arXiv:2104.02695v4 [math.FA] 3 Sep 2021 rnlto feeycmatstwihle nacoe ball closed a in lies which compact every of translation a hrfr oaseigysrne property. stronger seemingly a to therefore eso fHa ulst,wihfrsprbeBnc spaces Banach separable for which th introduc sets, set Darji knowledge, null Borel that author’s Haar mentioning the of worth of version also is best It the unanswered. To compactivorous? be fadol fte r o armar.Orami oso oec set some compactivorous show a to given is Bana aim separable) that, Our necessarily prove (not we meagre. in Haar sets not compactivorous are of they if only and if ercspace metric egein meagre td fcmatvru esi oiae ytefloigq following introducti the thorough by motivated a s provide is The 6, sets decades. for Chapter compactivorous on of [2], going as study been well has as them on [6], research and [4] in lsd ovxsti eaal aahspace Banach separable a in set convex closed, K that such ihcmatsupport compact with Abstract ntecneto aahsae.Let spaces. Banach of context the in ocutbepout flclycmatPls groups. Polish exten compact an locally prove of and products countable spaces to Banach characteri nonseparable some give and We separable . Borel of nonnegligibility E K Keywords: Classification: Subject 2020 2021 7, September obe to arnl.Let null. Haar in x rbblt probability fti rprycnb on n[] hr ti rvdda s a spa as Banach provided separable is a in it set set where Borel Borel [7], given A in a found that be ensures which can property this of nasprbeBnc pc,ti sasffiin odto wh condition sufficient a is this space, Banach separable a In . , ∈ in arnl eswr rtdfie neeyAeinPls group Polish Abelian every in defined first were sets null Haar h ups ftepeetppri osuytento fcomp of notion the study to is paper present the of purpose The X µ X ( X K and compactivorous ti lotimdaet e h oe opcioossets compactivorous Borel why see to immediate almost is It . hr sannmt,(eaiey pnsbe of open (relatively) nonempty, a is there ∩ E opcioosst arnl e,Ha egeset. meagre Haar set, null Haar set, compactivorous set A . K OPCIOOSST NBNC SPACES BANACH IN SETS COMPACTIVOROUS V x M nasprbeBnc space Banach separable a in ) E ∈ ∩ M > o all for V X nasprbeBnc space Banach separable a in E 0 n otnosfunction continuous a and 6= hc implies which , uhthat such E ecmatvru n osdraBrlpoaiiymeasure probability Borel a consider and compactivorous be ∅ x µ naBnc space Banach a in and f o n opc set compact any for if, ∈ ihcmatspoton support compact with K X ⊂ x bev htcoe ust of subsets closed that Observe . K + X ydfiiin hr sa pnset open an is there definition, By . K ∩ 1. AIERAVASINIDAVIDE µ V ∩ ( Introduction − V 62 piay,4B0 41 (secondary). 54H11 46B50, (primary), 46B20 6= x ⊂ E ∅ + X X E 1 easti aahspace Banach a in set a be E and scmatvru ffreeycmatset compact every for if compactivorous is Since . sHa egei hr xssacompact a exists there if meagre Haar is ) E f X > hr is there , x : K 0 is + . M X X ⊂ arnull Haar K K If . → uhthat such X ∩ K ∩ ain fti rpryi both in property this of sations hr xs noe set open an exist there , C X V > r V hc a etasae into translated be can which sntHa ul utit must null, Haar not is soe n oetyin nonemtpy and open is uhthat such ⊂ hsae.I particular, In spaces. ch ino h antheorem main the of sion 0 X c urnesteHaar the guarantees ich fteeeit Borel a exists there if E di 5 categorical a [5] in ed nt h oi.Our topic. the to on uhthat such µ h rtmention first The . eto:let uestion: fradius of r compactivorous are ed sflos a follows: as reads ei o arnull. Haar not is ce ( ffiin condition ufficient x usini still is question e re rils[3], articles urvey + haracterisations yChristensen by cioossets activorous f V E X − 0 = ) 1 and . E ( antbe cannot r x E leading , contains + C x ssaid is o all for E ea be ∈ ) X V is µ 2 DAVIDE RAVASINI

Sections 2 and 3 are devoted to the presentation of two of the main tools that will be needed in the proof of the main theorem, which is developed in Section 4. Section 5 shows a way to extend the main theorem to countable products of arbitrary, locally compact Polish groups. The standard notation of Banach space theory is used. If X is a Banach space, BX denotes the closed unit ball of X, whereas B(x, ε) denotes the open ball centered in x and with radius ε. Given a set S, span S denotes the smallest closed subspace which contains it. The and of a set S in a are denoted int S and cl S respectively S and 2 stands for the power set of a set S. ℓ1 is the Banach space of absolutely summable sequences, whereas c0 is the Banach space of sequences which converge to 0. All Banach spaces are assumed to be real.

