Introduction
A Ricci flow is a family of Riemannian metrics g ( t ) g(t) g ( t ) on a smooth manifold that solves
∂ g ∂ t = − 2 ric g , g ( 0 ) = g 0 . (1) \frac{\partial g}{\partial t}=-2\operatorname{ric}_{g},\qquad g(0)=g_{0}. \tag*{(1)} ∂ t ∂ g = − 2 ric g , g ( 0 ) = g 0 . (1)
A central problem is to determine whether the Ricci flow encounters a finite-time singularity or admits a long-time solution. Solutions to (1) that exist for all t > 0 t>0 t > 0 are called immortal Ricci flows. Key to the study of immortal Ricci flows is understanding their long-time asymptotics. In certain cases, immortal Ricci flows converge to expanding Ricci solitons, self-similar solutions to the Ricci flow that evolve only by homothetic expansion and pullback by diffeomorphisms [23 , 24 , 4 BL18, 25 ] .
The long-time behaviour of immortal Ricci flows on closed 3-manifolds admitting locally homogeneous metrics modelled on R 3 \mathbb{R}^{3} R 3 , Nil, and Sol is now characterised. Up to finite covers, these manifolds are torus bundles over S 1 S^{1} S 1 with local T 2 \mathbb{T}^{2} T 2 -actions related by group automorphisms prescribed by the monodromy. For each of the three geometries, the locally homogeneous Ricci flows on these manifolds were described in [16 ] . Lott–Šešum prove that if the initial metric is invariant under the local T 2 \mathbb{T}^{2} T 2 -actions, then its Ricci flow is immortal and satisfies type III curvature estimates [26 ] . Recent work by Lott shows that, for arbitrary immortal Ricci flows on these manifolds without symmetry, the blowdowns of the metrics lifted to the universal cover converge to homogeneous expanding Ricci solitons [25 ] .
These torus bundles belong to a broader class of manifolds with local symmetries, called twisted principal bundles : fibre bundles that have Lie group fibres and transition functions consisting of smoothly varying left translations and locally constant group automorphisms. The significance of this class of manifolds is illustrated by Lafuente–Thompson’s recent constructions of expanding Ricci solitons with constant scalar curvature. Starting in dimension four, their constructions yield many novel examples that are not locally homogeneous [27 ] . In a wider context, this type of local symmetry plays a key role in the study of Ricci flows that collapse with bounded curvature [24 , 7 ] .
Our first main theorem establishes long-time existence, together with type III estimates, for Ricci flows of invariant metrics on twisted principal bundles with nilpotent fibres over S 1 S^{1} S 1 in arbitrary dimension.
Theorem A. Let N N N be a nilpotent Lie group . Ricci flows of invariant metrics g ( t ) g(t) g ( t ) on twisted principal N N N -bundles over S 1 S^1 S 1 are immortal and satisfy the type III estimates
∣ Rm g ( t ) ∣ g ( t ) ≤ C t , diam ( S 1 , g ˉ ( t ) ) ≤ C 0 1 + t , \lvert\operatorname{Rm}_{g(t)}\rvert_{g(t)} \le\frac{C}{t}, \qquad\operatorname{diam}(S^1,\bar{g}(t)) \le C_0\sqrt{1+t}, ∣ Rm g ( t ) ∣ g ( t ) ≤ t C , diam ( S 1 , g ˉ ( t )) ≤ C 0 1 + t , where g ˉ ( t ) \bar{g}(t) g ˉ ( t ) is the quotient metric. Here C > 0 C>0 C > 0 is a constant depending only on dimension, and C 0 > 0 C_0>0 C 0 > 0 is a constant depending only on the initial metric.
Theorem A extends the long-time existence results and curvature estimates proved by Lauret for Ricci flows of left-invariant metrics on nilpotent Lie groups [19 ] and by Gindi–Streets for Ricci flows of invariant metrics on four-dimensional nilpotent principal bundles over S 1 S^1 S 1 [11 ] . Lauret’s proof relied in part on a technique called bracket flow , an equivalent formulation of homogeneous Ricci flow in which the Lie bracket evolves while the inner product on the Lie algebra remains fixed. The bracket flow was first developed for nilpotent Lie groups by Guzhvina [12 ] and later by Lauret for simply-connected nilmanifolds and homogeneous spaces [19 , 21 ] . Subsequent applications include Lafuente’s scalar curvature blowup criterion for homogeneous Ricci flows [18 ] and Böhm–Lafuente’s proof that every sequence of blowdowns of an immortal homogeneous Ricci flow subconverges to a homogeneous expanding Ricci soliton [BL18], with further applications to other geometric flows [22 , 30 ] .
To prove Theorem A, we extend the homogeneous bracket flow to twisted principal bundles. For simplicity, let us explain our construction in the particular case of Ricci flows ( M , g ) (M,g) ( M , g ) that are invariant under a free and proper isometric action of a connected Lie group G G G with arbitrary cohomogeneity . The G G G -invariant vector fields correspond to sections of a vector bundle T M / G → M / G TM/G \to M/G T M / G → M / G . This bundle is endowed with a Lie bracket on sections satisfying some natural axioms, which in this particular case is simply the bracket of G G G -invariant vector fields on M M M . Thus, T M / G TM/G T M / G is a Lie algebroid .
Sections of the adjoint bundle , the Lie algebra bundle associated to M → M / G M \to M/G M → M / G by the adjoint representation ad M = M × Ad g \operatorname{ad}M = M \times_{\operatorname{Ad}} \mathfrak{g} ad M = M × Ad g , correspond to invariant vertical vector fields. The differential of the submersion M → M / G M \to M/G M → M / G induces the short exact sequence of vector bundles,
0 → ad M → T M / G → T ( M / G ) → 0. (2) 0 \to\operatorname{ad}M \to TM/G \to T(M/G) \to0. \tag*{(2)} 0 → ad M → T M / G → T ( M / G ) → 0. (2)
Splittings of this sequence (2) correspond to principal connections on M → M / G M \to M/G M → M / G . The G G G -invariant metrics on M M M are in one-to-one correspondence with bundle metrics on T M / G TM/G T M / G , which are determined by a bundle metric on ad M \operatorname{ad}M ad M , a Riemannian metric on M / G M/G M / G , and a principal connection. The main idea for the higher-cohomogeneity bracket flow is to evolve the Lie algebroid bracket of T M / G TM/G T M / G , together with the Riemannian metric on M / G M/G M / G , and the connection, while keeping the bundle metric on ad M \operatorname{ad}M ad M fixed. When M / G M/G M / G is a point, T M / G TM/G T M / G is the Lie algebra g \mathfrak{g} g and we recover Lauret’s homogeneous bracket flow.
Our next theorem establishes equivalence between the Ricci flow of invariant metrics and the bracket flow (see Theorem 3.4 for a more precise statement).
Theorem B. On twisted principal bundles, the Ricci flow of invariant metrics and the bracket flow are equivalent as deformations of Riemannian Lie algebroids.
A key ingredient in the proof is Uhlenbeck’s trick of moving frames. Given an invariant Ricci flow, take a moving frame of invariant vertical vector fields that remains orthonormal along the flow. We then pull back the fixed Lie algebroid bracket by this change of frame, to obtain a curve of Lie algebroid brackets. Given either the Ricci flow or the bracket flow, the other is obtained by solving for a time-dependent change of frame determined by a fibrewise ordinary differential equation. Consequently, both flows have the same maximal time of existence, and curvature estimates obtained for the bracket flow hold under the Ricci flow.
To understand the convergence and long-time behaviour of the Ricci flows in Theorem A, we consider the normalised Ricci flow:
∂ g ∂ t = − 2 ric g − g , g ( 0 ) = g 0 . (3) \frac{\partial g}{\partial t}=-2\operatorname{ric}_{g}-g,\qquad g(0)=g_0. \tag*{(3)} ∂ t ∂ g = − 2 ric g − g , g ( 0 ) = g 0 . (3)
The two flows are related in the following sense: if g ( t ) g(t) g ( t ) is a normalised Ricci flow (3) , then g ˉ ( s ) = s g ( log s ) \bar{g}(s)=s g(\log s) g ˉ ( s ) = s g ( log s ) is a solution to the unnormalised Ricci flow (1) . Hence, immortal Ricci flows defined for all s > 0 s>0 s > 0 correspond to eternal normalised Ricci flows, that is, solutions to (3) that exist for all times. In this correspondence, g ( t ) = e − t g ˉ ( e t ) g(t)=e^{-t}\bar{g}(e^t) g ( t ) = e − t g ˉ ( e t ) and g 0 = g ˉ ( 1 ) g_0=\bar{g}(1) g 0 = g ˉ ( 1 ) . The type III estimates of Theorem A imply that the corresponding normalised Ricci flows have uniformly bounded Riemannian curvature and quotient diameter for all positive time. For this reason, we study the time translates g k ( t ) : = g ( t + t k ) g_k(t):=g(t+t_k) g k ( t ) := g ( t + t k ) of the normalised Ricci flow. If s k = e t k s_k=e^{t_k} s k = e t k , then s k − 1 g ˉ ( s k s ) = s g k ( log s ) s_k^{-1}\bar{g}(s_ks)=sg_k(\log s) s k − 1 g ˉ ( s k s ) = s g k ( log s ) , so the time translates describe the blowdowns of the unnormalised Ricci flow. Indeed, if g k ( t ) g_k(t) g k ( t ) converges to g ∞ ( t ) g_\infty(t) g ∞ ( t ) , then the blowdowns converge to s g ∞ ( log s ) sg_\infty(\log s) s g ∞ ( log s ) .
Gindi derives rigidity results for blowdown limits of the invariant Ricci flow on nilpotent principal bundles, assuming integrability of the horizontal distribution along the flow [10 Thm. 8.1] . For Heisenberg fibres, the limits are locally expanding Ricci solitons [10 Thm. 10.3] . For one-dimensional base, the horizontal distribution always remains integrable along the flow.
Our final theorem demonstrates that the normalised Ricci flow of invariant metrics on twisted principal bundles with nilpotent fibres over S 1 S^1 S 1 , once lifted to the universal cover, subconverges to a solvsoliton.
Theorem C. Let N N N be a simply-connected nilpotent Lie group. Let M → S 1 M\to S^1 M → S 1 be a twisted principal N N N -bundle. Let g ( t ) g(t) g ( t ) be a normalised Ricci flow of invariant metrics on M M M . Let g ~ ( t ) \widetilde{g}(t) g ( t ) be the lift of g ( t ) g(t) g ( t ) to the universal cover M ~ \widetilde{M} M . Fix a point p ~ ∈ M ~ \widetilde{p}\in\widetilde{M} p ∈ M . Then, there exists a sequence t k → ∞ t_k\to\infty t k → ∞ and a simply-connected solvable Lie group S S S equipped with a solvsoliton metric g ∞ g_\infty g ∞ such that
( M ~ , g ~ ( t k ) , p ~ ) ⟶ ( S , g ∞ , e ) (\widetilde{M},\widetilde{g}(t_k),\widetilde{p})\longrightarrow(S,g_\infty,e) ( M , g ( t k ) , p ) ⟶ ( S , g ∞ , e )
in the smooth pointed Cheeger–Gromov topology.
Recall that a solvsoliton is an invariant expanding Ricci soliton on a simply-connected solvable Lie group that evolves under Ricci flow by homothetic expansion and pullback by Lie group automorphisms [20 ] . To prove Theorem C, we first prove a compactness theorem for normalised Ricci flows of invariant metrics on twisted principal bundles over S 1 S^1 S 1 with simply-connected nilpotent fibres. The compactness argument combines the type III estimates of Theorem A with Cheeger–Gromov–Taylor’s injectivity radius estimate [8 Thm. 4.7] , Hamilton’s compactness theorem [14 ] , and Heber’s compactness of Killing fields [15 p. 330] . The assumptions that the fibres are nilpotent and simply-connected provide the volume non-collapsing needed for the Cheeger–Gromov–Taylor injectivity radius estimate. If the diameter of the base collapses, then the limit of the universal covers is homogeneous. We appeal to [BL18, BL23] to obtain a solvsoliton limit. If the base does not collapse, then we can derive properties of the limit using Gindi–Streets monotone energy functional [11 Def. 4.1] :
I ( g ) = ∫ S 1 ∣ L ∣ 2 + 1 2 d s . (4) \mathcal{I}(g)=\int_{S^1}|L|^2+\frac{1}{2}\,ds. \tag*{(4)} I ( g ) = ∫ S 1 ∣ L ∣ 2 + 2 1 d s . (4)
After obtaining a limit Ricci flow, we take the corresponding bracket flow. The properties obtained from (4) reduce this bracket flow to a homogeneous bracket flow on the fibres, with a fixed connection and quotient metric. We integrate a monotone quantity for homogeneous normalised bracket flow [19 ] :
F ( μ , h , g ˉ ) = ∫ S 1 R ^ μ + ∣ Ric ^ μ ∣ 2 d s . (5) \mathcal{F}(\mu,h,\bar{g})=\int_{S^1}\widehat{R}_{\mu}+\left|\widehat{\operatorname{Ric}}_{\mu}\right|^2\,ds. \tag*{(5)} F ( μ , h , g ˉ ) = ∫ S 1 R μ + Ric μ 2 d s . (5)
The monotonicity of F \mathcal{F} F shows that a further subsequential limit has nilsoliton fibres, while retaining the properties established by (4). Together, these conditions imply that the limit metric is a solvsoliton, proving Theorem C.
For twisted principal bundles with unipotent monodromy, we use the limit properties from the proof of Theorem C to show collapse of the quotient S 1 S^1 S 1 . Arguing by contradiction, we assume that the quotient diameter remains bounded below along a sequence of times of the normalised Ricci flow. The corresponding limit has unipotent holonomy and a parallel non-zero shape operator. These properties are contradictory, proving the following corollary (see Theorem 5.6 for a more precise statement).
Corollary D. Let N N N be a simply-connected nilpotent Lie group acting with cohomogeneity one on M M M , with orbit space M / N ≅ S 1 M/N\cong S^1 M / N ≅ S 1 . If g ( t ) g(t) g ( t ) is a normalised Ricci flow of N N N -invariant metrics on M M M , then
lim t → ∞ diam ( M / N , g ˉ ( t ) ) = 0 , \lim_{t\to\infty}\operatorname{diam}(M/N,\bar{g}(t))=0, t → ∞ lim diam ( M / N , g ˉ ( t )) = 0 , where g ˉ ( t ) \bar{g}(t) g ˉ ( t ) is the quotient metric.
For principal N-bundles over S 1 S^1 S 1 , the collapse of the base given in Theorem D is incompatible with the convergence through principal bundle isomorphisms assumed in [10 Thm. 1.16, Thm. 11.1] . As observed in [10 §1.1.5] , these results can be extended to twisted principal bundles, where monodromy can prevent the base from collapsing.
Example 1.1. Lott–Šešum provide the following example [26 eq. 3.9] . Take M M M to be the twisted principal R 2 \mathbb{R}^2 R 2 -bundle over S 1 S^1 S 1 with hyperbolic monodromy given by
H = ( 2 1 1 1 ) . H = \begin{pmatrix} 2 & 1 \\ 1 & 1 \end{pmatrix}. H = ( 2 1 1 1 ) .
Let x + , x − ∈ R 2 x_+,x_- \in\mathbb{R}^2 x + , x − ∈ R 2 be eigenvectors of H H H with corresponding eigenvalues λ = 1 2 ( 3 + 5 ) \lambda= \frac{1}{2}(3+\sqrt{5}) λ = 2 1 ( 3 + 5 ) and λ − 1 = 1 2 ( 3 − 5 ) \lambda^{-1} = \frac{1}{2}(3-\sqrt{5}) λ − 1 = 2 1 ( 3 − 5 ) . Let c = log λ c = \log\lambda c = log λ . Then
g ( t ) = 4 c 2 d y 2 + e − t ( e 2 c y d x + 2 + e − 2 c y d x − 2 ) g(t) = 4c^2dy^2 + e^{-t}\left(e^{2c y}dx_+^2 + e^{-2c y}dx_-^2\right) g ( t ) = 4 c 2 d y 2 + e − t ( e 2 cy d x + 2 + e − 2 cy d x − 2 )
is an invariant normalised Ricci flow on M → S 1 M \to S^1 M → S 1 with a constant quotient metric g ˇ ( t ) = 4 c 2 d y 2 \check{g}(t) = 4c^2dy^2 g ˇ ( t ) = 4 c 2 d y 2 .
1.1. Acknowledgements. The author would like to thank his supervisors, Ramiro Lafuente and Kyle Broder, for their guidance, helpful discussions, and feedback on earlier versions of this work. The author gratefully acknowledges access to OpenAI’s ChatGPT for Academic Researchers program through a research workspace awarded to Kyle Broder. This work was supported by an Australian Government Research Training Program (RTP) Scholarship.
1.2. AI disclosure. No AI tools were used in the proofs of Theorem A or Theorem B. ChatGPT 5.6 Sol, 6 Astra, and 6.1 Sol assisted in the proofs of Theorem C and Theorem D. The author takes full responsibility for the mathematical content of this paper. No text was written by AI.
Preliminaries
Lauret’s bracket flow and Uhlenbeck’s trick
To begin, we recall Lauret’s bracket flow on Lie groups and demonstrate its equivalence to the Ricci flow [19 , 21 ] . Let G G G be a connected Lie group and let g ( t ) : t ∈ [ 0 , T ) g(t): t \in[0,T) g ( t ) : t ∈ [ 0 , T ) be a Ricci flow of left-invariant metrics with initial metric g ( 0 ) = g 0 g(0)=g_0 g ( 0 ) = g 0 .
By a trick attributed to Karen Uhlenbeck, we simplify the Ricci flow equation by applying an appropriate change of frame (cf. [13 p. 155] ). At t = 0 t=0 t = 0 , choose a left-invariant g 0 g_0 g 0 -orthonormal frame { X i } \{X_i\} { X i } . Evolve this frame by
∂ t X i = Ric g ( t ) ( X i ) , \partial_t X_i = \operatorname{Ric}_{g(t)}(X_i), ∂ t X i = Ric g ( t ) ( X i ) ,
where Ric g ( t ) \operatorname{Ric}_{g(t)} Ric g ( t ) is the Ricci endomorphism. This provides a moving frame { X i ( t ) } \{X_i(t)\} { X i ( t )} on G G G which remains g ( t ) g(t) g ( t ) -orthonormal under Ricci flow:
∂ ∂ t g ( X i , X j ) = ( ∂ t g ) ( X i , X j ) + g ( ∂ t X i , X j ) + g ( X i , ∂ t X j ) = − 2 ric g ( t ) ( X i , X j ) + g ( Ric g ( t ) ( X i ) , X j ) + g ( X i , Ric g ( t ) ( X j ) ) = 0. \frac{\partial}{\partial t}g(X_i,X_j) = (\partial_tg)(X_i,X_j) + g(\partial_tX_i,X_j) + g(X_i,\partial_tX_j)
= -2\operatorname{ric}_{g(t)}(X_i,X_j) + g(\operatorname{Ric}_{g(t)}(X_i),X_j) + g(X_i,\operatorname{Ric}_{g(t)}(X_j)) = 0. ∂ t ∂ g ( X i , X j ) = ( ∂ t g ) ( X i , X j ) + g ( ∂ t X i , X j ) + g ( X i , ∂ t X j ) = − 2 ric g ( t ) ( X i , X j ) + g ( Ric g ( t ) ( X i ) , X j ) + g ( X i , Ric g ( t ) ( X j )) = 0.
The structure constants are μ i j k ( t ) X k ( t ) = [ X i ( t ) , X j ( t ) ] \mu_{ij}^k(t)X_k(t) = [X_i(t),X_j(t)] μ ij k ( t ) X k ( t ) = [ X i ( t ) , X j ( t )] . By differentiating this equation, we get the ODE for the structure constants
d μ i j k d t X k = [ Ric g ( t ) ( X i ) , X j ] + [ X i , Ric g ( t ) ( X j ) ] − Ric g ( t ) [ X i , X j ] . \frac{d\mu_{ij}^k}{dt}X_k = [\operatorname{Ric}_{g(t)}(X_i),X_j] + [X_i,\operatorname{Ric}_{g(t)}(X_j)] - \operatorname{Ric}_{g(t)}[X_i,X_j]. d t d μ ij k X k = [ Ric g ( t ) ( X i ) , X j ] + [ X i , Ric g ( t ) ( X j )] − Ric g ( t ) [ X i , X j ] .
To express this evolution invariantly, let ⟨ ⋅ , ⋅ ⟩ 0 \langle\cdot,\cdot\rangle_0 ⟨ ⋅ , ⋅ ⟩ 0 be the inner product on g \mathfrak{g} g given by evaluating g 0 g_0 g 0 at the identity. Take μ ∈ Λ 2 g ∗ ⊗ g \mu\in\Lambda^2\mathfrak{g}^* \otimes\mathfrak{g} μ ∈ Λ 2 g ∗ ⊗ g such that ( g , μ ) (\mathfrak{g},\mu) ( g , μ ) and ( g , [ ⋅ , ⋅ ] ) (\mathfrak{g},[\cdot,\cdot]) ( g , [ ⋅ , ⋅ ]) are isomorphic as Lie algebras. By the third theorem of Lie, there exists a connected and simply-connected Lie group G ~ μ \widetilde{G}_\mu G μ with Lie algebra ( g , μ ) (\mathfrak{g},\mu) ( g , μ ) . Let g ~ μ \widetilde{g}_\mu g μ be the left-invariant metric on G ~ μ \widetilde{G}_\mu G μ such that g ~ μ ( e ) = ⟨ ⋅ , ⋅ ⟩ 0 \widetilde{g}_\mu(e)=\langle\cdot,\cdot\rangle_0 g μ ( e ) = ⟨ ⋅ , ⋅ ⟩ 0 . Let Ric μ \operatorname{Ric}_\mu Ric μ be the Ricci endomorphism of ( G ~ μ , g ~ μ ) (\widetilde{G}_\mu,\widetilde{g}_\mu) ( G μ , g μ ) at the identity. Consider
d μ d t = δ μ ( Ric μ ) , μ ( 0 ) = [ ⋅ , ⋅ ] (6) \frac{d\mu}{dt} = \delta_\mu(\operatorname{Ric}_\mu), \qquad\mu(0)=[\cdot,\cdot] \tag*{(6)} d t d μ = δ μ ( Ric μ ) , μ ( 0 ) = [ ⋅ , ⋅ ] (6)
where δ μ : End ( g ) → Λ 2 g ∗ ⊗ g \delta_\mu:\operatorname{End}(\mathfrak{g}) \to\Lambda^2\mathfrak{g}^* \otimes\mathfrak{g} δ μ : End ( g ) → Λ 2 g ∗ ⊗ g is given by
δ μ ( α ) : = d d t ∣ t = 0 e − α t μ ( e α t ⋅ , e α t ⋅ ) = μ ( α ⋅ , ⋅ ) + μ ( ⋅ , α ⋅ ) − α μ ( ⋅ , ⋅ ) . \delta_\mu(\alpha) := \left.\frac{d}{dt}\right|_{t=0}e^{-\alpha t}\mu(e^{\alpha t}\cdot,e^{\alpha t}\cdot) = \mu(\alpha\cdot,\cdot) + \mu(\cdot,\alpha\cdot) - \alpha\mu(\cdot,\cdot). δ μ ( α ) := d t d t = 0 e − α t μ ( e α t ⋅ , e α t ⋅ ) = μ ( α ⋅ , ⋅ ) + μ ( ⋅ , α ⋅ ) − α μ ( ⋅ , ⋅ ) .
for all α ∈ End ( g ) \alpha\in\operatorname{End}(\mathfrak{g}) α ∈ End ( g ) . The ordinary differential equation (6) is called the bracket flow . As the time derivative d μ d t \frac{d\mu}{dt} d t d μ is tangent to the orbit GL ( g ) ⋅ [ ⋅ , ⋅ ] \operatorname{GL}(\mathfrak{g}) \cdot[\cdot,\cdot] GL ( g ) ⋅ [ ⋅ , ⋅ ] , there exists a unique smooth one-parameter family of linear transformations Φ ( t ) \Phi(t) Φ ( t ) satisfying μ ( t ) = Φ ( t ) ⋅ [ ⋅ , ⋅ ] \mu(t)=\Phi(t)\cdot[\cdot,\cdot] μ ( t ) = Φ ( t ) ⋅ [ ⋅ , ⋅ ] and solving
d Φ d t = − Ric μ Φ , Φ ( 0 ) = Id . \frac{d\Phi}{dt}=-\operatorname{Ric}_{\mu}\Phi,\qquad\Phi(0)=\operatorname{Id}. d t d Φ = − Ric μ Φ , Φ ( 0 ) = Id .
The maps Φ ( t ) : ( g , [ ⋅ , ⋅ ] ) → ( g , μ ( t ) ) \Phi(t):(\mathfrak{g},[\cdot,\cdot])\to(\mathfrak{g},\mu(t)) Φ ( t ) : ( g , [ ⋅ , ⋅ ]) → ( g , μ ( t )) are Lie algebra isomorphisms. They integrate uniquely to Lie group isomorphisms φ ~ ( t ) : G ~ → G ~ μ ( t ) \widetilde{\varphi}(t):\widetilde{G}\to\widetilde{G}_{\mu(t)} φ ( t ) : G → G μ ( t ) with d φ ~ ( t ) e = Φ ( t ) d\widetilde{\varphi}(t)_{e}=\Phi(t) d φ ( t ) e = Φ ( t ) . Now, if Γ \Gamma Γ is the kernel of the universal covering homomorphism, then define G μ ( t ) = G ~ μ ( t ) / φ ~ ( t ) ( Γ ) G_{\mu(t)}=\widetilde{G}_{\mu(t)}/\widetilde{\varphi}(t)(\Gamma) G μ ( t ) = G μ ( t ) / φ ( t ) ( Γ ) . The metrics g ~ μ ( t ) \widetilde{g}_{\mu(t)} g μ ( t ) descend to g μ ( t ) g_{\mu(t)} g μ ( t ) , and the Lie group isomorphisms φ ~ ( t ) \widetilde{\varphi}(t) φ ( t ) descend to φ ( t ) : G → G μ ( t ) \varphi(t):G\to G_{\mu(t)} φ ( t ) : G → G μ ( t ) . By pulling back g μ ( t ) g_{\mu(t)} g μ ( t ) by the Lie group isomorphisms φ ( t ) : G → G μ ( t ) \varphi(t):G\to G_{\mu(t)} φ ( t ) : G → G μ ( t ) , we obtain the Ricci flow of left-invariant metrics g ( t ) g(t) g ( t ) :
d d t ( φ ( t ) ∗ g μ ( t ) ) ( e ) = d d t ⟨ Φ ( t ) ⋅ , Φ ( t ) ⋅ ⟩ 0 = ⟨ Φ ′ ( t ) ⋅ , Φ ( t ) ⋅ ⟩ 0 + ⟨ Φ ( t ) ⋅ , Φ ′ ( t ) ⋅ ⟩ 0 = − ⟨ Ric μ ( t ) Φ ( t ) ⋅ , Φ ( t ) ⋅ ⟩ 0 − ⟨ Φ ( t ) ⋅ , Ric μ ( t ) Φ ( t ) ⋅ ⟩ 0 = − 2 ⟨ Ric μ ( t ) Φ ( t ) ⋅ , Φ ( t ) ⋅ ⟩ 0 = − 2 ric φ ( t ) ∗ g μ ( t ) ( e ) . \begin{aligned}
\frac{d}{dt}\left(\varphi(t)^*g_{\mu(t)}\right)(e)=\frac{d}{dt}\langle\Phi(t)\cdot,\Phi(t)\cdot\rangle_0 \\
&=\langle\Phi'(t)\cdot,\Phi(t)\cdot\rangle_0+\langle\Phi(t)\cdot,\Phi'(t)\cdot\rangle_0 \\
&=-\langle\operatorname{Ric}_{\mu(t)}\Phi(t)\cdot,\Phi(t)\cdot\rangle_0-\langle\Phi(t)\cdot,\operatorname{Ric}_{\mu(t)}\Phi(t)\cdot\rangle_0 \\
&=-2\langle\operatorname{Ric}_{\mu(t)}\Phi(t)\cdot,\Phi(t)\cdot\rangle_0 \\
&=-2\operatorname{ric}_{\varphi(t)^*g_{\mu(t)}}(e).
