<<

arXiv:2106.02924v2 [math.GR] 2 Jul 2021 (1) npriua,when particular, In and t 2 ] suppose 3]: [2, ity hoe . eoesteCuh–aepr nqaiy steon the as inequality, of Cauchy–Davenport the recovers 1.1 Theorem B11) yteUiest elwhpfo IC n yteTrijitz the by and UIUC, from Fellowship University the by RB21011), where oua a.Suppose map. modular ν hoe 1.1. Theorem groups: compact locally all to facoe set closed a of G When udmna euti diiecmiaoisi h Cauchy–Daven the is combinatorics additive in result fundamental A Jwsspotdb rodO eka eerhAad(ICCam (UIUC Award Research Beckman O. Arnold by supported was YJ 2020 h anrsl fti ae sagnrlzto fteCauchy–Daven the of generalization a is paper this of result main The Z = min /p H µ Z H ahmtc ujc Classification. Subject Abstract. agsoe rprcmatsbrusof compact proper over ranges G − ( a ieone. size has 1 G agsoe rprcmatsbrusof subgroups compact proper over ranges  h orsodn ih armaueon Haar right corresponding the sacci ru forder of cyclic a is ν ν G G µ ( ACYDVNOTTERMFOR THEOREM CAUCHY–DAVENPORT A E XY ( G X Let With . ( egnrlz h acyDvnottermt oal opc gr compact locally to theorem Cauchy–Davenport the generalize We XY ) ,Y X, ) G G + ) OAL OPC GROUPS COMPACT LOCALLY sunimodular, is ,Y X, ≥ µ ealclycmatgroup, compact locally a be IA IGADCIUMN TRAN CHIEU-MINH AND JING YIFAN α r oepysbesof subsets nonempty are µ µ G | G X G min ( inf = ( XY ( H Y + r oepycmatsbesof subsets compact nonempty are { ) ) Y µ ) x ≤   G 1. ∈ ≥ | X ( min X Introduction ∆ 1 min + ) G − { rmr 20;Scnay1B0 05D10. 11B30, Secondary 22D05; Primary p ( β x n take and , {| αν µ − ) 1 1 G X µ and G ( G | Y ( sup + X ( ) E β + ) − | H ) Y G αν , Z sup = sup − | µ e ∆ ker H /p β with µ G − µ G G ( Z G 1 µ 1 H G ( µ p , G E y h onigmaueon measure counting the and , o oeprime some for G ) ∈ µ G etHa esr on measure Haar left a ( ) } Y H G ( } . with Y ( . ∆ ) H ) µ , nk Fellowship. insky G  ∆ ) G ( G y ≤ G , ( ) and µ G ypoe proper ly ehave we , µ : µ G u eerhBoard Research pus ) G G G ( } XY ( ( , XY G E → ottheorem port otinequal- port p ) ) . ) then , R ) sasubset a is oups. > ≤ 0 sthe is 1 , G G , , A CAUCHY–DAVENPORT THEOREM FOR LOCALLY COMPACT GROUPS 2

The Cauchy–Davenport theorem and its generalizations reflect the expansion (or growth) phenomenon in locally compact groups; for example, when G = Z/pZ, X = Y , and |X| < p/2, it implies |X + X|/|X| ≥ 3/2. Prior to our work, Kneser obtained a generalization of the Cauchy–Davenport the- orem for locally compact [11], and Kemperman did so for discrete groups [9] and, more generally, for unimodular locally compact groups [10]; inequal- ity (1) is a restatement of the second result by Kemperman. The naive generalization involving only a left does not work for nonunimodular groups. Indeed, if G is connected and nonunimodular, one can easily construct nonempty compact X,Y ⊆ G with

