The practical understanding of real magnitudes and their uses in approxi
mations leads eventually to the idea of characterizing the field R of real numbers by a suitable list of axioms. It is not enough to require just that R be an ordered field, for there are many such fields, including the field Q of rational numbers and the formal power series field R«/» . The crucial feature is a completeness axiom, stated in §I.4: Every non-empty set of reals with an upper bound has a least upper bound. An ordered field with
5. Axioms for the Reals 1 03
this property is called a complete ordered field. The real numbers form such a field. The formal power series do not.
This completeness axiom implies the Archimedean law. For suppose to the contrary that there were positive reals
a >
0 andb >
0 such that no multiple na exceededb.
Then the set S of all multiples na, forn
a natural number, has an upper bound, namelyb.
By the completeness axiom, S then has some least upper bound, call itb *;
thusb * �
na for alln.
This also means thatb * � (n + l )a
for alln,
and so thatb * -a �
na for alln.
Thus
b * - a
is less thanb *
and is also an upper bound for S, a contradiction to the choice of
b *
as a least upper bound for S.The force of the completeness axiom is to insure that all the real numbers that ought to be there are there. For example, the irrational V2 must be there, as the least upper bound of the set 1 , 1 .4, 1 .4 1 , 1 .4 1 4, . . . of rationals (those approximating V2). Similarly ." must be there, as the least upper bound of the set of decimals 3, 3 . 1 , 3. 1 4, 3 . 1 4 1 , 3 . 1 4 1 5, 3 . 1 4 1 59, . . . . Indeed, since there is a rational between any two reals, any real can be expressed as the least upper bound of a set of rationals. The completeness axiom is also used in more sophisticated ways, for example in the proof of Rolle's theorem and of the mean value theorem of the calculus (Chapter V).
There are other equivalent forms of the completeness axiom. Instead of requiring that every bounded set of reals have a least upper bound, one may require any one of the following:
Dedekind Cut Axiom. If the set R of reals is the union of two disjoint non-empty subsets L and U such that
x
E L and y E U implyx <
y, then there is a real number r such thatx <
r for allx
E L and r<
y for all y E U.Cauchy Condition. Every Cauchy sequence of real numbers has a limit.
Weierstrass Condition. If a series
C( + C2 + . . . + Cn + .
. . of positive real numbers
Cn >
0 is such that there is an upper boundb
for all partial sumsSn
=C( + . . . + Cn ,
then the series converges.It is illuminating to establish the equivalence of the different complete
ness axioms. The Dedekind cut axiom is perhaps the more geometric (and has already appeared (§III.2) in the completeness axioms for the geometry of the plane): Any cut of the real line into a lower part L and an upper part U must be a cut
at
some real number r. Indeed, each real number determines j ust two such cuts, one with L consisting of allx <
r, the other with L consisting of allx <
r. In this way, real numbers can be completely described by cuts.The completeness axioms also serves to determine the real numbers uniquely, up to an isomorphism of ordered fields. We sketch the proof.
1 04
IV. Real Num bers Suppose that R ' is any complete ordered field, with unit elementI '
for multiplication. Then0 < I ',
for otherwiseI ' < 0,
which would giveo <
- I ' and soI '
=( - I '
)(- I ' ) > 0,
a contradiction. For each natural numbern > 0
the multiplen I '
must then be positive, and all these elementsn I '
are different elements in our field R '. Thereforecp ( n ) = n
I ' defines an injective mapcp :
N ... R'
. Because R ' is a field, thiscp
can be extended to a mapcp : Q
... R'
by settingcp ( n/m ) = cp ( n )jcp ( m )
wheneverm =1= 0;
in other words, R ' contains a copycp ( Q )
of the ordinary rational numbersQ.
Also,cp
preserves the order. By the Archimedean law, each elementr '
in R ' must then be a least upper bound of a set L ' of these rationals, indeed it is the least upper bound of the imagecp
(L) of the setL =
{n lm 1 m =1= 0, cp (n lm) < r ' } .
This set L has the special property that x
<
y and y E L imply x E L, so it is the lower half of a "cut" in the rationals. Since it is bounded, it has an (ordinary) real numberr
as least upper bound, andr
in turn determines L. The one-to-one correspondencer 1-+ r '
then maps the ordinary real numbers on the given complete ordered field R '. One can show that it preserves sums, products and order, hence it is the desired order isomorphism R == R '.
Note that this argument makes essential use of sets, such as the sets L of rationals. In this it is like the use of sets in the induction axioms to prove that the Peano postulates uniquely determine the natural numbers.
We now have two different axiomatic descriptions of the reals: Here, the real
field,
described as a complete ordered field, in §I.4 the realcontin
uum,
described as an unbounded ordered set with a denumerable dense subset. Each description determines the set of reals uniquely, up to an isomorphism of the structure concerned. However the structures are drasti
cally different. The real continuum has only the order structure, and there are many automorphisms of this structure. The real field has both order and algebraic structure, and its only automorphism (by a proof like that just above) is the identity automorphism. These differing structures on the same "thing" (here the reals) are much like the differing structures on the plane (without and with orientations, as in §III.8). These differing struc
tures furthermore reflect practical differences. Thus the ordered contin
uum handles comparisons of many items where one knows only which of two items is the larger, with no measure of "how much" larger.
In either structure, the real numbers form a "continuous" scale. Physi
cists (and others) sometimes suggest that an "atomic" or "discrete" scale would be more "real"; finitists propose finite scales of magnitude. These proposals turn out to be hard to execute-the continuous scale of reals works smoothly!