Documente Academic
Documente Profesional
Documente Cultură
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
Beginning students often misunderstand the definition of Cauchy sequences when they
first encounter it in an introductory real analysis course. In particular, many students
fail to understand that it involves far more than that the distance between successive
terms is tending to zero. Nevertheless, sequences which satisfy this weaker property
are interesting in their own right. We call them quasi-Cauchy.
Proof. For every k ≥ 1, fix y0k , y1k , . . . , ynkk in I with y0k = bk , ynkk = ak+1 , and
|yik − yi−1
k
| < 1k for 1 ≤ i ≤ n k . Then the sequence
328
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 117
The preceding theorem can be strengthened in the special case that I is bounded:
The notion of quasi-Cauchy makes perfect sense in any metric space (X, d) (af-
ter the purely typographical substitution of d(x, y) for |x − y|). The main difference
is that, for some important metric spaces, the distinction between quasi-Cauchy and
Cauchy vanishes. Let us coin a term for this phenomenon:
The intuition is that in a nonincremental space you can’t travel far via the arbitrarily
small increments between successive terms in a quasi-Cauchy sequence. Nonincre-
mental spaces exist: take the set of integers along with the Euclidean metric. Are there
any interesting examples? Yes: in [2] they are shown to arise naturally in certain classes
of topological algebras. Furthermore, they arise in the theory of ultrametric spaces:
Ultrametric spaces are also called non-Archimedean spaces or isosceles spaces. The
latter is because they have the interesting property that in them all triangles are isosce-
les. They arise naturally in the study of p-adic numbers and in other areas of analysis
(e.g., [1]). More recently, they have found applications in theoretical computer science
(e.g., [5]) and other areas. The following theorem is part of the folklore:
Proof. Suppose that (X, d) is an ultrametric space. Suppose that xn is a quasi-
Cauchy sequence in X . Suppose > 0 is given. Fix K > 0 such that n ≥ K implies
d(xn , xn+1 ) < . Suppose that m, n ≥ K with m ≤ n. Then repeated application of
the ultrametric inequality yields
The idea behind the proof is that for ultrametric spaces the relation between points
of being within of each other is a transitive relation. Thus for any > 0 the collection
of -balls forms a partition of the space.
A natural question is if the converse of this last theorem is true. The answer is no.
For a simple example, let X be a three-element set consisting of the vertices of any
nonisosceles triangle, and let d be the distance between the vertices. Then (X, d) is
not an ultrametric space, but it is nonincremental since any quasi-Cauchy sequence is
eventually constant and hence Cauchy as well. There is an interesting partial converse,
involving notions of metric equivalence.
In the special case that X 1 = X 2 = X (so that d1 and d2 are pseudometrics on the
same space) we will simply say that d1 and d2 are topologically or uniformly equivalent
if the identity mapping is either a homeomorphism or a uniform equivalence. Note
that all 3-element metric spaces (including equilateral ones) are uniformly equivalent
to each other. Thus the property of being an ultrametric space is not preserved by
uniform equivalence. On the other hand, it is easy to see that the property of being
nonincremental is preserved by uniform equivalence.
There is a partial converse to the previous theorem:
330
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 117
Note that d ∗ (x, y) < implies that x and y are -connected (with respect to d). It
is easy to check that d ∗ (x, y) ≥ 0, d ∗ (x, y) = d ∗ (y, x), and d ∗ (x, y) ≤ sup{d ∗ (x, z),
d ∗ (z, y)} for all x, y, z ∈ X . This last inequality follows from the observation that
the relation of being -connected is transitive. Thus d ∗ is an ultra-pseudometric (a
pseudometric which satisfies the ultrametric inequality). We claim that d ∗ and d are
equivalent. Note that establishing this claim automatically yields that d ∗ is in fact a
metric since any pseudometric equivalent to a metric is itself a metric.
Let Bd (x, ) = {y | d(x, y) < }, the -ball with respect to d centered at x. Define
Bd ∗ (x, ) in the analogous way. Note that d ∗ (x, y) ≤ d(x, y) for all x, y. This implies
that for any x ∈ X and any > 0,
Thus the d topology is finer than the d ∗ topology. On the other hand, suppose x ∈ X
and > 0. We need to show that there exists a δ > 0 with Bd ∗ (x, δ) ⊆ Bd (x, ).