µG(XY ) <µG(X). However, Kemperman observed the following intriguing statement involving both the left and right Haar measure in [10]: If G is connected, X,Y ⊆ G are nonempty and compact, then ν (X) µ (Y ) µ (G) (2) min G + G , G ≤ 1. (νG(XY ) µG(XY ) µG(XY )) Surprisingly, inequalities of this form is necessary for the purpose generalizing the Brunn–Minkowski inequality to an arbitrary locally , even if one only cares about unimodular groups; see [8] for details. One might then ask whether there is a common generalization of (1) and (2) reflecting the expansion phenomenon in an arbitrary locally compact groups. Our main result of the paper is a response to this question. The proof of Theorem 1.1 also yields some conclusion about the structure of the set XY . This can be seen as a partial result towards a generalization of the Kneser theorem to locally compact groups. See Example 4.1 and Proposition 4.2 for the relevant discussion. The paper is organized as follows. In Section 2, we will present some basic back- ground on locally compact groups. In Section 3, we prove Theorem 1.1. Many of the ideas we used can be trace back to the work Kemperman [9, 10]. In Section 4, we present our structural result and post some open problems.

2. Preliminaries on locally compact groups

Suppose µG is a left Haar measure on G. Then:

(1) (left-translation-invariant) µG(X)= µG(aX) for all a ∈ G and all measurable sets X ⊆ G. A CAUCHY–DAVENPORT THEOREM FOR LOCALLY COMPACT GROUPS 3

(2) (inner and outer regular) µG(X) = sup µG(K) = inf µG(U) with K ranging over compact subsets of X and U ranging over open subsets of G containing X. (3) (compactly finite) µG takes finite measure on compact subsets of G. (4) (measurability characterization) If there is an increasing sequence (Kn) of compact subsets of X, and a decreasing sequence (Un) of open subsets of G with X ⊆ Un for all n such that limn→∞ µG(Kn) = limn→∞ µG(Un), then X is measurable.

The notion of a right Haar measure νG is obtained by replacing (1) by right-translation- −1 invariant. Suppose µG is a left Haar measure on G. Let νG = µG , that is for every −1 Borel set X, νG(X)= µG(X ). It is easy to see that νG is a right Haar measure. The following classical result by Haar makes the above notions enduring features of locally compact group:

Fact 2.1. Let G be a locally compact group. Up to multiplication by a positive constant, there is a unique left Haar measure and of G. A similar statement holds for right Haar measure.

Given a locally compact group G, and µG is a left Haar measure on G. For every x ∈ G, recall that

>0 ∆G : G → R x x 7→ µG/µG

x is the modular function of G, where µG is a left Haar measure on G defined by x µG(X)= µG(Xx), for every measurable set X. When the image of ∆G is always 1, we say G is unimodular, which also means that a left Haar measure is also a right >0 >0 Haar measure. In general, ∆G(x) takes values in R , where R is the multiplicative group of positive together with the usual Euclidean topology. The next fact records some basic properties of the modular function.

Fact 2.2. Let G be a locally compact group with a left Haar measure µG and a right Haar measure νG.

(1) Suppose H is a normal closed subgroup of G, then ∆H = ∆G. In particular, if H = ker ∆G, then H is unimodular. >0 (2) The function ∆G : G → R is a continuous homomorphism. (3) For every x ∈ G and every measurable set X, we have µG(Xx) = ∆G(x)µG(X), −1 and νG(xX) = ∆G (x)νG(X). (4) There is a constant c such that G f dµG = c G f∆G dνG for every f ∈ Cc(G). −1 If νG = µG , then c =1. ´ ´ A CAUCHY–DAVENPORT THEOREM FOR LOCALLY COMPACT GROUPS 4

3. Proof of Theorem 1.1 In this section, we prove our main theorem. Proof of Theorem 1.1. Let ν (X) µ (Y ) (3) G + G =1+ ρ. νG(XY ) µG(XY ) We may assume ρ> 0, as when ρ ≤ 0 the conclusion is immediate. Now recall that the modular function ∆G is continuous, so there is x0 ∈ X and y0 ∈ Y such that

