Proceedings of the Edinburgh Mathematical Society (1983) 26, 289-295 ©

THE MEAN VALUE THEOREM AND ANALYTIC FUNCTIONS

by ELGIN H. JOHNSTON
(Received 20th January 1982)

It is well known that the mean value theorem (MVT) does not, in general, hold for
analytic functions. The most familiar example to this effect is f(z) = ez since
2ni zo
e -e°j=2nie for any z o eC. On the other hand, it is easy to show that the MVT
holds in C if /(z) is a polynomial of degree at most 2. Thus it is natural to ask what
conditions on a function /(z) analytic in a domain D are necessary and sufficient for
/(z) to satisfy the MVT in D. This is one of the questions answered in this paper.
Many authors have devised "substitutes" for the MVT that do apply to all analytic
functions. For example Samuelsson [5] (see also Robertson [3] and Novinger [2]) has
proved the following local version of the mean value theorem.

Theorem A. Let /(z) be analytic in a domain containing z0. Then there is a

neighbourhood N of z0 such that if Zj is any point in N, then there exists a point z with
an
| Z ~'K Z O + Z I ) | < T | Z O ~ 2 I | & such that

f(zi)-f(z0)=f'(z)(z1-z0).

Notice that the point z does not necessarily lie on the segment [zo^^. McLeod [1]
has proved a version of the MVT that involves a convex combination of derivatives on
[zo>zil-

Theorem B. Let /(z) be analytic in a domain D. If zo,z1eD and the segment

[ZOJZJSD, then there are points wo,w, e(zQ,zx) and there is a X (O^A^l) with

In this paper we look in a direction different from those in Theorems A and B and
instead ask for what analytic functions does the classical MVT hold. We actually ask for
more ... in particular when does a pair of analytic functions satisfy the generalised
MVT? This is made more precise in the following definition.

Definition 1. Let f(z) and g(z) be functions analytic in a domain D £ C, and suppose
that g(z) is one-to-one in D. Then /(z) and g(z) satisfy the generalised mean value
property (GMVP) on D if, whenever the line segment [zi,z 2 ]s/> (zY^z^), there is a
289

https://doi.org/10.1017/S001309150000434X Published online by Cambridge University Press

 290 ELGIN H. JOHNSTON

point oe(z,,z 2 ) such that

f(z2)-f(zi) J'(c)

In the above definition, the classical MVP arises if we take g(z) to be linear. Since the
GMVP is clearly satisfied if /(z) = constant, we eliminate this case from our
considerations.

Theorem 2. Let f{z) (^ constant) be analytic in a domain D and let g(z) be analytic
and one-to-one in D. Then the following are equivalent.
(i) f(z) and g(z) satisfy the GMVP on D.

u and
(1
° f\z) ~ g'(z) md f'{z) - g\z)
cis meromorphic functions in D.
(iii) One of the following statements holds; in (a) and (c) it is assumed that g is univalent
in D:
(a) /(z) and g(z) are nonconstant polynomials of degree at most 2,
(b) f(z) = Ag(z) + Bfor some complex constants A and B (A^O),
(c) f(z) = A cos ccz + B sin az + C and g(z) = D cos az + E sin a.z+ F where A, B, C, D, E, F
are complex constants and (|/l| + |B|)(|£>| + |£|)a=/=0.
Proof. (iii)=>(i). This is easily checked.
(i)=>(ii). Select aeD with /'(a)^0. Since neither the addition of constants to /(z) and g(z)
nor the translation of the variable affects the GMVP, we may assume a = 0=/(0)=g(0).
Since g'(0)^0, g~1(w) is analytic in a neighbourhood of w = 0, with g~1(0) = 0. Thus
f°g~l(w) is analytic in a neighbourhood of 0 and we can write

for |w|<£, where e is a suitable positive number. Letting w=g(z), we find that

f(z) = Alg(z) + A2(g(z))2+--- (1)

for \z\ < d, where 6 is a sufficiently small positive number.

