Introduction
The Novikov conjecture asserts that the higher signatures of closed oriented smooth manifolds are invariant under orientation-preserving homotopy equivalences. Noncommutative geometry provides an approach to the study of the Novikov conjecture via higher index theory. More precisely, for a countable discrete group Γ \Gamma Γ , the Novikov conjecture for closed oriented smooth manifolds with fundamental group Γ \Gamma Γ follows from the rational strong Novikov conjecture for Γ \Gamma Γ . The latter asserts that the rational Kasparov–Baum–Connes assembly map
K ∗ ( B Γ ) ⊗ Z Q ⟶ K ∗ ( C r ∗ Γ ) ⊗ Z Q K_{*}(B\Gamma) \otimes_{\mathbb{Z}} \mathbb{Q} \longrightarrow K_{*}(C_{r}^{*}\Gamma) \otimes_{\mathbb{Z}} \mathbb{Q} K ∗ ( B Γ ) ⊗ Z Q ⟶ K ∗ ( C r ∗ Γ ) ⊗ Z Q
is injective, where K ∗ ( B Γ ) K_{*}(B\Gamma) K ∗ ( B Γ ) denotes the K K K -homology of the classifying space B Γ B\Gamma B Γ for the group Γ \Gamma Γ and K ∗ ( C r ∗ Γ ) K_{*}(C_{r}^{*}\Gamma) K ∗ ( C r ∗ Γ ) denotes the K K K -theory of the reduced group C ∗ C^{*} C ∗ -algebra C r ∗ Γ C_{r}^{*}\Gamma C r ∗ Γ associated to Γ \Gamma Γ . Although the Novikov conjecture has been established for many classes of
Key words and phrases. smooth diffeomorphisms, higher-order derivative, coarse embedding, Novikov conjecture.
The first author is partially supported by Shanghai Pujiang Project, No. 25PJA033.
The second author was supported in part by Grant for Excellent Young Scholars in Tianyuan Mathematics. groups, it remains open in general for a natural and important class: countable subgroups of diffeomorphism groups of compact smooth manifolds.
Inspired by Connes’ theorem [5 ] on higher signatures associated with Gelfand–Fuchs classes of diffeomorphism groups, Gong, Wu and Yu [8 ] introduced a notion of geometric discreteness to describe the extent of derivative distortion for the volume-preserving subgroups of Diff ( M ) \operatorname{Diff}(M) Diff ( M ) . They proved that every such geometrically discrete subgroup satisfies the rational strong Novikov conjecture. Gong, Wu, Xie and Yu [7 ] subsequently removed the volume-preserving assumption by introducing μ \mu μ -discreteness for subgroups of Diff ( M ) \operatorname{Diff}(M) Diff ( M ) with respect to a Borel probability measure μ \mu μ on M M M . They also proved the rational strong Novikov conjecture for every countable μ \mu μ -discrete subgroup of Diff ( M ) \operatorname{Diff}(M) Diff ( M ) .
Both geometric discreteness and μ \mu μ -discreteness are defined in terms of the first derivatives of diffeomorphisms.
In this paper, we use higher-order derivatives to study the diffeomorphisms. In local coordinates, the ordinary derivative records only the linear term of the Taylor expansion of a diffeomorphism. However, the r r r -th order derivative encodes all partial derivatives of orders 1 , … , r 1,\ldots,r 1 , … , r , thereby retaining the nonlinear terms through degree r r r . This leads to the notion of higher-order μ \mu μ -discreteness. It turns out that lower-order μ \mu μ -discreteness implies higher-order μ \mu μ -discreteness, but the converse fails in general, as the example in § 4 shows. Thus, higher-order μ \mu μ -discreteness allows us to detect a strictly larger class of subgroups of Diff ( M ) \operatorname{Diff}(M) Diff ( M ) than lower-order μ \mu μ -discreteness does.
Since our framework is based on coarse embeddings of the relevant spaces into L p L^p L p -spaces, it yields not only the strong Novikov conjecture but also the coarse Baum–Connes conjecture for all higher-order μ \mu μ -discrete groups.
Recall that a map h : ( X , d X ) → ( Y , d Y ) h:(X,d_X)\to(Y,d_Y) h : ( X , d X ) → ( Y , d Y ) between metric spaces is a coarse embedding [9 ] if there exist nondecreasing functions ρ − , ρ + : [ 0 , ∞ ) → [ 0 , ∞ ) \rho_-,\rho_+:[0,\infty)\to[0,\infty) ρ − , ρ + : [ 0 , ∞ ) → [ 0 , ∞ ) , with ρ − ( t ) → ∞ \rho_-(t)\to\infty ρ − ( t ) → ∞ as t → ∞ t\to\infty t → ∞ , such that
ρ − ( d X ( x , y ) ) ≤ d Y ( h ( x ) , h ( y ) ) ≤ ρ + ( d X ( x , y ) ) , x , y ∈ X . \rho_-(d_X(x,y)) \le d_Y(h(x),h(y)) \le\rho_+(d_X(x,y)), \qquad x,y\in X. ρ − ( d X ( x , y )) ≤ d Y ( h ( x ) , h ( y )) ≤ ρ + ( d X ( x , y )) , x , y ∈ X .
The following is our main result.
Theorem 1.1. Let M M M be a closed smooth manifold, and Γ < Diff ( M ) \Gamma<\operatorname{Diff}(M) Γ < Diff ( M ) an ( r , p ) (r,p) ( r , p ) -th order μ \mu μ -discrete subgroup. Then the strong Novikov conjecture holds for Γ \Gamma Γ , i.e. the Baum–Connes assembly map
μ : K K ∗ Γ ( E Γ ) ⟶ K ∗ ( C ∗ ( Γ ) ) \mu:KK_*^\Gamma(E\Gamma)\longrightarrow K_*(C^*(\Gamma)) μ : K K ∗ Γ ( E Γ ) ⟶ K ∗ ( C ∗ ( Γ )) is injective, where E Γ E\Gamma E Γ is the universal space for proper actions of the group Γ \Gamma Γ .
Moreover, the coarse Baum–Connes conjecture holds for the underlying metric space ∣ Γ ∣ |\Gamma| ∣Γ∣ associated to any proper left-invariant metric on Γ \Gamma Γ , i.e. the coarse assembly map
μ ∣ Γ ∣ : lim R → ∞ K ∗ ( P R ( ∣ Γ ∣ ) ) ⟶ K ∗ ( C ∗ ( ∣ Γ ∣ ) ) \mu_{|\Gamma|}:\lim_{R\to\infty}K_*(P_R(|\Gamma|))\longrightarrow K_*(C^*(|\Gamma|)) μ ∣Γ∣ : R → ∞ lim K ∗ ( P R ( ∣Γ∣ )) ⟶ K ∗ ( C ∗ ( ∣Γ∣ )) is an isomorphism.
We emphasize that our result establishes integral rather than merely rational injectivity of the Baum–Connes assembly map.
In Section 2, we introduce higher-order μ \mu μ -discreteness and higher-order orbit map. The proof of our main theorem fits into a more general construction that goes beyond Hilbert–Hadamard spaces. Thus, we shall first introduce a general type of spaces L s 0 p ( Ω , μ , E ) L^p_{s_0}(\Omega,\mu,E) L s 0 p ( Ω , μ , E ) , where E E E is a snowflake metric bundle over Ω \Omega Ω satisfying suitable assumptions, s 0 s_0 s 0 is a base section, and p p p need not be 2. These spaces can be highly nonlinear and need not be Hilbert–Hadamard spaces. In § 3 , we use an approach different from that of Gong, Wu, Xie and Yu [7 ] to obtain a more general result than Theorem 1.1. Our main theorem then follows immediately. In the final section, we present examples for which the first-order criterion fails but a second-order criterion applies.
AI disclosure. ChatGPT-6 was used solely for language editing and grammatical refinement during the preparation of this manuscript. All mathematical proofs and core results were developed by the authors.
Higher-order μ \mu μ -discreteness and higher-order orbit map
In this section, we first recall the notions of higher-order tangent bundles and higher-order derivatives. We then introduce L p L^p L p -direct integrals of metric bundles, which provide a common setting for higher-order μ \mu μ -discreteness and higher-order orbit map studied later. For a countable subgroup Γ < Diff ( M ) \Gamma< \operatorname{Diff}(M) Γ < Diff ( M ) , higher-order μ \mu μ -discreteness is defined by measuring how far the higher-order derivatives of elements of Γ \Gamma Γ deviate from isometries.
Higher-order derivatives
The ordinary derivative map between tangent bundles captures only the first-order linearization of a smooth map: in local coordinates, it retains the linear term of the Taylor expansion and discards all quadratic and higher-order terms. In contrast, the r r r -th order derivative map retains the full Taylor data through order r r r . More precisely, its value at a point is determined by all partial derivatives of orders 1 , … , r 1,\ldots,r 1 , … , r , and therefore encodes the entire Taylor expansion of a smooth map truncated at degree r r r . Thus, at the first-order level one sees only infinitesimal linear behavior, whereas at higher-order levels one also sees the successive nonlinear terms of the Taylor expansion. This additional information allows us to study a broader class of subgroups of Diff ( M ) \operatorname{Diff}(M) Diff ( M ) . In this subsection, we review some standard constructions following [13 Section 12] .
Definition 2.1. Let M M M and N N N be smooth manifolds, and let r ≥ 0 r \ge0 r ≥ 0 be an integer. Two smooth curves
ϕ , ψ : R ⟶ N \phi,\psi:\mathbb{R}\longrightarrow N ϕ , ψ : R ⟶ N are said to have r r r -th order contact at 0 0 0 , written
ϕ ∼ r ψ , \phi\sim_r\psi, ϕ ∼ r ψ , if, for every f ∈ C ∞ ( N , R ) f\in C^\infty(N,\mathbb{R}) f ∈ C ∞ ( N , R ) and every k = 0 , 1 , … , r k=0,1,\ldots,r k = 0 , 1 , … , r ,
d k d t k ∣ t = 0 ( f ∘ ϕ − f ∘ ψ ) ( t ) = 0. \left.\frac{d^k}{dt^k}\right|_{t=0}(f\circ\phi-f\circ\psi)(t)=0. d t k d k t = 0 ( f ∘ ϕ − f ∘ ψ ) ( t ) = 0. Obviously, ∼ r \sim_r ∼ r is an equivalence relation. Two smooth maps
ϕ , ψ : M ⟶ N \phi,\psi:M\longrightarrow N ϕ , ψ : M ⟶ N are said to determine the same r r r -jet at x ∈ M x\in M x ∈ M if, for every smooth curve h : R → M h:\mathbb{R}\to M h : R → M with h ( 0 ) = x h(0)=x h ( 0 ) = x , the curves ϕ ∘ h \phi\circ h ϕ ∘ h and ψ ∘ h \psi\circ h ψ ∘ h have r r r -th order contact at zero, i.e.
ϕ ∘ h ∼ r ψ ∘ h . \phi\circ h\sim_r\psi\circ h. ϕ ∘ h ∼ r ψ ∘ h . In this case, we write
j x r ϕ = j x r ψ . j_x^r\phi=j_x^r\psi. j x r ϕ = j x r ψ . The equivalence class j x r ϕ j_x^r\phi j x r ϕ is called the r r r -jet of ϕ \phi ϕ at x x x , or an r r r -jet of M M M into N N N . The set of all such r r r -jets is denoted by
J r ( M , N ) . J^r(M,N). J r ( M , N ) . The points x x x and ϕ ( x ) \phi(x) ϕ ( x ) are called the source and target of j x r ϕ ∈ J r ( M , N ) j_x^r\phi\in J^r(M,N) j x r ϕ ∈ J r ( M , N ) respectively. By J x r ( M , N ) J_x^r(M,N) J x r ( M , N ) or J r ( M , N ) y J^r(M,N)_y J r ( M , N ) y , we mean the set of all r r r -jets of M M M into N N N with source x ∈ M x\in M x ∈ M or target y ∈ N y\in N y ∈ N respectively, and we write
J x r ( M , N ) y = J x r ( M , N ) ∩ J r ( M , N ) y . J_x^r(M,N)_y=J_x^r(M,N)\cap J^r(M,N)_y. J x r ( M , N ) y = J x r ( M , N ) ∩ J r ( M , N ) y . The space
T r M : = J r ( M , R ) 0 T^rM:=J^r(M,\mathbb{R})_0 T r M := J r ( M , R ) 0
is called the space of all ( 1 , r ) (1,r) ( 1 , r ) -covelocities on M M M .
Since R \mathbb{R} R is a vector space, T r ∗ M ⟶ π M T^{r*}M \mathbin{\overset{\pi}{\longrightarrow}} M T r ∗ M ⟶ π M is a vector bundle, where π ( j x r ϕ ) = x \pi(j_x^r\phi)=x π ( j x r ϕ ) = x and each fiber T x r ∗ M = J x r ( M , R ) 0 T_x^{r*}M=J_x^r(M,\mathbb{R})_0 T x r ∗ M = J x r ( M , R ) 0 is equipped with the operations:
j x r ϕ + j x r ψ = j x r ( ϕ + ψ ) and a ⋅ j x r ϕ = j x r ( a ⋅ ϕ ) , a ∈ R . j_x^r\phi+j_x^r\psi=j_x^r(\phi+\psi)\quad\text{and}\quad a\cdot j_x^r\phi=j_x^r(a\cdot\phi),\quad a\in\mathbb{R}. j x r ϕ + j x r ψ = j x r ( ϕ + ψ ) and a ⋅ j x r ϕ = j x r ( a ⋅ ϕ ) , a ∈ R .
It is called the ( 1 , r ) (1,r) ( 1 , r ) -covelocities bundle on M M M .
Consider the dual vector bundle
T ( r ) M ≔ ( T r ∗ M ) ∗ T^{(r)}M\coloneqq(T^{r*}M)^* T ( r ) M : = ( T r ∗ M ) ∗
of the ( 1 , r ) (1,r) ( 1 , r ) -covelocities bundle on M M M . For every smooth map ϕ : M → N \phi:M\to N ϕ : M → N , the jet composition
A ⟼ A ∘ j x r ϕ , x ∈ M , A ∈ T ϕ ( x ) r ∗ N , A\longmapsto A\circ j_x^r\phi,\quad x\in M,\quad A\in T_{\phi(x)}^{r*}N, A ⟼ A ∘ j x r ϕ , x ∈ M , A ∈ T ϕ ( x ) r ∗ N ,
defines a linear map
λ ( j x r ϕ ) : T ϕ ( x ) r ∗ N ⟶ T x r ∗ M , \lambda(j_x^r\phi):T_{\phi(x)}^{r*}N\longrightarrow T_x^{r*}M, λ ( j x r ϕ ) : T ϕ ( x ) r ∗ N ⟶ T x r ∗ M ,
and hence the dual map
( λ ( j x r ϕ ) ) ∗ : T x ( r ) M ⟶ T ϕ ( x ) ( r ) N . \bigl(\lambda(j_x^r\phi)\bigr)^*:T_x^{(r)}M\longrightarrow T_{\phi(x)}^{(r)}N. ( λ ( j x r ϕ ) ) ∗ : T x ( r ) M ⟶ T ϕ ( x ) ( r ) N .
We denote
D x ( r ) ϕ = ( λ ( j x r ϕ ) ) ∗ . D_x^{(r)}\phi=\bigl(\lambda(j_x^r\phi)\bigr)^*. D x ( r ) ϕ = ( λ ( j x r ϕ ) ) ∗ .
They determine a smooth bundle map
D ( r ) ϕ : T ( r ) M ⟶ T ( r ) N . D^{(r)}\phi:T^{(r)}M\longrightarrow T^{(r)}N. D ( r ) ϕ : T ( r ) M ⟶ T ( r ) N .
The assignments M ↦ T ( r ) M M\mapsto T^{(r)}M M ↦ T ( r ) M and ϕ ↦ D ( r ) ϕ \phi\mapsto D^{(r)}\phi ϕ ↦ D ( r ) ϕ define a covariant functor. In particular,
D ( r ) ( ψ ∘ ϕ ) = D ( r ) ψ ∘ D ( r ) ϕ D^{(r)}(\psi\circ\phi)=D^{(r)}\psi\circ D^{(r)}\phi D ( r ) ( ψ ∘ ϕ ) = D ( r ) ψ ∘ D ( r ) ϕ
for all smooth maps ϕ : M 1 → M 2 \phi:M_1\to M_2 ϕ : M 1 → M 2 and ψ : M 2 → M 3 \psi:M_2\to M_3 ψ : M 2 → M 3 . The bundle T ( r ) M T^{(r)}M T ( r ) M is called r r r -th order tangent bundle on M M M , the elements of T ( r ) M T^{(r)}M T ( r ) M are called r r r -th order tangent vectors on M M M , D ( r ) ϕ D^{(r)}\phi D ( r ) ϕ is called r r r -th order derivative map associated to ϕ : M → N \phi:M\to N ϕ : M → N , and D x ( r ) ϕ D_x^{(r)}\phi D x ( r ) ϕ is called the r r r -th order derivative of ϕ \phi ϕ at x x x .
The above construction generalizes the usual tangent bundle and derivative map since the following diagram commutes:
T ( 1 ) M → D ( 1 ) ϕ T ( 1 ) N ≅ ↓ ↓ ≅ T M → D ϕ T N \begin{CD}
T^{(1)}M @>{D^{(1)}\phi}>> T^{(1)}N \\
@V{\cong}VV @VV{\cong}V \\
TM @>{D_\phi}>> TN
\end{CD} T ( 1 ) M ≅ ↓ ⏐ T M D ( 1 ) ϕ D ϕ T ( 1 ) N ↓ ⏐ ≅ T N
where the vertical arrows are the canonical isomorphisms and D ϕ : T M → T N D_\phi:TM\to TN D ϕ : T M → T N is the usual derivative map.
Lp-direct integral and snowflake metric bundle
In this subsection, we introduce metric bundles and their L p L^p L p -direct integrals. We then focus on snowflake metric bundles, which provide a natural setting for the study of higher-order Riemannian metric bundles.
Let
π : E ⟶ Ω \pi:E\longrightarrow\Omega π : E ⟶ Ω
be a continuous surjective map between topological spaces. Assume that each fiber E z : = π − 1 ( z ) E_z:=\pi^{-1}(z) E z := π − 1 ( z ) carries a metric d z d_z d z compatible with its subspace topology. We call E → Ω E\to\Omega E → Ω a metric bundle if there exists a metric space ( X , d X ) (X,d_X) ( X , d X ) such that every point of Ω \Omega Ω has an open neighborhood U U U with a homeomorphism
Φ U : π − 1 ( U ) ⟶ U × X , \Phi_U:\pi^{-1}(U)\longrightarrow U\times X, Φ U : π − 1 ( U ) ⟶ U × X ,
whose restriction to ( E z , d z ) (E_z,d_z) ( E z , d z ) is an isometry onto { z } × X \{z\}\times X { z } × X for every z ∈ U z\in U z ∈ U . We call Φ U \Phi_U Φ U a local trivialization and ( X , d X ) (X,d_X) ( X , d X ) the typical fiber .
A metric bundle π : E → Ω \pi: E \to\Omega π : E → Ω is called measurable if its fiberwise distance function
δ : E × Ω E ⟶ [ 0 , ∞ ) , δ ( u , v ) : = d π ( u ) ( u , v ) , \delta: E \mathbin{\times_{\Omega}} E \longrightarrow[0,\infty), \qquad\delta(u,v) := d_{\pi(u)}(u,v), δ : E × Ω E ⟶ [ 0 , ∞ ) , δ ( u , v ) := d π ( u ) ( u , v ) ,
is measurable with respect to the restriction of B ( E ) ⊗ B ( E ) \mathcal{B}(E) \otimes\mathcal{B}(E) B ( E ) ⊗ B ( E ) to
E × Ω E : = { ( u , v ) ∈ E × E : π ( u ) = π ( v ) } , E \mathbin{\times_{\Omega}} E := \{(u,v) \in E \times E : \pi(u) = \pi(v)\}, E × Ω E := {( u , v ) ∈ E × E : π ( u ) = π ( v )} ,
where B ( E ) \mathcal{B}(E) B ( E ) denotes the Borel σ \sigma σ -algebra of E E E , and B ( E ) ⊗ B ( E ) \mathcal{B}(E) \otimes\mathcal{B}(E) B ( E ) ⊗ B ( E ) is the product σ \sigma σ -algebra. Let μ \mu μ be a Borel measure on Ω \Omega Ω . A section s : Ω → E s : \Omega\to E s : Ω → E is called μ \mu μ -measurable if s − 1 ( U ) s^{-1}(U) s − 1 ( U ) is μ \mu μ -measurable for every open subset U ⊆ E U \subseteq E U ⊆ E . The measurability of the fiberwise distance function implies that, for any two μ \mu μ -measurable sections s , t s,t s , t , the function
z ⟼ d z ( s ( z ) , t ( z ) ) z \longmapsto d_z\bigl(s(z),t(z)\bigr) z ⟼ d z ( s ( z ) , t ( z ) )
is μ \mu μ -measurable.