∆G(x0) = min ∆G(x) and ∆G(y0) = max ∆G(y) x∈X y∈Y ∗ −1 ∗ −1 Set X = x0 X and Y = Yy0 . Let G≥1 = {x ∈ G : ∆G(x) ≥ 1}, and G≤1 = {x ∈ ∗ ∗ ∗ ∗ G : ∆G(x) ≤ 1}. Then X ⊆ G≥1, Y ⊆ G≤1, idG ∈ X ∩ Y . By the continuity of ∆G, both G≥1 and G≤1 are closed. We will focus on the case when G is noncompact, as otherwise the group is uni- modular, and all the later arguments still work for this case by simply replacing (3) by µG(XY )= µG(X)+ µG(Y ) − ρ/µG(XY ) <µG(G). ′ ′ ′ Let Ω be the collection of pair of sets (X ,Y ) satisfies that X is νG-measurable, ′ Y is µG-measurable, ′ ∗ ∗ ′ ∗ ∗ νG(X \ (X Y ∩ G≥1))=0, µG(Y \ (X Y ∩ G≤1))=0, and ′ ′ ∗ ∗ (νG × µG){(x, y): x ∈ X ,y ∈ Y ,xy∈ / X Y } =0. The following claim tells us one can choose a pair from Ω with the largest possible measures. ′ ′ Claim 1. There is (X0,Y0) ∈ Ω, such that for every other (X ,Y ) ∈ Ω, either ′ ′ νG(X )+ µG(Y ) < νG(X0)+ µG(Y0), ′ ′ ′ or νG(X )+ µG(Y )= νG(X0)+ µG(Y0) and νG(X ) ≤ νG(X0). Proof of Claim 1. For any (X′,Y ′) ∈ Ω, we have

′ ′ ∗ ∗

½ ½ ½X (x) Y (y)(1 − X Y (xy)) dνG(x) dµG(y)=0. ¨G×G

Let Γν be the real linear vector space of all real valued νG-measurable functions on 1 G with a finite L -norm, and let Γµ be the real linear vector space of all real valued 1 µG-measurable functions on G with a finite L -norm. We equip Γν and Γµ with their weak topology; see [1]. Now we define Kν,Kµ the subsets of Γν , Γµ such that ∗ ∗ Kν = {f ∈ Γν : f(x) ∈ [0, 1] for all x, and f(x) = 0 when x∈ / X Y ∩ G≥1}, A CAUCHY–DAVENPORT THEOREM FOR LOCALLY COMPACT GROUPS 5 and ∗ ∗ Kµ = {f ∈ Γµ : f(x) ∈ [0, 1] for all x, and f(x) = 0 when x∈ / X Y ∩ G≤1}.

It is easy to check that Kν and Kµ are closed and sequentially compact subsets (with respect to the weak topology) of Γν and Γµ respectively. Now we consider the space Γν ×Γµ equipped with the product topology. Then Kν × ∗ ∗ Kµ is a closed and sequentially compact subset of Γν ×Γµ. As X Y is measurable, for each ε, there are finitely many bounded continuous functions φk, ψk for 1 ≤ k ≤ N, such that N

∗ ∗ (1 − ½X Y (xy)) − φk(x)ψk(y) dνG(x) dµG(y) < ε. ¨G×G i=1 X On the other hand, since

N (f,g) 7→ f(x)g(y)φ (x)ψ (y) dν (x) dµ (y) ¨ k k G G k G×G X=1 is a continuous function on Γν × Γµ, the function

∗ ∗ (4) Ψ(f,g)= f(x)g(y)(1 − ½X Y (xy)) dνG(x) dµG(y) ¨G×G is continuous on Kν × Kµ. Let Λ ⊆ Kν × Kµ be the collections of (f,g) such that Ψ(f,g) is 0. It is again easy

∗ ∗ ½ to check that Λ is closed and sequentially compact. Moreover, as (½X , Y ) ∈ Λ, we have Λ =6 ∅. Finally, let us consider the function

Φ(f,g)= f(x) dνG(x)+ g(x) dµG(x). ˆG ˆG

This is a continuous function on Γν × Γµ. As Λ is sequentially compact, there is a nonempty subset Λ′ ⊆ Λ such that for every (f ′,g′) ∈ Λ′, Φ(f ′,g′) = sup Φ(f,g). (f,g)∈Λ ′ As Λ is nonempty, closed, and sequentially compact, there is (f0,g0) ∈ Λ such that f dνG is maximized. ´ Let X0 = supp(f0) and Y0 = supp(g0). Clearly (X0,Y0) ∈ Ω, and for every (X′,Y ′) ∈ Ω we have