\end{aligned} d t d ( φ ( t ) ∗ g μ ( t ) ) ( e ) = d t d ⟨ Φ ( t ) ⋅ , Φ ( t ) ⋅ ⟩ 0 = ⟨ Φ ′ ( t ) ⋅ , Φ ( t ) ⋅ ⟩ 0 + ⟨ Φ ( t ) ⋅ , Φ ′ ( t ) ⋅ ⟩ 0 = − ⟨ Ric μ ( t ) Φ ( t ) ⋅ , Φ ( t ) ⋅ ⟩ 0 − ⟨ Φ ( t ) ⋅ , Ric μ ( t ) Φ ( t ) ⋅ ⟩ 0 = − 2 ⟨ Ric μ ( t ) Φ ( t ) ⋅ , Φ ( t ) ⋅ ⟩ 0 = − 2 ric φ ( t ) ∗ g μ ( t ) ( e ) .
The Ricci flow g ( t ) g(t) g ( t ) and the bracket flow g μ ( t ) g_{\mu(t)} g μ ( t ) differ only by pullback by time-dependent diffeomorphisms [21 Thm. 3.3] . The curvatures and all Riemannian invariants of the Ricci flow can be studied along the bracket flow.
Riemannian geometry on twisted principal bundles
Here, we define twisted principal bundles, the class of manifolds considered in our main theorems. The material in this subsection is based on [27 §2] and [26 §3.1] .
Definition 2.1. Let G G G be a Lie group. A twisted principal G G G -bundle is a fibre bundle π : M → B \pi:M\to B π : M → B with fibre G G G and structure group contained in Aut ( G ) × G L \operatorname{Aut}(G)\times G_L Aut ( G ) × G L where the topology on Aut ( G ) \operatorname{Aut}(G) Aut ( G ) is discrete. Here, G L G_L G L is the group of left-multiplications of G G G acting on itself.
The structure group of a twisted principal bundle need not commute with right group multiplication G R G_R G R , and so the fibrewise G G G -actions do not, in general, extend to a global action on M M M . Let ( U α , ψ α ) (U_\alpha,\psi_\alpha) ( U α , ψ α ) be a local trivialisation of M M M , and define the right G G G -action on M U α : = π − 1 ( U α ) M_{U_\alpha}:=\pi^{-1}(U_\alpha) M U α := π − 1 ( U α ) by
(7)
R g α : M U α × G → M U α , R g α ( p ) = ψ α − 1 ( ψ α ( p ) ⋅ g ) . (7) R^\alpha_g:M_{U_\alpha}\times G\to M_{U_\alpha},\qquad R^\alpha_g(p)=\psi_\alpha^{-1}\bigl(\psi_\alpha(p)\cdot g\bigr). \tag*{(7)} R g α : M U α × G → M U α , R g α ( p ) = ψ α − 1 ( ψ α ( p ) ⋅ g ) . (7)
If ( U β , ψ β ) (U_\beta,\psi_\beta) ( U β , ψ β ) is an overlapping local trivialisation with transition function τ α β ( x ) = φ α β ∘ L h ( x ) \tau_{\alpha\beta}(x)=\varphi_{\alpha\beta}\circ L_{h(x)} τ α β ( x ) = φ α β ∘ L h ( x ) where φ α β \varphi_{\alpha\beta} φ α β is constant, then
R g α ( p ) = ψ α − 1 ( ψ β ( p ) ⋅ φ α β ( g ) ) = R φ α β ( g ) β ( p ) . R^\alpha_g(p)=\psi_\alpha^{-1}\bigl(\psi_\beta(p)\cdot\varphi_{\alpha\beta}(g)\bigr)=R^\beta_{\varphi_{\alpha\beta}(g)}(p). R g α ( p ) = ψ α − 1 ( ψ β ( p ) ⋅ φ α β ( g ) ) = R φ α β ( g ) β ( p ) .
Example 2.2 (Möbius band). The Möbius band is a twisted principal R \mathbb{R} R -bundle over S 1 S^1 S 1 with structure group { ± Id } ⋉ R \{\pm\operatorname{Id}\}\ltimes\mathbb{R} { ± Id } ⋉ R . Cover S 1 S^1 S 1 by two trivialisations ( U α , ψ α ) (U_\alpha,\psi_\alpha) ( U α , ψ α ) and ( U β , ψ β ) (U_\beta,\psi_\beta) ( U β , ψ β ) with transition functions Id \operatorname{Id} Id and − Id -\operatorname{Id} − Id . Then, on one of the overlaps (connected components of U α ∩ U β ) U_\alpha\cap U_\beta) U α ∩ U β ) ,
R g α ( p ) = ψ α − 1 ( ψ α ( p ) + g ) = ψ β − 1 ( ψ β ( p ) − g ) = R − g β ( p ) . R^\alpha_g(p)=\psi_\alpha^{-1}\bigl(\psi_\alpha(p)+g\bigr)=\psi_\beta^{-1}\bigl(\psi_\beta(p)-g\bigr)=R^\beta_{-g}(p). R g α ( p ) = ψ α − 1 ( ψ α ( p ) + g ) = ψ β − 1 ( ψ β ( p ) − g ) = R − g β ( p ) .
Example 2.3 (Torus bundles). Let H ∈ SL 2 ( Z ) H\in\operatorname{SL}_2(\mathbb{Z}) H ∈ SL 2 ( Z ) . The following is a twisted principal T 2 \mathbb{T}^2 T 2 -bundle over S 1 S^1 S 1 ,
M = ( R × T 2 ) / ∼ , ( s + 1 , x ) ∼ ( s , H x ) . M=(\mathbb{R}\times\mathbb{T}^2)/\sim,\qquad(s+1,x)\sim(s,Hx). M = ( R × T 2 ) / ∼ , ( s + 1 , x ) ∼ ( s , H x ) .
If H H H is elliptic, i.e., ∣ tr H ∣ = 2 \lvert\operatorname{tr}H\rvert=2 ∣ tr H ∣ = 2 , but H H H is not elliptic, then M M M is finitely covered by T 3 \mathbb{T}^3 T 3 , so M M M admits a flat geometry. If H H H is parabolic, that is, ∣ tr H ∣ = 2 \lvert\operatorname{tr}H\rvert=2 ∣ tr H ∣ = 2 , but H H H is not elliptic, then M M M admits a Nil geometry. If H H H is hyperbolic, that is, ∣ tr H ∣ > 2 \lvert\operatorname{tr}H\rvert>2 ∣ tr H ∣ > 2 , then M M M admits a Sol geometry.
Although there may not be a globally defined G G G -action on M M M , the local actions (7) provide well-defined natural transformation groups G U ⊂ Diff ( M U ) G_U\subset\operatorname{Diff}(M_U) G U ⊂ Diff ( M U ) that act freely and transitively on the fibres. A choice of local trivialisation gives an isomorphism G ≅ G U G\cong G_U G ≅ G U . In lieu of a global action, these natural transformation groups are sufficient to provide a well-defined notion of G G G -invariance on M M M .
A tensor on a twisted principal bundle π : M → B \pi:M\to B π : M → B is called invariant if it is invariant under the actions of the local transformation groups G U G_U G U for all trivialising neighbourhoods M U ≃ U × G M_U \simeq U \times G M U ≃ U × G . The bundle of invariant vertical vector fields is a Lie algebra bundle called the adjoint bundle ad M → B \operatorname{ad}M \to B ad M → B . If { U α × G } \{U_\alpha\times G\} { U α × G } is a bundle atlas of M M M , then { U α × g } \{U_\alpha\times\mathfrak{g}\} { U α × g } is a bundle atlas of ad M \operatorname{ad}M ad M with transition functions given by the group homomorphism
Aut ( G ) ⋉ G L → Aut ( g ) , ( φ , g ) ↦ d φ e ∘ Ad ( g ) . \operatorname{Aut}(G) \ltimes G_L \to\operatorname{Aut}(\mathfrak{g}), \qquad(\varphi,g) \mapsto d\varphi_e \circ\operatorname{Ad}(g). Aut ( G ) ⋉ G L → Aut ( g ) , ( φ , g ) ↦ d φ e ∘ Ad ( g ) .
As the structure group of ad M \operatorname{ad}M ad M is contained in Aut ( g ) \operatorname{Aut}(\mathfrak{g}) Aut ( g ) , the Lie bracket on g \mathfrak{g} g induces a canonical section [ ⋅ , ⋅ ] ∈ Γ ( Λ 2 ad M ∗ ⊗ ad M ) [\cdot,\cdot] \in\Gamma(\Lambda^2\operatorname{ad}M^* \otimes\operatorname{ad}M) [ ⋅ , ⋅ ] ∈ Γ ( Λ 2 ad M ∗ ⊗ ad M ) whose restrictions to fibres are Lie brackets.
A horizontal distribution H ⊂ T M \mathcal{H} \subset TM H ⊂ T M on a twisted principal bundle is a distribution that is complementary to the vertical distribution V = ker π ∗ \mathcal{V}=\ker\pi_* V = ker π ∗ . An invariant horizontal distribution defines a principal connection . A principal connection is a vector bundle map
ω : T M → π ∗ ad M \omega:TM\to\pi^*\operatorname{ad}M ω : T M → π ∗ ad M
which when restricted to the vertical distribution is the canonical identification of V p \mathcal{V}_p V p with ad M π ( p ) \operatorname{ad}M_{\pi(p)} ad M π ( p ) . Vice versa, the kernel of a principal connection is a given invariant horizontal distribution. A horizontal lift of X ∈ X ( B ) X\in\mathfrak{X}(B) X ∈ X ( B ) is denoted X ˉ \bar{X} X ˉ . Given an invariant vertical vector field U U U , denote the corresponding section of the adjoint bundle as U a d : = ω ( U ) U_{\mathrm{ad}}:=\omega(U) U ad := ω ( U ) .
Equip M → B M\to B M → B with an invariant Riemannian metric g g g . The orthogonal complement of the vertical distribution determines a principal connection ω \omega ω , the restriction to invariant vertical vector fields determines a vector bundle metric h h h on ad M \operatorname{ad}M ad M , and the metric on the horizontal distribution descends to a Riemannian metric g ˇ \check{g} g ˇ on B B B . The invariant Riemannian metric uniquely decomposes as
g = ω ∗ h + π ∗ g ˇ = h ( ω ⋅ , ω ⋅ ) + g ˇ ( π ∗ ⋅ , π ∗ ⋅ ) . g=\omega^*h+\pi^*\check{g}=h(\omega\cdot,\omega\cdot)+\check{g}(\pi_*\cdot,\pi_*\cdot). g = ω ∗ h + π ∗ g ˇ = h ( ω ⋅ , ω ⋅ ) + g ˇ ( π ∗ ⋅ , π ∗ ⋅ ) .
Conversely, every such triple ( h , ω , g ˇ ) (h,\omega,\check{g}) ( h , ω , g ˇ ) uniquely determines an invariant Riemannian metric on M M M . There is a bijection
{ invariant Riemannian metrics g on M } ⟷ { vector bundle metrics h on ad M principal connections ω Riemannian metrics g ˇ on B } . (8) \{ \text{invariant Riemannian metrics }g\text{ on }M\}
\longleftrightarrow
\left\{
\begin{array}{l}
\text{vector bundle metrics }h\text{ on }\operatorname{ad}M\\
\text{principal connections }\omega\\
\text{Riemannian metrics }\check{g}\text{ on }B
\end{array}
\right\}. \tag*{(8)} { invariant Riemannian metrics g on M } ⟷ ⎩ ⎨ ⎧ vector bundle metrics h on ad M principal connections ω Riemannian metrics g ˇ on B ⎭ ⎬ ⎫ . (8)
Let U , V , W U,V,W U , V , W be vertical vector fields on M M M , and let X , Y , Z X,Y,Z X , Y , Z be vector fields on B B B . If U U U is invariant, denote U a d ∈ Γ ( ad M ) U_{\mathrm{ad}}\in\Gamma(\operatorname{ad}M) U ad ∈ Γ ( ad M ) as the corresponding section. A principal connection on M → B M\to B M → B determines horizontal lifts X ˉ \bar{X} X ˉ that are invariant under the local natural transformations G U G_U G U . This induces an affine connection on ad M \operatorname{ad}M ad M :
∇ a d : X ( B ) × Γ ( ad M ) → Γ ( ad M ) , ∇ X a d U a d : = [ X ˉ , U ] a d . (9) \nabla^{\mathrm{ad}}:\mathfrak{X}(B)\times\Gamma(\operatorname{ad}M)\to\Gamma(\operatorname{ad}M),\qquad\nabla^{\mathrm{ad}}_XU_{\mathrm{ad}}:=[\bar{X},U]_{\mathrm{ad}}. \tag*{(9)} ∇ ad : X ( B ) × Γ ( ad M ) → Γ ( ad M ) , ∇ X ad U ad := [ X ˉ , U ] ad . (9)
Equip M M M with an invariant Riemannian metric g g g . Let ∇ M \nabla^M ∇ M be the Levi-Civita connection of ( M , g ) (M,g) ( M , g ) , and define
∇ : X ( B ) × Γ ( ad M ) → Γ ( ad M ) , ∇ X U a d : = ( V ∇ X ˉ M U ) a d , \nabla:\mathfrak{X}(B)\times\Gamma(\operatorname{ad}M)\to\Gamma(\operatorname{ad}M),\qquad\nabla_XU_{\mathrm{ad}}:=(\mathcal{V}\nabla^M_{\bar{X}}U)_{\mathrm{ad}}, ∇ : X ( B ) × Γ ( ad M ) → Γ ( ad M ) , ∇ X U ad := ( V ∇ X ˉ M U ) ad ,
where V \mathcal{V} V denotes the vertical component of the vector field. This connection is metric on ( ad M , h ) (\operatorname{ad}M,h) ( ad M , h ) and is characterised by Koszul’s formula:
h ( ∇ X U a d , V a d ) = 1 2 ( X h ( U a d , V a d ) + h ( ∇ X a d U a d , V a d ) − h ( U a d , ∇ X a d V a d ) ) . h(\nabla_XU_{\mathrm{ad}},V_{\mathrm{ad}})=\frac{1}{2}\left(Xh(U_{\mathrm{ad}},V_{\mathrm{ad}})+h(\nabla^{\mathrm{ad}}_XU_{\mathrm{ad}},V_{\mathrm{ad}})-h(U_{\mathrm{ad}},\nabla^{\mathrm{ad}}_XV_{\mathrm{ad}})\right). h ( ∇ X U ad , V ad ) = 2 1 ( X h ( U ad , V ad ) + h ( ∇ X ad U ad , V ad ) − h ( U ad , ∇ X ad V ad ) ) .
As such, the metric connection ∇ \nabla ∇ may be defined intrinsically on ( ad M , h ) (\operatorname{ad}M,h) ( ad M , h ) without the need to compute the Levi-Civita connection of ( M , g ) (M,g) ( M , g ) .
The space of connections on a vector bundle is affine and modelled on the endomorphism-valued one-forms. There exists an L ∈ Ω 1 ( B ; End ( ad M ) ) L\in\Omega^1(B;\operatorname{End}(\operatorname{ad}M)) L ∈ Ω 1 ( B ; End ( ad M )) such that
∇ a d = ∇ − L . \nabla^{\mathrm{ad}}=\nabla-L. ∇ ad = ∇ − L .
By the torsion freeness of ∇ M \nabla^M ∇ M , this one-form is the shape operator of the fibres:
L X U a d = ( V ∇ U M X ˉ ) a d . L_XU_{\mathrm{ad}}=(\mathcal{V}\nabla^M_U\bar{X})_{\mathrm{ad}}. L X U ad = ( V ∇ U M X ˉ ) ad .
If { U i } \{U_i\} { U i } is a local orthonormal vertical frame on ( M , g ) (M,g) ( M , g ) and { X j } \{X_j\} { X j } is a local orthonormal frame on ( B , g ˇ ) (B,\check{g}) ( B , g ˇ ) , then the mean curvature vector of the fibres is given by
N = ∑ i H ∇ U i M U i = − ∑ j ( tr L X j ) X ˉ j . N=\sum_i\mathcal{H}\nabla^M_{U_i}U_i=-\sum_j(\operatorname{tr}L_{X_j})\bar{X}_j. N = i ∑ H ∇ U i M U i = − j ∑ ( tr L X j ) X ˉ j .
This is a basic horizontal vector field in T M TM T M . We also write N N N for its projection π ∗ N \pi_*N π ∗ N . The homogeneous mean curvature vector of the fibres is characterised by
h ( H , U a d ) = tr ( ad U a d ) . h(H,U_{\mathrm{ad}})=\operatorname{tr}(\operatorname{ad}U_{\mathrm{ad}}). h ( H , U ad ) = tr ( ad U ad ) .
Let ∇ ^ \widehat{\nabla} ∇ be the Levi-Civita connection of the fibres of ( M , g ) → ( B , g ˘ ) (M,g)\to(B,\breve{g}) ( M , g ) → ( B , g ˘ ) given by ∇ ^ U V = V ∇ U M V \widehat{\nabla}_U V=\mathcal{V}\nabla^M_U V ∇ U V = V ∇ U M V . Its restriction to invariant vertical vector fields is again invariant, thus providing the bundle map
∇ ^ : ad M ⊗ ad M → ad M , ∇ ^ U a d V a d : = ( V ∇ U M V ) a d . \widehat{\nabla}:\operatorname{ad}M\otimes\operatorname{ad}M\to\operatorname{ad}M,\qquad\widehat{\nabla}_{U_{\mathrm{ad}}}V_{\mathrm{ad}}:=(\mathcal{V}\nabla^M_U V)_{\mathrm{ad}}. ∇ : ad M ⊗ ad M → ad M , ∇ U ad V ad := ( V ∇ U M V ) ad .
Let ∇ ˘ \breve{\nabla} ∇ ˘ be the Levi-Civita connection of the base ( B , g ˘ ) (B,\breve{g}) ( B , g ˘ ) . The curvature of the principal connection ω \omega ω is given by
F ∈ Ω 2 ( B ; ad M ) , F X , Y : = − ω [ X ‾ , Y ‾ ] . F\in\Omega^2(B;\operatorname{ad}M),\qquad F_{X,Y}:=-\omega[\overline{X},\overline{Y}]. F ∈ Ω 2 ( B ; ad M ) , F X , Y := − ω [ X , Y ] .
O’Neill’s integrability tensor, viewed as an ad M \operatorname{ad}M ad M -valued two-form on B B B , is given by
A ∈ Ω 2 ( B ; ad M ) , A X , Y : = 1 2 ω [ X ‾ , Y ‾ ] = − 1 2 F X , Y . A\in\Omega^2(B;\operatorname{ad}M),\qquad A_{X,Y}:=\frac{1}{2}\omega[\overline{X},\overline{Y}]=-\frac{1}{2}F_{X,Y}. A ∈ Ω 2 ( B ; ad M ) , A X , Y := 2 1 ω [ X , Y ] = − 2 1 F X , Y .
For X , Y ∈ X ( B ) X,Y\in\mathcal{X}(B) X , Y ∈ X ( B ) and U a d ∈ Γ ( ad M ) U_{\mathrm{ad}}\in\Gamma(\operatorname{ad}M) U ad ∈ Γ ( ad M ) , write
A X ( Y ) : = A X , Y , ( A ∗ U a d ) ( X , Y ) : = h ( A X , Y , U a d ) , ( L U a d ) ( X ) : = L X U a d . A_X(Y):=A_{X,Y},\qquad(A^*U_{\mathrm{ad}})(X,Y):=h(A_{X,Y},U_{\mathrm{ad}}),\qquad(LU_{\mathrm{ad}})(X):=L_XU_{\mathrm{ad}}. A X ( Y ) := A X , Y , ( A ∗ U ad ) ( X , Y ) := h ( A X , Y , U ad ) , ( L U ad ) ( X ) := L X U ad .
We use g g g for the induced tensor inner products, obtained by full contraction with h h h and g ˘ \breve{g} g ˘ , and use the corresponding tensor norms.
Next, we recall O’Neill’s Ricci curvature formulas, expressed in the notation and tensor inner product conventions of this subsection [1 9.36] . In the following, we drop the subscript a d \mathrm{ad} ad for sections of the adjoint bundle.
Proposition 2.4 (O’Neill’s Ricci curvature formulas). Let M → B M\to B M → B be a twisted principal bundle equipped with an invariant metric g g g . Then,
ric g ( U , U ) = ric ^ g ( U , U ) + ∣ A ∗ U ∣ 2 + h ( L N U , U ) + h ( ( ∇ ∗ L ) U , U ) . (10) \operatorname{ric}_g(U,U)=\widehat{\operatorname{ric}}_g(U,U)+|A^*U|^2+h(L_NU,U)+h((\nabla^*L)U,U). \tag*{(10)} ric g ( U , U ) = ric g ( U , U ) + ∣ A ∗ U ∣ 2 + h ( L N U , U ) + h (( ∇ ∗ L ) U , U ) . (10) ric g ( U , X ‾ ) = − h ( ∇ ^ ∗ L X , U ) − h ( ( ∇ ∗ A ) X , U ) − g ( A X , L U ) + h ( A X , N , U ) . (11) \operatorname{ric}_g(U,\overline{X})=-h(\widehat{\nabla}^*L_X,U)-h((\nabla^*A)X,U)-g(A_X,LU)+h(A_{X,N},U). \tag*{(11)} ric g ( U , X ) = − h ( ∇ ∗ L X , U ) − h (( ∇ ∗ A ) X , U ) − g ( A X , LU ) + h ( A X , N , U ) . (11) ric g ( X ‾ , X ‾ ) = ric ˘ g ( X , X ) − 2 ∣ A X ∣ 2 − ∣ L X ∣ 2 + g ˘ ( ∇ ˘ X N , X ) . (12) \operatorname{ric}_g(\overline{X},\overline{X})=\breve{\operatorname{ric}}_g(X,X)-2|A_X|^2-|L_X|^2+\breve{g}(\breve{\nabla}_XN,X). \tag*{(12)} ric g ( X , X ) = ric ˘ g ( X , X ) − 2∣ A X ∣ 2 − ∣ L X ∣ 2 + g ˘ ( ∇ ˘ X N , X ) . (12)
Here ric ^ g \widehat{\operatorname{ric}}_g ric g is the Ricci tensor of the fibres, A ∗ A^* A ∗ is the adjoint of A A A with respect to h h h , and ∇ ∗ , ∇ ^ ∗ \nabla^*,\widehat{\nabla}^* ∇ ∗ , ∇ ∗ are the formal adjoints to the connections ∇ , ∇ ^ \nabla,\widehat{\nabla} ∇ , ∇ :
∇ ∗ η = − ∑ i ( ∇ X i η ) X i , ∇ ^ ∗ ξ = − ∑ j ( ∇ ^ U j ξ ) U j , \nabla^*\eta=-\sum_i(\nabla_{X_i}\eta)X_i,\qquad\widehat{\nabla}^*\xi=-\sum_j(\widehat{\nabla}_{U_j}\xi)U_j, ∇ ∗ η = − i ∑ ( ∇ X i η ) X i , ∇ ∗ ξ = − j ∑ ( ∇ U j ξ ) U j ,
for n n n -forms η \eta η on B B B with values in tensor powers of ad M \operatorname{ad}M ad M , and tensor fields ξ \xi ξ on ad M \operatorname{ad}M ad M , where { X i } \{X_i\} { X i } is a local orthonormal frame on ( B , g ˘ ) (B,\breve{g}) ( B , g ˘ ) , and { U i } \{U_i\} { U i } is a local orthonormal frame on ( ad M , h ) (\operatorname{ad}M,h) ( ad M , h ) .
:::
These equations simplify considerably when the base is one-dimensional. Notably, all one-dimensional distributions are integrable, so A = 0 A=0 A = 0 . To simplify notation, for a local unit vector field X X X on B B B , write L = L X L=L_X L = L X and ∇ ∗ L = − ( ∇ X L ) X = − ∇ L \nabla^*L=-(\nabla_XL)X=-\nabla L ∇ ∗ L = − ( ∇ X L ) X = − ∇ L .
Proposition 2.5 (O’Neill’s Riemannian curvature formulas when dim B = 1 \dim B=1 dim B = 1 ). Let M → B 1 M\to B^1 M → B 1 be a twisted principal bundle equipped with an invariant metric g g g . Then,
Rm g ( U , V , W , U ′ ) = Rm ^ g ( U , V , W , U ′ ) + h ( L U , W ) h ( L V , U ′ ) − h ( L V , W ) h ( L U , U ′ ) . (13) \operatorname{Rm}_g(U,V,W,U')=\widehat{\operatorname{Rm}}_g(U,V,W,U')+h(LU,W)h(LV,U')-h(LV,W)h(LU,U'). \tag*{(13)} Rm g ( U , V , W , U ′ ) = Rm g ( U , V , W , U ′ ) + h ( LU , W ) h ( L V , U ′ ) − h ( L V , W ) h ( LU , U ′ ) . (13) Rm g ( U , V , W , X ‾ ) = h ( ( ∇ ^ V L ) U , W ) − h ( ( ∇ ^ U L ) V , W ) . (14) \operatorname{Rm}_g(U,V,W,\overline{X})=h((\widehat{\nabla}_VL)U,W)-h((\widehat{\nabla}_UL)V,W). \tag*{(14)} Rm g ( U , V , W , X ) = h (( ∇ V L ) U , W ) − h (( ∇ U L ) V , W ) . (14) Rm g ( U , X ‾ , X ‾ , V ) = − h ( ( ∇ L ) U , V ) − h ( L 2 U , V ) . (15) \operatorname{Rm}_g(U,\overline{X},\overline{X},V)=-h((\nabla L)U,V)-h(L^2U,V). \tag*{(15)} Rm g ( U , X , X , V ) = − h (( ∇ L ) U , V ) − h ( L 2 U , V ) . (15)
where X ‾ \overline{X} X is unit and Rm ^ g \widehat{\operatorname{Rm}}_g Rm g is the Riemannian curvature tensor of the fibres.
Proposition 2.6 (O’Neill’s Ricci curvature formulas when dim B = 1 \dim B=1 dim B = 1 ). Let M → B 1 M \to B^{1} M → B 1 be a twisted principal bundle equipped with an invariant metric g g g . Then,
ric g ( U , U ) = ric ^ g ( U , U ) − ( tr L ) h ( L U , U ) − h ( ( ∇ L ) U , U ) (16) \operatorname{ric}_{g}(U,U)=\widehat{\operatorname{ric}}_{g}(U,U)-(\operatorname{tr}L)h(LU,U)-h((\nabla L)U,U) \tag*{(16)} ric g ( U , U ) = ric g ( U , U ) − ( tr L ) h ( LU , U ) − h (( ∇ L ) U , U ) (16)
ric g ( X ‾ , U ) = − h ( ∇ ^ ∗ L , U ) (17) \operatorname{ric}_{g}(\overline{X},U)=-h(\widehat{\nabla}^{*}L,U) \tag*{(17)} ric g ( X , U ) = − h ( ∇ ∗ L , U ) (17)
ric g ( X ‾ , X ‾ ) = − ∣ L ∣ 2 − tr ( ∇ L ) (18) \operatorname{ric}_{g}(\overline{X},\overline{X})=-\lvert L\rvert^{2}-\operatorname{tr}(\nabla L) \tag*{(18)} ric g ( X , X ) = − ∣ L ∣ 2 − tr ( ∇ L ) (18)
where X ‾ \overline{X} X is unit and ric ^ g \widehat{\operatorname{ric}}_{g} ric g is the Ricci tensor of the fibres.
Transitive Lie algebroids
In this subsection, we describe the geometry of transitive Lie algebroids and their relationship to the geometry of twisted principal bundles. For general theory, we refer to [28 ] , and for Riemannian Lie algebroids [5 ] .
The vector bundle of all invariant vector fields is called the Atiyah bundle T M / G → B TM/G \to B T M / G → B . This is perhaps misleading notation as there is no global G G G -action on M M M . Even so, this vector bundle is well-defined since the local transformation groups G U ⊂ Diff ( M U ) G_{U} \subset\operatorname{Diff}(M_{U}) G U ⊂ Diff ( M U ) are well-defined. If { U α × G } \{U_{\alpha} \times G\} { U α × G } is a bundle atlas of M M M , then the invariant vector fields on these trivialisations are identified by
T ( U α × G ) / G ≃ T U α ⊕ ( U α × g ) . T(U_{\alpha} \times G)/G \simeq TU_{\alpha} \oplus(U_{\alpha} \times\mathfrak{g}). T ( U α × G ) / G ≃ T U α ⊕ ( U α × g ) .
The differentials of the transition functions of M M M descend to the local quotients and satisfy the cocycle condition, providing the transition functions of T M / G → B TM/G \to B T M / G → B .
