pozn/tm/hamel.

tex November 6, 2015

Hamel basis and additive functions

1

Hamel basis
References: [Hei, Section 4.1], [Ku, Section 4.2, Chapter 11], [NS, Kapitola 4.7],
[A, Section 6F] 2

Existence of Hamel basis

Definition 1. Let V be a vector space over a field K. We say that B is a
Hamel basis in V if B is linearly independent and every vector v ∈ V can be
obtained as a linear combination of vectors from B.
This is equivalent to the condition that every x ∈ V can be written in
precisely one way as X
ci xi
i∈F

where F si finite, ci ∈ K and xi ∈ B for each i ∈ F .

It is also easy to see that for any vector space W and any map g : B → W
there exists exactly one linear map f : V → W such that f |B = g.
Theorem 1. Let V be a vector space over K. Let A be an linearly independent
subset of V . Then there exist a Hamel basis B of V such that A ⊆ B. (Any
linearly independent set is contained in a basis.)

Proof. Zorn’s lemma.

Corollary 1. Every vector space has a Hamel basis.
Proof. For V = {0} we have a basis B = ∅.
If V 6= {0}, we can take any non-zero element x ∈ V and use Theorem 1 for
A = {x}.
In some cases we are able to write down a basis explicitly, for example in
finitely-dimensional space or in the following example. However, the claim that
a Hamel basis exists for each vector space over any field already implies AC (see
[HR, Form 1A]).

Example 1. Let c00 be the space of all real sequences which have only finitely
many non-zero terms. Then {e(i) ; i ∈ N}, where the sequence e(i) is given by
(i)
en = δin , is a Hamel basis of this space.
1 Available at http://thales.doa.fmph.uniba.sk/sleziak/texty/rozne/pozn/tm/
2 See also: thales.doa.fmph.uniba.sk/sleziak/texty/rozne/AC/cont.pdf

1
Cardinality of Hamel basis
Proposition 1. If B1 , B2 are Hamel bases of a vector space V , then card B1 =
card B2 .
Because of the above result, it makes sense to define Hamel dimension of a
vector space V as the cardinality of any of its bases.

Hamel bases in linear normed spaces and Banach spaces

3

Cardinality. Recall that a subset A of a topological space X is called meagre

in X if it is a countable union of nowhere-dense sets. Baire category theorem:
If X is a complete metric space, then X is not meagre in X; i.e., X cannot be
obtained as a countable union of nowhere-dense sets. (Similar claim is true for
locally compact Hausdorff spaces.)
Theorem 2. Let X be an infinite-dimensional Banach space.
a) If S is a subspace of X which has countable Hamel basis, then X is meagre
in X.
b) Any Hamel basis of X is uncountable.
The proof uses Baire category theorem and the fact that every finitely-
dimensional subspace of a Banach space is closed (see [FHH+ , Proposition 1.36]).
The same argument can be used to show analogous result for completely metriz-
able topological vector spaces (see [AB, Corollary 5.23]).
The above result can be in fact improved: It can be shown that cardinality
of infinite-dimensional Banach space is at least c. We will give here a proof from
[L].
We first recall a few fact about almost disjoint families (see [BŠ, §III.1], [B,
Theorem 5.35], [JW, Theorems 17.17, 17.18]).
Definition 2. Let A = {Ai ; i ∈ I} be a system of subsets of X. We say that
A is an almost disjoint family or AD family on X, if card Ai = card X for each
i ∈ I and the intersection Ai ∩ Aj is finite for each i, j ∈ I, i 6= j.
Lemma 1. If X is an infinite countable set then there is an AD family on X
of cardinality c.
Proof. We will work with X = Q. (The obtained AD family can be transferred
to any infinite countable set.)
For every r ∈ R there is an injective sequence fr : N → Q of rational numbers,
which converges to r. Put Ar = fr [N]. It is easy to see that {Ar ; r ∈ R} is an
AD family.
3 I should mention that I’ve learned about some of these results (and their proofs)

from discussions at http://math.stackexchange.com. See http://math.stackexchange.com/

questions/74101/, http://math.stackexchange.com/questions/33282/ and http://math.
stackexchange.com/questions/79184/.

2
 There are several other nice proofs of the existence of an AD family with the
above properties. For instance, we could use nodes of an infinite binary tree.
See, for example, [G].4
Theorem 3. If X is an infinite-dimensional Banach space then Hamel dimen-
sion of X is at least c.