′ ′

′ ′ ½ νG(X )+ µG(Y )= ½X (x) dνG(x)+ Y (x) dµG(x) ˆG ˆG A CAUCHY–DAVENPORT THEOREM FOR LOCALLY COMPACT GROUPS 6

≤ f0(x) dνG(x)+ g0(x) dµG(x) ≤ νG(X0)+ µG(Y0). ˆG ˆG When the equality holds in the above inequality, we have

′ ′ νG(X )= ½X (x) dνG(x) ≤ f0(x) dνG(x) ≤ νG(X0), ˆG ˆG this proves the claim. 1

Now we fix such a pair (X0,Y0). It is worth noting that the structures of X0 and Y0 might be very different from the structures of the original sets X and Y , so the later proof on the structures of X0 and Y0 provides no information on X and Y . However, since X0Y0 ⊆ XY , understanding X0 and Y0 will help us to understand the structure of XY .

Let νX0 be the measure restrict to X0, that is νX0 (Z) = νG(X0 ∩ Z) when Z is measurable. Let supp(νX0 ) be the support of the measure νX0 , that is a set of elements x in G such that each open neighborhood U of x satisfies that νX0 (U) > 0, equivalently νG(X0 ∩ U) > 0. We similarly define µY0 and supp(µY0 ). Clearly

νG(X0) ≤ νG(supp(νX0 )) and µG(Y0) ≤ µG(supp(µY0 )). ∗ ∗ As X Y is closed, one can check that (supp(νX0 ), supp(µY0 )) ∈ Ω. Hence by Claim 1,

νG(X0) ≥ νG(supp(νX0 )), and µG(Y0) ≥ µG(supp(µY0 )). Therefore,

νG(X0)= νG(supp(νX0 )) and µG(Y0)= µG(supp(µY0 )).

By replacing X0, Y0 if necessary, we may assume that

(5) X0 = supp(νX0 ) and Y0 = supp(µY0 ). ∗ ∗ In particular, X0,Y0 are closed subsets of X Y , and hence compact. As X0 ⊆ G≥1

and Y0 ⊆ G≤1, we have ½ (6) µG(X0)= ½X0 (x) dµG = ∆G(x) X0 (x) dνG ≥ νG(X0), ˆG ˆG and

−1 ½ (7) νG(Y0)= ½Y0 (x) dνG = ∆G (x) Y0 (x) dµG ≥ µG(Y0). ˆG ˆG

Let H = X0 ∩ Y0, then H is compact, and belongs to ker ∆G.

Claim 2. X0H = X0 and HY0 = Y0. Proof of Claim 2. Observe that, for every (X′,Y ′) ∈ Ω, for every g ∈ X′ ∩ Y ′, we have the following property: (X′ ∪ X′g,Y ′ ∩ g−1Y ′) ∈ Ω, and (X′ ∩ X′g−1,Y ′ ∪ gY ′) ∈ Ω. A CAUCHY–DAVENPORT THEOREM FOR LOCALLY COMPACT GROUPS 7

This is because (X′ ∪ X′g)(Y ′ ∩ g−1Y ′) ⊆ X′Y ′ ⊆ X∗Y ∗, ′ ′ −1 ′ ′ and ∆G(g) = 1. Likewise for (X ∩ X g ,Y ∪ gY ). Now we fix h ∈ H, and consider pairs of sets −1 −1 (X0 ∪ X0h, Y0 ∩ h Y0), and (X0 ∩ X0h ,Y0 ∪ hY0). As both of the pairs are in Ω, by Claim 1, we have either −1 −1 νG(X0 ∩ X0h ) <µG(Y0 ∩ h Y0), or −1 −1 −1 νG(X0 ∩ X0h )= µG(Y0 ∩ h Y0) and νG(X0 ∩ X0h )=0. Also by Claim 1, either −1 −1 νG(X0 ∩ X0h ) >µG(Y0 ∩ h Y0), or −1 −1 −1 νG(X0 ∩ X0h )= µG(Y0 ∩ h Y0) and µG(Y0 ∩ h Y0)=0. −1 −1 Hence, the only possibility is νG(X0 ∩ X0h )= µG(Y0 ∩ h Y0) = 0. −1 −1 It remains to show that both X0 ∩ X0h and Y0 ∩ h Y0 are empty. Suppose −1 x ∈ X0 ∩ X0h , and hence xh∈ / X0. As X0 is compact, one can find an open ∅ neighborhood U of x such that Uh ∩ X0 = . This implies x0x∈ / supp(νX0 ), −1 1 contradictse the fact that X0 =e supp(νX0 ). Likewise, Y0 ∩ h Y0 is empty. Using Claim 2, wee are going to show that H is in fact a compacte group. Claim 3. H is a compact group.

