STEINHAUS’ LATTICE-POINT PROBLEM

FOR BANACH SPACES

TOMASZ KANIA AND TOMASZ KOCHANEK

Abstract. Steinhaus proved that given a positive integer n, one may find a circle sur-
arXiv:1302.6443v5 [math.FA] 15 Sep 2016

rounding exactly n points of the integer lattice. This statement has been recently extended
to Hilbert spaces by Zwoleński, who replaced the integer lattice by any infinite set that
intersects every ball in at most finitely many points. We investigate Banach spaces satis-
fying this property, which we call (S), and characterise them by means of a new geometric
property of the unit sphere which allows us to show, e.g., that all strictly convex norms
have (S), nonetheless, there are plenty of non-strictly convex norms satisfying (S). We also
study the corresponding renorming problem.

1. Introduction and statement of the main results

The following feature of the integer lattice in the Euclidean plane was probably first ob-
served by Steinhaus [11, Problem 24 on p. 17]: for any natural number n one may find
a circle surrounding exactly n lattice points. Zwoleński [12] generalised this fact to the
setting of Hilbert spaces in the following manner. He replaced the set of lattice points by a
more general quasi-finite set, i.e., an infinite subset A of a metric space X such that each
ball in X contains only finitely many elements of A. His result then reads as follows.
Theorem ([12]). Let A be a quasi-finite subset of a Hilbert space X. Then there exists a
dense subset Y ⊂ X such that for every y ∈ Y and n ∈ N there exists a ball B centred at
y with |A ∩ B| = n.
Let us then distill the property that we will term Steinhaus’ property (S). A metric space
X has this property if, by definition,
(S) for any quasi-finite set A ⊂ X there exists a dense set Y ⊂ X such that for all
y ∈ Y and n ∈ N there exists a ball B centred at y with |A ∩ B| = n.
We translate condition (S), formulated above, into three equivalent statements concerning
the geometry of the unit ball of a Banach space. Roughly speaking, they require that,
locally, the unit sphere of X does not look the same at any two distinct points. This
approach will be particularly beneficial, as it will allow us to identify spaces that share

2010 Mathematics Subject Classification. Primary 46B04, 46B20.

Key words and phrases. Steinhaus’ problem, lattice points, strictly convex space.
The first-named author acknowledges with thanks funding received from the European Research Council
/ ERC Grant Agreement No. 291497. The research of the second-named author was supported by the
Polish Ministry of Science and Higher Education in the years 2013–14, under Project No. IP2012011072.
1
2 T. KANIA AND T. KOCHANEK

that property with Hilbert spaces, yet of a very different nature. Our first main result then
reads as follows.
Theorem A. Let X be a Banach space. The following assertions are equivalent:
(S) X has Steinhaus’ property;
(S1 ) for any quasi-finite set A ⊂ X there exists a dense set Y ⊂ X such that for every
y ∈ Y there exists a ball B centred at y with |A ∩ B| = 1;
(S’) for all x, y ∈ X with x 6= y, kxk = kyk = 1 and each δ > 0 there exists a z ∈ X
with kzk < δ such that one of the vectors x + z and y + z has norm greater than
1, whereas the other has norm smaller than 1;
(S”) for all x, y ∈ X with x 6= y, kxk = kyk = 1 and each δ > 0 there exists a z ∈ X
with kzk < δ such that kx + zk =6 ky + zk.
In other words, condition (S”) means exactly that one cannot find a neighbourhood of
parallel line segments on the unit sphere of equal length. This seems to be a new geometric
property which, as we will see, is essentially weaker than strict convexity. Notice that, in
contrast to many other classical properties, property (S) is not inherited by subspaces and,
in a sense, is neither local nor global.
Properties (S’) and (S”) are related to another (weaker) property of ‘non-flatness’ of the
unit sphere:
(F) the unit sphere SX of X does not contain any flat faces, that is to say, there is no
non-empty subset of SX , open in the relative norm topology, that is contained in
a hyperplane.
Here by a hyperplane of X we understand a translation of a subspace of X of codimension 1,
i.e., a set of the form x + ker(x∗ ) for some x ∈ X and x∗ ∈ X ∗ . Note, p however, that (F)
does not imply (S”) that is witnessed by the norm k(x, y, z)k = max{ x2 + y 2 , |z|} for
(x, y, z) ∈ R3 (consider the points (1, 0, 0) and (1, 0, 21 )). However, whether every Banach
space admits a renorming satisfying (F) seems to be an attractive open problem.
We employ the announced equivalence to extend Zwoleński’s result to strictly convex
Banach spaces (Corollary 2). It is well-known that not every Banach space admits a strictly
convex renorming, just to mention the examples of `∞ (Γ) for any uncountable set Γ (see
[4] and [5, §4.5]) or the quotient space `∞ /c0 ([3]). This motivates the question of whether
strict convexity and property (S) are equivalent at the level of renormings, and a negative
answer is a part of our next result.
Theorem B. Assuming that the continuum is a real-valued measurable cardinal, there
exists a non-strictly convexifable Banach space whose norm satisfies (S). Moreover, for any
Banach space X we have:
(i) if dim X 6 2, then X has property (S) if and only if X is strictly convex;
(ii) if dim X > 2 and X admits a renorming with property (S), then it also admits
a non-strictly convex renorming with property (S).
Solovay ([10]) proved that the assertion that the continuum is a real-valued cardinal is
equiconsistent with the existence of a two-valued measurable cardinal number, therefore
 STEINHAUS’ LATTICE-POINT PROBLEM 3

