Arxiv:Math/0610904V3 [Math.MG] 2 Jul 2008

arXiv:math/0610904v3 [math.MG] 2 Jul 2008 ySna´ n skona h Blaschke-Santaló inequalit the 20]: 19, as [3, known is and Santaló by ojcue nuprbudadalwrbudfrtevalue the for bound lower a and of bound upper an conjectured n o-eomnmmi ahdimension each in minimum non-zero a and of in volume uous Mahler the as known is oeisda rplrbd.(ti loteui alo h nor the of ball unit by the defined also one is the (It to body. dual polar or dual its note ∗ 1.2 Conjecture v 1.1 Theorem ainudrGatN.0606795 No. Grant under dation an-amn 1]osre htteMhe oueo an of volume instance, Mahler For the n much that sphere. a observed round [19] is a Saint-Raymond minimum) than real shape the complicated more (or minimum conjectured the iiie mn etal ymti ovxbde ythe by bodies convex C cube symmetric centrally among minimized rs-oyoe u lob te oyoe when polytopes other by also but cross-polytope, alrcnetr stu pt nepnnilfco [5]: factor exponential an to up true is conjecture Mahler 17].) [9, bodi zonoids convex and i 1-unconditional [19] established including cases, been special has some conjecture Mahler the (Nonetheless, hsmtra sbsduo okspotdb h ainlSc National the by supported work upon based is material This ( -cube K If h oe on ssilaconjecture: a still is bound lower The h iiiainpolmi osdrdhre because harder considered is problem minimization The nantdppr oranadMla hwdta the that showed Milman and Bourgain paper, noted a In v ( ) K K suiul aiie yellipsoids. by maximized uniquely is ) n ⊂ C K nec xddmnin h pe on a proven was bound upper The dimension. fixed each in [ = n R n fnl nain,i civsafiiemaximum finite a achieves it invariant, affinely and ste o nyb h oueo h dual the of volume the by only not tied is n − sacnrlysmerccne oy let body, convex symmetric centrally a is 1 qaiywhen equality ftepedshrso nidfiieinrpoutsae Be sphere space. the in product integrals inner linking indefinite Gauss traditional an analyze of pseudospheres the of K K ojcueo h product the on conjecture imn h hwdta hr saconstant a is there that showed who Milman, where oa body polar Back,Santaló)(Blaschke, , ♦ 1 san is h ro ssavrino h as ikn nerlt obta to integral linking Gauss the of version a uses proof The eetbihavrino h oteekcnetr,wihi which conjecture, bottleneck the of version a establish We (Mahler) ] ⊆ n v . ( K .INTRODUCTION 1. γ K n n × cb.Tebtlnc ojcue(nislatgnrlform general least its (in conjecture bottleneck The -cube. samntncfco htbgn at begins that factor monotonic a is ( = ) K K ◦ . ◦ K h alrcnetr set htteMhe volume Mahler the that asserts conjecture Mahler The . smnmzdwhen minimized is K Vol nayfie ieso ,v n, dimension fixed any In )Tepouto h volumes the of product The .) rmteMhe ojcuet as ikn integrals linking Gauss to conjecture Mahler the From sa lisi.I ple oamr eea ojcueconc conjecture general more a to applies It ellipsoid. an is eateto ahmtc,Uiest fClfri,Davi California, of University Mathematics, of Department K )( . Vol K nayfie ieso n, dimension fixed any In ic ti ohcontin- both is it Since . v ( eiae om ahr nn atclroccasion particular no on father, my to Dedicated K K ◦ ( = ) ) K Vol n sa lisi.I mle h alrcnetr pt fact a to up conjecture Mahler the implies It ellipsoid. an is alr[15] Mahler . K )( n ec Foun- ience Vol slarge. is c π K 4 rgKuperberg Greg ( C uhta h alrcnetr stu pt atrof factor a to up true is conjecture Mahler the that such K K ◦ n ovre to converges and n ∗ ) ◦ de- a , ) es m is n y ftevlm fasmerccne body convex symmetric a of volume the of S n uhta o n n n etal-ymti ovxbody convex centrally-symmetric n, dimension any of and K n any for that such imntermsy htalMhe oue in volumes Mahler all that says theorem Milman ewl iiietevlm facranbody certain a of volume the minimize will We stant be- see symmetry; low.) central without holds Bourga also the theorem (Indeed, Milman factor. exponential some by differ only n uedfe nyb atrof factor sphere a a by of only volume differ Mahler cube the a that and Note known. currently of necks hoe 1.4 Theorem proofs the of all constant Although the construct Milman. technically by independently and stera beto td,w aernmdit renamed have we study, of object real the is hoe 1.3 Theorem en h subsets the Define H r a,otooa,adcnee tteorigin. the at centered and orthogonal, flat, are ftehyperboloids the of hi le on Then join. filled their ek ftepstv n eaieui suopee H pseudospheres unit negative and positive the of necks nwi ti vrnon-convex.) ever is it if know auei n ieto. epeiul hwdthat showed previously We direction.) any in vature in ( h body The pclk eto of section spacelike a H umnflsof Submanifolds − I.e. 1 − − R h ro fTheorem of proof The nti ril,w iletbihabte au ftecon- the of value better a establish will we article, this In K n nhproi space hyperbolic in and n httericuin r oooyeuvlne [13]. equivalences homotopy are inclusions their that and , hyhv obudr onswt eoetisccur- extrinsic zero with points boundary no have they , nayidfiieinrpoutspace product inner indefinite any in n + as h acltosaesmlr ewl also will we similar, are calculations the cause c unipisaprilslto oteMahler the to solution partial a implies turn n × ∗ K √ and ycmaigteMhe ouewt nte volume. another with volume Mahler the comparing by nacntn oe on nVol on bound lower constant a in R set httevlm facrandomain certain a of volume the that asserts ) + .Ti teghn euto oranand Bourgain of result a strengthens This 2. n K and ups that Suppose . K v ♦ ( − K K stefildji,dfie nSection in defined join, filled the is Bugi,Milman) (Bourgain, Mi theorem) (Main K and ) ± − smnmzd(o-nqey when (non-uniquely) minimized is = H I eeec 13, Reference (In . rigteji faytonecks two any of join the erning H K ,C 95616 CA s, { + ± ♥ ( ~ x Vol = and v a h le on ic h le join filled the Since join. filled the was ,~ H ( y K { 1.3 H ) K + ( N ) ∈ ~ n that , H x and − ≥ ♦ ,~ a ensmlfidb iir[16] Pisier by simplified been has K − 1 y . ) smnmzdwe N when minimized is c . ×

fti yeaecalled are type this of n ~ K x v e N Let K K · ◦ ( K ~ . C y − ◦ r ohpstvl curved. positively both are c rof or hr sacntn c constant a is There n