Proof of Claim 3. It suffices to show that H is a group. By Claim 2, as H = X0 ∩ Y0, for every h1, h2 ∈ H, we have h1h2 ∈ X0h2 = X0, and similarly h1h2 ∈ h1Y0 = Y0. Hence h1h2 ∈ X0 ∩ Y0 = H. Let h be in H. It is easy to see that hH ⊆ H is compact and closed under multiplication. Consider the collection C of all nonempty compact subsets of hH which is closed under multiplication. Ordering C by inclusion, then every chain in C has a lower bound in C by compactness. By Zorn’s lemma, hH contains a minimal nonempty compact subset H′ which is closed under multiplication. Then, for every h′ ∈ H′, h′H′ = H′h′ = H′. Thus H′ is a group. Since H′ ⊆ hH, this implies that −1 idG ∈ hH, hence h ∈ H. 1 The next claim shows that H is large. Claim 4. Let ∗ ∗ ∗ ∗ ∗ ∗ (νG(X )+ µG(Y ))νG(X Y )µG(X Y ) κ = ∗ ∗ ∗ ∗ ∗ ∗ . νG(X )µG(X Y )+ µG(Y )νG(X Y ) Then µG(H) ≥ ρκ. A CAUCHY–DAVENPORT THEOREM FOR LOCALLY COMPACT GROUPS 8

Proof of Claim 4. Since X0 ∪Y0 ⊆ X0Y0, by the inclusion–exclusion principle, as well as (6) and (7),

(8) µG(H) ≥ µG(X0)+ µG(Y0) − µG(X0Y0) ≥ νG(X0)+ µG(Y0) − µG(X0Y0), and

(9) νG(H) ≥ νG(X0)+ νG(Y0) − νG(X0Y0) ≥ νG(X0)+ µG(Y0) − νG(X0Y0). −1 Since H is a group, by the choice of νG, we have µG(H)= νG(H )= νG(H). As ∗ ∗ 1 νG(X ) 1 µG(Y ) 1 = ∗ ∗ ∗ ∗ + ∗ ∗ ∗ ∗ , κ νG(X )+ µG(Y ) νG(X Y ) νG(X )+ µG(Y ) µG(X Y ) ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ we have min{νG(X Y ),µG(X Y )} ≤ κ ≤ max{νG(X Y ),µG(X Y )}. This in particularly implies that

κ ≥ min{νG(X0Y0),µG(X0Y0)}. Therefore by Claim 1 and (8), (9), µ (H) ν (X ) µ (Y ) ν (X∗) µ (Y ∗) G ≥ G 0 + G 0 − 1 ≥ G + G − 1 κ κ κ κ κ ∗ ∗ νG(X ) µG(Y ) = ∗ ∗ + ∗ ∗ − 1= ρ, νG(X Y ) µG(X Y ) and this proves the claim. 1 ∗ ∗ −1 −1 −1 −1 −1 Since H = X0 ∩ Y0 ⊆ X ∪ Y ⊆ x0 XYy0 , we have y0 Hy0 ⊆ y0 x0 XY , and −1 −1 −1 x0Hx0 ⊆ XYy0 x0 . Thus −1 µG(H) ≤ min{∆G(y0) µG(XY ), ∆G(x0)νG(XY )}, which finishes the proof. 