Definition 2.7. A Lie algebroid ( E , B , [ ⋅ , ⋅ ] , # ) (E,B,[\cdot,\cdot],\#) ( E , B , [ ⋅ , ⋅ ] , # ) is a vector bundle E → B E \to B E → B equipped with a Lie bracket [ ⋅ , ⋅ ] [\cdot,\cdot] [ ⋅ , ⋅ ] on the sections, with an anchor # : E → T B \# : E \to TB # : E → T B . An anchor is a bundle map # : E → T B \# : E \to TB # : E → T B that preserves the Lie brackets, and the bracket satisfies the Leibniz identity
[ a , f b ] = f [ a , b ] + # a ( f ) b , a , b ∈ Γ ( E ) , f ∈ C ∞ ( B ) . [a,fb]=f[a,b]+\#a(f)b,\qquad a,b\in\Gamma(E),\qquad f\in C^{\infty}(B). [ a , f b ] = f [ a , b ] + # a ( f ) b , a , b ∈ Γ ( E ) , f ∈ C ∞ ( B ) .
A Lie algebroid is called transitive if its anchor is surjective.
Example 2.8. As the bracket of invariant vector fields on a twisted principal G G G -bundle M → B M \to B M → B is again invariant, the sections Γ ( T M / G ) \Gamma(TM/G) Γ ( T M / G ) inherit the Lie bracket of vector fields on M M M . The vector bundle T M / G TM/G T M / G equipped with this Lie bracket and the anchor π ∗ : T M / G → T B \pi_{*}:TM/G \to TB π ∗ : T M / G → T B is a Lie algebroid called the Atiyah algebroid ( T M / G , B , [ ⋅ , ⋅ ] , π ∗ ) (TM/G,B,[\cdot,\cdot],\pi_{*}) ( T M / G , B , [ ⋅ , ⋅ ] , π ∗ ) .
Definition 2.9. Let ( E 1 , B , [ ⋅ , ⋅ ] 1 , # 1 ) (E_{1},B,[\cdot,\cdot]_{1},\#_{1}) ( E 1 , B , [ ⋅ , ⋅ ] 1 , # 1 ) and ( E 2 , B , [ ⋅ , ⋅ ] 2 , # 2 ) (E_{2},B,[\cdot,\cdot]_{2},\#_{2}) ( E 2 , B , [ ⋅ , ⋅ ] 2 , # 2 ) be two Lie algebroids over the same base. A morphism of Lie algebroids is a vector bundle map Φ : E 1 → E 2 \Phi:E_{1}\to E_{2} Φ : E 1 → E 2 over the identity, such that
# 2 ∘ Φ = # 1 , Φ [ a , b ] 1 = [ Φ ( a ) , Φ ( b ) ] 2 , \#_{2}\circ\Phi=\#_{1},\qquad\Phi[a,b]_{1}=[\Phi(a),\Phi(b)]_{2}, # 2 ∘ Φ = # 1 , Φ [ a , b ] 1 = [ Φ ( a ) , Φ ( b ) ] 2 ,
for all a , b ∈ Γ ( E 1 ) a,b\in\Gamma(E_{1}) a , b ∈ Γ ( E 1 ) . An isomorphism of Lie algebroids is a bijective morphism of Lie algebroids.
The kernel of the anchor π ∗ \pi_{*} π ∗ of the Atiyah algebroid consists of the invariant vertical vector fields. Hence, Atiyah bundles fit into the short exact sequence of vector bundles
0 → ad M → ι T M / G → π ∗ T B → 0. (19) 0\to\operatorname{ad}M\xrightarrow{\iota}TM/G\xrightarrow{\pi_{*}}TB\to0. \tag*{(19)} 0 → ad M ι T M / G π ∗ T B → 0. (19)
A principal connection is a splitting σ : T B → T M / G \sigma:TB\to TM/G σ : T B → T M / G of the sequence (19) with π ∗ ∘ σ = Id T B \pi_{*}\circ\sigma=\operatorname{Id}_{TB} π ∗ ∘ σ = Id T B , or equivalently a vector bundle map ω : T M / G → ad M \omega:TM/G\to\operatorname{ad}M ω : T M / G → ad M with ω ∘ ι = Id ad M \omega\circ\iota=\operatorname{Id}_{\operatorname{ad}M} ω ∘ ι = Id ad M . The two descriptions are related by
ι ∘ ω + σ ∘ π ∗ = Id T M / G , ω ∘ σ = 0. \iota\circ\omega+\sigma\circ\pi_{*}=\operatorname{Id}_{TM/G},\qquad\omega\circ\sigma=0. ι ∘ ω + σ ∘ π ∗ = Id T M / G , ω ∘ σ = 0.
This is the decomposition of a vector field into its vertical and horizontal components:
U = U V + U H = V U + H U = ( ι ∘ ω ) ( U ) + ( σ ∘ π ∗ ) ( U ) . U=U^{V}+U^{H}=\mathcal{V}U+\mathcal{H}U=(\iota\circ\omega)(U)+(\sigma\circ\pi_{*})(U). U = U V + U H = V U + H U = ( ι ∘ ω ) ( U ) + ( σ ∘ π ∗ ) ( U ) .
The affine connection ∇ a d \nabla^{\mathrm{ad}} ∇ ad defined in (9) is the mixed term of the Lie algebroid bracket on T M / G TM/G T M / G :
[ U a d + X ‾ , V a d + Y ‾ ] = [ U a d , V a d ] + ∇ X ‾ a d V a d − ∇ Y ‾ a d U a d + [ X ‾ , Y ‾ ] . [U_{\mathrm{ad}}+\overline{X},V_{\mathrm{ad}}+\overline{Y}]=[U_{\mathrm{ad}},V_{\mathrm{ad}}]+\nabla^{\mathrm{ad}}_{\overline{X}}V_{\mathrm{ad}}-\nabla^{\mathrm{ad}}_{\overline{Y}}U_{\mathrm{ad}}+[\overline{X},\overline{Y}]. [ U ad + X , V ad + Y ] = [ U ad , V ad ] + ∇ X ad V ad − ∇ Y ad U ad + [ X , Y ] .
With respect to the splitting, the last term is
[ X ‾ , Y ‾ ] = σ [ X , Y ] − ι F X , Y = [ X ‾ , Y ‾ ] + 2 ι Λ X , Y . [\overline{X},\overline{Y}]=\sigma[X,Y]-\iota F_{X,Y}=[\overline{X},\overline{Y}]+2\iota\Lambda_{X,Y}. [ X , Y ] = σ [ X , Y ] − ι F X , Y = [ X , Y ] + 2 ι Λ X , Y .
Definition 2.10. A Riemannian Lie algebroid ( E , B , [ ⋅ , ⋅ ] , # , g ) (E,B,[\cdot,\cdot],\#,g) ( E , B , [ ⋅ , ⋅ ] , # , g ) is a Lie algebroid with a vector bundle metric g g g on E → B E \to B E → B .
Definition 2.11. An isometric isomorphism of Riemannian Lie algebroids is an isomorphism of Lie algebroids that is an isometry of metric vector bundles.
As invariant Riemannian metrics on twisted principal bundles are determined entirely by their evaluations along invariant vector fields, we have the bijection
{ invariant Riemannian metrics g on M } ⟷ { vector bundle metrics on T M / G } . (20) \{\text{invariant Riemannian metrics }g\text{ on }M\}\longleftrightarrow\{\text{vector bundle metrics on }TM/G\}. \tag*{(20)} { invariant Riemannian metrics g on M } ⟷ { vector bundle metrics on T M / G } . (20)
Next, we describe the local geometry of Riemannian Lie algebroids. A transitive Lie algebroid ( E , B , [ ⋅ , ⋅ ] , # ) (E,B,[\cdot,\cdot],\#) ( E , B , [ ⋅ , ⋅ ] , # ) has surjective anchor, and thus there is a Lie algebra bundle ker # → B \ker\#\to B ker # → B and a short exact sequence
0 → ker # → ι E → # T B → 0. 0\to\ker\#\xrightarrow{\iota}E\xrightarrow{\#}TB\to0. 0 → ker # ι E # T B → 0.
A splitting of this sequence σ \sigma σ and its vertical projection ω \omega ω induce an affine connection on ker # \ker\# ker # and a curvature
∇ X ker U ker = ω [ σ X , ι U ker ] , F X , Y = − ω [ σ X , σ Y ] , X ∈ X ( B ) , U ker ∈ Γ ( ker # ) . \nabla^{\ker}_{X}U_{\ker}=\omega[\sigma X,\iota U_{\ker}],\qquad F_{X,Y}=-\omega[\sigma X,\sigma Y],\qquad X\in\mathfrak{X}(B),\quad U_{\ker}\in\Gamma(\ker\#). ∇ X k e r U k e r = ω [ σ X , ι U k e r ] , F X , Y = − ω [ σ X , σ Y ] , X ∈ X ( B ) , U k e r ∈ Γ ( ker # ) .
Define the deformation coboundary operator [9 Equ. 1]
δ [ ⋅ , ⋅ ] ( α ) ( U , V ) = [ α U , V ] + [ U , α V ] − α [ U , V ] , α ∈ Γ ( End ( E ) ) , U , V ∈ Γ ( E ) . \delta_{[\cdot,\cdot]}(\alpha)(U,V)=[\alpha U,V]+[U,\alpha V]-\alpha[U,V],\qquad\alpha\in\Gamma(\operatorname{End}(E)),\quad U,V\in\Gamma(E). δ [ ⋅ , ⋅ ] ( α ) ( U , V ) = [ α U , V ] + [ U , α V ] − α [ U , V ] , α ∈ Γ ( End ( E )) , U , V ∈ Γ ( E ) .
This operator is skew-symmetric, R \mathbb{R} R -linear, and
δ [ ⋅ , ⋅ ] ( α ) ( U , f V ) = f δ [ ⋅ , ⋅ ] ( α ) ( U , V ) + ( # α U ) ( f ) V , f ∈ C ∞ ( B ) . \delta_{[\cdot,\cdot]}(\alpha)(U,fV)=f\delta_{[\cdot,\cdot]}(\alpha)(U,V)+(\#\alpha U)(f)V,\qquad f\in C^\infty(B). δ [ ⋅ , ⋅ ] ( α ) ( U , f V ) = f δ [ ⋅ , ⋅ ] ( α ) ( U , V ) + ( # α U ) ( f ) V , f ∈ C ∞ ( B ) .
If # ∘ α = 0 \#\circ\alpha=0 # ∘ α = 0 , then δ [ ⋅ , ⋅ ] ( α ) \delta_{[\cdot,\cdot]}(\alpha) δ [ ⋅ , ⋅ ] ( α ) is C ∞ ( B ) C^\infty(B) C ∞ ( B ) -linear. Vector bundle metrics g g g on E E E uniquely decompose as the same metric data ( h , ω , g ˇ ) (h,\omega,\check{g}) ( h , ω , g ˇ ) as in subsection 2.2, with ker # \ker\# ker # instead of ad M \operatorname{ad}M ad M :
g = h ( ω ⋅ , ω ⋅ ) + g ˇ ( # ⋅ , # ⋅ ) . g=h(\omega\cdot,\omega\cdot)+\check{g}(\#\cdot,\#\cdot). g = h ( ω ⋅ , ω ⋅ ) + g ˇ ( # ⋅ , # ⋅ ) .
Given a vector bundle metric on a transitive Lie algebroid E → B E\to B E → B , there is a unique connection, ∇ : Γ ( E ) × Γ ( E ) → Γ ( E ) \nabla:\Gamma(E)\times\Gamma(E)\to\Gamma(E) ∇ : Γ ( E ) × Γ ( E ) → Γ ( E ) satisfying
∇ f a b = f ∇ a b , ∇ a ( f b ) = f ∇ a b + # a ( f ) b , a , b ∈ Γ ( E ) , f ∈ C ∞ ( B ) , \nabla_{fa}b=f\nabla_a b,\qquad\nabla_a(fb)=f\nabla_a b+\#a(f)b,\qquad a,b\in\Gamma(E),\quad f\in C^\infty(B), ∇ f a b = f ∇ a b , ∇ a ( f b ) = f ∇ a b + # a ( f ) b , a , b ∈ Γ ( E ) , f ∈ C ∞ ( B ) ,
that is metric and torsion-free [5 p.62] . It is characterised by Koszul’s formula:
2 g ( ∇ a b , c ) = # a ( g ( b , c ) ) + # b ( g ( a , c ) ) − # c ( g ( a , b ) ) + g ( [ a , b ] , c ) − g ( [ b , c ] , a ) + g ( [ c , a ] , b ) , \begin{aligned}
2g(\nabla_a b,c)={}&\#a(g(b,c))+\#b(g(a,c))-\#c(g(a,b))\\
&+g([a,b],c)-g([b,c],a)+g([c,a],b),
\end{aligned} 2 g ( ∇ a b , c ) = # a ( g ( b , c )) + # b ( g ( a , c )) − # c ( g ( a , b )) + g ([ a , b ] , c ) − g ([ b , c ] , a ) + g ([ c , a ] , b ) ,
for a , b , c ∈ Γ ( E ) a,b,c\in\Gamma(E) a , b , c ∈ Γ ( E ) . Its curvature and Ricci tensor are given by
R ( a , b ) c = ∇ a ∇ b c − ∇ b ∇ a c − ∇ [ a , b ] c , ric g ( a , b ) = tr ( c ↦ R ( c , a ) b ) . R(a,b)c=\nabla_a\nabla_b c-\nabla_b\nabla_a c-\nabla_{[a,b]}c,\qquad\operatorname{ric}_g(a,b)=\operatorname{tr}(c\mapsto R(c,a)b). R ( a , b ) c = ∇ a ∇ b c − ∇ b ∇ a c − ∇ [ a , b ] c , ric g ( a , b ) = tr ( c ↦ R ( c , a ) b ) .
Write X ˉ = σ X \bar{X}=\sigma X X ˉ = σ X and U = ι U ker U=\iota U_{\ker} U = ι U k e r for X ∈ X ( B ) X\in\mathfrak{X}(B) X ∈ X ( B ) and U ker ∈ Γ ( ker # ) U_{\ker}\in\Gamma(\ker\#) U k e r ∈ Γ ( ker # ) . Denote the induced metric connection on ker # \ker\# ker # also by ∇ \nabla ∇ . As in subsection 2.2,
∇ X U ker : = ω ∇ X ˉ U , ∇ ^ U ker V ker : = ω ∇ U V , L X U ker : = ω ∇ U X ˉ . \nabla_XU_{\ker}:=\omega\nabla_{\bar{X}}U,\qquad\widehat{\nabla}_{U_{\ker}}V_{\ker}:=\omega\nabla_UV,\qquad L_XU_{\ker}:=\omega\nabla_U\bar{X}. ∇ X U k e r := ω ∇ X ˉ U , ∇ U k e r V k e r := ω ∇ U V , L X U k e r := ω ∇ U X ˉ .
Torsion freeness gives ∇ ker = ∇ − L \nabla^{\ker}=\nabla-L ∇ k e r = ∇ − L , and O’Neill’s integrability tensor is again A X , Y = − 1 2 F X , Y A_{X,Y}=-\frac{1}{2}F_{X,Y} A X , Y = − 2 1 F X , Y . The following proposition follows directly from Crainic–Fernandes [6 ] and Lie’s third theorem. The correspondence of metrics, Levi-Civita connections, and curvature follows from Koszul’s formula.
Proposition 2.12 (Local principal bundle realisation). Let ( E , B , [ ⋅ , ⋅ ] , # ) (E,B,[\cdot,\cdot],\#) ( E , B , [ ⋅ , ⋅ ] , # ) be a transitive Lie algebroid equipped with a vector bundle metric g g g . Let U ⊆ B U\subseteq B U ⊆ B be open and contractible. There is a simply-connected Lie group G G G and an isomorphism of Lie algebroids over U U U ,
E U ≅ T ( U × G ) / G E_U\cong T(U\times G)/G E U ≅ T ( U × G ) / G under which g ∣ U g|_U g ∣ U corresponds to an invariant Riemannian metric on the principal bundle U × G → U U\times G\to U U × G → U . The induced Levi-Civita connection, curvature and Ricci tensor agree with their definitions on E U E_U E U , that is, they do not depend on the choice of realisation.
Definition 2.13 (Deformations of Riemannian Lie algebroids (cf. [9 Def. 1] ).) Fix a vector bundle E → B E \to B E → B . A family of Riemannian Lie algebroids over an interval I ⊆ R I \subseteq\mathbb{R} I ⊆ R is a collection of Riemannian Lie algebroids
( E , B , [ ⋅ , ⋅ ] t , # t , g t ) (E,B,[\cdot,\cdot]_{t},\#_{t},g_{t}) ( E , B , [ ⋅ , ⋅ ] t , # t , g t ) varying smoothly with t ∈ I t \in I t ∈ I . A family of isometric isomorphisms of Riemannian Lie algebroids
Φ ( t ) : ( E , B , [ ⋅ , ⋅ ] t , # t , g t ) → ( E ′ , B ′ , [ ⋅ , ⋅ ] t ′ , # t ′ , g t ′ ) \Phi(t):(E,B,[\cdot,\cdot]_{t},\#_{t},g_{t})\to(E',B',[\cdot,\cdot]'_{t},\#'_{t},g'_{t}) Φ ( t ) : ( E , B , [ ⋅ , ⋅ ] t , # t , g t ) → ( E ′ , B ′ , [ ⋅ , ⋅ ] t ′ , # t ′ , g t ′ ) varying smoothly with t t t is called an equivalence. If an equivalence between two families of Riemannian Lie algebroids exists, then they are equivalent. A deformation of a Riemannian Lie algebroid ( E , B , [ ⋅ , ⋅ ] , # , g ) (E,B,[\cdot,\cdot],\#,g) ( E , B , [ ⋅ , ⋅ ] , # , g ) is a family of Riemannian Lie algebroids over an interval I I I containing 0 0 0 with initial value
( E , B , [ ⋅ , ⋅ ] 0 , # 0 , g 0 ) = ( E , B , [ ⋅ , ⋅ ] , # , g ) . (E,B,[\cdot,\cdot]_{0},\#_{0},g_{0})=(E,B,[\cdot,\cdot],\#,g). ( E , B , [ ⋅ , ⋅ ] 0 , # 0 , g 0 ) = ( E , B , [ ⋅ , ⋅ ] , # , g ) .
Example 2.14.
(i) A Ricci flow g ( t ) g(t) g ( t ) on a smooth manifold M M M is a family of Riemannian Lie algebroids with fixed bracket and fixed anchor:
( T M , M , [ ⋅ , ⋅ ] , Id T M , g ( t ) ) (TM,M,[\cdot,\cdot],\operatorname{Id}_{TM},g(t)) ( T M , M , [ ⋅ , ⋅ ] , Id T M , g ( t )) (ii) The Ricci flow of left-invariant metrics on a Lie group with initial metric g 0 g_{0} g 0 is a deformation of a Riemannian Lie algebroid with fixed bracket and fixed anchor
( g , ∗ , [ ⋅ , ⋅ ] , ∗ , ⟨ ⋅ , ⋅ ⟩ t ) . (\mathfrak{g},\ast,[\cdot,\cdot],\ast,\langle\cdot,\cdot\rangle_{t}). ( g , ∗ , [ ⋅ , ⋅ ] , ∗ , ⟨ ⋅ , ⋅ ⟩ t ) . Lauret’s bracket flow is another deformation of the same Riemannian Lie algebroid but instead with fixed metric and fixed anchor:
( g , ∗ , μ ( t ) , ∗ , ⟨ ⋅ , ⋅ ⟩ 0 ) . (\mathfrak{g},\ast,\mu(t),\ast,\langle\cdot,\cdot\rangle_{0}). ( g , ∗ , μ ( t ) , ∗ , ⟨ ⋅ , ⋅ ⟩ 0 ) . These two families of Riemannian Lie algebroids are equivalent [21 Thm. 3.3] . The equivalence is constructed by solving either ODE,
d Φ d t = − Φ Ric g , d Φ d t = − Ric μ Φ , \frac{d\Phi}{dt}=-\Phi\operatorname{Ric}_{g},\qquad\frac{d\Phi}{dt}=-\operatorname{Ric}_{\mu}\Phi, d t d Φ = − Φ Ric g , d t d Φ = − Ric μ Φ , with initial value Φ ( 0 ) = Id \Phi(0)=\operatorname{Id} Φ ( 0 ) = Id .
(iii) The Ricci flow of invariant metrics on a twisted principal G-bundle M → B M \to B M → B with initial metric g 0 g_{0} g 0 is a deformation of a Riemannian Lie algebroid with fixed bracket and fixed anchor:
( T M / G , B , [ ⋅ , ⋅ ] , π ∗ , g ( t ) ) . (TM/G,B,[\cdot,\cdot],\pi_{*},g(t)). ( T M / G , B , [ ⋅ , ⋅ ] , π ∗ , g ( t )) .
Bracket flows in higher cohomogeneity
In this section, we prove Theorem B. First, we write the Ricci flow of invariant metrics as a system of partial differential equations for the reduced data triple ( h , ω , g ˘ ) (h,\omega,\breve{g}) ( h , ω , g ˘ ) (8) . We define the bracket flow on twisted principal bundles and prove its equivalence to the Ricci flow as deformations of Lie algebroids. In the final subsection, we discuss moment maps on the adjoint bundle, in preparation for the proof of Theorem A.
Ricci flow equations
Let M → B M \to B M → B be a twisted principal G-bundle. Equip M M M with an invariant Riemannian metric g g g . Define the Ricci tensor and the Ricci endomorphism,
ric g : T M / G × T M / G → R , Ric g : T M / G → T M / G . \operatorname{ric}_{g}:TM/G\times TM/G\to\mathbb{R},\qquad\operatorname{Ric}_{g}:TM/G\to TM/G. ric g : T M / G × T M / G → R , Ric g : T M / G → T M / G .
With respect to the splitting given by the orthogonal complement, write
ric g a d : ad M × ad M → R , ric g a d ( U a d , U a d ) = ric g ( U , U ) . (21) \operatorname{ric}^{\mathrm{ad}}_{g}:\operatorname{ad}M\times\operatorname{ad}M\to\mathbb{R},\qquad\operatorname{ric}^{\mathrm{ad}}_{g}(U_{\mathrm{ad}},U_{\mathrm{ad}})=\operatorname{ric}_{g}(U,U). \tag*{(21)} ric g ad : ad M × ad M → R , ric g ad ( U ad , U ad ) = ric g ( U , U ) . (21)
ric g T B : T B × T B → R , ric g T B ( X , X ) = ric g ( X ‾ , X ‾ ) . (22) \operatorname{ric}^{TB}_{g}:TB\times TB\to\mathbb{R},\qquad\operatorname{ric}^{TB}_{g}(X,X)=\operatorname{ric}_{g}(\overline{X},\overline{X}). \tag*{(22)} ric g T B : T B × T B → R , ric g T B ( X , X ) = ric g ( X , X ) . (22)
Let V , H : T M / G → T M / G \mathcal{V},\mathcal{H}:TM/G\to TM/G V , H : T M / G → T M / G be the projections of invariant vector fields into their vertical and horizontal components, as given in § 2.2 . Decompose the Ricci endomorphism as the following endomorphisms of T M / G TM/G T M / G :
Ric g a d = V Ric g V , Ric g a d , T B = V Ric g H , Ric g T B , a d = H Ric g V , Ric g T B = H Ric g H . (23) \operatorname{Ric}^{\mathrm{ad}}_{g}=\mathcal{V}\operatorname{Ric}_{g}\mathcal{V},\qquad\operatorname{Ric}^{\mathrm{ad},TB}_{g}=\mathcal{V}\operatorname{Ric}_{g}\mathcal{H},\qquad\operatorname{Ric}^{TB,\mathrm{ad}}_{g}=\mathcal{H}\operatorname{Ric}_{g}\mathcal{V},\qquad\operatorname{Ric}^{TB}_{g}=\mathcal{H}\operatorname{Ric}_{g}\mathcal{H}. \tag*{(23)} Ric g ad = V Ric g V , Ric g ad , T B = V Ric g H , Ric g T B , ad = H Ric g V , Ric g T B = H Ric g H . (23)
Proposition 3.1. On twisted principal bundles, the Ricci flow of invariant metrics (1) is equivalent to the following system of partial differential equations for the triple ( h , ω , g ˇ ) (h,\omega,\check{g}) ( h , ω , g ˇ ) :
∂ h ∂ t = − 2 ric g a d , (24) \frac{\partial h}{\partial t}=-2\operatorname{ric}^{\mathrm{ad}}_{g}, \tag*{(24)} ∂ t ∂ h = − 2 ric g ad , (24) ∂ ω ∂ t = − 2 ω Ric g a d , T B , (25) \frac{\partial\omega}{\partial t}=-2\omega\operatorname{Ric}^{\mathrm{ad},TB}_{g}, \tag*{(25)} ∂ t ∂ ω = − 2 ω Ric g ad , T B , (25) ∂ g ˇ ∂ t = − 2 ric g T B . (26) \frac{\partial\check{g}}{\partial t}=-2\operatorname{ric}^{TB}_{g}. \tag*{(26)} ∂ t ∂ g ˇ = − 2 ric g T B . (26) where h h h is a vector bundle metric on ad M \operatorname{ad}M ad M , ω \omega ω is a principal connection, and g ˇ \check{g} g ˇ is a Riemannian metric on the base B B B (8).
Proof. Equations (24) and (26) are the restrictions of the Ricci flow equation to vertical and horizontal vectors. For equation (25), let g ( t ) g(t) g ( t ) be a Ricci flow of invariant metrics. Take a vector field X X X on B B B and take its horizontal lift X ˉ ( t ) \bar{X}(t) X ˉ ( t ) . For all fixed vertical vectors U U U , we know that g ( t ) ( U , X ˉ ( t ) ) = 0 g(t)(U,\bar{X}(t))=0 g ( t ) ( U , X ˉ ( t )) = 0 . Differentiating this we get
0 = − 2 ric g ( U , X ˉ ) + g ( U , ∂ t X ˉ ) . 0=-2\operatorname{ric}_{g}(U,\bar{X})+g(U,\partial_t\bar{X}). 0 = − 2 ric g ( U , X ˉ ) + g ( U , ∂ t X ˉ ) . By definition ω ( t ) ( X ˉ ( t ) ) = 0 \omega(t)(\bar{X}(t))=0 ω ( t ) ( X ˉ ( t )) = 0 . Differentiating this we get
0 = ∂ ω ∂ t ( X ˉ ) + ω ( ∂ t X ˉ ) = ∂ ω ∂ t ( X ˉ ) + 2 ω ( Ric g X ˉ ) , 0=\frac{\partial\omega}{\partial t}(\bar{X})+\omega(\partial_t\bar{X})=\frac{\partial\omega}{\partial t}(\bar{X})+2\omega(\operatorname{Ric}_{g}\bar{X}), 0 = ∂ t ∂ ω ( X ˉ ) + ω ( ∂ t X ˉ ) = ∂ t ∂ ω ( X ˉ ) + 2 ω ( Ric g X ˉ ) , while ω \omega ω restricted to vertical vector fields remains unchanged under the Ricci flow. This demonstrates equation (25). Conversely, differentiating
g = ω ∗ h + π ∗ g ˇ = h ( ω ⋅ , ω ⋅ ) + g ˇ ( π ∗ ⋅ , π ∗ ⋅ ) g=\omega^*h+\pi^*\check{g}=h(\omega\cdot,\omega\cdot)+\check{g}(\pi_*\cdot,\pi_*\cdot) g = ω ∗ h + π ∗ g ˇ = h ( ω ⋅ , ω ⋅ ) + g ˇ ( π ∗ ⋅ , π ∗ ⋅ ) recovers the Ricci flow equation. □
Bracket flow equations
Let M → B M\to B M → B be a twisted principal G-bundle. As in 2.1, given a Ricci flow of invariant metrics g ( t ) g(t) g ( t ) on M M M , we apply Uhlenbeck’s trick in the vertical directions by choosing a moving local frame { U j ( t ) } \{U_j(t)\} { U j ( t )} of the adjoint bundle that solves
∂ t U j = Ric g a d ( U j ) . \partial_tU_j=\operatorname{Ric}^{\mathrm{ad}}_{g}(U_j). ∂ t U j = Ric g ad ( U j ) .