Proof. We first construct inductively systems {xi ; i ∈ N} ⊆ X and {x∗i ; i ∈

N} ⊆ X ∗ such that x∗i (xj ) = δij and kxi k = 1.
Let us describe the inductive step in detail. Suppose we have already
constructed x1 , . . . , xk and x∗1 , . . . , x∗k fulfilling the above conditions. Choose
xk+1 ∈/ [x1 , . . . , xk ] such that kxk+1 = 1k. (This is possible, since X is infinite-
dimensional.) The map x∗k+1 : [x1 , . . . , xk+1 ] → R given by x∗k+1 (xi ) = δij is
linear map on a finitely-dimensional subspace, hence it is continuous. By Hahn-
Banach theorem it can be extended to a linear continuous function from X to
R.
The above conditions imply xk ∈ / (x∗k )−1 (0)
/ [{xj ; j ∈ N, j 6= k}], since xk ∈
and the later set is a closed subspace of X containing {xj ; j ∈ N, j 6= k}.
Now let A = {Ai ; i ∈ R} be an AD family on N. For each i ∈ R we define
X 1
ai = xj .
2j
j∈Ai

(Note that k 21j xj k ≤ 21j , which implies that the above series is Cauchy and thus
convergent.)
We will show that {ai ; i ∈ R} is an independent set. By Theorem 1 this
implies that Hamel dimension P of X is at least c.
Let us assume that i∈F ci ai = 0 for some finite set F , where all ci ’s are
non-zero. Let [
P := (Ai ∩ Aj ).
i,j∈F
i6=j

This set is finite, since A is an AD family. The above finite sum can be rewritten
as
X∞
dj xj = 0,
j=1
ci
where dj = whenever i ∈ F and j ∈ Ai \ P . Since each set Ai \ P is infinite,
2j
we have infinitely many non-zero coefficients in this sum. Thus we can rewrite
the last equation as X
xk = fi xi
i6=k

for some k and fi ∈ R, which contradicts the assumption that xk ∈

/ {xj ; j 6= k}.

4 Or http://math.stackexchange.com/q/162387, http://math.stackexchange.com/q/
278837.

3
Existence of unbounded linear functionals.
Proposition 2. If X is an infinite-dimensional linear normed space, then there
exist non-continuous linear function f : X → R.
Proof. Choose an infinite independent set {xn ; n ∈ N} such that kxn k = 1 for
each n ∈ N and a function f : X → R such that f (xn ) = n.

Continuity of coordinate functionals. If B is a Hamel basis of a vector

space X Pover R, and we define fb : x → R which assigns to x its b-th coordinate,
i.e., x = b∈B fb (x)b for each x ∈ X, then fb is a linear function from X to R.
Suppose that X is, moreover, a Banach space. We would like to know
whether the functions fb are continuous. We will show that at most finitely
many of them can be continuous.

Proposition 3. Let B be a Hamel basis of a Banach space X. Let fb , b ∈ B,

be the coordinate functionals. Then there is only finitely many b’s such that fb
is continuous.
Proof. Suppose that {bi ; i ∈ N} is an infinite subset of B such that each fbi is
continuous. W.l.o.g. we may assume that kbi k = 1.
Let
∞
X 1
x := bi .
i=1
2i

(Since X is complete, theP above sum converges.)

n
We also denote xn := i=1 21i bi . Since xn converges to x, we have fbk (x) =
lim fb (xn ) = 21k for each k ∈ N. Thus the point x has infinitely many non-zero
n→∞ k
coordinates, which contradicts the definition of Hamel basis.
We can give another proof based on Banach-Steinhaus theorem (uniform
boundedness principle). We show first the following:
Lemma 2. Let B be a Hamel basis of a Banach space X. Let fb , b ∈ B, be the
coordinate functionals. Let C = {b ∈ B; fb is continuous}. Then sup{kfb k; b ∈
C} < ∞.
Proof. For any x ∈ X there is at most finitely many b’s in C such that fb (x) 6= 0.
This implies that supb∈C |fb (x)| is finite. Banach-Steinhaus theorem this implies
sup{kfb k; b ∈ C} < ∞.

Proof of Proposition 3. Let B be any Hamel basis for X. For any choice of
constants cb , b ∈ B, is the set {cb fb ; b ∈ B} a Hamel basis as well. The
coordinate functionals for this new basis are gb = c1b fb . If the set C = {b ∈ B; fb
is continuous} is infinite, then by an appropriate choice of constant cb we can
obtain sup{kfb k; b ∈ C} = ∞, which contradicts the above lemma.