= ogo au o tis it for value good no , ∈ ~ x ) satmlk eto of section timelike a is K . R · + ± K ~ c ♦ y c n n 1 ♦ R n n N and = oteBourgain- the So . a h ovxhull convex the was n its and π 4 } . with , ( a ± n , b γ 1 ) n K n e N let and , } , K − ♦ n 2 ♦ ed not do We . yaytwo any by dimensions ftetwo the of , ⊂ + K n N and necks K + × ♦ + > and K in- be ◦ is − 0 . . 2

+ Because of the geometryof the necks N and N−, we previ- Conjecture 1.7 (Mahler). In any fixed dimension n, v(K) is ously called Theorem 1.4 “the bottleneck conjecture”. We will minimized among convex bodies K, whose interiors contain interpret Theorem 1.4 as a new isoperimetric-type inequality the origin, by a centered simplex ∆n. for convex bodies: Although we will only directly study centrally symmetric Corollary 1.5. If K Rn is a centrally symmetric convex ⊂ convex bodies, Corollary 1.6 yields a corollary for general body, then Vol K♦ is minimized when K is an ellipsoid. convex bodies using a standard technique. Corollary 1.5 follows immediately from Theorem 1.4 when Corollary 1.8. If K Rn is a convex body with the origin in K and K are positively curved, because necks K+ and K that ◦ − its interior, then ⊂ come from a symmetric convex body K are flat, orthogonal, and centered when K is an ellipsoid. By continuity, the bound 4n(n!)4πn π n must also hold for arbitrary K. v(K) = δn v(∆n), ≥ (2n)!2(n/2)!2 2e On the one hand, Vol K♦ is minimized when the Mahler volume v(K) is maximized. On the other hand, K♦ is only where moderately exponentially smaller than its superset K K◦. n × n 6 n Explicitly, let Bn be the round unit ball in R . Then 2e 8 n! e 2π δ = . π n 2n !2 n 2 !2 n 1 n+1 e 2n(n!)2πn ≤ ( ) ( / ) ( + ) → Vol Bn♦ = (2n)!(n/2)!2 Thus, our value for the Bourgain-Milman constant in the asymmetric case is π . because Bn is the join of two round, orthogonal n-balls of ra- 2e dius √2 and Remark. Even though the Bourgain-Milman inequality holds πn/2 in the asymmetric case, the bound in Corollary 1.6 does not. Vol Bn = . In particular, (n/2)! v(∆ ) 1/n e π In comparison, lim n = < . n ∞ v(C ) 4 4 4n → n v(Cn)= . n! The proof fails because when K is not centrally symmetric, + In conclusion: only K exists as a subset of K K◦. The region K♦ can be formed but its volume may be larger× than v(K). Nonetheless n Corollary 1.6. If K R is a centrally symmetric convex Theorem 1.4 does say something about asymmetric convex body, then ⊂ + + bodies. We can let N = K1 for one convex body K1, and + n let N be a natural isometric image of K for another convex 2n(n!)2πn π − 2 v(K) = γn v(Cn), body K . Then the statement is that Vol N is minimized ≥ (2n)!(n/2)!2 4 2 ♦ when K1 and K2 coincide, and are a centered ellipsoid. where 3 n Proof. Given a convex body K, then K K is its difference 4 n! 2 body, by definition the set of differences− between points in K. γn = √2. π ≤ (2n)!(n/2)!2 → It is a centrally symmetric convex body. Rogers and Shephard [18] showed that π Thus, our value of c in Theorem 1.3 is /4. Since the 2n Bourgain-Milman theorem is also often phrased in terms of Vol K K Vol K, − ≤ n v(Bn), the Mahler volume of a round ball, we can also write Corollary 1.6 as with equality if and only if K is a simplex. The polar body of 2n n! 2 K K is best described by a relation between the correspond- ( ) n − n v(K) v(Bn) > 2− v(Bn). ing norms on R : ≥ (2n)! ~x = ~x + ~x . Thus, our value for the Bourgain-Milmanconstant in this form (K K)◦ K◦ K◦ 1 2 || || − || || || ||− is 2 . Mahler’s conjecture would imply π . Corollary 1.6 also improves a previous non-asymptotic es- Since timate which was inspired by the bottleneck conjecture [12]. 1 There we established that if K Rn and n 4, then Vol K = d~x, n 1 n ⊂ ≥ ZS − n ~x K n || || v(K) (log2 n)− v(Bn). ≥ and since the integrand is convex, Jensen’s inequality tells us Corollary 1.6 is strictly better for all n 4. that There is also a version of the Mahler≥ conjecture for asym- n metric convex bodies that contain the origin in their interiors. Vol (K K)◦ 2− (Vol K◦), − ≤ 3 with equality if and only if K is centrally symmetric. Combin- For example, the positive and negative pseudospheres H± ing these estimates with Corollary 1.6 establishes the claim. of R(a,b) are pseudo-Riemannian with signature (a 1,b) and a 1 b− a The numerical comparison to the simplex case uses the ele- (a,b 1). They are diffeomorphic to S − R and R b 1 − × × mentary volume formula S − . The diffeomorphisms can be chosen so that every fiber of the first factor is spacelike and every fiber of the second n+1 (n + 1) factor is timelike. v(∆n)= . (n!)2 If A and B are two sets in Rn, we define their geometric join A B to be the union of all line segments that connect a point in∗A and a point in B. The geometric join is a kind of partial Rn Section 4 is a digression in which we will compute invariant convex hull. If M is a closed manifold of codimension ⊂ n n 1 1, we define the filling M of M to be the compact region in R linking forms for the sphere S − and for hyperbolic space n 1 that it encloses. If A B is such a manifold, then A B is the H − . The computations are similar to Lemma 3.2 used in ∗ ∗ the proof of Theorem 1.4. We also give a second derivation filled join of A and B. The vector space Rn Rn has a relevant inner product (or in the sphere case using elementary geometry and probability × instead of Stokes’ theorem. dot product), with signature (n,n), given by the formula ~x1 ~y2 +~x2 ~y1 (~x1,~y1) (~x2,~y2)= · · . (1) · 2 Acknowledgments + The hyperboloids H and H− from the introduction are the positive and negative unit pseudospheres with respect to this The author would like to thank Tom Ilmanen, Misha inner product. Note that this inner product does not have de- Kapovich, Boaz Klartag, Bruce Kleiner, George Kozlowski, terminant 1. We will use the standard volume structure on Peter Michor, Vitali Milman, Andrew Waldron, and his par- Rn Rn rather± than the one induced by 1. (It may have been ents for helpful discussions. The author would especially a bit× clearer to put K in a vector space V and then define this like to thank Semyon Alesker, Dennis DeTurck, and Herman inner product on V V . We will use the notation K Rn Gluck for helping to check the main result. Finally, the author ∗ because it is standard⊕ in finite-dimensional convex geometr⊂ y.) would like to thank David Kazhdan and Joseph Bernstein for introducing him to the Bourgain-Milman theorem in 1987. H− K− + 2. AREVIEW H