2. The rescaling property The following, simple property of compactivorous sets works as a motivation for the definition we are going to introduce later. Lemma 2.1. Let E be a compactivorous set in a Banach space. Then E is com- pactivorous by rescaling. That is: for each compact set K ⊂ X there are x ∈ X and δ > 0 such that x + δK ⊂ E. Proof. Let K be a compact subset of X. Since every compact set is contained in a convex compact set, we can assume that K is convex. Pick an V and x ∈ X such that K ∩ V 6= ∅ and x + K ∩ V ⊂ E. Without loss of generality, we can assume that 0 ∈ K ∩ V and that V = B(0,ε) for some ε > 0. Let δ ∈ (0, 1) be such that δK ⊂ V . Since K is convex, δK ⊂ K, hence δK ⊂ K ∩ V . Thus, x + δK ⊂ E.  Let F be a family of subsets in a separable Banach space X. Drawing inspiration from Lemma 2.1, we say that F has the rescaling property if the two following conditions are satisfied. (1) F is hereditary: if A ∈F and B ⊆ A, then B ∈F. (2) For each compact set K ⊂ X, there are δ > 0 and x ∈ X such that x + δK ∈F. Lemma 2.2. Let E be a subset of a Banach space X. E is compactivorous by rescaling if and only if every closed, separable subspace Y ⊆ X admits a family of subsets with the rescaling property FY such that every S ∈ FY can be translated into E. Proof. Suppose that E is compactivorous by rescaling and let Y be a separable subspace of X. For each compact set K ⊂ Y , choose δK > 0 such that δK · K can be translated into E and define

FY = {S ⊂ Y : there is a compact set K ⊂ Y such that S ⊆ δK · K}.

FY is the family with the rescaling property we were looking for. Conversely, let K ⊂ X be compact and define Y = span K. Y is a separable subspace of X and admits therefore a family of subsets FY with the rescaling property such that every S ∈FY can be translated into E. Hence, there are y ∈ Y and δ > 0 such that y + δK ∈FY and z ∈ X such that z + y + δK ⊂ E. Setting x = z + y, we have x + δK ⊂ E and since K is arbitrary, this concludes the proof.  COMPACTIVOROUS SETS IN BANACH SPACES 3

As we shall see later, Lemma 2.2 allows to reduce the study of compactivorous sets to separable Banach spaces by looking at families with the rescaling property instead of compactivorous sets themselves.

3. Saturated compact subsets of ℓ1

The second tool we need is the concept of saturated compact subsets of ℓ1. To have a first understanding of what they are it is important to recall the following, well-known characterisation of compact subsets of ℓ1.

Proposition 3.1. A set S ⊂ ℓ1 is totally bounded if and only if (1) S is pointwise bounded, i.e. for any n ≥ 1 the set {x(n): x ∈ S} is bounded; (2) For any ε> 0 there is n0 ≥ 1 such that ∞ |x(j)| <ε jX=n

for any x ∈ S and n ≥ n0.

In particular, a set K ⊂ ℓ1 is compact if and only if K is closed and satisfies conditions (1) and (2).

