Isometry groups of non-positively curved spaces: discrete subgroups

We study lattices in non-positively curved metric spaces. Borel density is established in that setting as well as a form of Mostow rigidity. A converse to the flat torus theorem is provided. Geometric arithmeticity results are obtained after a detour…

Authors: P.-E. Caprace, N. Monod

Isometry groups of non-positively curved spaces: discrete subgroups
ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS PIERRE-EMMANUEL CAPRA CE* AND NICOLAS MONOD ‡ Abstra t. W e study latties in non-p ositiv ely urv ed metri spaes. Borel densit y is established in that setting as w ell as a form of Mosto w rigidit y . A on v erse to the at torus theorem is pro vided. Geometri arithmetiit y results are obtained after a detour through sup errigidit y and arithmetiit y of abstrat latties. Residual niteness of latties is also studied. Riemannian symmetri spaes are  haraterised amongst CA T(0) spaes admitting latties in terms of the existene of parab oli isometries. 1. Intr odution Latties in semi-simple algebrai groups ha v e a tan talisingly ri h struture; they inlude arithmeti groups and more generally S-arithmeti groups o v er arbitrary  harateristis. The nature of these groups is shap ed in part b y the fat that they are realised as isometries of a anonial non-p ositiv ely urv ed spae: the asso iated Riemannian symmetri spae, or BruhatTits building, or a pro dut of b oth t yp es. Man y other groups of rather div erse origins share this prop ert y to o ur as latties in non-p ositiv ely urv ed spaes, singular or not:  The fundamen tal group of a losed Riemannian manifold of non-p ositiv e setionnal urv ature. Here the spae ated up on is the univ ersal o v ering, whi h is a Hadamard manifold.  Man y Gromo v-h yp erb oli groups admit a prop erly dison tin uous o ompat ation on some CA T( − 1 ) spae b y isometries. Amongst the examples arising in this w a y are h yp erb oli Co xeter groups [ Mou88 ℄, C ′ ( 1 6 ) and C ′ ( 1 4 ) - T (4) small anellation groups [ Wis04 ℄, 2 -dimensional 7 -systoli groups [ J06 ℄. It is in fat a w ell kno wn op en problem of M. Gromo v to onstrut an example of a Gromo v-h yp erb oli group whi h is not a CA T(0) group (see [ Gro93 , 7.B℄; also Remark 2.3(2) in Chapter I I I. Γ of [ BH99 ℄).  In [ BM00b ℄, striking examples of nitely presen ted simple groups are onstruted as latties in a pro dut of t w o lo ally nite trees. T ree latties w ere previously studied in [ BL01 ℄.  A minimal adjoin t KaMo o dy group o v er a nite eld, as dened b y J. Tits [ Tit87 ℄, is endo w ed with t w o B N -pairs whi h yield strongly transitiv e ations on a pair of t winned buildings. When the order of the ground eld is large enough, the Ka Mo o dy group is a lattie in the pro dut of these t w o buildings [ Rém99 ℄. Subsuming all the ab o v e examples, w e dene a CA T(0) lattie as a pair (Γ , X ) onsisting of a prop er CA T(0) spae X with o ompat isometry group Is( X ) and a lattie subgroup Key wor ds and phr ases. Lattie, arithmeti group, non-p ositiv e urv ature, CA T(0) spae, lo ally ompat group. *F.N.R.S. Resear h Asso iate. ‡ Supp orted in part b y the Swiss National Siene F oundation. 1 2 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD Γ < Is( X ) , i.e. a disrete subgroup of nite in v arian t o v olume (the ompat-op en top ology mak es Is( X ) a lo ally ompat seond oun table group whi h is th us anonially endo w ed with Haar measures). W e sa y that (Γ , X ) is uniform if Γ is o ompat in Is( X ) or, equiv a- len tly , if the quotien t Γ \ X is ompat; that ase orresp onds to Γ b eing a CA T(0) group in the usual terminology . Amongst CA T(0) latties, the most imp ortan t, and also the b est understo o d, notably through the w ork of G. Margulis, onsist undoubtedly of those arising from latties in semi- simple groups o v er lo al elds. It is therefore natural to address t w o sets of questions. (a) What pr op erties of these latti es ar e shar e d by al l CA T(0) latti es? (b) What pr op erties har aterise them within the lass of CA T(0) latti es? This artile is dev oted to the study of CA T(0) latties and en tres largely around the ab o v e questions, though w e also address the general question of the in terpla y b et w een the algebrai struture of a CA T(0) lattie and the geometri prop erties of the underlying spae. Some of the te hniques established in the presen t pap er ha v e b een used in a subsequen t in v estigation of latties in pro duts of KaMo o dy groups [ CM08d ℄. W e shall no w desrib e the main results of this artile; for man y of them, the ore of the text will on tain a stronger, more preise but p erhaps more p onderous v ersion. Our notation is standard, as realled in the Notation setion of the ompanion pap er [ CM08 ℄. W e refer to the latter for terminology and shall quote it freely . . Geometri Borel densit y . As a link b et w een the general theory exp osed in [ CM08 ℄ and the study of CA T(0) latties, w e prop ose the follo wing analogue of A. Borel's densit y theorem [ Bor60 ℄. Theorem 1.1. L et X b e a pr op er CA T(0) sp a e, G a lo  al ly  omp at gr oup ating  ontin- uously by isometries on X and Γ < G a latti e. Supp ose that X has no Eulide an fator. If G ats minimal ly without xe d p oint at innity, so do es Γ . This onlusion fails for spaes with a Eulidean fator. The theorem will b e established more generally for losed subgroups with nite in v arian t o v olume. It should b e ompared to (and an of ourse b e gainfully om bined with) a similar densit y prop ert y of normal subgroups established as Theorem 1.10 in [ CM08 ℄. Remark 1.2. Theorem 1.1 applies to general prop er CA T(0) spaes. It implies in partiular the lassial Borel densit y theorem (see the end of Setion 2 ). As with lassial Borel densit y , w e shall use this densit y statemen t to deriv e statemen ts ab out the en traliser, normaliser and radial of latties in Setion 2 . A more elemen tary v arian t of the ab o v e theorem sho ws that a large lass of groups ha v e rather restrited ations on prop er CA T(0) spaes; as an appliation, one sho ws: A ny isometri ation of R. Thompson 's gr oup F on any pr op er CA T(0) sp a e X has a xe d p oint in X , see Corollary 2.3 . Theorem 1.1 also pro vides additional information ab out the totally dis- onneted groups D j o urring in Theorem 1.6 in [ CM08 ℄. . ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 3 Latties: Eulidean fator, b oundary , irreduibilit y and Mosto w rigidit y . Reall that the Flat T orus the or em , originating in the w ork of GromollW olf [ GW71 ℄ and La wson Y au [ L Y72 ℄, asso iates Eulidean subspaes R n to an y subgroup Z n of a CA T(0) group, see [ BH99 ,  I I.7℄. (In the lassial setting, when the CA T(0) group is giv en b y a ompat non-p ositiv ely urv ed manifold, this amoun ts to the seemingly more symmetri statemen t that su h a subgroup exists if and only if there is a at torus is the manifold.) The on v erse is a w ell kno wn op en problem stated b y M. Gromo v in [ Gro93 ,  6 . B 3 ℄; for manifolds see S.-T. Y au, problem 65 in [ Y au82 ℄). P oin t ( i ) in the follo wing result is a (v ery partial) answ er; in the sp eial ase of o ompat Riemannian manifolds, this w as the main result of P . Eb erlein's artile [ Eb e83 ℄. Theorem 1.3. L et X b e a pr op er CA T(0) sp a e, G < Is( X ) a lose d sub gr oup ating minimal ly and  o  omp atly on X and Γ < G a nitely gener ate d latti e. Then: (i) If the Eulide an fator of X has dimension n , then Γ p ossesses a nite index sub- gr oup Γ 0 whih splits as Γ 0 ≃ Z n × Γ ′ . Mor e over, the dimension of the Eulide an fator is har aterise d as the maximal r ank of a fr e e A b elian normal sub gr oup of Γ . (ii) G has no xe d p oint at innity; the set of Γ -xe d p oints at innity is  ontaine d in the (p ossibly empty) b oundary of the Eulide an fator. P oin t ( ii ) is partiularly useful in onjution with the man y results assuming the absene of xed p oin ts at innit y in [ CM08 ℄. In addition, it is already a rst indiation that the mere existene of a (nitely generated) lattie is a serious restrition on a prop er CA T(0) spae ev en within the lass of o ompat minimal spaes. W e reall that E. Hein tze [ Hei74 ℄ pro dued simply onneted negativ ely urv ed Riemannian manifolds that are homogeneous (in partiular, o ompat) but ha v e a p oin t at innit y xed b y all isometries. Sine a CA T(0) lattie onsists of a group and a spae, there are t w o natural notions of irreduibilit y: of the group or of the spae. In the ase of latties in semi-simple groups, the t w o notions are kno wn to oinide b y a result of Margulis [ Mar91 , I I.6.7℄. W e pro v e that this is the ase for CA T(0) latties as ab o v e. Theorem 1.4. In the setting of The or em 1.3 , Γ is irr e duible as an abstr at gr oup if and only if for any nite index sub gr oup Γ 1 and any Γ 1 -e quivariant de  omp osition X = X 1 × X 2 with X i non- omp at, the pr oje tion of Γ 1 to b oth Is( X i ) is non-disr ete. The om bination of Theorem 1.4 , Theorem 1.3 and of an appropriate form of sup errigidit y allo w us to giv e a CA T(0) v ersion of Mosto w rigidit y for reduible spaes (Setion 4.E ). . Geometri arithmetiit y . W e no w exp ose results giving p erhaps unexp etedly strong onlusions for CA T(0) latties  b oth for the group and for the spae. These results w ere announed in [ CM08e ℄ in the ase of CA T(0) groups; the presen t setting of nitely generated latties is more general sine CA T(0) groups are nitely generated ( f. Lemma 3.3 b elo w). W e reall that an isometry g is parab oli if the translation length inf x ∈ X d ( g x, x ) is not a hiev ed. F or general CA T(0) spaes, parab oli isometries are not w ell understo o d; in fat, ruling out their existene an sometimes b e the essen tial diult y in rigidit y statemen ts. Theorem 1.5. L et (Γ , X ) b e an irr e duible nitely gener ate d CA T(0) latti e with X ge o desi-  al ly  omplete. Assume that X p ossesses some p ar ab oli isometry. 4 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD If Γ is r esidual ly nite, then X is a pr o dut of symmetri sp a es and BruhatTits build- ings. In p artiular, Γ is an arithmeti latti e unless X is a r e al or  omplex hyp erb oli sp a e. If Γ is not r esidual ly nite, then X stil l splits o a symmetri sp a e fator. Mor e over, the nite r esidual Γ D of Γ is innitely gener ate d and Γ / Γ D is an arithmeti gr oup. (Reall that the nite residual of a group is the in tersetion of all nite index subgroups.) W e single out a purely geometri onsequene. Corollary 1.6. L et (Γ , X ) b e a nitely gener ate d CA T(0) latti e with X ge o desi al ly  om- plete. Then X p ossesses a p ar ab oli isometry if and only if X ∼ = M × X ′ , wher e M is a symmetri sp a e of non- omp at typ e. Without the assumption of geo desi ompleteness, w e still obtain an arithmetiit y state- men t when the underlying spae admits some parab oli isometry that is neutral , i.e. whose displaemen t length v anishes. Neutral parab oli isometries are ev en less understo o d, not ev en for their dynamial prop erties (whi h an b e ompletely wild at least in Hilb ert spae [ Ede64 ℄); as for familiar examples, they are pro vided b y unip oten t elemen ts in semi- simple algebrai groups. Theorem 1.7. L et (Γ , X ) b e an irr e duible nitely gener ate d CA T(0) latti e. If X admits any neutr al p ar ab oli isometry, then either: (i) Is( X ) is a r ank one simple Lie gr oup with trivial  entr e; or: (ii) Γ has a normal sub gr oup Γ D suh that Γ / Γ D is an arithmeti gr oup. Mor e over, Γ D is either nite or innitely gener ate d. W e turn to another t yp e of statemen t of arithmetiit y/geometri sup errigidit y . Ha ving established an abstrat arithmetiit y theorem (presen ted b elo w as Theorem 1.9 ), w e an app eal to our geometri results and pro v e the follo wing. Theorem 1.8. L et (Γ , X ) b e an irr e duible nitely gener ate d CA T(0) latti e with X ge o desi-  al ly  omplete. Assume that Γ p ossesses some faithful nite-dimensional line ar r epr esenta- tion ( in har ateristi 6 = 2 , 3) . If X is r e duible, then Γ is an arithmeti latti e and X is a pr o dut of symmetri sp a es and BruhatTits buildings. Setion 6 on tains more results of this nature but also demonstrates b y a family of exam- ples that some of the in triaies in the more detailed statemen ts reet indeed the existene of more exoti pairs (Γ , X ) . . Abstrat arithmetiit y . When preparing for the pro of of our geometri arithmetiit y statemen ts, w e are led to study irreduible latties in pro duts of general top ologial groups in the abstrat. Building notably on ideas of Margulis, w e stablish the follo wing arithmeti- it y statemen t (for whi h w e reall that the quasi-en tre Q Z of a top ologial group is the subset of elemen ts with op en en traliser). Theorem 1.9. L et Γ < G = G 1 × · · · × G n b e an irr e duible nitely gener ate d latti e, wher e e ah G i is any lo  al ly  omp at gr oup. If Γ admits a faithful Zariski-dense r epr esentation in a semi-simple gr oup over some eld of har ateristi 6 = 2 , 3 , then the amenable r adi al R of G is  omp at and the quasi- entr e Q Z ( G ) is virtual ly  ontaine d in Γ · R . F urthermor e, up on r eplaing G by a nite index ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 5 sub gr oup, the quotient G/R splits as G + × Q Z ( G/R ) wher e G + is a semi-simple algebr ai gr oup and the image of Γ in G + is an arithmeti latti e. In partiular, the quasi-en tre Q Z ( G/R ) is disr ete . In shorter terms, this theorem states that up to a ompat extension, G is the diret pro dut of a semi-simple algebrai group b y a (p ossibly trivial) disrete group, and that the image of Γ in the non-disrete part is an arithmeti group. The assumption on the  harateristi an b e sligh tly w eak ened. In the ourse of the pro of, w e  haraterise all irreduible nitely generated latties in pro duts of the form G = S × D where S is a semi-simple Lie group and D a totally disonneted group (Theorem 5.18 ). In partiular, it turns our that D m ust neessarily b e lo ally pronite b y analyti. The orresp onding question for simple algebrai groups instead of Lie groups is also in v estigated (Theorem 5.20 ). . Unique geo desi extension. Complete simply onneted Riemannian manifolds of non- p ositiv e urv ature, sometimes also alled Hadamard manifolds, form a lassial family of prop er CA T(0) spaes to whi h the preeding results ma y b e applied. In fat, the natural lass to onsider in our on text onsists of those prop er CA T(0) spaes in whi h ev ery geo- desi segmen t extends uniquely to a bi-innite geo desi line. Clearly , this lass on tains all Hadamard manifolds, but it presumably on tains more examples. It is, ho w ev er, somewhat restrited with resp et to the main thrust of the presen t w ork sine it do es not allo w for, sa y , simpliial omplexes; aordingly , the onlusions of the theorem b elo w are also more stringen t. Theorem 1.10. L et X b e a pr op er CA T(0) sp a e with uniquely extensible ge o desis. Assume that Is( X ) ats  o  omp atly without xe d p oints at innity. (i) If X is irr e duible, then either X is a symmetri sp a e or Is( X ) is disr ete. (ii) If Is( X ) p ossesses a nitely gener ate d non-uniform latti e Γ whih is irr e duible as an abstr at gr oup, then X is a symmetri sp a e (without Eulide an fator). (iii) Supp ose that Is( X ) p ossesses a nitely gener ate d latti e Γ (if Γ is uniform, this is e quivalent to the  ondition that Γ is a disr ete  o  omp at gr oup of isometries of X ). If Γ is irr e duible (as an abstr at gr oup) and X is r e duible, then X is a symmetri sp a e (without Eulide an fator). In the sp eial ase of Hadamard manifolds, statemen t (i) w as kno wn under the assumption that Is( X ) satises the dualit y ondition (without assuming that Is( X ) ats o ompatly without xed p oin ts at innit y). This is due to P . Eb erlein (Prop osition 4.8 in [ Eb e82 ℄). Lik ewise, statemen t (iii) for manifolds is Prop osition 4.5 in [ Eb e82 ℄. More reen tly , F arbW ein b erger [ FW06 ℄ in v estigated analogous questions for aspherial manifolds. . Latties and the de Rham deomp osition. In [ CM08 ℄, w e pro v ed a de Rham deomp osition (1.i) X ′ ∼ = X 1 × · · · × X p × R n × Y 1 × · · · × Y q for prop er CA T(0) spaes X with nite-dimensional Tits b oundary and su h that Is( X ) has no xed p oin t at innit y , see A ddendum 1.8 in [ CM08 ℄. (Here X ′ ⊆ X is the anonial 6 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD minimal in v arian t subspae, and w e reall that X ′ = X e.g. when X is geo desially omplete and admits a o ompat lattie b y Lemma 3.13 in [ CM08 ℄.) It turns out that this de Rham deomp osition is an in v arian t of CA T(0) groups in the follo wing sense (see Corollary 4.14 ). Theorem 1.11. L et X b e a pr op er CA T(0) sp a e and Γ < Is( X ) b e a gr oup ating pr op erly dis ontinuously and  o  omp atly. Then any other suh sp a e admitting a pr op er  o  omp at Γ -ation has the same numb er of fators in ( 1.i ) and the Eulide an fator has same dimension. ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 7 Contents 1. In tro dution 1 2. An analogue of Borel densit y 8 2.A. Fixed p oin ts at innit y 8 2.B. Geometri densit y for subgroups of nite o v olume 9 2.C. The limit set of subgroups of nite o v olume 11 3. CA T(0) latties, I: the Eulidean fator 12 3.A. Preliminaries on latties 12 3.B. V ariations on Auslander's theorem 13 3.C. Latties, the Eulidean fator and xed p oin ts at innit y 14 4. CA T(0) latties, I I: pro duts 18 4.A. Irreduible latties in CA T(0) spaes 18 4.B. The h ull of a lattie 20 4.C. On the anonial disrete k ernel 21 4.D. Residually nite latties 22 4.E. Strong rigidit y for pro dut spaes 23 5. Arithmetiit y of abstrat latties 24 5.A. Sup errigid pairs 26 5.B. Boundary maps 29 5.C. Radial sup errigidit y 30 5.D. Latties with non-disrete ommensurators 31 5.E. Latties in pro duts of Lie and totally disonneted groups 32 5.F. Latties in general pro duts 34 6. Geometri arithmetiit y 37 6.A. CA T(0) latties and parab oli isometries 37 6.B. Arithmetiit y of linear CA T(0) latties 39 6.C. A family of examples 40 7. A few questions 42 Referenes 44 8 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD 2. An analogue of Borel density Before disussing our analogue of Borel's densit y theorem [ Bor60 ℄ in Setion 2.B b elo w, w e presen t a more elemen tary phenomenon based on o-amenabilit y . 2.A. Fixed p oin ts at innit y. Reall that a subgroup H of a top ologial group G is o-amenable if an y on tin uous ane G -ation on a on v ex ompat set (in a Hausdor lo ally on v ex top ologial v etor spae) has a xed p oin t whenev er it has an H -xed p oin t. The argumen ts of A damsBallmann [ AB98 ℄ imply the follo wing preliminary step to w ards Theorem 2.4 : Prop osition 2.1. L et G b e a top olo gi al gr oup with a  ontinuous isometri ation on a pr op er CA T(0) sp a e X without Eulide an fator. Assume that the G -ation is minimal and do es not have a glob al xe d p oint in ∂ X . Then any  o-amenable sub gr oup of G stil l has no glob al xe d p oint in ∂ X . Pr o of. Supp ose for a on tradition that a o-amenable subgroup H < G xes ξ ∈ ∂ X . Then G preserv es a probabilit y measure µ on ∂ X and w e obtain a on v ex funtion f : X → R b y in tegrating Busemann funtions against this measure; as in [ AB98 ℄, the o yle equation for Busemann funtions (see  2 in [ CM08 ℄) imply that f is G -in v arian t up to onstan ts. The argumen ts therein sho w that f is onstan t and that µ is supp orted on at p oin ts. Ho w ev er, in the absene of a Eulidean fator, the set of at p oin ts has a unique irumen tre when non-empt y [ AB98 , 1.7℄; this pro vides a G -xed p oin t, a on tradition.  