If { U i ( t ) } \{U_i(t)\} { U i ( t )} is initially orthonormal, then it remains orthonormal along the Ricci flow. Indeed, the metric coefficients of the vertical directions are constant,
∂ ∂ t g ( U i , U j ) = ∂ g ∂ t ( U i , U j ) + g ( ∂ t U i , U j ) + g ( U i , ∂ t U j ) = − 2 ric g ( U i , U j ) + g ( Ric g a d ( U i ) , U j ) + g ( U i , Ric g a d ( U j ) ) = 0. \begin{aligned}
\frac{\partial}{\partial t}g(U_i,U_j)=\frac{\partial g}{\partial t}(U_i,U_j)+g(\partial_tU_i,U_j)+g(U_i,\partial_tU_j) \\
&=-2\operatorname{ric}_{g}(U_i,U_j)+g(\operatorname{Ric}^{\mathrm{ad}}_{g}(U_i),U_j)+g(U_i,\operatorname{Ric}^{\mathrm{ad}}_{g}(U_j)) \\
&=0.
\end{aligned} ∂ t ∂ g ( U i , U j ) = ∂ t ∂ g ( U i , U j ) + g ( ∂ t U i , U j ) + g ( U i , ∂ t U j ) = − 2 ric g ( U i , U j ) + g ( Ric g ad ( U i ) , U j ) + g ( U i , Ric g ad ( U j )) = 0.
At t = 0 t=0 t = 0 , let X ˉ i \bar{X}_i X ˉ i be a local orthonormal horizontal frame. Write
[ U i + X ˉ i , U j + X ˉ j ] = [ U i , U j ] + ∇ X ˉ i a d U j − ∇ X ˉ j a d U i + [ X ˉ i , X ˉ j ] = c i j k ( X i , X j ) U k + [ X i , X j ] , \begin{aligned}
[U_i+\bar{X}_i,U_j+\bar{X}_j]=[U_i,U_j]+\nabla^{\mathrm{ad}}_{\bar{X}_i}U_j-\nabla^{\mathrm{ad}}_{\bar{X}_j}U_i+[\bar{X}_i,\bar{X}_j] \\
&=c^k_{ij}(X_i,X_j)U_k+[X_i,X_j],
\end{aligned} [ U i + X ˉ i , U j + X ˉ j ] = [ U i , U j ] + ∇ X ˉ i ad U j − ∇ X ˉ j ad U i + [ X ˉ i , X ˉ j ] = c ij k ( X i , X j ) U k + [ X i , X j ] ,
where c i j k c^k_{ij} c ij k are the vertical structure coefficients of the Lie algebroid bracket in the moving frame { X ˉ i , U i ( t ) } \{\bar{X}_i,U_i(t)\} { X ˉ i , U i ( t )} . They are given by the following sum of structure coefficients of the adjoint bundle μ i j k \mu^k_{ij} μ ij k , connection coefficients ω i k ∈ Ω 1 ( B ) \omega^k_i\in\Omega^1(B) ω i k ∈ Ω 1 ( B ) , and curvature coefficients F k ∈ Ω 2 ( B ) F^k\in\Omega^2(B) F k ∈ Ω 2 ( B ) :
c i j k ( X i , X j ) = μ i j k + ω j k ( X i ) − ω i k ( X j ) − F k ( X i , X j ) . c^k_{ij}(X_i,X_j)=\mu^k_{ij}+\omega^k_j(X_i)-\omega^k_i(X_j)-F^k(X_i,X_j). c ij k ( X i , X j ) = μ ij k + ω j k ( X i ) − ω i k ( X j ) − F k ( X i , X j ) .
The time derivative of the vertical structure coefficients in this choice of moving frame is
∂ c i j k ∂ t ( X i , X j ) U k = [ Ric g a d U i , U j + X ˉ j ] + [ U i + X ˉ i , Ric g a d U j ] − Ric g a d [ U i + X ˉ i , U j + X ˉ j ] = δ [ ⋅ , ⋅ ] ( Ric g a d ) ( U i + X ˉ i , U j + X ˉ j ) . \begin{aligned}
\frac{\partial c^k_{ij}}{\partial t}(X_i,X_j)U_k=[\operatorname{Ric}^{\mathrm{ad}}_{g}U_i,U_j+\bar{X}_j]+[U_i+\bar{X}_i,\operatorname{Ric}^{\mathrm{ad}}_{g}U_j]-\operatorname{Ric}^{\mathrm{ad}}_{g}[U_i+\bar{X}_i,U_j+\bar{X}_j] \\
&=\delta_{[\cdot,\cdot]}(\operatorname{Ric}^{\mathrm{ad}}_{g})(U_i+\bar{X}_i,U_j+\bar{X}_j).
\end{aligned} ∂ t ∂ c ij k ( X i , X j ) U k = [ Ric g ad U i , U j + X ˉ j ] + [ U i + X ˉ i , Ric g ad U j ] − Ric g ad [ U i + X ˉ i , U j + X ˉ j ] = δ [ ⋅ , ⋅ ] ( Ric g ad ) ( U i + X ˉ i , U j + X ˉ j ) .
Definition 3.2 (Bracket Flow). Let π : M → B \pi:M\to B π : M → B be a twisted principal G G G -bundle equipped with an invariant Riemannian metric g 0 g_0 g 0 . Write g 0 = ω 0 ∗ h 0 + π ∗ g ˘ 0 g_0=\omega_0^*h_0+\pi^*\breve{g}_0 g 0 = ω 0 ∗ h 0 + π ∗ g ˘ 0 where h 0 h_0 h 0 is a vector bundle metric on the adjoint bundle ad M → B \operatorname{ad}M\to B ad M → B , ω 0 \omega_0 ω 0 is a principal connection, and g ˘ 0 \breve{g}_0 g ˘ 0 is a Riemannian metric on the base B B B (see (8) , (20) ). A bracket flow is a deformation
( T M / G , B , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t ) ) , (27) (TM/G,B,[\cdot,\cdot]_{\mu(t)},\pi_*,g_{\mu(t)}), \tag*{(27)} ( T M / G , B , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t ) ) , (27) of the Riemannian Lie algebroid ( T M / G , B , [ ⋅ , ⋅ ] , π ∗ , g 0 ) (TM/G,B,[\cdot,\cdot],\pi_*,g_0) ( T M / G , B , [ ⋅ , ⋅ ] , π ∗ , g 0 ) with fixed anchor π ∗ \pi_* π ∗ , such that the metric
g μ ( t ) = ω μ ( t ) ∗ h 0 + π ∗ g ˘ μ ( t ) g_{\mu(t)}=\omega_{\mu(t)}^*h_0+\pi^*\breve{g}_{\mu(t)} g μ ( t ) = ω μ ( t ) ∗ h 0 + π ∗ g ˘ μ ( t ) and the bracket [ ⋅ , ⋅ ] μ ( t ) [\cdot,\cdot]_{\mu(t)} [ ⋅ , ⋅ ] μ ( t ) solve the following system of partial differential equations
∂ [ ⋅ , ⋅ ] μ ∂ t = δ [ ⋅ , ⋅ ] μ ( Ric g μ a d ) (28) \frac{\partial[\cdot,\cdot]_{\mu}}{\partial t}=\delta_{[\cdot,\cdot]_{\mu}}\left(\operatorname{Ric}^{\mathrm{ad}}_{g_{\mu}}\right) \tag*{(28)} ∂ t ∂ [ ⋅ , ⋅ ] μ = δ [ ⋅ , ⋅ ] μ ( Ric g μ ad ) (28) ∂ ω μ ∂ t = − 2 ω μ Ric g μ a d , T B (29) \frac{\partial\omega_{\mu}}{\partial t}=-2\omega_{\mu}\operatorname{Ric}^{\mathrm{ad},TB}_{g_{\mu}} \tag*{(29)} ∂ t ∂ ω μ = − 2 ω μ Ric g μ ad , T B (29) ∂ g ˘ μ ∂ t = − 2 ric g μ T B . (30) \frac{\partial\breve{g}_{\mu}}{\partial t}=-2\operatorname{ric}^{TB}_{g_{\mu}}. \tag*{(30)} ∂ t ∂ g ˘ μ = − 2 ric g μ T B . (30) Here, ric g μ ( t ) \operatorname{ric}_{g_{\mu(t)}} ric g μ ( t ) and Ric g μ ( t ) \operatorname{Ric}_{g_{\mu(t)}} Ric g μ ( t ) are the Ricci tensor and Ricci endomorphism of the evolving Riemannian Lie algebroids ( T M / G , B , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t ) ) (TM/G,B,[\cdot,\cdot]_{\mu(t)},\pi_*,g_{\mu(t)}) ( T M / G , B , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t ) ) , and the superscripts, a d \mathrm{ad} ad and T B TB T B , denote the projections with respect to splittings by orthogonal complements of g μ ( t ) g_{\mu(t)} g μ ( t ) (21–23).
Now, we prove that the bracket flow is equivalent to the Ricci flow as deformations of Riemannian Lie algebroids in the sense of Theorem 2.13.
Theorem 3.4 (Theorem B). Let π : M → B \pi:M\to B π : M → B be a twisted principal G G G -bundle equipped with an invariant metric g 0 = ( h 0 , ω 0 , g ˘ 0 ) g_0=(h_0,\omega_0,\breve{g}_0) g 0 = ( h 0 , ω 0 , g ˘ 0 ) . Ricci flows of invariant metrics on M M M with initial metric g 0 g_0 g 0 are in one-to-one correspondence with bracket flows (Theorem 3.2) with the same initial data. There exists a unique smooth family of vector bundle automorphisms Φ ( t ) \Phi(t) Φ ( t ) of T M / G TM/G T M / G over the identity that preserve the anchor and satisfy
g ( t ) = Φ ( t ) ∗ g μ ( t ) . g(t)=\Phi(t)^*g_{\mu(t)}. g ( t ) = Φ ( t ) ∗ g μ ( t ) . The vector bundle automorphisms solve the fibrewise ODEs
(i) ∂ ∂ t Φ = − Φ Ric g a d , Φ ( 0 ) = Id , \text{(i)}\quad\frac{\partial}{\partial t}\Phi=-\Phi\operatorname{Ric}^{\mathrm{ad}}_g,\qquad\Phi(0)=\operatorname{Id}, (i) ∂ t ∂ Φ = − Φ Ric g ad , Φ ( 0 ) = Id , (ii) ∂ ∂ t Φ = − Ric g μ a d Φ , Φ ( 0 ) = Id . \text{(ii)}\quad\frac{\partial}{\partial t}\Phi=-\operatorname{Ric}^{\mathrm{ad}}_{g_{\mu}}\Phi,\qquad\Phi(0)=\operatorname{Id}. (ii) ∂ t ∂ Φ = − Ric g μ ad Φ , Φ ( 0 ) = Id . Furthermore, the following conditions hold
(iii) h = h 0 ( Φ a d ⋅ , Φ a d ⋅ ) , \text{(iii)}\quad h=h_0(\Phi^{\mathrm{ad}}\cdot,\Phi^{\mathrm{ad}}\cdot), (iii) h = h 0 ( Φ ad ⋅ , Φ ad ⋅ ) , (iv) [ ⋅ , ⋅ ] μ = Φ [ Φ − 1 ⋅ , Φ − 1 ⋅ ] , \text{(iv)}\quad[\cdot,\cdot]_{\mu}=\Phi[\Phi^{-1}\cdot,\Phi^{-1}\cdot], (iv) [ ⋅ , ⋅ ] μ = Φ [ Φ − 1 ⋅ , Φ − 1 ⋅ ] , (v) ω μ = Φ a d ∘ ω ∘ Φ − 1 , \text{(v)}\quad\omega_{\mu}=\Phi^{\mathrm{ad}}\circ\omega\circ\Phi^{-1}, (v) ω μ = Φ ad ∘ ω ∘ Φ − 1 , (vi) g ˘ μ = g ˘ , \text{(vi)}\quad\breve{g}_{\mu}=\breve{g}, (vi) g ˘ μ = g ˘ , where Φ ι = ι Φ a d \Phi_{\iota}=\iota\Phi^{\mathrm{ad}} Φ ι = ι Φ ad and ι : ad M → T M / G \iota:\operatorname{ad}M\to TM/G ι : ad M → T M / G is the inclusion.
Proof. Let g ( t ) = ( h , ω , g ˇ ) ( t ) g(t)=(h,\omega,\check{g})(t) g ( t ) = ( h , ω , g ˇ ) ( t ) be a Ricci flow. Then by Theorem 3.1, ( h , ω , g ˇ ) ( t ) (h,\omega,\check{g})(t) ( h , ω , g ˇ ) ( t ) solves equations (24–26).
Solve the linear ODE (i) fibrewise. The solution Φ ( t ) \Phi(t) Φ ( t ) is smooth, unique, and has the same maximal time of existence as the Ricci flow. Furthermore, Φ ( t ) : T M / G → T M / G \Phi(t):TM/G\to TM/G Φ ( t ) : T M / G → T M / G remains invertible as
∂ ∂ t det Φ = − ( det Φ ) tr ( Ric g a d ) , det Φ ( 0 ) = det Id = 1. \frac{\partial}{\partial t}\det\Phi=-(\det\Phi)\operatorname{tr}\left(\operatorname{Ric}^{\mathrm{ad}}_{g}\right),\qquad\det\Phi(0)=\det\operatorname{Id}=1. ∂ t ∂ det Φ = − ( det Φ ) tr ( Ric g ad ) , det Φ ( 0 ) = det Id = 1.
Since π ∗ Ric a d = 0 \pi_{*}\operatorname{Ric}^{\mathrm{ad}}=0 π ∗ Ric ad = 0 and Φ ( 0 ) = Id \Phi(0)=\operatorname{Id} Φ ( 0 ) = Id , we find that π ∗ Φ = π ∗ \pi_{*}\Phi=\pi_{*} π ∗ Φ = π ∗ , and so Φ \Phi Φ preserves the vertical distribution defining Φ ι = ι Φ a d \Phi\iota=\iota\Phi^{\mathrm{ad}} Φ ι = ι Φ ad . The endomorphism Φ a d ( t ) : ad M → ad M \Phi^{\mathrm{ad}}(t):\operatorname{ad}M\to\operatorname{ad}M Φ ad ( t ) : ad M → ad M satisfies
∂ Φ a d ∂ t = − Φ a d ω Ric g ι . \frac{\partial\Phi^{\mathrm{ad}}}{\partial t}=-\Phi^{\mathrm{ad}}\omega\operatorname{Ric}_{g}\iota. ∂ t ∂ Φ ad = − Φ ad ω Ric g ι .
Consider the vector bundle metric h 0 ( Φ a d ⋅ , Φ a d ⋅ ) h_{0}(\Phi^{\mathrm{ad}}\cdot,\Phi^{\mathrm{ad}}\cdot) h 0 ( Φ ad ⋅ , Φ ad ⋅ ) on the adjoint bundle ad M → B \operatorname{ad}M\to B ad M → B . Its time derivative is
∂ ∂ t h 0 ( Φ a d ⋅ , Φ a d ⋅ ) = h 0 ( ∂ t Φ a d ⋅ , Φ a d ⋅ ) + h 0 ( Φ a d ⋅ , ∂ t Φ a d ⋅ ) = − h 0 ( Φ a d ω Ric g ι ⋅ , Φ a d ⋅ ) − h 0 ( Φ a d ⋅ , Φ a d ω Ric g ι ⋅ ) . \begin{aligned}
\frac{\partial}{\partial t}h_{0}(\Phi^{\mathrm{ad}}\cdot,\Phi^{\mathrm{ad}}\cdot)
&=h_{0}(\partial_{t}\Phi^{\mathrm{ad}}\cdot,\Phi^{\mathrm{ad}}\cdot)+h_{0}(\Phi^{\mathrm{ad}}\cdot,\partial_{t}\Phi^{\mathrm{ad}}\cdot)\\
&=-h_{0}(\Phi^{\mathrm{ad}}\omega\operatorname{Ric}_{g}\iota\cdot,\Phi^{\mathrm{ad}}\cdot)-h_{0}(\Phi^{\mathrm{ad}}\cdot,\Phi^{\mathrm{ad}}\omega\operatorname{Ric}_{g}\iota\cdot).
\end{aligned} ∂ t ∂ h 0 ( Φ ad ⋅ , Φ ad ⋅ ) = h 0 ( ∂ t Φ ad ⋅ , Φ ad ⋅ ) + h 0 ( Φ ad ⋅ , ∂ t Φ ad ⋅ ) = − h 0 ( Φ ad ω Ric g ι ⋅ , Φ ad ⋅ ) − h 0 ( Φ ad ⋅ , Φ ad ω Ric g ι ⋅ ) .
On the other hand, the vector bundle metric h ( t ) h(t) h ( t ) on ad M \operatorname{ad}M ad M under Ricci flow solves (24). Hence,
∂ h ∂ t = − 2 h ( ω Ric g ι ⋅ , ⋅ ) = − h ( ω Ric g ι ⋅ , ⋅ ) − h ( ⋅ , ω Ric g ι ⋅ ) . \begin{aligned}
\frac{\partial h}{\partial t}
&=-2h(\omega\operatorname{Ric}_{g}\iota\cdot,\cdot)\\
&=-h(\omega\operatorname{Ric}_{g}\iota\cdot,\cdot)-h(\cdot,\omega\operatorname{Ric}_{g}\iota\cdot).
\end{aligned} ∂ t ∂ h = − 2 h ( ω Ric g ι ⋅ , ⋅ ) = − h ( ω Ric g ι ⋅ , ⋅ ) − h ( ⋅ , ω Ric g ι ⋅ ) .
By uniqueness, h = h 0 ( Φ a d ⋅ , Φ a d ⋅ ) h=h_{0}(\Phi^{\mathrm{ad}}\cdot,\Phi^{\mathrm{ad}}\cdot) h = h 0 ( Φ ad ⋅ , Φ ad ⋅ ) , proving (iii). Next, we show that if g ( t ) = ( h , ω , g ˇ ) ( t ) g(t)=(h,\omega,\check{g})(t) g ( t ) = ( h , ω , g ˇ ) ( t ) is a Ricci flow, then
Φ [ Φ − 1 ⋅ , Φ − 1 ⋅ ] , g Φ : = ( h 0 , Φ a d ∘ ω ∘ Φ − 1 , g ˇ ) (31) \Phi[\Phi^{-1}\cdot,\Phi^{-1}\cdot],\qquad g_{\Phi}:=(h_{0},\Phi^{\mathrm{ad}}\circ\omega\circ\Phi^{-1},\check{g}) \tag*{(31)} Φ [ Φ − 1 ⋅ , Φ − 1 ⋅ ] , g Φ := ( h 0 , Φ ad ∘ ω ∘ Φ − 1 , g ˇ ) (31)
is a bracket flow. The bracket Φ [ Φ − 1 , Φ − 1 ] \Phi[\Phi^{-1},\Phi^{-1}] Φ [ Φ − 1 , Φ − 1 ] is compatible with the anchor π ∗ Φ − 1 = π ∗ \pi_{*}\Phi^{-1}=\pi_{*} π ∗ Φ − 1 = π ∗ and has initial value [ ⋅ , ⋅ ] [\cdot,\cdot] [ ⋅ , ⋅ ] . As ( Φ a d ) ∗ h 0 = h (\Phi^{\mathrm{ad}})^{*}h_{0}=h ( Φ ad ) ∗ h 0 = h and π ∗ Φ − 1 = π ∗ \pi_{*}\Phi^{-1}=\pi_{*} π ∗ Φ − 1 = π ∗ , the vector bundle metric determined by the triple in (31) is isometric to the Ricci flow g ( t ) g(t) g ( t ) :
g Φ : = ( Φ a d ∘ ω ∘ Φ − 1 ) ∗ h 0 + π ∗ g ˇ = ( Φ − 1 ) ∗ ( ω ∗ h + π ∗ g ˇ ) = ( Φ − 1 ) ∗ g ( t ) . g_{\Phi}:=(\Phi^{\mathrm{ad}}\circ\omega\circ\Phi^{-1})^{*}h_{0}+\pi^{*}\check{g}=(\Phi^{-1})^{*}(\omega^{*}h+\pi^{*}\check{g})=(\Phi^{-1})^{*}g(t). g Φ := ( Φ ad ∘ ω ∘ Φ − 1 ) ∗ h 0 + π ∗ g ˇ = ( Φ − 1 ) ∗ ( ω ∗ h + π ∗ g ˇ ) = ( Φ − 1 ) ∗ g ( t ) .
Naturality of the Ricci curvature gives us
Ric g Φ = Φ Ric g Φ − 1 . \operatorname{Ric}_{g_{\Phi}}=\Phi\operatorname{Ric}_{g}\Phi^{-1}. Ric g Φ = Φ Ric g Φ − 1 .
The Ricci curvature blocks satisfy
Ric g Φ a d = Φ Ric g a d Φ − 1 , Ric g Φ a d , T B = Φ Ric g a d , T B Φ − 1 , ric g Φ T B = ric g T B . \operatorname{Ric}^{\mathrm{ad}}_{g_{\Phi}}=\Phi\operatorname{Ric}^{\mathrm{ad}}_{g}\Phi^{-1},\qquad
\operatorname{Ric}^{\mathrm{ad},TB}_{g_{\Phi}}=\Phi\operatorname{Ric}^{\mathrm{ad},TB}_{g}\Phi^{-1},\qquad
\operatorname{ric}^{TB}_{g_{\Phi}}=\operatorname{ric}^{TB}_{g}. Ric g Φ ad = Φ Ric g ad Φ − 1 , Ric g Φ ad , T B = Φ Ric g ad , T B Φ − 1 , ric g Φ T B = ric g T B .
Then, the bracket evolves by
∂ ∂ t Φ [ Φ − 1 ⋅ , Φ − 1 ⋅ ] = − Φ Ric g a d [ Φ − 1 ⋅ , Φ − 1 ⋅ ] + Φ [ Ric g a d Φ − 1 ⋅ , Φ − 1 ⋅ ] + Φ [ Φ − 1 ⋅ , Ric g a d Φ − 1 ⋅ ] = − Ric g Φ a d Φ [ Φ − 1 ⋅ , Φ − 1 ⋅ ] + Φ [ Φ − 1 Ric g a d ⋅ , Φ − 1 ⋅ ] + Φ [ Φ − 1 ⋅ , Φ − 1 Ric g a d ⋅ ] = δ Φ [ Φ − 1 , Φ − 1 ] ( Ric g a d ) . \begin{aligned}
\frac{\partial}{\partial t}\Phi[\Phi^{-1}\cdot,\Phi^{-1}\cdot]
&=-\Phi\operatorname{Ric}^{\mathrm{ad}}_{g}[\Phi^{-1}\cdot,\Phi^{-1}\cdot]+\Phi[\operatorname{Ric}^{\mathrm{ad}}_{g}\Phi^{-1}\cdot,\Phi^{-1}\cdot]+\Phi[\Phi^{-1}\cdot,\operatorname{Ric}^{\mathrm{ad}}_{g}\Phi^{-1}\cdot]\\
&=-\operatorname{Ric}^{\mathrm{ad}}_{g_{\Phi}}\Phi[\Phi^{-1}\cdot,\Phi^{-1}\cdot]+\Phi[\Phi^{-1}\operatorname{Ric}^{\mathrm{ad}}_{g}\cdot,\Phi^{-1}\cdot]+\Phi[\Phi^{-1}\cdot,\Phi^{-1}\operatorname{Ric}^{\mathrm{ad}}_{g}\cdot]\\
&=\delta_{\Phi[\Phi^{-1},\Phi^{-1}]}(\operatorname{Ric}^{\mathrm{ad}}_{g}).
\end{aligned} ∂ t ∂ Φ [ Φ − 1 ⋅ , Φ − 1 ⋅ ] = − Φ Ric g ad [ Φ − 1 ⋅ , Φ − 1 ⋅ ] + Φ [ Ric g ad Φ − 1 ⋅ , Φ − 1 ⋅ ] + Φ [ Φ − 1 ⋅ , Ric g ad Φ − 1 ⋅ ] = − Ric g Φ ad Φ [ Φ − 1 ⋅ , Φ − 1 ⋅ ] + Φ [ Φ − 1 Ric g ad ⋅ , Φ − 1 ⋅ ] + Φ [ Φ − 1 ⋅ , Φ − 1 Ric g ad ⋅ ] = δ Φ [ Φ − 1 , Φ − 1 ] ( Ric g ad ) .
the connection evolves by
∂ ∂ t ( Φ a d ∘ ω ∘ Φ − 1 ) = − ( Φ a d Ric g a d ) ∘ ω ∘ Φ − 1 − Φ a d ∘ ( 2 ω Ric g a d , T B ) ∘ Φ − 1 + Φ a d ∘ ω ∘ ( Ric g a d Φ − 1 ) = − Φ a d ∘ ( 2 ω Ric g a d , T B ) ∘ Φ − 1 = − 2 ( Φ a d ∘ ω ∘ Φ − 1 ) Ric g Φ a d , T B , \begin{aligned}
\frac{\partial}{\partial t}(\Phi^{\mathrm{ad}}\circ\omega\circ\Phi^{-1})
&=-(\Phi^{\mathrm{ad}}\operatorname{Ric}^{\mathrm{ad}}_{g})\circ\omega\circ\Phi^{-1}-\Phi^{\mathrm{ad}}\circ(2\omega\operatorname{Ric}^{\mathrm{ad},TB}_{g})\circ\Phi^{-1}+\Phi^{\mathrm{ad}}\circ\omega\circ(\operatorname{Ric}^{\mathrm{ad}}_{g}\Phi^{-1})\\
&=-\Phi^{\mathrm{ad}}\circ(2\omega\operatorname{Ric}^{\mathrm{ad},TB}_{g})\circ\Phi^{-1}\\
&=-2(\Phi^{\mathrm{ad}}\circ\omega\circ\Phi^{-1})\operatorname{Ric}^{\mathrm{ad},TB}_{g_{\Phi}},
\end{aligned} ∂ t ∂ ( Φ ad ∘ ω ∘ Φ − 1 ) = − ( Φ ad Ric g ad ) ∘ ω ∘ Φ − 1 − Φ ad ∘ ( 2 ω Ric g ad , T B ) ∘ Φ − 1 + Φ ad ∘ ω ∘ ( Ric g ad Φ − 1 ) = − Φ ad ∘ ( 2 ω Ric g ad , T B ) ∘ Φ − 1 = − 2 ( Φ ad ∘ ω ∘ Φ − 1 ) Ric g Φ ad , T B ,
and the metric on the base evolves by
∂ g ˇ ∂ t = − 2 ric g T B = − 2 ric g Φ T B . \frac{\partial\check{g}}{\partial t}=-2\operatorname{ric}^{TB}_{g}=-2\operatorname{ric}^{TB}_{g_{\Phi}}. ∂ t ∂ g ˇ = − 2 ric g T B = − 2 ric g Φ T B .
Indeed, (31) is a bracket flow (Theorem 3.2).
Conversely, let ( T M / G , B , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t ) ) (TM/G,B,[\cdot,\cdot]_{\mu(t)},\pi_{*},g_{\mu(t)}) ( T M / G , B , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t ) ) be a bracket flow with initial metric g 0 g_{0} g 0 . Solve the linear ODE (ii) fibrewise. As before, the solution Φ ( t ) \Phi(t) Φ ( t ) is unique, smooth, invertible, preserves the anchor π ∗ \pi_* π ∗ , and has the same maximal time of existence as the bracket flow. Observe that the pullback of the bracket is fixed:
∂ ∂ t Φ − 1 [ Φ ⋅ , Φ ⋅ ] μ = δ Φ − 1 [ Φ ⋅ , Φ ⋅ ] μ ( Φ − 1 ∂ t Φ + Φ − 1 Ric g μ a d Φ ) = 0. \frac{\partial}{\partial t}\Phi^{-1}[\Phi\cdot,\Phi\cdot]_{\mu}
=\delta_{\Phi^{-1}[\Phi\cdot,\Phi\cdot]_{\mu}}\left(\Phi^{-1}\partial_t\Phi+\Phi^{-1}\operatorname{Ric}^{\mathrm{ad}}_{g_\mu}\Phi\right)=0. ∂ t ∂ Φ − 1 [ Φ ⋅ , Φ ⋅ ] μ = δ Φ − 1 [ Φ ⋅ , Φ ⋅ ] μ ( Φ − 1 ∂ t Φ + Φ − 1 Ric g μ ad Φ ) = 0.