Definition 2.2. Let π : E → Ω \pi: E \to\Omega π : E → Ω be a measurable metric bundle, and let μ \mu μ be a Borel measure on Ω \Omega Ω . Fix a μ \mu μ -measurable base section s 0 : Ω → E s_0 : \Omega\to E s 0 : Ω → E . For p ∈ [ 1 , + ∞ ) p \in[1,+\infty) p ∈ [ 1 , + ∞ ) , the L p L^p L p -direct integral of E E E over ( Ω , μ ) (\Omega,\mu) ( Ω , μ ) based at s 0 s_0 s 0 , denoted by
L s 0 p ( Ω , μ , E ) , L^p_{s_0}(\Omega,\mu,E), L s 0 p ( Ω , μ , E ) , is the space of equivalence classes of μ \mu μ -measurable sections s s s satisfying
∫ Ω d z ( s ( z ) , s 0 ( z ) ) p d μ ( z ) < ∞ . \int_{\Omega} d_z\bigl(s(z),s_0(z)\bigr)^p\,d\mu(z)<\infty. ∫ Ω d z ( s ( z ) , s 0 ( z ) ) p d μ ( z ) < ∞. Two sections are equivalent if and only if they agree μ \mu μ -almost everywhere. Moreover, the Minkowski inequality implies that the formula
d L p ( [ s ] , [ t ] ) = ( ∫ Ω d z ( s ( z ) , t ( z ) ) p d μ ( z ) ) 1 / p d_{L^p}\bigl([s],[t]\bigr)=\left(\int_{\Omega}d_z\bigl(s(z),t(z)\bigr)^p\,d\mu(z)\right)^{1/p} d L p ( [ s ] , [ t ] ) = ( ∫ Ω d z ( s ( z ) , t ( z ) ) p d μ ( z ) ) 1/ p defines a well-defined metric on L s 0 p ( Ω , μ , E ) L^p_{s_0}(\Omega,\mu,E) L s 0 p ( Ω , μ , E ) .
In general, the metric space L s 0 p ( Ω , μ , E ) L^p_{s_0}(\Omega,\mu,E) L s 0 p ( Ω , μ , E ) depends on the base section s 0 s_0 s 0 . For the bundles Riem ( r ) ( M ) → M \operatorname{Riem}^{(r)}(M) \to M Riem ( r ) ( M ) → M considered in the next subsection, when M M M is compact and μ \mu μ is finite, the resulting metric space is independent of the choice of continuous base section. Nevertheless, we retain the base section in the notation to make clear which section is pushed forward by diffeomorphisms in the orbit maps introduced in § 2.4 .
coincides isometrically with L f 0 p ( Ω , μ , X ) L^p_{f_0}(\Omega,\mu,X) L f 0 p ( Ω , μ , X ) . Here L f 0 p ( Ω , μ , X ) L^p_{f_0}(\Omega,\mu,X) L f 0 p ( Ω , μ , X ) is called the L p L^p L p -continuum product of X X X over ( Ω , μ ) (\Omega,\mu) ( Ω , μ ) based at f 0 f_0 f 0 . It consists of equivalence classes of μ \mu μ -measurable maps f : Ω → X f:\Omega\to X f : Ω → X satisfying
∫ Ω d X ( f ( z ) , f 0 ( z ) ) p d μ ( z ) < ∞ , \int_{\Omega} d_X(f(z),f_0(z))^p\,d\mu(z)<\infty, ∫ Ω d X ( f ( z ) , f 0 ( z ) ) p d μ ( z ) < ∞ ,
and is equipped with the metric
d L p ( [ f ] , [ g ] ) = ( ∫ Ω d X ( f ( z ) , g ( z ) ) p d μ ( z ) ) 1 / p . d_{L^p}([f],[g])=\left(\int_{\Omega}d_X(f(z),g(z))^p\,d\mu(z)\right)^{1/p}. d L p ([ f ] , [ g ]) = ( ∫ Ω d X ( f ( z ) , g ( z ) ) p d μ ( z ) ) 1/ p .
If, in addition, X X X is a Banach space equipped with the norm metric d X ( x , y ) = ∥ x − y ∥ d_X(x,y)=\lVert x-y\rVert d X ( x , y ) = ∥ x − y ∥ , then its separability ensures that every μ \mu μ -measurable map Ω → X \Omega\to X Ω → X is Bochner measurable (see [1 Appendix K] ). Thus, for the zero base section s 0 ( z ) = ( z , 0 ) s_0(z)=(z,0) s 0 ( z ) = ( z , 0 ) , the space L s 0 p ( Ω , μ , Ω × X ) L^p_{s_0}(\Omega,\mu,\Omega\times X) L s 0 p ( Ω , μ , Ω × X ) (i.e., L 0 p ( Ω , μ , X ) L^p_0(\Omega,\mu,X) L 0 p ( Ω , μ , X ) ) is canonically isometric to the classical Bochner space L p ( Ω , μ , X ) L^p(\Omega,\mu,X) L p ( Ω , μ , X ) .
We next introduce the notion of snowflake embedding. Let ( X , d X ) (X,d_X) ( X , d X ) and ( Y , d Y ) (Y,d_Y) ( Y , d Y ) be metric spaces. For α ∈ ( 0 , 1 ] \alpha\in(0,1] α ∈ ( 0 , 1 ] , a map h : X → Y h:X\to Y h : X → Y is called an α \alpha α -snowflake embedding if
h : ( X , d X α ) ⟶ ( Y , d Y ) h:(X,d_X^\alpha)\longrightarrow(Y,d_Y) h : ( X , d X α ) ⟶ ( Y , d Y )
is a bi-Lipschitz embedding. Equivalently, there exist constants 0 < c ≤ C < ∞ 0<c\le C<\infty 0 < c ≤ C < ∞ such that
c d X ( x , y ) α ≤ d Y ( h ( x ) , h ( y ) ) ≤ C d X ( x , y ) α c d_X(x,y)^\alpha\le d_Y(h(x),h(y))\le C d_X(x,y)^\alpha c d X ( x , y ) α ≤ d Y ( h ( x ) , h ( y )) ≤ C d X ( x , y ) α
for all $x,y\in X. If such a map exists, we say that X X X admits an α \alpha α -snowflake embedding into Y Y Y .
The study of snowflake embeddings has roots in Schoenberg’s work in the 1930s [18 ] . A major subsequent development was Assouad’s classical theorem, which established that every doubling metric space admits an α \alpha α -snowflake embedding into a finite-dimensional Euclidean space for every α ∈ ( 0 , 1 ) \alpha\in(0,1) α ∈ ( 0 , 1 ) [2 ] .
The theory later expanded beyond the doubling setting. Lang and Schlichenmaier studied Nagata dimension, a quantitative refinement of Gromov’s asymptotic dimension, and showed that metric spaces of finite Nagata dimension admit bi-Lipschitz embeddings into finite products of metric trees after snowflaking by sufficiently small exponents [14 Theorem 1.3] . Naor and Silberman subsequently showed, through a more subtle argument, that every metric space of finite Nagata dimension admits an α \alpha α -snowflake embedding into Hilbert space for every α ∈ ( 0 , 1 ) \alpha\in(0,1) α ∈ ( 0 , 1 ) [15 Lemma 5.1 and Theorem 5.2] .
The class of metric spaces with finite Nagata dimension includes all homogeneous Hadamard manifolds, pinched negatively curved Hadamard manifolds, doubling spaces, geodesic Gromov hyperbolic spaces that are doubling in the small, metric ( R ) (\mathbb{R}) ( R ) -trees, and Euclidean buildings, among others. Moreover, this class is closed under taking finite products and finite unions [14 ] .
These results motivate our introduction of snowflake metric bundles. By applying snowflake embeddings fiberwise and then assembling them, we shall obtain coarse embeddings of the associated L p L^p L p -direct integrals into suitable Banach spaces, leading to applications to the coarse Baum–Connes and Novikov conjectures. Bundles arising from higher-order Riemannian metrics, which we discuss later, provide natural examples of this framework.
Definition 2.4. Let π : E → Ω \pi:E\to\Omega π : E → Ω be a metric bundle with typical fiber ( X , d X ) (X,d_X) ( X , d X ) . The bundle is called a snowflake metric bundle if X X X admits an α \alpha α -snowflake embedding into separable Hilbert space for some α ∈ ( 0 , 1 ] \alpha\in(0,1] α ∈ ( 0 , 1 ] .
L p L^p L p -spaces and a broad class of nonlinear analogues, especially those arising from the higher-order Riemannian metric bundles introduced later.
Higher-order μ \mu μ -discreteness
Let T ( r ) M T^{(r)}M T ( r ) M be the r r r -th order tangent bundle of M M M . In this subsection, we first consider the bundle
Riem ( r ) ( M ) ⟶ M , \operatorname{Riem}^{(r)}(M) \longrightarrow M, Riem ( r ) ( M ) ⟶ M ,
whose fiber over each x ∈ M x \in M x ∈ M is the space of all inner products on T x ( r ) M T_x^{(r)}M T x ( r ) M , equipped with its canonical structure as a homogeneous Hadamard manifold. This is a snowflake metric bundle in the sense of Definition 2.4 . We then use the r r r -th order derivatives of diffeomorphisms to quantify how continuous sections of this bundle, viewed as Riemannian metrics on T ( r ) M T^{(r)}M T ( r ) M , are distorted. This yields a quantitative notion of higher-order metric distortion and leads naturally to the definition of higher-order μ \mu μ -discreteness for subgroups of Diff ( M ) \operatorname{Diff}(M) Diff ( M ) . This notion is a generalization of that introduced by Gong–Wu–Xie–Yu [7 ] .
Let V V V be an n n n -dimensional real vector space. Denote by
P ( V ) : = { g : V × V ⟶ R ∣ g is a positive-definite symmetric bilinear form } \mathcal{P}(V) := \{g : V \times V \longrightarrow\mathbb{R} \mid g\text{ is a positive-definite symmetric bilinear form}\} P ( V ) := { g : V × V ⟶ R ∣ g is a positive-definite symmetric bilinear form }
the space of all inner products on V V V . After choosing a linear isomorphism V ≅ R n V \cong\mathbb{R}^n V ≅ R n , the space P ( V ) \mathcal{P}(V) P ( V ) is identified with
P ( n ) : = { Q ∈ M n ( R ) ∣ Q T = Q and v T Q v > 0 for every v ≠ 0 } . \mathcal{P}(n) := \{Q \in M_n(\mathbb{R}) \mid Q^{\mathsf{T}} = Q\text{ and }v^{\mathsf{T}}Qv > 0\text{ for every }v \ne0\}. P ( n ) := { Q ∈ M n ( R ) ∣ Q T = Q and v T Q v > 0 for every v = 0 } .
The congruence action of GL ( n , R ) \operatorname{GL}(n,\mathbb{R}) GL ( n , R ) on P ( n ) \mathcal{P}(n) P ( n ) is transitive, and the stabilizer of the identity matrix I n I_n I n is O ( n ) O(n) O ( n ) . Consequently,
P ( V ) ≅ P ( n ) ≅ GL ( n , R ) / O ( n ) . \mathcal{P}(V) \cong\mathcal{P}(n) \cong\operatorname{GL}(n,\mathbb{R})/O(n). P ( V ) ≅ P ( n ) ≅ GL ( n , R ) / O ( n ) .
The standard Riemannian metric on P ( n ) \mathcal{P}(n) P ( n ) is given by
⟨ A , B ⟩ Q = Tr ( Q − 1 A Q − 1 B ) , A , B ∈ T Q P ( n ) ≅ Sym n ( R ) . \langle A,B\rangle_Q = \operatorname{Tr}(Q^{-1}AQ^{-1}B), \qquad A,B \in T_Q\mathcal{P}(n) \cong\operatorname{Sym}_n(\mathbb{R}). ⟨ A , B ⟩ Q = Tr ( Q − 1 A Q − 1 B ) , A , B ∈ T Q P ( n ) ≅ Sym n ( R ) .
Denote by d P ( n ) d_{\mathcal{P}(n)} d P ( n ) the corresponding Riemannian distance on P ( n ) \mathcal{P}(n) P ( n ) , i.e.,
d P ( n ) ( Q 1 , Q 2 ) = ∥ log ( Q 1 − 1 / 2 Q 2 Q 1 − 1 / 2 ) ∥ H S , Q 1 , Q 2 ∈ P ( n ) , d_{\mathcal{P}(n)}(Q_1,Q_2) = \left\lVert\log\left(Q_1^{-1/2}Q_2Q_1^{-1/2}\right)\right\rVert_{\mathrm{HS}}, \qquad Q_1,Q_2 \in\mathcal{P}(n), d P ( n ) ( Q 1 , Q 2 ) = log ( Q 1 − 1/2 Q 2 Q 1 − 1/2 ) HS , Q 1 , Q 2 ∈ P ( n ) ,
where ∥ ⋅ ∥ H S \lVert\cdot\rVert_{\mathrm{HS}} ∥ ⋅ ∥ HS denotes the Hilbert–Schmidt norm. With this metric, P ( n ) \mathcal{P}(n) P ( n ) is a homogeneous Hadamard manifold (see [4 , 8 ] ). With the induced metric d P ( V ) d_{\mathcal{P}(V)} d P ( V ) on P ( V ) \mathcal{P}(V) P ( V ) , P ( V ) \mathcal{P}(V) P ( V ) is also a homogeneous Hadamard manifold.
Let L : V → V ′ L: V \to V' L : V → V ′ be a linear isomorphism. For each g ∈ P ( V ′ ) g \in\mathcal{P}(V') g ∈ P ( V ′ ) , its pullback by L L L is the bilinear form L ∗ g L^*g L ∗ g on V V V defined by
( L ∗ g ) ( u , v ) = g ( L u , L v ) , u , v ∈ V . (L^*g)(u,v) = g(Lu,Lv), \qquad u,v \in V. ( L ∗ g ) ( u , v ) = g ( Lu , Lv ) , u , v ∈ V .
This form is symmetric and positive-definite: for every u ∈ V ∖ { 0 } u \in V \setminus\{0\} u ∈ V ∖ { 0 } , we have L u ≠ 0 Lu \ne0 Lu = 0 and hence
( L ∗ g ) ( u , u ) = g ( L u , L u ) > 0. (L^*g)(u,u) = g(Lu,Lu) > 0. ( L ∗ g ) ( u , u ) = g ( Lu , Lu ) > 0.
Thus pullback defines a bijection
L ∗ : P ( V ′ ) ⟶ P ( V ) , g ↦ L ∗ g , (1) L^*:\mathcal{P}(V') \longrightarrow\mathcal{P}(V), \qquad g \mapsto L^*g, \tag*{(1)} L ∗ : P ( V ′ ) ⟶ P ( V ) , g ↦ L ∗ g , (1)
whose inverse is ( L − 1 ) ∗ (L^{-1})^* ( L − 1 ) ∗ . By the congruence invariance of the canonical metrics on the spaces of inner products, this bijection is an isometry. Explicitly, for all g , g ′ ∈ P ( V ′ ) g,g' \in\mathcal{P}(V') g , g ′ ∈ P ( V ′ ) ,
d P ( V ) ( L ∗ g , L ∗ g ′ ) = d P ( V ′ ) ( g , g ′ ) . d_{\mathcal{P}(V)}(L^*g,L^*g') = d_{\mathcal{P}(V')}(g,g'). d P ( V ) ( L ∗ g , L ∗ g ′ ) = d P ( V ′ ) ( g , g ′ ) .
Let M M M be a smooth manifold of dimension n n n , and let r ≥ 1 r \ge1 r ≥ 1 be an integer. The rank of the r r r -th order tangent bundle
T ( r ) M = ( J r ( M , R ) 0 ) ∗ T^{(r)}M = \left(J^r(M,\mathbb{R})_0\right)^* T ( r ) M = ( J r ( M , R ) 0 ) ∗
is
m r : = rank ( T ( r ) M ) = ∑ k = 1 r ( n + k − 1 k ) = ( n + r r ) − 1. m_r := \operatorname{rank}(T^{(r)}M) = \sum_{k=1}^{r}\binom{n+k-1}{k} = \binom{n+r}{r}-1. m r := rank ( T ( r ) M ) = k = 1 ∑ r ( k n + k − 1 ) = ( r n + r ) − 1.
By a smooth local frame of T ( r ) M T^{(r)}M T ( r ) M over an open subset U ⊆ M U \subseteq M U ⊆ M , we mean a smooth vector-bundle isomorphism
ε : U × R m r ⟶ T ( r ) M ∣ U (2) \varepsilon: U \times\mathbb{R}^{m_r} \longrightarrow T^{(r)}M|_U \tag*{(2)} ε : U × R m r ⟶ T ( r ) M ∣ U (2)
covering id U \operatorname{id}_U id U . For each x ∈ U x \in U x ∈ U , write
ε x : R m r ⟶ T x ( r ) M , ε x ( v ) : = ε ( x , v ) . \varepsilon_x: \mathbb{R}^{m_r} \longrightarrow T_x^{(r)}M,\qquad\varepsilon_x(v):=\varepsilon(x,v). ε x : R m r ⟶ T x ( r ) M , ε x ( v ) := ε ( x , v ) .
For g ∈ P ( T x ( r ) M ) g \in\mathcal{P}(T_x^{(r)}M) g ∈ P ( T x ( r ) M ) , its pullback by ε x \varepsilon_x ε x is the inner product ε x ∗ g ∈ P ( R m r ) \varepsilon_x^*g \in\mathcal{P}(\mathbb{R}^{m_r}) ε x ∗ g ∈ P ( R m r ) given by
( ε x ∗ g ) ( u , v ) = g ( ε x ( u ) , ε x ( v ) ) , u , v ∈ R m r . (\varepsilon_x^*g)(u,v)=g(\varepsilon_x(u),\varepsilon_x(v)),\qquad u,v\in\mathbb{R}^{m_r}. ( ε x ∗ g ) ( u , v ) = g ( ε x ( u ) , ε x ( v )) , u , v ∈ R m r .
These pullbacks define local trivializations of the metric bundle of homogeneous Hadamard manifolds
Riem ( r ) ( M ) : = ⨆ x ∈ M P ( T x ( r ) M ) ⟶ M , \operatorname{Riem}^{(r)}(M):=\bigsqcup_{x\in M}\mathcal{P}(T_x^{(r)}M)\longrightarrow M, Riem ( r ) ( M ) := x ∈ M ⨆ P ( T x ( r ) M ) ⟶ M ,
whose fiber over x ∈ M x\in M x ∈ M is
Riem x ( r ) ( M ) = P ( T x ( r ) M ) . \operatorname{Riem}^{(r)}_x(M)=\mathcal{P}(T_x^{(r)}M). Riem x ( r ) ( M ) = P ( T x ( r ) M ) .
We thus obtain a snowflake metric bundle with typical fiber P ( R m r ) \mathcal{P}(\mathbb{R}^{m_r}) P ( R m r ) . A section
s ∈ ∏ x ∈ M Riem x ( r ) ( M ) s\in\prod_{x\in M}\operatorname{Riem}^{(r)}_x(M) s ∈ x ∈ M ∏ Riem x ( r ) ( M )
is continuous if and only if, for every smooth local frame ε \varepsilon ε over an open subset U ⊆ M U\subseteq M U ⊆ M , the map
U ⟶ P ( R m r ) , x ↦ ε x ∗ ( s ( x ) ) U\longrightarrow\mathcal{P}(\mathbb{R}^{m_r}),\qquad x\mapsto\varepsilon_x^*(s(x)) U ⟶ P ( R m r ) , x ↦ ε x ∗ ( s ( x ))
is continuous.
We now introduce an L p L^p L p -pseudometric on Diff ( M ) \operatorname{Diff}(M) Diff ( M ) that measures how far higher-order derivatives deviate from isometries.
Fix a continuous section g r : M → Riem ( r ) ( M ) g_r:M\to\operatorname{Riem}^{(r)}(M) g r : M → Riem ( r ) ( M ) and suppose that M M M is compact with a finite Borel measure μ \mu μ . For a linear map
A : T x ( r ) M ⟶ T y ( r ) M , A:T_x^{(r)}M\longrightarrow T_y^{(r)}M, A : T x ( r ) M ⟶ T y ( r ) M ,
its operator norm with respect to g r g_r g r is defined by
∥ A ∥ g r = sup u ∈ T x ( r ) M ∖ { 0 } g r , y ( A u , A u ) g r , x ( u , u ) . \lVert A\rVert_{g_r}=\sup_{u\in T_x^{(r)}M\setminus\{0\}}\frac{\sqrt{g_{r,y}(Au,Au)}}{\sqrt{g_{r,x}(u,u)}}. ∥ A ∥ g r = u ∈ T x ( r ) M ∖ { 0 } sup g r , x ( u , u ) g r , y ( A u , A u ) .
