Aslaksen-Multiple-Valued Complex Functions and Computer Algebre
Aslaksen-Multiple-Valued Complex Functions and Computer Algebre
1 Introduction
I recently taught a course on complex analysis. That forced me to think more
carefully about branches. Being interested in computer algebra, it was only
natural that I wanted to see how such programs dealt with these problems.
I was also inspired by a paper by Stoutemyer ([3]).
While programs like Derive, Maple, Mathematica and Reduce are very
powerful, they also have their fair share of problems. In particular, branches
are somewhat of an Achilles' heel for them. As is well-known, the complex
logarithm function is properly defined as a multiple-valued function. And
since the general power and exponential functions are defined in terms of
the logarithm function, they are also multiple valued. But for actual com-
putations, we need to make them single valued, which we do by choosing a
branch. In Section 2, we will consider some transformation rules for branches
of multiple-valued complex functions in painstaking detail.
The purpose of this short article is not to do a comprehensive comparative
study of different computer algebra system. (For an attempt at that, see [4].)
My goal is simply to make the readers aware of some of the problems, and
1991 Mathematics Subject Classification. Primary 30-01, 68Q40.
13
to encourage the readers to sit down and experiment with their favourite
programs.
I would like to t h a n k Willi-Hans Steeb and Michael Wester for helpful
comments.
If we write
l+2nlri) T M -~- e T M = e~ (2)
and
(el+2n=i) l+2n~i _-- e1+4~,~i-4,2~ 2 _ ee-4,2~ 2, (3)
it follows t h a t
e -4"~2 = 1. (4)
There are also a number of paradoxes involving square roots. Let me just
give two.
~-1 1 (6)
1 i
i2_-1.
14
In order to clarify such problems, we will take a fairly detailed look at some
properties of elementary transcendental functions.
For z = x +. iy, the complex exponential function is defined by
It satisfies the property e (z+~) = e~e~, but does it satisfy the property
(e~) ~ = e(~)? In order to answer this, we must look at the complex log-
arithm function.
We define the principal argument by z ]zie iArg(z) and Arg(z) E (-77, 77].
=
We do not define the principal argument of 0, and we will from now on assume
that z is different from 0. Notice that we have defined the principal argument
on the negative axis, too, but it is of course not continuous there. Extending
the definition of the principal argument to the negative numbers gives us as
a ready supply of counter examples.
We then define the principal logarithm Log(z) by Log(z) = log Izl +
i Arg(z), where log [z[ denotes the usual real logarithm of Iz[. We clearly
have e L°g(z) = z, but do we have Log(e ~) = z?
In order to study this, we will introduce the following terminology.
T h e o r e m 2 We have
In particular,
Proof: We have
We will next study whether the complex logarithm satisfies the property
Log(zw) = Log(z) + Log(w). To this end, we must first study Arg(uv). It is
easy to see that
ppe(z, w) -- ( A r g ( z w ) - A r g ( z ) - Arg(w))/(21r).
We can now define the right (left) half-plane as the set of points where csgn(z)
is positive (negative).
Hence
Log(l/z) = ~-_ Log(z), if z is not negative
( Log(z) + 27ri, if z is negative.
We are now ready to consider whether (eZ)w equals e ~w. The key issue is
that (e~) ~° involves the exponential function with base e" and not just e. So
while e ~° is a genuine single-valued function, we need to choose a branch in
order to make (eZ)w single valued.
T h e o r e m 9 We have
( e z ) w ~ eZW e-W27riImq(z).
17
Proof:
(e:) w = eLog(eZ) w = e(Re(z)+ilmr(z)) w = e(Z-i2~rlmq(z)) w = eZW e-Wi2~rlmq(z) []
el--4n21r2e--(1+2nTri)21rin = el--4n21r2e--2n~ri+4n2~r2 = e~
C o r o l l a r y 10 We have
(eZ)l/2 = (--1)Imq(z)eZ/2"
In particular,
(eZ)l/2 = eZ/2 if and only i f I m ( z ) • ( ( 4 n - 1)~r, (4n + 1)~-], n • Z.
( z w ) a : zawaea21rippe(z,w).
Proof:
T h e o r e m 13 We have
T h e o r e m 14 We have
if z is not negative
L-livE, if z is negative.
~/U;=sgn(z)/~,
Proof: If z is not negative, we have
1~ = e L°gcl/z)/2 = e -L°g(z)/2 = 1 / v / ~ ,
The last two results resolve the two square root paradoxes given at the
beginning of this section.
3 Computer tests
Computer algebra systems are in general much better at reducing the differ-
ence between two equivalent expressions to 0, than simplifying an expression
to a specific form. I therefore suggest that the readers experiments with
the following eight tests (adapted from [3]) using their favourite computer
algebra system.
19
Test 1
Test 2
Test 3
Test 4
Test 5
(a) Log(zw) - Log(z) - Log(w) should not simplify when z and w are
complex.
20
(b) log(zw) - log(z) - log(w) should simplify to 0 when z and w are both
positive.
Test 6
Test 7
Test 8
References
[1] Thomas Clausen, Aufgabe 53, J. Reine Angew. Math. 2 (1827), 286-287.
[2] Reinhold Remmert, Theory of Complex Functions, Graduate Texts in
Math., vol. 122, Springer-Verlag, 1991.
[3] David R. Stoutemyer, Crimes and misdemeanors in the computer alge-
bra trade, Notices Amer. Math. Soc. 38 (1991), 778-785.
[4] Michael Wester, A review of CAS mathematical capabilities, Computer
Algebra Nederland 13 (1994) 41-48.