Set g ( t ) = Φ ( t ) ∗ g μ ( t ) g(t)=\Phi(t)^*g_\mu(t) g ( t ) = Φ ( t ) ∗ g μ ( t ) , then
Φ ( t ) : ( T M / G , [ ⋅ , ⋅ ] , π ∗ , g ( t ) ) ⟶ ( T M / G , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t ) ) \Phi(t):(TM/G,[\cdot,\cdot],\pi_*,g(t))\longrightarrow(TM/G,[\cdot,\cdot]_{\mu(t)},\pi_*,g_\mu(t)) Φ ( t ) : ( T M / G , [ ⋅ , ⋅ ] , π ∗ , g ( t )) ⟶ ( T M / G , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t ))
is an isometric isomorphism of Riemannian Lie algebroids where g ( t ) g(t) g ( t ) is a Ricci flow:
∂ g ∂ t = ∂ ∂ t Φ ∗ g μ = ∂ ∂ t ( h 0 ( ω μ Φ ⋅ , ω μ Φ ⋅ ) + g ˇ μ ( π ∗ Φ ⋅ , π ∗ Φ ⋅ ) ) = h 0 ( ( ∂ t ω μ Φ + ω μ ∂ t Φ ) ⋅ , ω μ Φ ⋅ ) + h 0 ( ω μ Φ ⋅ , ( ∂ t ω μ Φ + ω μ ∂ t Φ ) ⋅ ) + ( ∂ t g ˇ μ ) ( π ∗ ⋅ , π ∗ ⋅ ) = − h 0 ( ω μ ( 2 Ric g μ a d , T B + Ric g μ a d ) Φ ⋅ , ω μ Φ ⋅ ) − h 0 ( ω μ Φ ⋅ , ω μ ( 2 Ric g μ a d , T B + Ric g μ a d ) Φ ⋅ ) \begin{aligned}
\frac{\partial g}{\partial t}=\frac{\partial}{\partial t}\Phi^*g_\mu\\
&=\frac{\partial}{\partial t}\left(h_0(\omega_\mu\Phi\cdot,\omega_\mu\Phi\cdot)+\check{g}_\mu(\pi_*\Phi\cdot,\pi_*\Phi\cdot)\right) \\
&=h_0((\partial_t\omega_\mu\Phi+\omega_\mu\partial_t\Phi)\cdot,\omega_\mu\Phi\cdot)+h_0(\omega_\mu\Phi\cdot,(\partial_t\omega_\mu\Phi+\omega_\mu\partial_t\Phi)\cdot)+(\partial_t\check{g}_\mu)(\pi_*\cdot,\pi_*\cdot) \\
&=-h_0(\omega_\mu(2\operatorname{Ric}^{\mathrm{ad},TB}_{g_\mu}+\operatorname{Ric}^{\mathrm{ad}}_{g_\mu})\Phi\cdot,\omega_\mu\Phi\cdot)-h_0(\omega_\mu\Phi\cdot,\omega_\mu(2\operatorname{Ric}^{\mathrm{ad},TB}_{g_\mu}+\operatorname{Ric}^{\mathrm{ad}}_{g_\mu})\Phi\cdot)
\end{aligned} ∂ t ∂ g = ∂ t ∂ Φ ∗ g μ = ∂ t ∂ ( h 0 ( ω μ Φ ⋅ , ω μ Φ ⋅ ) + g ˇ μ ( π ∗ Φ ⋅ , π ∗ Φ ⋅ ) ) = h 0 (( ∂ t ω μ Φ + ω μ ∂ t Φ ) ⋅ , ω μ Φ ⋅ ) + h 0 ( ω μ Φ ⋅ , ( ∂ t ω μ Φ + ω μ ∂ t Φ ) ⋅ ) + ( ∂ t g ˇ μ ) ( π ∗ ⋅ , π ∗ ⋅ ) = − h 0 ( ω μ ( 2 Ric g μ ad , T B + Ric g μ ad ) Φ ⋅ , ω μ Φ ⋅ ) − h 0 ( ω μ Φ ⋅ , ω μ ( 2 Ric g μ ad , T B + Ric g μ ad ) Φ ⋅ )
− 2 ric g μ T B ( π ∗ ⋅ , π ∗ ⋅ ) = − 2 h 0 ( ω μ Φ Ric g μ a d , T B ⋅ , ω μ Φ ⋅ ) − h 0 ( ω μ Φ Ric g a d ⋅ , ω μ Φ ⋅ ) \begin{aligned}
\qquad-2\operatorname{ric}^{TB}_{g_\mu}(\pi_*\cdot,\pi_*\cdot) \\
&=-2h_0(\omega_\mu\Phi\operatorname{Ric}^{\mathrm{ad},TB}_{g_\mu}\cdot,\omega_\mu\Phi\cdot)-h_0(\omega_\mu\Phi\operatorname{Ric}^{\mathrm{ad}}_g\cdot,\omega_\mu\Phi\cdot)
\end{aligned} − 2 ric g μ T B ( π ∗ ⋅ , π ∗ ⋅ ) = − 2 h 0 ( ω μ Φ Ric g μ ad , T B ⋅ , ω μ Φ ⋅ ) − h 0 ( ω μ Φ Ric g ad ⋅ , ω μ Φ ⋅ )
− 2 h 0 ( ω μ Φ ⋅ , ω μ Φ Ric g a d , T B ⋅ ) − h 0 ( ω μ Φ ⋅ , ω μ Φ Ric g a d ⋅ ) − 2 ric g T B = − 2 ric g T B × a d − 2 ric g a d − 2 ric g a d × T B − 2 ric g T B = − 2 ric g . \begin{aligned}
\qquad-2h_0(\omega_\mu\Phi\cdot,\omega_\mu\Phi\operatorname{Ric}^{\mathrm{ad},TB}_{g}\cdot)-h_0(\omega_\mu\Phi\cdot,\omega_\mu\Phi\operatorname{Ric}^{\mathrm{ad}}_g\cdot)-2\operatorname{ric}^{TB}_g \\
&=-2\operatorname{ric}^{TB}_g\times\mathrm{ad}-2\operatorname{ric}^{\mathrm{ad}}_g-2\operatorname{ric}^{\mathrm{ad}\times TB}_g-2\operatorname{ric}^{TB}_g \\
&=-2\operatorname{ric}_g.
\end{aligned} − 2 h 0 ( ω μ Φ ⋅ , ω μ Φ Ric g ad , T B ⋅ ) − h 0 ( ω μ Φ ⋅ , ω μ Φ Ric g ad ⋅ ) − 2 ric g T B = − 2 ric g T B × ad − 2 ric g ad − 2 ric g ad × T B − 2 ric g T B = − 2 ric g .
The second last equality follows from symmetry of the blocks. □
Moment maps on the adjoint bundle
Let G G G be a Lie group equipped with an invariant metric g g g . Let μ \mu μ be the bracket of the Lie algebra g \mathfrak{g} g , and let ⟨ ⋅ , ⋅ ⟩ \langle\cdot,\cdot\rangle ⟨ ⋅ , ⋅ ⟩ be the inner product on the Lie algebra given by evaluating g g g at the identity. The Ricci curvature endomorphism is
Ric μ g = M μ g − 1 2 B μ g − S g ( ad μ H μ g ) . (32) \operatorname{Ric}^{g}_{\mu}=M^{g}_{\mu}-\frac{1}{2}B^{g}_{\mu}-S^{g}(\operatorname{ad}_{\mu}H^{g}_{\mu}). \tag*{(32)} Ric μ g = M μ g − 2 1 B μ g − S g ( ad μ H μ g ) . (32)
Here M μ g M^{g}_{\mu} M μ g is the self-adjoint endomorphism of ( g , ⟨ ⋅ , ⋅ ⟩ ) (\mathfrak{g},\langle\cdot,\cdot\rangle) ( g , ⟨ ⋅ , ⋅ ⟩) given by
⟨ M μ g U , U ⟩ = − 1 2 ∑ i , j ⟨ μ ( U , V i ) , V j ⟩ 2 + 1 4 ∑ i , j ⟨ μ ( V i , V j ) , U ⟩ 2 \langle M^{g}_{\mu}U,U\rangle=-\frac{1}{2}\sum_{i,j}\langle\mu(U,V_i),V_j\rangle^2+\frac{1}{4}\sum_{i,j}\langle\mu(V_i,V_j),U\rangle^2 ⟨ M μ g U , U ⟩ = − 2 1 i , j ∑ ⟨ μ ( U , V i ) , V j ⟩ 2 + 4 1 i , j ∑ ⟨ μ ( V i , V j ) , U ⟩ 2
where { V i } \{V_i\} { V i } is an orthonormal basis for ( g , ⟨ ⋅ , ⋅ ⟩ ) (\mathfrak{g},\langle\cdot,\cdot\rangle) ( g , ⟨ ⋅ , ⋅ ⟩) . Consider the action of G L ( g ) \mathrm{GL}(\mathfrak{g}) GL ( g ) on Λ 2 g ∗ ⊗ g \Lambda^2\mathfrak{g}^*\otimes\mathfrak{g} Λ 2 g ∗ ⊗ g given by Φ ⋅ λ = Φ λ ( Φ − 1 ⋅ , Φ − 1 ⋅ ) \Phi\cdot\lambda=\Phi\lambda(\Phi^{-1}\cdot,\Phi^{-1}\cdot) Φ ⋅ λ = Φ λ ( Φ − 1 ⋅ , Φ − 1 ⋅ ) . Then,
4 ⟨ M μ g , α ⟩ = d d t ∣ t = 0 ∣ exp ( t α ) ⋅ μ ∣ 2 = − ⟨ δ μ ( α ) , μ ⟩ 4\langle M^{g}_{\mu},\alpha\rangle=\left.\frac{d}{dt}\right|_{t=0}\left|\exp(t\alpha)\cdot\mu\right|^2=-\langle\delta_{\mu}(\alpha),\mu\rangle 4 ⟨ M μ g , α ⟩ = d t d t = 0 ∣ exp ( t α ) ⋅ μ ∣ 2 = − ⟨ δ μ ( α ) , μ ⟩
for all α ∈ End ( g ) \alpha\in\operatorname{End}(\mathfrak{g}) α ∈ End ( g ) . Because of this, we refer to this endomorphism M μ g M^{g}_{\mu} M μ g as the moment map of the G L ( g ) \mathrm{GL}(\mathfrak{g}) GL ( g ) action on Λ 2 g ∗ ⊗ g \Lambda^2\mathfrak{g}^*\otimes\mathfrak{g} Λ 2 g ∗ ⊗ g . The second term of (32) is the Killing form endomorphism given by
⟨ B μ g U , V ⟩ = tr ad μ U ad μ V . \langle B^{g}_{\mu}U,V\rangle=\operatorname{tr}\operatorname{ad}_{\mu}U\operatorname{ad}_{\mu}V. ⟨ B μ g U , V ⟩ = tr ad μ U ad μ V .
The third term of (32) is the symmetric part of the mean curvature operator ad μ H μ g \operatorname{ad}_{\mu}H^{g}_{\mu} ad μ H μ g , where ⟨ H μ g , U ⟩ = tr ad μ U \langle H^{g}_{\mu},U\rangle=\operatorname{tr}\operatorname{ad}_{\mu}U ⟨ H μ g , U ⟩ = tr ad μ U for all U ∈ g U\in\mathfrak{g} U ∈ g . When G G G is nilpotent, B μ g B^{g}_{\mu} B μ g and H μ g H^{g}_{\mu} H μ g vanish as all the eigenvalues of nilpotent matrices are zero, so (32) becomes just
Ric μ g = M μ g . \operatorname{Ric}^{g}_{\mu}=M^{g}_{\mu}. Ric μ g = M μ g .
Let M → B M\to B M → B be a principal G G G -bundle equipped with an invariant metric g = ( h , ω , g ˇ ) g=(h,\omega,\check{g}) g = ( h , ω , g ˇ ) . Let μ \mu μ be the bracket on the adjoint bundle. On the fibres, G G G acts freely and transitively, so the Ricci curvature of the fibres is given again by (32). With a fixed background metric h h h on ad M \operatorname{ad}M ad M , we write
Ric ^ μ = M μ − 1 2 B μ − S ( ad μ H μ ) . \widehat{\operatorname{Ric}}_{\mu}=M_{\mu}-\frac{1}{2}B_{\mu}-S(\operatorname{ad}_{\mu}H_{\mu}). Ric μ = M μ − 2 1 B μ − S ( ad μ H μ ) .
Here one can think of M μ ∈ End ( ad M ) M_{\mu}\in\operatorname{End}(\operatorname{ad}M) M μ ∈ End ( ad M ) as the moment map of the gauge group action (the vector bundle automorphisms over the identity) on sections of Λ 2 ad M ∗ ⊗ ad M \Lambda^2\operatorname{ad}M^*\otimes\operatorname{ad}M Λ 2 ad M ∗ ⊗ ad M :
Φ ⋅ λ = Φ λ ( Φ − 1 ⋅ , Φ − 1 ⋅ ) , \Phi\cdot\lambda=\Phi\lambda(\Phi^{-1}\cdot,\Phi^{-1}\cdot), Φ ⋅ λ = Φ λ ( Φ − 1 ⋅ , Φ − 1 ⋅ ) ,
where Φ ∈ Aut ( ad M ) \Phi\in\operatorname{Aut}(\operatorname{ad}M) Φ ∈ Aut ( ad M ) and λ ∈ Γ ( Λ 2 ad M ∗ ⊗ ad M ) \lambda\in\Gamma(\Lambda^{2}\operatorname{ad}M^{*} \otimes\operatorname{ad}M) λ ∈ Γ ( Λ 2 ad M ∗ ⊗ ad M ) . Indeed,
h ( M μ , α ) = − 1 4 h ( δ μ ( α ) , μ ) (33) h(M_{\mu},\alpha)=-\frac{1}{4}h(\delta_{\mu}(\alpha),\mu) \tag*{(33)} h ( M μ , α ) = − 4 1 h ( δ μ ( α ) , μ ) (33)
where δ μ : End ( ad M ) → Γ ( Λ 2 ad M ∗ ⊗ ad M ) \delta_{\mu}:\operatorname{End}(\operatorname{ad}M)\to\Gamma(\Lambda^{2}\operatorname{ad}M^{*}\otimes\operatorname{ad}M) δ μ : End ( ad M ) → Γ ( Λ 2 ad M ∗ ⊗ ad M ) is given by
δ μ ( α ) = μ ( α ⋅ , ⋅ ) + μ ( ⋅ , α ⋅ ) − α μ ( ⋅ , ⋅ ) . \delta_{\mu}(\alpha)=\mu(\alpha\cdot,\cdot)+\mu(\cdot,\alpha\cdot)-\alpha\mu(\cdot,\cdot). δ μ ( α ) = μ ( α ⋅ , ⋅ ) + μ ( ⋅ , α ⋅ ) − α μ ( ⋅ , ⋅ ) .
More generally, let [ ⋅ , ⋅ ] μ [\cdot,\cdot]_{\mu} [ ⋅ , ⋅ ] μ be a Lie algebroid bracket on T M / G TM/G T M / G with anchor π ∗ \pi_{*} π ∗ . Let μ ∈ Γ ( Λ 2 ad M ∗ ⊗ ad M ) \mu\in\Gamma(\Lambda^{2}\operatorname{ad}M^{*}\otimes\operatorname{ad}M) μ ∈ Γ ( Λ 2 ad M ∗ ⊗ ad M ) be the restriction of [ ⋅ , ⋅ ] μ [\cdot,\cdot]_{\mu} [ ⋅ , ⋅ ] μ . A choice of principal connection on M M M determines an affine connection ∇ ad μ \nabla^{\operatorname{ad}\mu} ∇ ad μ on the adjoint bundle ad M \operatorname{ad}M ad M :
∇ X ad μ U ad = ω [ X ‾ , U ] μ . \nabla^{\operatorname{ad}\mu}_{X}U_{\operatorname{ad}}=\omega[\overline{X},U]_{\mu}. ∇ X ad μ U ad = ω [ X , U ] μ .
By the Jacobi identity, we find that μ \mu μ is parallel with respect to this connection,
( ∇ X ad μ μ ) ( U ad , V ad ) = ∇ X ad μ μ ( U ad , V ad ) − μ ( ∇ X ad μ U ad , V ad ) − μ ( U ad , ∇ X ad μ V ad ) = [ X ‾ , [ U , V ] μ ] μ − [ [ X ‾ , U ] μ , V ] μ − [ U , [ X ‾ , V ] μ ] μ = 0. \begin{aligned}
(\nabla^{\operatorname{ad}\mu}_{X}\mu)(U_{\operatorname{ad}},V_{\operatorname{ad}})&=\nabla^{\operatorname{ad}\mu}_{X}\mu(U_{\operatorname{ad}},V_{\operatorname{ad}})-\mu(\nabla^{\operatorname{ad}\mu}_{X}U_{\operatorname{ad}},V_{\operatorname{ad}})-\mu(U_{\operatorname{ad}},\nabla^{\operatorname{ad}\mu}_{X}V_{\operatorname{ad}})\\
&=[\overline{X},[U,V]_{\mu}]_{\mu}-[[\overline{X},U]_{\mu},V]_{\mu}-[U,[\overline{X},V]_{\mu}]_{\mu}\\
&=0.
\end{aligned} ( ∇ X ad μ μ ) ( U ad , V ad ) = ∇ X ad μ μ ( U ad , V ad ) − μ ( ∇ X ad μ U ad , V ad ) − μ ( U ad , ∇ X ad μ V ad ) = [ X , [ U , V ] μ ] μ − [[ X , U ] μ , V ] μ − [ U , [ X , V ] μ ] μ = 0.
Equip the adjoint bundle ad M \operatorname{ad}M ad M with a metric h h h . If ∇ \nabla ∇ is a metric connection on ( ad M , h ) (\operatorname{ad}M,h) ( ad M , h ) , then define L ∈ Ω 1 ( B ; End ( ad M ) ) L\in\Omega^{1}(B;\operatorname{End}(\operatorname{ad}M)) L ∈ Ω 1 ( B ; End ( ad M )) by ∇ ad μ = ∇ − L \nabla^{\operatorname{ad}\mu}=\nabla-L ∇ ad μ = ∇ − L . As ∇ ad μ μ = 0 \nabla^{\operatorname{ad}\mu}\mu=0 ∇ ad μ μ = 0 , we have
( ∇ X μ ) = − δ μ ( L X ) , ( ∇ X , Y 2 μ ) = − δ μ ( ( ∇ X L ) Y ) + δ δ μ ( L X ) μ ( L Y ) . (\nabla_{X}\mu)=-\delta_{\mu}(L_{X}),\qquad(\nabla^{2}_{X,Y}\mu)=-\delta_{\mu}((\nabla_{X}L)Y)+\delta_{\delta_{\mu}(L_{X})}\mu(L_{Y}). ( ∇ X μ ) = − δ μ ( L X ) , ( ∇ X , Y 2 μ ) = − δ μ (( ∇ X L ) Y ) + δ δ μ ( L X ) μ ( L Y ) .
Proposition 3.5. Let M → B M\to B M → B be a twisted principal bundle. Equip ad M → B \operatorname{ad}M\to B ad M → B with a bundle metric h h h . Let μ ∈ Γ ( Λ 2 ad M ∗ ⊗ ad M ) \mu\in\Gamma(\Lambda^{2}\operatorname{ad}M^{*}\otimes\operatorname{ad}M) μ ∈ Γ ( Λ 2 ad M ∗ ⊗ ad M ) satisfy the Jacobi identity. Then
h ( ∇ X M μ , α ) = − 1 2 h ( δ μ ( α ) , ∇ X μ ) (34) h(\nabla_{X}M_{\mu},\alpha)=-\frac{1}{2}h(\delta_{\mu}(\alpha),\nabla_{X}\mu) \tag*{(34)} h ( ∇ X M μ , α ) = − 2 1 h ( δ μ ( α ) , ∇ X μ ) (34)
h ( ∇ X ∇ Y M μ , α ) = − 1 2 h ( δ ∇ X μ ( α ) , ∇ Y μ ) − 1 2 h ( δ μ ( α ) , ∇ X ∇ Y μ ) (35) h(\nabla_{X}\nabla_{Y}M_{\mu},\alpha)=-\frac{1}{2}h(\delta_{\nabla_{X}\mu}(\alpha),\nabla_{Y}\mu)-\frac{1}{2}h(\delta_{\mu}(\alpha),\nabla_{X}\nabla_{Y}\mu) \tag*{(35)} h ( ∇ X ∇ Y M μ , α ) = − 2 1 h ( δ ∇ X μ ( α ) , ∇ Y μ ) − 2 1 h ( δ μ ( α ) , ∇ X ∇ Y μ ) (35)
for all h h h -symmetric α ∈ End ( ad M ) \alpha\in\operatorname{End}(\operatorname{ad}M) α ∈ End ( ad M ) and for all vector fields X , Y X,Y X , Y on B B B . Here M μ M_{\mu} M μ is defined by (33).
Proof. If α ∈ End ( ad M ) \alpha\in\operatorname{End}(\operatorname{ad}M) α ∈ End ( ad M ) is h h h -symmetric, then δ ( α ) ∈ End ( Λ 2 ad M ∗ ⊗ ad M ) \delta(\alpha)\in\operatorname{End}(\Lambda^{2}\operatorname{ad}M^{*}\otimes\operatorname{ad}M) δ ( α ) ∈ End ( Λ 2 ad M ∗ ⊗ ad M ) is too. As ∇ X δ μ ( α ) = δ μ ( ∇ X α ) + δ ∇ X μ ( α ) \nabla_{X}\delta_{\mu}(\alpha)=\delta_{\mu}(\nabla_{X}\alpha)+\delta_{\nabla_{X}\mu}(\alpha) ∇ X δ μ ( α ) = δ μ ( ∇ X α ) + δ ∇ X μ ( α ) , we can write
4 h ( ∇ X M μ , α ) = 4 X h ( M μ , α ) − 4 h ( M μ , ∇ X α ) = − X h ( δ μ ( α ) , μ ) + h ( δ μ ( ∇ X α ) , μ ) = − h ( δ ∇ X μ ( α ) , μ ) − h ( δ μ ( α ) , ∇ X μ ) = − 2 h ( δ μ ( α ) , ∇ X μ ) . \begin{aligned}
4h(\nabla_{X}M_{\mu},\alpha)&=4Xh(M_{\mu},\alpha)-4h(M_{\mu},\nabla_{X}\alpha)\\
&=-Xh(\delta_{\mu}(\alpha),\mu)+h(\delta_{\mu}(\nabla_{X}\alpha),\mu)\\
&=-h(\delta_{\nabla_{X}\mu}(\alpha),\mu)-h(\delta_{\mu}(\alpha),\nabla_{X}\mu)\\
&=-2h(\delta_{\mu}(\alpha),\nabla_{X}\mu).
\end{aligned} 4 h ( ∇ X M μ , α ) = 4 X h ( M μ , α ) − 4 h ( M μ , ∇ X α ) = − X h ( δ μ ( α ) , μ ) + h ( δ μ ( ∇ X α ) , μ ) = − h ( δ ∇ X μ ( α ) , μ ) − h ( δ μ ( α ) , ∇ X μ ) = − 2 h ( δ μ ( α ) , ∇ X μ ) .
For equation (35), we substitute in equation (34):
2 h ( ∇ X ∇ Y M μ , α ) = 2 X h ( ∇ Y M μ , α ) − 2 h ( ∇ Y M μ , ∇ X α ) = − X h ( δ μ ( α ) , ∇ Y μ ) + h ( δ μ ( ∇ X α ) , ∇ Y μ ) = − h ( δ ∇ X μ ( α ) , ∇ Y μ ) − h ( δ μ ( α ) , ∇ X ∇ Y μ ) . \begin{aligned}
2h(\nabla_{X}\nabla_{Y}M_{\mu},\alpha)&=2Xh(\nabla_{Y}M_{\mu},\alpha)-2h(\nabla_{Y}M_{\mu},\nabla_{X}\alpha)\\
&=-Xh(\delta_{\mu}(\alpha),\nabla_{Y}\mu)+h(\delta_{\mu}(\nabla_{X}\alpha),\nabla_{Y}\mu)\\
&=-h(\delta_{\nabla_{X}\mu}(\alpha),\nabla_{Y}\mu)-h(\delta_{\mu}(\alpha),\nabla_{X}\nabla_{Y}\mu).
\end{aligned} 2 h ( ∇ X ∇ Y M μ , α ) = 2 X h ( ∇ Y M μ , α ) − 2 h ( ∇ Y M μ , ∇ X α ) = − X h ( δ μ ( α ) , ∇ Y μ ) + h ( δ μ ( ∇ X α ) , ∇ Y μ ) = − h ( δ ∇ X μ ( α ) , ∇ Y μ ) − h ( δ μ ( α ) , ∇ X ∇ Y μ ) .
Type III estimates
In this section, we prove Theorem A. To do this, we use the bracket flow. Under the bracket flow, we prove Riemannian curvature and diameter estimates using the maximum principle. By Theorem B, these estimates hold for the Ricci flow.