its consistency cannot be proved in ZFC alone (assuming of course that ZFC itself is
consistent). Interestingly, our construction in this universe is possible because the real-
valued measurability of the continuum implies the failure of the Continuum Hypothesis
([2, p. 131]) and we take advantage not only of pleasant measure-theoretic properties of
the continuum but also of the existence of an uncountable cardinal number below it.
It seems unlikely that real-measurability of the continuum is really necessary to show
that there exist Banach spaces with (S) but which do not have a strictly convex renorming.
This leaves the question of possibility of such constructions in ZFC open.

2. Proof of Theorem A
Proof of Theorem A. Since the implications (S) ⇒ (S1 ) and (S’) ⇒ (S”) hold true trivially,
it is enough to prove that (S1 ) ⇒ (S’), (S’) ⇒ (S) and (S”) ⇒ (S’).
(S1 ) ⇒ (S’): Suppose that (S1 ) holds. Fix any δ > 0 and x, y ∈ X with x 6= y,
kxk = kyk = 1. Consider any quasi-finite set A ⊂ X such that A ∩ (1 + δ)BX = {x, y},
where BX stands for the closed unit ball of X. According to (S1 ), there is a u ∈ X,
kuk < δ/2, such that for some r > 0 the open ball B(u, r) contains exactly one element of
A. Suppose there is an a ∈ A \ {x, y} belonging to B(u, r). Then
δ δ
r > ka − uk > kak − kuk > (1 + δ) − =1+ ,
2 2
hence kx − uk < r, that is x ∈ B(u, r); a contradiction. Consequently, B(u, r) contains
exactly one of the points x and y, say x ∈ B(u, r) and y 6∈ B(u, r). Then
δ δ
1− < kx − uk < r 6 ky − uk < 1 + .
2 2
Suppose that r 6 1, r = 1 − ε with some ε ∈ [0, δ/2) and take any number ρ satisfying
n δ o
0 < ρ < min r − kx − uk, − ε .
2
Obviously, we may find v ∈ X with kvk 6 ε + ρ such that ky − (u + v)k > r + ε + ρ > 1.
Then we also have
kx − (u + v)k 6 kx − uk + kvk < r − ρ + kvk 6 1.
Therefore, setting z = −(u+v) completes the proof of our claim, since we have the estimate
ku + vk < ε + ρ + δ/2 < δ. We proceed similarly in the case where r > 1 so the proof of
(S1 ) ⇒ (S’) is then complete.
(S’) ⇒ (S): Let X be a Banach space X that satisfies (S’) and let A ⊂ X be a quasi-finite
set. For any n ∈ N set

Gn = x ∈ X : |A ∩ B(x, r)| = n for some r > 0 .
It is evident, in view of the definition of a quasi-finite set, that each Gn is an open subset
of X. We shall prove that it is also dense.
4 T. KANIA AND T. KOCHANEK