+ + In this section, we review some relevant definitions and re- K K♦ K sults in Reference 13 and set up our new arguments. As in the introduction, let K Rn be a centrally symmetric convex ⊂ K body such that K and K◦ are both positively curved. (Note − that K is positively curved if and only if K◦ is.) Rn Rn Let V be an (non-degenerate) indefinite inner product space; we will write all inner products as dot products. Recall + n n that a non-zero vector~x V is spacelike if~x ~x > 0, timelike if Figure 1: The geometry of K , K−, and K♦ in R R . ~x ~x < 0, and null if~x ~x =∈ 0. (The definitions· of spacelike and × timelike· can be switched· in some contexts.) The positive and negative unit pseudospheres of V are the solutions sets to the In this section, we will establish the facts that K+ is a neck + equations ~x ~x = 1. Every indefinite inner product space is of H (to be defined precisely below), that K− is a neck of · ± R(a,b) + isometric to the standard inner product space with inner H−, and that their filled join K♦ = K K− is a starlike body + ∗ product that lies between H and H−. The overall geometry is shown in Figure 1. As the figure suggests, K is sometimes all of K ♦ × ~x ~y = x1y1 + x2y2 + + xaya K◦; for example, when K = Cn. (A useful exercise is to work · ··· out the geometry of K when K =C . Inthis case, K+ and K xa+1ya+1 xa+2ya+2 xa+bya+b ♦ 2 − − − −···− are non-planar octagons that visit all 16 vertices of K K , × ◦ with signature (a,b), for some a and b. which is a 4-cube.) But in other cases, K♦ is significantly Recall that a pseudo-Riemannian manifold M is a smooth smaller than K K◦. manifold with a smooth field of non-degenerate inner prod- Every point in×K+ is a pair consisting of a point~x K and a ∈ ucts, which is called a metric. The metric on M restricts to a dual vector~y K◦ that represents a supporting hyperplane of metric on every submanifold N M; however, the restriction ~x. In this situation,∈ ~x ∂K and ~y ∂K . Moreover, since K ⊂ ∈ ∈ ◦ may or may not be non-degenerate. A submanifold N of a and K◦ are both smooth,~x uniquely and smoothly determines pseudo-Riemannian manifold M is spacelike, timelike, or null ~y and vice versa. Therefore K+ is diffeomorphic to ∂K, andin n 1 if all of its non-zero tangent vectors are, respectively, space- particular, is diffeomorphic to the sphere S − . Likewise K− like, timelike, or null in their respective tangent spaces. is as well. 4

Lemma 2.1. K+ is a spacelike submanifold of H+ and K is that ρ Φ is a covering map and therefore a bijection by the − ◦ a timelike submanifold of H−. first claim. This is equivalent to the assertion of the lemma. It is elementary that ρ Φ is smooth on the smooth points Proof. (Sketch) The proof is from Reference 13. The con- of its domain. ◦ vex body K has a unique osculating ellipsoid E at the point ρ Φ + The claim that has degree 1 is not really necessary, be- ~x. Because E is an ellipsoid, E is a flat ellipsoid surface cause Sa+b 1 is simply◦ connected unless a b 1, in which + n n − = = in H R R , and in particular is a spacelike manifold. + ⊂ × case it is easy to show that N N− is a starlike quadrilateral. Since E osculates K at~x, it follows that One way to show it is that ρ ∗Φ varies continuously as N+ ◦ + + + and N− are varied. If we take N and N− to be orthogonal, T(~x,~y)K = T(~x,~y)E . + flat, and centered, then it can be seen directly that N N− is starlike and indeed convex, which implies that ρ Φ∗ has Thus, K+ is spacelike, and likewise K is timelike. − degree 1. ◦ In addition, K+ is a spacelike section of H+, in two senses. To confirm that ρ Φ has non-zero Jacobian away from N+ + ◦ Since H+ is a pseudo-Riemannian manifold of signature (n and N−, let ~x N and~y N−, and let + − ∈ ∈ 1,n), K is a spacelike section in the sense of having maximal + ~v1,~v2,...,~va 1 T~xN ~w1,~w2,...,~wb 1 T~yN− dimension. But also, the normal exponential map from any − ∈ − ∈ flat section K+ of H+ (such as E+ if E is an ellipsoid) makes be bases. Suppose that they are chosen so that H+ a bundle of hyperbolic spaces over a sphere. Then K+ ~x,~v1,~v2,...,~va 1 ~y,~w1,~w2,...,~wb 1 is a section of this bundle, because it is both spacelike and − − isotopic to K+.(K+ is isotopic to E+ by deforming K to E, are both positively oriented bases of maximal timelike and + and then all flat sections are isotopic.) Likewise K− has the spacelike planes. Given coordinates on N and N− that in- same properties in H−. duce these tangent bases, the Jacobian of ρ Φ at the point ◦ If V is any indefinite inner product space and H± are its (~x,~y,t) is positive and negative unit pseudospheres, define a neck of H± to be a spacelike or timelike submanifold N± which is a sec- J(~x,~y,t)= tion in both of the above senses. If V has signature (a,b), then + det (1 t)~v1,...,(1 t)~va 1,t~w1,...,t~wb 1,~y ~x,(1 t)~y+t~x N has dimension a 1 and N− has dimension b 1. − − − − − − − R(a,b) − (1 t)~v+t~w a+b For the main result, we will let V = . Each space- || − ||2 R(a,b) a 1 b 1 like a-plane in has a canonical orientation, as does each (1 t) − t − det ~v1,...,~va 1,~w1,...,~wb 1,~y,~x timelike b-plane. The group of isometries of R(a,b) that pre- = − − − , (1 t)~v+t~w a+b serve its total orientation is SO(a,b). The subgroup that sep- || − ||2 arately preserves the orientations of maximal spacelike and and so does not vanish. timelike plane is ISO(a,b); it is a connected Lie group. + + Let N and N− be necks of H and H−. π + + (H ) Lemma 2.2. The geometric join N N− is the boundary of a starlike body in Rn Rn. ∗ × π(N+) Proof. The argument is a refinement of the one in Refer- + ence 13, where it was argued that K K− is starlike except possibly on a set of measure 0. ∗ + + Let N ⊛ N− denote the abstract join of N and N− as defined in topology (and usually written N+ N ), and let ∗ − + (a,b) π + π + Φ : N ⊛ N− R Figure 2: (N ) is starlike and encloses the hole of (H ). → be the obvious continuous map with image N+ N , given by ∗ − the formula Finally at a point ~y N−, the map Φ has a local conelike structure. It is (tautologically)∈ smooth and non-singular along Φ(~x,~y,t) = (1 t)~x+t~y. the tangent space T N , while the direction of~y is annihilated − ~y − by the radial projection ρ. In the remaining directions, the Let map Φ is locally a conical extension of the restriction of the (a,b) a+b 1 projection ρ : R S − → (a,b) π : R T N− + ~y ⊥ be the tautological radial projection. → ~y h i The composition ρ Φ is a map between (a + b 1)- to the other neck N+. Because N+ is a spacelike section, spheres. We claim that◦ it has degree 1, that it is smooth− with π(N+) must be starlike. This implies that its cone extension + non-zero Jacobian except on N and N−, and that it is a local is a homeomorphism, and therefore that ρ Φ is a local home- homeomorphism at N+ and N . The last two claims imply omorphism at~y. The same argument applies◦ to~x N+. − ∈ 5