Com bining the ab o v e with the splitting metho ds used in Theorem 4.3 in [ CM08 ℄, w e reord a onsequene sho wing that the exat onlusions of the A damsBallmann theo- rem [ AB98 ℄ hold under m u h w eak er assumptions than the amenabilit y of G . Corollary 2.2. L et G b e a top olo gi al gr oup with a  ontinuous isometri ation on a pr op er CA T(0) sp a e X . Assume that G  ontains two  ommuting  o-amenable sub gr oups. Then either G xes a p oint at innity or it pr eserves a Eulide an subsp a e in X . W e emphasise that one an easily onstrut a w ealth of examples of highly non-amenable groups satisfying these assumptions. F or instane, giv en any group Q , the restrited wreath pro dut G = Z ⋉ L n ∈ Z Q on tains the pair of omm uting o-amenable groups H + = L n ≥ 0 Q and H − = L n< 0 Q , see [ MP03 ℄. (In fat, one an ev en arrange for H ± to b e onjugated up on replaing Z b y the innite dihedral group.) F or similar reasons, w e dedue the follo wing xed-p oin t prop ert y for R. Thompson's group F :=  g i , i ∈ N | g − 1 i g j g i = g j +1 ∀ j > i  ; this xed-p oin t result explains wh y the strategy prop osed in [ F ar08 ℄ to dispro v e amenabilit y of F with the A damsBallmann theorem annot w ork. Corollary 2.3. A ny F -ation by isometries on any pr op er CA T(0) sp a e X has a xe d p oint in X . Pr o of of Cor ol lary 2.2 . W e assume that G has no xed p oin t at innit y . By Prop osition 4.1 in [ CM08 ℄, there is a minimal non-empt y losed on v ex G -in v arian t subspae. Up on on- sidering the Eulidean deomp osition [ BH99 , I I.6.15℄ of the latter, w e an assume that X is G -minimal and without Eulidean fator and need to sho w that G xes a p oin t in X . Let H ± < G b e the omm uting o-amenable groups. In view of Prop osition 2.1 , b oth at without xed p oin t at innit y . In partiular, w e ha v e an ation of H = H + × H − without xed p oin t at innit y and the splitting theorem from [ Mon06 ℄ pro vides us with a ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 9 anonial subspae X + × X − ⊆ X with omp onen t-wise and minimal H -ation. All of ∂ X + is xed b y H − , whi h means that this b oundary is empt y . Sine X is prop er, it follo ws that X + is b ounded and hene redued to a p oin t b y minimalit y . Th us H + xes a p oin t in X ⊆ X and o-amenabilit y implies that G xes a probabilit y measure µ on X . If µ w ere supp orted on ∂ X , the pro of of Prop osition 2.1 w ould pro vide a G -xed p oin t at innit y , whi h is absurd. Therefore µ ( X ) > 0 . No w  ho ose a b ounded set B ⊆ X large enough so that µ ( B ) > µ ( X ) / 2 . Then an y G -translate of B m ust meet B . It follo ws that G has a b ounded orbit and hene a xed p oin t as laimed.  Pr o of of Cor ol lary 2.3 . W e refer to [ CFP96 ℄ for a detailed in tro dution to the group F . In partiular, F an b e realised as the group of all orien tation-preserving pieewise ane homeomorphisms of the in terv al [0 , 1] that ha v e dy adi breakp oin ts and slop es 2 n with n ∈ Z . Giv en a subset A ⊆ [0 , 1] w e denote b y F A < F the subgroup supp orted on A . W e laim that whenev er A has non-empt y in terior, F A is o-amenable in F . The argumen t is analogous to [ MP03 ℄ and to [ GM07 ,  4.F℄; indeed, in view of the alternativ e denition of F just realled, one an  ho ose a sequene { g n } in F su h that g n A on tains [1 /n, 1 − 1 /n ] and th us F g n A on tains F [1 /n, 1 − 1 /n ] . Consider the ompat spae of means on F /F A , namely nitely additiv e measures, endo w ed with the w eak-* top ology from the dual of ℓ ∞ ( F /F A ) . An y aum ulation p oin t µ of the sequene of Dira masses at g − 1 n F A will b e in v arian t under the union F ′ of the groups F [1 /n, 1 − 1 /n ] . No w F ′ is the k ernel of the deriv ativ e homomorphism F → 2 Z × 2 Z at the pair of p oin ts { 0 , 1 } . In partiular, F ′ is o-amenable in F and th us the F ′ -in v ariane of µ implies that there is also a F -in v arian t mean on F /F A , whi h is one of the  haraterisations of o-amenabilit y [ Eym72 ℄. Let no w X b e an y prop er CA T(0) spae with an F -ation b y isometries. W e an assume that F has no xed p oin t at innit y and therefore w e an also assume that X is minimal b y Prop osition 4.1 in [ CM08 ℄. The ab o v e laim pro vides us with man y pairs of omm uting o-amenable subgroups up on taking disjoin t sets of non-empt y in terior. Therefore, Corol- lary 2.2 sho ws that X ∼ = R n for some n . In partiular the isometry group is linear. Sine F is nitely generated (b y g 0 and g 1 in the ab o v e presen tation, ompare also [ CFP96 ℄), Malev's theorem [ Mal40 ℄ implies that the image of F is residually nite. The deriv ed subgroup of F (whi h iniden tally oinides with the group F ′ in tro dued ab o v e) b eing simple [ CFP96 ℄, it follo ws that it ats trivially . It remains only to observ e that t w o omm uting isometries of R n alw a ys ha v e a ommon xed p oin t in R n , whi h is a matter of linear algebra.  The ab o v e reasoning an b e adapted to yield similar results for bran h groups and related groups; w e shall address these questions elsewhere. 2.B. Geometri densit y for subgroups of nite o v olume. The follo wing geometri densit y theorem generalises Borel's densit y (see Prop osition 2.8 b elo w) and on tains Theo- rem 1.1 from the In tro dution. Theorem 2.4. L et G b e a lo  al ly  omp at gr oup with a  ontinuous isometri ation on a pr op er CA T(0) sp a e X without Eulide an fator. If G ats minimal ly and without glob al xe d p oint in ∂ X , then any lose d sub gr oup with nite invariant  ovolume in G stil l has these pr op erties. Remark 2.5. F or a related statemen t without the assumption on the Eulidean fator of X or on xed p oin ts at innit y , see Theorem 3.14 b elo w. Pr o of. Retain the notation of the theorem and let Γ < G b e a losed subgroup of nite in v arian t o v olume. In partiular, Γ is o-amenable and th us has no xed p oin ts at innit y 10 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD b y Prop osition 2.1 . By Prop osition 4.1 in [ CM08 ℄, there is a minimal non-empt y losed on v ex Γ -in v arian t subset Y ⊆ X and it remains to sho w Y = X . Cho ose a p oin t x 0 ∈ X and dene f : X → R b y f ( x ) = Z G/ Γ  d ( x, g Y ) − d ( x 0 , g Y )  dg . This in tegral on v erges b eause the in tegrand is b ounded b y d ( x, x 0 ) . The funtion f is on tin uous, on v ex (b y [ BH99 , I I.2.5(1)℄) and quasi-in v arian t in the sense that it satises (2.i) f ( hx ) = f ( x ) − f ( hx 0 ) ∀ h ∈ G. Sine G ats minimally and without xed p oin t at innit y , this implies that f is onstan t (see Setion 2 in [ AB98 ℄; alternativ ely , when ∂ X is nite-dimensional, it follo ws from The- orem 1.10 in [ CM08 ℄ sine ( 2.i ) implies that f is in v arian t under the deriv ed subgroup G ′ ). In partiular, d ( x, g Y ) is ane for all g . It follo ws that for all x ∈ X the losed set Y x =  z ∈ X : d ( z , Y ) = d ( x, Y )  is on v ex. W e laim that it is parallel to Y in the sense that d ( z , Y ) = d ( y , Y x ) for all z ∈ Y x and all y ∈ Y . Indeed, on the one hand d ( z , Y ) is onstan t o v er z ∈ Y x b y denition, and on the other hand d ( y , Y x ) is onstan t b y minimalit y of Y sine d ( · , Y x ) is a on v ex Γ -in v arian t funtion. In partiular, Y x is Γ -equiv arian tly isometri to Y via nearest p oin t pro jetion (ompare [ BH99 , I I.2.12℄) and ea h Y x is Γ -minimal. A t this p oin t, Remarks 39 in [ Mon06 ℄ sho w that there is an isometri Γ -in v arian t splitting X ∼ = Y × T . It remains to sho w that the spae of omp onen ts T is redued to a p oin t. Let th us s, t ∈ T and let m b e their midp oin t. Applying the ab o v e reasoning to the  hoie of minimal set Y 0 orresp onding to Y × { m } , w e dedue again that the distane to Y 0 is an ane funtion on X . Ho w ev er, this funtion is preisely the distane funtion d ( · , m ) in T omp osed with the pro jetion X → T . Being non-negativ e and ane on [ s, t ] , it v anishes on that segmen t and hene s = t .  Remark 2.6. When Γ is o ompat in G , the pro of an b e shortened b y in tegrating just d ( x, g Y ) in the denition of f ab o v e. Corollary 2.7. L et X b e a pr op er CA T(0) sp a e without Eulide an fator suh that G = Is( X ) ats minimal ly without xe d p oint at innity, and let Γ < G b e a lose d sub gr oup with nite invariant  ovolume. Then: (i) Γ has trivial amenable r adi al. (ii) The  entr aliser Z G (Γ) is trivial. (iii) If Γ is nitely gener ate d, then is has nite index in its normaliser N G (Γ) and the latter is a nitely gener ate d latti e in G . Pr o of. (i) and (ii) follo w b y the same argumen t as in the pro of of Theorem 1.10 in [ CM08 ℄. F or (iii) w e follo w [ Mar91 , Lemma I I.6.3℄. Sine Γ is losed and oun table, it is disrete b y Baire's ategory theorem and th us is a lattie in G . Sine it is nitely generated, its automorphism group is oun table. By (ii), the normaliser N G (Γ) maps injetiv ely to Aut(Γ) and hene is oun table as w ell. Th us N G (Γ) , b eing losed in G , is disrete b y applying Baire again. Sine it on tains the lattie Γ , it is itself a lattie and the index of Γ in N G (Γ) is nite. Th us N G (Γ) is nitely generated.  ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 11 As p oin ted out b y P . de la Harp e, p oin t( ii ) implies in partiular that an y lattie in G is ICC (whi h means b y denition that all its non-trivial onjugay lasses are innite). As is w ell kno wn, this is the riterion ensuring that the t yp e I I 1 v on Neumann algebra asso iated to the lattie is a fator [ T ak02 ,  V.7℄. Finally , w e indiate wh y Theorem 2.4 implies the lassial Borel densit y theorem of [ Bor60 ℄. It sues to justify the follo wing: Prop osition 2.8. L et k by a lo  al eld (A r hime de an or not), G a semi-simple k -gr oup without k -anisotr opi fators, X the symmetri sp a e or BruhatTits building asso iate d to G = G ( k ) and L < G any sub gr oup. If the L -ation on X is minimal without xe d p oint at innity, then L is Zariski-dense. Pr o of. Let ¯ L b e the ( k -p oin ts of the) Zariski losure of L . Then ¯ L is semi-simple; this follo ws e.g. from a v ery sp eial ase of Corollary 5.8 in [ CM08 ℄, whi h guaran tees that the radial of ¯ L is trivial. In the Ar himedean ase, w e ma y app eal to Karp elevi hMosto w theorem (see [ Kar53 ℄ or [ Mos55 ℄): an y semi-simple subgroup has a totally geo desi orbit in the symmetri spae. So the only semi-simple subgroup ating minimally is G itself. In the non-Ar himedean ase, w e ould app eal to E. Landv ogt funtorialit y theorem [ Lan00 ℄ whi h w ould nish the pro of. Ho w ev er, there is an alternativ e diret and elemen tary argu- men t whi h a v oids app ealing to lo . it. and go es as follo ws. First notie that, b y the same argumen t as in the pro of of Theorem 7.4 in [ CM08 ℄ p oin t ( iv ), the k -rank of a semi-simple subgroup ating minimally equals the k -rank of G (this holds in all ases, not only in the non-Ar himedean one). Therefore, the inlusion of spherial buildings B ¯ L → B G pro vided b y the group inlusion ¯ L → G has the prop ert y that B ¯ L is a top-dimensional sub-building of B G . An elemen tary argumen t (see [ KL06 , Lemma 3.3℄) sho ws that the union Y of all apartmen ts of X b ounded b y a sphere in B ¯ L is a losed on v ex subset of X . Clearly Y is ¯ L -in v arian t, hene Y = X b y minimalit y . Therefore B ¯ L = B G , whi h nally implies that ¯ L = G .  2.C. The limit set of subgroups of nite o v olume. Let X b e a omplete CA T(0) spae and G a group ating b y isometries on X . Reall that the limit set Λ G of G is the in tersetion of the b oundary ∂ X with the losure of the orbit G.x 0 in X = X ⊔ ∂ X of an y x 0 ∈ X , this set b eing indep enden t of x 0 . Prop osition 2.9. L et G b e a lo  al ly  omp at gr oup ating  ontinuously by isometries on a  omplete CA T(0) sp a e X . If Γ < G is any lose d sub gr oup with nite invariant  ovolume, then ΛΓ = Λ G . Consider the follo wing immediate orollary , whi h in the sp eial ase of Hadamard man- ifolds follo ws from the duality  ondition , see 1.9.16 and 1.9.32 in [ Eb e96 ℄. Corollary 2.10. L et G b e a lo  al ly  omp at gr oup with a  ontinuous ation by isometries on a pr op er CA T(0) sp a e. If the G -ation is  o  omp at, then any latti e in G has ful l limit set in ∂ X .  Pr o of of Pr op osition 2.9 . W e observ e that for an y non-empt y op en set U ⊆ G there is a ompat set C ⊆ G su h that U − 1 Γ C = G . Indeed, (using an idea of Selb erg, ompare Lemma 1.4 in [ Bor60 ℄), it sues to tak e C so large that µ  Γ C ) > µ (Γ \ G ) − µ (Γ U ) , 12 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD where µ denotes an in v arian t measure on Γ \ G ; an y righ t translate of Γ U in Γ \ G will then meet Γ C . No w let ξ ∈ Λ G and x 0 ∈ X . F or an y neigh b ourho o d V of ξ in ∂ X , w e shall onstrut an elemen t in ΛΓ ∩ V . Let U ⊆ G b e a ompat neigh b ourho o d of the iden tit y in G su h that U ξ ⊆ V and let { g n } b e a sequene of elemen ts of G with g n x 0 on v erging to ξ (one uses nets if X is not separable). In view of the ab o v e observ ation, there are sequenes { u n } in U and { c n } in C su h that u n g n c − 1 n ∈ Γ . The p oin ts g n c − 1 n x 0 remain at b ounded distane of g n x 0 as n → ∞ , and th us on v erge to ξ . Therefore,  ho osing an aum ulation p oin t u of { u n } in U , w e see that uξ is an aum ulation p oin t of { u n g n c − 1 n x 0 } , whi h is a sequene in Γ x 0 .  F or future use, w e observ e a v arian t of the ab o v e reasoning yielding a more preise fat in a simpler situation: Lemma 2.11. L et G b e a lo  al ly  omp at gr oup with a  ontinuous  o  omp at ation by isometries on a pr op er CA T(0) sp a e X . L et Γ < G b e a latti e and c : R + → X a ge o desi r ay suh that G xes c ( ∞ ) . Then ther e is a se quen e { γ i } in Γ suh that γ i c ( i ) r emains b ounde d over i ∈ N . Pr o of. F or the same reason as ab o v e, there is a ompat set U ⊆ G su h that G = U Γ U − 1 . Cho ose no w { g i } su h that g i c ( i ) remains b ounded and write g i = u i γ i v − 1 i with u i , v i ∈ U . W e ha v e d ( γ i c ( i ) , c (0)) = d ( g i v i c ( i ) , u i c (0)) ≤ d ( g i v i c ( i ) , g i c ( i )) + d ( g i c ( i ) , u i c (0)) ≤ d ( v i c ( i ) , c ( i )) + d ( g i c ( i ) , c (0)) + d ( u i c (0) , c (0)) . This is b ounded indep enden tly of i b eause d ( v i c ( i ) , c ( i )) ≤ d ( v i c (0) , c (0)) sine c ( ∞ ) is G -xed.  W e shall also need the follo wing: Lemma 2.12. A lo  al ly  omp at gr oup  ontaining a nitely gener ate d sub gr oup whose lo- sur e has nite  ovolume is  omp atly gener ate d. Pr o of. Denoting the losure of the giv en nitely generated subgroup b y Γ , w e an write G = U Γ C as in the pro of of Prop osition 2.9 with b oth U and C ompat. Sine Γ is a lo ally ompat group on taining a nitely generated dense subgroup, it is ompatly generated and the onlusion follo ws.  3. CA T(0) la tties, I: the Eulidean f a tor 3.A. Preliminaries on latties. W e b egin this setion with a few w ell kno wn basi fats ab out general latties. Prop osition 3.1. L et G b e a lo  al ly  omp at se  ond  ountable gr oup and N ✁ G b e a lose d normal sub gr oup. (i) Given a lose d  o  omp at sub gr oup Γ < G , the pr oje tion of Γ on G/ N is lose d if and only if Γ ∩ N is  o  omp at in N . (ii) Given a latti e Γ < G , the pr oje tion of Γ on G/ N is disr ete if and only if Γ ∩ N is a latti e in N . Pr o of. See Theorem 1.13 in [ Rag72 ℄.  The seond w ell kno wn result is straigh tforw ard to establish: ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 13 Lemma 3.2. L et G = H × D b e a lo  al ly  omp at gr oup. Given a latti e Γ < G and a  omp at op en sub gr oup Q < D , the sub gr oup Γ Q := Γ ∩ ( H × Q ) is a latti e in H × Q , whih is  ommensur ate d by Γ . If mor e over G/ Γ is  omp at, then so is ( H × Q ) / Γ Q .  (As w e shall see in Lemma 5.15 b elo w, there is a form of on v erse.) Let X b e a prop er CA T(0) spae and G = Is( X ) b e its isometry group. Giv en a disrete group Γ ating prop erly and o ompatly on X , then the quotien t G \ X is ompat and the image of Γ in G is a o ompat lattie (note that the k ernel of the map Γ → Is( X ) is nite). Con v ersely , if the quotien t G \ X is ompat, then an y o ompat lattie of G is a disrete group ating prop erly and o ompatly on X . Lemma 3.3. In the ab ove setting, G is  omp atly gener ate d and Γ is nitely gener ate d. Pr o of. F or la k of nding a lassial referene, w e refer to Lemma 22 in [ MMS04 ℄).  3.B. V ariations on Auslander's theorem. Lemma 3.4. L et A = R n ⋊ O ( n ) and S b e a semi-simple Lie gr oup without  omp at fator. A ny latti e Γ in G = A × S has a nite index sub gr oup Γ 0 whih splits as a dir e t pr o dut Γ 0 ∼ = Γ A × Γ ′ , wher e Γ A = Γ ∩ ( A × 1) is a latti e in ( A × 1) . Pr o of. Let V = R n denote the translation subgroup of A and U denote the losure of the pro jetion of Γ to S . The subgroup U < S is losed of nite o v olume; therefore it is either disrete or it on tains a semi-simple subgroup of p ositiv e dimension b y Borel's densit y theorem (in fat one ould b e more preise using the Main Result of [ Pra77 ℄, but this is not neessary for the presen t purp oses). On the other hand, Auslander's theorem [ Rag72 , Theorem 8.24℄ ensures that the iden tit y omp onen t of the pro jetion of Γ in S × A/V is soluble, from whi h it follo ws that U has a onneted soluble normal subgroup. Th us U is disrete. Therefore, b y Prop osition 3.1 , the group Γ A = Γ ∩ ( A × 1) is a lattie in ( A × 1) . In partiular Γ A is virtually Ab elian [ Th u97 , Corollary 4.1.13℄. Sine the pro jetion of Γ to S is a lattie in S , it is nitely generated [ Rag72 , 6.18℄. Therefore Γ p ossesses a nitely generated subgroup Λ on taining Γ A and whose pro jetion to S oinides with the pro jetion of Γ . Notie that Λ is a lattie in S × A b y [ Sim96 , Theorem 23.9.3℄; therefore Λ has nite index in Γ , whi h sho ws that Γ is nitely generated. Sine Γ A is normal in Γ , the pro jetion Γ A of Γ to A normalises the lattie Γ A and is th us virtually Ab elian. Hene Γ A is a nitely generated virtually Ab elian group whi h normalises Γ A . Therefore Γ A has a nite index subgroup whi h splits as a diret pro dut of the form Γ A × C , and the preimage Γ ′ of C in Γ is a normal subgroup whi h in tersets Γ A trivially . In partiular the group Γ ′ · Γ A ∼ = Γ ′ × Γ A is a nite index normal subgroup of Γ , as desired.  Lemma 3.5. L et Γ b e a gr oup  ontaining a sub gr oup of the form Γ 0 ∼ = Γ 0 S × Γ 0 A , wher e Γ 0 S is isomorphi to a latti e in a semi-simple Lie gr oup with trivial  entr e and no  omp at fator, and Γ 0 A is amenable. If Γ  ommensur ates Γ 0 , then Γ  ommensur ates b oth Γ 0 S and Γ 0 A . Pr o of. Let Γ 1 ∼ = Γ 1 S × Γ 1 A b e a onjugate of Γ 0 in Γ . The pro jetion of Γ 0 ∩ Γ 1 to Γ 0 S is a nite index subgroup of Γ 0 S . By Borel densit y theorem, it m ust therefore ha v e trivial amenable radial. In partiular the pro jetion of Γ 0 ∩ Γ 1 A to Γ 0 S is trivial. Therefore the image of the pro jetion of Γ 0 ∩ Γ 1 S (resp. Γ 0 ∩ Γ 1 A ) to Γ 0 S (resp. Γ 0 A ) is of nite index. The desired assertion follo ws.  Prop osition 3.6. L et A = R n ⋊ O ( n ) , S b e a semi-simple Lie gr oup with trivial  entr e and no  omp at fator, D b e a total ly dis onne te d lo  al ly  omp at gr oup and G = S × A × D . 14 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD Then any nitely gener ate d latti e Γ < G has a nite index sub gr oup Γ 0 whih splits as a dir e t pr o dut Γ 0 ∼ = Γ A × Γ ′ , wher e Γ A ⊆ Γ ∩ (1 × A × D ) is a nitely gener ate d virtual ly A b elian sub gr oup whose pr oje tion to A is a latti e. Pr o of. Let Q < D b e a ompat op en subgroup. By Lemma 3.2 , the in tersetion Γ 0 = Γ ∩ ( S × A × Q ) is a lattie in S × A × Q , whi h is ommensurated b y Γ . Sine Q is ompat, the pro jetion of Γ 0 to S × A is a lattie, to whi h w e ma y apply Lemma 3.4 . Up on replaing Γ 0 b y a nite index subgroup (whi h amoun ts to replaing Q b y an op en subgroup), this yields t w o normal subgroups Γ 0 S , Γ 0 A < Γ 0 and a deomp osition Γ 0 = Γ 0 S · Γ 0 A , where Γ 0 S ∩ Γ 0 A ⊆ Q and Γ 0 A = Γ 0 ∩ (1 × A × Q ) is a nitely generated virtually Ab elian group whose pro jetion to A is a lattie. By virtue of Lemma 3.5 , w e dedue that the image of the pro jetion of Γ to A om- mensurates a lattie in A . But the ommensurator of an y lattie in A is virtually Ab elian. Therefore, up on replaing Γ b y a nite index subgroup, it follo ws that the pro jetion of Γ to A normalises the pro jetion of Γ 0 A . W e no w dene Γ A = \ γ ∈ Γ γ Γ 0 A γ − 1 . Then the pro jetion of Γ A oinides with the pro jetion of Γ 0 A sine A is Ab elian; in parti- ular it is still a lattie. F urthermore, the subgroup Γ A is normal in Γ . W e no w pro eed as in the pro of of Lemma 3.4 . Sine the pro jetion of Γ to A is nitely generated and virtually Ab elian, w e ma y th us nd in this group a virtual omplemen t to the image of the pro jetion of Γ A . Let Γ ′ b e the preimage of this omplemen t in Γ . Then, up on replaing Γ b y a nite index subgroup, the group Γ ′ is normal in Γ and Γ = Γ A · Γ ′ . Sine Γ A is normal as w ell, the omm utator [Γ A , Γ ′ ] is on tained in the in tersetion Γ A ∩ Γ ′ , whi h is trivial b y onstrution. This nally sho ws that Γ ∼ = Γ A × Γ ′ , as desired.  