Assume, in search of a contradiction, that there is an open ball U = B(x0 , r0 ) in X not

intersecting Gn . Rescaling U if necessary, we may suppose that A ∩ U = ∅. With any
point x ∈ U we associate two integers m(x) < n and k(x) > 2 defined as follows: Since
x 6∈ Gn , there is the largest non-negative integer m(x) < n for which there exists q > 0
with |A ∩ B(x, q)| = m(x). Then, for every s > q we have either |A ∩ B(x, s)| = m(x) or
|A ∩ B(x, s)| > n. Define

s = inf t > 0 : |A ∩ B(x, t)| > n .
Then exactly m(x) points a1 , . . . , am(x) ∈ A lie in the ball B(x, sr0 ), whereas at least two
such points lie on the boundary of B(x, s); let us call them b1 , . . . , bk , where k > 2. In this
way we define k(x) = k.
Now, we shall use an infinite descent argument to obtain a desired contradiction. Let
a1 , . . . , am , b1 , . . . , bk be as above for x = x0 , where m = m(x0 ) and k = k(x0 ). Pick any
δ > 0 such that
{ai : 1 6 i 6 m} ⊆ B(x0 + u, s) ∩ A ⊆ {ai , bj : 1 6 i 6 m, 1 6 j 6 k}
for every u ∈ X with kuk < δ.
Define ρ = max{kai − x0 k : 1 6 i 6 m} < s and set γ = s − ρ. Each of the vectors
(bj − x0 )/s (j = 1, . . . , k) lies in the unit sphere. Applying the hypothesis (S’) to any two
of them (e.g., to j = 1, 2), we obtain a point z ∈ X with
kszk < min{δ, γ/2}
such that one of the vectors: bj − x0 − sz (j = 1, 2) has norm greater than s, whereas the
other has norm smaller than s. By decreasing δ, if necessary, we may also assume that the
point x := x0 + sz still lies in U . Therefore the ball B(x, s) with the centre in U contains
all ai ’s (1 6 i 6 m) and at least one but not all among bj ’s (1 6 j 6 k). Observe also that
by our choice of z, we have
kai − xk 6 kai − x0 k + kszk < ρ + γ/2 = s − γ/2 for each 1 6 i 6 m
and
kbj − xk > kbj − x0 k − kszk > s − γ/2 for each 1 6 j 6 k.
Therefore, by suitably rescaling the ball B(x, s), we obtain a new ball centred at x which
contains all of ai ’s and whose boundary contains some but not all of bj ’s. This shows that
we have either m(x) > m(x0 ) or k(x) < k(x0 ). This construction (with x0 replaced by x)
will ultimately lead to a contradiction, as we finally arrive at a point u ∈ U with m(u) > n
or k(u) < 2. Therefore, all the sets Gn (n ∈ N) are open and dense.
By the Baire Category Theorem, the set Y = ∞
T
n=1 Gn is dense in X and, obviously, for
each y ∈ Y and n ∈ N there is a ball B centred at y with |A ∩ B| = n. This completes the
proof of (S).
(S”) ⇒ (S’): Assume the negation of (S’) and choose distinct unit vectors x, y ∈ X and
δ > 0 so that there is no vector z ∈ X with kzk < δ for which exactly one of the vectors
x + z and y + z lies inside the unit ball of X. For every u ∈ SX define

Vu = z ∈ SX : ku + αzk < 1 for some α > 0
 STEINHAUS’ LATTICE-POINT PROBLEM 5

and

λu (z) = min δ, inf{α > 0 : ku + αzk > 1} (z ∈ Vu ).
By the assumption, we have Vx = Vy and λx (z) = λy (z) for every z ∈ Vx , which means that
the unit sphere looks locally the same at x and y (via the translation by y − x), namely,
(1) y − x + (B(x, δ) ∩ SX ) = B(y, δ) ∩ SX .
Pick η > 0 so small that
x+z y+z
(2) x− <δ and y− <δ if kzk < η.
kx + zk ky + zk
Now, using (S”), choose a vector z ∈ X with kzk < η so that kx + zk =6 ky + zk. We have
then two possibilities: either kx + zk 6 1 and ky + zk 6 1, or kx + zk > 1 and ky + zk > 1.
We shall consider the former case; for the latter one the argument is similar.
With no loss of generality we can assume that kx + zk > ky + zk. Consider the function
g : [0, ∞) → [0, ∞) given by
g(α) = kx + z + α(y − x)k
which is convex, as can be easily verified. In view of (1) and (2), we have
x+z
y−x+ = 1,
kx + zk
that is, g(kx + zk) = kx + zk. We have also g(0) = kx + zk and g(1) = ky + zk < kx + zk.
This is a contradiction with the convexity of g, as the arguments: 0, kx + zk and 1 lie in
this order on the real line. 