+ Lemma 2.2 implies that N and N− admit a filled join necessarily 1. Our Gauss linking form will be invariant under the non-compact group SO(a,b) rather than the usual rotation + N♦ = N N−. group SO(a + b). ∗ If X and Y are two manifolds, then differentials on X Y The lemma also means that the volume of N♦ can be ex- are doubly graded according to their degrees in the X and×Y pressed as an integral, by the method of infinitesimals: directions. The full differential d also splits as d = dx + dy according to differentiation along X and Y separately. This (a 1)!(b 1)! a 1 b 1 Vol N♦ = − − ~x ~y d~x∧ − d~y∧ − . (2) refines the de Rham complex Ω∗(X Y) into a double com- (a + b)! Z(~x,~y) ∧N+ ∧N ∧ + × ∈ × − plex. In our case, X = H and Y = H−. As before,we willuse + Remark. Equation (2) is different in several ways from the the vector coordinates~x H and~y H− using the inclusions R(a,b) ∈ ∈ formula in Reference 13. The most important change is that of H± in . the constant factor was not correct previously. In addition, the Lemma 3.1. Let ω be an ISO(a,b)-invariant differential form previous formula was for the volume of K , which is here + + ♥ on H H− of degree (a 1,b 1), and let N and N− be called K♦. Finally the old formula had d~x and d~y instead of closed,× smooth submanifolds− or− cycles of dimension a and b their wedge powers. While this was plausible notation, the + in H and H−. Also suppose that a,b > 1. Then wedge power notation is more literally correct and will be important to the proof of Theorem 1.4. ω ZN+ N Equation (2) is technically only true up to sign, but we can × − make it exactly true with suitable orientations for the necks. is invariant under chain homotopy of N+ and N if and only The differential forms in equation (2) are vector-valued and − if d d ω = 0. the wedge products are really “double wedges” in the algebra y x Remarks. Since dx and dy anticommute, the condition on ω is Λ (a,b) Ω (a,b) A = ∗(R ) ∗(R ). symmetric in ~x and ~y. In the context of double complexes, ω ⊗ can be called “weakly closed”. Even though both factors are graded-commutative algebras, Our form ω will actually be SO(a,b)-invariant. But we define multiplication in A by tensoring them as ungraded SO(a,b) is not connected when a,b > 0. Its connected sub- algebras. If ω1,ω2 A have degrees ( j1,k1) and ( j2,k2), then ∈ group ISO(a,b) is more natural for this lemma (and not really different). j1 j2+k1k2 ω1 ω2 = ( 1) ω2 ω1. ∧ − ∧ DeTurck and Gluck [6] show that the double integral of a ω For example, similar form (see Section 4) is chain-homotopy invariant if and only if it is weakly closed and ~x ~y = ~y ~x ~x d~y = d~y ~x d~x d~y = d~y d~x. ∧ − ∧ ∧ − ∧ ∧ ∧ dxω = dyσ dyω = dxκ Equation (2) is established by assuming simplex shapes for the for some forms σ and κ. This criterion does not require ω to infinitesimal elements d~x and d~y of N+ and N , so that the − be ISO(a,b)-invariant. corresponding infinitesimal element of N♦ is also a simplex (subtended by~x, d~x,~y, and d~y). Strictly speaking, the integral Proof. We will first consider the “if” direction, which is the on the right side of (2) is not scalar, but rather has a value one that we will need. Let in Λa+b(R(a,b)). However, it can be read as a scalar because Λa+b R(a,b) ( ) is 1-dimensional; we endow it with the standard ωx = ω. Z basis vector N− Then ~e1 ~e2 ~ea b. ∧ ∧···∧ + dωx = dxω ZN− 3. PROOF OF THEOREM 1.4 is a differential form on H+ of degree a. By Stokes’ theorem, The idea of the proof is to find a topological lower bound it is invariant under chain homotopy of N−, because dy anni- ω for the integral in equation (2). We can compactify the union hilates the integrand. In particular, d x is ISO(a,b)-invariant. + But there are no non-zero ISO(a,b)-invariant a-forms on H H− by radially projecting it onto the Euclidean-unit + a + + ∪ a+b 1 (a,b) H . They are sections of Λ (T ∗H ). Each fiber of T ∗H is a sphere S − in R , similar to the proof of Lemma 2.2. a+b 1 representation of the point stabilizer ISO(a 1,b) and is iso- The image is dense; the points in S − that are missed are − null vectors. Thus we have a compactification morphic to the defining representation V. The exterior power Λa(V) has no invariant vectors unless dimV = a. This can be + a+b 1 checked by passing to the action of the complexified Lie alge- H H− ∼= S − . ∪ bra so(a + b 1,C) on Λa(V ) C. The claim is then a basic Our lower bound will be proportional to a Gauss-type integral calculation in− the representation⊗ theory of complex simple Lie for the linking number of N+ and N in H+ H , which is algebras. Namely, Λa(V) C is an irreducible representation − ∪ − ⊗ 6 of so(a + b 1,C) unless a = b 1, in which case it is a di- yield the differential conditions rect sum of− two irreducible representations.− The only case with a b 0 in which it is the trivial representation is when ~x d~x = 0 ~y d~y = 0. (4) , > · · b = 1, which we have excluded by hypothesis. + The form ω is explicitly invariant under the diagonal action Thus, ωx is a closed form on H , so of SO(a,b), which is transitive on pairs of vectors (~x,~y) + ∈ H H− with a prescribed dot product. (Indeed, the given ωx = ω × + + form of ω is the general expression for an invariant form of ZN ZN N− × degree (a 1,b 1).) Therefore we can check the condition + − − is invariant under chain homotopy of N . By the same argu- dydxω = 0 for the standard vectors ment, it is also invariant under chain homotopy of N−. This concludes the “if” part of the lemma. ~x = (1,0,0,...,0) ~y = (sinhα,0,0,...,coshα,...,0). + + To sketch the “only if” part, suppose that N0 and N1 are + Here the two non-zero coordinates of ~y are y1 and ya+1 and two necks of H that agree except for a locus which is the α boundary of an a-dimensional blister L+. (By a blister, we is a hyperbolic angle (or rapidity). Then the conditions (4) mean any manifold with cusp boundary that connects two simplify to smooth manifolds that are only partly disjoint.) Likewise, sup- dx1 = 0 (sinhα)dy1 (coshα)dya+1 = 0. pose that N0− and N1− are two necks of H− that differ only at − a b-dimensional blister L . By two applications of Stokes’ − Let d~x and d~y be d~x and d~y with the first and (a + 1)st coordi- theorem, nates deleted. Then equation (3) becomes (term by term) e e p+q 0 = ∑ ( 1) ω = dydxω. a ω + + ( 1) dydx = − ZNp Nq ZL L p,q 0,1 × − × − − ∈ 2 a 1 b 1 (coshα) φ ′′dy1 dxa+1 d~x∧ − d~y∧ − + ∧ ∧ ∧ But then, the blisters L and L− can be arbitrarily small and a 1 b 1 can lie in any direction, so that in the limit, + (coshα)φ ′dya 1 dxa 1 e d~x∧ − e d~y∧ − + ∧ + ∧ ∧ a 1 b 1 ω b(coshα)φ ′dxa 1 dya 1 de~x∧ − de~y∧ − dydx = 0. − + ∧ + ∧ ∧ a 1 b 1 + a(sinhα)φ ′dy1 dxa 1 d~x∧ e− d~y∧ e− ∧ + ∧ ∧ a 1 b 1 + abφd~x∧ − d~y∧ − = 0 e e Lemma 3.2. If ∧ e e a 1 b 1 This expression so far uses the relations ω = φ(~x ~y)~x ~y d~x∧ − d~y∧ − · ∧ ∧ ∧ + dx1 = 0 dy1 dya+1 = 0 is an SO(a,b)-invariant differential form on H H− in ∧ (a,b) × R , then dydxω = 0 if and only if f (α)= φ(sinh(α)) sat- to eliminate terms in the expansion. If we apply the remaining isfies the ordinary differential equation relation