4
 It is easy to show that finitely many of coordinate functionals can be con-
tinuous. If X is a Banach space with a basis B and x1 , . . . , xn ∈ / X, then
[x1 , . . . , xn ] ⊕ X is a Banach space with a basis {x1 , . . . , xn } ∪ B and there are
at least n continuous coordinate functionals.
Also in the space c00 from Example 1 with sup-norm all coordinate func-
tionals are continuous. The space c00 is, of course, not complete.

Cauchy functional equation

References: [Ku, Section 5.2, Chapter 12], [S, Section 2.1], [Ka, Chapter 1], [Kh,
Chapter 7], [Her, Section 5.1], [A, Appendix to Chapter 6]
Let us study the functions f : R → R fulfilling

f (x + y) = f (x) + f (y). (1)

The equation (1) is called Cauchy equation and functions fulfilling (1) are called
additive functions.
It is easy to show that
Lemma 3. If a function f : R → R fulfills (1), then

f (qx) = qf (x)

holds for every q ∈ Q, x ∈ R.

This shows, that the additive functions are precisely the linear maps if we
consider R as a vector space over Q.
Lemma 3 implies that
Theorem 4. Every continuous solution (1) is of the form f (x) = ax for some
a ∈ R.

Non-linear solutions
Using the existence of Hamel basis in R (as a vector space over Q) we can show
that
Theorem 5. There exist non-linear solution of (1), i.e. functions f : R → R
that fulfill (1) but are not of the form f (x) = ax.
Theorem 6. If f is a non-linear solution of (1), then the graph of this function

G(f ) = {(x, f (x)); x ∈ R}

is dense in R2 .
The proof can be found e.g. in [Her, Theorem 5.4].
Theorems 4 and 6 suggest that well-behaved solutions of (1) are linear and
that non-linear solutions have to be, in some sense, pathological. Let us mention
a one more result in this direction.

5
Theorem 7. Every measurable solution of (1) is linear.
An elegant proof is given in [Her, Theorem 5.5].
This last result means that by showing the existence of non-continuous so-
lutions of (1) we have also obtained the existence of non-measurable sets.

References
[A] Benno Artmann. The Concept of Number: From Quaternions to Mon-
ads and Topological Fields. John Wiley, New York, 1988.
[AB] Charalambos D. Aliprantis and Kim C. Border. Infinite Dimensional
Analysis, A Hitchhiker’s Guide. Springer, Berlin, 3rd edition, 2006.
[B] Lev Bukovský. The Structure of Real Line. Springer, Basel, 2011.
[BŠ] Bohuslav Balcar and Petr Štěpánek. Teorie množin. Academia, Praha,
2001.
[FHH+ ] Marián Fabian, Petr Habala, Petr Hájek, Vicente Montesinos, and Va-
clav Zizler. Banach Space Theory. The Basis for Linear and Nonlinear
Analysis. Springer, New York, 2011. CMS Books in Mathematics.
[G] Stefan Geschke. Almost disjoint and independent families. Poznmky k
prednke, http://thales.doa.fmph.uniba.sk/sleziak/texty/.
[Hei] Christopher Heil. A Basis Theory Primer. Springer, New York, 2011.
[Her] Horst Herrlich. The Axiom of Choice. Springer-Verlag, Berlin, 2006.
Lecture Notes in Mathematics 1876.
[HR] Paul Howard and Jean E. Rubin. Consequences of the axiom of choice.
Mathematical Surveys and Monographs. 59. Providence, RI: American
Mathematical Society (AMS), 1998.
[JW] Winfried Just and Martin Weese. Discovering modern set theory II:
Set-theoretic tools for every mathematician. Amer. Math. Soc., Provi-
dence, RI, 1997. Graduate Studies in Mathematics 18.
[Ka] Pl. Kannappan. Functional Equations and Inequalities with Applica-
tions. Springer, New York, 2009. Springer Monographs in Mathematics.
[Kh] A. B. Kharazishvili. Strange functions in real analysis. Marcel Dekker,
New York, 2000.
[Ku] Marek Kuczma. An introduction to the theory of functional equa-
tions and inequalities. Cauchy’s euqation and Jensen’s inequality.
Birkhäuser, Basel, 2nd edition, 2009.
[L] H. Elton Lacey. The Hamel dimension of any infinite dimensional sep-
arable Banach space is c. Amer. Math. Monthly, 80(3):298, 1973.

6
[NS] A. Naylor and G. Sell. Teória lineárnych operátorov v technických a
prı́rodných vedách (Linear Operator Theory in Engineering and Sci-
ence). Alfa, Bratislava.
[S] Christopher G. Small. Functional Equations and How to Solve Them.
Springer, New York, 2007. Problem Books in Mathematics.