3. Examples
In this section we will demonstrate some applications of Theorem A in concrete situations.
We begin with a strengthening of Zwoleński’s result.
Given two elements x, y in a real vector space X, we denote by xy the line segment
between x and y, i.e., xy = {λx + (1 − λ)y : λ ∈ [0, 1]}.
Proposition 1. Let X be a Banach space and suppose that x, y ∈ X are distinct unit
vectors. If xy 6⊆ SX , then for each δ > 0 there is z ∈ X with kzk < δ such that one of the
vector x + z, y + z has norm greater than 1 whereas the other one has norm strictly less
than 1.
Proof. Let δ > 0 and x, y ∈ X with x 6= y, kxk = kyk = 1 be given. Then each point inside
the segment xy, joining x and y, has norm smaller than 1, whereas each point lying on the
straight line passing through x and y, but outside xy, has norm larger than 1. Therefore,
any point z ∈ X satisfying 0 < kzk < δ and x + z ∈ xy does the job. 
Corollary 2. Every strictly convex Banach space X satisfies (S).
6 T. KANIA AND T. KOCHANEK

Now, we will see that strictly convex spaces do not exhaust the whole class of Banach
spaces satisfying Steinhaus’ condition. In fact, these two classes differ already in dimension
three. The following construction will also serve as a base for the proof of Theorem B.
Example 3. We claim that there exists a norm ||| · ||| in R3 such that (R3 , ||| · |||) contains
`2∞ isometrically (and hence is not strictly convex), nonetheless it satisfies condition (S).
We are indebted to the referee for suggesting the following example which significantly
simplified our original construction.
First, observe that the negation of (S”) easily implies that there are two different points
x and y on the unit sphere and δ > 0 so that kx + zk = kw + zk whenever kzk < δ and
w ∈ xy. In other words, if a given Banach space fails Steinhaus’ condition, then there must
be a ‘neighbourhood’ of segments on the unit sphere. Having this in mind we set
B = {(x1 , x2 , x3 ) ∈ [−1, 1]3 : |x3 | 6 f (x1 , x2 )},
where f : [−1, 1]2 → [0, 1] is any continuous function satisfying the equations f (0, 0) = 1
and f (−x1 , −x2 ) = f (x1 , x2 ) which vanishes on the boundary of [−1, 1]2 and is strictly
concave on (−1, 1)2 . For example, we can take f (x1 , x2 ) = (1 − |x1 |)p (1 − |x2 |)p with
0 < p < 21 . Then, let ||| · ||| be the norm on R3 defined as the Minkowski functional of
B. Since there are only four segments lying on the unit sphere (the edges of the square
[−1, 1]2 ×{0}), the Banach space (R3 , ||| · |||) satisfies Steinhaus’ condition due to the remark
above.
It is worth noticing a simple geometrical feature of B which makes ||| · ||| satisfy condition
(S’). Namely, considering any two different points x = (t, 1, 0) and y = (u, 1, 0) with
0 6 t < u < 1 we see that the curve lying on B that starts at x and is parallel to the x2 x3 -
plane is flatter at the point x than its counterpart at the point y. Therefore, for a given
δ > 0, one can take a vector z ∈ R3 with kzk < δ of the form z = (0, v, w) to guarantee that
exactly one (more precisely: the latter one) of the vectors x+z, y+z goes outside of B. For
any other two points our claim is either trivial or analogous. The upper part of the ball B
defined as above with p = 13 , as well as some contour lines illustrating the above-mentioned
flattening effect, are depicted in the two figures below.
 STEINHAUS’ LATTICE-POINT PROBLEM 7