For ϕ , ψ ∈ Diff ( M ) \phi,\psi\in\operatorname{Diff}(M) ϕ , ψ ∈ Diff ( M ) and x ∈ M x\in M x ∈ M , define
κ r , g r ( ϕ , ψ , x ) = max { ∥ D ϕ − 1 ( x ) ( r ) ( ψ − 1 ∘ ϕ ) ∥ g r , ∥ D ψ − 1 ( x ) ( r ) ( ϕ − 1 ∘ ψ ) ∥ g r } . \kappa_{r,g_r}(\phi,\psi,x)=\max\left\{\left\lVert D_{\phi^{-1}(x)}^{(r)}(\psi^{-1}\circ\phi)\right\rVert_{g_r},\left\lVert D_{\psi^{-1}(x)}^{(r)}(\phi^{-1}\circ\psi)\right\rVert_{g_r}\right\}. κ r , g r ( ϕ , ψ , x ) = max { D ϕ − 1 ( x ) ( r ) ( ψ − 1 ∘ ϕ ) g r , D ψ − 1 ( x ) ( r ) ( ϕ − 1 ∘ ψ ) g r } .
The two linear maps appearing here are mutually inverse, since
D ψ − 1 ( x ) ( r ) ( ϕ − 1 ∘ ψ ) ∘ D ϕ − 1 ( x ) ( r ) ( ψ − 1 ∘ ϕ ) = id T ϕ − 1 ( x ) ( r ) M . D_{\psi^{-1}(x)}^{(r)}(\phi^{-1}\circ\psi)\circ D_{\phi^{-1}(x)}^{(r)}(\psi^{-1}\circ\phi)=\operatorname{id}_{T_{\phi^{-1}(x)}^{(r)}M}. D ψ − 1 ( x ) ( r ) ( ϕ − 1 ∘ ψ ) ∘ D ϕ − 1 ( x ) ( r ) ( ψ − 1 ∘ ϕ ) = id T ϕ − 1 ( x ) ( r ) M .
Consequently,
κ r , g r ( ϕ , ψ , x ) ≥ 1. \kappa_{r,g_r}(\phi,\psi,x)\geq1. κ r , g r ( ϕ , ψ , x ) ≥ 1.
Fix 1 ≤ p < ∞ 1\leq p<\infty 1 ≤ p < ∞ throughout the following discussion.
Define
d μ , g r ( r , p ) : Diff ( M ) × Diff ( M ) ⟶ [ 0 , ∞ ) d_{\mu,g_r}^{(r,p)}:\operatorname{Diff}(M)\times\operatorname{Diff}(M)\longrightarrow[0,\infty) d μ , g r ( r , p ) : Diff ( M ) × Diff ( M ) ⟶ [ 0 , ∞ )
by
d μ , g r ( r , p ) ( ϕ , ψ ) = ( ∫ M ( log κ r , g r ( ϕ , ψ , x ) ) p d μ ( x ) ) 1 / p . d_{\mu,g_r}^{(r,p)}(\phi,\psi)=\left(\int_M\bigl(\log\kappa_{r,g_r}(\phi,\psi,x)\bigr)^p\,d\mu(x)\right)^{1/p}. d μ , g r ( r , p ) ( ϕ , ψ ) = ( ∫ M ( log κ r , g r ( ϕ , ψ , x ) ) p d μ ( x ) ) 1/ p .
The integral is well-defined. In fact, the two maps
x ⟼ D ϕ − 1 ( x ) ( r ) ( ψ − 1 ∘ ϕ ) and x ⟼ D ψ − 1 ( x ) ( r ) ( ϕ − 1 ∘ ψ ) x\longmapsto D_{\phi^{-1}(x)}^{(r)}(\psi^{-1}\circ\phi)\qquad\text{and}\qquad x\longmapsto D_{\psi^{-1}(x)}^{(r)}(\phi^{-1}\circ\psi) x ⟼ D ϕ − 1 ( x ) ( r ) ( ψ − 1 ∘ ϕ ) and x ⟼ D ψ − 1 ( x ) ( r ) ( ϕ − 1 ∘ ψ )
depend continuously on x x x . Since g r g_r g r is continuous and the operator norm of a linear map between finite-dimensional inner-product spaces depends continuously on both the linear map and the inner products, the function
x ⟼ κ r , g r ( ϕ , ψ , x ) x \longmapsto\kappa_{r,g_r}(\phi,\psi,x) x ⟼ κ r , g r ( ϕ , ψ , x )
is continuous. Hence, since M M M is compact and μ \mu μ is finite, it follows that
log κ r , g r ( ϕ , ψ , ⋅ ) ∈ L p ( M , μ ) . \log\kappa_{r,g_r}(\phi,\psi,\cdot) \in L^p(M,\mu). log κ r , g r ( ϕ , ψ , ⋅ ) ∈ L p ( M , μ ) .
Therefore, d μ , g r ( r , p ) ( ϕ , ψ ) d_{\mu,g_r}^{(r,p)}(\phi,\psi) d μ , g r ( r , p ) ( ϕ , ψ ) is finite and well-defined.
Lemma 2.6. d μ , g r ( r , p ) d_{\mu,g_r}^{(r,p)} d μ , g r ( r , p ) is a pseudometric on Diff ( M ) \operatorname{Diff}(M) Diff ( M ) , i.e., for any ϕ 1 , ϕ 2 , ϕ 3 ∈ Diff ( M ) \phi_1,\phi_2,\phi_3 \in\operatorname{Diff}(M) ϕ 1 , ϕ 2 , ϕ 3 ∈ Diff ( M ) , we have
d μ , g r ( r , p ) ( ϕ 1 , ϕ 1 ) = 0 d_{\mu,g_r}^{(r,p)}(\phi_1,\phi_1)=0 d μ , g r ( r , p ) ( ϕ 1 , ϕ 1 ) = 0 ;
d μ , g r ( r , p ) ( ϕ 1 , ϕ 2 ) = d μ , g r ( r , p ) ( ϕ 2 , ϕ 1 ) d_{\mu,g_r}^{(r,p)}(\phi_1,\phi_2)=d_{\mu,g_r}^{(r,p)}(\phi_2,\phi_1) d μ , g r ( r , p ) ( ϕ 1 , ϕ 2 ) = d μ , g r ( r , p ) ( ϕ 2 , ϕ 1 ) ;
d μ , g r ( r , p ) ( ϕ 1 , ϕ 3 ) ≤ d μ , g r ( r , p ) ( ϕ 1 , ϕ 2 ) + d μ , g r ( r , p ) ( ϕ 2 , ϕ 3 ) d_{\mu,g_r}^{(r,p)}(\phi_1,\phi_3)\le d_{\mu,g_r}^{(r,p)}(\phi_1,\phi_2)+d_{\mu,g_r}^{(r,p)}(\phi_2,\phi_3) d μ , g r ( r , p ) ( ϕ 1 , ϕ 3 ) ≤ d μ , g r ( r , p ) ( ϕ 1 , ϕ 2 ) + d μ , g r ( r , p ) ( ϕ 2 , ϕ 3 ) .
Proof. The first two axioms are immediate from the definition of d μ , g r ( r , p ) d_{\mu,g_r}^{(r,p)} d μ , g r ( r , p ) . To verify the last axiom, we observe that for ϕ 1 , ϕ 2 , ϕ 3 ∈ Diff ( M ) \phi_1,\phi_2,\phi_3 \in\operatorname{Diff}(M) ϕ 1 , ϕ 2 , ϕ 3 ∈ Diff ( M ) , functoriality gives
D ϕ 1 − 1 ( x ) ( r ) ( ϕ 3 − 1 ∘ ϕ 1 ) = D ϕ 2 − 1 ( x ) ( r ) ( ϕ 3 − 1 ∘ ϕ 2 ) ∘ D ϕ 1 − 1 ( x ) ( r ) ( ϕ 2 − 1 ∘ ϕ 1 ) , D_{\phi_1^{-1}(x)}^{(r)}(\phi_3^{-1}\circ\phi_1)=D_{\phi_2^{-1}(x)}^{(r)}(\phi_3^{-1}\circ\phi_2)\circ D_{\phi_1^{-1}(x)}^{(r)}(\phi_2^{-1}\circ\phi_1), D ϕ 1 − 1 ( x ) ( r ) ( ϕ 3 − 1 ∘ ϕ 1 ) = D ϕ 2 − 1 ( x ) ( r ) ( ϕ 3 − 1 ∘ ϕ 2 ) ∘ D ϕ 1 − 1 ( x ) ( r ) ( ϕ 2 − 1 ∘ ϕ 1 ) ,
D ϕ 3 − 1 ( x ) ( r ) ( ϕ 1 − 1 ∘ ϕ 3 ) = D ϕ 2 − 1 ( x ) ( r ) ( ϕ 1 − 1 ∘ ϕ 2 ) ∘ D ϕ 3 − 1 ( x ) ( r ) ( ϕ 2 − 1 ∘ ϕ 3 ) . D_{\phi_3^{-1}(x)}^{(r)}(\phi_1^{-1}\circ\phi_3)=D_{\phi_2^{-1}(x)}^{(r)}(\phi_1^{-1}\circ\phi_2)\circ D_{\phi_3^{-1}(x)}^{(r)}(\phi_2^{-1}\circ\phi_3). D ϕ 3 − 1 ( x ) ( r ) ( ϕ 1 − 1 ∘ ϕ 3 ) = D ϕ 2 − 1 ( x ) ( r ) ( ϕ 1 − 1 ∘ ϕ 2 ) ∘ D ϕ 3 − 1 ( x ) ( r ) ( ϕ 2 − 1 ∘ ϕ 3 ) .
The submultiplicativity of operator norms therefore implies
κ r , g r ( ϕ 1 , ϕ 3 , x ) ≤ κ r , g r ( ϕ 1 , ϕ 2 , x ) ⋅ κ r , g r ( ϕ 2 , ϕ 3 , x ) . \kappa_{r,g_r}(\phi_1,\phi_3,x)\le\kappa_{r,g_r}(\phi_1,\phi_2,x)\cdot\kappa_{r,g_r}(\phi_2,\phi_3,x). κ r , g r ( ϕ 1 , ϕ 3 , x ) ≤ κ r , g r ( ϕ 1 , ϕ 2 , x ) ⋅ κ r , g r ( ϕ 2 , ϕ 3 , x ) .
Hence
∣ log κ r , g r ( ϕ 1 , ϕ 3 , x ) ∣ ≤ log κ r , g r ( ϕ 1 , ϕ 2 , x ) + log κ r , g r ( ϕ 2 , ϕ 3 , x ) , \left|\log\kappa_{r,g_r}(\phi_1,\phi_3,x)\right|\le\log\kappa_{r,g_r}(\phi_1,\phi_2,x)+\log\kappa_{r,g_r}(\phi_2,\phi_3,x), ∣ log κ r , g r ( ϕ 1 , ϕ 3 , x ) ∣ ≤ log κ r , g r ( ϕ 1 , ϕ 2 , x ) + log κ r , g r ( ϕ 2 , ϕ 3 , x ) ,
and the Minkowski inequality yields
d μ , g r ( r , p ) ( ϕ 1 , ϕ 3 ) ≤ d μ , g r ( r , p ) ( ϕ 1 , ϕ 2 ) + d μ , g r ( r , p ) ( ϕ 2 , ϕ 3 ) . d_{\mu,g_r}^{(r,p)}(\phi_1,\phi_3)\le d_{\mu,g_r}^{(r,p)}(\phi_1,\phi_2)+d_{\mu,g_r}^{(r,p)}(\phi_2,\phi_3). d μ , g r ( r , p ) ( ϕ 1 , ϕ 3 ) ≤ d μ , g r ( r , p ) ( ϕ 1 , ϕ 2 ) + d μ , g r ( r , p ) ( ϕ 2 , ϕ 3 ) .
:::
For τ ∈ Diff ( M ) \tau\in\operatorname{Diff}(M) τ ∈ Diff ( M ) , the pseudometric satisfies the covariance relation
d τ ∗ μ , g r ( r , p ) ( ϕ , ψ ) = d μ , g r ( r , p ) ( τ − 1 ∘ ϕ , τ − 1 ∘ ψ ) , d_{\tau_*\mu,g_r}^{(r,p)}(\phi,\psi)=d_{\mu,g_r}^{(r,p)}(\tau^{-1}\circ\phi,\tau^{-1}\circ\psi), d τ ∗ μ , g r ( r , p ) ( ϕ , ψ ) = d μ , g r ( r , p ) ( τ − 1 ∘ ϕ , τ − 1 ∘ ψ ) ,
where τ ∗ μ \tau_*\mu τ ∗ μ denotes the push-forward of μ \mu μ under τ \tau τ , defined by
( τ ∗ μ ) ( E ) = μ ( τ − 1 ( E ) ) (\tau_*\mu)(E)=\mu(\tau^{-1}(E)) ( τ ∗ μ ) ( E ) = μ ( τ − 1 ( E ))
for every Borel set E ⊆ M E\subseteq M E ⊆ M .
We now restrict the pseudometric to a countable subgroup Γ < Diff ( M ) \Gamma<\operatorname{Diff}(M) Γ < Diff ( M ) . For every γ ∈ Γ \gamma\in\Gamma γ ∈ Γ ,
d μ , g r ( r , p ) ( γ , 1 Γ ) = ( ∫ M ( log max { ∥ D γ − 1 ( x ) ( r ) γ ∥ g r , ∥ D x ( r ) ( γ − 1 ) ∥ g r } ) p d μ ( x ) ) 1 / p . d_{\mu,g_r}^{(r,p)}(\gamma,1_\Gamma)=\left(\int_M\left(\log\max\left\{\left\|D_{\gamma^{-1}(x)}^{(r)}\gamma\right\|_{g_r},\left\|D_x^{(r)}(\gamma^{-1})\right\|_{g_r}\right\}\right)^p\,d\mu(x)\right)^{1/p}. d μ , g r ( r , p ) ( γ , 1 Γ ) = ( ∫ M ( log max { D γ − 1 ( x ) ( r ) γ g r , D x ( r ) ( γ − 1 ) g r } ) p d μ ( x ) ) 1/ p .
Since
D x ( r ) ( γ − 1 ) = ( D γ − 1 ( x ) ( r ) γ ) − 1 , D_x^{(r)}(\gamma^{-1})=\left(D_{\gamma^{-1}(x)}^{(r)}\gamma\right)^{-1}, D x ( r ) ( γ − 1 ) = ( D γ − 1 ( x ) ( r ) γ ) − 1 ,
this pseudometric measures, in an L p L^p L p -sense, how far the r r r -th order derivative of γ \gamma γ is from being an isometry.
We introduce the following higher-order generalization of μ \mu μ -discreteness given in [7 ] .
Definition 2.7. We say that a countable subgroup
Γ < Diff ( M ) \Gamma< \operatorname{Diff}(M) Γ < Diff ( M )
is ( r , p ) (r,p) ( r , p ) -th order μ \mu μ -discrete if
inf τ ∈ Γ d τ ∗ μ , g r ( r , p ) ( γ , 1 Γ ) ⟶ ∞ as γ → ∞ in Γ . \inf_{\tau\in\Gamma} d^{(r,p)}_{\tau_*\mu,g_r}(\gamma,1_\Gamma) \longrightarrow\infty\qquad\text{as } \gamma\to\infty\text{ in }\Gamma. τ ∈ Γ inf d τ ∗ μ , g r ( r , p ) ( γ , 1 Γ ) ⟶ ∞ as γ → ∞ in Γ.
Equivalently, for every R > 0 R>0 R > 0 , the set
{ γ ∈ Γ : inf τ ∈ Γ d τ ∗ μ , g r ( r , p ) ( γ , 1 Γ ) ≤ R } \left\{\gamma\in\Gamma:\inf_{\tau\in\Gamma}d^{(r,p)}_{\tau_*\mu,g_r}(\gamma,1_\Gamma)\le R\right\} { γ ∈ Γ : τ ∈ Γ inf d τ ∗ μ , g r ( r , p ) ( γ , 1 Γ ) ≤ R }
is finite.
Remark 2.8. The definition of ( r , p ) (r,p) ( r , p ) -th order μ \mu μ -discreteness is independent of the choice of the continuous section g r g_r g r . Indeed, let g r ′ g'_r g r ′ be another continuous section. Since M M M is compact, there exists C ≥ 1 C\ge1 C ≥ 1 such that
C − 1 g r ≤ g r ′ ≤ C g r C^{-1}g_r\le g'_r\le Cg_r C − 1 g r ≤ g r ′ ≤ C g r
as quadratic forms on each fiber. Consequently,
C − 1 κ r , g r ( ϕ , ψ , x ) ≤ κ r , g r ′ ( ϕ , ψ , x ) ≤ C κ r , g r ( ϕ , ψ , x ) C^{-1}\kappa_{r,g_r}(\phi,\psi,x)\le\kappa_{r,g'_r}(\phi,\psi,x)\le C\kappa_{r,g_r}(\phi,\psi,x) C − 1 κ r , g r ( ϕ , ψ , x ) ≤ κ r , g r ′ ( ϕ , ψ , x ) ≤ C κ r , g r ( ϕ , ψ , x )
for all ϕ , ψ ∈ Diff ( M ) \phi,\psi\in\operatorname{Diff}(M) ϕ , ψ ∈ Diff ( M ) and x ∈ M x\in M x ∈ M . Taking logarithms and applying the reverse triangle inequality in L p L^p L p gives
∣ d ν , g r ( r , p ) ( ϕ , ψ ) − d ν , g r ′ ( r , p ) ( ϕ , ψ ) ∣ ≤ ν ( M ) 1 / p log C \left|d^{(r,p)}_{\nu,g_r}(\phi,\psi)-d^{(r,p)}_{\nu,g'_r}(\phi,\psi)\right|\le\nu(M)^{1/p}\log C d ν , g r ( r , p ) ( ϕ , ψ ) − d ν , g r ′ ( r , p ) ( ϕ , ψ ) ≤ ν ( M ) 1/ p log C
for every finite Borel measure ν \nu ν on M M M . Since ( τ ∗ μ ) ( M ) = μ ( M ) (\tau_*\mu)(M)=\mu(M) ( τ ∗ μ ) ( M ) = μ ( M ) , this bound is uniform for ν = τ ∗ μ \nu=\tau_*\mu ν = τ ∗ μ , τ ∈ Γ \tau\in\Gamma τ ∈ Γ . Thus the divergence condition in the definition is unchanged when g r g_r g r is replaced by g r ′ g'_r g r ′ .
Moreover, it is straightforward to see that, when r = 1 r=1 r = 1 and p = 2 p=2 p = 2 , our definition reduces to the notion of μ \mu μ -discreteness introduced in [7 ] .
Furthermore, for any fixed p ∈ [ 1 , ∞ ) p\in[1,\infty) p ∈ [ 1 , ∞ ) and finite Borel measure μ \mu μ , the ( r 1 , p ) (r_1,p) ( r 1 , p ) -th order μ \mu μ -discreteness implies ( r 2 , p ) (r_2,p) ( r 2 , p ) -th order μ \mu μ -discreteness whenever r 1 < r 2 r_1<r_2 r 1 < r 2 . Let 1 ≤ r 1 < r 2 1\le r_1<r_2 1 ≤ r 1 < r 2 , and let ι : T ( r 1 ) M ↪ T ( r 2 ) M \iota:T^{(r_1)}M\hookrightarrow T^{(r_2)}M ι : T ( r 1 ) M ↪ T ( r 2 ) M be the natural inclusion obtained by dualizing the jet projection j x r 2 f ↦ j x r 1 f j^{r_2}_x f\mapsto j^{r_1}_x f j x r 2 f ↦ j x r 1 f . By [13 Section 12.14] , we have
D x ( r 2 ) ϕ ∘ ι x = ι ϕ ( x ) ∘ D x ( r 1 ) ϕ , x ∈ M , ϕ ∈ Diff ( M ) . D^{(r_2)}_x\phi\circ\iota_x=\iota_{\phi(x)}\circ D^{(r_1)}_x\phi,\qquad x\in M,\ \phi\in\operatorname{Diff}(M). D x ( r 2 ) ϕ ∘ ι x = ι ϕ ( x ) ∘ D x ( r 1 ) ϕ , x ∈ M , ϕ ∈ Diff ( M ) .