Proposition 3.1 suggests we introduce the following definition: given x ∈ ℓ1, the modulus of decay of x is the sequence ωx ∈ c0 given by ∞

ωx(n)= |x(j)|. jX=n

Notice that, for each x ∈ ℓ1, the sequence ωx is nonnegative and decreasing, that

ωx(1) = ||x||ℓ1 and that the function ϕ : ℓ1 → c0 which assigns to each x ∈ ℓ1 its modulus of decay is continuous. If K ⊂ ℓ1 is compact, Proposition 3.1 guarantees that the sequence ΩK given by

ΩK (n) = sup ωx(n) x∈K belongs to c0, and we shall call it the modulus of decay of the set K. We say that a compact set K ⊂ ℓ1 is saturated if it consists exactly of all elements x ∈ ℓ1 such that ωx ≤ ΩK . Saturated compact sets are clearly symmetric and convex. However, they are far from being a countable product of intervals. Consider for instance ∞ 1 1 K = − , ⊂ ℓ .  n2 n2  1 nY=1 K is a compact set whose modulus of decay is − π2 π2 n 1 1 Ω (1) = , Ω (n)= − if n> 1, K 6 K 6 j2 Xj=1 2 but the element x = (π /6, 0, 0,... ) does not belong to K, even though ωx ≤ ΩK . Finally, it is also important to keep in mind that to every nonnegative and decreasing sequence z ∈ c0 there corresponds a saturated compact subset of ℓ1 whose modulus of decay is z, namely the set K(z) = {x ∈ ℓ1 : ωx ≤ z}. K(z) is compact by Proposition 3.1 and ΩK(z) = z by definition. In particular, this implies that every compact set is contained in a saturated compact set with the same modulus of decay. 4 DAVIDE RAVASINI

To conclude this preliminary part, the next lemma describes what happens if we intersect a saturated compact set with a closed ball centered around the origin.

Lemma 3.2. Let K ⊂ ℓ1 be a saturated compact set with modulus of decay ΩK .

For any r > 0, the compact set L = rBℓ1 ∩ K is also saturated and its modulus of decay ΩL is given exactly by ΩL(n) = min{r, ΩK(n)}.

Proof. Let x ∈ ℓ1 be such that ωx(n) ≤ min{r, ΩK (n)}. Then x ∈ K because K is saturated, and ||x||ℓ1 = ωx(1) ≤ r. Thus x ∈ rBℓ1 ∩ K. Conversely, if x ∈ L, then ωx(n) ≤ ΩK (n) for every n because L ⊆ K, moreover ωx(n) ≤ ωx(1) ≤ r, therefore ωx(n) ≤ min{r, ΩK(n)} for each n. 

4. The main theorem We are now ready to work towards a characterisation of compactivorous sets. As anticipated, we are actually going to study families with the rescaling property and perform a translation of the obtained results back in the original setting of compactivorous sets. We start with a lemma. Lemma 4.1. Let X be a Banach space, let Y be a closed subspace and denote with π : X → X/Y the quotient map. For each r> 0, the correspondence X −1 fr : int(rBX/Y ) → 2 , fr(x)= π (x) ∩ rBX , assumes closed, convex and nonempty values and it is lower hemicontinuous.

Proof. The only nontrivial assertion to verify is that fr is lower hemicontinuous for every r> 0. To see this, define a second correspondence X −1 gr : int(rBX/Y ) → 2 , gr(x)= π (x) ∩ int(rBX ). gr has nonempty images because int(rBX/Y )= π int(rBX ) and, since the corre- −1 spondence π is lower hemicontinuous and int(rBX ) is open, gr is also lower hemi- continuous. We want to show that fr is the closure of gr, i.e. that cl gr(x) = fr(x) for each x ∈ int(rBX/Y ). Then the lower hemicontinuity of f follows  from [1], −1 −1 Lemma 17.22. Clearly, we have cl π (x) ∩ int(rBX ) ⊆ π (x) ∩ rBX for every −1 −1 x ∈ int(rBX/Y ). Let y ∈ π (x) ∩ rBX and fix a vector z ∈ π (x) ∩ int(rBX ). −n −n For all positive integers n, define the elements xn = (1 − 2 )y +2 z. It is easy −1 to check that xn ∈ π (x) ∩ int(rBX ) for each n and the sequence (xn) converges −1 to y, hence y ∈ cl π (x) ∩ int(rBX ) and the equality cl g(x) = f(x) holds.    The next theorem characterises families with the rescaling property in a very strong sense. Theorem 4.2. Let X be a separable Banach space and let F be a hereditary family of subsets of X. The two following assertions are equivalent. (1) F has the rescaling property. (2) There is r> 0 with the property that, for every compact set K ⊂ rBX , one can find an x ∈ X such that x + K ∈F. Proof. First, let us observe that (2)⇒(1) is obvious, because for every compact set K there is δ > 0 such that δK fits inside rBX . Further, let T : X → Z be an isomorphism between X and another separable Banach space Z and consider the hereditary family of subsets T (F) = {T (S): S ∈ F}. It is trivial to check that F satisfies condition (1) in the statement of the theorem if and only if T (F) does, COMPACTIVOROUS SETS IN BANACH SPACES 5 and the same holds for condition (2). Thus, it suffices to focus our attention on quotient spaces of ℓ1. From now on, we assume that X = ℓ1/Y , where Y is a closed subspace of ℓ1, and that π : ℓ1 → X is the quotient map. Arguing by way of contradiction, suppose that, for each positive integer n, there −1 is a compact set Kn ⊂ int(n BX ) such that no translation of Kn belongs to F. Consider for each n the correspondence