4. Concluding Remarks In [11], Kneser proved a stronger result than the abelian version of Theorem 1.1: If G is a locally compact abelian group equipped with a Haar measure µG, and X,Y are nonempty compact subsets of G, then there is an open subgroup H such that

(i) µG(XY ) ≥ µG(X)+ µG(Y ) − µG(H). (ii) XY = XYH. It was shown by Olson [13] that the statement (ii) in the Kneser theorem cannot be extended to nonabelian groups, even if we replace it by a weaker condition that either XY = XYH, or XY = XHY , or XY = HXY . In a recent breakthrough, DeVos [4] characterized finite sets X,Y in a possibly nonabelian group with |XY | < |X| + |Y |. As a corollary of his main result, he A CAUCHY–DAVENPORT THEOREM FOR LOCALLY COMPACT GROUPS 9 obtained a generalization of the Kneser theorem to discrete groups, with a weakening version of statement (ii): for every g ∈ XY, there is z ∈ G such that g(zHz−1) ⊆ XY. It would be very interesting if such a result can be obtained for general locally compact groups. However, the following example given by Kemperman suggested the problem will be difficult:

Example 4.1. Let H ≤ ker ∆G be a compact group. Let X be an arbitrary compact subset of H, and Y = H∪W x where W is an arbitrary compact set with H∩W x = ∅. Assume ∆G(x) is sufficiently small, then we always have ν (X) µ (Y ) ν (X) µ (H)+ µ (W )∆ (x) G + G = G + G G G > 1. νG(XY ) µG(XY ) νG(H)+ νG(XW ) µG(H)+ µG(XW )∆G(x) As X, W are chosen arbitrarily, the structure of XW is hard to control. 1 On the other hand, from our proof of Theorem 1.1, we have a structural control on the “majority” of XY :

Proposition 4.2. Let G be a locally compact group, µG a left Haar measure on −1 R>0 G, νG = µG the corresponding right Haar measure on G, and ∆G : G → is the modular map. Suppose X,Y are nonempty compact subsets of G, and set α = infx∈X ∆G(x), β = supy∈Y ∆G(y). Then

(i) there is an open subgroup H of ker ∆G such that ν (X) µ (Y ) µ (H) µ (G) min G + G 1 − G , G ≤ 1, ν (XY ) µ (XY ) αν (X)+ β−1µ (Y ) µ (XY ) (  G G   G G  G ) (ii) there is D ⊆ XY with −1 min{β µG(D), ανG(D)} ν (X) µ (Y ) (αν (X)+ β−1µ (Y ))ν (XY )µ (XY ) ≤ µ (H) − G + G − 1 G G G G G ν (XY ) µ (XY ) ν (X)µ (XY )+ µ (Y )ν (XY )  G G  G G G G such that the following hold: for every g ∈ XY \ D, there exists z ∈ G such that g(zHz−1) ⊆ XY with the above H. ∗ ∗ Proof. Let X ,Y ,X0,Y0,H be the sets defined in the proof of Theorem 1.1. Set D = XY \ x0X0Y0y0. Let a ∈ x0X0, b ∈ Y0y0, and g = ab. Note that we have −1 −1 H ⊆ X0Y0 ⊆ x0 XYy0 , where x0 ∈ X0 and y0 ∈ Y0. Then −1 −1 −1 −1 −1 ∗ ∗ −1 −1 g XY = b a XY = b a x0X Y y0 ⊇ b a x0X0Y0y0. −1 −1 −1 −1 −1 −1 By Claim 2, Hby0 ⊆ Y0. As idG ∈ a x0X0, b ∈ b a x0X0. Thus b Hb ⊆ −1 −1 −1 b a x0X0Y0y0 ⊆ g XY . A CAUCHY–DAVENPORT THEOREM FOR LOCALLY COMPACT GROUPS 10

It remains to show that D has “small measure”. Without loss of generality, we ∗ ∗ ∗ ∗ ∗ ∗ assume µG(X Y ) = min{µG(X Y ), νG(X Y )}. Let κ be as in Claim 4. Then ∗ ∗ µG(X Y ) − κ ≤ 0. By (8) and (9), we have

µ (X∗Y ∗) − µ (x−1Dy−1) ν (X∗) µ (Y ∗) µ (H) G G 0 0 ≥ G + G − G κ κ κ κ ∗ ∗ νG(X ) µG(Y ) µG(H) ≥ ∗ ∗ + ∗ ∗ − . νG(X Y ) µG(X Y ) κ This implies

−1 −1 −1 ∗ ∗ β µG(D)= µG(x0 Dy0 ) ≤ µG(H) − ρκ + µG(X Y ) − κ = µG(H) − ρκ with ρ defined in (3). This proves statement (ii). 