Remark 3.7. In the setting of Prop osition 3.6 , assume that an y ompat subgroup of D normalised b y Γ is trivial. Then Γ A ⊆ 1 × A × 1 and the pro jetion of Γ to S × D is disrete. Indeed, the denition of Γ A giv en in the pro of sho ws that it is on tained in 1 × A × γ Qγ − 1 for all γ ∈ Γ and under the urren t assumptions the in tersetion T γ Qγ − 1 is trivial. The laim ab out the pro jetion to S × D follo ws from Prop osition 3.1 . 3.C. Latties, the Eulidean fator and xed p oin ts at innit y. Giv en a prop er CA T(0) spae X and a disrete group Γ ating prop erly and o ompatly , it is a w ell kno wn op en question, going ba k to M. Gromo v [ Gro93 ,  6 . B 3 ℄, to determine whether the presene of an n -dimensional at in X implies the existene of a free Ab elian group of rank n in Γ . (In the manifold ase, see problem 65 on Y au's list [ Y au82 ℄.) Here w e prop ose the follo wing theorem; the sp eial ase where X/ Γ is a ompat Riemannian manifold is the main result of Eb erlein's artile [ Eb e83 ℄ (ompare also the earlier Theorem 5.2 in [ Eb e80 ℄). Theorem 3.8. L et X b e a pr op er CA T(0) sp a e suh that G = Is( X ) ats  o  omp atly. Supp ose that X ∼ = R n × X ′ . (i) A ny nitely gener ate d latti e Γ < G has a nite index sub gr oup Γ 0 whih splits as a dir e t pr o dut Γ 0 ∼ = Z n × Γ ′ . (ii) If mor e over X is G -minimal ( e.g. if X is ge o desi al ly  omplete), then Z n ats trivial ly on X ′ and as a latti e on R n ; the pr oje tion of Γ to Is( X ′ ) is disr ete. W e reall that o ompat latties are automatially nitely generated in the ab o v e set- ting, Lemma 3.3 . The follo wing example sho ws that, without the assumption that G ats minimally , the pro jetion of Γ to Is( X ′ ) should not b e exp eted to ha v e disrete image: ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 15 Example 3.9 . Let X b e the losed submanifold of R 3 dened b y X = { ( x, y , z ) ∈ R 3 | 1 ≤ z ≤ 2 } and onsider the follo wing Riemannian metri on X : ds 2 = dx 2 + z 2 dy 2 + dz 2 . One readily v eries that it is non-p ositiv ely urv ed; th us X is a CA T(0) manifold. Clearly X splits o a one-dimensional Eulidean fator along the x -axis. Moreo v er the group H ∼ = R 2 of all translations along the xy -plane preserv es X and ats o ompatly . Let Γ b e the subgroup of H generated b y a and b , where a : ( x, y , z ) 7→ ( x, y , z ) + ( √ 2 , 1 , 0) and b : ( x, y , z ) 7→ ( x, y , z ) + (1 , √ 2 , 0) . Then Γ ∼ = Z 2 is a o ompat lattie in Is( X ) , but no non-trivial subgroup of Γ ats trivially on the y z -fator of X . The pro jetion of Γ to the isometry group of that fator is not disrete (see Prop osition 3.1 (ii)). The ab o v e result is the on v erse to the Flat T orus Theorem when it is stated as in [ BH99 , I I.7.1℄. In partiular w e dedue that the dimension of the Eulidean de Rham fator is an in v arian t of Γ . In the manifold ase, again, this is the main p oin t of [ Eb e83 ℄. Corollary 3.10. L et X b e a pr op er CA T(0) sp a e suh that G = Is( X ) ats  o  omp atly and minimal ly. L et Γ < G b e a nitely gener ate d latti e. Then the dimension of the Eulide an fator of X e quals the maximal r ank of a fr e e A b elian normal sub gr oup of Γ . In order to apply Theorem 1.6 in [ CM08 ℄ and A ddendum 1.8 in [ CM08 ℄ to w ards The- orem 3.8 , w e will need the follo wing. Theorem 3.11. L et X b e a pr op er CA T(0) sp a e suh that G = Is( X ) ats  o  omp atly and  ontains a nitely gener ate d latti e. Then X  ontains a  anoni al lose d  onvex G -invariant G -minimal subset X ′ 6 = ∅ whih has no Is( X ′ ) -xe d p oint at innity. Consider the immediate orollary . Corollary 3.12. L et X b e a pr op er CA T(0) sp a e suh that G = Is( X ) ats  o  omp atly and minimal ly. If G  ontains a nitely gener ate d latti e, then G has no xe d p oint at innity.  This sho ws that the mere existene of a nitely generated lattie imp oses restritions on o ompat CA T(0) spaes; m u h more detailed results in that spirit will b e giv en in Setion 6 . W e do not kno w whether the statemen t of Corollary 3.12 remains true without the nite generation assumption on the lattie (see Problem 7.3 b elo w). Example 3.13 . W e emphasise that the full isometry group of a o ompat prop er CA T(0) spae ma y ha v e global xed p oin ts at innit y; in fat, the spae migh t ev en b e homogeneous, as it is the ase for E. Hein tze's manifolds [ Hei74 ℄ men tioned earlier. An ev en simpler w a y to onstrut o ompat prop er CA T(0) spae with this prop ert y is to mimi Example 7.6 in [ CM08 ℄: Start from a regular tree T , assuming for deniteness that the v aleny is three. Replae ev ery v ertex b y a ongruen t op y of an isoseles triangle that is not equilateral, in su h a w a y that its distinguished v ertex alw a ys p oin ts to a xed p oin t at innit y (of the initial tree). Then the stabiliser H in Is( T ) of that p oin t at innit y still ats faithfully and o ompatly on the mo died spae T ′ ; the onstrution is so that the isometry group of T ′ is in fat redued to H . 16 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD W e shall also establish a strengthening of Corollary 3.12 , whi h an b e view ed as a form of Borel (or geometri) densit y theorem without assumption ab out xed p oin ts at innit y . Theorem 3.14. L et X b e a pr op er CA T(0) sp a e suh that G = Is( X ) ats  o  omp atly and minimal ly. Assume ther e is a nitely gener ate d latti e Γ < G . Then Γ ats minimal ly on X and mor e over al l Γ -xe d p oints at innity ar e  ontaine d in the b oundary of the (p ossibly trivial) Eulide an fator of X . W e no w turn to the pro ofs. In the ase of a disrete o ompat group Γ = G , a v ersion of the follo wing w as rst established b y BurgerS hro eder [ BS87 ℄ (as p oin ted out in [ AB98 , Corollary 2.7℄). Prop osition 3.15. L et X b e a pr op er CA T(0) sp a e, G < I s ( X ) a lose d sub gr oup whose ation on X is  o  omp at and Γ < G a nitely gener ate d latti e. Then ther e exists a Γ - invariant lose d  onvex subset Y ⊆ X whih splits Γ -e quivariantly as Y = E × W , wher e E is a (p ossibly 0 -dimensional) Eulide an sp a e on whih Γ ats by tr anslations and suh that ∂ E  ontains the xe d p oint set of G in ∂ X . Pr o of. W e an assume that there are G -xed p oin ts at innit y , sine otherwise there is nothing to pro v e. W e laim that for an y G -xed p oin t ξ there is a geo desi line σ : R → X with σ (+ ∞ ) = ξ su h that an y γ ∈ Γ mo v es σ to within a b ounded distane of itself  and hene to a parallel line b y on v exit y of the metri. Indeed, let c : R + → X b e a geo desi ra y with c ( ∞ ) = ξ and let { γ i } b e as in Lemma 2.11 . Then, b y ArzelàAsoli, there is a subsequene I ⊆ N and a geo desi line σ : R → X su h that σ ( t ) = lim i ∈ I γ i c ( t + i ) for all t . Sine ea h g ∈ G has b ounded displaemen t along c , the sequene { γ i g γ − 1 i } i ∈ I is b ounded and th us w e an assume that it on v erges for all g (realling that G is seond oun table, but w e shall only onsider g ∈ Γ an yw a y). Sine Γ is disrete and nitely generated, w e an further restrit I so that there is γ ∞ ∈ Γ su h that γ i γ γ − 1 i = γ ∞ γ γ − 1 ∞ ∀ γ ∈ Γ , i ∈ I . Sine d ( γ γ − 1 ∞ σ ( t ) , γ − 1 ∞ σ ( t )) = lim i ∈ I d ( γ − 1 i γ ∞ γ γ − 1 ∞ γ i c ( t + i ) , c ( t + i )) = lim i ∈ I d ( γ c ( t + i ) , c ( t + i )) ≤ d ( γ c (0) , c (0)) , It no w follo ws that ev ery γ ∈ Γ has b ounded displaemen t length along the geo desi γ − 1 ∞ σ . Th us the same holds for the geo desi σ whi h is therefore (b y on v exit y) translated to a parallel line b y ea h elemen t of Γ as laimed. Consider a at E ⊆ X that is maximal for the prop ert y that ea h elemen t of Γ has onstan t displaemen t length on E . Let Y b e the union of all ats that are at nite distane from E . One sho ws that Y splits as Y ∼ = E × W for some losed on v ex W ⊆ X using the Sandwi h Lemma [ BH99 , I I.2.12℄ and Lemma I I.2.15 of [ BH99 ℄ just lik e in Setion 3.B in [ CM08 ℄. The denition of Y sho ws that Γ preserv es Y as w ell as its splitting and ats on the E o ordinate b y translations. It remains to sho w that an y G -xed p oin t ξ ∈ ∂ X b elongs to ∂ E . First, ξ ∈ ∂ Y sine ∂ Y = ∂ X b y Corollary 2.10 ; w e th us represen t ξ b y a ra y c : R + → Y . Let no w σ b e a geo desi line as pro vided b y the laim. W e an assume that σ lies in Y b eause it w as onstruted from Γ -translates of c and Y is Γ -in v arian t. One an write σ = ( σ E , σ W ) where σ E , σ W are linearly re-parametrised geo desis in E and W , see [ BH99 , I.5.3℄. W e need to pro v e that σ W has zero sp eed. Sine an y giv en γ ∈ Γ has onstan t displaemen t along σ and on ea h of the parallel opies of E individually , its displaemen t is onstan t on the union of ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 17 all parallel opies of E visited b y σ , whi h is E × σ W ( R ) . The latter b eing again a at, the maximalit y of E sho ws that σ W is onstan t.  Pr o of of The or em 3.11 . Let Y = E × W ⊆ X b e as in Prop osition 3.15 . Reall that ∂ Y = ∂ X b y Corollary 2.10 . W e laim that ∂ X has irumradius > π / 2 . Indeed, it w ould otherwise ha v e a G -xed irumen tre b y Prop osition 3.1 in [ CM08 ℄, but this irumen tre annot b elong to ∂ E sine E is Eulidean; this on tradits Prop osition 3.15 . W e no w apply Corollary 3.10 in [ CM08 ℄. This yields a anonial G -in v arian t losed on v ex subset X ′ , whi h is minimal with resp et to the prop ert y that ∂ X ′ = ∂ X . It follo ws in partiular b y Corollary 2.10 that Γ ats minimally on X ′ . Let no w X ′ = E ′ × X ′ 0 b e the anonial splitting, where E ′ is the maximal Eulidean fator [ BH99 , I I.6.15℄. On the one hand, sine X ′ is Γ -minimal, Prop osition 3.15 applied to X ′ sho ws that G has no xed p oin ts in ∂ X ′ 0 sine E ′ is maximal as a Eulidean fator. On the other hand, Is( E ′ ) xes no p oin t at innit y on E ′ . W e dedue that Is( X ′ ) ∼ = Is( E ′ ) × Is( X ′ 0 ) has indeed no xed p oin t at innit y .  End of pr o of of The or em 3.14 . Arguing as in the pro of of Theorem 3.11 , w e establish that X is Γ -minimal. Let X = X ′ × E b e the anonial splitting, where E is the maximal Eulidean fator. Sine an y isometry of X deomp oses uniquely as isometries of E and X ′ (I I.6.15 in [ BH99 ℄), is sues to sho w that Γ has no xed p oin t in ∂ X ′ . This follo ws from Prop osition 2.1 applied to the G -ation on X ′ .  End of pr o of of The or em 3.8 . Assume rst that X is G -minimal, realling that this is the ase if X is geo desially omplete b y Lemma 3.13 in [ CM08 ℄. In view of Corollary 3.12 , w e an apply Theorem 1.6 in [ CM08 ℄ and w e are therefore in the setting of Prop osition 3.6 . Sine the group Γ A pro vided b y that prop osition on tains a nite index subgroup isomorphi to Z n , w e ha v e already established ( i ) under the additional minimalit y assumption. In order to sho w ( ii ), it sues b y Remark 3.7 to pro v e that an y ompat subgroup of G normalised b y Γ is trivial. This follo ws from the fat that X is Γ -minimal, as established in Theorem 3.14 . It remains to pro v e ( i ) without the assumption that X is G -minimal. Let Y ⊆ X b e the G -minimal set pro vided b y Theorem 3.11 and let Y ∼ = R m × Y ′ b e its Eulidean deomp osition. Then m ≥ n b eause of the  haraterisation of the Eulidean fator in terms of Cliord isometries [ BH99 , I I.6.15℄; indeed, an y (non-trivial) Cliord isometry of X restrits non-trivially to Y b eause Y has nite o-diameter. The k ernel F ✁ Γ of the Γ -ation on Y is nite and th us w e an assume that it is en tral up on replaing Γ with a nite index subgroup. P assing to a further nite index subgroup, w e kno w from the minimal ase that Γ /F splits as Γ /F = Z m × Λ ′ . Let Γ Z m , Γ ′ ✁ Γ b e the pre-images in Γ of those fators. Th us w e an write Γ = Γ Z m · Γ ′ with Γ Z m ∩ Γ ′ ⊆ F . It is straigh tforw ard that a nite en tral extension of Z m is virtually Z m (see e.g. [ BH99 , I I.7.9℄). Therefore Γ on tains a nite index subgroup isomorphi to Z m × Γ ′ and the result follo ws sine m ≥ n .  Pr o of of Cor ol lary 3.10 . Notie that a splitting Γ 0 ∼ = Z n × Γ ′ with Γ 0 normal and n maximal pro vides a normal subgroup Z n ✁ Γ sine Z n is  harateristi in Γ 0 . Therefore, giv en Theorem 3.8 , it only remains to see that a normal Z n ✁ Γ of maximal rank fores X to ha v e a Eulidean fator of dimension at least n . Otherwise, the pro jetion of Γ to the non- Eulidean fator X ′ w ould b e a lattie b y Theorem 3.8 ( ii ) and on tain an innite normal amenable subgroup, on traditing Corollary 2.7 ( i ).  18 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD Finally , w e reord that Theorem 1.3 is on tained in Theorem 3.8 and Corollary 3.10 for (i), and Corollary 3.12 and Theorem 3.14 for (ii). 4. CA T(0) la tties, I I: pr oduts 4.A. Irreduible latties in CA T(0) spaes. Reall from that a (top ologial) group is alled irreduible if no (op en) nite index subgroup splits non-trivially as a diret pro dut of (losed) subgroups. F or example, an y lo ally ompat group ating on tin uously , prop erly , minimally , without xed p oin t at innit y on an irreduible prop er CA T(0) spae is irreduible b y Theorem 1.10 in [ CM08 ℄. In partiular, an abstrat group Γ is irreduible if it do es not virtually split. This ter- minology is inspired b y the onept of irreduibilit y for losed manifolds, whi h means that no nite o v er of the manifold splits non-trivially . Of ourse, the univ ersal o v er of su h a manifold an still split. Indeed, one gets man y lassial CA T(0) groups b y onsidering irre- duible latties in pro duts of simple Lie groups or more generally of semi-simple algebrai groups o v er v arious lo al elds. The latter onept of irreduibilit y for latties is dened as follo ws: A lattie Γ < G = G 1 × · · · × G n in a pro dut of lo ally ompat groups is alled an irreduible lattie if its pro jetions to an y subpro dut of the G i 's are dense and ea h G i is non-disrete. The p oin t of this notion (and of the nearly onfusing terminology) is that it prev en ts Γ and its nite index subgroups from splitting as a pro dut of latties in G i . Moreo v er, if all G i 's are en tre-free simple Lie (or algebrai) groups without ompat fators, the irreduibilit y of Γ as a lattie is equiv alen t to its irreduibilit y as a group in and for itself; this is a result of Margulis [ Mar91 , I I.6.7℄. As w e shall see in Theorem 4.2 b elo w, a v ersion of this equiv alene holds for latties in the isometry group of a CA T(0) spae. Remark 4.1. (i) The non-disreteness of G i is often omitted from this denition; the dierene is inessen tial sine the notion of a lattie is trivial for disrete groups. Notie ho w ev er that our denition ensures that all G i are non-ompat and that n ≥ 2 . (ii) One v eries that any lattie Γ < G = G 1 × G 2 is an irreduible lattie in the pro dut G ∗ < G of the losures G ∗ i < G i of its pro jetions to G i (pro vided these pro jetions are non-disrete). The follo wing geometri v ersion of Margulis' riterion on tains Theorem 1.4 from the In tro dution. Theorem 4.2. L et X b e a pr op er CA T(0) sp a e, G < Is( X ) a lose d sub gr oup ating  o  omp atly on X , and Γ < G a nitely gener ate d latti e. (i) If Γ is irr e duible as an abstr at gr oup, then for any nite index sub gr oup Γ 0 < Γ and any Γ 0 -e quivariant splitting X = X 1 × X 2 with X 1 and X 2 non- omp at, the pr oje tion of Γ 0 to b oth Is( X i ) is non-disr ete. (ii) If in addition the G -ation is minimal, then the  onverse statement holds as wel l. Remark 4.3. Reall that the G -minimalit y is automati if X is geo desially omplete (Lemma 3.13 in [ CM08 ℄). Statemen t (ii) fails ompletely without minimalit y (as witnessed for instane b y the unosmop olitan mien of an equiv arian t mane). Pr o of of The or em 4.2 . Supp ose Γ irreduible. Let X ′ ⊆ X b e the anonial subspae pro- vided b y Theorem 3.11 . By Theorem 3.8 , the spae X ′ has no Eulidean fator unless X = R and Γ is virtually yli, in whi h ase the desired statemen t is empt y . ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 19 W e rst deal with the ase when G ats minimally on X ; b y Theorem 2.4 this amoun ts to assume X = X ′ . Supp ose for a on tradition that for Γ 0 and X ′ = X ′ 1 × X ′ 2 as in the statemen t, the pro jetion G 1 of Γ 0 to Is( X ′ 1 ) is disrete. Let G 2 b e the losure of the pro jetion of Γ 0 to Is( X ′ 2 ) and notie that b oth G i are ompatly generated sine Γ and hene also Γ 0 is nitely generated. The pro jetion Γ 2 of Γ 0 ∩ (1 × G 2 ) to G 2 is a lattie (b y Lemma 3.2 or b y Prop osition 3.1 ); b eing normal, it is o ompat and hene nitely generated. By Theorem 3.14 , the group Γ 0 , and hene also G 2 , ats minimally and without xed p oin t at innit y on X ′ 2 . Therefore Corollary 2.7 ( ii ) implies that the en traliser Z G 2 (Γ 2 ) is trivial. But Γ 2 is disrete, normal in G 2 , and nitely generated. Hene Z G 2 (Γ 2 ) is op en and th us G 2 is disrete. Therefore, the pro dut G 1 × G 2 , whi h on tains Γ 0 , is a lattie in Is( X ′ 1 ) × Is( X ′ 2 ) and th us in G . No w the index of Γ 0 in G 1 × G 2 is nite and th us Γ 0 splits virtually , a on tradition. W e no w ome ba k to the general ase X ′ ⊆ X and supp ose that X p ossesses a Γ 0 - equiv arian t splitting X = X 1 × X 2 . The group H = Is( X 1 ) × I s ( X 2 ) < Is( X ) on tains Γ 0 ; hene its ation on X ′ is minimal without xed p oin t at innit y b y Corollary 3.12 . There- fore, the splitting theorem [ Mon06 , Theorem 9℄ implies that X ′ p ossesses a Γ 0 -equiv arian t splitting X ′ = X ′ 1 × X ′ 2 indued b y X = X 1 × X 2 via H . Up on replaing Γ 0 b e a nite index subgroup, the preeding paragraph th us yields a splitting Γ 0 /F ∼ = G 1 × G 2 of the image of Γ 0 in Is( X ′ ) , where F denotes the k ernel of the Γ 0 -ation on X ′ . Sine F is nite, so is the pro jetion to Is( X 3 − i ) of the preimage b G i of G i in Γ , for i = 1 , 2 . Therefore up on passing to a nite index subgroup w e ma y and shall assume that b G i ats trivially on Is( X 3 − i ) . No w the subgroup of Is( X 1 ) × Is ( X 2 ) generated b y b G 1 and b G 2 splits as b G 1 × b G 2 and is ommensurable to Γ 0 , a on tradition. Con v ersely , supp ose no w that the G -ation is minimal and that Γ = Γ ′ × Γ ′′ splits non- trivially (after p ossibly ha ving replaed it b y a nite index subgroup). If X = R n , then reduibilit y of Γ fores n ≥ 2 and w e are done in view of the struture of Bieb erba h groups. If X is not Eulidean but has a Eulidean fator, then Theorem 3.8 ( ii ) pro vides a disrete pro jetion of Γ to the non-Eulidean fator Is( X ′ ) ; furthermore, X ′ is indeed non-ompat as desired sine otherwise b y minimalit y it is redued to a p oin t, on trary to our assumption. If on the other hand X has no Eulidean fator, then Γ ats minimally and without xed p oin t at innit y b y Theorem 3.11 . Then the desired splitting is pro vided b y the splitting theorem [ Mon06 , Theorem 9℄. Both pro jetions of Γ are disrete, indeed isomorphi to Γ ′ resp etiv ely Γ ′′ b eause the ited splitting theorem ensures omp onen t wise ation of Γ .  W e no w briey turn to uniquely geo desi spaes and to the analogues in this setting of some of P . Eb erlein's results for Hadamard manifolds. Theorem 4.4. L et X b e a pr op er CA T(0) sp a e with uniquely extensible ge o desis suh that Is( X ) ats  o  omp atly on X . If Is( X ) admits a nitely gener ate d non-uniform irr e duible latti e, then X is a symmetri sp a e (without Eulide an fator). Pr o of. The ation of Is( X ) is minimal b y Lemma 3.13 in [ CM08 ℄ and without xed p oin t at innit y b y Corollary 3.12 . Th us w e an apply Theorem 1.6 in [ CM08 ℄ and A ddendum 1.8 in [ CM08 ℄. Notie that Is( X ) itself is non-disrete sine it on tains a non-uniform lattie; moreo v er, if it admits more than one fator in the deomp osition of Theorem 1.6 in [ CM08 ℄, then the latter are all non-disrete b y Theorem 4.2 . Therefore, w e an apply Theorem 7.10 20 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD in [ CM08 ℄ to all fators of X . It remains only to justify that X has no Eulidean fa- tor; otherwise, Auslander's theorem (ompare also Theorem 3.8 ) implies X = R , whi h is inompatible with the fat that Γ is non-uniform.  