−1 ℓ1 −1 −1 fn : int(n BX ) → 2 , fn(x)= π (x) ∩ n Bℓ1 , which is lower hemicontinuous by Lemma 4.1. For each n, let sn be a continuous selection of fn, which exists thanks to Michael’s selection theorem (see [8], Theorem ′′ −1 3.2 ) and observe that sn(Kn) is a compact set in n Bℓ1 . Define ∞

K = {0} ∪ sn(Kn). n[=1

K is compact and it comes therefore with its modulus of decay ΩK . Let L be the saturated compact set with the same modulus of decay and let H ⊂ ℓ1 be another saturated compact set whose modulus of decay ΩH satisfies Ω (j) lim H =+∞. j→∞ ΩK (j) Choose δ > 0 and observe that δH is another saturated compact set whose modulus of decay is δΩH . By our choice of H, there is the smallest index j0 such that

δΩH (j) > ΩK (j) for all j ≥ j0. If we set t = ΩK (j0) and let y ∈ tBℓ1 ∩ L, then by Lemma 3.2 we have ωy(j) ≤ min{t, ΩK (j)} ≤ δΩH (j) for all positive integers −1 j, which means that tBℓ1 ∩ L ⊆ δH. Let n be such that n ≤ t and observe −1 that sn(Kn) ⊆ n Bℓ1 ∩ L ⊆ tBℓ1 ∩ L ⊆ δH. If we apply the quotient map, we get that Kn = π sn(Kn) ⊆ δπ(H). This implies that there is no x ∈ X such that x + δπ(H) ∈F , otherwise we would conclude that x + Kn ∈F, because F is hereditary. As π(H) is compact and δ is arbitrary, this shows that (1) fails.  We are finally ready to combine all the results we have obtained so far in the characterisation theorem for compactivorous sets. Theorem 4.3. Let E be a subset in a Banach space X. Then the following asser- tions are equivalent. (1) E is compactivorous. (2) For every compact set K ⊂ X, there are x ∈ X and δ > 0 such that x + δK ⊂ E. (3) There is r> 0 with the property that, for every compact set K ⊂ rBX , one can find an x ∈ X such that x + K ⊂ E. Proof. The implication (1)⇒(2) is the content of Lemma 2.1, whereas (3)⇒(1) is obvious: if condition (3) is satisfied, then it suffices to intersect every compact set with a sufficiently small open ball to prove that E is compactivorous. It remains to show that (2)⇒(3). If E satisfies (2) and Y is a closed, separable subspace of X, then by Lemma 2.2 there is a family F of subsets of Y with the rescaling property such that every element of F can be translated into E. By Theorem 4.2, there is r > 0 with the property that, for each compact set K ⊂ rBY , there is y ∈ Y such that y + K ∈F. In particular, every compact subset of rBY can be translated into E. 6 DAVIDE RAVASINI

