L'Hospital's Rule for Complex-Valued Functions

Author(s): D. S. Carter
Source: The American Mathematical Monthly, Vol. 65, No. 4 (Apr., 1958), pp. 264-266
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2310244 .
Accessed: 25/09/2014 08:57

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.

http://www.jstor.org

This content downloaded from 134.184.144.237 on Thu, 25 Sep 2014 08:57:56 AM

All use subject to JSTOR Terms and Conditions
 MATHEMATICAL NOTES
EDITED BY Roy DUBISCH, FresnoState College
thisdepartment
Materialfor shouldbesenttoRoyDubisch,Department
ofMathematics,
FresnoStateCollege,Fresno26, California.
L'HOSPITAL'S RULE FOR COMPLEX-VALUED FUNCTIONS
D. S. CARTER, University
of California,
Los AlamosScientific
Laboratory
L'Hospital's rule forreal functionsmay be stated in the form:
Let f(x) and g(x), and theirderivatives
f'(x), g'(x), be continuouson an open
interval(0, a), and let
limf(x) = lim g(x) = 0.
x-O o-'O

If theratiof'/g' is definedon (0, a) (in thesense thatg' does notvanish) and has a
finitelimit at x =0, thenthe ratio offunctions,f/g, is definedand has the same
limit.*
A simple example shows that this rule is not generallyvalid when f and g
are complex-valued functions of a real variable. Taking f:=x, g = xe-ilx, we have
xei=
x + i

which vanishes as x->O, while f/g = eilx has no limit. The more pathological
example f= x, g = x(e-ilx -1) shows that f/g need not even be defined.
However, by placing additional restrictions on the functions, we can obtain
generalizations of the rule which cover many cases. For example, the rule is
valid provided

I. The ratio Ig'f/fgf' is defined (in the sense that the derivative Igf' of fgf
exists and does not vanish) and bounded on (0, a).

This is the simplest of a class of conditions obtained by introducing the

difference, 5(x), between f'/g' and its limit
L = limf'(x)/g'(x).
x po

Multiplying the equation S(x) =f'(x)/g'(x) -L by g'(x) and integrating, we have

rx
f(x) - Lg(x) = f b(y)g'(y)dy,

in which the integral may be improper. On dividing by g(x), it is clear that f/g
has the limit L provided
* Thisruleis equallyvalidiff'/g' has an infinite
limitat x = 0, sincetheconditions
onf and g
are thesame.This is nottrueoftheconditions derivedbelowforcomplex-valued Here
functions.
therolesoffandg mustbe interchanged explicitlytocovercasesofinfinite limits.
264

This content downloaded from 134.184.144.237 on Thu, 25 Sep 2014 08:57:56 AM

All use subject to JSTOR Terms and Conditions
1958] MATHEMATICAL NOTES 265

II . The ratio

R(x) [ (y)g'(y)dy]/ g(x)

is definedon (0, a) and vanishes as x- )O. We will show that each of the two
followingconditions,as well as I, impliesCondition II and is thereforesufficient
forvalidity:
I
III. The ratio r(x) 6(x)jjg'(x) /Jg(x)j' is definedon (0, a) and vanishes
as x--O, whereIgj ' is thederivativeof gI .
IV. Ig(x) is monotoneand thereal and imaginarypartsof 6(x) are ofbounded
j

variationon (0, a).

The firststep in the proofis to show that g cannot vanish on (0, a). For Con-
ditions I and III this followsimmediatelyfromRolle's theoremand the fact
that I gj is continuousand vanishes at x=0, while Igj' is definedand does not
vanish. For Condition IV it followsfromthe monotonicityof I gj: If g() = 0,
0 < t <a, then g vanishes identicallyon (0, t), which contradictsthe hypothesis
that g'50.
To showthat III impliesII, we notethat j gj' is givenby Ig|'= (gg'+g'g)
/21gj, where the complex conjugate g enjoys the same continuitypropertiesas
g. Hence Igj ' is continuous; and if we definer(O) =0, it followsthat r(x) is con-
tinuous on the closed interval [0, a/2]. Let p be an upper bound forr on that
interval. Then
x
fXj(y)j g'(y)jdy_ pf jg(y)jIdy<pIg(x)jI

and on taking the limit e->O we see that

x
f 5(y)I
I I g'(y)I dy

is defined,continuous, and has the continuous derivative I|8jg'j on (0, a/2).

Hence the ratio

Rl(x) =[ 6(y) I Ig'(y)Idy]/ I g(x)I

satisfiesthe conditionsof l'Hospital's rule forreal functions.Applyingthe rule
to R1, we findthat Condition II is satisfied,since IRI < IR, I.
Condition I clearly implies III, since 6(x) vanishes as x->O. Hence I implies
II.
To show that IV impliesII, we integratethe numeratorof R(x) by parts and
use the monotonicityof IgI to obtain the estimate

This content downloaded from 134.184.144.237 on Thu, 25 Sep 2014 08:57:56 AM

All use subject to JSTOR Terms and Conditions
266 MATHEMATICAL NOTES [April

|R(x) I < I 6(x) |+ db(y)|.

Hence R vanishes with x and II is satisfied.

Similar argumentsshow that Conditions 1, III, and IV are also valid forthe
Stolz extensionof 1'Hospital's rule, in which the conditions
limf(x) = lim g(X) = 0

are replaced by
limIf(x)I =lim I g(x) =3?
x-0 X-0

ON A DETERMINANTAL INEQUALITY
MARVIN MARCUS,* University
of BritishColumbia
Recently L. K. Hua [2] proved the followinginterestinginequality:
Let A and B be n-squarecomplexmatricesand assume
(1) I-A*A and I-B*B
are bothpositivesemidefinite.
Then
(2) | d(I - A*B) 12 > d(I - A*A)d(I - B*B)

and A * is theconjugatetransposeof A.
whered is thedeterminant
We prove herean extensionof the inequality (2). Let X>,aej, j be respectively
the eigenvalues of I-A *B, A *A and B*B so indexed that

l'XiINJ
'!? II?v+jj, b-
Xi+1,I Ijl a,?>aj?1, f?3> #j+1
flj forj
frj1 =1,'*n ,n1 - 1.
THEOREM. If I -A *A and I - B *B are bothpositivesemidefinite
thenfor each
k satisfying1 < k< n,
k k

(3) T | | 1 1 (1 - cq)(l -
An-ij+i
j=1 j=1

We firstestablish an inequality.
LEMMA. If u and v are complex n vectorsand

(4) Ilu+vII <2

then
(5) ~~~~I
1-(U2 V) 12 > (1 - IJUI12)(1 - IIVI12).

* This workwas completedundera National ResearchCouncil-NationalBureauof Stand-

thehelpfulsug-
The authoralso wishesto acknowledge
ardsPostdoctoralResearchAssociateship.
gestionsofDr. Olga Taussky-Todd.

This content downloaded from 134.184.144.237 on Thu, 25 Sep 2014 08:57:56 AM

All use subject to JSTOR Terms and Conditions