The follo wing related result is due to P . Eb erlein in the manifold ase (Prop osition 4.5 in [ Eb e82 ℄). W e shall establish another result of the same v ein later without assuming that geo desis are uniquely extensible (see Theorem 6.6 ). Theorem 4.5. L et X b e a pr op er CA T(0) sp a e with uniquely extensible ge o desis and Γ < Is( X ) b e a disr ete  o  omp at gr oup of isometries. If Γ is irr e duible (as an abstr at gr oup) and X is r e duible, then X is a symmetri sp a e (without Eulide an fator). Pr o of. One follo ws line-b y-line the pro of of Theorem 4.4 . The only dierene is that, in the presen t on text, the non-disreteness of the isometry group of ea h irreduible fator of X follo ws from Theorem 4.2 sine X is assumed reduible.  W e an no w onlude the pro of of Theorem 1.10 from the In tro dution. The rst state- men t w as established in Theorem 7.10 in [ CM08 ℄. The seond follo ws from Theorem 4.4 and the third from Theorem 4.5 in the uniform ase, and from Theorem 4.4 in the non-uniform one.  4.B. The h ull of a lattie. Let X b e a prop er CA T(0) spae X su h that Is( X ) ats o ompatly on X . Let Γ < Is( X ) b e a nitely generated lattie; note that the ondition of nite generation is redundan t if Γ is o ompat b y Lemma 3.3 . Theorem 3.11 pro vides a anonial Is( X ) -in v arian t subspae X ′ ⊆ X su h that G = I s ( X ′ ) has no xed p oin t at innit y . In this setion w e shall dene the h ull H Γ of the lattie Γ ; this is a lo ally ompat group H Γ < Is( X ′ ) anonially atta hed to the situation and on taining the image of Γ in Is( X ′ ) . F or simpliit y , w e rst treat the sp eial ase where Is( X ) ats minimally; th us X ′ = X and G = I s( X ) . Applying Theorem 1.6 in [ CM08 ℄ and A ddendum 1.8 in [ CM08 ℄, w e see in partiular that Γ p ossesses a anonial nite index normal subgroup Γ ∗ = Γ ∩ G ∗ whi h is the k ernel of the Γ -ation b y p erm utation on the set of fators in the deomp osition giv en b y A ddendum 1.8 in [ CM08 ℄. In the lassial ase when X is a symmetri spae, the losure of the pro jetion of Γ to the isometry group Is( X i ) of ea h fator is an op en subgroup of nite index , as so on as X is reduible. This is no longer true in general, ev en in the ase of Eulidean buildings. In fat, the same Γ ma y (and generally do es) o ur as lattie in inreasingly large am bien t groups Γ < G < G ′ < G ′′ < · · · . In order to address this issue, w e dene the h ull as follo ws. Consider the losed subgroup H Γ ∗ < G whi h is the diret pro dut of the losure of the images of Γ ∗ in ea h of the fators in the deomp osition of Theorem 1.6 in [ CM08 ℄. Then set H Γ = Γ · H Γ ∗ . In other w ords, w e ha v e inlusions Γ < H Γ < G. The losed subgroup H Γ ∗ is nothing but the h ull of the lattie Γ ∗ . It oinides with H ∗ Γ = H Γ ∩ G ∗ . In partiular H ∗ Γ = H Γ ∗ is a diret pro dut of irreduible groups satisfying all the restritions of Theorem 1.10 in [ CM08 ℄ (exept for the p ossible Eulidean motion fator), and the image of Γ ∗ in ea h of these fators is dense. Remark 4.6. Notie that Γ is alw a ys a lattie in H Γ (b y [ Rag72 , Lemma 1.6℄). W e emphasise that H Γ is non-disrete and that Γ ∗ is an irreduible lattie in H Γ ∗ (in the ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 21 sense of  4.A ) as so on as Γ is irreduible as a group and X is reduible; this follo ws from Theorem 4.2 . W e no w dene the h ull H Γ < G in the general situation G = Is( X ′ ) with X ′ ⊆ X giv en b y Theorem 3.11 . Sine Is( X ) \ X is o ompat, it follo ws that X ′ is r -dense in X for some r > 0 and the anonial map Is( X ) → G is prop er. Let F Γ ✁ Γ b e the nite k ernel of the indued map Γ → G and write Γ ′ := Γ /F Γ . Then the h ull of Γ is dened b y H Γ := H Γ ′ (reduing to the ab o v e ase). In other w ords, Γ sits in H Γ only mo dulo the anonial nite k ernel F Γ . In fat, F Γ is ev en anonially atta hed to Γ view ed as an abstrat group. Lemma 4.7. F Γ is a (ne  essarily unique) maximal nite normal sub gr oup of Γ . Mor e over, X ′ is Γ ′ -minimal. Pr o of. The Γ ′ -ation on X ′ is minimal b y an appliation of Theorem 3.14 and therefore ev ery nite normal subgroup of Γ ′ is trivial. Sine moreo v er the Γ -ation on X ′ is prop er, it follo ws that a normal subgroup of Γ is nite if and only if it lies in F Γ .  F or later referenes, w e reord the follo wing exp eted fat. Lemma 4.8. Assume that Γ is irr e duible. If X ′ is r e duible, then H Γ  ontains the identity  omp onent of G := Is( X ′ ) . In fat ( H Γ ) ◦ = G ◦ is a semi-simple Lie gr oup with trivial  entr e and no  omp at fator. Pr o of. By Theorem 3.8 , the h yp otheses on Γ imply that X ′ has no Eulidean fator. Th us ea h almost onneted fator of G ∗ is a simple Lie group with trivial en tre and no ompat fator. The pro jetion of Γ ∗ to ea h of these fators is non-disrete b y Theorem 4.2 and the assumption made on X ′ . Its losure is semi-simple and Zariski dense b y Theorem 2.4 and Prop osition 2.8 . The result follo ws.  4.C. On the anonial disrete k ernel. Let G = G 1 × G 2 b e a lo ally ompat group and Γ < G b e an irreduible lattie. It follo ws from irreduibilit y that the pro jetion to G i of the k ernel of the pro jetion Γ → G j 6 = i is a normal subgroup of G i . In other w ords, w e ha v e a anonial disrete normal subgroup Γ i ✁ G i dened b y Γ 1 = Pro j G 1  Γ ∩ ( G 1 × 1)  (and lik ewise for Γ 2 ) whi h w e all the anonial disrete k ernel of G i (dep ending on Γ ). W e observ e that the image Γ = Γ / (Γ 1 · Γ 2 ) of Γ in the anonial quotien t G 1 / Γ 1 × G 2 / Γ 2 is still an irreduible lattie (see Prop osi- tion 3.1 (ii)) and has the additional prop ert y that it pro jets injetiv ely in to b oth fators. In this subsetion, w e ollet some basi fats on latties in (pro duts of ) totally dison- neted lo ally ompat groups, adapting ideas of M. Burger and Sh. Mozes (see Prop osi- tions 2.1 and 2.2 in [ BM00b ℄). Prop osition 4.9. L et Γ < G = G 1 × G 2 b e an irr e duible latti e. Assume that G 2 is total ly dis onne te d,  omp atly gener ate d and without non-trivial  omp at normal sub gr oup. If Γ is r esidual ly nite, then  anoni al the disr ete kernel Γ 2 = Γ ∩ (1 × G 2 )  ommutes with the disr ete r esidual G ( ∞ ) 2 . Reall that the disrete residual G ( ∞ ) of a top ologial group G is b y denition the in tersetion of all op en normal subgroups. It is imp ortan t to remark that, b y Corollary 6.13 in [ CM08 ℄ the disrete residual of a non-disrete ompatly generated lo ally ompat group without non-trivial ompat normal subgroup is neessarily non-trivial. 22 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD Pr o of of Pr op osition 4.9 . By a sligh t abuse of notation, w e shall iden tify G 2 with the sub- group 1 × G 2 of G . Giv en a nite index normal subgroup Γ 0 ✁ Γ , the in tersetion Γ 0 ∩ Γ 2 is a disrete normal subgroup of G 2 (b y irreduibilit y), on tained as a nite index subgroup in Γ 2 . Th us G 2 ats b y onjugation on the nite quotien t Γ 2 / Γ 0 ∩ Γ 2 . In partiular the k ernel of this ation is a nite index losed normal subgroup, whi h is th us op en. Therefore, the disrete residual G ( ∞ ) 2 ats trivially on Γ 2 / Γ 0 ∩ Γ 2 . In other w ords, this means that [Γ 2 , G ( ∞ ) 2 ] ⊆ Γ 0 ∩ Γ 2 . Assume no w that Γ is residually nite. The preeding argumen t then sho ws that the omm utator [Γ 2 , G ( ∞ ) 2 ] is trivial, as desired.  Prop osition 4.10. L et Γ < G = G 1 × G 2 b e a  o  omp at latti e in a pr o dut of  omp atly gener ate d lo  al ly  omp at gr oups. Assume that G 2 is total ly dis onne te d and that the  en- tr aliser in G 1 of any  o  omp at latti e of G 1 is trivial. If the disr ete kernel Γ 2 = Γ ∩ (1 × G 2 ) is trivial, then the quasi- entr e Q Z ( G 2 ) is top olo gi al ly lo  al ly nite. Pr o of. Let S ⊆ Q Z ( G 2 ) b e a nite subset of the quasi-en tre. Then G 2 p ossesses a ompat op en subgroup U whi h en tralises S . By Lemma 3.2 the group Γ U = Γ ∩ ( G 1 × U ) is a o ompat lattie in G 1 × U . In partiular, there is a nite generating set T ⊆ Γ U . By a lemma of Selb erg [ Sel60 ℄, the group Z Γ ( T ) is a o ompat lattie in Z G ( T ) . But Z G ( T ) = Z G (Γ U ) ⊆ 1 × G 2 sine the pro jetion of Γ U to G 1 is a o ompat lattie. Sine the disrete k ernel Γ ∩ (1 × G 2 ) is trivial b y h yp othesis, the en traliser Z Γ ( T ) is trivial and, hene, Z G ( T ) is ompat. By onstrution S is on tained in Z G ( T ) , whi h yields the desired result.  4.D. Residually nite latties. Theorem 4.11. L et X b e a pr op er CA T(0) sp a e suh that Is( X ) ats  o  omp atly and minimal ly. L et Γ < Is ( X ) b e a nitely gener ate d latti e. Assume that Γ is irr e duible and r esidual ly nite. Then we have the fol lowing (se e Se tion 4.B for the notation): (i) Γ ∗ ats faithful ly on e ah irr e duible fator of X . (ii) If Γ is  o  omp at and X is r e duible, then for any lose d sub gr oup G < Is( X )  ontaining H Γ ∗ , we have Q Z ( G ) = Q Z ( G ∗ ) = 1 . F urthermor e so c( G ∗ ) is a dir e t pr o dut of r non-disr ete lose d sub gr oups, e ah of whih is har ateristi al ly simple, wher e r is the numb er of irr e duible fators of X . (W e emphasise that the irreduibilit y assumption onerns Γ as an abstrat group; ompare ho w ev er Remark 4.6 .) Pr o of. If X is irreduible, there is nothing to pro v e. W e assume heneforth that X is reduible. In view of Theorem 3.8 , X has no Eulidean fator. Moreo v er, Corollary 3.12 implies that Is( X ) xes no p oin t at innit y . In partiular, Γ and H Γ ∗ at minimally without xed p oin t at innit y b y Theorem 2.4 . Let H 1 , . . . , H r b e the irreduible fators of H Γ ∗ ; th us r oinides with the n um b er of irreduible fators of X . In view of Theorem 4.2 , the group Γ ∗ is an irreduible lattie in this pro dut. By Corollary 1.11 in [ CM08 ℄ and Theorem 2.4 , ea h H i is either a en tre-free simple Lie group or totally disonneted with trivial amenable radial. If H 1 is a simple Lie group, then it has no non-trivial disrete normal subgroup and hene (Γ ∗ ) 1 := Γ ∗ ∩ ( H 1 × 1 × · · · × 1) = 1 . ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 23 If H 1 is totally disonneted, then b y Prop osition 4.9 the anonial disrete k ernel (Γ ∗ ) 1 omm utes with the disrete residual H ( ∞ ) 1 , whi h is non-trivial b y Corollary 6.13 in [ CM08 ℄. Th us Z H 1 ( H ( ∞ ) 1 ) = 1 b y Theorem 1.10 in [ CM08 ℄ and hene (Γ ∗ ) 1 = 1 . Assertion (i) no w follo ws from a straigh tforw ard indution on r . Assume next that Γ is o ompat. Let G 1 , . . . , G r b e the irreduible fators of G ∗ . By Lemma 4.8 and Prop osition 4.10 , and in view of P art (i), for ea h totally disonneted fa- tor G i , the quasi-en tre Q Z ( G i ) is top ologially lo ally nite. Its losure is th us amenable, hene trivial b y Theorem 1.10 in [ CM08 ℄. Moreo v er, the quasi-en tre of ea h almost on- neted fator is trivial as w ell b y Lemma 4.8 . Clearly the pro jetion of the quasi-en tre of G ∗ to the irreduible fator G i is on tained in Q Z ( G i ) . This sho ws that Q Z ( G ∗ ) is trivial. Hene so is Q Z ( G ) , sine it on tains Q Z ( G ∗ ) as a nite index subgroup and sine G has no non-trivial nite normal subgroup b y Corollary 5.8 in [ CM08 ℄. No w the desired onlusion follo ws from Prop osition 6.11 in [ CM08 ℄.  4.E. Strong rigidit y for pro dut spaes. In [ CM08 ℄, w e presen ted a few sup errigidit y results (Setion 8 in [ CM08 ℄). Sup errigidit y should on tain, in partiular, strong rigidit y à la Mosto w. This is indeed the on ten t of Theorem 4.12 b elo w, where an isomorphism of latties is sho wn to extend to an am bien t group. Ho w ev er, in on trast to the lassial ase of symmetri spaes, whi h are homogeneous, the full isometry group do es not in general determine the spae sine CA T(0) spaes are in general not homogeneous. Another dierene is that the h ull of a lattie, as desrib ed in Setion 4.B , is generally smaller than the full isometry group of the am bien t CA T(0) spae. In view of the denition of the h ull, the follo wing statemen t is non-trivial only when X (or an in v arian t subspae) is reduible; this is exp eted sine w e w an t to use sup errigidit y for irreduible latties in pro duts. Theorem 4.12. L et X, Y b e pr op er CA T(0) sp a es and Γ , Λ disr ete  o  omp at gr oups of isometries of X , r esp e tively Y , not splitting (virtual ly) a Z n fator. Then any isomorphism Γ ∼ = Λ determines an isomorphism H Γ ∼ = H Λ suh that the fol- lowing  ommutes: Γ   / / Λ   H Γ ∼ = / / H Λ Theorem 4.12 pro vides a partial answ er to Question 21 in [ FHT08 ℄. Remark 4.13. The assumption on Z n fators is equiv alen t to exluding Eulidean fa- tors from X (or its anonial in v arian t subspae) b y Theorem 3.8 . On the one hand, this assumption is really neessary for the theorem to hold, ev en for symmetri spaes, sine one an t wist the pro dut using a Γ -ation on the Eulidean fator when H 1 (Γ) 6 = 0 (om- pare [ L Y72 ,  4℄). On the other hand, sine Bieb erba h groups are ob viously Mosto w-rigid, Theorem 4.12 together with Theorem 3.8 giv e us as omplete as p ossible a desription of the situation with Z n fators. Pr o of. Let X ′ ⊆ X b e the subset pro vided b y Theorem 3.11 . W e retain the notation F Γ ✁ Γ and Γ ′ = Γ /F Γ < Is( X ′ ) from Setion 4.B and reall from Lemma 4.7 that F Γ dep ends only on Γ as abstrat group and that X ′ is Γ ′ -minimal. W e dene Y ′ , F Λ and Λ ′ in the same w a y and ha v e the orresp onding lemma. In partiular, it follo w that the isomorphism Γ ∼ = Λ 24 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD desends to Γ ′ ∼ = Λ ′ . Therefore, w e an and shall assume from no w on that X and Y are minimal and Γ < H Γ < Is( X ) , Λ < H Λ < Is( Y ) . By Theorem 3.8 , w e kno w that X, Y ha v e no Eulidean fator. Th us Γ , Λ ha v e no xed p oin t at innit y b y Theorem 3.14 . W e laim that Γ has a nite index subgroup Γ † whi h deomp oses as a pro dut Γ † = Γ 1 × · · · × Γ s of irreduible fators, with s maximal for this prop ert y . Indeed, otherwise w e ould apply the splitting theorem of [ Mon06 ℄ to a  hain a nite index subgroups and on tradit the prop erness of X . W e write Λ † = Λ 1 × · · · × Λ s for the orresp onding groups in Λ . Com bining the splitting theorem with A ddendum 1.8 in [ CM08 ℄, it follo ws from the denition of the h ull that it is suien t to pro v e the statemen t for s = 1 . W e assume heneforth that Γ , and hene also Λ , is irreduible. F urthermore, if X and Y are b oth irreduible, then H Γ = Γ and H Λ = Λ and the desired statemen t is empt y . W e no w assume that X is reduible. By Theorem 2.4 , the lattie Γ (resp. Λ ) ats minimally without xed p oin t at innit y on X (resp. Y ). Theorem 8.4 in [ CM08 ℄ yields a on tin uous morphism f : H Γ ∗ → H Λ ∗ , whi h sho ws in partiular (b y the splitting theorem [ Mon06 ℄) that Y is reduible as w ell. A seond appliation of Theorem 8.4 in [ CM08 ℄ yields a seond on tin uous morphism f ′ : H Λ ∗ → H Γ ∗ . Notie that the resp etiv e restritions to Γ ∗ and Λ ∗ oinides with the giv en isomorphism and its in v erse. In partiular f ′ ◦ f (resp. f ′ ◦ f ) is the iden tit y on Γ (resp. Λ ). By denition of the h ull, it follo ws that f ′ ◦ f (resp. f ′ ◦ f ) is the iden tit y on H Γ ∗ (resp. H Λ ∗ ). The desired result nally follo ws, sine there is a anonial isomorphism Γ / Γ ∗ ∼ = H Γ /H Γ ∗ and sine the ation of H Γ /H Γ ∗ on H Γ ∗ is anonially determined b y the ation of Γ / Γ ∗ on Γ ∗ .  The ab o v e pro of sho ws in partiular that amongst spaes that are Γ -minimal without Eulidean fator, the n um b er of irreduible fators dep ends only up on the group Γ . If w e om bine this with Theorem 3.14 , Corollary 3.10 and Theorem 3.8 ( ii ), w e obtain that the n um b er of fators in the de Rham deomp osition (4.i) X ′ ∼ = X 1 × · · · × X p × R n × Y 1 × · · · × Y q of A ddendum 1.8 in [ CM08 ℄ is an in v arian t of the group: Corollary 4.14. L et X b e a pr op er CA T(0) sp a e and Γ < Is( X ) b e a gr oup ating pr op erly dis ontinuously and  o  omp atly. Then any other suh sp a e admitting a pr op er  o  omp at Γ -ation has the same numb er of fators in ( 4.i ) and the Eulide an fator has same dimension.  (W e reall that minimalit y is automati when X is geo desially omplete: Lemma 3.13 in [ CM08 ℄.) 5. Arithmetiity of abstra t la tties The main goal of this setion is to pro v e Theorem 1.9 , whi h w e no w state in a sligh tly more general form. F ollo wing G. Margulis [ Mar91 , IX.1.8℄, w e shall sa y that a simple algebrai group G dened o v er a eld k is admissible if none of the follo wing holds:  c har( k ) = 2 and G is of t yp e A 1 , B n , C n or F 4 .  c har( k ) = 3 and G is of t yp e G 2 . A semi-simple group will b e said admissible if all its fators are. Theorem 5.1. L et Γ < G = G 1 × · · · × G n b e an irr e duible nitely gener ate d latti e, wher e e ah G i is any lo  al ly  omp at gr oup. If Γ admits a faithful Zariski-dense r epr esentation in an admissible semi-simple gr oup (over any eld), then the amenable r adi al R of G is  omp at and the quasi- entr e Q Z ( G ) ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 25 is virtual ly  ontaine d in Γ · R . F urthermor e, up on r eplaing G by a nite index sub gr oup, the quotient G/R splits as G + × Q Z ( G/R ) wher e G + is a semi-simple algebr ai gr oup and the image of Γ in G + is an arithmeti latti e. Sine the pro jetion map G → G/R is prop er, the statemen t of Theorem 5.1 implies in partiular that Q Z ( G/R ) is disrete. Corollary 5.2. L et G = G 1 × · · · × G n b e a pr o dut of lo  al ly  omp at gr oups. Assume that G admits a nitely gener ate d irr e duible latti e with a faithful Zariski-dense r epr esentation in a semi-simple gr oup over some eld of har ateristi 6 = 2 , 3 . Then G is a  omp at extension of a dir e t pr o dut of a semi-simple algebr ai gr oup by a disr ete gr oup.  T o b e more preise, the arithmetiit y onlusion of Theorem 5.1 means the follo wing. There exists a global eld K , a onneted semi-simple K -anisotropi K -group H and a nite set Σ of v aluations of K su h that: (i) The quotien t Γ := Γ / Γ ∩ ( R · Q Z ( G )) is ommensurable with the arithmeti group H ( K (Σ)) , where K (Σ) is the ring of Σ -in tegers of K . Moreo v er, Σ on tains all Ar himedean v aluations v for whi h H is K v -isotropi, where K v denotes the v -ompletion of K . In parti- ular, b y BorelHarish-Chandra and BehrHarder redution theory , the diagonal em b edding realises H ( K (Σ)) as a lattie in the pro dut Q v ∈ Σ H ( K v ) . (ii) The group G + is isomorphi to Q v ∈ Σ H ( K v ) + and this isomorphism implemen ts the ommensurabilit y of Γ with H ( K (Σ)) . F or ba kground referenes, inluding on H ( K v ) + , see [ Mar91 , I.3℄. In on trast to statemen ts in [ Mon05 ℄, there is no assumption on the subgroup struture of the fators G i in Theorem 5.1 , whi h ma y not ev en b e irreduible. The nature of the linear represen tation is ho w ev er more restrited. Another impro v emen t is that no (w eak) o ompatness assumption is made on Γ . In partiular, under the same algebrai restritions on the fators G i as in lo . it. , w e obtain the follo wing arithmetiit y vs. non-linearit y alternativ e for all nitely generated latties. Corollary 5.3. L et Γ < G = G 1 × · · · × G n b e an irr e duible nitely gener ate d latti e, wher e e ah G i is a lo  al ly  omp at gr oup suh that every non-trivial lose d normal sub gr oup is  o  omp at. Then one of the fol lowing holds: (i) Every nite-dimensional line ar r epr esentation of Γ in har ateristi 6 = 2 , 3 has vir- tual ly soluble image. (ii) G is a semi-simple algebr ai gr oup and Γ is an arithmeti latti e. The h yp othesis made on ea h fator G i ma y b e used to desrib e to some exten t its struture indep enden tly of the existene of a lattie in G ; one an in partiular sho w [ CM08a ℄ that ea h G i is monolithi, th us extending the lassial result of Wilson [ Wil71 ℄ to lo ally ompat groups. Ho w ev er, w e will not app eal to this preliminary desription of the G i when pro ving Corollary 5.3 : the strutural information will instead b e obtained a p osteriori . Remark 5.4. In [ Mon05 ℄, the onlusion ( i ) w as replaed b y niteness of the image. This follo ws from the urren t onlusion in the more restrited setting of lo . it. thanks to Y. Shalom's sup errigidit y for  haraters [ Sha00 ℄, unless of ourse G i admits (virtually) a non-zero on tin uous homomorphism to R (after all in the urren t setting w e an ha v e G i = R ). It is part of the assumptions in [ Mon05 ℄ that no su h homomorphism exists, so that Corollary 5.3 indeed generalises lo . it . 