Hence, upon identifying T ( r 1 ) M T^{(r_1)}M T ( r 1 ) M with its image under ι \iota ι , D ( r 2 ) ϕ D^{(r_2)}\phi D ( r 2 ) ϕ restricts to D ( r 1 ) ϕ D^{(r_1)}\phi D ( r 1 ) ϕ for every ϕ ∈ Diff ( M ) \phi\in\operatorname{Diff}(M) ϕ ∈ Diff ( M ) . Since μ \mu μ -discreteness is independent of the choice of reference metric, we may take g r 1 = g r 2 ∣ T ( r 1 ) M g_{r_1}=g_{r_2}|_{T^{(r_1)}M} g r 1 = g r 2 ∣ T ( r 1 ) M . Since D ( r 1 ) ϕ D^{(r_1)}\phi D ( r 1 ) ϕ is the restriction of D ( r 2 ) ϕ D^{(r_2)}\phi D ( r 2 ) ϕ for every ϕ ∈ Diff ( M ) \phi\in\operatorname{Diff}(M) ϕ ∈ Diff ( M ) , each of the two operator norms defining κ r 1 , g r 1 \kappa_{r_1,g_{r_1}} κ r 1 , g r 1 is bounded by the corresponding norm at order r 2 r_2 r 2 . Thus,
κ r 1 , g r 1 ( γ , 1 Γ , x ) ≤ κ r 2 , g r 2 ( γ , 1 Γ , x ) , \kappa_{r_1,g_{r_1}}(\gamma,1_\Gamma,x)\le\kappa_{r_2,g_{r_2}}(\gamma,1_\Gamma,x), κ r 1 , g r 1 ( γ , 1 Γ , x ) ≤ κ r 2 , g r 2 ( γ , 1 Γ , x ) ,
and hence
inf τ ∈ Γ d τ ∗ μ , g r 1 ( r 1 , p ) ( γ , 1 Γ ) ≤ inf τ ∈ Γ d τ ∗ μ , g r 2 ( r 2 , p ) ( γ , 1 Γ ) . \inf_{\tau\in\Gamma}d^{(r_1,p)}_{\tau_*\mu,g_{r_1}}(\gamma,1_\Gamma)\le\inf_{\tau\in\Gamma}d^{(r_2,p)}_{\tau_*\mu,g_{r_2}}(\gamma,1_\Gamma). τ ∈ Γ inf d τ ∗ μ , g r 1 ( r 1 , p ) ( γ , 1 Γ ) ≤ τ ∈ Γ inf d τ ∗ μ , g r 2 ( r 2 , p ) ( γ , 1 Γ ) .
Therefore, ( r 1 , p ) (r_1,p) ( r 1 , p ) -th order μ \mu μ -discreteness implies ( r 2 , p ) (r_2,p) ( r 2 , p ) -th order μ \mu μ -discreteness. However, the converse fails in general, as the example in § 4 shows.
Higher-order orbit maps
Throughout this subsection, we fix 1 ≤ p < ∞ 1\le p<\infty 1 ≤ p < ∞ and let M M M be a closed smooth manifold equipped with a finite Borel measure μ \mu μ . We construct the higher-order orbit map from Diff ( M ) \operatorname{Diff}(M) Diff ( M ) into the L p L^p L p -direct integral of the snowflake metric bundle Riem ( r ) ( M ) \operatorname{Riem}^{(r)}(M) Riem ( r ) ( M ) . We then prove that, for any countable subgroup Γ < Diff ( M ) \Gamma< \operatorname{Diff}(M) Γ < Diff ( M ) equipped with a proper left-invariant metric, the restriction of this orbit map to Γ \Gamma Γ is a coarse embedding if and only if Γ \Gamma Γ is ( r , p ) (r,p) ( r , p ) -th order μ \mu μ -discrete.
Construction 2.9. We fix a continuous base section
g r : M ⟶ Riem ( r ) ( M ) , x ↦ g r , x . g_r:M\longrightarrow\operatorname{Riem}^{(r)}(M),\qquad x\mapsto g_{r,x}. g r : M ⟶ Riem ( r ) ( M ) , x ↦ g r , x .
Let ϕ ∈ Diff ( M ) \phi\in\operatorname{Diff}(M) ϕ ∈ Diff ( M ) . Since
D ( r ) ( ϕ − 1 ∘ ϕ ) = D ( r ) id M = id T ( r ) M , D^{(r)}(\phi^{-1}\circ\phi)=D^{(r)}\operatorname{id}_M=\operatorname{id}_{T^{(r)}M}, D ( r ) ( ϕ − 1 ∘ ϕ ) = D ( r ) id M = id T ( r ) M ,
the linear map
D x ( r ) ϕ : T x ( r ) M ⟶ T ϕ ( x ) ( r ) M D_x^{(r)}\phi:T_x^{(r)}M\longrightarrow T_{\phi(x)}^{(r)}M D x ( r ) ϕ : T x ( r ) M ⟶ T ϕ ( x ) ( r ) M
is an isomorphism whose inverse is
D ϕ ( x ) ( r ) ( ϕ − 1 ) . D_{\phi(x)}^{(r)}(\phi^{-1}). D ϕ ( x ) ( r ) ( ϕ − 1 ) .
For every section
s : M ⟶ Riem ( r ) ( M ) , x ↦ s x , s:M\longrightarrow\operatorname{Riem}^{(r)}(M),\qquad x\mapsto s_x, s : M ⟶ Riem ( r ) ( M ) , x ↦ s x ,
of Riem ( r ) ( M ) \operatorname{Riem}^{(r)}(M) Riem ( r ) ( M ) , define its r r r -th order push-forward by ϕ \phi ϕ to be the section ϕ ∗ ( r ) s \phi_*^{(r)}s ϕ ∗ ( r ) s ,
( ϕ ∗ ( r ) s ) y : = ( D y ( r ) ( ϕ − 1 ) ) ∗ s ϕ − 1 ( y ) , (\phi_*^{(r)}s)_y:=\left(D_y^{(r)}(\phi^{-1})\right)^*s_{\phi^{-1}(y)}, ( ϕ ∗ ( r ) s ) y := ( D y ( r ) ( ϕ − 1 ) ) ∗ s ϕ − 1 ( y ) ,
where the pullback on the right-hand side is given by (1) . Equivalently, for u , v ∈ T y ( r ) M u,v\in T_y^{(r)}M u , v ∈ T y ( r ) M ,
( ϕ ∗ ( r ) s ) y ( u , v ) = s ϕ − 1 ( y ) ( D y ( r ) ( ϕ − 1 ) ( u ) , D y ( r ) ( ϕ − 1 ) ( v ) ) . (\phi_*^{(r)}s)_y(u,v)=s_{\phi^{-1}(y)}\left(D_y^{(r)}(\phi^{-1})(u),D_y^{(r)}(\phi^{-1})(v)\right). ( ϕ ∗ ( r ) s ) y ( u , v ) = s ϕ − 1 ( y ) ( D y ( r ) ( ϕ − 1 ) ( u ) , D y ( r ) ( ϕ − 1 ) ( v ) ) .
In particular, if y = ϕ ( x ) y=\phi(x) y = ϕ ( x ) , then
( ϕ ∗ ( r ) s ) ϕ ( x ) = ( ( D x ( r ) ϕ ) − 1 ) ∗ s x . (\phi_*^{(r)}s)_{\phi(x)}=\left(\left(D_x^{(r)}\phi\right)^{-1}\right)^*s_x. ( ϕ ∗ ( r ) s ) ϕ ( x ) = ( ( D x ( r ) ϕ ) − 1 ) ∗ s x .
The r r r -th order push-forward satisfies
id ∗ ( r ) s = s and ( ϕ ∘ ψ ) ∗ ( r ) s = ϕ ∗ ( r ) ( ψ ∗ ( r ) s ) \operatorname{id}_*^{(r)}s=s\quad\text{and}\quad(\phi\circ\psi)_*^{(r)}s=\phi_*^{(r)}(\psi_*^{(r)}s) id ∗ ( r ) s = s and ( ϕ ∘ ψ ) ∗ ( r ) s = ϕ ∗ ( r ) ( ψ ∗ ( r ) s )
for all ϕ , ψ ∈ Diff ( M ) \phi,\psi\in\operatorname{Diff}(M) ϕ , ψ ∈ Diff ( M ) . Moreover, this push-forward preserves continuity and Borel measurability. Indeed, let ε \varepsilon ε be a smooth local frame over U ⊆ M U\subseteq M U ⊆ M given by (2) . It induces a smooth local frame over ϕ ( U ) \phi(U) ϕ ( U ) by
ε ~ y = D ϕ − 1 ( y ) ( r ) ϕ ∘ ε ϕ − 1 ( y ) , y ∈ ϕ ( U ) . \widetilde{\varepsilon}_y=D_{\phi^{-1}(y)}^{(r)}\phi\circ\varepsilon_{\phi^{-1}(y)},\qquad y\in\phi(U). ε y = D ϕ − 1 ( y ) ( r ) ϕ ∘ ε ϕ − 1 ( y ) , y ∈ ϕ ( U ) .
By the definition of ϕ ∗ ( r ) \phi_*^{(r)} ϕ ∗ ( r ) , the local representation of ϕ ∗ ( r ) s \phi_*^{(r)}s ϕ ∗ ( r ) s in this frame is
ϕ ( U ) ⟶ P ( R m r ) , y ⟼ ε ~ y ∗ ( ( ϕ ∗ ( r ) s ) y ) = ε ϕ − 1 ( y ) ∗ ( s ϕ − 1 ( y ) ) . \phi(U)\longrightarrow\mathcal{P}(\mathbb{R}^{m^r}),\qquad
y\longmapsto\widetilde{\varepsilon}_y^*((\phi_*^{(r)}s)_y)=\varepsilon_{\phi^{-1}(y)}^*(s_{\phi^{-1}(y)}). ϕ ( U ) ⟶ P ( R m r ) , y ⟼ ε y ∗ (( ϕ ∗ ( r ) s ) y ) = ε ϕ − 1 ( y ) ∗ ( s ϕ − 1 ( y ) ) .
This is the composition of the local representation x ↦ ε x ∗ ( s x ) x\mapsto\varepsilon_x^*(s_x) x ↦ ε x ∗ ( s x ) with ϕ − 1 \phi^{-1} ϕ − 1 . Since ϕ − 1 \phi^{-1} ϕ − 1 is continuous, ϕ ∗ ( r ) s \phi_*^{(r)}s ϕ ∗ ( r ) s is continuous whenever s s s is continuous, and Borel measurable whenever s s s is Borel measurable.
Since the pointwise distance between continuous sections is continuous, compactness of M M M and finiteness of μ \mu μ give
∫ M d P ( T x ( r ) M ) ( ( ϕ ∗ ( r ) g r ) x , g r , x ) p d μ ( x ) < ∞ . \int_M d_{\mathcal{P}(T_x^{(r)}M)}\left((\phi_*^{(r)}g_r)_x,g_{r,x}\right)^p\,d\mu(x)<\infty. ∫ M d P ( T x ( r ) M ) ( ( ϕ ∗ ( r ) g r ) x , g r , x ) p d μ ( x ) < ∞.
Consequently,
[ ϕ ∗ ( r ) g r ] ∈ L g r p ( M , μ , Riem ( r ) ( M ) ) [\phi_*^{(r)}g_r]\in L_{g_r}^p\left(M,\mu,\operatorname{Riem}^{(r)}(M)\right) [ ϕ ∗ ( r ) g r ] ∈ L g r p ( M , μ , Riem ( r ) ( M ) )
for every ϕ ∈ Diff ( M ) \phi\in\operatorname{Diff}(M) ϕ ∈ Diff ( M ) . We define the ( r , p ) (r,p) ( r , p ) -th order orbit map
F μ , g r ( r , p ) : Diff ( M ) ⟶ L g r p ( M , μ , Riem ( r ) ( M ) ) F_{\mu,g_r}^{(r,p)}:\operatorname{Diff}(M)\longrightarrow L_{g_r}^{p}\left(M,\mu,\operatorname{Riem}^{(r)}(M)\right) F μ , g r ( r , p ) : Diff ( M ) ⟶ L g r p ( M , μ , Riem ( r ) ( M ) )
by
F μ , g r ( r , p ) ( ϕ ) = [ ϕ ∗ ( r ) g r ] . F_{\mu,g_r}^{(r,p)}(\phi)=\left[\phi_*^{(r)}g_r\right]. F μ , g r ( r , p ) ( ϕ ) = [ ϕ ∗ ( r ) g r ] .
We are now ready to give an equivalent characterization of higher-order μ \mu μ -discreteness in terms of the higher-order orbit map.
Proposition 2.10. Assume that M M M is a closed smooth manifold and μ \mu μ is a finite Borel measure on M M M . For any countable subgroup Γ < Diff ( M ) \Gamma<\operatorname{Diff}(M) Γ < Diff ( M ) , the following statements are equivalent:
(1) Γ \Gamma Γ is ( r , p ) (r,p) ( r , p ) -th order μ \mu μ -discrete;
(2) for every proper left-invariant metric d Γ d_\Gamma d Γ on Γ \Gamma Γ and every continuous section g r : M → Riem ( r ) ( M ) g_r:M\to\operatorname{Riem}^{(r)}(M) g r : M → Riem ( r ) ( M ) , the restricted orbit map F μ , g r ( r , p ) ∣ Γ F_{\mu,g_r}^{(r,p)}\big|_\Gamma F μ , g r ( r , p ) Γ is a coarse embedding with respect to d Γ d_\Gamma d Γ .
Proof. The finite case is immediate, so assume that Γ \Gamma Γ is infinite.
For a continuous base section g r : M → Riem ( r ) ( M ) g_r:M\to\operatorname{Riem}^{(r)}(M) g r : M → Riem ( r ) ( M ) , we write
F = F μ , g r ( r , p ) ∣ Γ . F=F_{\mu,g_r}^{(r,p)}\big|_\Gamma. F = F μ , g r ( r , p ) Γ .
For γ , γ ~ ∈ Γ \gamma,\widetilde{\gamma}\in\Gamma γ , γ ∈ Γ , set
ψ = γ ~ − 1 γ , τ = γ ~ − 1 . \psi=\widetilde{\gamma}^{-1}\gamma,\qquad\tau=\widetilde{\gamma}^{-1}. ψ = γ − 1 γ , τ = γ − 1 .
By functoriality,
τ ∗ ( r ) ( γ ∗ ( r ) g r ) = ψ ∗ ( r ) g r , τ ∗ ( r ) ( γ ~ ∗ ( r ) g r ) = g r . \tau_*^{(r)}\left(\gamma_*^{(r)}g_r\right)=\psi_*^{(r)}g_r,\qquad
\tau_*^{(r)}\left(\widetilde{\gamma}_*^{(r)}g_r\right)=g_r. τ ∗ ( r ) ( γ ∗ ( r ) g r ) = ψ ∗ ( r ) g r , τ ∗ ( r ) ( γ ∗ ( r ) g r ) = g r .
Hence, we obtain
d L p ( F ( γ ) , F ( γ ~ ) ) p = ∫ M d P ( T τ ( y ) ( r ) M ) ( ( γ ∗ ( r ) g r ) y , ( γ ~ ∗ ( r ) g r ) y ) p d μ ( y ) = ∫ M d P ( T τ ( y ) ( r ) M ) ( ( τ ∗ ( r ) ( γ ∗ ( r ) g r ) ) τ ( y ) , ( τ ∗ ( r ) ( γ ~ ∗ ( r ) g r ) ) τ ( y ) ) p d μ ( y ) = ∫ M d P ( T τ ( y ) ( r ) M ) ( ( ψ ∗ ( r ) g r ) τ ( y ) , g r , τ ( y ) ) p d μ ( y ) = ∫ M d P ( T x ( r ) M ) ( ( ψ ∗ ( r ) g r ) x , g r , x ) p d ( τ ∗ μ ) ( x ) . (3) \begin{aligned}
d_{L^p}\left(F(\gamma),F(\widetilde{\gamma})\right)^p
&=\int_M d_{\mathcal{P}\left(T_{\tau(y)}^{(r)}M\right)}
\left(\left(\gamma_*^{(r)}g_r\right)_y,\left(\widetilde{\gamma}_*^{(r)}g_r\right)_y\right)^p\,d\mu(y)\\
&=\int_M d_{\mathcal{P}\left(T_{\tau(y)}^{(r)}M\right)}
\left(\left(\tau_*^{(r)}\left(\gamma_*^{(r)}g_r\right)\right)_{\tau(y)},
\left(\tau_*^{(r)}\left(\widetilde{\gamma}_*^{(r)}g_r\right)\right)_{\tau(y)}\right)^p\,d\mu(y)\\
&=\int_M d_{\mathcal{P}\left(T_{\tau(y)}^{(r)}M\right)}
\left(\left(\psi_*^{(r)}g_r\right)_{\tau(y)},g_{r,\tau(y)}\right)^p\,d\mu(y)\\
&=\int_M d_{\mathcal{P}\left(T_x^{(r)}M\right)}
\left(\left(\psi_*^{(r)}g_r\right)_x,g_{r,x}\right)^p\,d(\tau_*\mu)(x). \tag*{(3)}
\end{aligned} d L p ( F ( γ ) , F ( γ ) ) p = ∫ M d P ( T τ ( y ) ( r ) M ) ( ( γ ∗ ( r ) g r ) y , ( γ ∗ ( r ) g r ) y ) p d μ ( y ) = ∫ M d P ( T τ ( y ) ( r ) M ) ( ( τ ∗ ( r ) ( γ ∗ ( r ) g r ) ) τ ( y ) , ( τ ∗ ( r ) ( γ ∗ ( r ) g r ) ) τ ( y ) ) p d μ ( y ) = ∫ M d P ( T τ ( y ) ( r ) M ) ( ( ψ ∗ ( r ) g r ) τ ( y ) , g r , τ ( y ) ) p d μ ( y ) = ∫ M d P ( T x ( r ) M ) ( ( ψ ∗ ( r ) g r ) x , g r , x ) p d ( τ ∗ μ ) ( x ) . (3)
We next express the integrand in terms of singular values. Set
m r = rank ( T ( r ) M ) . m_r=\operatorname{rank}\left(T^{(r)}M\right). m r = rank ( T ( r ) M ) .
For each x ∈ M x\in M x ∈ M , choose a g r , ψ − 1 ( x ) g_{r,\psi^{-1}(x)} g r , ψ − 1 ( x ) -orthonormal basis of T ψ − 1 ( x ) ( r ) M T_{\psi^{-1}(x)}^{(r)}M T ψ − 1 ( x ) ( r ) M and a g r , x g_{r,x} g r , x -orthonormal basis of T x ( r ) M T_x^{(r)}M T x ( r ) M . Let A x A_x A x be the matrix of D ψ − 1 ( x ) ( r ) ψ D_{\psi^{-1}(x)}^{(r)}\psi D ψ − 1 ( x ) ( r ) ψ under these bases, and let
σ 1 ( x ) ≥ ⋯ ≥ σ m r ( x ) > 0 \sigma_1(x)\ge\cdots\ge\sigma_{m_r}(x)>0 σ 1 ( x ) ≥ ⋯ ≥ σ m r ( x ) > 0
be its singular values. These values are independent of the chosen orthonormal bases; no smooth choice of bases is needed. Since
D x ( r ) ( ψ − 1 ) = ( D ψ − 1 ( x ) ( r ) ψ ) − 1 , D_x^{(r)}\left(\psi^{-1}\right)=\left(D_{\psi^{-1}(x)}^{(r)}\psi\right)^{-1}, D x ( r ) ( ψ − 1 ) = ( D ψ − 1 ( x ) ( r ) ψ ) − 1 ,
the definition of κ r , g r \kappa_{r,g_r} κ r , g r gives
κ r , g r ( ψ , 1 Γ , x ) = max { σ 1 ( x ) , σ m r ( x ) − 1 } , log κ r , g r ( ψ , 1 Γ , x ) = max 1 ≤ j ≤ m r ∣ log σ j ( x ) ∣ . (4) \kappa_{r,g_r}(\psi,1_\Gamma,x)=\max\left\{\sigma_1(x),\sigma_{m_r}(x)^{-1}\right\},
\qquad
\log\kappa_{r,g_r}(\psi,1_\Gamma,x)=\max_{1\le j\le m_r}\left|\log\sigma_j(x)\right|. \tag*{(4)} κ r , g r ( ψ , 1 Γ , x ) = max { σ 1 ( x ) , σ m r ( x ) − 1 } , log κ r , g r ( ψ , 1 Γ , x ) = 1 ≤ j ≤ m r max ∣ log σ j ( x ) ∣ . (4)
In the chosen basis of T x ( r ) M T_x^{(r)}M T x ( r ) M , the matrix of g r , x g_{r,x} g r , x is I m r I_{m_r} I m r . By the definition of the r r r -th order push-forward, the matrix of ( ψ ∗ ( r ) g r ) x (\psi_*^{(r)}g_r)_x ( ψ ∗ ( r ) g r ) x is
Q x : = A x − T A x − 1 = ( A x A x T ) − 1 . Q_x:=A_x^{-T}A_x^{-1}=(A_xA_x^T)^{-1}. Q x := A x − T A x − 1 = ( A x A x T ) − 1 .