It is clear that the pair of functions /(z) and g(z) satisfy the GMVP on D if and only
if the pair f(z) — kg(z) and g(z) do (k = constant). So taking F(z)=f(z) — Alg(z), we may
assume that F{z) and g{z) satisfy the GMVP on D.
If F{z) is a constant function, then f(z) = A1g(z) + B, and At^0 since f(z) is not
identically constant. In this case it is easy to check that (ii) holds.
Suppose F(z) is not constant. Since F(0) = F'(0) = 0 it follows that F(z) takes the value
0 at z = 0 with multiplicity n ^ 2 . Thus [4,p. 216] there is a disc A(0,n)sA(0,<5) such that
F'(z) ^= 0 in A'(0, rj) and there is a neighbourhood N of 0 with N £ A(0, n) such that F(z) is
an n-to-one mapping on N. In fact each weF(N) — {0} is taken on at n distinct points of

https://doi.org/10.1017/S001309150000434X Published online by Cambridge University Press

 THE MEAN VALUE THEOREM AND ANALYTIC FUNCTIONS 291

N — {0}. If n ^ 3 , then we can find distinct points zuz2,z3eN — {0} with F(z1) = F(z2)
= F(z3). By the GMVP, each of the segments {zuz2), (zuz3), {z2,z3) must contain a zero
of F'(z). However these segments are all contained in A(0, rf) and 0 is the only zero of
F'(z) in A(0,r\). Thus 0e(zl,z2)r\(z2,z3)n(zl,z3). This is impossible for distinct points
z1;z2,z3. It follows that n = 2. Thus F(z) is a two-to-one mapping on N and, if zuz2 are
distinct points of AT —{0}, then

F(zl) = F(z2) implies that 0e(z x ,z 2 ), i.e. z 1 /z 2 <0. (2)

Let / be any line through 0 and let Hx, H2 be the two open half planes determined by
/. Let Ni = HtnN (i=l,2). It follows from (2) that F(z) is univalent on each of NUN2.
Furthermore, for i = 1,2, F(AT;) = F(iV) — y where y = F{lnN) is a simple analytic arc with
one endpoint 0. We can then define analytic functions hi-.F(N) — y-*Ni (i = l,2) with
(hj°F)(z) = z (zeNi). Since O^JV; (i=l,2), neither hy nor h2 takes the values 0. Thus
hY{w)lh2(w) is analytic on F(N) — y and by (2), hl(w)/h2(w)<0 on F(N) — y. Thus
h1(w)/h2(w) = k (k some negative constant). Now let Fbe a line through 0 and assume the
acute angle formed by / and T is less than n/10. Let Ht and H2 be the half planes
determined by T, and labeled so that Hx n H t is a sector of angle measure greater than
97i/10. Analagous to the previous development, define JV; = J?,-nN (i=l,2), y = F(Nnl)
and /j;:F(JV)-y->JV,. (i = 1,2) with (£,.<>F)(z) = z (zeJVf). Then

'IfliV, rv JVf) = M F(JV( n iV,)

(i = l,2), and it follows that fi1(w)/fi2(w) = k on F(N)-y. Now let zoeN1nN2. Then there
is a point z'oeN2nNl with F(z0) = F(z'o). We then have

k = Z° = /

Hence /c= — 1.
From the above argument we may conclude that if zuz2 are distinct points in N with
F(z1) = F(z2), then zl = —z2. Conversely, if Zj = —z2, we must have F(z1) = F{ — z2) = F(z2),
showing that F(z) is an even function. Thus

and

Since A, =f'{0)/g'(0) and a = 0 was chosen without loss of generality, we see that (ii)
holds for all zeD at which f'(z)^0. Since g'(z) is never 0 on D, it follows that
f"\z)lf\z) and/ (5) (z)//'(z) have only removable singularities on D. Thus (ii) holds in D.
(ii)=>(iii). We consider three cases

https://doi.org/10.1017/S001309150000434X Published online by Cambridge University Press

 292 ELGIN H. JOHNSTON

Case a. If f'"(z)/f'(z)=g'"(z)/g'(z) = O on D, then it follows that / and g are

polynomials of degree one or two. Observe that f(5Kz)/f'(x)=g(5)(z)/g'(z) is also satisfied
by such / and g.