26 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD 5.A. Sup errigid pairs. F or on v eniene, w e shall use the follo wing terminology . Let J b e a top ologial group and Λ < J an y subgroup. W e all the pair (Λ , J ) sup errigid if for an y lo al eld k and an y onneted absolutely almost simple adjoin t k -group H , ev ery abstrat homomorphism Λ → H ( k ) with un b ounded Zariski-dense image extends to a on tin uous homomorphism of J . Prop osition 5.5. L et (Λ , J ) b e a sup errigid p air with J lo  al ly  omp at and Λ nitely gener ate d with losur e of nite  ovolume in J . If Λ admits a faithful r epr esentation in an admissible semi-simple gr oup (over any eld) with Zariski-dense image, then the amenable r adi al R of J is  omp at and the quasi- entr e Q Z ( J ) is virtual ly  ontaine d in Λ · R . F urthermor e, up on r eplaing J by a nite index sub gr oup, the quotient J /R splits as J + × Q Z ( J /R ) wher e J + is a semi-simple algebr ai gr oup and the image of Λ in J + is an arithmeti latti e. (W e p oin t out again that in partiular the diret fator Q Z ( J /R ) is disrete.) One migh t exp et that Theorem 5.1 ould no w b e pro v ed b y establishing in omplete generalit y that nitely generated irreduible latties in pro duts of lo ally ompat groups form a sup errigid pair. F or uniform latties, or more generally w eakly o ompat square- summable latties, this is indeed true and w as pro v ed in [ Mon06 ℄. W e do not ha v e a pro of in general and shall es hew this diult y b y giving rst an indep enden t pro of of the ompatness of the amenable radial (Corollary 5.14 b elo w) and using the residual niteness of nitely generated linear groups b efore pro eeding with Prop osition 5.5 . Nev ertheless, w e do ha v e a general pro of as so on as the groups are totally disonneted. Theorem 5.6. L et Γ < G = G 1 × · · · × G n b e an irr e duible nitely gener ate d latti e, wher e e ah G i is any lo  al ly  omp at gr oup. If G is total ly dis onne te d, then (Γ , G ) is a sup errigid p air. (As w e shall see in Prop osition 5.11 , one an drop the nite generation assumption in the simpler ase where Γ pro jets faithfully to some fator G i .) The pro ofs will use the follo wing fat established in [ CM08a ℄. Prop osition 5.7. L et G b e a  omp atly gener ate d lo  al ly  omp at gr oup and { N v | v ∈ Σ } b e a  ol le tion of p airwise distint lose d normal sub gr oups of G suh that for e ah v ∈ Σ , the quotient H v = G/ N v is quasi-simple, non-disr ete and non- omp at. If T v ∈ Σ N v = 1 then Σ is nite and G has a har ateristi lose d  o  omp at sub gr oup whih splits as a nite dir e t pr o dut of | Σ | top olo gi al ly simple gr oups.  W e reall for the ab o v e statemen t that a group is alled quasi-simple if it p ossesses a o ompat normal subgroup whi h is top ologially simple and on tained in ev ery non-trivial losed normal subgroup. Pr o of of Pr op osition 5.5 . W e will largely follo w the ideas of Margulis, deduing arithmeti- it y from sup errigidit y [ Mar91 , Chapter IX℄. It is assumed that the reader has a op y of [ Mon05 ℄ at hand, sine it on tains a similar reasoning under dieren t h yp otheses. The  harateristi assumption in lo . it. will b e replaed b y the urren t admissibilit y assump- tion. The group J (and hene also all nite index subgroups and fators) is ompatly generated b y Lemma 2.12 . Let τ : Λ → H b e the giv en faithful represen tation. Up on replaing Λ and J b y nite index subgroups and p ost-omp osing τ with the pro jetion map H → H / Z ( H ) , w e shall assume heneforth that H is adjoin t and Zariski-onneted. The represen tation ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 27 τ : Λ → H need no longer b e faithful, but it still has nite k ernel. As in [ Mon05 , (3.3)℄, in view of the assumption that Λ is nitely generated, w e ma y assume that H is dened o v er a nitely generated eld K . This is the rst of t w o plaes where the admissibilit y assumption is used in lo . it. (follo wing VI I I.3.22 and IX.1.8 in [ Mar91 ℄). By Tits' alternativ e [ Tit72 ℄, the amenable radial of Λ is soluble-b y-lo ally-nite and th us lo ally nite sine τ (Λ) is Zariski-dense and H is semi-simple. The nite generation of K implies that this radial is in fat nite (see e.g. Corollary 4.8 in [ W eh73 ℄), th us trivial b y Zariski-densit y sine H is adjoin t. (The nite generation of K is essen tial in p ositiv e  harateristi sine for algebraially losed elds there is alw a ys a lo ally nite Zariski-dense subgroup [ BGM04 ℄.) It no w follo ws that if J is a ompat extension of a disrete group, then the latter has trivial amenable radial and th us all the onlusions of Prop osition 5.5 hold trivially . Therefore, w e assume heneforw ard that J is not ompat-b y-disrete. Let H = H 1 × · · · × H k b e the deomp osition of H in to its simple fators. W e shall w ork with the fators H i one at a time. Let τ i : Λ → H i b e the indued represen tation of Λ . Notie that τ i need not b e faithful; ho w ev er, it has Zariski-dense (and in partiular innite) image. W e let Σ i denote the set of all (inequiv alen t represen tativ es of ) v aluations v of K su h that the image of τ i (Λ) is not relativ ely ompat in H i ( K v ) (for the Hausdor top ology); observ e that this image is still Zariski-dense. Then Σ i is non-empt y sine τ i (Λ) is innite, see [ BG07 , Lemma 2.1℄. By h yp othesis, there exists a on tin uous represen tation J → H i ( K v ) for ea h v ∈ Σ i , extending the giv en Λ -represen tation. W e denote b y N v ✁ J the k ernel of this represen tation. Let I ⊆ { 1 , . . . , k } b e the set of all those indies i su h that J / N v is non-disrete for ea h v ∈ Σ i . W e laim that the set I is non-empty. Indeed, for ea h index j 6∈ I , there exists v j ∈ Σ j su h that N v j is op en in J . Th us the k ernel J + = \ j 6∈ I N v j of the on tin uous represen tation J → Q j 6∈ I H j ( K v j ) is op en. By assumption the losure of Λ in J has nite o v olume. Therefore, for ea h op en subgroup F < J , the losure of Λ ∩ F has nite o v olume in F . It follo ws in partiular that Λ ∩ F is innite unless F is ompat. These onsiderations apply to the op en subgroup J + < J . Sine J is not ompat-b y- disrete, w e dedue that Λ ∩ J + is innite. Therefore the restrition to Λ of the represen tation J → Q j 6∈ I H j ( K v j ) has innite k ernel and, hene, it do es not fator through τ : Λ → H ( K ) . In partiular it annot oinide with the giv en represen tation τ : Λ → H . Th us I is non- empt y . W e laim that for e ah i ∈ I , the set Σ i is nite. Let i ∈ I and v ∈ Σ i . The argumen ts of [ Mon05 , (3.7)℄ sho w that the isomorphi image of J / N v in H i ( K v ) on tains H i ( K v ) + . These argumen ts use again the admissibilit y assumption b eause the app eal to a result of R. Pink [ Pin98 ℄; the fat that the latter hold in the admissible ase is expliit in the table pro vided in Prop osition 1.6 of [ Pin98 ℄. F urthermore, it follo ws from Tits' simpliit y theorem [ Tit64 ℄ om bined with [ BT73 , 6.14℄ that ea h J / N v is quasi-simple. Moreo v er, an appliation of [ BT73 , 8.13℄ sho ws that the v arious quotien ts 28 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD ( J / N v ) v ∈ Σ i are pairwise non-isomorphi. In partiular the normal subgroups ( N v ) v ∈ Σ i are pairwise distint. Let D i = T v ∈ Σ i N v and reall that J /D i is ompatly generated. Pro jeting ea h N v to J /D i , w e obtain a family of pairwise distint normal subgroups of J /D i indexed b y Σ i su h that ea h orresp onding quotien t is quasi-simple, non-disrete and non-ompat. Therefore, the desired laim follo ws from Prop osition 5.7 . In partiular, app ealing again to [ BT73 , Corollaire 8.13℄, w e obtain a on tin uous map J → Q v ∈ Σ i H i ( K v ) whi h w e denote again b y τ J i . The k ernel of τ J i is D i . Up on replaing J and Λ b y nite index subgroups w e ma y assume that the image of τ J i oinides in fat with Q v ∈ Σ i H i ( K v ) + , ompare [ Mon05 , (3.9)℄. W e laim that R := J + ∩ D is  omp at and that J = J + · D , wher e D is dene d by D = T i ∈ I D i . W e rst sho w that R = J + ∩ D is ompat. Assume for a on tradition that this is not the ase. Then, giv en a ompat op en subgroup U of J , the in tersetion Λ 0 = Λ ∩ ( U · R ) is innite: this follo ws from the same argumen t as ab o v e, using the assumption that the losure of Λ has nite o v olume. F or ea h index j 6∈ I , w e ha v e J + ⊆ N v j and w e dedue that the image of Λ 0 in H j ( K v j ) is nite, sine it is on tained in the image of U . Equiv alen tly , the subgroup τ j (Λ 0 ) < H j ( K ) is nite. It follo ws in partiular that τ i (Λ 0 ) is innite for some i ∈ I . By [ BG07 , Lemma 2.1℄, there exists v ∈ Σ i su h that the image of Λ 0 in H i ( K v ) is un b ounded. This is absurd sine D ⊆ N v and hene the image of Λ 0 in H i ( K v ) is on tained in the image of the ompat subgroup U . This sho ws that the in tersetion R is indeed ompat. A t this p oin t w e kno w that the quotien t J /D is isomorphi to a subgroup of the pro dut Y i ∈ I Y v ∈ Σ i H i ( K v ) + whi h pro jets surjetiv ely on to ea h fator of the form Q v ∈ Σ i H i ( K v ) + . Using again the Goursat-t yp e argumen t as in Prop osition 5.7 , w e nd that J /D is indeed isomorphi to a nite pro dut of non-ompat non-disrete simple groups H i ( K v ) + . In partiular the quotien t J /D has no non-trivial op en normal subgroup. Sine J + is op en and normal in J , w e dedue that J = J + · D , thereb y establishing the laim. By the v ery nature of the statemen t, w e ma y replae J b y the quotien t J /R without an y loss of generalit y , sine R is ompat. In view of this further simpliation, the preeding laim implies that J ∼ = J + × D . In partiular D is disrete. It no w follo ws as in [ Mon05 , (3.11)℄ that K is a global eld, and that the image of Λ in the semi-simple group J /D is an arithmeti lattie (ompare [ Mon05 , (3.13)℄). Therefore, b y Prop osition 3.1 , the in tersetion Λ ∩ D is a lattie in D and, hene, the disrete normal subgroup D is virtually on tained in Λ . As J ∼ = J + × D and J + has trivial quasi-en tre, it follo ws that the quasi-en tre of J oinides with D . This nishes the pro of.  F or later use, w e single out a (simpler) v ersion of an argumen t referred to ab o v e. Lemma 5.8. L et H b e an admissible  onne te d absolutely almost simple adjoint k -gr oup H , wher e k is a lo  al eld. L et J b e a lo  al ly  omp at gr oup with a  ontinuous unb ounde d Zariski-dense homomorphism τ : J → H ( k ) . Then any  omp at normal sub gr oup of J is  ontaine d in the kernel of τ . ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 29 Pr o of. Let K ✁ J b e a ompat normal subgroup. The Zariski losure of τ ( K ) is normalised b y the Zariski-dense group τ ( J ) and therefore it is either H ( k ) or trivial. W e assume the former sine otherwise w e are done. W e laim that w e an assume k non-Ar himedean. Otherwise, either k = R or k = C . In the rst ase, τ ( K ) oinides with its Zariski losure b y W eyl's algebraiit y theorem [ Vin94 , 4.2.1℄ so that H ( k ) is ompat in whi h ase the lemma is v oid b y the un b oundedness assumption. In the seond ase, one an redue to the ase τ ( J ) ⊆ H ( R ) as in [ Mon05 , (3.5)℄ and th us τ ( K ) = 1 as b efore sine H ( R ) is also simple; the laim is pro v ed. F ollo wing no w an idea from [ Sha00 , p. 41℄ (see also the explanations in Setion (3.7) of [ Mon05 ℄), one uses [ Pin98 ℄ to dedue that τ ( K ) is op en up on p ossibly replaing k b y a losed subeld (the admissibilit y assumption en ters as in the pro of of Prop osition 5.5 ). W e an still denote this subeld b y k b eause it aommo dates the whole image τ ( J ) , see again [ Mon05 , (3.7)℄. No w τ ( J ) is an un b ounded op en subgroup and hene on tains H ( k ) + b y a result of J. Tits (see [ Pra82 ℄; this also follo ws from the Ho w eMo ore theorem [ HM79 ℄ whi h ho w ev er is p osterior to Tits' result). This implies that the ompat group τ ( K ) is trivial sine H ( k ) + is simple b y [ Tit64 ℄.  5.B. Boundary maps. W e reord t w o statemen ts extrated from Margulis' w ork in the form most on v enien t for us. Prop osition 5.9. L et J b e a se  ond  ountable lo  al ly  omp at gr oup with a me asur e lass pr eserving ation on a standar d pr ob ability sp a e B . L et Λ < J b e a dense sub gr oup with a Zariski-dense unb ounde d r epr esentation τ : Γ → H ( k ) to a  onne te d absolutely almost k -simple adjoint gr oup H over an arbitr ary lo  al eld k . If ther e is a pr op er k -sub gr oup L < H and a Λ -e quivariant non-essential ly- onstant me a- sur able map B → H ( k ) / L ( k ) , then τ extends to a  ontinuous homomorphism J → H ( k ) . Pr o of. The argumen t is giv en b y A'Camp oBurger in the  harateristi zero ase at the end of Setion 7 in [ AB94 ℄ (pp. 1819). This referene onsiders homogeneous spaes for B but this restrition is nev er used. The general statemen t is referred to in [ Bur95 ℄ and details are giv en in [ Bon04 ℄.  Prop osition 5.10. L et Γ b e a  ountable gr oup with a Zariski-dense unb ounde d r epr esenta- tion Γ → H ( k ) to a  onne te d absolutely almost k -simple adjoint gr oup H over an arbitr ary lo  al eld k . L et B b e a standar d pr ob ability sp a e with a me asur e lass pr eserving Γ -ation that is amenable in Zimmer's sense [ Zim84 ℄ and suh that the diagonal ation on B 2 is er go di. Then ther e is a pr op er k -sub gr oup L < H and a Γ -e quivariant non-essential ly- onstant me asur able map B → H ( k ) / L ( k ) . Pr o of. Again, this is pro v ed in [ AB94 ℄ for the  harateristi zero ase (and B homogeneous) and the neessary adaptations to the general ase are explained in [ Bon04 ℄.  W e shall need these sp ei statemen ts b elo w. They rst app eared within the pro of of Margulis' ommensurator sup errigidit y , whi h an adapted as follo ws using [ Bur95 ℄ and Lemma 8.3 in [ CM08 ℄, pro viding a rst step to w ards Theorem 5.6 . Prop osition 5.11. L et G = G 1 × G 2 b e a pr o dut of lo  al ly  omp at σ - omp at gr oups and Λ < G b e an irr e duible latti e. Assume that the pr oje tion of Λ to G 1 is inje tive and that G 2 admits a  omp at op en sub gr oup. Then the p air (Λ , G ) is sup errigid. Pr o of. W e laim that one an assume G seond oun table. As explained in [ Mon06 , Prop osi- tion 61℄, σ -ompatness implies the existene of a ompat normal subgroup K ✁ G meeting 30 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD Λ trivially and su h that G/K is seond oun table. Applying the statemen t to G/K to- gether with the image of Λ therein yields the general statemen t sine the pro jetion of Λ to G/K is an isomorphism; this pro v es the laim. Let τ : Λ → H ( k ) b e as in the denition of sup errigid pairs and let U < G 2 b e a ompat op en subgroup. Set Λ U = Λ ∩ ( G 1 × U ) . By the injetivit y assumption and Lemma 3.2 , w e an onsider Λ U as a lattie in G 1 whi h is ommensurated b y (the image of the pro jetion of ) Λ . W e distinguish t w o ases. Assume rst that τ (Λ U ) is un b ounded in the lo ally ompat group H ( k ) . W e ma y then apply Margulis' ommensurator sup errigidit y in its general form prop osed b y M. Burger [ Bur95 , Theorem 2.A℄, see [ Bon04 ℄ for details. This yields a on tin uous map J → H ( k ) fatoring through G 1 and extending the giv en Λ -represen tation, as desired. Assume no w that τ (Λ U ) is b ounded, whi h is equiv alen t to Λ U xing a p oin t in the symmetri spae or BruhatTits building asso iated to H ( k ) . Then Lemma 8.3 in [ CM08 ℄ yields a on tin uous map J → H ( k ) fatoring through G 2 .  5.C. Radial sup errigidit y. Theorem 5.12. L et G b e a lo  al ly  omp at gr oup, R ✁ G its amenable r adi al, Γ < G a nitely gener ate d latti e and F the losur e of the image of Γ in G/R . Then any Zariski-dense unb ounde d r epr esentation of Γ in any  onne te d absolutely almost simple adjoint k -gr oup H over any lo  al eld k arises fr om a  ontinuous r epr esentation of F via the map Γ → F . (In partiular, the pair (Γ / (Γ ∩ R ) , F ) is sup errigid.) Pr o of. Notie that G is σ -ompat sine it on tains a nitely generated, hene oun table, lattie. (In fat G is ev en ompatly generated b y Lemma 2.12 .) Set J = G/R . There exists a standard probabilit y J -spae B on whi h the Γ -ation is amenable and su h that the diagonal Γ -ation on B 2 is ergo di; it sues to  ho ose B to b e the P oisson b oundary of a symmetri random w alk with full supp ort on J . Indeed: (i) The J -ation is amenable as w as sho wn b y Zimmer [ Zim78 ℄; this implies that the G -ation is amenable sine R is an amenable group and th us that the Γ -ation is amenable sine Γ is losed in G (see [ Zim84 , 5.3.5℄). (ii) The diagonal ation of an y losed nite o v olume subgroup F < J on B 2 is ergo di in view of the er go diity with  o eients of J , and hene the same holds for dense subgroups of F . F or detailed ba kground on this strengthening of ergo diit y in tro dued in [ BM02 ℄ and on the P oisson b oundary in general, w e refer the reader to [ Ka  03 ℄. Let no w k b e a lo al eld, H a onneted absolutely almost simple k -group and Γ → H ( k ) a Zariski-dense un b ounded represen tation. W e an apply Prop osition 5.10 and obtain a prop er subgroup L < H and a Γ -equiv arian t map B → H ( k ) / L ( k ) . W riting Λ for the image of Γ in J , w e an therefore apply Prop osition 5.9 with F instead of J and the onlusion follo ws.  Remark 5.13. An examination of this pro of sho ws that one has also the follo wing related result. Let J b e a seond oun table lo ally ompat group and Λ ⊆ J a dense oun table subgroup whose ation on J b y left m ultipliation is amenable. Then the pair (Λ , J ) is sup errigid. Indeed, one an again argue with Prop ositions 5.9 and 5.10 b eause it is easy to  he k that in the presen t situation an y amenable J -spae is also amenable for Λ view ed as a disrete group. Related ideas w ere used b y R. Zimmer in [ Zim87 ℄. Corollary 5.14. L et G b e a lo  al ly  omp at gr oup and Γ < G a nitely gener ate d latti e. If Γ admits a faithful Zariski-dense r epr esentation in an admissible semi-simple gr oup (over any eld), then the amenable r adi al of G is  omp at. ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 31 Pr o of. Let R b e the amenable radial of G , F b e the losure of the image of Γ in G/R and J < G the preimage of F in G . The on ten t of Theorem 5.12 is that the pair (Γ , J ) is sup errigid. Sine in addition Γ is losed and of nite o v olume in J (see [ Rag72 , Lemma 1.6℄), w e ma y apply Prop osition 5.5 and dedue that the amenable radial of J is ompat. The onlusion follo ws sine R < J .  