Setup
Let M → S 1 M \to S^{1} M → S 1 be a twisted principal G-bundle. Consider the Ricci flow of invariant metrics on M M M in the following gauge,
∂ g ∂ t = − 2 ric g + L N g , g ( 0 ) = g 0 (36) \frac{\partial g}{\partial t}=-2\operatorname{ric}_{g}+\mathcal{L}_{N}g,\qquad g(0)=g_{0} \tag*{(36)} ∂ t ∂ g = − 2 ric g + L N g , g ( 0 ) = g 0 (36)
where N = − ( tr L ) X ‾ N=-(\operatorname{tr}L)\overline{X} N = − ( tr L ) X is the mean curvature vector of the fibres. This choice simplifies computations as it eliminates the first order terms in our maximum principle arguments. The Ricci curvature of invariant metrics on twisted principal bundles with one-dimensional base is given in Theorem 2.6. Then, in this gauge (36) , the equivalent formulation of the bracket flow (Theorem 3.2)
( T M / G , B , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t ) ) (TM/G,B,[\cdot,\cdot]_{\mu}(t),\pi_{*},g_{\mu}(t)) ( T M / G , B , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t ))
where
g μ ( t ) = ω μ ( t ) ∗ h 0 + π ∗ g ˇ μ ( t ) , g_{\mu}(t)=\omega_{\mu}(t)^{*}h_{0}+\pi^{*}\check{g}_{\mu}(t), g μ ( t ) = ω μ ( t ) ∗ h 0 + π ∗ g ˇ μ ( t ) ,
is the following system
∂ [ ⋅ , ⋅ ] μ ∂ t = δ [ ⋅ , ⋅ ] μ ( Ric ^ μ − ∇ L ) , (37) \frac{\partial[\cdot,\cdot]_{\mu}}{\partial t}=\delta_{[\cdot,\cdot]_{\mu}}\left(\widehat{\operatorname{Ric}}_{\mu}-\nabla L\right), \tag*{(37)} ∂ t ∂ [ ⋅ , ⋅ ] μ = δ [ ⋅ , ⋅ ] μ ( Ric μ − ∇ L ) , (37)
∂ ω μ ∂ t ( X ‾ ) = 2 ω μ ( ∇ ^ ∗ L ) , (38) \frac{\partial\omega_{\mu}}{\partial t}(\overline{X})=2\omega_{\mu}(\widehat{\nabla}^{*}L), \tag*{(38)} ∂ t ∂ ω μ ( X ) = 2 ω μ ( ∇ ∗ L ) , (38)
∂ g ˇ μ ∂ t = 2 ∣ L ∣ 2 g ˇ μ . (39) \frac{\partial\check{g}_{\mu}}{\partial t}=2|L|^{2}\check{g}_{\mu}. \tag*{(39)} ∂ t ∂ g ˇ μ = 2∣ L ∣ 2 g ˇ μ . (39)
Here Ric ^ μ \widehat{\operatorname{Ric}}_{\mu} Ric μ is the Ricci endomorphism of the fibres. We do not use a subscript g μ g_{\mu} g μ as the Ricci curvature of the fibres only depends on the fixed vector bundle metric h h h on a d M adM a d M , and the moving Lie bracket μ ( t ) \mu(t) μ ( t ) on a d M adM a d M . Here, X ‾ \overline{X} X is the horizontal lift of a unit vector field X X X on the base, L = L X L=L_{X} L = L X is the shape operator, ∇ L = ( ∇ X L ) X \nabla L=(\nabla_{X}L)X ∇ L = ( ∇ X L ) X is its covariant derivative, ∇ ^ \widehat{\nabla} ∇ is the Levi-Civita connection of the fibres, and its adjoint applied to the shape operator is
∇ ^ ∗ L = ∇ ^ ∗ ( L X ) = − ∑ j ( ∇ ^ U j L X ) U j \widehat{\nabla}^{*}L=\widehat{\nabla}^{*}(L_{X})=-\sum_{j}(\widehat{\nabla}_{U_{j}}L_{X})U_{j} ∇ ∗ L = ∇ ∗ ( L X ) = − j ∑ ( ∇ U j L X ) U j
for local orthonormal frames { U i } \{U_{i}\} { U i } of ( a d M , h 0 ) (adM,h_{0}) ( a d M , h 0 ) . It follows from the proof of Theorem B that (37–39) is equivalent to (36) . By Theorem 2.5, the Riemannian curvature of ( T M / G , B , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t ) ) (TM/G,B,[\cdot,\cdot]_{\mu}(t),\pi_{*},g_{\mu}(t)) ( T M / G , B , [ ⋅ , ⋅ ] μ ( t ) , π ∗ , g μ ( t )) is bounded above by
∣ Rm g μ ∣ ≤ C ( ∣ μ ∣ 2 + ∣ L ∣ 2 + ∣ ∇ L + L 2 ∣ ) (40) |\operatorname{Rm}_{g_{\mu}}|\leq C\left(|\mu|^{2}+|L|^{2}+|\nabla L+L^{2}|\right) \tag*{(40)} ∣ Rm g μ ∣ ≤ C ( ∣ μ ∣ 2 + ∣ L ∣ 2 + ∣∇ L + L 2 ∣ ) (40)
for some constant C > 0 C>0 C > 0 . Here μ ( t ) \mu(t) μ ( t ) is the Lie bracket on a d M adM a d M given by the restriction of [ ⋅ , ⋅ ] μ [\cdot,\cdot]_{\mu} [ ⋅ , ⋅ ] μ to vertical vector fields. In the following subsections, we estimate each of these terms. First, we compute the evolution of the shape operator, the Levi-Civita connection ∇ \nabla ∇ on a d M adM a d M , and the connection ∇ a d μ \nabla^{ad\mu} ∇ a d μ on a d M adM a d M induced by the principal connection:
∇ X a d μ U a d = ω μ [ X ‾ , U ] μ , \nabla^{ad\mu}_{X}U_{ad}=\omega_{\mu}[\overline{X},U]_{\mu}, ∇ X a d μ U a d = ω μ [ X , U ] μ ,
Lemma 4.1. Let M → S 1 M \to S^{1} M → S 1 be a twisted principal G-bundle. Fix a vector bundle metric h h h on the adjoint bundle a d M adM a d M and let ( [ ⋅ , ⋅ ] μ , ω μ , g ˇ μ ) ( t ) ([\cdot,\cdot]_{\mu},\omega_{\mu},\check{g}_{\mu})(t) ([ ⋅ , ⋅ ] μ , ω μ , g ˇ μ ) ( t ) be a bracket flow on M M M solving (37–39). The connections and the shape operator evolve by
∂ t ( ∇ X a d μ ) = ∇ X a d μ ( Ric ^ μ − ∇ L ) − 2 ad μ ( ∇ ^ ∗ L ) − ∣ L ∣ 2 ∇ X a d μ , \partial_{t}(\nabla^{ad\mu}_{X})=\nabla^{ad\mu}_{X}(\widehat{\operatorname{Ric}}_{\mu}-\nabla L)-2\operatorname{ad}_{\mu}(\widehat{\nabla}^{*}L)-|L|^{2}\nabla^{ad\mu}_{X}, ∂ t ( ∇ X a d μ ) = ∇ X a d μ ( Ric μ − ∇ L ) − 2 ad μ ( ∇ ∗ L ) − ∣ L ∣ 2 ∇ X a d μ , ∂ t L = − ∇ X ( Ric ^ μ − ∇ L ) + 2 ad μ ( ∇ ^ ∗ L ) s y m − ∣ L ∣ 2 L , \partial_{t}L=-\nabla_{X}(\widehat{\operatorname{Ric}}_{\mu}-\nabla L)+2\operatorname{ad}_{\mu}(\widehat{\nabla}^{*}L)^{\mathrm{sym}}-|L|^{2}L, ∂ t L = − ∇ X ( Ric μ − ∇ L ) + 2 ad μ ( ∇ ∗ L ) sym − ∣ L ∣ 2 L , ∂ t ( ∇ X ) = − [ L , Ric ^ μ − ∇ L ] − 2 ad μ ( ∇ ^ ∗ L ) s k e w − ∣ L ∣ 2 ∇ X . \partial_{t}(\nabla_{X})=-[L,\widehat{\operatorname{Ric}}_{\mu}-\nabla L]-2\operatorname{ad}_{\mu}(\widehat{\nabla}^{*}L)^{\mathrm{skew}}-|L|^{2}\nabla_{X}. ∂ t ( ∇ X ) = − [ L , Ric μ − ∇ L ] − 2 ad μ ( ∇ ∗ L ) skew − ∣ L ∣ 2 ∇ X . Here X X X is a unit vector field on S 1 S^{1} S 1 , [ ⋅ , ⋅ ] [\cdot,\cdot] [ ⋅ , ⋅ ] is the commutation of endomorphisms of a d M adM a d M , and the superscripts sym and skew denote the symmetric and skew-symmetric components of an endomorphism of a d M adM a d M with respect to h h h .
Proof. If X ‾ \overline{X} X is horizontal and X X X is unit at t = 0 t=0 t = 0 , for this to persist, the vector field X ‾ \overline{X} X must evolve by
∂ t X ‾ = − 2 ∇ ^ ∗ L − ∣ L ∣ 2 X ‾ , ∂ t X = − ∣ L ∣ 2 X \partial_{t}\overline{X}=-2\widehat{\nabla}^{*}L-|L|^{2}\overline{X},\qquad\partial_{t}X=-|L|^{2}X ∂ t X = − 2 ∇ ∗ L − ∣ L ∣ 2 X , ∂ t X = − ∣ L ∣ 2 X
Fix a time-independent vertical vector U U U . To take the time derivative of ∇ X a d U a d \nabla^{\mathrm{ad}}_X U_{\mathrm{ad}} ∇ X ad U ad , we write
∇ X a d U a d = ω μ [ X ˉ , U ] μ , \nabla^{\mathrm{ad}}_X U_{\mathrm{ad}}=\omega_\mu[\bar{X},U]_\mu, ∇ X ad U ad = ω μ [ X ˉ , U ] μ ,
so then we get
∂ t ∇ X a d U a d = ∂ ω μ ∂ t [ X ˉ , U ] μ + ω μ ∂ [ ⋅ , ⋅ ] μ ∂ t ( X , U ) + ω μ [ ∂ t X ˉ , U ] μ = − 2 μ ( ∇ ^ ∗ L , U a d ) + ( ∇ X a d ( Ric ^ μ − ∇ L ) ) U a d − ∣ L ∣ 2 ∇ X a d U a d . \begin{aligned}
\partial_t\nabla^{\mathrm{ad}}_X U_{\mathrm{ad}}&=\frac{\partial\omega_\mu}{\partial t}[\bar{X},U]_\mu+\omega_\mu\frac{\partial[\cdot,\cdot]_\mu}{\partial t}(X,U)+\omega_\mu[\partial_t\bar{X},U]_\mu\\
&=-2\mu(\widehat{\nabla}^{*}L,U_{\mathrm{ad}})+\left(\nabla^{\mathrm{ad}}_X(\widehat{\operatorname{Ric}}_\mu-\nabla L)\right)U_{\mathrm{ad}}-\lvert L\rvert^2\nabla^{\mathrm{ad}}_XU_{\mathrm{ad}}.
\end{aligned} ∂ t ∇ X ad U ad = ∂ t ∂ ω μ [ X ˉ , U ] μ + ω μ ∂ t ∂ [ ⋅ , ⋅ ] μ ( X , U ) + ω μ [ ∂ t X ˉ , U ] μ = − 2 μ ( ∇ ∗ L , U ad ) + ( ∇ X ad ( Ric μ − ∇ L ) ) U ad − ∣ L ∣ 2 ∇ X ad U ad .
With respect to a chosen h h h -orthonormal frame, the connection matrix of ∇ a d \nabla^{\mathrm{ad}} ∇ ad splits into a skew-symmetric part and a symmetric part. The skew-symmetric part is the connection matrix of the metric connection ∇ \nabla ∇ . The symmetric part is − L -L − L . Because h h h does not depend on time, we get
∂ t ∇ X + ∣ L ∣ 2 ∇ X = ( ∂ t ∇ X a d + ∣ L ∣ 2 ∇ X a d ) s k e w , ∂ t L + ∣ L ∣ 2 L = − ( ∂ t ∇ X a d + ∣ L ∣ 2 ∇ X a d ) s y m \partial_t\nabla_X+\lvert L\rvert^2\nabla_X=\left(\partial_t\nabla^{\mathrm{ad}}_X+\lvert L\rvert^2\nabla^{\mathrm{ad}}_X\right)^{\mathrm{skew}},\qquad\partial_tL+\lvert L\rvert^2L=-\left(\partial_t\nabla^{\mathrm{ad}}_X+\lvert L\rvert^2\nabla^{\mathrm{ad}}_X\right)^{\mathrm{sym}} ∂ t ∇ X + ∣ L ∣ 2 ∇ X = ( ∂ t ∇ X ad + ∣ L ∣ 2 ∇ X ad ) skew , ∂ t L + ∣ L ∣ 2 L = − ( ∂ t ∇ X ad + ∣ L ∣ 2 ∇ X ad ) sym
in a fixed orthonormal frame. In particular, we get the splitting into symmetric and skew-symmetric parts
∇ X a d ( Ric ^ μ − ∇ L ) = ∇ X ( Ric ^ μ − ∇ L ) − [ L , Ric ^ μ − ∇ L ] . \nabla^{\mathrm{ad}}_X(\widehat{\operatorname{Ric}}_\mu-\nabla L)=\nabla_X(\widehat{\operatorname{Ric}}_\mu-\nabla L)-[L,\widehat{\operatorname{Ric}}_\mu-\nabla L]. ∇ X ad ( Ric μ − ∇ L ) = ∇ X ( Ric μ − ∇ L ) − [ L , Ric μ − ∇ L ] .
Lemma 4.2. Let M → S 1 M\to S^1 M → S 1 be a twisted principal G G G -bundle. Fix a vector bundle metric h h h on the adjoint bundle ad M \operatorname{ad}M ad M , and let ( [ ⋅ , ⋅ ] μ , ω μ , g ˇ μ ) ( t ) ([\cdot,\cdot]_\mu,\omega_\mu,\check{g}_\mu)(t) ([ ⋅ , ⋅ ] μ , ω μ , g ˇ μ ) ( t ) be a bracket flow on M M M solving (37–39). Take a vector ( U a d ) x ∈ ad M x (U_{\mathrm{ad}})_x\in\operatorname{ad}M_x ( U ad ) x ∈ ad M x and parallel transport it to a local section U a d U_{\mathrm{ad}} U ad via the metric connection ∇ ( t ) \nabla(t) ∇ ( t ) induced by the Levi-Civita connection. If U a d ( t ) U_{\mathrm{ad}}(t) U ad ( t ) evolves by
∂ t U a d = Q U a d , \partial_tU_{\mathrm{ad}}=QU_{\mathrm{ad}}, ∂ t U ad = Q U ad , where Q ( t ) Q(t) Q ( t ) is a skew-symmetric endomorphism of ( ad M , h ) (\operatorname{ad}M,h) ( ad M , h ) solving
∇ X Q = [ L , Ric ^ μ − ∇ L ] + 2 ( ad μ ( ∇ ^ ∗ L ) ) s k e w , Q ( x , t ) = 0 , \nabla_XQ=[L,\widehat{\operatorname{Ric}}_\mu-\nabla L]+2\left(\operatorname{ad}_\mu(\widehat{\nabla}^{*}L)\right)^{\mathrm{skew}},\qquad Q(x,t)=0, ∇ X Q = [ L , Ric μ − ∇ L ] + 2 ( ad μ ( ∇ ∗ L ) ) skew , Q ( x , t ) = 0 , then U a d ( t ) U_{\mathrm{ad}}(t) U ad ( t ) is ∇ ( t ) \nabla(t) ∇ ( t ) -parallel.
Proof. From Theorem 4.1 ,
∂ t ( ∇ X U a d ) = ∂ t ( ∇ X ) U a d + ∇ X ∂ t U a d = ∂ t ( ∇ X ) U a d + ∇ X ( Q U a d ) = ∂ t ( ∇ X ) U a d + ( ∇ X Q ) U a d + Q ∇ X U a d = − ∣ L ∣ 2 ∇ X U a d + Q ∇ X U a d . \begin{aligned}
\partial_t(\nabla_XU_{\mathrm{ad}})&=\partial_t(\nabla_X)U_{\mathrm{ad}}+\nabla_X\partial_tU_{\mathrm{ad}}\\
&=\partial_t(\nabla_X)U_{\mathrm{ad}}+\nabla_X(QU_{\mathrm{ad}})\\
&=\partial_t(\nabla_X)U_{\mathrm{ad}}+(\nabla_XQ)U_{\mathrm{ad}}+Q\nabla_XU_{\mathrm{ad}}\\
&=-\lvert L\rvert^2\nabla_XU_{\mathrm{ad}}+Q\nabla_XU_{\mathrm{ad}}.
\end{aligned} ∂ t ( ∇ X U ad ) = ∂ t ( ∇ X ) U ad + ∇ X ∂ t U ad = ∂ t ( ∇ X ) U ad + ∇ X ( Q U ad ) = ∂ t ( ∇ X ) U ad + ( ∇ X Q ) U ad + Q ∇ X U ad = − ∣ L ∣ 2 ∇ X U ad + Q ∇ X U ad .
This is a homogeneous linear ODE for ∇ X U a d \nabla_XU_{\mathrm{ad}} ∇ X U ad with initial value ∇ X U a d ( 0 ) = 0 \nabla_XU_{\mathrm{ad}}(0)=0 ∇ X U ad ( 0 ) = 0 , and thus, it will be zero for all time. Indeed, if U a d ( t ) U_{\mathrm{ad}}(t) U ad ( t ) is initially parallel, it will remain parallel along the bracket flow. □
Lie bracket estimate
Proposition 4.3. Let N n N^n N n be a nilpotent Lie group. Let M → S 1 M\to S^1 M → S 1 be a twisted principal N N N -bundle. Fix a vector bundle metric h h h on ad M \operatorname{ad}M ad M . If ( [ ⋅ , ⋅ ] μ , ω , g ˇ ) ( t ) ([\cdot,\cdot]_\mu,\omega,\check{g})(t) ([ ⋅ , ⋅ ] μ , ω , g ˇ ) ( t ) is a bracket flow on M M M solving (37–39), then for t > 0 t>0 t > 0 , the Lie bracket on the adjoint bundle ad M \operatorname{ad}M ad M satisfies
∣ μ ∣ 2 ≤ 2 n t . \lvert\mu\rvert^2\leq\frac{2n}{t}. ∣ μ ∣ 2 ≤ t 2 n .
Proof. Take X ( t ) X(t) X ( t ) to be a unit vector field on ( S 1 , g ˇ ( t ) ) (S^1,\check{g}(t)) ( S 1 , g ˇ ( t )) . All norms are given with respect to h h h . The square norm of the Lie bracket ∣ μ ∣ 2 \lvert\mu\rvert^2 ∣ μ ∣ 2 is a function of time and the base S 1 S^1 S 1 . By the Jacobi identity, the bracket μ \mu μ is ∇ a d μ \nabla^{\mathrm{ad}}\mu ∇ ad μ -parallel,
0 = ∇ X a d μ = ∇ X μ + δ μ ( L ) . 0=\nabla^{\mathrm{ad}}_X\mu=\nabla_X\mu+\delta_\mu(L). 0 = ∇ X ad μ = ∇ X μ + δ μ ( L ) .
In a ∇ ( t ) \nabla(t) ∇ ( t ) -parallel frame of ad M \operatorname{ad}M ad M , the Laplacian is
Δ S 1 ∣ μ ∣ 2 = X 2 ∣ μ ∣ 2 = 2 ∣ ∇ X μ ∣ 2 + 2 h ( ∇ X ∇ X μ , μ ) = 2 ∣ δ μ ( L ) ∣ 2 + 2 h ( δ δ μ ( L ) ( L ) − δ μ ∇ L , μ ) = 4 ∣ δ μ ( L ) ∣ 2 − 2 h ( δ μ ∇ L , μ ) . \Delta_{S^1}|\mu|^2 = X^2|\mu|^2
= 2|\nabla_X\mu|^2 + 2h(\nabla_X\nabla_X\mu,\mu)
= 2|\delta_\mu(L)|^2 + 2h(\delta_{\delta_\mu(L)}(L)-\delta_\mu\nabla L,\mu)
= 4|\delta_\mu(L)|^2 - 2h(\delta_\mu\nabla L,\mu). Δ S 1 ∣ μ ∣ 2 = X 2 ∣ μ ∣ 2 = 2∣ ∇ X μ ∣ 2 + 2 h ( ∇ X ∇ X μ , μ ) = 2∣ δ μ ( L ) ∣ 2 + 2 h ( δ δ μ ( L ) ( L ) − δ μ ∇ L , μ ) = 4∣ δ μ ( L ) ∣ 2 − 2 h ( δ μ ∇ L , μ ) .
In a fixed frame, the time derivative is
∂ ∂ t ∣ μ ∣ 2 = 2 h ( ∂ t μ , μ ) = 2 h ( δ μ ( Ric ^ μ ) , μ ) − 2 h ( δ μ ∇ L , μ ) . \frac{\partial}{\partial t}|\mu|^2 = 2h(\partial_t\mu,\mu)
= 2h(\delta_\mu(\widehat{\operatorname{Ric}}_\mu),\mu)-2h(\delta_\mu\nabla L,\mu). ∂ t ∂ ∣ μ ∣ 2 = 2 h ( ∂ t μ , μ ) = 2 h ( δ μ ( Ric μ ) , μ ) − 2 h ( δ μ ∇ L , μ ) .
Hence, the heat operator is
( ∂ ∂ t − Δ S 1 ) ∣ μ ∣ 2 = 2 h ( δ μ ( Ric ^ μ ) , μ ) − 4 ∣ δ μ ( L ) ∣ 2 = − 8 h ( M μ , Ric ^ μ ) − 4 ∣ δ μ ( L ) ∣ 2 . \left(\frac{\partial}{\partial t}-\Delta_{S^1}\right)|\mu|^2
=2h(\delta_\mu(\widehat{\operatorname{Ric}}_\mu),\mu)-4|\delta_\mu(L)|^2
=-8h(M_\mu,\widehat{\operatorname{Ric}}_\mu)-4|\delta_\mu(L)|^2. ( ∂ t ∂ − Δ S 1 ) ∣ μ ∣ 2 = 2 h ( δ μ ( Ric μ ) , μ ) − 4∣ δ μ ( L ) ∣ 2 = − 8 h ( M μ , Ric μ ) − 4∣ δ μ ( L ) ∣ 2 .
where M μ M_\mu M μ is the moment map. As N is nilpotent, the Ricci curvature of the fibres is Ric ^ μ = M μ \widehat{\operatorname{Ric}}_\mu=M_\mu Ric μ = M μ , while the scalar curvature of the fibres is R ^ μ = ⟨ M μ , Id ⟩ = − 1 4 ∣ μ ∣ 2 \widehat{R}_\mu=\langle M_\mu,\operatorname{Id}\rangle=-\frac{1}{4}|\mu|^2 R μ = ⟨ M μ , Id ⟩ = − 4 1 ∣ μ ∣ 2 . By Cauchy–Schwarz,
( ∂ ∂ t − Δ S 1 ) ∣ μ ∣ 2 ≤ − 8 ∣ M μ ∣ 2 ≤ − 8 n ⟨ M μ , Id ⟩ 2 = − 1 2 n ∣ μ ∣ 4 . \left(\frac{\partial}{\partial t}-\Delta_{S^1}\right)|\mu|^2
\leq-8|M_\mu|^2
\leq-\frac{8}{n}\langle M_\mu,\operatorname{Id}\rangle^2
=-\frac{1}{2n}|\mu|^4. ( ∂ t ∂ − Δ S 1 ) ∣ μ ∣ 2 ≤ − 8∣ M μ ∣ 2 ≤ − n 8 ⟨ M μ , Id ⟩ 2 = − 2 n 1 ∣ μ ∣ 4 .
By the maximum principle,
∣ μ ∣ 2 ≤ 2 n t . |\mu|^2\leq\frac{2n}{t}. ∣ μ ∣ 2 ≤ t 2 n .
Shape operator estimate
Proposition 4.4. Let N be a nilpotent Lie group. Let M → S 1 M\to S^1 M → S 1 be a twisted principal N-bundle. Fix a vector bundle metric h h h on ad M \operatorname{ad}M ad M . If ( [ ⋅ , ⋅ ] μ , ω , g ˇ ) ( t ) ([\cdot,\cdot]_\mu,\omega,\check{g})(t) ([ ⋅ , ⋅ ] μ , ω , g ˇ ) ( t ) is a bracket flow on M M M solving (37–39), then for t > 0 t>0 t > 0 ,
∣ L ∣ 2 ≤ 1 2 t . |L|^2\leq\frac{1}{2t}. ∣ L ∣ 2 ≤ 2 t 1 . Proof. Take X ( t ) X(t) X ( t ) to be a unit vector field on ( S 1 , g ˇ ( t ) ) (S^1,\check{g}(t)) ( S 1 , g ˇ ( t )) . All norms are given with respect to h h h . In a ∇ ( t ) \nabla(t) ∇ ( t ) -parallel frame, the Laplacian of the square norm is
Δ S 1 ∣ L ∣ 2 = X 2 ∣ L ∣ 2 = 2 ∣ ∇ L ∣ 2 + 2 h ( ∇ 2 L , L ) . \Delta_{S^1}|L|^2=X^2|L|^2=2|\nabla L|^2+2h(\nabla^2L,L). Δ S 1 ∣ L ∣ 2 = X 2 ∣ L ∣ 2 = 2∣∇ L ∣ 2 + 2 h ( ∇ 2 L , L ) . From Theorem 4.1 , in a fixed frame, the time derivative is
∂ ∂ t ∣ L ∣ 2 = 2 h ( ∂ t L , L ) = − 2 h ( ∇ Ric ^ μ , L ) + 2 h ( ∇ 2 L , L ) + 4 h ( ad μ ( ∇ ^ ∗ L ) , L ) − 2 ∣ L ∣ 4 . \frac{\partial}{\partial t}|L|^2=2h(\partial_tL,L)
=-2h(\nabla\widehat{\operatorname{Ric}}_\mu,L)+2h(\nabla^2L,L)+4h(\operatorname{ad}_\mu(\widehat{\nabla}^{*}L),L)-2|L|^4. ∂ t ∂ ∣ L ∣ 2 = 2 h ( ∂ t L , L ) = − 2 h ( ∇ Ric μ , L ) + 2 h ( ∇ 2 L , L ) + 4 h ( ad μ ( ∇ ∗ L ) , L ) − 2∣ L ∣ 4 . If ∇ ^ \widehat{\nabla} ∇ is the Levi-Civita connection of the fibres, by torsion-freeness,
ad μ ( ∇ ^ ∗ L ) = ∇ ^ ∇ ^ ∗ L − ∇ ^ ⋅ ( ∇ ^ ∗ L ) . \operatorname{ad}_\mu(\widehat{\nabla}^{*}L)=\widehat{\nabla}_{\widehat{\nabla}^{*}L}-\widehat{\nabla}_{\cdot}(\widehat{\nabla}^{*}L). ad μ ( ∇ ∗ L ) = ∇ ∇ ∗ L − ∇ ⋅ ( ∇ ∗ L ) . Since ∇ ^ ∇ ^ ∗ L ∈ End ( ad M ) \widehat{\nabla}_{\widehat{\nabla}^{*}L}\in\operatorname{End}(\operatorname{ad}M) ∇ ∇ ∗ L ∈ End ( ad M ) is skew-symmetric, we get
h ( ad μ ( ∇ ^ ∗ L ) , L ) = − h ( ∇ ^ ⋅ ∇ ^ ∗ L , L ) = − ∣ ∇ ^ ∗ L ∣ 2 + h ( ∇ ^ ∗ L , L H ) , h(\operatorname{ad}_\mu(\widehat{\nabla}^{*}L),L)
=-h(\widehat{\nabla}_{\cdot}\widehat{\nabla}^{*}L,L)
=-|\widehat{\nabla}^{*}L|^2+h(\widehat{\nabla}^{*}L,LH), h ( ad μ ( ∇ ∗ L ) , L ) = − h ( ∇ ⋅ ∇ ∗ L , L ) = − ∣ ∇ ∗ L ∣ 2 + h ( ∇ ∗ L , L H ) , where H = ∑ i ∇ ^ U ^ i U i H=\sum_i\widehat{\nabla}_{\widehat{U}_i}U_i H = ∑ i ∇ U i U i is the mean curvature vector of the fibres as homogeneous spaces. The heat operator is then
( ∂ ∂ t − Δ S 1 ) ∣ L ∣ 2 = − 2 h ( ∇ Ric ^ μ , L ) − 2 ∣ ∇ L ∣ 2 − 4 ∣ ∇ ^ ∗ L ∣ 2 + 4 h ( ∇ ^ ∗ L , L H ) − 2 ∣ L ∣ 4 . (41) \left(\frac{\partial}{\partial t}-\Delta_{S^1}\right)|L|^2
=-2h(\nabla\widehat{\operatorname{Ric}}_\mu,L)-2|\nabla L|^2-4|\widehat{\nabla}^{*}L|^2+4h(\widehat{\nabla}^{*}L,LH)-2|L|^4. \tag*{(41)} ( ∂ t ∂ − Δ S 1 ) ∣ L ∣ 2 = − 2 h ( ∇ Ric μ , L ) − 2∣∇ L ∣ 2 − 4∣ ∇ ∗ L ∣ 2 + 4 h ( ∇ ∗ L , L H ) − 2∣ L ∣ 4 . (41)
As N N N is nilpotent, H = 0 H=0 H = 0 and R i c ^ μ = M μ \widehat{\mathrm{Ric}}_\mu=M_\mu Ric μ = M μ . Moreover, by Theorem 3.5
− 2 h ( ∇ M μ , L ) = − ∣ δ μ ( L ) ∣ 2 . -2h(\nabla M_\mu,L)=-|\delta_\mu(L)|^2. − 2 h ( ∇ M μ , L ) = − ∣ δ μ ( L ) ∣ 2 .
The desired inequality follows from the maximum principle. □
Diameter estimate
Proposition 4.5. Let N N N be a nilpotent Lie group. Let M → S 1 M\to S^1 M → S 1 be a twisted principal N N N -bundle. Fix a vector bundle metric h h h on ad M \operatorname{ad}M ad M . If ( [ ⋅ , ⋅ ] μ , ω , g ˇ ) ( t ) ([\cdot,\cdot]_\mu,\omega,\check{g})(t) ([ ⋅ , ⋅ ] μ , ω , g ˇ ) ( t ) is a bracket flow on M M M solving {(37–39)}, then for t > 0 t>0 t > 0
diam ( S 1 , g ˇ ( t ) ) ≤ C 0 1 + t , \operatorname{diam}(S^1,\check{g}(t))\le C_0\sqrt{1+t}, diam ( S 1 , g ˇ ( t )) ≤ C 0 1 + t , where C 0 C_0 C 0 is some positive constant depending only on the initial metric.
Proof. By equation (41) in the proof of Theorem 4.4,
∣ L ∣ 2 ≤ 1 2 t + c 0 . |L|^2\le\frac{1}{2t+c_0}. ∣ L ∣ 2 ≤ 2 t + c 0 1 .
where c 0 − 1 = max { 1 , sup S 1 ∣ L ( 0 ) ∣ 2 } c_0^{-1}=\max\{1,\sup_{S^1}|L(0)|^2\} c 0 − 1 = max { 1 , sup S 1 ∣ L ( 0 ) ∣ 2 } . The derivative of the diameter is bounded by
d d t diam ( S 1 , g ˇ ( t ) ) = d d t 1 2 ∫ S 1 d s = 1 2 ∫ S 1 ∣ L ∣ 2 d s ≤ 1 2 t + c 0 diam ( S 1 , g ˇ ( t ) ) . \frac{d}{dt}\operatorname{diam}(S^1,\check{g}(t))
=\frac{d}{dt}\frac{1}{2}\int_{S^1}ds
=\frac{1}{2}\int_{S^1}|L|^2\,ds
\le\frac{1}{2t+c_0}\operatorname{diam}(S^1,\check{g}(t)). d t d diam ( S 1 , g ˇ ( t )) = d t d 2 1 ∫ S 1 d s = 2 1 ∫ S 1 ∣ L ∣ 2 d s ≤ 2 t + c 0 1 diam ( S 1 , g ˇ ( t )) .