Case b. If f'"(z)/f'(z)=g'"(z)/g'(z) = k ± 0 on D, then it follows immediately that f(z) =

A cos a.z + B sin az + C and g{z) = D cos az + E sin az + F where \A\ + \B\ ± 0, \D\ + \E\ =£ 0 and
a2=-k. Note that f{5){z)/f'(z)=g(5){z)/g'{z) is also true for such / and g.

Case c. If f'"(z)/f'(z)=g'"(z)/g'(z) = u(z) where u(z) is not constant on D, then since

g'(z)= 0 we see that u(z) is analytic on D. Thus

and differentiation gives

g<5>(z) = g"'(z)«(z) + 2g"(z)u'(z) +g'(z)u"(z).

Dividing by g'(z) we have

Similarly,

v
/'(z) """ ' f'(z)

Since u'(z) ^ 0 in D it follows that

f'{z) g'(z)

for all zeD. Hence

and so,

Thus /'(z) = i4^(z) and f(z) = Ag{z) + B. Q

As an immediate corollary we can characterise those analytic functions that satisfy the
MVT on D.

https://doi.org/10.1017/S001309150000434X Published online by Cambridge University Press

 THE MEAN VALUE THEOREM AND ANALYTIC FUNCTIONS 293

Corollary 3. Let f(z) be analytic on a domain D. Then f(z) satisfies the MVT on D
(i.e. the GMVP with g(z) = z) if and only if f(z) is a polynomial of degree at most 2.

Remark 4. Since the arguments in the proof of Theorem 2 were all local, it is clear
that the univalence hypothesis on g can be dropped. If we instead require only that g(z)
is not constant on D. then the proof of Theorem 2 shows that (ii) holds at all points
where f'{z)g'(z) £ 0. It follows that the equalities in (ii) actually hold in all of D as
equalities between meromorphic functions. The equation

when g(z1)=g{z2) {zl±z2) can then be interpreted as saying that there is a point
ce(zl,z2) such that the order of the pole of {f(zx)—f{z))/(g{z1)—g(z)) at z2 is the same as
the order of the pole of f'(z)/g'(z) at c. Indeed this is easily checked for those functions
f(z) and g(z) listed in (iii).

Remark 5. Although the condition / ' 5)(z)//'(z) = g(5\z)/g'(z) was used in only one case of
the proof of (ii)=>(iii), the condition cannot be dropped. For example, it is easy to check that if
/(z) = z4 and g(z) = z~1 are analytic on a domain D, then /'"(z)//'(z) =
g'"(z)lg'(z). However f(z) and g{z) will not satisfy the GMVP on D. (Note that
on D).

Remark 6. When discussing the MVT it seems appropriate to mention Rolle's

Theorem. We will say that a non constant function /(z) analytic on D satisfies the Rolle
Property (RP) on D if [zi,z 2 ]£D (zi^zi) a n d f{zi)=f{z2) imply the existence of a
point cs(zuz2) such that /'(c) = 0. It is clear that if f'{zo)j=0 (z0eD) then /(z) satisfies
RP in a neighbourhood of z0. This illustrates a major difference between RP and the
MVP. A function may satisfy RP locally on a domain D without satisfying RP on the
whole domain D, while if functions /(z) and g[z) satisfy the GMVP on any open subset
of D, then they satisfy the GMVP on D. However, since only local arguments were used
in the proof of Theorem 2, the same arguments can be used to prove the following
result.