5.D. Latties with non-disrete ommensurators. The follo wing useful tri k allo ws to realise the ommensurator of an y lattie in a lo ally ompat group G as a lattie in a pro dut G × D . A similar reasoning in the sp eial ase of automorphism groups of trees ma y b e found in [ BG02 , Theorem 6.6℄. Lemma 5.15. L et Λ b e a gr oup and Γ < Λ a sub gr oup  ommensur ate d by Λ . L et D b e the  ompletion of Λ with r esp e t to the left or right uniform strutur e gener ate d by the Λ - onjugates of Γ . Then D is a total ly dis onne te d lo  al ly  omp at gr oup. If furthermor e G is a lo  al ly  omp at gr oup  ontaining Λ as a dense sub gr oup suh that Γ is disr ete (r esp. is a latti e) in G , then the diagonal emb e dding of Λ in G × D is disr ete (r esp. is an irr e duible latti e). The ab o v e lemma is in some sense a on v erse to Lemma 3.2 . In the sp eial ase where one starts with a lattie satisfying a faithfulness ondition, this relation b eomes ev en stronger. Lemma 5.16. L et G, H b e lo  al ly  omp at gr oups and Λ < G × H a latti e. Assume that the pr oje tion of Λ to G is faithful and that b oth pr oje tions ar e dense. L et U < H b e a  omp at op en sub gr oup, set Γ = Λ ∩ ( G × U ) as in L emma 3.2 and  onsider the gr oup D as in L emma 5.15 (up on viewing Λ as a sub gr oup of G ). Dene the  omp at normal sub gr oup K ✁ H as the  or e K = T h ∈ H hU h − 1 of U in G . Then the map Λ → D indu es an isomorphism of top olo gi al gr oups H /K ∼ = D . Pr o of of L emma 5.15 . One v eries readily the ondition giv en in [ Bou60 ℄ (TG I I I,  3, No 4, Théorème 1) ensuring that the ompletion satises the axioms of a group top ology . W e emphasise that it is part of the denition of the ompletion that D is Hausdor; in other w ords D is obtained b y rst ompleting Λ with resp et to the group top ology as dened ab o v e, and then dividing out the normal subgroup onsisting of those elemen ts whi h are not separated from the iden tit y . Let U denote the losure of the pro jetion of Γ to D . By denition U is op en. Notie that it is ompat sine it is a quotien t of the pronite ompletion of Γ b y onstrution. In partiular D is lo ally ompat. By a sligh t abuse of notation, let us iden tify Γ and Λ with their images in D . W e laim that U ∩ Λ = Γ . Indeed, let { γ n } n ≥ 0 b e a sequene of elemen ts of Γ su h that lim n γ n = λ ∈ Λ . Sine λ Γ λ − 1 is a neigh b ourho o d of the iden tit y in Λ (with resp et to the top ology indued from D ), it follo ws that γ n λ − 1 ∈ λ Γ λ − 1 for n large enough. Th us λ ∈ γ n Γ = Γ . Assume no w that Γ is disrete and  ho ose a neigh b ourho o d V of the iden tit y in G su h that Γ ∩ V = 1 . In view of the preeding laim the pro dut V × U is a neigh b ourho o d of the iden tit y in G × D whi h meets Λ trivially , thereb y sho wing that Λ is disrete. Assume nally that Γ is a lattie in G and let F b e a fundamen tal domain. Then F × U is a fundamen tal domain for Λ in G × D , whi h has nite v olume sine a Haar measure for G × D ma y b e obtained b y taking the pro dut of resp etiv e Haar measures for G and D . Th us Λ has nite o v olume in G × D .  Pr o of of L emma 5.16 . In order to onstrut a on tin uous homomorphism π : H → D , it sues to  he k that an y net in Λ whose image in H on v erges to the iden tit y also on v erges 32 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD to the iden tit y in D ; this follo ws from the denitions of Γ and D sine the net is ev en tually in an y Λ -onjugate of U . Notie that π has dense image. W e laim that the k ernel of π is T λ ∈ Λ λU λ − 1 . Indeed, if on the one hand k ∈ ke r( π ) is the limit of the images in H of a net { λ i } in Λ , then for an y λ w e ha v e ev en tually λ i ∈ λ − 1 Γ λ ⊆ λ − 1 ( G × U ) λ so that indeed k ∈ λ − 1 U λ sine U is losed. Con v ersely , if k ∈ T λ ∈ Λ λU λ − 1 is limit of images of { λ i } , then, sine U is op en, for an y λ the image of λ i is ev en tually in λU λ − 1 , hene in λ Γ λ − 1 so that π ( λ i ) → 1 . This pro v es the laim. No w it follo ws that k er( π ) is indeed the ore K of the statemen t sine U is ompat. The fat that π is on to and op en follo ws from the existene of a ompat op en subgroup in H .  Theorem 5.17. L et G b e a lo  al ly  omp at gr oup and Γ < G b e a latti e. Assume that G p ossesses a nitely gener ate d dense sub gr oup Λ suh that Γ < Λ < Comm G (Γ) . If Λ admits a faithful Zariski-dense r epr esentation in an admissible semi-simple gr oup (over any eld), then the amenable r adi al R of G is  omp at and the quasi- entr e Q Z ( G ) is virtual ly  ontaine d in Γ · R . F urthermor e, up on r eplaing G by a nite index sub gr oup, the quotient G/R splits as G + × Q Z ( G/R ) wher e G + is a semi-simple algebr ai gr oup and the image of Γ in G + is an arithmeti latti e. Pr o of. Let J = G × D , where D is the totally disonneted lo ally ompat group pro vided b y Lemma 5.15 . As a totally disonneted group, it has n umerous ompat op en subgroups (for instane the losure of Γ ). W e shall view Λ as an irreduible lattie in J . The pro jetion of Λ to G is faithful b y onstrution. By Prop osition 5.11 , the pair (Λ , J ) is sup errigid. This allo ws us to apply Prop osition 5.5 . Sine the amenable radial R G of G is on tained in the amenable radial R J of J , it is ompat. F urthermore, the quasi-en tre of G is on tained in the quasi-en tre of J and the en tre-free group G/R G is a diret fator of J + × Q Z ( J /R J ) ; the desired onlusions follo w.  5.E. Latties in pro duts of Lie and totally disonneted groups. Theorem 5.18. L et Γ < G = S × D b e a nitely gener ate d irr e duible latti e, wher e S is a  onne te d semi-simple Lie gr oup with trivial  entr e and D is a total ly dis onne te d lo  al ly  omp at gr oup. L et Γ D ✁ D b e the  anoni al disr ete kernel of D . Then D / Γ D is a pr onite extension of a semi-simple algebr ai gr oup Q and the image of Γ in S × Q , whih is isomorphi to Γ / Γ D , is an arithmeti latti e. Corollary 5.19. In p artiular, D is lo  al ly pr onite by analyti.  A family of examples will b e onstruted in Setion 6.C b elo w, sho wing that the statemen t annot b e simplied ev en in a geometri setting (see Remark 6.7 ). Pr o of of The or em 5.18 . By the v ery nature of the statemen t, w e an fator out the anonial disrete k ernel. Therefore, w e shall assume heneforth that the pro jetion map Γ → S is injetiv e. W e an also assume that S has no ompat fators. Sine S is onneted with trivial en tre, there is a Zariski onneted semi-simple adjoin t R -group H without R - anisotropi fators su h that S = H ( R ) . Notie that the injetivit y of Γ → S is preserv ed when passing to nite index subgroups. By Prop osition 5.11 , the pair (Γ , G ) is sup errigid. W e an therefore apply Prop osition 5.5 . In partiular, D has ompat amenable radial and therefore, in view of the statemen t of Theorem 5.18 , w e an assume that this radial is trivial. Giv en the onlusion of Prop osi- tion 5.5 , it only remains to sho w that the quasi-en tre Q Z ( G ) of G is trivial. W e no w kno w that Q Z ( G ) is virtually on tained in Γ ; sine on the other hand S has trivial quasi-en tre, ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 33 Q Z ( G ) ⊆ 1 × D . In other w ords, Q Z ( G ) is on tained in the disrete k ernel Γ D , whi h has b een rendered trivial. This ompletes the pro of.  W e ha v e treated Theorem 5.18 as a p ort of all on the w a y to Theorem 5.1 . In fat, one an also desrib e latties in pro duts of groups with a simple algebrai fator o v er an arbitrary lo al eld and in most ases without assuming nite generation a priori . W e reord the follo wing statemen t, whi h will not b e used b elo w. Theorem 5.20. L et k b e any lo  al eld and G an admissible  onne te d absolutely almost simple adjoint k -gr oup. L et H b e any  omp atly gener ate d lo  al ly  omp at gr oup admitting a  omp at op en sub gr oup. L et Γ < G ( k ) × H b e an irr e duible latti e. In  ase k has p ositive har ateristi and the k -r ank of G is one, we assume Γ  o  omp at. Then H / Γ H is a  omp at extension of a semi-simple algebr ai gr oup Q and the image of Γ in G ( k ) × Q is an arithmeti latti e. There is no assumption whatso ev er on the ompatly generated lo ally ompat group H b ey ond admitting a ompat op en subgroup; reall that the latter is automati if H is totally disonneted [ Bou71 , I I I  4 No 6℄. Notie that a p osteriori it follo ws from arithmetiit y that Γ is nitely generated; in the pro of b elo w, nite generation will b e established in t w o steps. Pr o of of The or em 5.20 . W e fator out the anonial disrete k ernel Γ H and assume hene- forth that it is trivial. This do es not aet the other assumptions and th us w e  ho ose some ompat op en subgroup U < H . W e write G = G ( k ) and onsider Γ U = Γ ∩ ( G × U ) as in Lemma 3.2 . Sine w e fatored out the anonial disrete k ernel, w e an onsider Γ U as a lattie in G ommensurated b y the dense subgroup Γ < G . Moreo v er, Γ U is nitely generated; indeed, either w e ha v e sim ultaneously rank k ( G ) = 1 and c har( k ) > 0 , in whi h ase w e assumed Γ o ompat, so that Γ U is o ompat in the ompatly generated group G ( k ) (again Lemma 3.2 ) and hene nitely generated [ Mar91 , I.0.40℄; or else, Γ U is kno wn to b e nitely generated b y applying, as the ase ma y b e, either Kazhdan's prop ert y , or the theory of fundamen tal domains, or the o ompatness of p -adi latties  w e refer to Margulis, Setions (3.1) and (3.2) of Chapter IX in [ Mar91 ℄. W e an no w apply Margulis' arthmetiit y [ Mar91 , 1.(1)℄ and dedue that G is dened o v er a global eld K and that Γ U is ommensurable to G ( K ( S )) for some nite set of plaes S ; in short Γ U is S -arithmeti. (The idea to obtain rst this preliminary arithmetiit y of Γ U w as suggested b y M. Burger.) It follo ws that Γ is rational o v er the global eld K , see Theorem 3.b in [ Bor66 ℄ ( lo . it. is form ulated for the Lie group ase; see [ W or07 , Lemma 7.3℄ in general). Sine the pair (Γ , G × H ) is sup errigid (for instane b y Prop osition 5.11 ), only the a priori la k of nite generation for Γ prev en ts us from applying Prop osition 5.5 . Ho w ev er, a go o d part of the pro of of that prop osition is already seured here sine Γ has b een sho wn to b e rational o v er a global eld. W e no w pro eed to explain ho w to adapt the remaining part of that pro of to the urren t setting. W e use those elemen ts of notation in tro dued in the pro of of Prop osition 5.5 that do not onit with presen t notation and review all uses of nite generation that are either expliit in the pro of of Prop osition 5.5 or impliit through referenes to [ Mon05 ℄. The ompat generation of G × H is an assumption rather than a onsequene of Lemma 2.12 . W e also used nite generation in order to pass to a nite index subgroup of Γ on tained in G ( K v ) + for all v aluations v ∈ Σ . W e shall p ostp one this step, so that the whole argu- men tation pro vides us with maps from G × H to a pro dut Q of fators that lie in-b et w een 34 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD G ( K v ) + and G ( K v ) . In partiular all these fators are quasi-simple and w e an still app eal to Prop osition 5.7 as b efore. Notie ho w ev er that at the v ery end of the pro of, one nite generation is gran ted, w e an in v ok e the argumen t that G ( K v ) / G ( K v ) + is virtually torsion Ab elian [ BT73 , 6.14℄ and th us redue again to the ase where Γ is on tained in G ( K v ) + . W e no w justify that the image of Γ in G × Q is disrete b eause previously this follo w ed from [ Mon05 , (3.13)℄ whi h relies on nite generation. If Γ w ere not disrete, an appliation of [ BG07 , Lemma 2.1℄ w ould pro vide a v aluation v / ∈ Σ with Γ un b ounded in G ( K v ) , whi h is absurd. W e are no w in a situation where G × H maps to G × Q with o ompat nite o v olume image and injetiv ely on Γ ; therefore the disreteness of the image of Γ implies that this map is prop er and hene H is a ompat extension of Q . Pushing forw ard the measure on ( G × H ) / Γ , w e see that the image of Γ in G × Q is a lattie. No w Γ is nitely generated (see ab o v e referenes to [ Mar91 , IX℄) and th us the pro of is ompleted as in Prop osition 5.5 . The disrete fator o urring in the onlusion of the latter prop osition is trivial for the same reason as in the pro of of Theorem 5.18 .  5.F. Latties in general pro duts. W e b egin with the sp eial ase of totally disonneted groups. Pr o of of The or em 5.6 . An issue that w e need to deal with is that the pro jetion of Γ to G 1 is a priori not faithful. In order to irum v en t this diult y , w e pro eed to a preliminary onstrution. Let ι : Γ → b Γ b e the anonial map to the pronite ompletion of Γ and denote its k ernel b y Γ (f ) ; in other w ords, Γ (f ) is the nite residual of Γ . Let b G 1 denote the lo ally ompat group whi h is dened as the losure of the image of Γ in G 1 × b Γ under the pro dut map pro j 1 × ι , where pro j 1 : G → G 1 is the anonial pro jetion. Sine pro j 1 (Γ) is dense in G 1 and b Γ is ompat, the anonial map b G 1 → G 1 is surjetiv e. In other w ords, the group b G 1 is a ompat extension of G 1 . W e no w dene G ′ 1 = G 2 × · · · × G n and b G = b G 1 × G ′ 1 . Then Γ admits a diagonal em b edding in to b G through whi h the injetion of Γ in G fators. W e will heneforth iden tify Γ with its image in b G and onsider Γ as an irreduible lattie of b G . W e laim that the p air (Γ , b G ) is sup errigid . The argumen t is a v ariation on the pro of of Prop osition 5.11 . Let τ : Γ → H ( k ) b e as in the denition of sup errigid pairs. Sine τ (Γ) is nitely generated and linear, it is residually nite [ Mal40 ℄. This means that τ fators through Γ := Γ / Γ (f ) . Let U < G ′ 1 b e a ompat op en subgroup, Γ U = Γ ∩ ( b G 1 × U ) and Γ U = Γ U / (Γ U ∩ Γ (f ) ) . By onstrution and Lemma 3.2 , w e an onsider Γ U as a lattie in b G 1 ommensurated b y Γ . Arguing as in Prop osition 5.11 , when τ (Γ U ) is un b ounded one applies ommensurator sup errigidit y yielding a on tin uous map J → H ( k ) and extending the map Γ → H ( k ) and hene also τ . When τ (Γ U ) is b ounded, one applies Lemma 8.3 in [ CM08 ℄ instead and the resulting extension fators through G ′ 1 . This pro v es the laim. In order to onlude that the pair (Γ , G ) is also sup errigid, it no w sues to apply Lemma 5.8 .  Corollary 5.21. The or em 5.1 holds in the p artiular  ase of total ly dis onne te d gr oups. Pr o of. Theorem 5.6 pro vides the h yp othesis needed for Prop osition 5.5 .  W e no w turn to the general ase Γ < G = G 1 × · · · × G n of Theorem 5.1 . The main part of the remaining pro of will onsist of a areful analysis of ho w the lattie Γ migh t sit ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 35 in v arious subpro duts hidden in the fators G i or their nite index subgroups one the amenable radial has b een trivialised. It will turn out that Γ is virtually a diret pro dut Γ ′ × Γ ′′ , where Γ ′ is an irreduible lattie in a pro dut S ′ × D ′ with S ′ a semi-simple Lie (virtual) subpro dut of G and D ′ a totally disonneted subgroup of G whose p osition will b e laried; as for Γ ′′ , it is an irreduible lattie in a semi-simple Lie group S ′′ that turns out to satisfy the assumptions of Margulis' arithmetiit y . Of ourse, an y of the ab o v e fators migh t w ell b e trivial. Completion of the pr o of of The or em 5.1 . The amenable radial is ompat b y Corollary 5.14 and hene w e an assume that it is trivial. The group G (and hene also all nite index subgroups and fators) is ompatly generated b y Lemma 2.12 . Up on regrouping the last n − 1 fators and in view of the denition of an irreduible lattie (see p. 18 ), w e an assume G = G 1 × G 2 . W e apply the solution to Hilb ert's fth problem (ompare Theorem 5.6 in [ CM08 ℄) and write G i = S i × D i after replaing G and Γ with nite index subgroups. Here S i are onneted semi-simple en tre-free Lie groups without ompat fators and D i totally disonneted ompatly generated with trivial amenable radial. Set S = S 1 × S 2 and D = D 1 × D 2 . Th us Γ is a lattie in G = S × D . Notie that if S is trivial, then G is totally disonneted and w e are done b y Theorem 5.6 . W e assume heneforth that S is non-trivial. The main remaining obstale is that the lattie Γ need not b e irreduible with resp et to the pro dut deomp osition G = S × D . Observe that the losur e pro j D (Γ) of the pr oje tion of Γ to D has trivial amenable r adi al. Indeed pro j D i (Γ) is dense in Γ i for i = 1 , 2 , hene the pro jetion pro j D (Γ) → D i has dense image. The desired laim follo ws sine G , and hene D i , has trivial amenable radial. Let U < D b e a ompat op en subgroup and set Γ U = Γ ∩ ( S × U ) . By Lemma 3.2 , the pro jetion pro j S (Γ U ) of Γ U to S is a lattie whi h is ommensurated b y pro j S (Γ) . The lattie pro j S (Γ U ) p ossesses a nite index subgroup whi h admits a anonial splitting in to nitely man y irreduible groups Γ 1 × · · · × Γ r , ompare Theorem 4.2 . F urthermore ea h Γ i is an irreduible lattie in a semi-simple subgroup S i < S whi h is obtained b y regrouping some of the simple fators of S . Sine the pro jetion of Γ to ea h G 1 and G 2 , and hene to S 1 and S 2 , is dense, it follo ws that the pro jetion of Γ to ea h simple fator of S is dense. W e no w onsider the pro jetion of Γ to the v arious fators S i . In view of the preeding remark and the fat that Γ i is an irreduible lattie in S i , it follo ws that pro j S i (Γ) is either dense in S i or disrete and on tains Γ i with nite index, see [ Mar91 , IX.2.7℄. Let no w S ′ = h S i | pro j S i (Γ) is non-disrete i and S ′′ = h S i | pro j S i (Γ) is disrete i . W e laim that the pr oje tion of Γ to S ′ is dense. If this failed, then b y [ Mar91 , IX.2.7℄ there w ould b e a subpro dut of some simple fators of S ′ on whi h the pro jetion of Γ is a lattie. Sine ea h Γ i is irreduible, this subpro dut is a regrouping S i 1 × · · · × S i p of some fators S i . No w the pro jetion of Γ is a lattie in this subgroup, hene it on tains the pro dut Γ i 1 × · · · × Γ i p with nite index and th us pro jets disretely to ea h S i j . This on tradits the denition of S ′ and pro v es the laim. Our next laim is that Γ has a nite index sub gr oup whih splits as Γ ′ × Γ ′′ , wher e Γ ′′ = pro j S ′′ (Γ) and Γ ′ is a latti e in S ′ × D . 36 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD In order to establish this, w e dene Γ ′ = Ker(pro j : Γ → S ′′ ) and Γ ′′ = \ γ ∈ Γ γ Γ U γ − 1 . Notie that Γ ′ and Γ ′′ are b oth normal subgroups of Γ . Sine pro j D (Γ ′′ ) is a ompat subgroup of D normalised b y pro j D (Γ) , whi h has trivial amenable radial, it follo ws that Γ ′′ ⊂ S ′ × S ′′ × 1 . Therefore, the in tersetion Γ ′ ∩ Γ ′′ is a normal subgroup of Γ on tained in S ′ × 1 × 1 . In view of the preeding laim, w e dedue that Γ ′ ∩ Γ ′′ = 1 . Th us h Γ ′ ∪ Γ ′′ i ✁ Γ is isomorphi to Γ ′ × Γ ′′ . Sine Γ ′ U = Γ ′ ∩ Γ U pro jets to a lattie in S ′ whi h omm utes with the pro jetion of Γ ′′ , w e dedue moreo v er that pro j S ′ (Γ ′′ ) = 1 , or equiv alen tly that Γ ′′ < 1 × S ′′ × 1 . Sine the pro jetion of Γ to S ′′ has disrete image b y denition, it follo ws from Prop o- sition 3.1 that Γ ′ < S ′ × 1 × D pro jets on to a lattie in S ′ × D . On the other hand, the v ery denition of S ′′ implies pro j S ′′ (Γ) on tains pro j S ′′ (Γ U ) , and hene also pro j S ′′ (Γ ′′ ) , as a nite index subgroup. In partiular, this sho ws that Γ ′ × Γ ′′ is a lattie in S ′ × S ′′ × D . Sine it is on tained in the lattie Γ , w e nally dedue that the index of Γ ′ × Γ ′′ in Γ is nite. W e observ e that w e ha v e in partiular obtained a lattie Γ ′′ < S ′′ with S ′′ non-simple and Γ ′′ irreduible (unless b oth Γ ′′ and S ′′ are trivial), b eause the pro jetion of Γ to an y simple Lie group fator is dense: indeed, an y simple fator m ust b e a fator of some G i and Γ pro jets densely on G i . It follo ws from Margulis' arithmetiit y theorem [ Mar91 , Theorem 1.(1')℄ that Γ ′′ is an arithmeti lattie in S ′′ . T urning to the other lattie, w e remark that Γ ′ admits a faithful Zariski-dense represen- tation in a semi-simple group, obtained b y reduing the giv en represen tation of Γ . F urther- more, notie that the pro jetion of Γ to S ′ oinides (virtually) with the pro jetion of Γ . In partiular it has dense image. Therefore, setting D ′ = pro j D (Γ ′ ) , w e ma y no w view Γ ′ as an irreduible lattie in S ′ × D ′ . W e ma y th us apply Theorem 5.18 . Notie that the same argumen t as b efore sho ws that D ′ has trivial amenable radial. W e laim that the  anoni al disr ete kernel Γ ′ D ′ is in fat a dir e t fator of D ′ . Indeed, sine Γ is residually nite b y Malev's theorem [ Mal40 ℄, Prop osition 4.9 ensures that Γ ′ D ′ en tralises the disrete residual D ′ ( ∞ ) . In partiular D ′ ( ∞ ) ∩ Γ ′ D ′ = 1 sine D ′ has trivial amenable radial. F urthermore, sine D ′ / Γ ′ D ′ is a semi-simple group, its disrete residual has nite index. In partiular, up on replaing D ′ b y a nite index subgroup w e ha v e D ′ ∼ = D ′ ( ∞ ) × Γ ′ D ′ as desired. It also follo ws that Γ ′ D ′ itself admits a Zariski-dense represen tation in a semi-simple group. It remains to onsider again the pro jetion maps pro j D i : D → D i . Restriting these maps to D ′ and using the fat that pro j D i ( D ′ ) is dense, w e obtain that D i ∼ = pro j D i ( D ′ ( ∞ ) ) × D ′ i , where D ′ i = pro j D i (Γ ′ D ′ ) . The nal onlusion follo ws b y applying Corollary 5.21 to the irreduible lattie Γ ′ D ′ < D ′ 1 × D ′ 2 .  