Remark 4. The above example shows that there is a Banach space X satisfying (S’), but
containing a pair of distinct vectors x, y ∈ X, with kxk = kyk = 1, such that for some
δ > 0 it is impossible to increase kxk and decrease kyk by adding to x and y the same
vector z ∈ X with kzk < δ. In other words, condition (S’) cannot be strengthened by
claiming which one of x + z and y + z has norm greater than 1.
Remark 5. Of course, if X is a non-strictly convex Banach space with dim X = 2, then
condition (S’) fails to hold. Therefore, Steinhaus’ condition is equivalent to strict convexity
in the class of Banach space with dimension at most 2.
The next corollary demonstrates that the classical L1 (µ)-spaces for atomless measures
µ also satisfy Steinhaus’ condition, giving thus another example of a non-strictly convex
space with this property. Recall that a set A in a measure space is called an atom if
µ(A) > 0 and µ(B) ∈ {0, µ(A)} for every measurable subset B of A.
Proposition 6. Let (Ω, Σ, µ) be a measure space. Then, the space L1 (µ) satisfies (S) if
and only if Ω contains at most one atom (up to measure-zero sets).
Proof. By Luther’s theorem [8], there is a decomposition µ = µ1 + µ2 with µ1 being semi-
finite (i.e., for each A ∈ Σ with µ1 (A) = ∞, there is a subset B ∈ Σ of A such that
0 < µ1 (B) < ∞) and µ2 being degenerate (i.e., the range of µ2 is contained in {0, ∞}).
The space L1 (µ) is then isometrically isomorphic to L1 (µ1 ) (for any f ∈ L1 (µ) we have
µ2 ({x : f (x) 6= 0}) = 0, thus the identity map yields the desired isometry). Therefore, we
consider only the case where µ is semi-finite.
First, suppose that (Ω, Σ, µ) is atomless. Fix two functions f, g ∈ L1 (µ) with f 6= g and
kf k = kgk = 1, and let δ > 0 be given. Interchanging f and g, if necessary, we may assume
that there is a set F ∈ Σ such that 0 < µ(F ) < ∞ and f (ω) > g(ω) for ω ∈ F . Since
∞ n
[ 1o
F = ω ∈ F : f (ω) > g(ω) + ,
n=1
n
we may also suppose that for some ε > 0 and all ω ∈ F we have f (ω) > g(ω) + ε.
Approximating f and g by step functions we may find a measurable set F 0 ⊂ F with
µ(F 0 ) > 0 and some cf , cg ∈ R such that
ε ε
|f (ω) − cf | < and |g(ω) − cg | < (ω ∈ F 0 ).
5 5
Hence, cf > cg + 35 ε and mf > Mg + 15 ε, where mf = ess inf f (F 0 ) and Mg = ess sup g(F 0 ).
We have three possibilities:
(i) mf > 0 and Mg > 0,
(ii) mf > 0 and Mg < 0,
(iii) mf 6 0 and Mg < 0.
With no loss of generality suppose that either (i) or (ii) occurs (the case (iii) is analogous
to (i)). Then there is a positive number d such that |mf − d| < mf and |Mg − d| > |Mg |;
indeed, in the former case we shall take any d ∈ (2Mg , 2mf ), while in the latter one any
sufficiently small d does the job.
8 T. KANIA AND T. KOCHANEK

Now, observe that for almost all ω ∈ F 0 we have

(3) |f (ω) − d| < |f (ω)| and |g(ω) − d| > |g(ω)|.
Indeed, for the first inequality note that in the case where f (ω) > d > 0 it holds trivially
true, while in the opposite case we have
|f (ω) − d| = d − f (ω) 6 d − mf 6 |d − mf | < mf 6 |f (ω)|.
For the other one observe that since Mg < d (recall |Mg − d| > |Mg |), we have g(ω) < d,
thus in the case where g(ω) > 0 we have
|g(ω) − d| = d − g(ω) > d − Mg = |d − Mg | > |Mg | > g(ω) = |g(ω)|,
whereas in the case where g(ω) < 0 this inequality is trivial.
By the Darboux property of finite atomless measures ([9, Théorème, p. 16], see also [6,
§215]), there is a measurable set H ⊂ F 0 with 0 < µ(H) < δ/d. Then kd · 1H k < δ,
where 1H stands for the characteristic function of H, while inequalities (3) imply that
kf − d · 1H k < 1 and kg − d · 1H k > 1. This proves assertion (S’), and hence also (S).
In the case where there is exactly one atom A ⊂ Ω (up to measure-zero sets), either
µ(Ω \ A) = 0, which means that L1 (µ) ∼ = R isometrically, or there exists an atomless part
B ⊂ Ω of positive measure so that Ω = A ∪ B. In the latter case, fix any f, g ∈ L1 (µ)
with f 6= g and kf k = kgk = 1. First, assume that f |B = g|B outside a set of measure
zero. Both f and g are constant almost everywhere R on A; denote those constant Rvalues as
cf and cg , respectively. Since kf k = |cf |µ(A) + B |f | dµ and kgk = |cg |µ(A) + B |f | dµ,
we have |cf | = |cg | and cf 6= cg . So, assuming that cf > 0 and cg < 0, for any given δ > 0
we have kf + δ · 1A k > 1 and kg + δ · 1A k < 1. In the case where f |B and g|B do not
coincide almost everywhere, we repeat the argument from the first part of the proof for the
atomless measure space (B, Σ0 , µ|Σ0 ), where Σ0 = {B ∩ C : C ∈ Σ}. Consequently, L1 (µ)
has property (S) whenever the underlying measure space contains at most one atom.
Finally, suppose Ω contains two disjoint atoms, say A1 and A2 . Consider the functions
f = 21 (1A1 + 1A2 ) and g = 41 1A1 + 34 1A2 . Obviously, for δ ∈ (0, 14 ) there is no function h
with khk < δ so that exactly one of f + h and g + h has norm larger than 1, thus in this
case (S) fails to hold. 
Theorem A gives an immediate answer to the question about Steinhaus’ property for
C0 (K)-spaces, and it is unsurprisingly negative except the trivial case where the considered
space is one-dimensional.
Corollary 7. Let K be a locally compact Hausdorff space that contains at least two points.
Then the space C0 (K) consisting of scalar-valued functions on K vanishing at infinity does
not have property (S).
Proof. Pick any two distinct points u, v ∈ K, and their disjoint neighbourhoods U and V .
Since K is completely regular, there is a continuous map ϕ : K → [0, 1] such that ϕ(u) = 1
and ϕ|K\U = 0. Similarly, since K \ U is also completely regular, there is a continuous
 STEINHAUS’ LATTICE-POINT PROBLEM 9