The eigenvalues of A x A x T A_xA_x^T A x A x T are σ 1 ( x ) 2 , … , σ m r ( x ) 2 \sigma_1(x)^2,\ldots,\sigma_{m_r}(x)^2 σ 1 ( x ) 2 , … , σ m r ( x ) 2 , so those of Q x Q_x Q x are σ 1 ( x ) − 2 , … , σ m r ( x ) − 2 \sigma_1(x)^{-2},\ldots,\sigma_{m_r}(x)^{-2} σ 1 ( x ) − 2 , … , σ m r ( x ) − 2 . The distance formula for the metric d P ( m r ) d_{P(m_r)} d P ( m r ) therefore yields
d P ( T x ( r ) M ) ( ( ψ ∗ ( r ) g r ) x , g r , x ) = d P ( m r ) ( Q x , I m r ) = ( ∑ j = 1 m r ( log ( σ j ( x ) − 2 ) ) 2 ) 1 / 2 = 2 ( ∑ j = 1 m r ( log σ j ( x ) ) 2 ) 1 / 2 . (5) \begin{aligned}
d_{P(T_x^{(r)}M)}\left((\psi_*^{(r)}g_r)_x,g_{r,x}\right)
&=d_{P(m_r)}(Q_x,I_{m_r})\\
&=\left(\sum_{j=1}^{m_r}\left(\log(\sigma_j(x)^{-2})\right)^2\right)^{1/2}\\
&=2\left(\sum_{j=1}^{m_r}\left(\log\sigma_j(x)\right)^2\right)^{1/2}. \tag*{(5)}
\end{aligned} d P ( T x ( r ) M ) ( ( ψ ∗ ( r ) g r ) x , g r , x ) = d P ( m r ) ( Q x , I m r ) = ( j = 1 ∑ m r ( log ( σ j ( x ) − 2 ) ) 2 ) 1/2 = 2 ( j = 1 ∑ m r ( log σ j ( x ) ) 2 ) 1/2 . (5)
Combining (4) and (5) , we obtain
2 log κ r , g r ( ψ , 1 Γ , x ) ≤ d P ( T x ( r ) M ) ( ( ψ ∗ ( r ) g r ) x , g r , x ) ≤ 2 m r log κ r , g r ( ψ , 1 Γ , x ) . (6) 2\log\kappa_{r,g_r}(\psi,1_\Gamma,x)\le d_{P(T_x^{(r)}M)}\left((\psi_*^{(r)}g_r)_x,g_{r,x}\right)\le2\sqrt{m_r}\log\kappa_{r,g_r}(\psi,1_\Gamma,x). \tag*{(6)} 2 log κ r , g r ( ψ , 1 Γ , x ) ≤ d P ( T x ( r ) M ) ( ( ψ ∗ ( r ) g r ) x , g r , x ) ≤ 2 m r log κ r , g r ( ψ , 1 Γ , x ) . (6)
Since κ r , g r ( ψ , 1 Γ , x ) ≥ 1 \kappa_{r,g_r}(\psi,1_\Gamma,x)\ge1 κ r , g r ( ψ , 1 Γ , x ) ≥ 1 , all terms in this pointwise comparison are nonnegative. Hence, it follows from (3) and (6) that
2 p d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) p = 2 p ∫ M ( log κ r , g r ( ψ , 1 Γ , x ) ) p d ( τ ∗ μ ) ( x ) ≤ ∫ M d P ( T x ( r ) M ) ( ( ψ ∗ ( r ) g r ) x , g r , x ) p d ( τ ∗ μ ) ( x ) = d L p ( F ( γ ) , F ( γ ~ ) ) p ≤ ( 2 m r ) p ∫ M ( log κ r , g r ( ψ , 1 Γ , x ) ) p d ( τ ∗ μ ) ( x ) = ( 2 m r ) p d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) p . \begin{aligned}
2^p d_{\tau_*\mu,g_r}^{(r,p)}(\psi,1_\Gamma)^p
&=2^p\int_M\left(\log\kappa_{r,g_r}(\psi,1_\Gamma,x)\right)^p\,d(\tau_*\mu)(x)\\
&\le\int_M d_{P(T_x^{(r)}M)}\left((\psi_*^{(r)}g_r)_x,g_{r,x}\right)^p\,d(\tau_*\mu)(x)\\
&=d_{L^p}\left(F(\gamma),F(\widetilde{\gamma})\right)^p\\
&\le(2\sqrt{m_r})^p\int_M\left(\log\kappa_{r,g_r}(\psi,1_\Gamma,x)\right)^p\,d(\tau_*\mu)(x)\\
&=(2\sqrt{m_r})^p d_{\tau_*\mu,g_r}^{(r,p)}(\psi,1_\Gamma)^p.
\end{aligned} 2 p d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) p = 2 p ∫ M ( log κ r , g r ( ψ , 1 Γ , x ) ) p d ( τ ∗ μ ) ( x ) ≤ ∫ M d P ( T x ( r ) M ) ( ( ψ ∗ ( r ) g r ) x , g r , x ) p d ( τ ∗ μ ) ( x ) = d L p ( F ( γ ) , F ( γ ) ) p ≤ ( 2 m r ) p ∫ M ( log κ r , g r ( ψ , 1 Γ , x ) ) p d ( τ ∗ μ ) ( x ) = ( 2 m r ) p d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) p .
Taking p p p -th roots therefore yields
2 d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) ≤ d L p ( F ( γ ) , F ( γ ~ ) ) ≤ 2 m r d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) . (7) 2d_{\tau_*\mu,g_r}^{(r,p)}(\psi,1_\Gamma)\le d_{L^p}\left(F(\gamma),F(\widetilde{\gamma})\right)\le2\sqrt{m_r}d_{\tau_*\mu,g_r}^{(r,p)}(\psi,1_\Gamma). \tag*{(7)} 2 d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) ≤ d L p ( F ( γ ) , F ( γ ) ) ≤ 2 m r d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) . (7)
( 2 ) ⟹ ( 1 ) (2)\Longrightarrow(1) ( 2 ) ⟹ ( 1 ) . Suppose that for a proper left-invariant metric d Γ d_\Gamma d Γ on Γ \Gamma Γ and a continuous section g r : M → Riem ( r ) ( M ) g_r:M\to\operatorname{Riem}^{(r)}(M) g r : M → Riem ( r ) ( M ) , the restricted orbit map F = F μ , g r ( r , p ) ∣ Γ F=F_{\mu,g_r}^{(r,p)}\big|_\Gamma F = F μ , g r ( r , p ) Γ is a coarse embedding with lower control function ρ − \rho_- ρ − , where ρ − ( t ) → ∞ \rho_-(t)\to\infty ρ − ( t ) → ∞ as t → ∞ t\to\infty t → ∞ . By (7) and the left invariance of d Γ d_\Gamma d Γ ,
ρ − ( d Γ ( ψ , 1 Γ ) ) ≤ d L p ( F ( γ ~ ψ ) , F ( γ ~ ) ) ≤ 2 m r d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) . \rho_-\left(d_\Gamma(\psi,1_\Gamma)\right)\le d_{L^p}\left(F(\widetilde{\gamma}\psi),F(\widetilde{\gamma})\right)\le2\sqrt{m_r}d_{\tau_*\mu,g_r}^{(r,p)}(\psi,1_\Gamma). ρ − ( d Γ ( ψ , 1 Γ ) ) ≤ d L p ( F ( γ ψ ) , F ( γ ) ) ≤ 2 m r d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) .
Thus,
inf τ ∈ Γ d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) ≥ ρ ( d Γ ( ψ , 1 Γ ) ) 2 m r ⟶ ∞ as d Γ ( ψ , 1 Γ ) → ∞ , \inf_{\tau\in\Gamma} d^{(r,p)}_{\tau_*\mu,g_r}(\psi,1_\Gamma) \ge\frac{\rho\bigl(d_\Gamma(\psi,1_\Gamma)\bigr)}{2\sqrt{m_r}} \longrightarrow\infty\quad\text{as } d_\Gamma(\psi,1_\Gamma)\to\infty, τ ∈ Γ inf d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) ≥ 2 m r ρ ( d Γ ( ψ , 1 Γ ) ) ⟶ ∞ as d Γ ( ψ , 1 Γ ) → ∞ ,
i.e., Γ \Gamma Γ is ( r , p ) (r,p) ( r , p ) -th order μ \mu μ -discrete.
(1) ⟹ \Longrightarrow ⟹ (2). Suppose that Γ \Gamma Γ is ( r , p ) (r,p) ( r , p ) -th order μ \mu μ -discrete. Let d Γ d_\Gamma d Γ be any proper left-invariant metric on Γ \Gamma Γ , and let g r : M → Riem ( r ) ( M ) g_r:M\to\operatorname{Riem}^{(r)}(M) g r : M → Riem ( r ) ( M ) be any continuous section.
For R ≥ 0 R\ge0 R ≥ 0 , set
C R = max ψ ∈ Γ d Γ ( ψ , 1 Γ ) ≤ R sup x ∈ M log κ r , g r ( ψ , 1 Γ , x ) . C_R=\max_{\substack{\psi\in\Gamma\\d_\Gamma(\psi,1_\Gamma)\le R}}\sup_{x\in M}\log\kappa_{r,g_r}(\psi,1_\Gamma,x). C R = ψ ∈ Γ d Γ ( ψ , 1 Γ ) ≤ R max x ∈ M sup log κ r , g r ( ψ , 1 Γ , x ) .
Properness of d Γ d_\Gamma d Γ , compactness of M M M , and continuity of κ r , g r \kappa_{r,g_r} κ r , g r imply that C R C_R C R is finite and nondecreasing. Since ( τ ∗ μ ) ( M ) = μ ( M ) (\tau_*\mu)(M)=\mu(M) ( τ ∗ μ ) ( M ) = μ ( M ) , (7) gives
d L p ( F ( γ ) , F ( γ ~ ) ) ≤ 2 m r μ ( M ) 1 / p C R whenever d Γ ( γ , γ ~ ) ≤ R . d_{L^p}\bigl(F(\gamma),F(\widetilde{\gamma})\bigr)\le2\sqrt{m_r}\mu(M)^{1/p}C_R \quad\text{whenever } d_\Gamma(\gamma,\widetilde{\gamma})\le R. d L p ( F ( γ ) , F ( γ ) ) ≤ 2 m r μ ( M ) 1/ p C R whenever d Γ ( γ , γ ) ≤ R .
Thus ρ + ( R ) : = 2 m r μ ( M ) 1 / p C R \rho_+(R):=2\sqrt{m_r}\mu(M)^{1/p}C_R ρ + ( R ) := 2 m r μ ( M ) 1/ p C R is an upper control function for F F F .
For the lower control, define
a ( ψ ) = inf τ ∈ Γ d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) , ρ − ( R ) = 2 inf ψ ∈ Γ d Γ ( ψ , 1 Γ ) ≥ R a ( ψ ) . a(\psi)=\inf_{\tau\in\Gamma}d^{(r,p)}_{\tau_*\mu,g_r}(\psi,1_\Gamma),\qquad
\rho_-(R)=2\inf_{\substack{\psi\in\Gamma\\d_\Gamma(\psi,1_\Gamma)\ge R}}a(\psi). a ( ψ ) = τ ∈ Γ inf d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) , ρ − ( R ) = 2 ψ ∈ Γ d Γ ( ψ , 1 Γ ) ≥ R inf a ( ψ ) .
Since Γ \Gamma Γ is infinite and d Γ d_\Gamma d Γ is proper, ρ − \rho_- ρ − is finite-valued and nondecreasing. Moreover, ( r , p ) (r,p) ( r , p ) -th order μ \mu μ -discreteness implies that ρ − ( R ) → ∞ \rho_-(R)\to\infty ρ − ( R ) → ∞ as R → ∞ R\to\infty R → ∞ . For ψ = γ ~ − 1 γ \psi=\widetilde{\gamma}^{-1}\gamma ψ = γ − 1 γ and τ = γ ~ − 1 \tau=\widetilde{\gamma}^{-1} τ = γ − 1 , (7) yields
ρ − ( d Γ ( γ , γ ~ ) ) ≤ 2 a ( ψ ) \rho_-\bigl(d_\Gamma(\gamma,\widetilde{\gamma})\bigr)\le2a(\psi) ρ − ( d Γ ( γ , γ ) ) ≤ 2 a ( ψ )
≤ 2 d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) ≤ d L p ( F ( γ ) , F ( γ ~ ) ) . \begin{aligned}
\le2d^{(r,p)}_{\tau_*\mu,g_r}(\psi,1_\Gamma) \\
&\le d_{L^p}\bigl(F(\gamma),F(\widetilde{\gamma})\bigr).
\end{aligned} ≤ 2 d τ ∗ μ , g r ( r , p ) ( ψ , 1 Γ ) ≤ d L p ( F ( γ ) , F ( γ ) ) .
Hence F F F is a coarse embedding. □
Embedding into Banach spaces
In this section, we first show that the L p L^p L p -direct integrals of snowflake metric bundles over compact spaces equipped with finite Borel measures admit coarse embeddings into L p ′ L^{p'} L p ′ -spaces for suitable p ′ ∈ [ 1 , ∞ ) p'\in[1,\infty) p ′ ∈ [ 1 , ∞ ) . We then deduce that the strong Novikov conjecture (respectively, the coarse Baum–Connes conjecture) holds for every countable group (respectively, discrete metric space with bounded geometry) that admits a coarse embedding into the L p L^p L p -direct integral of a snowflake metric bundle over a compact space. As a corollary, the main theorem will follow by combining this result with Proposition 2.10 .
Embedding of L p L^p L p -direct integrals
We work in the more general setting of metacompact base spaces. We first show that a metric bundle with separable typical fiber over such a space admits a measurable trivialization. This allows us to identify its L p L^p L p -direct integral isometrically with an L p L^p L p -continuum product.
Lemma 3.1. Let Ω \Omega Ω be a metacompact space equipped with a Borel measure μ \mu μ , and let π : E → Ω \pi:E\to\Omega π : E → Ω be a metric bundle with a separable typical fiber ( X , d X ) (X,d_X) ( X , d X ) . Then E E E is a measurable metric bundle. Moreover, for every p ∈ [ 1 , + ∞ ) p\in[1,+\infty) p ∈ [ 1 , + ∞ ) and μ \mu μ -measurable base section s 0 : Ω → E s_0:\Omega\to E s 0 : Ω → E , there exists an isometric bijection
Ψ ^ : L s 0 p ( Ω , μ , E ) ⟶ L s ^ 0 p ( Ω , μ , X ) , \widehat{\Psi}:L^p_{s_0}(\Omega,\mu,E)\longrightarrow L^p_{\widehat{s}_0}(\Omega,\mu,X), Ψ : L s 0 p ( Ω , μ , E ) ⟶ L s 0 p ( Ω , μ , X ) , where s ^ 0 : Ω → X \widehat{s}_0:\Omega\to X s 0 : Ω → X is a μ \mu μ -measurable map associated to s 0 s_0 s 0 .
Proof. By metacompactness, there exists a point-finite open cover { U i } i ∈ Λ \{U_i\}_{i\in\Lambda} { U i } i ∈ Λ of Ω \Omega Ω (that is, each point belongs to only finitely many U i U_i U i ) with local trivializations
Φ i : π − 1 ( U i ) ⟶ U i × X . \Phi_i:\pi^{-1}(U_i)\longrightarrow U_i\times X. Φ i : π − 1 ( U i ) ⟶ U i × X .
For each z ∈ Ω z\in\Omega z ∈ Ω , let
Λ ( z ) = { i ∈ Λ : z ∈ U i } . \Lambda(z)=\{i\in\Lambda:z\in U_i\}. Λ ( z ) = { i ∈ Λ : z ∈ U i } .
This set is finite and nonempty. For n ≥ 1 n\ge1 n ≥ 1 , set
G n = { z ∈ Ω : ∣ Λ ( z ) ∣ ≥ n } , B n = G n ∖ G n + 1 . G_n=\{z\in\Omega:|\Lambda(z)|\ge n\},\qquad B_n=G_n\setminus G_{n+1}. G n = { z ∈ Ω : ∣Λ ( z ) ∣ ≥ n } , B n = G n ∖ G n + 1 .
Every set G n G_n G n is open, since
G n = ⋃ F ⊆ Λ ∣ F ∣ = n ⋂ i ∈ F U i . G_n=\bigcup_{\substack{F\subseteq\Lambda\\ |F|=n}}\bigcap_{i\in F}U_i. G n = F ⊆ Λ ∣ F ∣ = n ⋃ i ∈ F ⋂ U i .
Thus each B n B_n B n is Borel. Since every Λ ( z ) \Lambda(z) Λ ( z ) is finite and nonempty,
Ω = ⨆ n = 1 ∞ B n . \Omega=\bigsqcup_{n=1}^{\infty}B_n. Ω = n = 1 ⨆ ∞ B n .
Hence the sets B n B_n B n form a countable Borel partition of Ω \Omega Ω , even though Λ \Lambda Λ may be uncountable.
By Zermelo’s well-ordering theorem, fix a well-order on Λ \Lambda Λ . For each z ∈ Ω z\in\Omega z ∈ Ω , let i ( z ) i(z) i ( z ) be the least element of Λ ( z ) \Lambda(z) Λ ( z ) with respect to this order. Define
Ψ : E ⟶ Ω × X , Ψ ( u ) : = Φ i ( π ( u ) ) ( u ) . \Psi:E\longrightarrow\Omega\times X,\qquad\Psi(u):=\Phi_{i(\pi(u))}(u). Ψ : E ⟶ Ω × X , Ψ ( u ) := Φ i ( π ( u )) ( u ) .
This is a bijection preserving base points, with inverse
Ψ − 1 ( z , x ) : = Φ i ( z ) − 1 ( z , x ) . \Psi^{-1}(z,x):=\Phi_{i(z)}^{-1}(z,x). Ψ − 1 ( z , x ) := Φ i ( z ) − 1 ( z , x ) .
Let π ^ : Ω × X → X \hat{\pi}:\Omega\times X\to X π ^ : Ω × X → X denote the projection onto the second factor. By construction,
d X ( π ^ ( Ψ ( u ) ) , π ^ ( Ψ ( v ) ) ) = d z ( u , v ) , u , v ∈ E z . (8) d_X\bigl(\hat{\pi}(\Psi(u)),\hat{\pi}(\Psi(v))\bigr)=d_z(u,v),\qquad u,v\in E_z. \tag*{(8)} d X ( π ^ ( Ψ ( u )) , π ^ ( Ψ ( v )) ) = d z ( u , v ) , u , v ∈ E z . (8)
We claim that, for every n ≥ 1 n\ge1 n ≥ 1 , the restriction
Ψ n : = Ψ ∣ π − 1 ( B n ) : π − 1 ( B n ) ⟶ B n × X \Psi_n:=\left.\Psi\right|_{\pi^{-1}(B_n)}:\pi^{-1}(B_n)\longrightarrow B_n\times X Ψ n := Ψ ∣ π − 1 ( B n ) : π − 1 ( B n ) ⟶ B n × X
is a homeomorphism. Indeed, for z ∈ B n z\in B_n z ∈ B n , the set
V z : = ⋂ i ∈ Λ ( z ) U i V_z:=\bigcap_{i\in\Lambda(z)}U_i V z := i ∈ Λ ( z ) ⋂ U i
is an open neighborhood of z z z . If z ′ ∈ V z ∩ B n z'\in V_z\cap B_n z ′ ∈ V z ∩ B n , then Λ ( z ) ⊆ Λ ( z ′ ) \Lambda(z)\subseteq\Lambda(z') Λ ( z ) ⊆ Λ ( z ′ ) and both sets have exactly n n n elements. Thus Λ ( z ′ ) = Λ ( z ) \Lambda(z')=\Lambda(z) Λ ( z ′ ) = Λ ( z ) , so i ( z ′ ) = i ( z ) i(z')=i(z) i ( z ′ ) = i ( z ) . Consequently,
Ψ ∣ π − 1 ( V z ∩ B n ) : π − 1 ( V z ∩ B n ) ⟶ ( V z ∩ B n ) × X \left.\Psi\right|_{\pi^{-1}(V_z\cap B_n)}:\pi^{-1}(V_z\cap B_n)\longrightarrow(V_z\cap B_n)\times X Ψ ∣ π − 1 ( V z ∩ B n ) : π − 1 ( V z ∩ B n ) ⟶ ( V z ∩ B n ) × X
is the restriction of Φ i ( z ) \Phi_{i(z)} Φ i ( z ) and is therefore a homeomorphism. As z z z ranges over B n B_n B n , the sets ( V z ∩ B n ) (V_z\cap B_n) ( V z ∩ B n ) and ( V z ∩ B n ) × X (V_z\cap B_n)\times X ( V z ∩ B n ) × X form open covers of π − 1 ( B n ) \pi^{-1}(B_n) π − 1 ( B n ) and B n × X B_n\times X B n × X , respectively, in their subspace topologies. Hence both Ψ n \Psi_n Ψ n and Ψ n − 1 \Psi_n^{-1} Ψ n − 1 are continuous, proving the claim.