Theorem 7. If f(z) is analytic in a domain D, f'(zo) = O and f(z) satisfies the RP on D,

then f(z + z0) is an even function in a neighbourhood of 0. Furthermore, if f is not
identically constant, then f"(z0) =fc 0.

A somewhat longer argument can be used to prove the following more informative
result.

Theorem 8. / / /(z) is analytic, nonconstant and satisfies RP on a convex domain D,

then f(z) takes no value with multiplicity greater than two in D.

https://doi.org/10.1017/S001309150000434X Published online by Cambridge University Press

 294 ELGIN H. JOHNSTON

Proof. Suppose /(z) takes the value a with multiplicity at least 3. Then three cases
might arise:

Case 1. There is a point zoeD with f{zo) = a and f'(zo)=f"(zo) = O.

Case 2. There are distinct points zo,zy&D with f(zo)=f(zl) = a and f'(zo) = O.

Case 3. There are distinct points zo,z1,z2eD with f(zo)=f(zl)=f(z2) = a.

In Case 1 we arrive at a contradiction by applying Theorem 7. In Cases 2 and 3 it

can be shown that f'(z) = 0 on D, contradicting the fact that / is not identically
constant.
If Case 2 holds, then by the Open Mapping Theorem and Theorem 7 we may find a
neighbourhood N of z0 such that f(z + z0) is even in N — zo = {z — zo:zeN} and such
that /(z) is a two-to-one mapping on N, with /'(z)^0 on N — {z0}. Depending on
whether / ' ( z J ^ O or /'(zi) = 0 we may find a neighbourhood AT of zx on which /(z) is a
one-to-one mapping or a two-to-one mapping. In either case both f{N) and f(N') are
neighbourhoods of a. Let / be the line determined by z0 and zy. Let Hi and H2 be the
open half planes determined by I and let JVf =//,-n AT, N'i = H,nN' (i=l,2). Since
), f(N\) are both open, since /(A^) contains a neighbourhood of a and since
aef(N\), it follows that /(A^) n f(N\) is a non empty open set with a
Thus we can find sequences { ( ^ J E A ^ and {rjk}zN\ with the following properties:

(i) lim £k = z0 and lim t]k = zl

Since A^ and N2 are open, we may assume, by a subsequence argument that the
segments {[£4,>7J} are pairwise disjoint. By (ii) and RP, there is a point pke(Zk,r]k) with
f'(Pk) — ® (fe = 1,2,...). Since the segments {[^, T/J} are disjoint, the pk's
produced are distinct. By another subsequence argument we may assume \imk^oopk = p
exists where petz^z^zD. But then the Identity Theorem [4,p. 209] implies /'(z) = 0
on D, contradicting the assumption that / is not identically constant. Thus Case 2
cannot occur.
The argument that Case 3 cannot occur is similar and is left to the reader.
I wish to thank the referee for many useful suggestions.

REFERENCES

1. R. M. MCLEOD, Mean value theorems for vector valued functions, Proc. Edinburgh Math.
Soc. (2) 14 (1965), 197-209.
2. W. P. NOVINGER, A local mean value theorem for analytic functions with smooth boundary
values, Glasgow Math. J. 15 (1974), 27-29.

https://doi.org/10.1017/S001309150000434X Published online by Cambridge University Press

 THE MEAN VALUE THEOREM AND ANALYTIC FUNCTIONS 295

3. J. M. ROBERTSON, A local mean value theorem for the complex plane, Proc. Edinburgh Math.
Soc. (2)16(1968/69), 329-331.
4. WALTER RUDIN, Real and Complex Analysis, (McGraw-Hill, New York, 1966).
5. AKE SAMUELSON, A local mean value theorem for analytic functions, American Math. Monthly
80 (1973), 45^46.

IOWA STATE UNIVERSITY

AMES, IOWA 50010

https://doi.org/10.1017/S001309150000434X Published online by Cambridge University Press