map ϕ1 : K \ U → [0, 1/2] such that ϕ1 (v) = 1/2 and ϕ1 |K\(U ∪V ) = 0. Then the mapping
ψ : K → [0, 1] defined by

ϕ(x) for x ∈ U,
ψ(x) =
ϕ1 (x) for x ∈ K \ U,
is continuous and, of course, ϕ 6= ψ. So, both functions ϕ and ψ belong to the unit sphere
of C0 (K), but for any δ ∈ (0, 1/2) condition (S’) is violated. 

4. Proof of Theorem B
Here, we shall construct a Banach space with property (S) but without any strictly
convex renorming. Assume that the continuum c is a real-valued cardinal number. This
implies that there is an atomless, c-complete probability measure µ defined on the power
set of a set Ω with the cardinality of the continuum (see, e.g., [7, §543B(c)])—here, by a λ-
complete measure µ (λ is an uncountable cardinal) we understand a measure satisfying the
following condition: for every cardinal κ < λ and for every family (Aα )α<κ of measurable
sets, their union A is measurable and
n [
o
µ(A) = sup µ Aα : F ⊂ κ finite .
α∈F

The statement that the continuum is a real-valued cardinal is equiconsistent with the
existence of a two-valued measurable cardinal ([10]), which is stronger than the consistency
of ZFC alone. Banach and Kuratowski ([2, p. 131]) proved that if such a measure exists,
then the Continuum Hypothesis fails to hold, hence there exists at least one uncountable
cardinal below the continuum. We will show that for any set Γ with |Γ| < c, the Bochner
space L1 (µ, `∞ (Γ)) satisfies Steinhaus’ condition. In particular, if Γ is uncountable, such
space does not have a strictly convex renorming as it contains `∞ (Γ) embedded via constant
functions and this space does not have such a renorming by a result of Day ([4], see also
[5, §4.5]).
Theorem 8. Assume that c is a real-valued cardinal number and let Γ be a set with car-
dinality less than c. Then the Bochner space X = L1 (µ, `∞ (Γ)) has property (S) for some
atomless, probability measure µ.
Proof. As c is assumed to be a real-valued cardinal number, there exists an atomless prob-
ability measure space (Ω, ℘(Ω), µ), where Ω is a set with the cardinality of the continuum
and µ is c-complete. Then µ is the required measure.
Let f 6= g be two norm-one elements of X. Since members of X are equivalence classes
of the relation of equality almost everywhere, let us work with concrete representatives
f, g : Ω → `∞ (Γ). Fix δ > 0. There exists n0 ∈ N such that µ(Fn0 ) > 0 where
n 1o
Fn0 = ω ∈ Ω : kf (ω) − g(ω)k`∞ (Γ) > .
n0
10 T. KANIA AND T. KOCHANEK