f + (a + b)(tanhα) f + ab f = 0. (sinhα)dy1 (coshα)dya 1 = 0 ′′ ′ − + and collect differential factors, we conclude that all of the Proof. We compute: terms are proportional to

a 1 b 1 a 1 b 1 dxω = (~y d~x)φ ′(~x ~y)~x ~y d~x∧ − d~y∧ − dy dx d~x∧ − d~y∧ − , · · ∧ ∧ ∧ 1 a+1 a b 1 ∧ ∧ ∧ φ(~x ~y)~y d~x∧ d~y∧ − . − · ∧ ∧ and the scalar factor yields thee equatione Then 2 (coshα) φ ′′ + (a + b + 1)(sinhα)φ ′ + abφ = 0. a 1 b 1 dydxω =(~x d~y)(~y d~x)φ ′′~x ~y d~x∧ − d~y∧ − · · ∧ ∧ ∧ However, the notation of this differential equation is mis- a 1 b 1 φ α + (d~y d~x)φ ′~x ~y d~x∧ − d~y∧ − leading, because is not directly a function of . Rather, · ∧ ∧ ∧ φ φ a 1 b = (t), where (~y d~x)φ ′~x d~x∧ − d~y∧ − · ∧ ∧ a b 1 α (~x d~y)φ ′~y d~x∧ d~y∧ − t =~x ~y = sinh . − · ∧ ∧ · a b φd~x∧ d~y∧ = 0. (3) If we let f (α)= φ(sinhα), then − ∧

In addition, the constraints f ′′ + (a + b)(tanhα) f ′ + ab f = 0

~x ~x = 1 ~y ~y = 1 is the corresponding ODE for f . · · − 7

Let f be the solution to the differential equation 1.0

f ′′ + (a + b)(tanhα) f ′ + ab f = 0 0.5 with f (0)= 1 and f ′(0)= 0. Lemmas 3.1 and 3.2 say that