Suppose no such δ exists. Then for every integer n ≥ 1 we can find a point an ∈
Bd ∗ (x, 1/n) with an ∈ Bd (x, ). Since d ∗ (x, an ) < 1/n we can fix a (1/n)-chain
x0 , x1 , . . . xk connecting x to an . Thus x0 = x, xk = an , and d(xi , xi+1 ) < 1/n for
i < k. Call
x0 , x1 , . . . , xk−1 , xk , xk−1 , . . . , x1 , x0
the (1/n)-circuit. It leads from x to a point outside of Bd (x, ) and then back again
with steps of length < 1/n. Now define a sequence yi by starting with the 1-circuit,
following it by the (1/2)-circuit, then the (1/3)-circuit, etc. Since the circuits begin
and end at x, d(yi , yi+1 ) = 0 if yi and yi+1 straddle the boundary between successive
circuits. On the other hand, if yi and yi+1 are both inside a circuit (say the (1/n)-
circuit) then d(yi , yi+1 ) < 1/n. Note that n → ∞ as i → ∞. Thus yi is a quasi-
Cauchy sequence in (X, d). It is not Cauchy since for any K > 0 you can find i, j ≥
K with d(yi , y j ) ≥ : pick i ≥ K so that yi = x is the start of a circuit and pick
j > i so that y j is the midpoint of that circuit. This contradicts the fact that (X, d) is
nonincremental, so the required δ must exist after all.
At this stage we know that if (X, d) is a metric space then d uniformly equivalent to
an ultrametric implies that d is nonincremental which in turn implies that d is equiva-
lent to an ultrametric. It is natural to ask if either of these implications can be reversed.
We will give two examples to show that they can’t.
For the first example, note that any discrete metric space is equivalent to an ultramet-
ric space: if (X, d) is discrete (for any point x there is an > 0 with Bd (x, ) = {x})
then d ∗ given by x = y implies d ∗ (x, y) = 1 is an equivalent ultrametric on X . If we
let X be the set of partial sums of the harmonic series with d the induced Euclidean
metric on X , then X is clearly discrete but fails to be nonincremental. Thus being
equivalent to an ultrametric does not imply being nonincremental.
The second example is more involved. Suppose that (X, d) is uniformly equiva-
lent to an ultrametric space. Then there exists an ultrametric d ∗ on X with d and d ∗
uniformly equivalent. Suppose that > 0 is given. Then there exists γ > 0 such that
Now let
6
1 t t t
t
t q q q
t
t t t -
1 2 3 q q q
Figure 1.
Consider X as a subset of the Cartesian plane and let d be the induced Euclidean
metric. X consists of infinitely many columns where each column is distance 1 from
its neighboring columns. Thus any quasi-Cauchy sequence is eventually confined
to a single column. But each column consists of only finitely many points. So any
quasi-Cauchy sequence in eventually constant. Thus every quasi-Cauchy sequence is
Cauchy, so (X, d) is nonincremental. On the other hand, (X, d) fails to satisfy (∗) with
= 1/2: For any δ > 0 you can pick n with 1/n < δ. Then the points p1 = (n, 0)
and p2 = (n, 1) are clearly δ-connected but d( p1 , p2 ) = 1 > . Thus, (X, d) is not
uniformly equivalent to an ultrametric space.
To put these results in context: papers have appeared charactering ultrametric spaces
up to topological equivalence [3] and up to uniform equivalence [4]. Nonincremental
spaces are strictly in between. Whether this is a useful concept or a curiosity remains
to be seen.
332
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 117
ACKNOWLEDGMENTS. We would like to thank the referees for their many helpful suggestions and com-
ments.
REFERENCES
DAVID BURTON graduated from the U.S. Naval Academy in 1981. After serving in the submarine fleet for
several years, he returned to school and earned a Ph.D. in mathematics from Vanderbilt in 1997, specializing
in functional analysis. He is currently chair of the Mathematics and Computer Science Department at the
Franciscan University of Steubenville.
Department of Mathematics and Computer Science, Franciscan University of Steubenville,
Steubenville OH 43952
dburton@franciscan.edu
JOHN COLEMAN received his B.S. from Kent State University in 1985 and his Ph.D. from the University
of Colorado at Boulder in 1992, specializing in universal algebra. His main research interest is in the interplay
between algebra and topology in topological algebras. He currently teaches mathematics and occasionally
computer science at the Franciscan University of Steubenville.
Department of Mathematics and Computer Science, Franciscan University of Steubenville,
Steubenville OH 43952
jcoleman@franciscan.edu