In Example 4.1, while µG(XW x) is small, νG(XW x)= νG(XW ) can be very large. It suggests that one of the µG(D) and νG(D) can be large, hence to get an upper bound on min{µG(D), νG(D)} (as in Proposition 4.2) might be the best thing we can get. Nevertheless, we believe the bound in Proposition 4.2 is not sharp. In fact, we conjecture that when G is unimodular, D = ∅. Another direction is to consider the inverse problems. The Vosper theorem [15] and the Freiman 2.4k theorem [5] characterize subsets of Z/pZ where the equality in the Cauchy–Davenport theorem happens or nearly happens. For unimodular groups, the corresponding questions of characterizing

µG(XY )= µG(X)+ µG(Y ), or µG(XY ) ≤ µG(X)+ µG(Y )+ δ, were asked by Griesmer [6], by Kemperman [10], and by Tao [14]. The answer for connected locally compact groups were recently obtained by the authors [7]. One can also ask the similar questions when G contains subgroups of finite positive measure, and the equality in Theorem 1.1 nearly happens: suppose G is unimodular, H ≤ G, when will we have

µG(XY ) ≤ µG(X)+ µG(Y ) − µG(H)+ δ?

If G is not unimodular, when will we have ν (X) µ (Y ) µ (H) G + G − G ≥ 1 − δ? νG(XY ) µG(XY ) min{µG(XY ), νG(XY )}

For nonunimodular G, the special case when µG(XY ) = µG(Y ) is characterized by Macbeath [12]. A CAUCHY–DAVENPORT THEOREM FOR LOCALLY COMPACT GROUPS 11

References 1. V. I. Bogachev and O. G. Smolyanov, Topological vector spaces and their applications, Springer Monographs in Mathematics, Springer, Cham, 2017. MR 3616849 2. Augustin Louis Cauchy, Recherches sur les nombres,J. Ecole´ Polytech 9 (1813), 99–116. 3. Harold Davenport, On the addition of residue classes, J. London Math. Soc. 10 (1935), 30–32. 4. Matt DeVos, The structure of critical product sets, arXiv:1301.0096 (2013). 5. G. A. Fre˘ıman, Foundations of a structural theory of set addition, American Mathematical Society, Providence, R. I., 1973, Translated from the Russian, Translations of Mathematical Monographs, Vol 37. MR 0360496 6. John T. Griesmer, Semicontinuity of structure for small sumsets in compact abelian groups, Discrete Anal. (2019), Paper No. 18, 46. MR 4042161 7. Yifan Jing and Chieu-Minh Tran, Minimal and nearly minimal measure expansions in connected unimodular groups, arXiv:2006.01824 (2021). 8. Yifan Jing, Chieu-Minh Tran, and Ruixiang Zhang, A nonabelian Brunn-Minkowski inequality, arXiv:2101.07782 (2021). 9. J. H. B. Kemperman, On complexes in a semigroup, Nederl. Akad. Wetensch. Proc. Ser. A. 59 (1956), 247–254. MR 0085263 10. , On products of sets in a locally compact group, Fund. Math. 56 (1964), 51–68. MR 202913 11. Martin Kneser, Summenmengen in lokalkompakten abelschen Gruppen, Math. Z. 66 (1956), 88–110. MR 81438 12. A. M. Macbeath, On the measure of product sets in a , J. London Math. Soc. 35 (1960), 403–407. MR 126501 13. John E. Olson, On the symmetric difference of two sets in a group, European J. Combin. 7 (1986), no. 1, 43–54. MR 850143 14. Terence Tao, An inverse theorem for an inequality of Kneser, Proc. Steklov Inst. Math. 303 (2018), no. 1, 193–219, Published in Russian in Tr. Mat. Inst. Steklova 303 (2018), 209–238. MR 3920221 15. A. G. Vosper, The critical pairs of subsets of a group of prime order, J. London Math. Soc. 31 (1956), 200–205. MR 77555

Department of Mathematics, University of Illinois at Urbana-Champaign, Urbana IL, USA Email address: [email protected] Department of Mathematics, University of Notre Dame, Notre Dame IN, USA Email address: [email protected]