Pr o of of Cor ol lary 5.3 . Sine Γ is nitely generated and irreduible, all G i are ompatly generated (alternativ ely , apply Lemma 2.12 ). W e laim that all pro jetions Γ → G i are injetiv e. Indeed, if not, then (b y indution on n ) there is j su h that the anonial disrete k ernel Γ G j is non-trivial. It is then o ompat, whi h implies that the pro jetion G j / Γ G j × Y i 6 = j G i − → Y i 6 = j G i ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 37 is prop er. This is on tradits the fat that the image of Γ in the left hand side ab o v e is disrete whilst it is dense in the righ t hand side, pro ving the laim. Supp ose giv en a linear represen tation of Γ in  harateristi 6 = 2 , 3 whose image is not virtually soluble. Arguing as in [ Mon05 ℄, w e an redue to the ase where w e ha v e a Zariski- dense represen tation τ : Γ → H ( K ) in a non-trivial onneted adjoin t absolutely simple group H o v er a nitely generated eld K . Sine τ (Γ) is innite, w e an  ho ose a ompletion k of K for whi h τ (Γ) is un b ounded [ BG07 , 2.1℄. P art of the argumen t in [ Mon05 ℄ is dev oted to pro ving that the represen tation is a p os- teriori faithful. One an adapt the en tire pro of to the presen t setting, but w e prop ose an alternativ e line of reasoning using an amenabilit y theorem from [ BS06 ℄. Supp ose to w ards a on tradition that the k ernel Γ 0 ✁ Γ of τ is non-trivial. Sine the pro jetions are in- jetiv e, the losure N i of the image of Γ 0 in G i is a non-trivial losed subgroup, whi h is normal b y irreduibilit y and hene is o ompat. Then Theorem 1.3 in [ BS06 ℄ implies that Γ / Γ 0 is amenable, on traditing the fat that τ (Γ) is not virtually soluble in view of Tits' alternativ e [ Tit72 ℄. A t this p oin t w e an onlude b y Theorem 5.1 .  6. Geometri arithmetiity 6.A. CA T(0) latties and parab oli isometries. W e no w sp eialise the v arious arith- metiit y results of Setion 5 to the ase of latties in CA T(0) spaes and om bine them with some of our geometri results. Reall that a parab oli isometry is alled neutral if it has zero translation length; the follo wing on tains Theorem 1.7 from the In tro dution. Theorem 6.1. L et X b e a pr op er CA T(0) sp a e with  o  omp at isometry gr oup and Γ < G := Is( X ) b e a nitely gener ate d latti e. Assume that Γ is irr e duible and that G  ontains a neutr al p ar ab oli isometry. Then one of the fol lowing assertions holds: (i) G is a non- omp at simple Lie gr oup of r ank one with trivial  entr e. (ii) Ther e is a sub gr oup Γ D ⊆ Γ normalise d by G , whih is either nite or innitely gener ate d and suh that the quotient Γ / Γ D is an arithmeti latti e in a pr o dut of semi-simple Lie and algebr ai gr oups. Pr o of. Let X ′ ⊆ X b e the anonial subspae pro vided b y Theorem 3.11 ; notie that X ′ still admits a neutral parab oli isometry . Theorem 1.6 in [ CM08 ℄ and its addendum no w apply to X ′ . The spae X ′ has no Eulidean fator: indeed, otherwise Theorem 3.8 w ould imply X ′ = R , whi h has no parab oli isometries. The k ernel of the Γ -ation on X ′ is nite and w e will inlude it in the subgroup Γ D b elo w. W e distinguish t w o ases aording as X ′ has one or more fators. In the rst ase, Is( X ′ ) annot b e totally disonneted sine otherwise Corollary 6.3 in [ CM08 ℄ p oin t ( i ) rules out neutral parab oli isometries. Th us Is( X ′ ) is a non-ompat simple Lie group with trivial en tre. If its real rank is one, w e are in ase (i); otherwise, Γ is arithmeti b y Margulis' arithmetiit y theorem [ Mar91 , Theorem 1.(1')℄ and w e are in ase (ii). F or the rest of the pro of w e treat the ase of sev eral fators for X ′ ; let Γ ∗ and let H Γ ∗ b e as in Setion 4.B . Note that H Γ ∗ ats o ompatly on ea h irreduible fator of X ′ . F urthermore, ea h irreduible fator of H Γ ∗ is non-disrete b y Theorem 4.2 . Therefore H Γ ∗ is a pro dut of the form S × D (p ossibly with one trivial fator), where S is a semi-simple Lie group with trivial en tre and D is a ompatly generated totally disonneted group without disrete fator. 38 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD By Corollary 6.3 in [ CM08 ℄ p oin t ( i ) , the existene of a neutral parab oli isometry in G implies that Is( X ′ ) is not totally disonneted. Lemma 4.8 ensures that the iden tit y omp onen t of Is( X ′ ) is in fat on tained in H Γ ∗ . Therefore, up on passing to a nite index subgroup, the iden tit y omp onen t of Is( X ′ ) oinides with S . If D is trivial, then H Γ ∗ = S is a onneted semi-simple Lie group on taining Γ as an irreduible lattie. Sine S is non-simple, it has higher rank and w e ma y app eal again to Margulis' arithmetiit y theorem; th us w e are done in this ase. Otherwise, D is non-trivial and w e ma y then apply Theorem 5.18 . It remains to  he k that the normal subgroup Γ D < Γ , if non-trivial, is not nitely generated. But w e kno w that Γ D is a disrete normal subgroup of D . By Theorem 2.4 , the lattie Γ , and hene also H Γ ∗ , ats minimally without xed p oin t at innit y on ea h irreduible fator of X ′ . Therefore, Corollary 5.8 in [ CM08 ℄ ensures that D has no nitely generated disrete normal subgroup, as desired.  Here is another v ariation, of a more geometri a v our; this time, it is not required that there b e a neutr al parab oli isometry: Theorem 6.2. L et X b e a pr op er ge o desi al ly  omplete CA T(0) sp a e with  o  omp at isom- etry gr oup and Γ < Is( X ) b e a nitely gener ate d latti e. Assume that Γ is irr e duible and r esidual ly nite. If G := Is ( X )  ontains any p ar ab oli isometry, then X is a pr o dut of symmetri sp a es and BruhatTits buildings. In p artiular, Γ is an arithmeti latti e unless X is a r e al or  omplex hyp erb oli sp a e. Pr o of. W e main tain the notation of the previous pro of and follo w the same argumen ts. W e do not kno w a priori whether there exists a neutral parab oli isometry . Ho w ev er, under the presen t assumption that X is geo desially omplete, Corollary 6.3 in [ CM08 ℄ p oin t ( iii ) sho ws that the existene of any parab oli isometry is enough to ensure that Is( X ′ ) is not totally disonneted. Th us the onlusion of Theorem 6.1 holds. In ase (i), Theorem 7.4 in [ CM08 ℄ p oin t ( iii ) ensures that X is a rank one symmetri spae and w e are done. W e no w assume that (ii) holds and dene D as in the pro of of Theorem 6.1 . The anonial disrete k ernel Γ D is trivial b y Theorem 4.11 . Sine D has no non-trivial ompat normal subgroup b y Corollary 5.8 in [ CM08 ℄, it follo ws from Theorem 5.18 that D is a totally disonneted semi-simple algebrai group. Therefore, the desired result is a onsequene of Theorem 7.4 in [ CM08 ℄ p oin t ( iii ).  F or the reord, w e prop ose a v arian t of Theorem 6.2 : Theorem 6.3. L et X b e a pr op er ge o desi al ly  omplete CA T(0) sp a e with  o  omp at isom- etry gr oup and Γ < Is( X ) b e a nitely gener ate d latti e. Assume that Γ is irr e duible and that every normal sub gr oup of Γ is nitely gener ate d. If G := Is ( X )  ontains any p ar ab oli isometry, then X is a pr o dut of symmetri sp a es and BruhatTits buildings of total r ank ≥ 2 . In p artiular, Γ is an arithmeti latti e. Pr o of. As for Theorem 6.2 , w e an apply Theorem 6.1 . W e laim that ase (i) is ruled out under the urren t assumptions. Indeed, a lattie in a simple Lie group of rank one is relativ ely h yp erb oli (see [ F ar98 ℄ or [ Osi06 ℄) and as su h has n umerous innitely generated normal subgroups (and is ev en SQ-univ ersal, see [ Gro87 ℄ or [ Del96 ℄ for the h yp erb oli ase and [ AMO07 ℄ for the general relativ e ase). In ase (ii) the disrete k ernel Γ D is trivial and rank one is exluded as in ase (i) if the group is Ar himedean; if it is non-Ar himedean, then there are no non-uniform nitely generated latties (see [ BL01 ℄) and th us Γ is again Gromo v-h yp erb oli whi h on tradits the assumption on normal, subgroups as b efore.  ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 39 W e an no w omplete the pro of of some results stated in the In tro dution. Pr o of of The or em 1.5 . If Γ is residually nite, then Theorem 6.2 yields the desired onlu- sion; it therefore remains to onsider the ase where Γ is not residually nite. W e follo w the b eginning of the pro of of Theorem 6.2 un til the in v o ation of Theorem 4.11 , sine the latter no longer applies. Ho w ev er, w e still kno w that there is a non-trivial Lie fator in Is( X ′ ) and therefore w e apply Theorem 5.18 in order to obtain the desired onlusion ab out the lattie Γ . As for the symmetri spae fator of the spae, it is pro vided b y Theorem 7.4 in [ CM08 ℄ p oin t ( iii ) .  Pr o of of Cor ol lary 1.6 . One impliation is giv en b y Theorem 1.5 . F or the on v erse, it sues to reall that unip oten t elemen ts exist in all semi-simple Lie groups of p ositiv e real rank.  6.B. Arithmetiit y of linear CA T(0) latties. W e start b y onsidering CA T(0) latties with a linear non-disrete linear ommensurator: Theorem 6.4. L et X b e a pr op er ge o desi al ly  omplete CA T(0) sp a e with  o  omp at isom- etry gr oup and Γ < Is( X ) b e a nitely gener ate d latti e. Assume that Is( X ) p ossesses a nitely gener ate d sub gr oup Λ  ontaining Γ as a sub gr oup of innite index, and  ommensu- r ating Γ . If X is irr e duible and Λ p ossesses a faithful nite-dimensional line ar r epr esentation (in har ateristi 6 = 2 , 3 ), then X is a symmetri sp a e or a BruhatTits building; in p artiular Γ is an arithmeti latti e. Remark 6.5. Sev eral examples of irreduible CA T(0) spaes X of dimension > 1 ad- mitting a disrete o ompat group of isometries with a non-disrete ommensurator in Is( X ) ha v e b een onstruted b y F. Haglund [ Hag98 ℄ and A. Thomas [ Tho06 ℄ (see also [ Hag , Théorème A℄ and [ BT ℄). In all ases that spae X is endo w ed with w alls; in partiular X is the union of t w o prop er losed on v ex subspaes. This implies in partiular that X is not a Eulidean building. Therefore, Theorem 6.4 has the follo wing onsequene: in the aforemen tioned examples of Haglund and Thomas, either the  ommensur ator of the latti e is nonline ar, or it is the union of a tower of latti es . In fat, as omm uniated to us b y F. Haglund, for most of these latties the ommensurator on tains ellipti elemen ts of in- nite order; this implies righ t a w a y that the ommensurator is not an asending union of latties and, hene, it is nonlinear. Note on the other hand that it is already kno wn that Is( X ) is mostly nonlinear in these examples, sine it on tains losed subgroups isomorphi to the full automorphism group of regular trees. Pr o of of The or em 6.4 . Sine X is irreduible and the ase X = R satises the onlusions of the theorem, w e assume heneforth that X has no Eulidean fator. The Is( X ) -ation on X is minimal b y Lemma 3.13 in [ CM08 ℄ and has no xed p oin t at innit y b y Corollary 3.12 . In partiular, w e an apply Theorem 1.1 in [ CM08 ℄: either Is( X ) is totally disonneted or it is simple Lie group with trivial en tre and X is the asso iated symmetri spae. In the latter ase, Margulis' arithmetiit y theorem nishes the pro of. W e assume heneforth that Is( X ) is totally disonneted. Let G denote the losure of Λ in Is( X ) . Note that G ats minimally without xed p oin t at innit y , sine it on tains a subgroup, namely Γ , whi h p ossesses these prop erties b y Theorem 2.4 . In partiular G has trivial amenable radial b y Theorem 1.10 in [ CM08 ℄ and th us the same holds for the dense subgroup Λ < G . In partiular an y faithful represen tation of Λ to an algebrai group yields a faithful represen tation of Λ to an adjoin t semi-simple algebrai group with Zariski-dense image, to whi h w e an apply Theorem 5.17 . As w e sa w, 40 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD the group G has no non-trivial ompat (in fat amenable) normal subgroup and furthermore G is irreduible sine X is so, see Theorem 1.10 in [ CM08 ℄. The fat that the lattie Γ has innite index in Λ rules out the disrete ase. Therefore G is a simple algebrai group and Γ an arithmeti lattie. It remains to dedue that X has the desired geometri shap e. This will follo w from Theorem 7.4 in [ CM08 ℄p oin t ( iii ) pro vided w e sho w that ∂ X is nite-dimensional and that G has full limit set. The rst fat holds sine X is o ompat; the seond is pro vided b y Corollary 2.10 .  Remark 6.5 illustrates that Theorem 6.4 fails dramatially if one assumes only that Γ is linear. Ho w ev er, passing no w to the ase where X is reduible, the linearit y of Γ is enough to establish arithmetiit y , indep enden tly of an y assumption on ommensurators, the result announed in Theorem 1.8 in the In tro dution. Theorem 6.6. L et X b e a pr op er ge o desi al ly  omplete CA T(0) sp a e with  o  omp at isom- etry gr oup and Γ < Is( X ) b e a nitely gener ate d latti e. Assume that Γ is irr e duible and p ossesses some faithful line ar r epr esentation (in har ateristi 6 = 2 , 3 ). If X is r e duible, then Γ is an arithmeti latti e and X is a pr o dut of symmetri sp a es and BruhatTits buildings. Pr o of. In view of Theorem 3.8 , w e an assume that X has no Eulidean fator. The Is( X ) - ation on X is minimal b y Lemma 3.13 in [ CM08 ℄ and has no xed p oin t at innit y b y Corollary 3.12 . In partiular, w e an apply Theorem 1.1 in [ CM08 ℄ to obtain deomp osi- tions of Is( X ) and X in whi h the fators of X orresp onding to onneted fators of Is( X ) are isometri to symmetri spaes. There is no loss of generalit y in assuming Γ ∗ = Γ in the notation of Setion 4.B . Let no w G b e the h ull of Γ . By Remark 4.6 , the group Γ is an irreduible lattie in G . Sine Is( X ) ats minimally without xed p oin t at innit y , it follo ws from Corollary 2.7 that Γ has trivial amenable radial. In partiular an y faithful represen tation of Γ to an algebrai group yields a faithful represen tation of Γ to an adjoin t semi-simple algebrai group with Zariski-dense image, to whi h w e an apply Theorem 5.1 . The group G has no non-trivial ompat normal subgroup e.g. b y minimalit y . F urther- more the disrete fator is trivial b y Theorem 4.2 . Therefore G is a simple algebrai group and Γ an arithmeti lattie. It remains to dedue that X has the desired geometri shap e and this follo ws exatly as in the pro of of Theorem 6.4 .  6.C. A family of examples. W e shall no w onstrut a family of latties Γ < G = S × D as in the statemen t of Theorem 5.18 (see also Theorem 6.2 ) with the follo wing additional prop erties: (i) There is a prop er CA T(0) spae Y with D < Is( Y ) su h that the D -ation is o ompat, minimal and without xed p oin t at innit y . In partiular, setting X = X S × Y , where X S denotes the symmetri spae asso iated to S , the Γ -ation on X is prop erly dison tin uous (in fat free), o ompat, minimal, without xed p oin t at innit y . (ii) The anonial disrete k ernel Γ D ✁ D is innite (in fat, it is a free group of oun table rank). (iii) The pronite k ernel of D / Γ D → Q is non-trivial. Remark 6.7. Sine D is minimal, it has no ompat normal subgroup and th us w e see that the pronite extension app earing in Theorem 5.18 annot b e eliminated. ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 41 W e b egin with a general onstrution: Let g b e (the geometri realisation of ) a lo ally nite graph (not redued to a single p oin t) and let Q < Is( g ) a losed subgroup whose ation is v ertex-transitiv e. In partiular, Q is a ompatly generated totally disonneted lo ally ompat group. W e p oin t out that an y ompatly generated totally disonneted lo ally ompat group an b e realised as ating on su h a graph b y onsidering S hreier graphs g , see [ Mon01 ,  11.3℄; the k ernel of this ation is ompat and arbitrary small. On the other hand, if Q is a non-Ar himedean semi-simple group, one an also tak e v ery expliit graphs dra wn on the BruhatTits building of Q , e.g. the 1 -sk eleton (this part is inspired b y [ BM00a , 1.8℄, see also [ BMZ04 ℄). Let moreo v er C b e an innite pronite group and  ho ose a lo ally nite ro oted tree t with a lev el-transitiv e C -ation for whi h ev ery innite ra y has trivial stabiliser. F or instane, one an  ho ose the oset tree asso iated to an y nested sequene of op en subgroups with trivial in tersetion, see the pro of of Théorème 15 in  6 on p. 82 in [ Ser77 ℄. W e dene a lo ally nite graph h with a C × Q -ation as the 1 -sk eleton of the square omplex t × g . Let a = e h b e the univ ersal o v er of h , Λ = π 1 ( h ) and dene the totally disonneted lo ally ompat group D b y the orresp onding extension 1 − → Λ − → D − → C × Q − → 1 . Prop osition 6.8. Ther e exists a pr op er CA T(0) sp a e Y suh that D sits in Is( Y ) as a lose d sub gr oup whose ation is  o  omp at, minimal and without xe d p oint at innity. Pr o of. One v eries readily the follo wing: Lemma 6.9. L et a b e (the ge ometri r e alisation of ) a lo  al ly nite simpliial tr e e and D < Is( a ) any sub gr oup. L et x ∈ a b e a vertex and let Y b e the  ompletion of the metri sp a e obtaine d by assigning to e ah e dge of a the length 2 − r , wher e r is the  ombinatorial distan e fr om this e dge to the ne ar est p oint of the orbit D .x . Then Y is a pr op er CA T(0) sp a e with a  o  omp at  ontinuous isometri D -ation. Mor e- over, if the D -ation on a was minimal or without xe d p oint at innity, then the  orr e- sp onding statement holds for the D -ation on Y .  Apply the lemma to the tree a = e h onsidered earlier. W e laim that the D -ation on a is minimal. Clearly it sues to sho w that the Λ -ation is minimal. Note that Λ ats transitiv ely on ea h bre of p : e h → h . Th us it is enough to sho w that the on v ex h ull of a giv en bre meets ev ery other bre. Consider t w o distint v erties v , v ′ ∈ h . The pro dut nature of h mak es it lear that v and v ′ are b oth on tained in a ommon minimal lo op based at v . This lo op lifts to a geo desi line in e h whi h meets the resp etiv e bres of v and v ′ alternativ ely and p erio dially . In partiular, this onstrution yields a geo desi segmen t joining t w o p oin ts in the bre ab o v e v and on taining a p oin t sitting ab o v e v ′ , whene the laim. Sine Λ ats freely and minimally on the tree a whi h is not redued to a line, it follo ws that Λ xes no end of a . Th us the lemma pro vides a prop er CA T(0) spae Y with a o ompat minimal isometri D -ation, without xed p oin t at innit y . It remains to sho w that D < Is( Y ) is losed. This holds b eause the totally disonneted groups Is( a ) and Is( Y ) are isomorphi; indeed, the anonial map a → Y indues a on tin uous surjetiv e homomorphism Is( a ) → Is( Y ) , whi h is th us op en.  