The sets π − 1 ( B n ) \pi^{-1}(B_n) π − 1 ( B n ) and B n × X B_n\times X B n × X are Borel, since π \pi π and the projection Ω × X → Ω \Omega\times X\to\Omega Ω × X → Ω are continuous. They give countable Borel decompositions of E E E and Ω × X \Omega\times X Ω × X , respectively. As each Ψ n \Psi_n Ψ n is a homeomorphism, both Ψ \Psi Ψ and Ψ − 1 \Psi^{-1} Ψ − 1 are Borel measurable.
Since X X X is separable, the function ( x , y ) ↦ d X ( x , y ) (x,y)\mapsto d_X(x,y) ( x , y ) ↦ d X ( x , y ) is measurable with respect to the product σ \sigma σ -algebra B ( X ) ⊗ B ( X ) \mathcal{B}(X)\otimes\mathcal{B}(X) B ( X ) ⊗ B ( X ) , and hence the fiberwise distance function of Ω × X \Omega\times X Ω × X is measurable. Together with the Borel measurability of Ψ \Psi Ψ and its preservation of distances within each fiber, this implies that the fiberwise distance function of E E E is also measurable.
For a section s : Ω → E s : \Omega\to E s : Ω → E , set
s ^ = π ^ ∘ Ψ ∘ s . \widehat{s}=\widehat{\pi}\circ\Psi\circ s. s = π ∘ Ψ ∘ s .
Since X X X is separable, a map f : Ω → X f : \Omega\to X f : Ω → X is μ \mu μ -measurable if and only if the associated section z ↦ ( z , f ( z ) ) z\mapsto(z,f(z)) z ↦ ( z , f ( z )) of the trivial bundle Ω × X → Ω \Omega\times X\to\Omega Ω × X → Ω is μ \mu μ -measurable. Together with the Borel measurability of Ψ \Psi Ψ and Ψ − 1 \Psi^{-1} Ψ − 1 , this shows that
s ⟼ s ^ s\longmapsto\widehat{s} s ⟼ s
is a bijection between μ \mu μ -measurable sections of E E E and μ \mu μ -measurable maps Ω → X \Omega\to X Ω → X . Moreover,
s = t μ -a.e. ⟺ s ^ = t ^ μ -a.e. s=t\quad\mu\text{-a.e.}\quad\Longleftrightarrow\quad\widehat{s}=\widehat{t}\quad\mu\text{-a.e.} s = t μ -a.e. ⟺ s = t μ -a.e.
Combining with (8) these observations, we obtain
∫ Ω d z ( s ( z ) , t ( z ) ) p d μ ( z ) = ∫ Ω d X ( s ^ ( z ) , t ^ ( z ) ) p d μ ( z ) \int_{\Omega}d_z(s(z),t(z))^p\,d\mu(z)=\int_{\Omega}d_X(\widehat{s}(z),\widehat{t}(z))^p\,d\mu(z) ∫ Ω d z ( s ( z ) , t ( z ) ) p d μ ( z ) = ∫ Ω d X ( s ( z ) , t ( z ) ) p d μ ( z )
for any two μ \mu μ -measurable sections s , t : Ω → E s,t:\Omega\to E s , t : Ω → E . It therefore induces a well-defined isometric bijection
Ψ ^ : L s 0 p ( Ω , μ , E ) ⟶ L s ^ 0 p ( Ω , μ , X ) , Ψ ^ ( [ s ] ) : = [ s ^ ] . \widehat{\Psi}:L^p_{s_0}(\Omega,\mu,E)\longrightarrow L^p_{\widehat{s}_0}(\Omega,\mu,X),\qquad\widehat{\Psi}([s]):=[\widehat{s}]. Ψ : L s 0 p ( Ω , μ , E ) ⟶ L s 0 p ( Ω , μ , X ) , Ψ ([ s ]) := [ s ] .
□ \square □
We emphasize that every snowflake metric bundle over a metacompact space is measurable by Lemma 3.1, since its typical fiber is separable. Consequently, the construction of L p L^p L p -direct integrals in Definition 2.2 applies to every snowflake metric bundle. The following result is a generalization of [19 Proposition 12.3] , which shows that, although L p L^p L p -direct integrals are highly nonlinear, they nevertheless admit coarse embeddings into Banach spaces.
Proposition 3.2. Let Ω \Omega Ω be a metacompact space equipped with a σ \sigma σ -finite Borel measure μ \mu μ , and let π : E → Ω \pi:E\to\Omega π : E → Ω be a snowflake metric bundle with typical fiber ( X , d X ) (X,d_X) ( X , d X ) . Then for every p ∈ [ 1 , ∞ ) p\in[1,\infty) p ∈ [ 1 , ∞ ) , there exists p ′ ∈ [ 1 , ∞ ) p'\in[1,\infty) p ′ ∈ [ 1 , ∞ ) such that for any μ \mu μ -measurable base section s 0 : Ω → E s_0:\Omega\to E s 0 : Ω → E , the L p L^p L p -direct integral L s 0 p ( Ω , μ , E ) L^p_{s_0}(\Omega,\mu,E) L s 0 p ( Ω , μ , E ) admits a coarse embedding into an L p ′ L^{p'} L p ′ -space L p ′ ( Ω ′ , μ ′ ) L^{p'}(\Omega',\mu') L p ′ ( Ω ′ , μ ′ ) .
Proof. Fix p ∈ [ 1 , ∞ ) p\in[1,\infty) p ∈ [ 1 , ∞ ) .
Since X X X admits a snowflake embedding into a separable Hilbert space, X X X is separable. By Lemma 3.1, it therefore suffices to show that
L f 0 p ( Ω , μ , X ) L^p_{f_0}(\Omega,\mu,X) L f 0 p ( Ω , μ , X )
admits a coarse embedding into an L p ′ L^{p'} L p ′ -space L p ′ ( Ω ′ , μ ′ ) L^{p'}(\Omega',\mu') L p ′ ( Ω ′ , μ ′ ) , where f 0 : Ω → X f_0:\Omega\to X f 0 : Ω → X is a μ \mu μ -measurable map.
Choose a snowflake embedding
h : X ⟶ ℓ 2 ( N ) . h:X\longrightarrow\ell^2(\mathbb{N}). h : X ⟶ ℓ 2 ( N ) .
Thus there exist α ∈ ( 0 , 1 ] \alpha\in(0,1] α ∈ ( 0 , 1 ] and constants 0 < c ≤ C < ∞ 0<c\le C<\infty 0 < c ≤ C < ∞ such that
c d X ( x , y ) α ≤ ∥ h ( x ) − h ( y ) ∥ ≤ C d X ( x , y ) α , x , y ∈ X . (9) c d_X(x,y)^\alpha\le\lVert h(x)-h(y)\rVert\le C d_X(x,y)^\alpha,\qquad x,y\in X. \tag*{(9)} c d X ( x , y ) α ≤ ∥ h ( x ) − h ( y )∥ ≤ C d X ( x , y ) α , x , y ∈ X . (9)
Set
p ′ = p α . p'=\frac{p}{\alpha}. p ′ = α p .
Then p ′ ∈ [ 1 , ∞ ) p' \in[1,\infty) p ′ ∈ [ 1 , ∞ ) and α p ′ = p \alpha p'=p α p ′ = p .
Let m m m denote the Lebesgue measure on [ 0 , 1 ] [0,1] [ 0 , 1 ] , and set
Y = L p ′ ( [ 0 , 1 ] , m ) . Y=L^{p'}([0,1],m). Y = L p ′ ([ 0 , 1 ] , m ) .
The Gaussian construction (see Proposition 6.4.12 in [1 ] ) gives a linear isometric embedding
I : ℓ 2 ( N ) ⟶ Y . \mathcal{I}:\ell^2(\mathbb{N})\longrightarrow Y. I : ℓ 2 ( N ) ⟶ Y .
Indeed, let { g n } n ≥ 1 \{g_n\}_{n\geq1} { g n } n ≥ 1 be independent standard real Gaussian random variables on ( [ 0 , 1 ] , m ) ([0,1],m) ([ 0 , 1 ] , m ) . These variables belong to L p ′ ( [ 0 , 1 ] , m ) L^{p'}([0,1],m) L p ′ ([ 0 , 1 ] , m ) , and every finite real linear combination ∑ j a j g j \sum_j a_jg_j ∑ j a j g j has the same distribution as ( ∑ j a j 2 ) 1 / 2 g 1 \left(\sum_j a_j^2\right)^{1/2}g_1 ( ∑ j a j 2 ) 1/2 g 1 . Hence
∥ ∑ j a j g j ∥ L p ′ = ∥ g 1 ∥ L p ′ ( ∑ j ∣ a j ∣ 2 ) 1 / 2 . \left\|\sum_j a_jg_j\right\|_{L^{p'}}=\|g_1\|_{L^{p'}}\left(\sum_j |a_j|^2\right)^{1/2}. j ∑ a j g j L p ′ = ∥ g 1 ∥ L p ′ ( j ∑ ∣ a j ∣ 2 ) 1/2 .
Let { e n } n ≥ 1 \{e_n\}_{n\geq1} { e n } n ≥ 1 denote the canonical orthonormal basis of ℓ 2 ( N ) \ell^2(\mathbb{N}) ℓ 2 ( N ) . The assignment e n ↦ g n / ∥ g 1 ∥ L p ′ e_n\mapsto g_n/\|g_1\|_{L^{p'}} e n ↦ g n /∥ g 1 ∥ L p ′ extends to a linear isometric embedding I : ℓ 2 ( N ) → Y \mathcal{I}:\ell^2(\mathbb{N})\to Y I : ℓ 2 ( N ) → Y .
For a μ \mu μ -measurable map f : Ω → X f:\Omega\to X f : Ω → X representing an element of L f 0 p ( Ω , μ , X ) L^p_{f_0}(\Omega,\mu,X) L f 0 p ( Ω , μ , X ) , define f ~ : Ω → Y \widetilde f:\Omega\to Y f : Ω → Y by
f ~ ( z ) = I ( h ( f ( z ) ) − h ( f 0 ( z ) ) ) , z ∈ Ω . \widetilde f(z)=\mathcal{I}\bigl(h(f(z))-h(f_0(z))\bigr),\qquad z\in\Omega. f ( z ) = I ( h ( f ( z )) − h ( f 0 ( z )) ) , z ∈ Ω.
The continuity of h h h and I \mathcal{I} I , together with the separability of Y Y Y , implies that f ~ \widetilde f f is Bochner measurable with respect to μ \mu μ . Moreover, it follows from (9) that
∫ Ω ∥ f ~ ( z ) ∥ Y p ′ d μ ( z ) = ∫ Ω ∥ h ( f ( z ) ) − h ( f 0 ( z ) ) ∥ 2 p ′ d μ ( z ) ≤ C p ′ ∫ Ω d X ( f ( z ) , f 0 ( z ) ) p d μ ( z ) < ∞ . \begin{aligned}
\int_\Omega\|\widetilde f(z)\|_Y^{p'}\,d\mu(z)=\int_\Omega\|h(f(z))-h(f_0(z))\|_2^{p'}\,d\mu(z) \\
&\leq C^{p'}\int_\Omega d_X(f(z),f_0(z))^p\,d\mu(z)<\infty.
\end{aligned} ∫ Ω ∥ f ( z ) ∥ Y p ′ d μ ( z ) = ∫ Ω ∥ h ( f ( z )) − h ( f 0 ( z )) ∥ 2 p ′ d μ ( z ) ≤ C p ′ ∫ Ω d X ( f ( z ) , f 0 ( z ) ) p d μ ( z ) < ∞.
Thus f ~ \widetilde f f represents an element of the Bochner space L p ′ ( Ω , μ , Y ) L^{p'}(\Omega,\mu,Y) L p ′ ( Ω , μ , Y ) . Note that the class [ f ~ ] [\widetilde f] [ f ] depends only on [ f ] [f] [ f ] . Hence the map
ψ : L f 0 p ( Ω , μ , X ) ⟶ L p ′ ( Ω , μ , Y ) , ψ ( [ f ] ) : = [ f ~ ] , \psi:L^p_{f_0}(\Omega,\mu,X)\longrightarrow L^{p'}(\Omega,\mu,Y),\qquad\psi([f]):=[\widetilde f], ψ : L f 0 p ( Ω , μ , X ) ⟶ L p ′ ( Ω , μ , Y ) , ψ ([ f ]) := [ f ] ,
is well-defined.
For any [ f ] , [ g ] ∈ L f 0 p ( Ω , μ , X ) [f],[g]\in L^p_{f_0}(\Omega,\mu,X) [ f ] , [ g ] ∈ L f 0 p ( Ω , μ , X ) , the linearity and isometry of I \mathcal{I} I give
∥ ψ ( [ f ] ) − ψ ( [ g ] ) ∥ L p ′ ( Ω , μ , Y ) p ′ = ∫ Ω ∥ f ~ ( z ) − g ~ ( z ) ∥ Y p ′ d μ ( z ) = ∫ Ω ∥ h ( f ( z ) ) − h ( g ( z ) ) ∥ 2 p ′ d μ ( z ) . \begin{aligned}
\|\psi([f])-\psi([g])\|_{L^{p'}(\Omega,\mu,Y)}^{p'}=\int_\Omega\|\widetilde f(z)-\widetilde g(z)\|_Y^{p'}\,d\mu(z) \\
&=\int_\Omega\|h(f(z))-h(g(z))\|_2^{p'}\,d\mu(z).
\end{aligned} ∥ ψ ([ f ]) − ψ ([ g ]) ∥ L p ′ ( Ω , μ , Y ) p ′ = ∫ Ω ∥ f ( z ) − g ( z ) ∥ Y p ′ d μ ( z ) = ∫ Ω ∥ h ( f ( z )) − h ( g ( z )) ∥ 2 p ′ d μ ( z ) .
It follows from (9) that
c d L p ( [ f ] , [ g ] ) α ≤ ∥ ψ ( [ f ] ) − ψ ( [ g ] ) ∥ L p ′ ( Ω , μ , Y ) ≤ C d L p ( [ f ] , [ g ] ) α . c\,d_{L^p}([f],[g])^\alpha\leq\|\psi([f])-\psi([g])\|_{L^{p'}(\Omega,\mu,Y)}\leq C\,d_{L^p}([f],[g])^\alpha. c d L p ([ f ] , [ g ] ) α ≤ ∥ ψ ([ f ]) − ψ ([ g ]) ∥ L p ′ ( Ω , μ , Y ) ≤ C d L p ([ f ] , [ g ] ) α .
Thus, ψ \psi ψ is a coarse embedding.
Finally, set
Ω ′ = Ω × [ 0 , 1 ] and μ ′ = μ ⊗ m , \Omega'=\Omega\times[0,1]\quad\text{and}\quad\mu'=\mu\otimes m, Ω ′ = Ω × [ 0 , 1 ] and μ ′ = μ ⊗ m ,
where μ ⊗ m \mu\otimes m μ ⊗ m denotes the product measure. By the Fubini isomorphism for Bochner spaces (see [11 Proposition 1.2.24] ), there is a canonical isometric isomorphism
L p ′ ( Ω , μ , Y ) ⟶ ≅ L p ′ ( Ω ′ , μ ′ ) . L^{p'}(\Omega,\mu,Y) \overset{\cong}{\longrightarrow} L^{p'}(\Omega',\mu'). L p ′ ( Ω , μ , Y ) ⟶ ≅ L p ′ ( Ω ′ , μ ′ ) .
Composing ψ \psi ψ with this isomorphism gives the required coarse embedding. This completes the proof.
The preceding argument generally changes the exponent from p p p to p ′ = p / α p'=p/\alpha p ′ = p / α .
Proof of the main theorem
We first briefly recall the definitions of the strong Novikov conjecture and the coarse Baum–Connes conjecture.
Let Γ \Gamma Γ be a countable discrete group equipped with a proper left-invariant metric, and denote the so-obtained metric space by ∣ Γ ∣ |\Gamma| ∣Γ∣ . Let X X X be a discrete metric space with bounded geometry.
For R > 0 R>0 R > 0 , let P R ( Γ ) P_R(\Gamma) P R ( Γ ) denote the Rips complex with vertex set Γ \Gamma Γ , in which a finite subset spans a simplex if and only if its diameter is at most R R R . The Rips complex P R ( Γ ) P_R(\Gamma) P R ( Γ ) is equipped with the spherical metric (cf. [24 ] ). The Rips complex P R ( X ) P_R(X) P R ( X ) is defined similarly.
The group Γ \Gamma Γ acts on P R ( Γ ) P_R(\Gamma) P R ( Γ ) by γ ⋅ ( ∑ g a g g ) = ∑ g a g γ g \gamma\cdot\left(\sum_g a_g g\right)=\sum_g a_g\gamma g γ ⋅ ( ∑ g a g g ) = ∑ g a g γ g for ∑ g a g g ∈ P R ( Γ ) \sum_g a_g g\in P_R(\Gamma) ∑ g a g g ∈ P R ( Γ ) and γ ∈ Γ \gamma\in\Gamma γ ∈ Γ . This is a proper action by isometries.
The Baum–Connes and coarse Baum–Connes assembly maps are, respectively,
μ Γ : lim R → ∞ K ∗ Γ ( P R ( Γ ) ) ⟶ K ∗ ( C r ∗ ( Γ ) ) , \mu_\Gamma:\lim_{R\to\infty}K_*^\Gamma(P_R(\Gamma))\longrightarrow K_*(C_r^*(\Gamma)), μ Γ : R → ∞ lim K ∗ Γ ( P R ( Γ )) ⟶ K ∗ ( C r ∗ ( Γ )) ,
μ X : lim R → ∞ K ∗ ( P R ( X ) ) ⟶ K ∗ ( C ∗ ( X ) ) . \mu_X:\lim_{R\to\infty}K_*(P_R(X))\longrightarrow K_*(C^*(X)). μ X : R → ∞ lim K ∗ ( P R ( X )) ⟶ K ∗ ( C ∗ ( X )) .
where K ∗ Γ ( P R ( Γ ) ) K_*^\Gamma(P_R(\Gamma)) K ∗ Γ ( P R ( Γ )) denotes its equivariant K K K -homology, K ∗ ( P R ( X ) ) K_*(P_R(X)) K ∗ ( P R ( X )) denotes the non-equivariant K K K -homology of P R ( X ) P_R(X) P R ( X ) , C r ∗ ( Γ ) C_r^*(\Gamma) C r ∗ ( Γ ) is the reduced group C ∗ C^* C ∗ -algebra of Γ \Gamma Γ , and C ∗ ( X ) C^*(X) C ∗ ( X ) is the Roe algebra of X X X . Both direct limits are induced by inclusions of Rips complexes as R R R increases. In the language of equivariant K K KK K K -theory, the direct limit lim R → ∞ K ∗ Γ ( P R ( Γ ) ) \lim_{R\to\infty}K_*^\Gamma(P_R(\Gamma)) lim R → ∞ K ∗ Γ ( P R ( Γ )) is canonically identified with K K ∗ Γ ( E Γ ) KK_*^\Gamma(E\Gamma) K K ∗ Γ ( E Γ ) , as in [12 Section 3] .
The strong Novikov conjecture for Γ \Gamma Γ and the coarse Baum–Connes conjecture for X X X assert, respectively, that μ Γ \mu_\Gamma μ Γ is injective and μ X \mu_X μ X is an isomorphism. Kasparov and Yu [12 ] proved that the strong Novikov conjecture holds for every countable discrete group admitting a coarse embedding into an ℓ p \ell^p ℓ p -space for some 1 ≤ p < ∞ 1\leq p<\infty 1 ≤ p < ∞ . Wang, Xie, Yu and Zhu [20 ] showed that the coarse Baum–Connes conjecture holds for any discrete metric space with bounded geometry that admits a coarse embedding into an ℓ p \ell^p ℓ p -space for some 1 ≤ p < ∞ 1\leq p<\infty 1 ≤ p < ∞ .
We need the following useful lemma.