+ 1 a 1 b 1 ℓ(N ,N−)= f (sinh− ~x ~y)~x ~y d~x∧ − d~y∧ − 1.5 1.0 0.5 0.5 1.0 1.5 ZN+ N · ∧ ∧ ∧ − − − × − 0.5 − is invariant under homotopy of the necks, and is therefore the same for all necks. By Lemma 3.3, f (α) < f (0)= 1 when 1.0 α = 0. Therefore − 6 + + ℓ(N ,N−) w(N ,N−). ≤ Figure 3: The even solution f (α) with a = b = 9. Equality is achieved if and only if ~x ~y = 0 always; in other + · words, when N and N− are orthogonal, flat, centered necks, as desired. Lemma 3.1 is false in the stated generality when a = 1 or α Lemma 3.3. If f ( ) satisfies the differential equation b = 1. Suppose that b = 1 (say). In this case, H− is a two- sheeted hyperboloid and a valid neck N consists of one point α − f ′′ + (a + b)(tanh ) f ′ + ab f = 0 on each sheet. Lemma 3.1 does hold, by a modified argument, if we either require N to be a neck, or if we assume an even and f (0)= 0, then f (α) < f (0) when α = 0. − ′ | | | | 6 solution f to the ODE in Lemma 3.2. However, we will give Figure 3 shows an example. the old geometric proof from Reference 13 and explain how it is actually the same proof. Let ~y be the difference between Proof. The differential equation for f is that of a damped har- the two points of N−; it is necessarily a timelike vector. After monic oscillator, where the damping is forwards in time for a Lorentz transformation, positive time α > 0, and backwards in time for negative time α < 0. Therefore if f ′(0), the oscillator will lose energy in ~y = (0,0,0,...,0,y) both directions and never again reach f (0). In detail, let ± with y 2, with equality when N− is centered. Then + ≥ v(N ,N−) is proportional to the area enclosed by the pro- 2 2 π + + ( f ′) + ab f jection of N onto the first a coordinates. Meanwhile H E = a 1 Ra + 2 projects onto the outside of the unit sphere S − in , and N must also wind around the hole, as in Figure 2. The enclosed be the energy of the oscillator. Then area is minimized when N+ is the boundary of the hole, i.e., + 2 when N is a flat, centered neck orthogonal to ~y. The inte- E′ = f ′′ f ′ + ab f ′ f = (a + b)(tanhα)( f ′) . + − gral v(N ,N−) also has a factor of y, and is minimized when α y = 2, which occurs when N− is centered. Thus E′ 0 when > 0 and vice-versa, with equality only ρ ≤ α α This completes the proof in all cases. We remark that if when f ′ = 0. Thus, E( ) < E(0) for = 0. Moreover, a a 1 6 is the radial projection of R onto S − and µ is Haar measure on Sa 1, then the pullback π (ρ (µ)) is the same form ω that 2E − ∗ ∗ f , is defined in Lemmas 3.1 and 3.2. The scalar kernel is | |≤ r ab 1 with equality when f 0, as is the case when α 0. These f (α)= , ′ = = (coshα)a relations together establish the lemma. which explicitly satisfies the equation and estimate of Proof of Theorem 1.4. When a,b > 1, we combine the lem- Lemma 3.3. mas to establish the theorem. We want to find the necks N+ and N− that minimize the integral 4. OTHER LINKING FORMS + a 1 b 1 w(N ,N−)= ~x ~y d~x∧ − d~y∧ − . + ZN N− ∧ ∧ ∧ × The calculation of Lemma 3.2 can be adapted to compute n 1 We orient the necks so that the integrand — not just the in- Gauss linking forms in the round (n 1)-sphere S − , or in n 1 − tegral — is positive. This is possible by the geometry estab- hyperbolic space H − . If M is one of these two geometries, (a 1,b 1) 2 lished in the proof of Lemma 2.2; namely, the vector spaces the form ω is now an element of Ω − − (M× ∆), where + 2 \ T~xN + ~x and T~yN− + ~y are maximal timelike and space- ∆ is the diagonal of M and a + b = n. The goal is to choose h i h i × like subspaces of R(a,b). ω so that it is invariant under Isom(M) and so that its double 8 integral is the linking number between (a 1)-dimensional Then the resulting ODE in f is: − and (b 1)-dimensional submanifolds N1 and N2 in M, − f ′′ + (a + b)(cothα) f ′ + ab f = 0. lk(N1,N2)= ω, ZN N There are two solutions, one even α and the other odd in α. 1× 2 When a = b = 2, the even and odd solutions are: or a proportionality. This goal is closely related to the problem of finding the coshα (sinhα) α(coshα) f (α)= f (α)= − . propagator of generalized electromagnetism, or the general- (sinhα)3 (sinhα)3 ized Biot-Savart law, in curved spaces. This question was re- cently considered by DeTurck and Gluck in S3 and H3 [7]. It If we rescale the odd solution to match the Gauss formula in is not exactly the same question as ours, because if ω is such flat R3 as α 0, we obtain a propagator, then it satisfies a Laplace-type equation instead → of an exterior derivative equation and a symmetry condition. coshα f (α)= . However, the propagator has to be an invariant linking form 4π(sinhα)3 by Ampere’s law. 2 This solution is also equivalent to the one given by Gluck and The configuration space M× ∆ is no longer a Cartesian product, but it is foliated both horizontally\ and vertically, so DeTurck. (It is the analytic continuation of the linking form 2 on the spherical manifold RP3. We could also analytically the de Rham complex Ω∗(M× ∆) still refines to a double complex. Lemma 3.1 still holds,\ except that the chain homo- continue equation (6) to obtain a different solution.) topies on N1 and N2 must be disjoint. This allows the integral There is also a geometric argument for equation (6). Let 3 3 ω ω N1 and N2 be two knots in S , and let ~p S be a point that of to be proportionalto lk(N1,N2), provided that dydx = 0. ∈ No other invariants are available for the allowed chain homo- does not lie on N1, N2, or N1. The cone C = C(N1,~p) over − 3 topies. N1 with vertex ~p is a well-defined geometric chain in S , us- n 1 (n,0) ing geodesic segments to connect p to points in N . Thus the We can suppose that M = S − is the unit sphere in R , ~ 1 n 1 (n 1,1) linking number lk N N equals the homological intersection or that M = H − is the positive unit pseudosphere in R − . ( 1, 2) In either case, we can write C(N1,~p) N2 of C with N2. This is still so if we choose ~p at random on∩ S3 with respect to Haar measure. Therefore if we ω φ a 1 b 1 = (~x ~y)~x ~y d~x∧ − d~y∧ − divide N1 and N2 into infinitesimal segments d~x and d~y, the · ∧ ∧ ∧ expected value of the homological intersection, and perform a calculation similar to the one in Lemma 3.2. The calculations are similar enough that we will just give the ω = E~p C(d~x,~p) d~y , (7) conclusions. ∩ n 1 If M = S − , then we let is an invariant linking form. ~x = (1,0,0,...,0) ~y = (cosα,sinα,0,0,...,0)