By integrating, we get
diam ( S 1 , g ˇ ( t ) ) ≤ diam ( S 1 , g ˇ ( 0 ) ) 1 + 2 t c 0 ≤ C 0 1 + t . \operatorname{diam}(S^1,\check{g}(t))
\le\operatorname{diam}(S^1,\check{g}(0))\sqrt{1+\frac{2t}{c_0}}
\le C_0\sqrt{1+t}. diam ( S 1 , g ˇ ( t )) ≤ diam ( S 1 , g ˇ ( 0 )) 1 + c 0 2 t ≤ C 0 1 + t .
□
Jacobi operator estimate
Proposition 4.6. Let N n N^n N n be a nilpotent Lie group. Let M → S 1 M\to S^1 M → S 1 be a twisted principal N N N -bundle. Fix a vector bundle metric h h h on ad M \operatorname{ad}M ad M . If ( [ ⋅ , ⋅ ] μ , ω , g ˇ ) ( t ) ([\cdot,\cdot]_\mu,\omega,\check{g})(t) ([ ⋅ , ⋅ ] μ , ω , g ˇ ) ( t ) is a bracket flow on M M M solving {(37–39)}, then for t > 0 t>0 t > 0
∣ L ∣ 2 + ∣ ∇ L ∣ ≤ C t |L|^2+|\nabla L|\le\frac{C}{t} ∣ L ∣ 2 + ∣∇ L ∣ ≤ t C where C = C ( n ) > 0 C=C(n)>0 C = C ( n ) > 0 is some constant depending only on dimension.
Proof of Theorem 4.6. We apply the maximum principle to ∣ L ∣ 2 + ∣ ∇ L ∣ |L|^2+|\nabla L| ∣ L ∣ 2 + ∣∇ L ∣ . First, we compute the heat operator of ∣ ∇ L ∣ 2 |\nabla L|^2 ∣∇ L ∣ 2 . Take X ( t ) X(t) X ( t ) to be a unit vector field on ( S 1 , g ˇ ( t ) ) (S^1,\check{g}(t)) ( S 1 , g ˇ ( t )) . All norms are with respect to h h h . In a ∇ ( t ) \nabla(t) ∇ ( t ) -parallel frame, the Laplacian is
Δ S 1 ∣ ∇ L ∣ 2 = X 2 ∣ ∇ L ∣ 2 = 2 ∣ ∇ 2 L ∣ 2 + 2 h ( ∇ 3 L , ∇ L ) . \Delta_{S^1}|\nabla L|^2=X^2|\nabla L|^2=2|\nabla^2L|^2+2h(\nabla^3L,\nabla L). Δ S 1 ∣∇ L ∣ 2 = X 2 ∣∇ L ∣ 2 = 2∣ ∇ 2 L ∣ 2 + 2 h ( ∇ 3 L , ∇ L ) .
In a fixed frame, the time derivative is
∂ ∂ t ∣ ∇ L ∣ 2 = 2 h ( ∂ t ∇ L , ∇ L ) = 2 h ( ( ∂ t ∇ X ) L , ∇ L ) + 2 h ( ∇ X ∂ t L , ∇ L ) . \frac{\partial}{\partial t}|\nabla L|^2=2h(\partial_t\nabla L,\nabla L)=2h((\partial_t\nabla_X)L,\nabla L)+2h(\nabla_X\partial_tL,\nabla L). ∂ t ∂ ∣∇ L ∣ 2 = 2 h ( ∂ t ∇ L , ∇ L ) = 2 h (( ∂ t ∇ X ) L , ∇ L ) + 2 h ( ∇ X ∂ t L , ∇ L ) .
By Theorem 4.1, the connection evolves by
( ∂ t ∇ X ) L = − [ [ L , R i c ^ μ − ∇ L ] , L ] − 2 [ ad μ ( ∇ ^ ∗ L ) s k e w , L ] − ∣ L ∣ 2 ∇ L , (\partial_t\nabla_X)L=-[[L,\widehat{\mathrm{Ric}}_\mu-\nabla L],L]-2[\operatorname{ad}_\mu(\widehat{\nabla}^{*}L)^{\mathrm{skew}},L]-|L|^2\nabla L, ( ∂ t ∇ X ) L = − [[ L , Ric μ − ∇ L ] , L ] − 2 [ ad μ ( ∇ ∗ L ) skew , L ] − ∣ L ∣ 2 ∇ L ,
while the shape operator evolves by
∇ X ∂ t L = − ∇ 2 R i c ^ μ + ∇ 3 L + 2 ∇ X ad μ ( ∇ ^ ∗ L ) s y m − X ( ∣ L ∣ 2 ) L − ∣ L ∣ 2 ∇ L . \nabla_X\partial_tL=-\nabla^2\widehat{\mathrm{Ric}}_\mu+\nabla^3L+2\nabla_X\operatorname{ad}_\mu(\widehat{\nabla}^{*}L)^{\mathrm{sym}}-X(|L|^2)L-|L|^2\nabla L. ∇ X ∂ t L = − ∇ 2 Ric μ + ∇ 3 L + 2 ∇ X ad μ ( ∇ ∗ L ) sym − X ( ∣ L ∣ 2 ) L − ∣ L ∣ 2 ∇ L .
In a fixed frame, the time derivative is
∂ ∂ t ∣ ∇ L ∣ 2 = − 2 h ( [ [ L , R i c ^ μ ] , L ] , ∇ L ) − 2 ∣ [ L , ∇ L ] ∣ 2 − 4 h ( [ ad μ ( ∇ ^ ∗ L ) s k e w , L ] , ∇ L ) − 2 h ( ∇ 2 R i c ^ μ , ∇ L ) + 2 h ( ∇ 3 L , ∇ L ) + 4 h ( ∇ X ad μ ( ∇ ^ ∗ L ) , ∇ L ) − 4 h ( L , ∇ L ) 2 − 4 ∣ L ∣ 2 ∣ ∇ L ∣ 2 . \begin{aligned}
\frac{\partial}{\partial t}|\nabla L|^2={}&-2h([[L,\widehat{\mathrm{Ric}}_\mu],L],\nabla L)-2|[L,\nabla L]|^2-4h([\operatorname{ad}_\mu(\widehat{\nabla}^{*}L)^{\mathrm{skew}},L],\nabla L)\\
&-2h(\nabla^2\widehat{\mathrm{Ric}}_\mu,\nabla L)+2h(\nabla^3L,\nabla L)+4h(\nabla_X\operatorname{ad}_\mu(\widehat{\nabla}^{*}L),\nabla L)\\
&-4h(L,\nabla L)^2-4|L|^2|\nabla L|^2.
\end{aligned} ∂ t ∂ ∣∇ L ∣ 2 = − 2 h ([[ L , Ric μ ] , L ] , ∇ L ) − 2∣ [ L , ∇ L ] ∣ 2 − 4 h ([ ad μ ( ∇ ∗ L ) skew , L ] , ∇ L ) − 2 h ( ∇ 2 Ric μ , ∇ L ) + 2 h ( ∇ 3 L , ∇ L ) + 4 h ( ∇ X ad μ ( ∇ ∗ L ) , ∇ L ) − 4 h ( L , ∇ L ) 2 − 4∣ L ∣ 2 ∣∇ L ∣ 2 .
Since N n N^n N n is nilpotent, Ric ^ μ = M μ \widehat{\operatorname{Ric}}_\mu=M_\mu Ric μ = M μ , so ∣ Ric ^ μ ∣ ≤ C ∣ μ ∣ 2 \lvert\widehat{\operatorname{Ric}}_\mu\rvert\le C\lvert\mu\rvert^2 ∣ Ric μ ∣ ≤ C ∣ μ ∣ 2 for some C = C ( n ) > 0 C=C(n)>0 C = C ( n ) > 0 . By Koszul’s formula ∣ ∇ ∗ ^ L ∣ ≤ C ∣ μ ∣ ∣ L ∣ \lvert\widehat{\nabla^*}L\rvert\le C\lvert\mu\rvert\lvert L\rvert ∣ ∇ ∗ L ∣ ≤ C ∣ μ ∣ ∣ L ∣ , and so
− 2 h ( [ [ L , Ric ^ μ ] , L ] , ∇ L ) − 4 h ( [ ad μ ( ∇ ∗ ^ L ) s k e w , L ] , ∇ L ) ≤ C ∣ μ ∣ 2 ∣ L ∣ 2 ∣ ∇ L ∣ . -2h([[\!L,\widehat{\operatorname{Ric}}_\mu],L],\nabla L)-4h([\operatorname{ad}_\mu(\widehat{\nabla^*}L)^{\mathrm{skew}},L],\nabla L)\le C\lvert\mu\rvert^2\lvert L\rvert^2\lvert\nabla L\rvert. − 2 h ([[ L , Ric μ ] , L ] , ∇ L ) − 4 h ([ ad μ ( ∇ ∗ L ) skew , L ] , ∇ L ) ≤ C ∣ μ ∣ 2 ∣ L ∣ 2 ∣ ∇ L ∣ .
By Theorem 3.5, the second spatial variation of the moment map is
− 2 h ( ∇ 2 M μ , ∇ L ) = h ( δ δ μ ( L ) ( ∇ L ) , δ μ ( L ) ) − ∣ δ μ ∇ L ∣ 2 + h ( δ μ ( ∇ L ) , δ δ μ ( L ) ( L ) ) . -2h(\nabla^2M_\mu,\nabla L)=h(\delta_{\delta_\mu(L)}(\nabla L),\delta_\mu(L))-\lvert\delta_\mu\nabla L\rvert^2+h(\delta_\mu(\nabla L),\delta_{\delta_\mu(L)}(L)). − 2 h ( ∇ 2 M μ , ∇ L ) = h ( δ δ μ ( L ) ( ∇ L ) , δ μ ( L )) − ∣ δ μ ∇ L ∣ 2 + h ( δ μ ( ∇ L ) , δ δ μ ( L ) ( L )) .
Clearly, ∣ δ μ ( L ) ∣ ≤ 3 ∣ L ∣ ∣ μ ∣ \lvert\delta_\mu(L)\rvert\le3\lvert L\rvert\lvert\mu\rvert ∣ δ μ ( L )∣ ≤ 3 ∣ L ∣ ∣ μ ∣ , so by Cauchy–Schwarz
− 2 h ( ∇ 2 M μ , ∇ L ) ≤ 54 ∣ μ ∣ 2 ∣ L ∣ 2 ∣ ∇ L ∣ − ∣ δ μ ( ∇ L ) ∣ 2 . -2h(\nabla^2M_\mu,\nabla L)\le54\lvert\mu\rvert^2\lvert L\rvert^2\lvert\nabla L\rvert-\lvert\delta_\mu(\nabla L)\rvert^2. − 2 h ( ∇ 2 M μ , ∇ L ) ≤ 54 ∣ μ ∣ 2 ∣ L ∣ 2 ∣ ∇ L ∣ − ∣ δ μ ( ∇ L ) ∣ 2 .
Next, as ∇ μ = − δ μ ( L ) \nabla\mu=-\delta_\mu(L) ∇ μ = − δ μ ( L ) ,
∣ ∇ ∇ ∗ ^ L ∣ ≤ C ( ∣ ∇ μ ∣ ∣ L ∣ + ∣ μ ∣ ∣ ∇ L ∣ ) ≤ C ( 3 ∣ μ ∣ ∣ L ∣ 2 + ∣ μ ∣ ∣ ∇ L ∣ ) . \begin{aligned}
\lvert\nabla\widehat{\nabla^*}L\rvert\le C(\lvert\nabla\mu\rvert\lvert L\rvert+\lvert\mu\rvert\lvert\nabla L\rvert) \\
&\le C(3\lvert\mu\rvert\lvert L\rvert^2+\lvert\mu\rvert\lvert\nabla L\rvert).
\end{aligned} ∣ ∇ ∇ ∗ L ∣ ≤ C (∣ ∇ μ ∣ ∣ L ∣ + ∣ μ ∣ ∣ ∇ L ∣) ≤ C ( 3 ∣ μ ∣ ∣ L ∣ 2 + ∣ μ ∣ ∣ ∇ L ∣) .
We know that
∇ X ad μ ( ∇ ∗ ^ L ) = ad ∇ X μ ( ∇ ∗ ^ L ) + ad μ ( ∇ X ∇ ∗ ^ L ) , \nabla_X\operatorname{ad}_\mu(\widehat{\nabla^*}L)=\operatorname{ad}_{\nabla_X\mu}(\widehat{\nabla^*}L)+\operatorname{ad}_\mu(\nabla_X\widehat{\nabla^*}L), ∇ X ad μ ( ∇ ∗ L ) = ad ∇ X μ ( ∇ ∗ L ) + ad μ ( ∇ X ∇ ∗ L ) ,
so
∣ ∇ X ad μ ( ∇ ∗ ^ L ) ∣ ≤ ∣ ∇ μ ∣ ∣ ∇ ∗ ^ L ∣ + ∣ μ ∣ ∣ ∇ X ∇ ∗ ^ L ∣ ≤ C ∣ μ ∣ 2 ( ∣ L ∣ 2 + ∣ ∇ L ∣ ) . \begin{aligned}
\lvert\nabla_X\operatorname{ad}_\mu(\widehat{\nabla^*}L)\rvert\le\lvert\nabla\mu\rvert\lvert\widehat{\nabla^*}L\rvert+\lvert\mu\rvert\lvert\nabla_X\widehat{\nabla^*}L\rvert\\
&\le C\lvert\mu\rvert^2(\lvert L\rvert^2+\lvert\nabla L\rvert).
\end{aligned} ∣ ∇ X ad μ ( ∇ ∗ L )∣ ≤ ∣ ∇ μ ∣ ∣ ∇ ∗ L ∣ + ∣ μ ∣ ∣ ∇ X ∇ ∗ L ∣ ≤ C ∣ μ ∣ 2 (∣ L ∣ 2 + ∣ ∇ L ∣) .
By Propositions 4.3 and 4.4, the heat operator of the square norm is bounded by
( ∂ ∂ t − Δ S 1 ) ∣ ∇ L ∣ 2 ≤ − ∣ δ μ ( ∇ L ) ∣ 2 + C ∣ μ ∣ 2 ∣ ∇ L ∣ ( ∣ L ∣ 2 + ∣ ∇ L ∣ ) − 4 h ( L , ∇ L ) 2 − 4 ∣ L ∣ 2 ∣ ∇ L ∣ 2 − 2 ∣ ∇ 2 L ∣ 2 ≤ − ∣ δ μ ( ∇ L ) ∣ 2 − 4 h ( L , ∇ L ) 2 − 4 ∣ L ∣ 2 ∣ ∇ L ∣ 2 − 2 ∣ ∇ 2 L ∣ 2 + C t ∣ ∇ L ∣ ( ∣ L ∣ 2 + ∣ ∇ L ∣ ) . \begin{aligned}
\left(\frac{\partial}{\partial t}-\Delta_{S^1}\right)\lvert\nabla L\rvert^2\le-\lvert\delta_\mu(\nabla L)\rvert^2+C\lvert\mu\rvert^2\lvert\nabla L\rvert(\lvert L\rvert^2+\lvert\nabla L\rvert)-4h(L,\nabla L)^2-4\lvert L\rvert^2\lvert\nabla L\rvert^2-2\lvert\nabla^2L\rvert^2 \\
&\le-\lvert\delta_\mu(\nabla L)\rvert^2-4h(L,\nabla L)^2-4\lvert L\rvert^2\lvert\nabla L\rvert^2-2\lvert\nabla^2L\rvert^2+\frac{C}{t}\lvert\nabla L\rvert(\lvert L\rvert^2+\lvert\nabla L\rvert).
\end{aligned} ( ∂ t ∂ − Δ S 1 ) ∣ ∇ L ∣ 2 ≤ − ∣ δ μ ( ∇ L ) ∣ 2 + C ∣ μ ∣ 2 ∣ ∇ L ∣ (∣ L ∣ 2 + ∣ ∇ L ∣) − 4 h ( L , ∇ L ) 2 − 4 ∣ L ∣ 2 ∣ ∇ L ∣ 2 − 2 ∣ ∇ 2 L ∣ 2 ≤ − ∣ δ μ ( ∇ L ) ∣ 2 − 4 h ( L , ∇ L ) 2 − 4 ∣ L ∣ 2 ∣ ∇ L ∣ 2 − 2 ∣ ∇ 2 L ∣ 2 + t C ∣ ∇ L ∣ (∣ L ∣ 2 + ∣ ∇ L ∣) .
The heat operator of the norm is then
( ∂ ∂ t − Δ S 1 ) ∣ ∇ L ∣ ≤ − ∣ δ μ ( ∇ L ) ∣ 2 2 ∣ ∇ L ∣ − 2 h ( L , ∇ L ) 2 ∣ ∇ L ∣ − 2 ∣ L ∣ 2 ∣ ∇ L ∣ + C t ( ∣ L ∣ 2 + ∣ ∇ L ∣ ) , \left(\frac{\partial}{\partial t}-\Delta_{S^1}\right)\lvert\nabla L\rvert\le-\frac{\lvert\delta_\mu(\nabla L)\rvert^2}{2\lvert\nabla L\rvert}-\frac{2h(L,\nabla L)^2}{\lvert\nabla L\rvert}-2\lvert L\rvert^2\lvert\nabla L\rvert+\frac{C}{t}(\lvert L\rvert^2+\lvert\nabla L\rvert), ( ∂ t ∂ − Δ S 1 ) ∣ ∇ L ∣ ≤ − 2 ∣ ∇ L ∣ ∣ δ μ ( ∇ L ) ∣ 2 − ∣ ∇ L ∣ 2 h ( L , ∇ L ) 2 − 2 ∣ L ∣ 2 ∣ ∇ L ∣ + t C (∣ L ∣ 2 + ∣ ∇ L ∣) ,
and by adding (41) and positive terms, we get
( ∂ ∂ t − Δ S 1 ) ( ∣ L ∣ 2 + ∣ ∇ L ∣ ) ≤ − 2 ∣ ∇ L ∣ 2 − 2 ∣ L ∣ 2 ∣ ∇ L ∣ − 2 ∣ L ∣ 4 + C t ( ∣ L ∣ 2 + ∣ ∇ L ∣ ) . \left(\frac{\partial}{\partial t}-\Delta_{S^1}\right)(\lvert L\rvert^2+\lvert\nabla L\rvert)\le-2\lvert\nabla L\rvert^2-2\lvert L\rvert^2\lvert\nabla L\rvert-2\lvert L\rvert^4+\frac{C}{t}(\lvert L\rvert^2+\lvert\nabla L\rvert). ( ∂ t ∂ − Δ S 1 ) (∣ L ∣ 2 + ∣ ∇ L ∣) ≤ − 2 ∣ ∇ L ∣ 2 − 2 ∣ L ∣ 2 ∣ ∇ L ∣ − 2 ∣ L ∣ 4 + t C (∣ L ∣ 2 + ∣ ∇ L ∣) .
By the maximum principle, ∣ L ∣ 2 + ∣ ∇ L ∣ \lvert L\rvert^2+\lvert\nabla L\rvert ∣ L ∣ 2 + ∣ ∇ L ∣ is bounded by solutions to
d y d t = − y 2 + C t y , \frac{dy}{dt}=-y^2+\frac{C}{t}y, d t d y = − y 2 + t C y ,
with initial value greater than ∣ L ∣ 2 + ∣ ∇ L ∣ \lvert L\rvert^2+\lvert\nabla L\rvert ∣ L ∣ 2 + ∣ ∇ L ∣ at t = ε > 0 t=\varepsilon>0 t = ε > 0 . Fix some c 0 > sup S 1 × [ 0 , ε ] ( ∣ L ∣ 2 + ∣ ∇ L ∣ ) c_0>\sup_{S^1\times[0,\varepsilon]}(\lvert L\rvert^2+\lvert\nabla L\rvert) c 0 > sup S 1 × [ 0 , ε ] (∣ L ∣ 2 + ∣ ∇ L ∣) . The solution to the ODE with initial value y ( ε ) = c 0 y(\varepsilon)=c_0 y ( ε ) = c 0 is
y = C + 1 t + ( C + 1 c 0 − ε ) ( t ε ) − C . y=\frac{C+1}{t+\left(\frac{C+1}{c_0}-\varepsilon\right)\left(\frac{t}{\varepsilon}\right)^{-C}}. y = t + ( c 0 C + 1 − ε ) ( ε t ) − C C + 1 .
For small ε > 0 \varepsilon>0 ε > 0 , we get
y ≤ C + 1 t y\le\frac{C+1}{t} y ≤ t C + 1
where C = C ( n ) > 0 C=C(n)>0 C = C ( n ) > 0 . The desired inequality for ∣ ∇ L + L 2 ∣ \lvert\nabla L+L^2\rvert ∣ ∇ L + L 2 ∣ follows from the triangle inequality. □
This completes the proof of Theorem A.
Convergence
In this section, we prove Theorem C and Theorem D. First, we give compactness results for Ricci flows of invariant metrics on twisted principal bundles with simply-connected nilpotent fibres and S 1 S^1 S 1 base.
Compactness
Theorem 5.1. Fix n ∈ N n \in\mathbb{N} n ∈ N and K , d > 0 K,d>0 K , d > 0 . Let I ⊆ R I \subseteq\mathbb{R} I ⊆ R be an open interval containing 0. Let M k → S 1 M_k \to S^1 M k → S 1 be a sequence of twisted principal N k N_k N k -bundles where N k N_k N k is nilpotent, simply-connected and dim N k = n \dim N_k=n dim N k = n . For every k ∈ N k \in\mathbb{N} k ∈ N , let g k ( t ) g_k(t) g k ( t ) be a normalised Ricci flow of invariant metrics on M k M_k M k defined for t ∈ I t \in I t ∈ I such that
∣ Rm g k ( t ) ∣ g k ( t ) ≤ K , diam ( S 1 , g ˇ k ( t ) ) ≤ d . \lvert\operatorname{Rm}_{g_k(t)}\rvert_{g_k(t)} \leq K,\qquad\operatorname{diam}(S^1,\check{g}_k(t)) \leq d. ∣ Rm g k ( t ) ∣ g k ( t ) ≤ K , diam ( S 1 , g ˇ k ( t )) ≤ d . Take a sequence of points in the universal covers p ~ k ∈ M ~ k \widetilde{p}_k \in\widetilde{M}_k p k ∈ M k .
(a) Then, the lifted flows g ~ k ( t ) \widetilde{g}_k(t) g k ( t ) subconverge to a normalised Ricci flow
( M ~ k , g ~ k ( t ) , p ~ k ) → ( M ~ ∞ , g ~ ∞ ( t ) , p ~ ∞ ) (\widetilde{M}_k,\widetilde{g}_k(t),\widetilde{p}_k) \to(\widetilde{M}_\infty,\widetilde{g}_\infty(t),\widetilde{p}_\infty) ( M k , g k ( t ) , p k ) → ( M ∞ , g ∞ ( t ) , p ∞ ) in the pointed smooth Cheeger–Gromov topology on compact time intervals in I I I , and M ~ ∞ \widetilde{M}_\infty M ∞ is a principal N ∞ N_\infty N ∞ -bundle over R \mathbb{R} R for some simply-connected nilpotent Lie group N ∞ N_\infty N ∞ , and g ~ ∞ ( t ) \widetilde{g}_\infty(t) g ∞ ( t ) is N ∞ N_\infty N ∞ -invariant.
(b) If in addition, there exists a constant c > 0 c>0 c > 0 such that for all k ∈ N k \in\mathbb{N} k ∈ N ,
0 < c ≤ diam ( S 1 , g ˇ k ( t ) ) , 0<c\leq\operatorname{diam}(S^1,\check{g}_k(t)), 0 < c ≤ diam ( S 1 , g ˇ k ( t )) , then ( M ~ ∞ , g ~ ∞ ( t ) ) (\widetilde{M}_\infty,\widetilde{g}_\infty(t)) ( M ∞ , g ∞ ( t )) is the lift to the universal cover of some normalised Ricci flow of invariant metrics g ∞ ( t ) g_\infty(t) g ∞ ( t ) on a twisted principal N ∞ N_\infty N ∞ -bundle M ∞ → S 1 M_\infty\to S^1 M ∞ → S 1 .
(c) Assuming also that there exists a constant l > 0 l>0 l > 0 and generators τ k \tau_k τ k of the deck transformation groups M ~ k → M k \widetilde{M}_k \to M_k M k → M k such that
dist g ~ k ( 0 ) ( p ~ k , τ k ( p ~ k ) ) ≤ l , \operatorname{dist}_{\widetilde{g}_k(0)}(\widetilde{p}_k,\tau_k(\widetilde{p}_k))\leq l, dist g k ( 0 ) ( p k , τ k ( p k )) ≤ l , then, the flows g k ( t ) g_k(t) g k ( t ) subconverge
( M k , g k ( t ) , p k ) → ( M ∞ , g ∞ ( t ) , p ∞ ) (M_k,g_k(t),p_k) \to(M_\infty,g_\infty(t),p_\infty) ( M k , g k ( t ) , p k ) → ( M ∞ , g ∞ ( t ) , p ∞ ) in the pointed smooth Cheeger–Gromov topology on compact time intervals in I I I . Here M ∞ → S 1 M_\infty\to S^1 M ∞ → S 1 is a twisted principal N ∞ N_\infty N ∞ -bundle and g ∞ ( t ) g_\infty(t) g ∞ ( t ) is a normalised Ricci flow of invariant metrics whose lift to the universal cover is ( M ~ ∞ , g ~ ∞ ( t ) ) (\widetilde{M}_\infty,\widetilde{g}_\infty(t)) ( M ∞ , g ∞ ( t )) .
Proof of Theorem 5.1. The metrics g k ( t ) g_k(t) g k ( t ) are complete as the base is compact and the metrics on the fibres are homogeneous. The lifts to the universal cover g ~ k ( t ) \widetilde{g}_k(t) g k ( t ) are complete too.
Part (a). First, we prove an injectivity radius bound on ( M ~ k , g ~ k ( 0 ) ) (\widetilde{M}_k,\widetilde{g}_k(0)) ( M k , g k ( 0 )) and then apply Hamilton’s compactness theorem [14 ] . For brevity, write g ~ k : = g ~ k ( 0 ) \widetilde{g}_k:=\widetilde{g}_k(0) g k := g k ( 0 ) , and use horizontal transport to identify
( M ~ k , g ~ k , p ~ k ) ≅ ( R × N k , d s 2 + h k ( s ) , ( 0 , e ) ) . (42) (\widetilde{M}_k,\widetilde{g}_k,\widetilde{p}_k)\cong(\mathbb{R}\times N_k,ds^2+h_k(s),(0,e)). \tag*{(42)} ( M k , g k , p k ) ≅ ( R × N k , d s 2 + h k ( s ) , ( 0 , e )) . (42)
where h k ( s ) h_k(s) h k ( s ) is a one-parameter family of right-invariant metrics on N k N_k N k . The Lie exponential maps of N k N_k N k are diffeomorphisms. As ad X \operatorname{ad}_X ad X is nilpotent for all X ∈ n k X\in\mathfrak{n}_k X ∈ n k , the differential after right translation back to the identity
( R exp ( − X ) ) ∗ d ( exp ) X = e ad X − Id ad X = Id + 1 2 ad X + 1 6 ( ad X ) 2 + ⋯ (R_{\exp(-X)})_*d(\exp)_X=\frac{e^{\operatorname{ad}_X}-\operatorname{Id}}{\operatorname{ad}_X}=\operatorname{Id}+\frac{1}{2}\operatorname{ad}_X+\frac{1}{6}(\operatorname{ad}_X)^2+\cdots ( R e x p ( − X ) ) ∗ d ( exp ) X = ad X e ad X − Id = Id + 2 1 ad X + 6 1 ( ad X ) 2 + ⋯
[32 Theorem 2.14.3] is unipotent. For all r > 0 r>0 r > 0 ,
Vol h exp ( B ( 0 , r ) ) = ∫ B ( 0 , r ) ∣ det d ( exp ) X ∣ d x = Vol ( n , h e ) B ( 0 , r ) = ω n r n \operatorname{Vol}_{h}\exp(B(0,r))=\int_{B(0,r)}\lvert\det d(\exp)_X\rvert\,dx=\operatorname{Vol}_{(\mathfrak{n},h_e)}B(0,r)=\omega_n r^n Vol h exp ( B ( 0 , r )) = ∫ B ( 0 , r ) ∣ det d ( exp ) X ∣ d x = Vol ( n , h e ) B ( 0 , r ) = ω n r n
where d x dx d x is the Euclidean measure on ( n , h e ) (\mathfrak{n},h_e) ( n , h e ) , and ω n \omega_n ω n is the volume of the n n n -dimensional Euclidean unit ball. Given X ∈ n X\in\mathfrak{n} X ∈ n , by right-invariance the length of the curve γ X ( t ) = exp ( t X ) \gamma_X(t)=\exp(tX) γ X ( t ) = exp ( tX ) satisfies
dist h ( e , exp X ) ≤ Length h ( γ X ) = ∫ 0 1 ∣ γ ˙ X ( t ) ∣ h d t = ∫ 0 1 ∣ ( R exp ( t X ) ) ∗ X ∣ d t = ∫ 0 1 ∣ X ∣ h e d t = ∣ X ∣ h e . \operatorname{dist}_{h}(e,\exp X)\leq\operatorname{Length}_{h}(\gamma_X)=\int_0^1\lvert\dot{\gamma}_X(t)\rvert_h\,dt=\int_0^1\lvert(R_{\exp(tX)})_*X\rvert\,dt=\int_0^1\lvert X\rvert_{h_e}\,dt=\lvert X\rvert_{h_e}. dist h ( e , exp X ) ≤ Length h ( γ X ) = ∫ 0 1 ∣ γ ˙ X ( t ) ∣ h d t = ∫ 0 1 ∣( R e x p ( tX ) ) ∗ X ∣ d t = ∫ 0 1 ∣ X ∣ h e d t = ∣ X ∣ h e .