Lemma 3.3. Let 1 ≤ p < ∞ 1\leq p<\infty 1 ≤ p < ∞ , and let ( Γ , d Γ ) (\Gamma,d_\Gamma) ( Γ , d Γ ) be either a countable discrete group equipped with a proper left-invariant metric or a discrete metric space with bounded geometry. If ( Γ , d Γ ) (\Gamma,d_\Gamma) ( Γ , d Γ ) admits a coarse embedding into an L p L^p L p -space, then it admits a coarse embedding into ℓ p ( N ) \ell^p(\mathbb{N}) ℓ p ( N ) .
Proof. Suppose that
h : Γ ⟶ L p ( Ω , μ ) h:\Gamma\longrightarrow L^p(\Omega,\mu) h : Γ ⟶ L p ( Ω , μ )
is a coarse embedding. Thus, there exist nondecreasing functions
ρ − , ρ + : [ 0 , ∞ ) ⟶ [ 0 , ∞ ) \rho_-,\rho_+:[0,\infty)\longrightarrow[0,\infty) ρ − , ρ + : [ 0 , ∞ ) ⟶ [ 0 , ∞ )
such that lim t → ∞ ρ − ( t ) = ∞ \lim_{t\to\infty}\rho_-(t)=\infty lim t → ∞ ρ − ( t ) = ∞ and
ρ − ( d Γ ( γ , γ ~ ) ) ≤ ∥ h ( γ ) − h ( γ ~ ) ∥ L p ≤ ρ + ( d Γ ( γ , γ ~ ) ) \rho_-\bigl(d_\Gamma(\gamma,\widetilde{\gamma})\bigr)\leq\lVert h(\gamma)-h(\widetilde{\gamma})\rVert_{L^p}\leq\rho_+\bigl(d_\Gamma(\gamma,\widetilde{\gamma})\bigr) ρ − ( d Γ ( γ , γ ) ) ≤ ∥ h ( γ ) − h ( γ ) ∥ L p ≤ ρ + ( d Γ ( γ , γ ) )
for all γ , γ ~ ∈ Γ \gamma,\widetilde{\gamma} \in\Gamma γ , γ ∈ Γ .
We first show that all finite subsets of Γ \Gamma Γ admit uniformly controlled coarse embeddings into ℓ p ( N ) \ell^p(\mathbb{N}) ℓ p ( N ) . Let E ⊆ Γ E \subseteq\Gamma E ⊆ Γ be finite. Since simple functions are dense in L p ( Ω , μ ) L^p(\Omega,\mu) L p ( Ω , μ ) , for every γ ∈ E \gamma\in E γ ∈ E we may choose a simple function s γ ∈ L p ( Ω , μ ) s_\gamma\in L^p(\Omega,\mu) s γ ∈ L p ( Ω , μ ) such that
∥ h ( γ ) − s γ ∥ L p < 1 2 . \|h(\gamma)-s_\gamma\|_{L^p}<\frac{1}{2}. ∥ h ( γ ) − s γ ∥ L p < 2 1 .
Because E E E is finite, after passing to a common refinement of the finitely many measurable partitions involved, there exist pairwise disjoint measurable sets
A 1 , … , A N ⊆ Ω , 0 < μ ( A j ) < ∞ , A_1,\ldots,A_N\subseteq\Omega,\qquad0<\mu(A_j)<\infty, A 1 , … , A N ⊆ Ω , 0 < μ ( A j ) < ∞ ,
and scalars c γ , j c_{\gamma,j} c γ , j such that
s γ = ∑ j = 1 N c γ , j 1 A j for every γ ∈ E . s_\gamma=\sum_{j=1}^{N}c_{\gamma,j}1_{A_j}\qquad\text{for every }\gamma\in E. s γ = j = 1 ∑ N c γ , j 1 A j for every γ ∈ E .
Define
F E : E ⟶ ℓ N p ⊆ ℓ p ( N ) F_E:E\longrightarrow\ell^p_N\subseteq\ell^p(\mathbb{N}) F E : E ⟶ ℓ N p ⊆ ℓ p ( N )
by
F E ( γ ) = ( μ ( A 1 ) 1 / p c γ , 1 , … , μ ( A N ) 1 / p c γ , N ) . F_E(\gamma)=\left(\mu(A_1)^{1/p}c_{\gamma,1},\ldots,\mu(A_N)^{1/p}c_{\gamma,N}\right). F E ( γ ) = ( μ ( A 1 ) 1/ p c γ , 1 , … , μ ( A N ) 1/ p c γ , N ) .
For all γ , γ ~ ∈ E \gamma,\widetilde{\gamma}\in E γ , γ ∈ E , we then have
∥ F E ( γ ) − F E ( γ ~ ) ∥ ℓ p p = ∑ j = 1 N μ ( A j ) ∣ c γ , j − c γ ~ , j ∣ p = ∥ s γ − s γ ~ ∥ L p p . \begin{aligned}
\|F_E(\gamma)-F_E(\widetilde{\gamma})\|_{\ell^p}^{p}
=\sum_{j=1}^{N}\mu(A_j)|c_{\gamma,j}-c_{\widetilde{\gamma},j}|^{p} \\
&=\|s_\gamma-s_{\widetilde{\gamma}}\|_{L^p}^{p}.
\end{aligned} ∥ F E ( γ ) − F E ( γ ) ∥ ℓ p p = j = 1 ∑ N μ ( A j ) ∣ c γ , j − c γ , j ∣ p = ∥ s γ − s γ ∥ L p p .
Hence
∥ F E ( γ ) − F E ( γ ~ ) ∥ ℓ p = ∥ s γ − s γ ~ ∥ L p . \|F_E(\gamma)-F_E(\widetilde{\gamma})\|_{\ell^p}
=\|s_\gamma-s_{\widetilde{\gamma}}\|_{L^p}. ∥ F E ( γ ) − F E ( γ ) ∥ ℓ p = ∥ s γ − s γ ∥ L p .
Moreover, the reverse triangle inequality gives
∣ ∥ s γ − s γ ~ ∥ L p − ∥ h ( γ ) − h ( γ ~ ) ∥ L p ∣ ≤ ∥ ( s γ − h ( γ ) ) − ( s γ ~ − h ( γ ~ ) ) ∥ L p ≤ ∥ s γ − h ( γ ) ∥ L p + ∥ s γ ~ − h ( γ ~ ) ∥ L p < 1. \begin{aligned}
\left|\|s_\gamma-s_{\widetilde{\gamma}}\|_{L^p}
-\|h(\gamma)-h(\widetilde{\gamma})\|_{L^p}\right|
\leq\|(s_\gamma-h(\gamma))-(s_{\widetilde{\gamma}}-h(\widetilde{\gamma}))\|_{L^p} \\
&\leq\|s_\gamma-h(\gamma)\|_{L^p}
+\|s_{\widetilde{\gamma}}-h(\widetilde{\gamma})\|_{L^p}
<1.
\end{aligned} ∣ ∥ s γ − s γ ∥ L p − ∥ h ( γ ) − h ( γ ) ∥ L p ∣ ≤ ∥ ( s γ − h ( γ )) − ( s γ − h ( γ )) ∥ L p ≤ ∥ s γ − h ( γ ) ∥ L p + ∥ s γ − h ( γ ) ∥ L p < 1.
Consequently,
max { 0 , ρ − ( d Γ ( γ , γ ~ ) ) − 1 } ≤ ∥ F E ( γ ) − F E ( γ ~ ) ∥ ℓ p ≤ ρ + ( d Γ ( γ , γ ~ ) ) + 1. \max\{0,\rho_-(d_\Gamma(\gamma,\widetilde{\gamma}))-1\}
\leq\|F_E(\gamma)-F_E(\widetilde{\gamma})\|_{\ell^p}
\leq\rho_+(d_\Gamma(\gamma,\widetilde{\gamma}))+1. max { 0 , ρ − ( d Γ ( γ , γ )) − 1 } ≤ ∥ F E ( γ ) − F E ( γ ) ∥ ℓ p ≤ ρ + ( d Γ ( γ , γ )) + 1.
Define
ρ ~ − ( t ) = max { 0 , ρ − ( t ) − 1 } , ρ ~ + ( t ) = ρ + ( t ) + 1. \widetilde{\rho}_-(t)=\max\{0,\rho_-(t)-1\},\qquad
\widetilde{\rho}_+(t)=\rho_+(t)+1. ρ − ( t ) = max { 0 , ρ − ( t ) − 1 } , ρ + ( t ) = ρ + ( t ) + 1.
These functions are independent of the finite subset E E E , and
lim t → ∞ ρ ~ − ( t ) = ∞ . \lim_{t\to\infty}\widetilde{\rho}_-(t)=\infty. t → ∞ lim ρ − ( t ) = ∞.
Thus, all finite subsets of Γ \Gamma Γ admit uniformly controlled coarse embeddings into ℓ p ( N ) \ell^p(\mathbb{N}) ℓ p ( N ) .
In either case, every bounded subset of Γ \Gamma Γ is finite. We may apply [16 Theorem 1.3] to conclude that Γ \Gamma Γ admits a coarse embedding into ℓ p ( N ) \ell^p(\mathbb{N}) ℓ p ( N ) . □
We now present a general result, which will lead to the proof of the main theorem.
Theorem 3.4. Let Ω \Omega Ω be a metacompact space equipped with a σ \sigma σ -finite Borel measure μ \mu μ , and let π : E → Ω \pi:E\to\Omega π : E → Ω be a snowflake metric bundle . Then the strong Novikov conjecture
(respectively, the coarse Baum–Connes conjecture) holds for every countable group (respectively, discrete metric space with bounded geometry) that admits a coarse embedding into the L p L^{p} L p -direct integral
L s 0 p ( Ω , μ , E ) L_{s_0}^{p}(\Omega,\mu,E) L s 0 p ( Ω , μ , E )
for some μ \mu μ -measurable base section s 0 : Ω → E s_0:\Omega\to E s 0 : Ω → E and some p ∈ [ 1 , + ∞ ) p\in[1,+\infty) p ∈ [ 1 , + ∞ ) .
Proof. Let Γ \Gamma Γ be a countable discrete group equipped with a proper left-invariant metric (respectively, a discrete metric space with bounded geometry). Suppose that Γ \Gamma Γ admits a coarse embedding into the L p L^{p} L p -direct integral
L s 0 p ( Ω , μ , E ) L_{s_0}^{p}(\Omega,\mu,E) L s 0 p ( Ω , μ , E )
for some p ∈ [ 1 , ∞ ) p\in[1,\infty) p ∈ [ 1 , ∞ ) and some μ \mu μ -measurable base section s 0 : Ω → E s_0:\Omega\to E s 0 : Ω → E . Combining Proposition 3.2 and Lemma 3.3, we obtain a coarse embedding of Γ \Gamma Γ into ℓ p ′ ( N ) \ell^{p'}(\mathbb{N}) ℓ p ′ ( N ) for some p ′ ∈ [ 1 , ∞ ) p'\in[1,\infty) p ′ ∈ [ 1 , ∞ ) . The desired conclusion now follows from the results of Kasparov–Yu [12 ] and Wang–Xie–Yu–Zhu [20 ] . □
Finally, we are ready to prove Theorem 1.1 .
Proof of Theorem 1.1 . The desired result follows immediately from Proposition 2.10 and Theorem 3.4 . □
Remark 3.5. We note that Theorem 3.4 may be of independent interest and have applications beyond the proof of our main theorem. Indeed, Theorem 1.1 concerns only coarse embeddings arising from orbit maps, whereas Theorem 3.4 applies to general coarse embeddings.
Examples
Throughout this section, we use the L 2 L^{2} L 2 -norm. We shall present an example demonstrating that second-order μ \mu μ -discreteness has strictly greater detecting power than its first-order counterpart. The same argument also yields, for every r ≥ 1 r\ge1 r ≥ 1 , examples that are ( r + 1 , 2 ) (r+1,2) ( r + 1 , 2 ) -th order μ 0 \mu_0 μ 0 -discrete for some Borel probability measure μ 0 \mu_0 μ 0 , but are not ( r , 2 ) (r,2) ( r , 2 ) -th order μ \mu μ -discrete for any finite Borel measure μ \mu μ .
Example 4.1. There exists a finitely generated amenable subgroup
Γ < Diff ( S 2 ) \Gamma<\operatorname{Diff}(S^2) Γ < Diff ( S 2 ) with the following properties:
Γ \Gamma Γ is not ( 1 , 2 ) (1,2) ( 1 , 2 ) -th order μ \mu μ -discrete for any finite Borel measure μ \mu μ on S 2 S^2 S 2 ;
Γ \Gamma Γ is ( 2 , 2 ) (2,2) ( 2 , 2 ) -th order μ 0 \mu_0 μ 0 -discrete for a Borel probability measure μ 0 \mu_0 μ 0 on S 2 S^2 S 2 .
Proof. Let
Γ = { g ( a , b , c ) = ( 1 a c 0 1 b 0 0 1 ) : a , b , c ∈ Z } \Gamma=\left\{g_{(a,b,c)}=
\begin{pmatrix}
1 & a & c\\
0 & 1 & b\\
0 & 0 & 1
\end{pmatrix}:a,b,c\in\mathbb{Z}\right\} Γ = ⎩ ⎨ ⎧ g ( a , b , c ) = 1 0 0 a 1 0 c b 1 : a , b , c ∈ Z ⎭ ⎬ ⎫
be the discrete Heisenberg group. Its multiplication law is
g ( a , b , c ) ⋅ g ( a ′ , b ′ , c ′ ) = g ( a + a ′ , b + b ′ , c + c ′ + a b ′ ) . g_{(a,b,c)}\cdot g_{(a',b',c')}=g_{(a+a',b+b',c+c'+ab')}. g ( a , b , c ) ⋅ g ( a ′ , b ′ , c ′ ) = g ( a + a ′ , b + b ′ , c + c ′ + a b ′ ) .
It is a non-abelian nilpotent group generated by
X : = g ( 1 , 0 , 0 ) and Y : = g ( 0 , 1 , 0 ) . X:=g_{(1,0,0)}\quad\text{and}\quad Y:=g_{(0,1,0)}. X := g ( 1 , 0 , 0 ) and Y := g ( 0 , 1 , 0 ) .
In particular, Γ \Gamma Γ is amenable.
Every matrix A ∈ G L ( 3 , R ) A \in\mathrm{GL}(3,\mathbb{R}) A ∈ GL ( 3 , R ) determines a smooth map
ϕ A : S 2 ⟶ S 2 , ϕ A ( v ) : = A v ∥ A v ∥ . \phi_A:S^2 \longrightarrow S^2,\qquad\phi_A(v):=\frac{Av}{\lVert Av\rVert}. ϕ A : S 2 ⟶ S 2 , ϕ A ( v ) := ∥ A v ∥ A v .
Since ϕ A ∘ ϕ B = ϕ A B \phi_A \circ\phi_B=\phi_{AB} ϕ A ∘ ϕ B = ϕ A B for all A , B ∈ G L ( 3 , R ) A,B \in\mathrm{GL}(3,\mathbb{R}) A , B ∈ GL ( 3 , R ) and ϕ I = id S 2 \phi_I=\operatorname{id}_{S^2} ϕ I = id S 2 , we obtain a group homomorphism
G L ( 3 , R ) ⟶ Diff ( S 2 ) , A ⟼ ϕ A . \mathrm{GL}(3,\mathbb{R}) \longrightarrow\operatorname{Diff}(S^2),\qquad A\longmapsto\phi_A. GL ( 3 , R ) ⟶ Diff ( S 2 ) , A ⟼ ϕ A .
The kernel of this homomorphism is { t I : t > 0 } \{tI:t>0\} { t I : t > 0 } . Since
Γ ∩ { t I : t > 0 } = { I } , \Gamma\cap\{tI:t>0\}=\{I\}, Γ ∩ { t I : t > 0 } = { I } ,
the restriction of this homomorphism to Γ \Gamma Γ is injective. We henceforth identify Γ \Gamma Γ with its image in Diff ( S 2 ) \operatorname{Diff}(S^2) Diff ( S 2 ) .
Step one. We show that Γ \Gamma Γ is not ( 1 , 2 ) (1,2) ( 1 , 2 ) -th order μ \mu μ -discrete for any finite Borel measure μ \mu μ on S 2 S^2 S 2 .
Set
p = ( 1 , 0 , 0 ) ∈ S 2 . p=(1,0,0)\in S^2. p = ( 1 , 0 , 0 ) ∈ S 2 .
Every element of Γ \Gamma Γ fixes both p p p and − p -p − p . Consider the open hemisphere
U + : = { ( x 1 , x 2 , x 3 ) ∈ S 2 : x 1 > 0 } U_+:=\{(x_1,x_2,x_3)\in S^2:x_1>0\} U + := {( x 1 , x 2 , x 3 ) ∈ S 2 : x 1 > 0 }
and the coordinate parametrization
κ : R 2 ⟶ U + , κ ( y , z ) = ( 1 , y , z ) 1 + y 2 + z 2 . \kappa:\mathbb{R}^2\longrightarrow U_+,\qquad\kappa(y,z)=\frac{(1,y,z)}{\sqrt{1+y^2+z^2}}. κ : R 2 ⟶ U + , κ ( y , z ) = 1 + y 2 + z 2 ( 1 , y , z ) .
For any g ( a , b , c ) ∈ Γ g_{(a,b,c)}\in\Gamma g ( a , b , c ) ∈ Γ , the local expression of ϕ g ( a , b , c ) \phi_{g_{(a,b,c)}} ϕ g ( a , b , c ) near p p p is
Φ a , b , c : = κ − 1 ∘ ϕ g ( a , b , c ) ∘ κ , \Phi_{a,b,c}:=\kappa^{-1}\circ\phi_{g_{(a,b,c)}}\circ\kappa, Φ a , b , c := κ − 1 ∘ ϕ g ( a , b , c ) ∘ κ ,
where
Φ a , b , c ( y , z ) = ( y + b z 1 + a y + c z , z 1 + a y + c z ) . \Phi_{a,b,c}(y,z)=\left(\frac{y+bz}{1+ay+cz},\frac{z}{1+ay+cz}\right). Φ a , b , c ( y , z ) = ( 1 + a y + cz y + b z , 1 + a y + cz z ) .
Since
D 0 Φ a , b , c = ( 1 b 0 1 ) , D_0\Phi_{a,b,c}=
\begin{pmatrix}
1 & b\\
0 & 1
\end{pmatrix}, D 0 Φ a , b , c = ( 1 0 b 1 ) ,
it follows that the derivative of ϕ g ( a , b , c ) \phi_{g_{(a,b,c)}} ϕ g ( a , b , c ) at p p p detects only the parameter b b b . In particular, since X a = g ( a , 0 , 0 ) X^a=g_{(a,0,0)} X a = g ( a , 0 , 0 ) ,
D p ( 1 ) ϕ X a = id T p ( 1 ) S 2 , a ∈ Z . D_p^{(1)}\phi_{X^a}=\operatorname{id}_{T_p^{(1)}S^2},\qquad a\in\mathbb{Z}. D p ( 1 ) ϕ X a = id T p ( 1 ) S 2 , a ∈ Z .
Using the parametrization − κ -\kappa − κ near − p -p − p gives the same local expression, and hence
D − p ( 1 ) ϕ X a = id T − p ( 1 ) S 2 , a ∈ Z . D_{-p}^{(1)}\phi_{X^a}=\operatorname{id}_{T_{-p}^{(1)}S^2},\qquad a\in\mathbb{Z}. D − p ( 1 ) ϕ X a = id T − p ( 1 ) S 2 , a ∈ Z .
Fix a finite Borel measure μ \mu μ on S 2 S^2 S 2 and a continuous section
g 1 : S 2 ⟶ Riem ( 1 ) ( S 2 ) . g_1:S^2\longrightarrow\operatorname{Riem}^{(1)}(S^2). g 1 : S 2 ⟶ Riem ( 1 ) ( S 2 ) .
For every a ∈ Z a\in\mathbb{Z} a ∈ Z , define
h a ( x ) = log κ 1 , g 1 ( X a , 1 Γ , x ) . h_a(x)=\log\kappa_{1,g_1}(X^a,1_\Gamma,x). h a ( x ) = log κ 1 , g 1 ( X a , 1 Γ , x ) .
The function h a h_a h a is continuous and bounded on S 2 S^2 S 2 . Since X a X^a X a fixes p p p and − p -p − p , and its derivatives at these points are the identity, we have
h a ( p ) = h a ( − p ) = 0. h_a(p)=h_a(-p)=0. h a ( p ) = h a ( − p ) = 0.
For each integer m ≥ 1 m \ge1 m ≥ 1 , set
τ m = g ( m , m , m 3 ) . \tau_m = g_{(m,m,m^3)}. τ m = g ( m , m , m 3 ) .