Remark 6.10. The ab o v e onstrution giv es an example of a prop er CA T(0) spae with a totally disonneted o ompat and minimal group of isometries su h that not all p oin t stabilisers are op en. Consider indeed the p oin ts added when ompleting. Their stabilisers 42 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD map to Q under D → ( C × Q ) and hene annot b e op en. In other w ords, the ation is not smo oth in the terminology of [ Cap07 ℄. Notie ho w ev er that the set of p oin ts with op en stabiliser is neessarily a dense on v ex in v arian t set. W e shall no w sp eialise this general onstrution to yield our family of examples. Let K , H , Σ , K (Σ) b e as desrib ed after Theorem 5.1 on p. 25 . W e write Σ f , Σ ∞ ⊆ Σ for the subsets of nite/innite plaes and assume that b oth are non-empt y . Let S = Q v ∈ Σ f H ( K v ) + and Q = Q v ∈ Σ ∞ H ( K v ) + . The group ∆ = H ( K (Σ)) ∩ ( S × Q ) is an irreduible o ompat lattie in S × Q . Let C b e an y pronite group with a dense inlusion ∆ → C . W e no w em b ed ∆ diagonally in S × C × Q ; learly ∆ is a o ompat lattie. Let Γ < G = S × D b e its pre-image. Then Γ is a o ompat lattie sine it on tains the disrete k ernel of the anonial map G → S × C × Q . It is learly irreduible and therefore pro vides an example that the struture of the desription in the onlusion of Theorem 5.18 annot b e simplied. F urthermore, the normal subgroup app earing in Theorem 6.1 (ii) is also una v oidable. W e end this setion with a few supplemen tary remarks on the preeding onstrution: (i) If the pronite group C has no disrete normal subgroup, then Γ D = π 1 ( h ) oinides with the quasi-en tre of D . This w ould b e the ase for example if C = H ( K v ) and H is almost K -simple of higher rank, where v is a non-Ar himedean v aluation su h that H is K v -anisotropi. In partiular, in that situation Γ D is the unique maximal disrete normal subgroup of D and the quotien t D / Γ D has a unique maximal ompat normal subgroup. Th us the group G admits a unique deomp osition as in Theorem 5.18 in this ase. (ii) W e emphasise that, ev en though D / Γ D deomp oses as a diret pro dut C × Q in the ab o v e onstrution, the group D admits no non-trivial diret pro dut de- omp osition, sine it ats minimally without xed p oin t at innit y on a tree (see Theorem 1.10 in [ CM08 ℄). (iii) The fat that D / Γ D deomp oses as a diret pro dut C × Q is not a oinidene. In fat, this is happ ens alw a ys pro vided that ev ery o ompat lattie in S has the Congruene Subgroup Prop ert y (CSP). Indeed, giv en a ompat op en subgroup U of D / Γ D , the in tersetion Γ U of Γ ∩ ( S × U ) is an irreduible lattie in S × U with trivial anonial disrete k ernels. By (CSP), up on replaing Γ U b y a nite index subgroup (whi h amoun ts to replae U b y an op en subgroup), the pronite ompletion c Γ U splits as the pro dut o v er all primes p of the pro- p ompletions [ (Γ U ) p , whi h are just-innite. Th us the anonial surjetiv e map c Γ U → U sho ws that U is a diret pro dut. This implies that the maximal ompat normal subgroup of D / Γ D is a diret fator. (iv) A ording to a onjeture of Serre's (fo otnote on page 489 in [ Ser70 ℄), if S has higher rank then ev ery irreduible lattie in S has (CSP). (See [ Rag04 ℄ for a reen t surv ey on this onjeture.) 7. A few questions W e onlude b y olleting some further questions that w e ha v e enoun tered while w orking on this pap er. It is w ell kno wn that the Tits b oundary of a prop er CA T(0) spae with o ompat isometry group is neessarily nite-dimensional (see [ Kle99 , Theorem C℄). It is quite p ossible that the same onlusion holds under a m u h w eak er assumption. ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 43 Question 7.1. Let X b e a prop er CA T(0) spae su h that Is( X ) has full limit set. Is the b oundary ∂ X nite-dimensional? A p ositiv e answ er to this question w ould sho w in partiular that the seond set of as- sumptions  denoted (b)  in Theorem 7.4 in [ CM08 ℄ is in fat redundan t. Let G b e a simple Lie group ating on tin uously b y isometries on a prop er CA T(0) spae X . The Karp elevi hMosto w theorem ensures that there exists a on v ex orbit when X is a symmetri spae of non-ompat t yp e. This statemen t, ho w ev er, annot b e generalised to arbitrary X in view of Example 7.7 in [ CM08 ℄. Question 7.2. Let G b e a simple Lie group ating on tin uously b y isometries on a prop er CA T(0) spae X . If the ation is o ompat, do es there exist a on v ex orbit? It is sho wn in Theorem 7.4 in [ CM08 ℄ p oin t ( iii ) that if X is geo desially omplete, then the answ er is p ositiv e. It is go o d to k eep in mind Example 7.6 in [ CM08 ℄, whi h sho ws that the natural analogue of this question for a simple algebrai group o v er a non- Ar himedean lo al eld has a negativ e answ er. More optimistially , one an ask for a on v ex orbit whenev er the simple Lie group ats on a omplete (not neessarily prop er) CA T(0) spae, but assuming the ation non-ev anesen t (in the sense of [ Mon06 ℄). A p ositiv e answ er w ould imply sup errigidit y statemen ts up on applying it to spaes of equiv arian t maps. Man y of our statemen ts on CA T(0) latties require the assumption of nite generation. One should of ourse w onder for ea h of them whether it remains v alid without this as- sumption. One instane where this question is esp eially striking is the follo wing (see Corollary 3.12 ). Question 7.3. Let X b e a prop er CA T(0) spae whi h is minimal and o ompat. Assume that Is( X ) on tains a lattie. Is it true that Is( X ) has no xed p oin t at innit y? In a forthoming artile [ CM08b ℄, w e shall establish a p ositiv e answ er to this question b y in v estigating the rle of unimo dularit y for the full isometry group. W e ha v e seen in Corollary 7.12 in [ CM08 ℄ that if the isometry group of a prop er CA T(0) spae X is non-disrete in a strong sense, then Is( X ) omes lose to b eing a diret pro dut of top ologially simple groups. Question 7.4. Retain the assumptions of Corollary 7.12 in [ CM08 ℄. Is it true that so c( G ∗ ) is a pro dut of simple groups? Is it o ompat in G , or at least do es G ha v e ompat Ab elianisation? Clearly Corollary 7.12 in [ CM08 ℄ redues the question to the ase where Is( X ) is totally disonneted. One an also ask if so c( G ∗ ) is ompatly generated (whi h is the ase e.g if it is o ompat in G ). If so, w e obtain additional information b y applying Prop osition 6.12 in [ CM08 ℄. In the ab o v e situation one furthermore exp ets that the geometry of X is eno ded in the struture of Is( X ) . In preise terms, w e prop ose the follo wing. Question 7.5. Retain the assumptions of Corollary 7.12 in [ CM08 ℄. It it true that an y prop er o ompat ation of G on a prop er CA T(0) spae Y yields an equiv arian t isometry ∂ X → ∂ Y b et w een the Tits b oundaries? Or an equiv arian t homeomorphism b et w een the b oundaries with resp et to the ne top ology? The disussion around Corollary 5.3 ( f. Remark 5.4 ) suggests the follo wing. 44 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD Question 7.6. Let Γ < G = G 1 × · · · × G n b e an irreduible nitely generated lattie, where ea h G i is a lo ally ompat group. Do es ev ery  harater Γ → R extend on tin uously to G ? Y. Shalom [ Sha00 ℄ pro v ed that this is the ase when Γ is o ompat and in some other situations. Referenes [AB94℄ Norb ert A'Camp o and Mar Burger, R ése aux arithmétiques et  ommensur ateur d'apr ès G. A. Mar gulis , In v en t. Math. 116 (1994), no. 1-3, 125. [AB98℄ Sot A dams and W erner Ballmann, A menable isometry gr oups of Hadamar d sp a es , Math. Ann. 312 (1998), no. 1, 183195. [AMO07℄ Goulnara Arzhan tsev a, Ashot Minasy an, and Denis Osin, The SQ-universality and r esidual pr op- erties of r elatively hyp erb oli gr oups , J. Algebra 315 (2007), no. 1, 165177. [BG02℄ Nadia Benakli and Y air Glasner, A utomorphism gr oups of tr e es ating lo  al ly with ane p ermu- tations , Geom. Dediata 89 (2002), 124. [BG07℄ Emman uel Breuillard and T sa hik Gelander, A top olo gi al Tits alternative , Ann. of Math. (2) 166 (2007), no. 2, 427474. [BGM04℄ Janez Bernik, Rob ert Guralni k, and Mitja Mastnak, R e dution the or ems for gr oups of matri es , Linear Algebra Appl. 383 (2004), 119126. [BH99℄ Martin R. Bridson and André Haeiger, Metri sp a es of non-p ositive urvatur e , Grundlehren der Mathematis hen Wissens haften 319, Springer, Berlin, 1999. [BL01℄ Hyman Bass and Alexander Lub otzky , T r e e latti es , Progress in Mathematis, v ol. 176, Birkhäuser Boston In., Boston, MA, 2001, With app endies b y Bass, L. Carb one, Lub otzky , G. Rosen b erg and J. Tits. [BM00a℄ Mar Burger and Shahar Mozes, Gr oups ating on tr e es: fr om lo  al to glob al strutur e , Inst. Hautes Études Si. Publ. Math. (2000), no. 92, 113150 (2001). [BM00b℄ , L atti es in pr o dut of tr e es , Inst. Hautes Études Si. Publ. Math. (2000), no. 92, 151194 (2001). [BM02℄ Mar Burger and Niolas Mono d, Continuous b ounde d  ohomolo gy and appli ations to rigidity the ory , Geom. F unt. Anal. 12 (2002), no. 2, 219280. [BMZ04℄ Mar Burger, Shahar Mozes, and Rob ert J. Zimmer, Line ar r epr esentations and arithmetiity for latti es in pr o duts of tr e es , FIM preprin t, 2004. [Bon04℄ P atri k Bon vin, Str ong b oundaries and  ommensur ator sup er-rigidity , Geom. F unt. Anal. 14 (2004), no. 4, 843848, App endix to T op olo gi al simpliity,  ommensur ator sup er-rigidity and non-line arities of Ka-Mo o dy gr oups b y B. Rém y . [Bor60℄ Armand Borel, Density pr op erties for  ertain sub gr oups of semi-simple gr oups without  omp at  omp onents , Ann. of Math. (2) 72 (1960), 179188. [Bor66℄ , Density and maximality of arithmeti sub gr oups , J. Reine Angew. Math. 224 (1966), 7889. [Bou60℄ Niolas Bourbaki, Éléments de mathématique. Pr emièr e p artie. (F asiule III.) Livr e III; T op olo- gie génér ale. Chap. 3: Gr oup es top olo giques. Chap. 4: Nombr es r é els , T roisième édition revue et augmen tée, A tualités Si. Indust., No. 1143. Hermann, P aris, 1960. [Bou71℄ , Éléments de mathématique. Top olo gie génér ale. Chapitr es 1 à 4 , Hermann, P aris, 1971. [BS87℄ Mar Burger and Viktor S hro eder, A menable gr oups and stabilizers of me asur es on the b oundary of a Hadamar d manifold , Math. Ann. 276 (1987), no. 3, 505514. [BS06℄ Uri Bader and Y eh uda Shalom, F ator and normal sub gr oup the or ems for latti es in pr o duts of gr oups , In v en t. Math. 163 (2006), no. 2, 415454. [BT℄ Angela Barnhill and Anne Thomas, Density of  ommensur ators for right-angle d buildings , Preprin t. [BT73℄ Armand Borel and Jaques Tits, Homomorphismes abstr aits de gr oup es algébriques simples , Ann. of Math. (2) 97 (1973), 499571. [Bur95℄ Mar Burger, R igidity pr op erties of gr oup ations on CA T(0) -sp a es , Pro eedings of the In terna- tional Congress of Mathematiians, V ol. 1, 2 (Züri h, 1994) (Basel), Birkhäuser, 1995, pp. 761769. [Cap07℄ Pierre-Emman uel Caprae, A menable gr oups and hadamar d sp a es with a total ly dis onne te d isometry gr oup , Preprin t, a v ailable at , 2007. ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 45 [CFP96℄ James W. Cannon, William J. Flo yd, and W alter R. P arry , Intr o dutory notes on Rihar d Thomp- son 's gr oups , Enseign. Math. (2) 42 (1996), no. 3-4, 215256. [CM08a℄ Pierre-Emman uel Caprae and Niolas Mono d, De  omp osing lo  al ly  omp at gr oups into simple pie  es , Preprin t, arXiv:math/0811.4101, 2008. [CM08b℄ , Fixe d p oints and amenability in non-p ositive urvatur e , In preparation, 2008. [CM08℄ , Isometry gr oups of non-p ositively urve d sp a es: Strutur e the ory , Preprin t, 2008. [CM08d℄ , A latti e in mor e than two KaMo o dy gr oups is arithmeti , Preprin t, arXiv:math/0812.1383, 2008. [CM08e℄ , Some pr op erties of non-p ositively urve d latti es , C. R. Math. A ad. Si. P aris 346 (2008), no. 15-16, 857862. [Del96℄ Thomas Delzan t, Sous-gr oup es distingués et quotients des gr oup es hyp erb oliques. , Duk e Math. J. 83 (1996), no. 3, 661682. [Eb e80℄ P atri k Eb erlein, L atti es in sp a es of nonp ositive urvatur e , Ann. of Math. (2) 111 (1980), no. 3, 435476. [Eb e82℄ , Isometry gr oups of simply  onne te d manifolds of nonp ositive urvatur e II , A ta Math. 149 (1982), no. 1-2, 4169. [Eb e83℄ , Eulide an de Rham fator of a latti e of nonp ositive urvatur e , J. Dieren tial Geom. 18 (1983), no. 2, 209220. [Eb e96℄ , Ge ometry of nonp ositively urve d manifolds , Chiago Letures in Mathematis, Univ ersit y of Chiago Press, Chiago, IL, 1996. [Ede64℄ Mi hael Edelstein, On non-exp ansive mappings of Banah sp a es , Pro . Cam bridge Philos. So . 60 (1964), 439447. [Eym72℄ Pierre Eymard, Moyennes invariantes et r epr ésentations unitair es , Springer-V erlag, Berlin, 1972, Leture Notes in Mathematis, V ol. 300. [F ar98℄ Benson F arb, R elatively hyp erb oli gr oups , Geom. F unt. Anal. 8 (1998), no. 5, 810840. [F ar08℄ Daniel F arley , The ation of Thompson 's gr oup on a CA T(0) b oundary , Groups Geom. Dyn. 2 (2008), no. 2, 185222. [FHT08℄ Benson F arb, G. Christopher Hrusk a, and Anne Thomas, Pr oblems on automorphism gr oups of nonp ositively urve d p olyhe dr al  omplexes and their latti es , T o app ear in F ests hrift in honor of Rob ert Zimmer's 60th Birthda y , 2008. [FW06℄ Benson F arb and Shm uel W ein b erger, Isometries, rigidity and universal  overs , Annals of Math. (to app ear), 2006. [GM07℄ Y air Glasner and Niolas Mono d, A menable ations, fr e e pr o duts and a xe d p oint pr op erty , Bull. Lond. Math. So . 39 (2007), no. 1, 138150. [Gro87℄ Mikhaïl Gromo v, Hyp erb oli gr oups , Essa ys in group theory , Math. Si. Res. Inst. Publ., v ol. 8, Springer, New Y ork, 1987, pp. 75263. [Gro93℄ , Asymptoti invariants of innite gr oups , Geometri group theory , V ol. 2 (Sussex, 1991) (Cam bridge), London Math. So . Leture Note Ser., v ol. 182, Cam bridge Univ. Press, 1993, pp. 1295. [GW71℄ Detlef Gromoll and Joseph A. W olf, Some r elations b etwe en the metri strutur e and the algebr ai strutur e of the fundamental gr oup in manifolds of nonp ositive urvatur e , Bull. Amer. Math. So . 77 (1971), 545552. [Hag℄ F rédéri Haglund, Finite index sub gr oups of gr aph pr o duts , Geom. Dediata (to app ear). [Hag98℄ , R ése aux de Coxeter-Davis et  ommensur ateurs , Ann. Inst. F ourier (Grenoble) 48 (1998), no. 3, 649666. [Hei74℄ Ernst Hein tze, On homo gene ous manifolds of ne gative urvatur e , Math. Ann. 211 (1974), 2334. [HM79℄ Roger E. Ho w e and Calvin C. Mo ore, Asymptoti pr op erties of unitary r epr esentations , J. F unt. Anal. 32 (1979), no. 1, 7296. [J06℄ T adeusz Jan uszkiewiz and Jaek wi  atk o wski, Simpliial nonp ositive urvatur e , Publ. Math. Inst. Hautes Études Si. (2006), no. 104, 185. [Ka  03℄ V adim A. Ka  mano vi h, Double er go diity of the Poisson b oundary and appli ations to b ounde d  ohomolo gy , Geom. F unt. Anal. 13 (2003), no. 4, 852861. [Kar53℄ F ridrikh I. Karp elevi£, Surfa es of tr ansitivity of a semisimple sub gr oup of the gr oup of motions of a symmetri sp a e , Doklady Ak ad. Nauk SSSR (N.S.) 93 (1953), 401404. [KL06℄ Brue Kleiner and Bernhard Leeb, R igidity of invariant  onvex sets in symmetri sp a es , In v en t. Math. 163 (2006), no. 3, 657676. 46 PIERRE-EMMANUEL CAPRA CE AND NICOLAS MONOD [Kle99℄ Brue Kleiner, The lo  al strutur e of length sp a es with urvatur e b ounde d ab ove , Math. Z. 231 (1999), no. 3, 409456. [Lan00℄ Erasm us Landv ogt, Some funtorial pr op erties of the Bruhat-Tits building , J. Reine Angew. Math. 518 (2000), 213241. [L Y72℄ H. Blaine La wson, Jr. and Shing-T ung Y au, Comp at manifolds of nonp ositive urvatur e , J. Dif- feren tial Geometry 7 (1972), 211228. [Mal40℄ Anatoly I. Mal'ev, On isomorphi matrix r epr esentations of innite gr oups , Re. Math. [Mat. Sb ornik℄ N.S. 8 (50) (1940), 405422. [Mar91℄ Gregory A. Margulis, Disr ete sub gr oups of semisimple Lie gr oups , Ergebnisse der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematis and Related Areas (3)℄, v ol. 17, Springer- V erlag, Berlin, 1991. [MMS04℄ Igor Miney ev, Niolas Mono d, and Y eh uda Shalom, Ide al bi ombings for hyp erb oli gr oups and appli ations , T op ology 43 (2004), no. 6, 13191344. [Mon01℄ Niolas Mono d, Continuous b ounde d  ohomolo gy of lo  al ly  omp at gr oups , Leture Notes in Math- ematis, v ol. 1758, Springer-V erlag, Berlin, 2001. [Mon05℄ , A rithmetiity vs. nonline arity for irr e duible latti es , Geom. Dediata 112 (2005), 225 237. [Mon06℄ , Sup errigidity for irr e duible latti es and ge ometri splitting , J. Amer. Math. So . 19 (2006), no. 4, 781814. [Mos55℄ George D. Mosto w, Some new de  omp osition the or ems for semi-simple gr oups , Mem. Amer. Math. So . 1955 (1955), no. 14, 3154. [Mou88℄ Gab or Moussong, Hyp erb oli  oxeter gr oups , Ph.D. thesis, Ohio State Univ ersit y , 1988. [MP03℄ Niolas Mono d and Sorin P opa, On  o-amenability for gr oups and von Neumann algebr as , C. R. Math. A ad. Si. So . R. Can. 25 (2003), no. 3, 8287. [Osi06℄ Denis Osin, R elatively hyp erb oli gr oups: intrinsi ge ometry, algebr ai pr op erties, and algorithmi pr oblems , Mem. Amer. Math. So . 179 (2006), no. 843, vi+100. [Pin98℄ Ri hard Pink, Comp at sub gr oups of line ar algebr ai gr oups , J. Algebra 206 (1998), no. 2, 438504. [Pra77℄ Gopal Prasad, Str ong appr oximation for semi-simple gr oups over funtion elds , Ann. of Math. (2) 105 (1977), no. 3, 553572. [Pra82℄ , Elementary pr o of of a the or em of Bruhat-Tits-Rousse au and of a the or em of Tits , Bull. So . Math. F rane 110 (1982), no. 2, 197202. [Rag72℄ Madabusi San tanam Ragh unathan, Disr ete sub gr oups of Lie gr oups , Springer-V erlag, New Y ork, 1972, Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 68. [Rag04℄ , The  ongruen e sub gr oup pr oblem , Pro . Indian A ad. Si. Math. Si. 114 (2004), no. 4, 299308. [Rém99℄ Bertrand Rém y , Constrution de r ése aux en thé orie de KaMo o dy , C. R. A ad. S. P aris 329 (1999), 475478. [Sel60℄ A tle Selb erg, On dis ontinuous gr oups in higher-dimensional symmetri sp a es. , Con trib. F untion Theory , In t. Collo qu. Bom ba y , Jan. 1960, 147-164 (1960)., 1960. [Ser70℄ Jean-Pierre Serre, L e pr oblème des gr oup es de  ongruen e p our SL 2 , Ann. of Math. (2) 92 (1970), 489527. [Ser77℄ , A rbr es, amalgames, SL 2 , So iété Mathématique de F rane, P aris, 1977, A v e un sommaire anglais, Rédigé a v e la ollab oration de Hyman Bass, Astérisque, No. 46. [Sha00℄ Y eh uda Shalom, R igidity of  ommensur ators and irr e duible latti es , In v en t. Math. 141 (2000), no. 1, 154. [Sim96℄ Mi hel Simonnet, Me asur es and pr ob abilities , Univ ersitext, Springer-V erlag, New Y ork, 1996, With a forew ord b y Charles-Mi hel Marle. [T ak02℄ Masami hi T ak esaki, The ory of op er ator algebr as. I , Enylopaedia of Mathematial Sienes, v ol. 124, Springer-V erlag, Berlin, 2002, Reprin t of the rst (1979) edition, Op erator Algebras and Non-omm utativ e Geometry , 5. [Tho06℄ Anne Thomas, L atti es ating on right-angle d buildings , Algebr. Geom. T op ol. 6 (2006), 12151238 (eletroni). [Th u97℄ William P . Th urston, Thr e e-dimensional ge ometry and top olo gy. Vol. 1 , Prineton Mathematial Series, v ol. 35, Prineton Univ ersit y Press, Prineton, NJ, 1997, Edited b y Silvio Levy . [Tit64℄ Jaques Tits, A lgebr ai and abstr at simple gr oups , Ann. of Math. (2) 80 (1964), 313329. [Tit72℄ , F r e e sub gr oups in line ar gr oups , J. Algebra 20 (1972), 250270. ISOMETR Y GR OUPS OF NON-POSITIVEL Y CUR VED SP A CES: DISCRETE SUBGR OUPS 47 [Tit87℄ , Uniqueness and pr esentation of Ka-Mo o dy gr oups over elds , J. Algebra 105 (1987), no. 2, 542573. [Vin94℄ Èrnest B. Vin b erg (ed.), Lie gr oups and Lie algebr as, III , Enylopaedia of Mathematial Sienes, v ol. 41, Springer-V erlag, Berlin, 1994, Struture of Lie groups and Lie algebras, A translation of Curr ent pr oblems in mathematis. F undamental dir e tions. V ol. 41 (Russian), Ak ad. Nauk SSSR, V seso yuz. Inst. Nau hn. i T ekhn. Inform., Moso w, 1990, T ranslation b y V. Mina hin. [W eh73℄ Bertram A. F. W ehrfritz, Innite line ar gr oups. An a  ount of the gr oup-the or eti pr op erties of innite gr oups of matri es , Springer-V erlag, New Y ork, 1973, Ergebnisse der Matematik und ihrer Grenzgebiete, Band 76. [Wil71℄ John S. Wilson, Gr oups with every pr op er quotient nite. , Pro . Cam b. Philos. So . 69 (1971), 373391. [Wis04℄ Daniel T. Wise, Cubulating smal l  an el lation gr oups , Geom. F unt. Anal. 14 (2004), no. 1, 150 214. [W or07℄ Kevin W ortman, Quasi-isometri rigidity of higher r ank S -arithmeti latti es. , Geom. T op ol. 11 (2007), 9951048. [Y au82℄ Shing T ung Y au, Pr oblem se tion , Seminar on Dieren tial Geometry , Ann. of Math. Stud., v ol. 102, Prineton Univ. Press, Prineton, N.J., 1982, pp. 669706. [Zim78℄ Rob ert J. Zimmer, A menable er go di gr oup ations and an appli ation to Poisson b oundaries of r andom walks , J. F untional Analysis 27 (1978), no. 3, 350372. [Zim84℄ , Er go di the ory and semisimple gr oups , Birkhäuser V erlag, Basel, 1984. [Zim87℄ , A menable ations and dense sub gr oups of Lie gr oups , J. F unt. Anal. 72 (1987), no. 1, 5864. UCL, 1348 Louv ain-la-Neuve, Belgium E-mail addr ess : pe.apraeulouvain.be EPFL, 1015 La usanne, Switzerland E-mail addr ess : niolas.monodepfl.h

Original Paper

Loading high-quality paper...

Comments & Academic Discussion

Loading comments...

Leave a Comment