This justifies the introduction of the following notation: given a closed, separable subspace Y ⊆ X, we denote by ρ(E, Y ) the supremum of all r> 0 such that every compact subset of rBY can be translated into E. Observe that if Y and Z are closed, separable subspaces of X such that Y ⊆ Z, then ρ(E,Z) ≤ ρ(E, Y ). Since every compact set lies in a closed, separable subspace, to ensure that E satisfies (3) it remains to prove that there is r > 0 such that ρ(E, Y ) > r for all closed, separable subspaces Y . Arguing by contradiction, suppose that for each positive −1 integer n there is a closed, separable subspace Yn such that ρ(E, Yn) ≤ n . Define ∞ Y = span Y .  n n[=1 Y is a closed, separable subspace of X and therefore it has its ρ(E, Y ) > 0. At the −1 same time though, we have Y ⊇ Yn for each n, hence ρ(E, Y ) ≤ ρ(E, Yn) ≤ n for each n, which is nonsense.  Corollary 4.4. Let X be a Banach space, Y a closed subspace and E a subset of X/Y . Let π : X → X/Y be the quotient map. Then E is compactivorous if and only if π−1(E) is compactivorous. Proof. Assume that π−1(E) is compactivorous and let s be a continuous selection of the correspondence π−1 : X/Y → 2X (we use once again [8], Theorem 3.2′′). If K ⊂ X/Y is compact, so is s(K), hence there are x ∈ X and δ > 0 such that x + δs(K) ⊂ π−1(E). Applying π, one gets π(x)+ δK ⊂ E. As K is arbitrary, this shows that E is compactivorous. Conversely, let E be compactivorous and let K ⊂ X be compact. Then there are x ∈ X and δ > 0 such that π(x)+ δπ(K) ⊂ E, which implies x + δK ⊂ π−1(E). The conclusion follows once again because K is arbitrary. 

5. Generalising to topological groups The notion of a compactivorous set can be extended to arbitrary Hausdorff topological groups. Let G be a Hausdorff topological group, let e be its unit and let E be a subset of G. We say that E is compactivorous if for every compact set K ⊆ G there are an open set V ⊆ G and g,h ∈ G such that K ∩ V 6= ∅ and g(K ∩ V )h ⊆ E. We shall also say that E is strongly compactivorous if there is a closed neighbourhood U of e such that, for every compact set K ⊆ U, there are g,h ∈ G such that gKh ⊆ E. Clearly, every strongly compactivorous set is compactivorous. Let us say that G is fattening if also the converse holds, i.e. if every compactivorous subset of G is strongly compactivorous. Locally compact groups are trivially fattening and that every Banach space is fattening is the content of Theorem 4.3. The next example shows that there are in fact nonfattening groups. Proposition 5.1. Let X be an infinite-dimensional Banach space with the Schur property. Then X is nonfattening when endowed with the weak topology.

Proof. We show that the unit ball BX is compactivorous but not strongly com- pactivorous. Let K be a weakly compact set in X. Since X has the Schur property, by the Eberlein-Šmulian theorem K is also norm-compact. The restrictions of both the weak and the norm topology to K make it a compact, Hausdorff space, hence they must agree. BX has nonempty interior in the norm topology, which means that it is compactivorous in the norm topology. Let U be a norm-open set such COMPACTIVOROUS SETS IN BANACH SPACES 7 that K ∩ U is nonempty and can be translated into BX . Let V be a weakly open set such that K ∩ U = K ∩ V . Then K ∩ V can be translated into BX and, since K is arbitrary, this shows that BX is compactivorous in the weak topology. However, BX cannot be strongly compactivorous, as every neighbourhood of 0 is unbounded and contains therefore compact sets of arbitrarily large diameter.  Finally, we prove that countable products of locally compact Polish groups are fattening. ∞ Theorem 5.2. Let {Gn}n=1 be a countable family of locally compact Polish groups and set ∞

G = Gn. nY=1 Then G is fattening. Proof. We start by fixing some notation. Let d be a complete metric which gener- ates the topology of G and, for each n ≥ 1, let pn : G → Gn be the usual projection map. Let en be the unit of Gn and e = (e1,e2,... ) be the unit of G. Let E be a −1 compactivorous subset of G, set Bk = {g ∈ G : d(g,e) ≤ k } for each positive integer k and, looking for a contradiction, assume that for each k there is a compact set Kk ⊆ Bk such that aKkb does not lie entirely in E for any a,b ∈ G. The set ∞

K = {e} ∪ Kk k[=1 is compact and so are the sets pn(K) for all n ≥ 1. Since Gn is locally compact, there is an open neighbourhood Un of pn(K) such that cl(Un) is compact. Define the compact set ∞