f (α)= φ(cosα). sinβ α −α d~x d~x sin | | Then the resulting ODE in f is: d~y sinβ α d~y sin | | f ′′ + (a + b)(cotα) f ′ ab f = 0. (5) α − In this case, the equation has the boundary condition that f is β non-singular at π. With this boundary condition, we obtain π Figure 4: The window of a geometric intersection in S3. a 1 b 1 ω = f (Vol S − )(Vol S − )lk(N1,N2), ZN N 2 1× 2 where Vol Sn denotes the volume of the round unit n-sphere. The expectation (7) can be computed explicitly. Define an- The constant factor is obtained by taking N1 and N2 to be or- gles α and β by n 1 thogonal great spheres in S − as a test case. Explicitly, if a = b = 2, cosα =~x ~y cosβ =~x ~p. · · (π α)(cosα) + (sinα) The cone C(d~x,~p) is an infinitesimally thin triangle which f (α)= − . (6) 4π2(sinα3) intersects d~y with some probability. If there is an intersection, its sign does not depend ~p, but only on the sign of This solution is equivalent to the one given by DeTurck and ~x ~y d~x d~y. There is no intersection when β < α. For Gluck [7]. ∧ ∧β α∧ n 1 fixed > , the probability of an intersection, times its sign, If M = H − , then we let is given by the formula ~x = (1,0,0,...,0) ~y = (coshα,sinhα,0,0,...,0) (sinβ α)(sinβ) f (α)= φ(coshα). − ~x ~y d~x d~y dβ. 2π2(sinα)3 ∧ ∧ ∧ ∧ 9

This can be seen by supposing that d~x and d~y are orthogonalto convex body K. This extra geometric property of K+ could be each other and to both~x and ~y. Then the intersection window important for other problems in convex geometry. for ~p is a rectangle which is Reference 13 discusses a different generalized bottleneck conjecture which is still open when a,b > 2. Instead of mini- sinβ α sinβ − d~x by d~y . mizing sinα | | sinα | | a 1 b 1 (See Figure 4.) The factor ~x ~y also produces a numerical ~x ~y d~x∧ − d~y∧ − , ∧ ZN+ N ∧ ∧ ∧ factor of sinα which must be cancelled. If we include varia- × − tion of β, the region for ~p is a brick of thickness dβ. To get we can minimize an overall expectation, we must integrate the volume of this β 3 π2 a 1 a 1 brick with respect to and divide by Vol S = 2 . Using (~x d~x∧ − ) (~y d~y∧ − ) Z + + N1 N2 ∧ · ∧ π (π α)(cosα) + (sinα) × (sinβ α)(sinβ)dβ = − , Zα − 2 for two necks of H+. The dot product on Λa(R(a,b)) in this integral is induced from the one on R(a,b). As before, we con- the final answer is jecture that the minimum occurs when the necks are flat and (π α)(cosα) + (sinα) centered, and in this case coincident rather than orthogonal. ω = − ~x ~y d~x d~y. + + 4π2(sinα)3 ∧ ∧ ∧ When N = K for a symmetric convex body K, the two integrals are equal. This formula agrees with equation (6). We can consider variations of the Mahler conjecture for complex and quaternionic convex bodies, i.e., the unit balls of complex and quaternionic norms. The natural conjecture 5. OPEN PROBLEMS is that the unit balls of the complex and quaternionic ℓ1 and ℓ∞ norms have the least Mahler volume. (The ℓ∞ ball is a 5.1. Odds and ends polydisk, a Cartesian power of the Euclidean unit ball or disk in the complex numbers or quaternions.) Hanner polytopes If the convex body K or its dual K◦ is not positively curved, also generalize analogously to Hanner bodies. If K is a com- + + then K is a weakly spacelike section of H , and K− is a plex or quaternionic convex body, then we can define not only z weakly timelike section of H−. In other words, both will have the necks K±, but more generally the neck K for any scalar some null tangents. In general if N± are weak necks of H± in z = 1. This extra geometry could lead to a variant of Theo- (a,b) + | | R , their join N N− might no longer be embeddedexcept rem 1.4 and an improvement to Corollary 1.6. ∗ + on a measure 0. For example, K K− is not embedded if K is a polytope. ∗ Nonetheless, formula (2) is still the volume of a region N♦. 5.2. Philosophy The necks are Lipschitz by virtue of being weakly spacelike ∞ and timelike, so the integrand is L . Moreover, the weak We can attempt to apply the philosophy behind Theo- necks can be approximated by strictly spacelike and timelike rem 1.4 and Corollary 1.6 to other problems in asymptotic necks, so the integrand is never negative; no volume is ever convex geometry. Suppose that we want to bound some subtracted. We conjecture that in this generality, flat necks affinely invariant function f (K) on convex bodies. Typically uniquely minimize the volume of N♦. This is immediate from f (K) is maximized (respectively, minimized) by ellipsoids; the proof of Theorem 1.4 for strict necks, but weak necks in- the question is to find a lower bound (respectively, an upper troduce new complications. bound) to say that all convex bodies K are “not too far from We do not know whether K♦ is always convex when K is round”. Our philosophy is to find another roundness statistic a centrally symmetric convex body. In light of the Mahler g(K) f (K) (respectively g(K) f (K)) which is minimized ≤ ≥ conjecture, we conjecture that K♦ has the most volume when (respectively maximized) when K is an ellipsoid. This yields K = Cn. (We do not even know this when n = 2.) A related a bound for the original statistic f (K). phenomenon is the fact that K = K K when K is a Hanner For example, the isotropic constant conjecture asserts that ♦ × ◦ polytope [10]. We conjecture that they are the only symmetric if K Rn is a convex body, its isotropic constant L(K) is convex bodies with this property, in light of the conjecture boundedabove⊂ by some constant C, independentof dimension that they are the only convex bodies with minimum Mahler [4]. (See also Reference 1). The centrally symmetric case of volume. the conjecture implies the general case [11]. The constant can For two general necks N±, N♦ need not be convex and its be defined by minimizing an expectation: volume is unbounded even in fixed dimensions. An upper bound may be possible if we suppose that N lies between def E [~x ~x] ♦ L(K)2 = min K′ , + + · 2/n H and H−. When a = b = n, then H and H− are contact K′=T(K) n(Vol K′) manifolds, and the necks that come from convex bodies are Legendrian. We conjecture the converse (and it may not be where the minimum is taken over all affine positions K′ = hard to prove), that every Legendrian neck of H+ is K+ for a T(K), and the expectation is taken with respect to a random 10 point~x K′. Itisalso nothardto showthat L(K) is minimized is maximized when K is an ellipsoid. We can even conjecture when K∈is an ellipsoid. (This result is due to Blaschke [2], but that the argument is very simple. We can minimize EK [~x ~x] over all K with fixed volume just by moving the measure· of K as qK(c)= PK K◦ [~x ~y c ~x K ~y K◦ ] close to the origin as possible.) × · ≥ || || || || is maximized when K is an ellipsoid; this would imply the Conjecture 5.1. If K Rn is centrally symmetric, then ⊂ conjecture for pK (c). It can be shown, using the technique of 2 osculating ellipsoids, that EK K [(~x ~y) ] × ◦ · + n 1 is maximized when K is an ellipsoid. (Vol K )(Vol Bn 1)ε −2 n 1 qK(1 ε)= − + o(ε −2 ), − v(K) A number of convex geometers, including Ball and Gi- annopolous, have long known that Conjecture 5.1 implies + and that pK(1 ε) has similar asymptotics. Here Vol K is the the isotropic constant conjecture. It is not hard to show [8, +− Prop. 1.2.3] that volume of K as a spacelike, and therefore Riemannian, submanifold of Rn Rn with the inner product 1. So we conjec- × + 2 2 2 2 2/n 2/n ture that ellipsoids maximize the ratio (Vol K )/v(K), even EK K [(~x ~y) ] n L(K) L(K◦) (Vol K) (Vol K◦) . × ◦ · ≥ though they maximize the denominator. A related conjecture The centrally symmetric isotropic constant conjecture then is the following: follows by combining Conjecture 5.1, Theorem 1.3, and + + R(a,b) Blaschke’s inequality applied to L(K ). On the negative side, Conjecture 5.2. If N H is a neck, then its Rie- ◦ +⊂ ⊂ + Conjecture 5.1 also sharpens the Blaschke-Santalótheorem, mannian volume Vol N is maximized when N is flat. so it is more difficult and perhaps less likely. We can propose even stronger and less likely conjectures We do not know if Conjecture 5.2 is hard to prove, or even than Conjecture 5.1. We conjecture that for every 0 < c < 1, genuinely an open problem, in the context of current methods the probability in differential geometry [14]. Note that if N+ is only a weak neck, then Vol N+ can van- + pK(c)= PK K [~x ~y c] ish. For example, if K is a polytope, then Vol K = 0. × ◦ · ≥