For each γ ∈ Γ let

n 1o
Gγ = ω ∈ Fn0 : |f (ω)(γ) − g(ω)(γ)| > .
n0
Since µ is defined on the power set of Ω, there is no problem with measurability of the
sets Gγ (γ ∈ Γ). Also, as µ is c-complete and |Γ| < c, the set Gγ0 has positive measure for
some γ0 ∈ Γ. Interchanging f with g, if necessary, we may suppose that the set
n 1o
F = ω ∈ Gγ0 : f (ω)(γ0 ) > g(ω)(γ0 ) +
n0
has positive measure. Now, we proceed as in the proof of Proposition 6.
Approximating the functions ω 7→ f (ω)(γ0 ) and ω 7→ g(ω)(γ0 ) (ω ∈ Ω) by step functions
we may find a set F 0 ⊂ F with µ(F 0 ) > 0 and some cf , cg ∈ R such that for almost all
ω ∈ F 0 we have
1 1
|f (ω)(γ0 ) − cf | < and |g(ω)(γ0 ) − cg | < .
5n0 5n0
Hence, cf > cg + 5n3 0 and mf > Mg + 5n1 0 , where mf = ess inf f (F 0 ) and Mg = ess sup g(F 0 ).
We have then three possibilities:
(i) mf > 0 and Mg > 0,
(ii) mf > 0 and Mg < 0,
(iii) mf 6 0 and Mg < 0,
which we tackle completely analogously as in the proof of Proposition 6 (here f (ω)(γ0 ) and
g(ω)(γ0 ) play the rôle of f (ω) and g(ω), respectively.) Therefore (assuming either (i) or
(ii) holds true), we observe that for some d > 0 and almost all ω ∈ F 0 we have
(4) |f (ω)(γ0 ) − d| < |f (ω)(γ0 )| and |g(ω)(γ0 ) − d| > |g(ω)(γ0 )|.
Since µ is atomless, there is a measurable set H ⊂ F 0 with 0 < µ(H) < δ/d. Then
kd·δγ0 ·1H kX < δ, where δγ0 ∈ `∞ (Γ) stands for the element that is zero apart from the
γ0th coordinate where it assumes value 1. Hence, (4) imply that kf − d·δγ0 ·1H kX < 1 and
kg − d·δγ0 ·1H kX > 1, as desired. 
Remark 9. Let us note that one may replace L1 (µ, `∞ (Γ)) in the statement of Theorem 8
with L1 (µ, Y ) where Y is any subspace of `∞ (Γ) containing c0 (Γ).
Corollary 10. Under the assumptions of Theorem 8, for every uncountable set Γ with
cardinality less than the continuum, the Banach space X = L1 (µ, `∞ (Γ)) has (S), yet it
lacks a strictly convex renorming.
We have thus proved the first assertion of Theorem B; clause (i) has been also already
observed; see Remark 5. It remains to prove clause (ii).
Proof of Theorem B (continued). Assume that a Banach space X with dim X > 2 has
a norm k · k satisfying (S); we can assume that this norm is in fact strictly convex, as
otherwise we are done. In the case where dim X = 3, the assertion is proved by Example 3,
so assume that dim X > 3. Choose any subspace Y ⊂ X of codimension 2 so that we have
 STEINHAUS’ LATTICE-POINT PROBLEM 11

X = Y ⊕ R ⊕ R and every element x ∈ X may be typically written as (y, α, β) with y ∈ Y ,

α, β ∈ R. (In fact, the symbols R formally stand for some fixed one-dimensional subspaces
of X.) Note that Y is a strictly convex space of dimension at least 2. Let k·k0 be a new
norm on X given by the decomposition X = (Y ⊕`1 R) ⊕`2 R, that is
p
kxk0 = (kyk + |α|)2 + |β|2 (x = (y, α, β)).

As any two finite direct sums of the same normed spaces are isomorphic, this defines
an equivalent norm on X which obviously fails to be strictly convex. Next, we shall show
that it has property (S).
For, suppose x1 = (y1 , α1 , β1 ) and x2 = (y2 , α2 , β2 ) are two distinct points from the unit
sphere of (X, k·k0 ). If β1 6= β2 , then (k(y1 , α1 )k, β1 ) and (k(y2 , α2 )k, β2 ) are two distinct
points on the unit circle, where the norm symbol stands for the `1 -norm on Y ⊕ R. Thus,
by manipulating the coordinates α and β we obtain a vector z of the form (0, α, β), and
of arbitrarily small length, so that kx + zk0 6= ky + zk0 .
Now, suppose that β1 = β2 and hence ky1 k + |α1 | = ky2 k + |α2 |. If y1 = y2 , then it
must be α1 = −α2 6= 0, whence we easily find a desired vector z being of the form (0, α, 0).
So, assume we have y1 6= y2 . In this case, we can find z with the aid of following simple
observation:

Claim. Since Y is strictly convex and dim Y > 2, for every pair of distinct vectors y1 , y2 ∈ Y
and every δ > 0 there exists z ∈ Y such that kzk < δ and ky1 + zk − ky1 k = 6 ky2 + zk − ky2 k.
Indeed, if the vectors y1 and y2 are linearly independent, we take z = ηy1 for suitably small
η > 0. Then, ky1 + zk − ky1 k = ηky1 k and this is equal to ky2 + zk − ky2 k if and only
if ky2 + zk = ky2 k + kzk, which is impossible as the norm is strictly convex. In the case
where y2 = γy1 for some γ ∈ R, we pick any vector z that is linearly independent of y1
and satisfies kzk < δ. Then, assuming with no loss of generality that |γ| > 1, the required
condition becomes kγy1 + zk = 6 (|γ| − 1)ky1 k + ky1 + zk which again follows from the strict
convexity of Y . The claim has been thus proved.
Now, take a vector z ∈ Y as in the above claim. Then we have:
ky1 + (z, 0, 0)k0 6= ky2 + (z, 0, 0)k0 ⇐⇒
ky1 + zk + |α1 | =
6 ky2 + zk + |α2 | ⇐⇒
(because ky1 k0 = ky2 k0 and β1 = β2 )
ky1 + zk − ky1 k =
6 ky2 + zk − ky2 k,

which is true. Therefore, we have checked that (X, k·k0 ) satisfies condition (S”). 
Remark 11. The last proof shows that in all dimensions > 4 there exist easy examples
of non-strictly convex Banach spaces satisfying (S). However, in the ‘`1 -`2 -sum argument’
above we heavily used the assumption that dim X > 4, so seemingly one cannot avoid a bit
more involved geometric argument in the case dim X = 3 (as in Example 3).
12 T. KANIA AND T. KOCHANEK

References
[1] F. Albiac and N.J. Kalton, Topics in Banach Space Theory, Graduate Texts in Mathematics 233,
Springer, 2006.
[2] S. Banach and K. Kuratowski, Sur une généralisation du problème de la mesure. Fund. Math. 14
(1929), 127–131.
[3] J. Bourgain, `∞ /c0 has no equivalent strictly convex norm, Proc. Amer. Math. Soc. 78 (1980), 225–
226.
[4] M.M. Day, Strict convexity and smoothness, Trans. Amer. Math. Soc. 78 (1955), 516–528.
[5] J. Diestel, Geometry of Banach Spaces – Selected Topics, Lecture Notes in Math. 485, Springer–Verlag,
Berlin–Heidelberg–New York 1975.
[6] D.H. Fremlin, Measure Theory: Broad Foundation (vol. 2), Torres Fremlin 2001.
[7] D.H. Fremlin, Measure Theory: Set-Theoretic Measure Theory (vol. 5), Torres Fremlin 2008.
[8] N.Y. Luther, A decomposition of measures, Canadian J. Math. 20 (1968), 953–958.
[9] W. Sierpiński, Sur les fonctions d’ensemble additives et continues, Fund. Math. 3 (1922), No. 1,
240–246.
[10] R.M. Solovay, Real-valued measurable cardinals, Axiomatic set theory (Proc. Sympos. Pure Math.,
Vol. XIII, Part I, Univ. California, Los Angeles, Calif., 1967) Amer. Math. Soc., Providence, R.I.,
1971, pp. 397–428.
[11] H. Steinhaus, One Hundred Problems in Elementary Mathematics, Dover Publ. 1965.
[12] P. Zwoleński, Some generalization of Steinhaus’ lattice points problem, Colloq. Math. 123 (2011),
129–132.

School of Mathematical Sciences, Western Gateway Building, University College Cork,

Cork, Ireland and Mathematics Institute, University of Warwick, Gibbet Hill Rd, Coven-
try, CV4 7AL, England
E-mail address: [email protected], [email protected]

Institute of Mathematics, Polish Academy of Sciences, Śniadeckich 8, 00-656 Warsaw,

Poland and Institute of Mathematics, University of Warsaw, Banacha 2, 02-097 Warsaw,
Poland
E-mail address: [email protected]