of the Roots of Certain Types of Polynomials,M

American Mathematical Society, XXIV (1922), I63-I8O •