[1] Keith Ball, Isometric problems in ℓp and sections of convex sets, arXiv:math.MG/0312475. Ph.D. thesis, Trinity College, 1986. [12] Greg Kuperberg, A low-technology estimate in convex geome- [2] Wilhelm Blaschke, Uber¨ affine Geometry XIV: eine minimum try, Internat. Math. Res. Notices 1992 (1992), no. 9, 181–183, Aufgabe für Legendres trägheits Ellipsoid, Ber. verh. sächs. arXiv:math.MG/9211216. Akad. d. Wiss. 70 (1918), 72–75. [13] , The bottleneck conjecture, Geom. Topol. 3 (1999), [3] , Vorlesungen über Differentialgeometrie II., Springer 119–135, arXiv:math.MG/9811119. Wien, Berlin-Heidelberg-New York, 1923. [14] Zhenyang Li and Yuguang Shi, Maximal slice in anti-de Sitter [4] Jean Bourgain, On high-dimensional maximal functions associ- space, arXiv:math.DG/0609278. ated to convex bodies, Amer. J. Math. 108 (1986), no. 6, 1467– [15] Kurt Mahler, Ein übertragungsprinzip für konvexe körper, Ca- 1476. sopis Pêst. Mat. Fys. 68 (1939), 93–102. [5] Jean Bourgain and Vitaly D. Milman, New volume ratio proper- [16] Gilles Pisier, The volume of convex bodies and Banach space ties for convex symmetric bodies in Rn, Invent. Math. 88 (1987), geometry, Cambridge Tracts in Mathematics, vol. 94, Cam- no. 2, 319–340. bridge University Press, 1989. [6] Dennis DeTurck and Herman Gluck, personal communication. [17] Shlomo Reisner, Zonoids with minimal volume-product, Math. [7] , The Gauss linking integral on the 3-sphere and in hy- Z. 192 (1986), no. 3, 339–346. perbolic 3-space, arXiv:math.GT/0406276. [18] C. A. Rogers and G. C. Shephard, The difference body of a con- [8] Apostolos Giannopoulos, Notes on isotropic convex bodies, vex body, Arch. Math. 8 (1957), 220–233. preprint. [19] Jean Saint-Raymond, Sur le volume des corps convexes symét- [9] Yehoram Gordon, Mathieu Meyer, and Shlomo Reisner, riques, Initiation seminar on analysis, 20th year: 1980/1981 Zonoids with minimal volume-product—a new proof, Proc. (G. Choquet, M. Rogalsky, and J. Saint-Raymond, eds.), Exp. Amer. Math. Soc. 104 (1988), no. 1, 273–276. No. 11, 25, Univ. Paris VI, Paris, 1981. [10] Olof Hanner, Intersections of translates of convex bodies, Math. [20] Luis A. Santaló, Un invariante afin para los cuerpos convexos Scand. 4 (1956), 65–87. del espacio de n dimensiones, Portugaliae Math. 8 (1949), 155– [11] Boaz Klartag, An isomorphic version of the slicing 161. problem, J. Funct. Anal. 218 (2005), no. 2, 372–394,