If x = ( x 1 , x 2 , x 3 ) ∈ S 2 x=(x_1,x_2,x_3)\in S^2 x = ( x 1 , x 2 , x 3 ) ∈ S 2 , then
ϕ τ m ( x ) = ( x 1 + m x 2 + m 3 x 3 , x 2 + m x 3 , x 3 ) ( x 1 + m x 2 + m 3 x 3 ) 2 + ( x 2 + m x 3 ) 2 + x 3 2 . \phi_{\tau_m}(x)=\frac{(x_1+mx_2+m^3x_3,x_2+mx_3,x_3)}{\sqrt{(x_1+mx_2+m^3x_3)^2+(x_2+mx_3)^2+x_3^2}}. ϕ τ m ( x ) = ( x 1 + m x 2 + m 3 x 3 ) 2 + ( x 2 + m x 3 ) 2 + x 3 2 ( x 1 + m x 2 + m 3 x 3 , x 2 + m x 3 , x 3 ) .
It follows that
ϕ τ m ( x ) ⟶ { sgn ( x 3 ) p , x 3 ≠ 0 , sgn ( x 2 ) p , x 3 = 0 , x 2 ≠ 0 , sgn ( x 1 ) p , x 2 = x 3 = 0 , as m → ∞ . \phi_{\tau_m}(x)\longrightarrow
\begin{cases}
\operatorname{sgn}(x_3)p, & x_3\ne0,\\
\operatorname{sgn}(x_2)p, & x_3=0,\ x_2\ne0,\\
\operatorname{sgn}(x_1)p, & x_2=x_3=0,
\end{cases}
\qquad\text{as }m\to\infty. ϕ τ m ( x ) ⟶ ⎩ ⎨ ⎧ sgn ( x 3 ) p , sgn ( x 2 ) p , sgn ( x 1 ) p , x 3 = 0 , x 3 = 0 , x 2 = 0 , x 2 = x 3 = 0 , as m → ∞.
In every case, the limit belongs to { p , − p } \{p,-p\} { p , − p } . Consequently, for each a ∈ Z a\in\mathbb{Z} a ∈ Z and x ∈ S 2 x\in S^2 x ∈ S 2 ,
h a ( ϕ τ m ( x ) ) ⟶ 0 as m → ∞ . h_a\left(\phi_{\tau_m}(x)\right)\longrightarrow0
\qquad\text{as }m\to\infty. h a ( ϕ τ m ( x ) ) ⟶ 0 as m → ∞.
Since h a h_a h a is bounded and μ \mu μ is finite, the dominated convergence theorem gives
d ( τ m ) ∗ μ , g 1 ( 1 , 2 ) ( X a , 1 Γ ) 2 = ∫ S 2 h a ( x ) 2 d ( ( τ m ) ∗ μ ) ( x ) = ∫ S 2 h a ( ϕ τ m ( x ) ) 2 d μ ( x ) ⟶ 0 , as m → ∞ . \begin{aligned}
d^{(1,2)}_{(\tau_m)_*\mu,g_1}(X^a,1_\Gamma)^2
=\int_{S^2}h_a(x)^2\,d((\tau_m)_*\mu)(x) \\
&=\int_{S^2}h_a\left(\phi_{\tau_m}(x)\right)^2\,d\mu(x)\longrightarrow0,
\qquad\text{as }m\to\infty.
\end{aligned} d ( τ m ) ∗ μ , g 1 ( 1 , 2 ) ( X a , 1 Γ ) 2 = ∫ S 2 h a ( x ) 2 d (( τ m ) ∗ μ ) ( x ) = ∫ S 2 h a ( ϕ τ m ( x ) ) 2 d μ ( x ) ⟶ 0 , as m → ∞.
Thus
inf τ ∈ Γ d τ ∗ μ , g 1 ( 1 , 2 ) ( X a , 1 Γ ) = 0 for every a ∈ Z . \inf_{\tau\in\Gamma}d^{(1,2)}_{\tau_*\mu,g_1}(X^a,1_\Gamma)=0
\qquad\text{for every }a\in\mathbb{Z}. τ ∈ Γ inf d τ ∗ μ , g 1 ( 1 , 2 ) ( X a , 1 Γ ) = 0 for every a ∈ Z .
Since X a → ∞ X^a\to\infty X a → ∞ in Γ \Gamma Γ as a → + ∞ a\to+\infty a → + ∞ , Γ \Gamma Γ is not ( 1 , 2 ) (1,2) ( 1 , 2 ) -th order μ \mu μ -discrete.
Step two. We show that Γ \Gamma Γ is ( 2 , 2 ) (2,2) ( 2 , 2 ) -th order μ 0 \mu_0 μ 0 -discrete for a Borel probability measure μ 0 \mu_0 μ 0 on S 2 S^2 S 2 .
Since every γ ∈ Γ \gamma\in\Gamma γ ∈ Γ fixes p p p , the second-order derivative D p ( 2 ) ϕ γ D_p^{(2)}\phi_\gamma D p ( 2 ) ϕ γ is an automorphism of T p ( 2 ) S 2 T_p^{(2)}S^2 T p ( 2 ) S 2 , i.e. D p ( 2 ) ϕ γ ∈ GL ( T p ( 2 ) S 2 ) D_p^{(2)}\phi_\gamma\in\operatorname{GL}(T_p^{(2)}S^2) D p ( 2 ) ϕ γ ∈ GL ( T p ( 2 ) S 2 ) .
We next show that, for a given element g ( a , b , c ) ∈ Γ g_{(a,b,c)}\in\Gamma g ( a , b , c ) ∈ Γ , the second-order derivative of ϕ g ( a , b , c ) \phi_{g_{(a,b,c)}} ϕ g ( a , b , c ) at p p p detects all three parameters a , b , c a,b,c a , b , c , whereas the first-order derivative at p p p detects only b b b .
With respect to the local coordinates ( y , z ) (y,z) ( y , z ) determined by κ \kappa κ , let
E : = ( e y , e z , e y y , e y z , e z z ) \mathcal{E}:=(e_y,e_z,e_{yy},e_{yz},e_{zz}) E := ( e y , e z , e y y , e y z , e z z )
be the ordered basis of
T p ( 2 ) S 2 = ( J p 2 ( S 2 , R ) 0 ) ∗ , T_p^{(2)}S^2=\left(J_p^2(S^2,\mathbb{R})_0\right)^*, T p ( 2 ) S 2 = ( J p 2 ( S 2 , R ) 0 ) ∗ ,
where the linear functionals in E \mathcal{E} E are defined by
e y ( j p 2 f ) = ∂ ( f ∘ κ ) ∂ y ( 0 , 0 ) , e_y(j_p^2f)=\frac{\partial(f\circ\kappa)}{\partial y}(0,0), e y ( j p 2 f ) = ∂ y ∂ ( f ∘ κ ) ( 0 , 0 ) ,
e z ( j p 2 f ) = ∂ ( f ∘ κ ) ∂ z ( 0 , 0 ) , e_z(j_p^2f)=\frac{\partial(f\circ\kappa)}{\partial z}(0,0), e z ( j p 2 f ) = ∂ z ∂ ( f ∘ κ ) ( 0 , 0 ) ,
e y y ( j p 2 f ) = ∂ 2 ( f ∘ κ ) ∂ y 2 ( 0 , 0 ) , e_{yy}(j_p^2f)=\frac{\partial^2(f\circ\kappa)}{\partial y^2}(0,0), e y y ( j p 2 f ) = ∂ y 2 ∂ 2 ( f ∘ κ ) ( 0 , 0 ) ,
e y z ( j p 2 f ) = ∂ 2 ( f ∘ κ ) ∂ y ∂ z ( 0 , 0 ) , e_{yz}(j_p^2f)=\frac{\partial^2(f\circ\kappa)}{\partial y\,\partial z}(0,0), e y z ( j p 2 f ) = ∂ y ∂ z ∂ 2 ( f ∘ κ ) ( 0 , 0 ) ,
e z z ( j p 2 f ) = ∂ 2 ( f ∘ κ ) ∂ z 2 ( 0 , 0 ) . e_{zz}(j_p^2f)=\frac{\partial^2(f\circ\kappa)}{\partial z^2}(0,0). e z z ( j p 2 f ) = ∂ z 2 ∂ 2 ( f ∘ κ ) ( 0 , 0 ) .
for every smooth function f f f defined near p p p with f ( p ) = 0 f(p)=0 f ( p ) = 0 . We denote by
[ L ] E [L]_{\mathcal{E}} [ L ] E
the 5 × 5 5 \times5 5 × 5 matrix of L ∈ GL ( T p ( 2 ) S 2 ) L \in\operatorname{GL}(T_p^{(2)}S^2) L ∈ GL ( T p ( 2 ) S 2 ) with respect to the ordered basis E \mathcal{E} E . By the definition of second-order derivatives, a routine application of the chain rule yields
D p ( 2 ) φ g ( a , b , c ) ( e y ) = e y , D_p^{(2)}\varphi_{g_{(a,b,c)}}(e_y)=e_y, D p ( 2 ) φ g ( a , b , c ) ( e y ) = e y ,
D p ( 2 ) φ g ( a , b , c ) ( e z ) = b e y + e z , D_p^{(2)}\varphi_{g_{(a,b,c)}}(e_z)=be_y+e_z, D p ( 2 ) φ g ( a , b , c ) ( e z ) = b e y + e z ,
D p ( 2 ) φ g ( a , b , c ) ( e y y ) = − 2 a e y + e y y , D_p^{(2)}\varphi_{g_{(a,b,c)}}(e_{yy})=-2ae_y+e_{yy}, D p ( 2 ) φ g ( a , b , c ) ( e y y ) = − 2 a e y + e y y ,
D p ( 2 ) φ g ( a , b , c ) ( e y z ) = − ( c + a b ) e y − a e z + b e y y + e y z , D_p^{(2)}\varphi_{g_{(a,b,c)}}(e_{yz})=-(c+ab)e_y-ae_z+be_{yy}+e_{yz}, D p ( 2 ) φ g ( a , b , c ) ( e y z ) = − ( c + ab ) e y − a e z + b e y y + e y z ,
D p ( 2 ) φ g ( a , b , c ) ( e z z ) = − 2 b c e y − 2 c e z + b 2 e y y + 2 b e y z + e z z . D_p^{(2)}\varphi_{g_{(a,b,c)}}(e_{zz})=-2bce_y-2ce_z+b^2e_{yy}+2be_{yz}+e_{zz}. D p ( 2 ) φ g ( a , b , c ) ( e z z ) = − 2 b c e y − 2 c e z + b 2 e y y + 2 b e y z + e z z .
Therefore
[ D p ( 2 ) φ g ( a , b , c ) ] E = ( 1 b − 2 a − ( c + a b ) − 2 b c 0 1 0 − a − 2 c 0 0 1 b b 2 0 0 0 1 2 b 0 0 0 0 1 ) . \left[D_p^{(2)}\varphi_{g_{(a,b,c)}}\right]_{\mathcal{E}}=
\begin{pmatrix}
1 & b & -2a & -(c+ab) & -2bc\\
0 & 1 & 0 & -a & -2c\\
0 & 0 & 1 & b & b^2\\
0 & 0 & 0 & 1 & 2b\\
0 & 0 & 0 & 0 & 1
\end{pmatrix}. [ D p ( 2 ) φ g ( a , b , c ) ] E = 1 0 0 0 0 b 1 0 0 0 − 2 a 0 1 0 0 − ( c + ab ) − a b 1 0 − 2 b c − 2 c b 2 2 b 1 .
Thus,
a = − ( [ D p ( 2 ) φ g ( a , b , c ) ] E ) 2 , 4 , b = ( [ D p ( 2 ) φ g ( a , b , c ) ] E ) 1 , 2 , c = − 1 2 ( [ D p ( 2 ) φ g ( a , b , c ) ] E ) 2 , 5 . (10) \begin{aligned}
a&=-\left(\left[D_p^{(2)}\varphi_{g_{(a,b,c)}}\right]_{\mathcal{E}}\right)_{2,4},\\
b&=\left(\left[D_p^{(2)}\varphi_{g_{(a,b,c)}}\right]_{\mathcal{E}}\right)_{1,2},\\
c&=-\frac{1}{2}\left(\left[D_p^{(2)}\varphi_{g_{(a,b,c)}}\right]_{\mathcal{E}}\right)_{2,5}. \tag*{(10)}
\end{aligned} a b c = − ( [ D p ( 2 ) φ g ( a , b , c ) ] E ) 2 , 4 , = ( [ D p ( 2 ) φ g ( a , b , c ) ] E ) 1 , 2 , = − 2 1 ( [ D p ( 2 ) φ g ( a , b , c ) ] E ) 2 , 5 . (10)
Now, let μ 0 \mu_0 μ 0 be the Dirac probability measure concentrated at p p p , i.e., for every Borel set E ⊆ S 2 E \subseteq S^2 E ⊆ S 2 ,
μ 0 ( E ) = { 1 , p ∈ E , 0 , p ∉ E . \mu_0(E)=
\begin{cases}
1, & p\in E,\\
0, & p\notin E.
\end{cases} μ 0 ( E ) = { 1 , 0 , p ∈ E , p ∈ / E .
Fix any continuous section g 2 : S 2 → Riem ( 2 ) ( S 2 ) g_2:S^2\to\operatorname{Riem}^{(2)}(S^2) g 2 : S 2 → Riem ( 2 ) ( S 2 ) .
Since E \mathcal{E} E and g 2 , p g_{2,p} g 2 , p are fixed, there exists a constant C ≥ 1 C\geq1 C ≥ 1 such that
∣ ( [ L ] E ) i , j ∣ ≤ C ∥ L ∥ g 2 , 1 ≤ i , j ≤ 5 , \left|\left([L]_{\mathcal{E}}\right)_{i,j}\right|\leq C\lVert L\rVert_{g_2},\qquad1\leq i,j\leq5, ( [ L ] E ) i , j ≤ C ∥ L ∥ g 2 , 1 ≤ i , j ≤ 5 ,
for every L ∈ GL ( T p ( 2 ) S 2 ) L\in\operatorname{GL}(T_p^{(2)}S^2) L ∈ GL ( T p ( 2 ) S 2 ) . Applying this estimate to the expressions (10) for a a a , b b b , c c c gives
∣ a ∣ ≤ C ∥ D p ( 2 ) φ g ( a , b , c ) ∥ g 2 , |a|\leq C\lVert D_p^{(2)}\varphi_{g_{(a,b,c)}}\rVert_{g_2}, ∣ a ∣ ≤ C ∥ D p ( 2 ) φ g ( a , b , c ) ∥ g 2 ,
∣ b ∣ ≤ C ∥ D p ( 2 ) φ g ( a , b , c ) ∥ g 2 , |b|\leq C\lVert D_p^{(2)}\varphi_{g_{(a,b,c)}}\rVert_{g_2}, ∣ b ∣ ≤ C ∥ D p ( 2 ) φ g ( a , b , c ) ∥ g 2 ,
∣ c ∣ ≤ C 2 ∥ D p ( 2 ) φ g ( a , b , c ) ∥ g 2 . |c|\leq\frac{C}{2}\lVert D_p^{(2)}\varphi_{g_{(a,b,c)}}\rVert_{g_2}. ∣ c ∣ ≤ 2 C ∥ D p ( 2 ) φ g ( a , b , c ) ∥ g 2 .
Consequently,
∥ D p ( 2 ) φ g ( a , b , c ) ∥ g 2 ≥ C − 1 max { ∣ a ∣ , ∣ b ∣ , ∣ c ∣ } . \lVert D_p^{(2)}\varphi_{g_{(a,b,c)}}\rVert_{g_2}\geq C^{-1}\max\{|a|,|b|,|c|\}. ∥ D p ( 2 ) φ g ( a , b , c ) ∥ g 2 ≥ C − 1 max { ∣ a ∣ , ∣ b ∣ , ∣ c ∣ } .
For every N > 0 N>0 N > 0 , the set
{ g ( a , b , c ) ∈ Γ : ∣ a ∣ ≤ N , ∣ b ∣ ≤ N , ∣ c ∣ ≤ N } \{g_{(a,b,c)}\in\Gamma:|a|\leq N,\ |b|\leq N,\ |c|\leq N\} { g ( a , b , c ) ∈ Γ : ∣ a ∣ ≤ N , ∣ b ∣ ≤ N , ∣ c ∣ ≤ N }
is finite. Hence
max { ∣ a ∣ , ∣ b ∣ , ∣ c ∣ } ⟶ ∞ as g ( a , b , c ) ⟶ ∞ in Γ . \max\{|a|,|b|,|c|\}\longrightarrow\infty\qquad\text{as }g_{(a,b,c)}\longrightarrow\infty\text{ in }\Gamma. max { ∣ a ∣ , ∣ b ∣ , ∣ c ∣ } ⟶ ∞ as g ( a , b , c ) ⟶ ∞ in Γ.
It follows that
∥ D p ( 2 ) φ γ ∥ g 2 ⟶ ∞ as γ ⟶ ∞ in Γ . \lVert D_p^{(2)}\varphi_\gamma\rVert_{g_2}\longrightarrow\infty\qquad\text{as }\gamma\longrightarrow\infty\text{ in }\Gamma. ∥ D p ( 2 ) φ γ ∥ g 2 ⟶ ∞ as γ ⟶ ∞ in Γ.
Since every τ ∈ Γ \tau\in\Gamma τ ∈ Γ fixes p p p , we have
τ ∗ μ 0 = μ 0 . \tau^*\mu_0=\mu_0. τ ∗ μ 0 = μ 0 .
Thus, we obtain
inf τ ∈ Γ d τ ∗ μ 0 , g 2 ( 2 , 2 ) ( γ , 1 Γ ) = d μ 0 , g 2 ( 2 , 2 ) ( γ , 1 Γ ) = log κ 2 , g 2 ( γ , 1 Γ , p ) = log max { ∥ D p ( 2 ) ϕ γ ∥ g 2 , ∥ D p ( 2 ) ϕ γ − 1 ∥ g 2 } ≥ log ∥ D p ( 2 ) ϕ γ ∥ g 2 ⟶ ∞ as γ → ∞ . \begin{aligned}
\inf_{\tau\in\Gamma}d_{\tau^*\mu_0,g_2}^{(2,2)}(\gamma,1_\Gamma)
&=d_{\mu_0,g_2}^{(2,2)}(\gamma,1_\Gamma)\\
&=\log\kappa_{2,g_2}(\gamma,1_\Gamma,p)\\
&=\log\max\left\{\left\|D_p^{(2)}\phi_\gamma\right\|_{g_2},\left\|D_p^{(2)}\phi_{\gamma^{-1}}\right\|_{g_2}\right\}\\
&\geq\log\left\|D_p^{(2)}\phi_\gamma\right\|_{g_2}\longrightarrow\infty\quad\text{as }\gamma\to\infty.
\end{aligned} τ ∈ Γ inf d τ ∗ μ 0 , g 2 ( 2 , 2 ) ( γ , 1 Γ ) = d μ 0 , g 2 ( 2 , 2 ) ( γ , 1 Γ ) = log κ 2 , g 2 ( γ , 1 Γ , p ) = log max { D p ( 2 ) ϕ γ g 2 , D p ( 2 ) ϕ γ − 1 g 2 } ≥ log D p ( 2 ) ϕ γ g 2 ⟶ ∞ as γ → ∞.
Therefore, Γ \Gamma Γ is ( 2 , 2 ) (2,2) ( 2 , 2 ) -th order μ 0 \mu_0 μ 0 -discrete. □
Remark 4.2. The same method also yields, for each n ≥ 3 n\geq3 n ≥ 3 , a Property ( T ) (T) ( T ) subgroup Γ ≅ Z n ⋊ SL ( n , Z ) \Gamma\cong\mathbb{Z}^n\rtimes\operatorname{SL}(n,\mathbb{Z}) Γ ≅ Z n ⋊ SL ( n , Z ) of Diff ( S n ) \operatorname{Diff}(S^n) Diff ( S n ) that is ( 2 , 2 ) (2,2) ( 2 , 2 ) -th order μ 0 \mu_0 μ 0 -discrete for some Borel probability measure μ 0 \mu_0 μ 0 , but not ( 1 , 2 ) (1,2) ( 1 , 2 ) -th order μ \mu μ -discrete for any finite Borel measure μ \mu μ on S n S^n S n . These examples show that higher-order μ \mu μ -discreteness detects a larger class of subgroups of Diff ( M ) \operatorname{Diff}(M) Diff ( M ) than its lower-order counterpart.
References
Department of Mathematics, Center of Noncommutative Geometry, Shanghai University of Finance and Economics
Email address: jintaodeng@sufe.edu.cn
Department of Mathematics, Liaoning University, Liaoning, China 110036
Email address: gengtian.ncg@gmail.com