L = cl(Un). nY=1 Since E is compactivorous, there must be a positive integer m, h1,h2 ∈ G and an open set V ⊆ G of the form V = V1 ×···× Vm × Gm+1 × · · · , where Vn ⊆ Gn is open for all n ∈ {1,...,m}, such that L ∩ V 6= ∅ and h1(L ∩ V )h2 ⊆ E. In particular, Un ∩ Vn 6= ∅ for all n ∈ {1,...,m}. For every n ∈ {1,...,m}, choose −1 gn ∈ Un ∩ Vn and observe that g (Un ∩ Vn) is an open neighbourhood of en. ∞ n Moreover, {pn(Bk)}k=1 is a nested neighbourhood basis of en for every n ≥ 1, −1 hence for each n ∈{1,...,m} there exists k0(n) such that pn(Bk) ⊆ gn (Un ∩ Vn) −1 for all k ≥ k0(n), which implies that pn(Kk) ⊆ gn (Un ∩Vn) for all k ≥ k0(n). Let j −1 be a sufficiently large index such that pn(Kj ) ⊆ gn (Un ∩Vn) for all n ∈{1,...,m} and set g = (g1,...,gm,em+1,... ). Then ∞ m ∞ gK ⊆ g p (K ) = g p (K ) × p (K ) ⊆ j  n j  n n j n j nY=1 nY=1 n=Ym+1 m ∞ ⊆ (Un ∩ Vn) × cl(Un) ⊆ L ∩ V. nY=1 n=Ym+1

However, this is a contradiction, as it implies that h1gKjh2 ⊆ E, against the assumption on Kj.  Corollary 5.3. The Fréchet space Rω and the Zω are fattening. 8 DAVIDE RAVASINI

6. Open questions The following questions arose naturally during the writing of this paper. (1) What other Hausdorff topological groups are fattening? What can be said about the weak topologies of Banach spaces without the Schur property or about Polish groups in general? (2) Esterle, Matheron and Moreau were the first to ask in [7] whether every closed, convex set which is not Haar null in a separable Banach space must be compactivorous. We observe that it is enough to ask the same question in ℓ1. Indeed, let Y be a closed subspace of ℓ1, let C ⊂ ℓ1/Y be a closed, convex subset and let π : ℓ1 → ℓ1/Y be the quotient map. Provided that the answer is affirmative, one obtains the following chain of implications: C is not Haar null ⇒ π−1(C) is not Haar null ⇒ π−1(C) is compactivorous ⇒ C is compactivorous. Using invariance under isomorphisms, the result would then become true in all separable Banach spaces. Acknowledgements The author would like to thank professors Eva Kopecká and Christian Bargetz for their precious suggestions. The work of the author is supported by the Austrian Science Fund (FWF): P 32523-N. References [1] C.D. Aliprantis and K.C. Border. Infinite Dimensional Analysis. A Hitch- hiker’s Guide. 3rd ed. Springer, 2006. [2] Y. Benyamini and J. Lindenstrauss. Geometric Nonlinear Functional Analysis, Volume 1. Vol. 48. American Mathematical Society, Colloquium Pubblications, 2000. [3] V.I. Bogachev. “Negligible Sets in Infinite-Dimensional Spaces”. Analysis Math- ematica 44 (2018), pp. 299–323. doi: https://doi.org/10.1007/s10476-018-0503-7. [4] J.P.R. Christensen. “On sets of Haar measure zero in Abelian Polish groups”. Israel Journal of Mathematics 13.3–4 (1972), pp. 255–260. [5] Udayan B. Darji. “On Haar meager sets”. Topology and its Applications 160.18 (Dec. 2013), pp. 2396–2400. doi: https://doi.org/10.1016/j.topol.2013.07.034. [6] Márton Elekes and Donát Nagy. “Haar null and Haar meager sets: a survey and new results”. Bulletin of the London Mathematical Society 52.4 (2020), pp. 561–619. doi: https://doi.org/10.1112/blms.12340. [7] J. Esterle, É. Matheron, and P. Moreau. “Haar negligibility of positive cones in Banach spaces”. St. Petersburg Mathematical Journal 27.5 (2016), pp. 731– 756. doi: https://doi.org/10.1090/spmj/1414. [8] Ernest Michael. “Continuous Selections I”. Annals of Mathematics 63.2 (Mar. 1956), pp. 361–382. doi: https://doi.org/10.2307/1969615.

Institut für Mathematik, Universität Innsbruck Technikerstraße 13, 6020 Innsbruck, Austria Email address: [email protected]