In particular,
Vol h B ( e , r ) ≥ Vol h exp ( B ( 0 , r ) ) = ω n r n . \operatorname{Vol}_{h}B(e,r)\geq\operatorname{Vol}_{h}\exp(B(0,r))=\omega_n r^n. Vol h B ( e , r ) ≥ Vol h exp ( B ( 0 , r )) = ω n r n .
Set r 0 = min { 1 , π 4 K } r_0=\min\left\{1,\frac{\pi}{4\sqrt{K}}\right\} r 0 = min { 1 , 4 K π } and define the regions
Ω k = { ( s , exp X ) : ∣ s ∣ < r 0 / 2 , ∣ X ∣ h k ( s ) < r 0 / 2 } . \Omega_k=\left\{(s,\exp X):\lvert s\rvert<r_0/2,\ \lvert X\rvert_{h_k(s)}<r_0/2\right\}. Ω k = { ( s , exp X ) : ∣ s ∣ < r 0 /2 , ∣ X ∣ h k ( s ) < r 0 /2 } .
As Ω k ⊆ B g ~ k ( p ~ k , r 0 ) \Omega_k \subseteq B_{\widetilde{g}_k}(\widetilde{p}_k,r_0) Ω k ⊆ B g k ( p k , r 0 ) ,
Vol g ~ k B g ~ k ( p ~ k , r 0 ) ≥ Vol g ~ k Ω k = ∫ − r 0 / 2 r 0 / 2 Vol h k ( s ) B h k ( s ) ( 0 , r 0 / 2 ) exp ( B h k ( s ) ( 0 , r 0 / 2 ) ) d s = ω n ( r 0 2 ) n r 0 . \operatorname{Vol}_{\widetilde{g}_k} B_{\widetilde{g}_k}(\widetilde{p}_k,r_0) \ge\operatorname{Vol}_{\widetilde{g}_k} \Omega_k = \int_{-r_0/2}^{r_0/2} \operatorname{Vol}_{h_k(s)} B_{h_k(s)}(0,r_0/2) \exp\left(B_{h_k(s)}(0,r_0/2)\right)\,ds = \omega_n\left(\frac{r_0}{2}\right)^n r_0. Vol g k B g k ( p k , r 0 ) ≥ Vol g k Ω k = ∫ − r 0 /2 r 0 /2 Vol h k ( s ) B h k ( s ) ( 0 , r 0 /2 ) exp ( B h k ( s ) ( 0 , r 0 /2 ) ) d s = ω n ( 2 r 0 ) n r 0 .
From the curvature bound, the Cheeger–Gromov–Taylor injectivity radius estimate [8 Thm. 4.7] gives
inj ( p ~ k , g ~ k ) ≥ r 0 2 1 1 + ( V 2 r 0 − K / Vol g ~ k B g ~ k ( p ~ k , r 0 ) ) \operatorname{inj}(\widetilde{p}_k,\widetilde{g}_k) \ge\frac{r_0}{2}\frac{1}{1+\left(V_{2r_0}^{-K}/\operatorname{Vol}_{\widetilde{g}_k} B_{\widetilde{g}_k}(\widetilde{p}_k,r_0)\right)} inj ( p k , g k ) ≥ 2 r 0 1 + ( V 2 r 0 − K / Vol g k B g k ( p k , r 0 ) ) 1
where V r − K V_r^{-K} V r − K is the volume of a ball of radius r r r in the simply connected space form of constant curvature − K -K − K . This provides a lower bound
inj ( p ~ k , g ~ k ) ≥ r 0 2 1 1 + ( V 2 r 0 − K / ω n ( r 0 2 ) n r 0 ) = : i ( n , K ) > 0. \operatorname{inj}(\widetilde{p}_k,\widetilde{g}_k) \ge\frac{r_0}{2}\frac{1}{1+\left(V_{2r_0}^{-K}/\omega_n\left(\frac{r_0}{2}\right)^n r_0\right)} =: i(n,K)>0. inj ( p k , g k ) ≥ 2 r 0 1 + ( V 2 r 0 − K / ω n ( 2 r 0 ) n r 0 ) 1 =: i ( n , K ) > 0.
By Hamilton’s compactness theorem [14 ] and Shi’s derivative estimates [29 ] , there is a subsequence such that
( M ~ k , g ~ k ( t ) , p ~ k ) → ( M ~ ∞ , g ~ ∞ ( t ) , p ~ ∞ ) (43) (\widetilde{M}_k,\widetilde{g}_k(t),\widetilde{p}_k) \to(\widetilde{M}_\infty,\widetilde{g}_\infty(t),\widetilde{p}_\infty) \tag*{(43)} ( M k , g k ( t ) , p k ) → ( M ∞ , g ∞ ( t ) , p ∞ ) (43)
in the pointed smooth Cheeger–Gromov topology on compact time intervals. Here g ~ ∞ ( t ) \widetilde{g}_\infty(t) g ∞ ( t ) is a complete normalised Ricci flow.
Fix t = 0 t=0 t = 0 and take g ~ k : = g ~ k ( 0 ) \widetilde{g}_k:=\widetilde{g}_k(0) g k := g k ( 0 ) . On M ~ k \widetilde{M}_k M k there is a global right N k N_k N k -action, and hence a Lie algebra of globally defined Killing fields of ( M ~ k , g ~ k ) (\widetilde{M}_k,\widetilde{g}_k) ( M k , g k ) . For every k k k , choose a vertical basis of Killing fields
n k : = span { V 1 , k # , … , V n , k # } (44) \mathfrak{n}_k := \operatorname{span}\{V_{1,k}^{\#},\ldots,V_{n,k}^{\#}\} \tag*{(44)} n k := span { V 1 , k # , … , V n , k # } (44)
that are g ~ k \widetilde{g}_k g k -orthonormal at p ~ k \widetilde{p}_k p k . The 1-jet norm of a vector field is
∣ U ∣ ∗ : = ( ∣ U ∣ 2 + ∣ ∇ U ∣ 2 ) 1 / 2 . |U|^{*}:=\left(|U|^2+|\nabla U|^2\right)^{1/2}. ∣ U ∣ ∗ := ( ∣ U ∣ 2 + ∣∇ U ∣ 2 ) 1/2 .
If U # U^{\#} U # is a vertical Killing field that is non-zero at p ∈ M p\in M p ∈ M , then its 1-jet norm satisfies the inequality
∣ U # ∣ p ∗ ≤ ( 1 + 2 ∣ L ∣ p 2 + ∣ A ∣ p 2 + 9 4 ∣ μ ∣ p 2 ) 1 / 2 ∣ U # ( p ) ∣ (45) |U^{\#}|_p^{*}\le\left(1+2|L|_p^2+|A|_p^2+\frac{9}{4}|\mu|_p^2\right)^{1/2}|U^{\#}(p)| \tag*{(45)} ∣ U # ∣ p ∗ ≤ ( 1 + 2∣ L ∣ p 2 + ∣ A ∣ p 2 + 4 9 ∣ μ ∣ p 2 ) 1/2 ∣ U # ( p ) ∣ (45)
where L L L is the shape operator of the fibres, A A A is O’Neill’s integrability tensor, and μ \mu μ is the bracket of the Lie algebra of Killing fields. As A = 0 A=0 A = 0 for all ( M ~ k , g ~ k ( t ) ) (\widetilde{M}_k,\widetilde{g}_k(t)) ( M k , g k ( t )) , to show that the 1-jet norms ∣ V i , k # ∣ p ~ k ∗ |V_{i,k}^{\#}|_{\widetilde{p}_k}^{*} ∣ V i , k # ∣ p k ∗ are bounded, we must bound the shape operators and the structure constants
( μ k ) i j m V m , k # = [ V i , k # , V j , k # ] . (\mu_k)_{ij}^{m}V_{m,k}^{\#}=[V_{i,k}^{\#},V_{j,k}^{\#}]. ( μ k ) ij m V m , k # = [ V i , k # , V j , k # ] .
Again, fix t = 0 t=0 t = 0 and take g ~ k : = g ~ k ( 0 ) \widetilde{g}_k:=\widetilde{g}_k(0) g k := g k ( 0 ) . In the identification (42), let ∇ k \nabla^k ∇ k be the Levi-Civita connections. The shape operators of the fibres are
L k U = ∇ U k ∂ s L_kU=\nabla_U^k\partial_s L k U = ∇ U k ∂ s
where U U U is vertical. The shape operator satisfies the Riccati equation
∇ ∂ s k L k + L k 2 + R k ( ⋅ , ∂ s ) ∂ s = 0. (46) \nabla_{\partial_s}^k L_k+L_k^2+R_k(\cdot,\partial_s)\partial_s=0. \tag*{(46)} ∇ ∂ s k L k + L k 2 + R k ( ⋅ , ∂ s ) ∂ s = 0. (46)
Along a horizontal geodesic, take a parallel vertical unit vector field V V V . The function φ ( s ) = h k ( L k V , V ) \varphi(s)=h_k(L_kV,V) φ ( s ) = h k ( L k V , V ) is defined for all s ∈ R s\in\mathbb{R} s ∈ R and satisfies
d φ d s ≤ K − φ 2 . \frac{d\varphi}{ds}\le K-\varphi^2. d s d φ ≤ K − φ 2 .
If φ ( s 0 ) < − K \varphi(s_0)<-\sqrt{K} φ ( s 0 ) < − K , comparison with ψ ′ = K − ψ 2 \psi'=K-\psi^2 ψ ′ = K − ψ 2 forces blow-up in finite positive time, contradicting smoothness on all of R \mathbb{R} R . If φ ( s 0 ) > K \varphi(s_0)>\sqrt{K} φ ( s 0 ) > K , then a contradiction follows by applying the previous argument to − φ ( 2 s 0 − s ) -\varphi(2s_0-s) − φ ( 2 s 0 − s ) . Hence,
− K ≤ h k ( L k V , V ) ≤ K (47) -\sqrt{K}\le h_k(L_kV,V)\le\sqrt{K} \tag*{(47)} − K ≤ h k ( L k V , V ) ≤ K (47)
for all vertical unit vectors V V V . The scalar curvatures of the nilpotent fibres are
scal ( h k ( s ) ) = − 1 4 ∣ μ k ∣ h k ( s ) 2 . \operatorname{scal}(h_k(s))=-\frac{1}{4}|\mu_k|_{h_k(s)}^2. scal ( h k ( s )) = − 4 1 ∣ μ k ∣ h k ( s ) 2 .
The traced Gauss equation and (47) imply
∣ scal ( h k ( s ) ) ∣ ≤ n ( n − 1 ) K + ∣ ( tr L k ) 2 − ∣ L k ∣ 2 ∣ ≤ 2 n 2 K . |\operatorname{scal}(h_k(s))| \le n(n-1)K + |(\operatorname{tr}L_k)^2 - |L_k|^2| \le2n^2K. ∣ scal ( h k ( s )) ∣ ≤ n ( n − 1 ) K + ∣ ( tr L k ) 2 − ∣ L k ∣ 2 ∣ ≤ 2 n 2 K .
so the structure constants are bounded
∣ μ k ∣ 2 ≤ 8 n 2 K . (48) |\mu_k|^2 \le8n^2K. \tag*{(48)} ∣ μ k ∣ 2 ≤ 8 n 2 K . (48)
By (45) , (47) , and (48) , at p ~ k \tilde p_k p ~ k the 1-jet norms of the Killing fields { V i , k # } \{V_{i,k}^{\#}\} { V i , k # } given in (44) are bounded by
1 ≤ ∣ V i , k # ∣ p ~ k ∗ ≤ B ( n , K ) . (49) 1 \le\left|V_{i,k}^{\#}\right|_{\tilde p_k}^{*} \le B(n,K). \tag*{(49)} 1 ≤ V i , k # p ~ k ∗ ≤ B ( n , K ) . (49)
Since the 1-jet norms are bounded, the sequences ( V i , k # ) k ∈ N (V_{i,k}^{\#})_{k\in\mathbb{N}} ( V i , k # ) k ∈ N subconverge in the C 1 C^1 C 1 -topology to g ~ ∞ \tilde g_\infty g ~ ∞ -Killing fields V i , ∞ # V_{i,\infty}^{\#} V i , ∞ # that are orthonormal at p ~ ∞ \tilde p_\infty p ~ ∞ [15 p. 330] . By the same argument as in [BL18, Prop. 9.8], the brackets converge
[ V i , k # , V j , k # ] → [ V i , ∞ # , V j , ∞ # ] . [V_{i,k}^{\#},V_{j,k}^{\#}] \to[V_{i,\infty}^{\#},V_{j,\infty}^{\#}]. [ V i , k # , V j , k # ] → [ V i , ∞ # , V j , ∞ # ] .
This means that
n ∞ : = span { V 1 , ∞ # , … , V n , ∞ # } \mathfrak n_\infty:= \operatorname{span}\{V_{1,\infty}^{\#},\ldots,V_{n,\infty}^{\#}\} n ∞ := span { V 1 , ∞ # , … , V n , ∞ # }
is an n n n -dimensional Lie algebra of Killing fields on ( M ~ ∞ , g ~ ∞ ) (\widetilde M_\infty,\tilde g_\infty) ( M ∞ , g ~ ∞ ) . This Lie algebra is nilpotent since brackets of length n + 1 n+1 n + 1 vanish for every n k \mathfrak n_k n k , and this property passes to the limit. Since ( M ~ ∞ , g ~ ∞ ) (\widetilde M_\infty,\tilde g_\infty) ( M ∞ , g ~ ∞ ) is complete, its Killing fields are complete. Hence n ∞ \mathfrak n_\infty n ∞ integrates to a global isometric action on ( M ~ ∞ , g ~ ∞ ) (\widetilde M_\infty,\tilde g_\infty) ( M ∞ , g ~ ∞ ) .
Next, we pass the projections M ~ k \widetilde M_k M k onto R \mathbb R R to the limit. The shape operator satisfies
L k = 1 2 h k − 1 d h k d s . L_k = \frac{1}{2}h_k^{-1}\frac{dh_k}{ds}. L k = 2 1 h k − 1 d s d h k .
From the bound (47) and by integrating, we find that the fibre metrics are uniformly equivalent on compact intervals
e − 2 ∣ s ∣ K h k ( 0 ) ≤ h k ( s ) ≤ e 2 ∣ s ∣ K h k ( 0 ) . (50) e^{-2|s|\sqrt{K}}h_k(0) \le h_k(s) \le e^{2|s|\sqrt{K}}h_k(0). \tag*{(50)} e − 2∣ s ∣ K h k ( 0 ) ≤ h k ( s ) ≤ e 2∣ s ∣ K h k ( 0 ) . (50)
The Riccati equation (46) gives
∣ ∇ ∂ s k L k ∣ ≤ ( n + 1 ) K . |\nabla_{\partial_s^k}L_k| \le(n+1)K. ∣ ∇ ∂ s k L k ∣ ≤ ( n + 1 ) K .
Differentiating (46) and using Shi’s estimates [29 ] gives, inductively,
∣ ( ∇ ∂ s k r ) L k ∣ ≤ C r . (51) |(\nabla_{\partial_s^k}^{r})L_k| \le C_r. \tag*{(51)} ∣ ( ∇ ∂ s k r ) L k ∣ ≤ C r . (51)
Vertical derivatives are controlled by invariance, Koszul’s formula, (47) , and the bracket bound (48) . Hence, all derivatives are uniformly bounded on compact s s s -intervals. Let s k : M ~ k → R s_k:\widetilde M_k\to\mathbb R s k : M k → R be the projections such that s k ( s , n ) = s s_k(s,n)=s s k ( s , n ) = s via the identifications (42) . Its gradient and Hessian are
∇ s k = ∂ s , Hess ( s k ) = g ~ k ( L k ⋅ , ⋅ ) , \nabla s_k=\partial_s,\qquad\operatorname{Hess}(s_k)=\tilde g_k(L_k\cdot,\cdot), ∇ s k = ∂ s , Hess ( s k ) = g ~ k ( L k ⋅ , ⋅ ) ,
where L k L_k L k vanishes in the horizontal direction. Let ( Φ k ) k ∈ N (\Phi_k)_{k\in\mathbb N} ( Φ k ) k ∈ N be the exhausting pointed embeddings of the Cheeger–Gromov convergence (43) . The estimates (47) , (48) , (51) give bounds on compact sets for s k ∘ Φ k s_k\circ\Phi_k s k ∘ Φ k and all their derivatives. Passing to a subsequence, s k ∘ Φ k s_k\circ\Phi_k s k ∘ Φ k converges to s ∞ s_\infty s ∞ smoothly on compact subsets, with
s ∞ ( p ~ ∞ ) = 0 , ∣ ∇ s ∞ ∣ g ~ ∞ = 1. s_\infty(\tilde p_\infty)=0,\qquad|\nabla s_\infty|_{\tilde g_\infty}=1. s ∞ ( p ~ ∞ ) = 0 , ∣∇ s ∞ ∣ g ~ ∞ = 1.
The gradient field ∇ s ∞ \nabla s_\infty ∇ s ∞ is complete. Let θ ∞ s \theta_\infty^s θ ∞ s be the flow of ∇ s ∞ \nabla s_\infty ∇ s ∞ and set Σ ∞ = s ∞ − 1 ( 0 ) \Sigma_\infty=s_\infty^{-1}(0) Σ ∞ = s ∞ − 1 ( 0 ) . The flow from
R × Σ ∞ → M ~ ∞ , ( s , x ) ↦ θ ∞ s ( x ) (52) \mathbb R\times\Sigma_\infty\to\widetilde M_\infty,\qquad(s,x)\mapsto\theta_\infty^s(x) \tag*{(52)} R × Σ ∞ → M ∞ , ( s , x ) ↦ θ ∞ s ( x ) (52)
is a diffeomorphism. Indeed, Σ ∞ \Sigma_\infty Σ ∞ is connected and the inherited metric is complete. Set Σ k = s k − 1 ( 0 ) \Sigma_k=s_k^{-1}(0) Σ k = s k − 1 ( 0 ) and let θ k u \theta_k^u θ k u be the local flow of ∇ Φ k ∗ g ~ k ( s k ∘ Φ k ) \nabla^{\Phi_k^*\tilde g_k}(s_k\circ\Phi_k) ∇ Φ k ∗ g ~ k ( s k ∘ Φ k ) . This gradient field is unit:
d d u ( s k ∘ Φ k ) ( θ k u ( x ) ) = ∣ ∇ ( s k ∘ Φ k ) ∣ 2 = 1. \frac{d}{du}(s_k\circ\Phi_k)(\theta_k^u(x))=|\nabla(s_k\circ\Phi_k)|^2=1. d u d ( s k ∘ Φ k ) ( θ k u ( x )) = ∣∇ ( s k ∘ Φ k ) ∣ 2 = 1.
By integrating, we get
( s k ∘ Φ k ) ( θ k u ( x ) ) = ( s k ∘ Φ k ) ( x ) + u . (s_k\circ\Phi_k)(\theta_k^u(x))=(s_k\circ\Phi_k)(x)+u. ( s k ∘ Φ k ) ( θ k u ( x )) = ( s k ∘ Φ k ) ( x ) + u .
Indeed, the map
Φ ^ k ( x ) : = Φ k ∘ θ k − ( s k ∘ Φ k ) ( x ) ( x ) , \hat\Phi_k(x):=\Phi_k\circ\theta_k^{-(s_k\circ\Phi_k)(x)}(x), Φ ^ k ( x ) := Φ k ∘ θ k − ( s k ∘ Φ k ) ( x ) ( x ) ,
defined on exhausting subsets of Σ ∞ \Sigma_\infty Σ ∞ , maps Σ ∞ \Sigma_\infty Σ ∞ to Σ k \Sigma_k Σ k . The map x ↦ θ k − ( s k ∘ Φ k ) ( x ) ( x ) x \mapsto\theta_k^{-(s_k\circ\Phi_k)(x)}(x) x ↦ θ k − ( s k ∘ Φ k ) ( x ) ( x ) converges smoothly on compact sets of Σ ∞ \Sigma_\infty Σ ∞ to the identity, so
Φ ^ k ∗ ( g ~ k ∣ Σ k ) → g ~ ∞ ∣ Σ ∞ , Φ ^ k ( p ~ ∞ ) = p ~ k . \widehat{\Phi}_k^*(\widetilde{g}_k|_{\Sigma_k}) \to\widetilde{g}_\infty|_{\Sigma_\infty}, \qquad\widehat{\Phi}_k(\widetilde{p}_\infty)=\widetilde{p}_k. Φ k ∗ ( g k ∣ Σ k ) → g ∞ ∣ Σ ∞ , Φ k ( p ∞ ) = p k .
These maps give smooth pointed Cheeger–Gromov convergence of the fibres. Each ( Σ k , g ~ k ∣ Σ k ) (\Sigma_k,\widetilde{g}_k|_{\Sigma_k}) ( Σ k , g k ∣ Σ k ) is homogeneous and diffeomorphic to R n \mathbb{R}^n R n , so by [BL18, Thm. 8.3] Σ ∞ \Sigma_\infty Σ ∞ is diffeomorphic to R n \mathbb{R}^n R n too.
Recall that the Lie algebra of Killing fields n ∞ \mathfrak{n}_\infty n ∞ integrates to a global isometric right action by a connected and simply connected Lie group N ~ ∞ \widetilde{N}_\infty N ∞ . As V i , k # ( s k ) = 0 V_{i,k}^{\#}(s_k)=0 V i , k # ( s k ) = 0 , the limit satisfies V i , ∞ # ( s ∞ ) = 0 V_{i,\infty}^{\#}(s_\infty)=0 V i , ∞ # ( s ∞ ) = 0 , so the N ~ ∞ \widetilde{N}_\infty N ∞ -action preserves the level sets of s ∞ s_\infty s ∞ . Let N ∞ N_\infty N ∞ be the quotient of N ~ ∞ \widetilde{N}_\infty N ∞ by the effective kernel. As
T p ~ ∞ Σ ∞ = span { V 1 , ∞ # ( p ~ ∞ ) , … , V n , ∞ # ( p ~ ∞ ) } , T_{\widetilde{p}_\infty}\Sigma_\infty=\operatorname{span}\{V_{1,\infty}^{\#}(\widetilde{p}_\infty),\ldots,V_{n,\infty}^{\#}(\widetilde{p}_\infty)\}, T p ∞ Σ ∞ = span { V 1 , ∞ # ( p ∞ ) , … , V n , ∞ # ( p ∞ )} ,
the effective kernel is discrete. Hence, Lie ( N ∞ ) = n ∞ \operatorname{Lie}(N_\infty)=\mathfrak{n}_\infty Lie ( N ∞ ) = n ∞ , so N ∞ N_\infty N ∞ is nilpotent and dim N ∞ = n \dim N_\infty=n dim N ∞ = n .
The orbit map
N ∞ → Σ ∞ , n ↦ p ~ ∞ ⋅ n (53) N_\infty\to\Sigma_\infty,\qquad n\mapsto\widetilde{p}_\infty\cdot n \tag*{(53)} N ∞ → Σ ∞ , n ↦ p ∞ ⋅ n (53)
is a local diffeomorphism because its differential at the identity is the fundamental vector field point map n ∞ → T p ~ ∞ Σ ∞ \mathfrak{n}_\infty\to T_{\widetilde{p}_\infty}\Sigma_\infty n ∞ → T p ∞ Σ ∞ such that X ↦ X # X\mapsto X^{\#} X ↦ X # . Pull back the metric g ~ ∞ ∣ Σ ∞ \widetilde{g}_\infty|_{\Sigma_\infty} g ∞ ∣ Σ ∞ by this map to get a right-invariant metric h ∞ ( 0 ) h_\infty(0) h ∞ ( 0 ) on N ∞ N_\infty N ∞ . Indeed, the orbit map is a local isometry from a complete connected Riemannian manifold, and thus is a covering. This means that the N ∞ N_\infty N ∞ -action on Σ ∞ \Sigma_\infty Σ ∞ is transitive. The action is also free. This is because stabilisers act orthogonally on the Lie algebra. Since N ∞ N_\infty N ∞ is connected and nilpotent, the Lie exponential map is surjective. If n ∈ N ∞ n\in N_\infty n ∈ N ∞ is a stabiliser, then there exists an X ∈ n ∞ X\in\mathfrak{n}_\infty X ∈ n ∞ such that n = exp X n=\exp X n = exp X . Hence, Ad ( n ) = exp ( ad X ) \operatorname{Ad}(n)=\exp(\operatorname{ad}X) Ad ( n ) = exp ( ad X ) is orthogonal and unipotent, so it is the identity. This means that n n n is central. By transitivity, n n n fixes all of Σ ∞ \Sigma_\infty Σ ∞ . As [ V i , ∞ # , ∇ s ∞ ] = 0 [V_{i,\infty}^{\#},\nabla s_\infty]=0 [ V i , ∞ # , ∇ s ∞ ] = 0 , the N ∞ N_\infty N ∞ -actions commute with θ ∞ \theta^\infty θ ∞ , so n n n fixes all of M ~ ∞ \widetilde{M}_\infty M ∞ . But the N ∞ N_\infty N ∞ action is effective, so n = e n=e n = e .
As the action on Σ ∞ \Sigma_\infty Σ ∞ is free and transitive, and Σ ∞ \Sigma_\infty Σ ∞ is diffeomorphic to R n \mathbb{R}^n R n , we get N ∞ N_\infty N ∞ is simply-connected. From (52) and (53) , we get the diffeomorphism
R × N ∞ → M ~ ∞ , ( s , n ) ↦ θ ∞ s ( p ~ ∞ ⋅ n ) , \mathbb{R}\times N_\infty\to\widetilde{M}_\infty,\qquad(s,n)\mapsto\theta_\infty^s(\widetilde{p}_\infty\cdot n), R × N ∞ → M ∞ , ( s , n ) ↦ θ ∞ s ( p ∞ ⋅ n ) ,
under which
( M ~ ∞ , g ~ ∞ , p ~ ∞ ) ≅ ( R × N ∞ , d s 2 + h ∞ ( s ) , ( 0 , e ) ) . (\widetilde{M}_\infty,\widetilde{g}_\infty,\widetilde{p}_\infty)\cong(\mathbb{R}\times N_\infty,ds^2+h_\infty(s),(0,e)). ( M ∞ , g ∞ , p ∞ ) ≅ ( R × N ∞ , d s 2 + h ∞ ( s ) , ( 0 , e )) .
Here h ∞ ( s ) h_\infty(s) h ∞ ( s ) is right-invariant, and the action is free and transitive on the fibres. The limit is a principal N ∞ N_\infty N ∞ -bundle over R \mathbb{R} R equipped with an invariant metric. This proves the first assertion.
The vertical Killing fields of (44) are independent of time and Killing along the flow g ~ k ( t ) \widetilde{g}_k(t) g k ( t ) . The smooth convergence on compact sets gives the same action along the limit flow g ~ ∞ ( t ) \widetilde{g}_\infty(t) g ∞ ( t ) .
Part (b). Next, assume that the base diameter is bounded below by some c > 0 c>0 c > 0 . At time t = 0 t=0 t = 0 , set
ℓ k = 2 diam ( S 1 , g ˘ k ( 0 ) ) , 2 c ≤ ℓ k ≤ 2 d . \ell_k=2\operatorname{diam}(S^1,\breve{g}_k(0)),\qquad2c\leq\ell_k\leq2d. ℓ k = 2 diam ( S 1 , g ˘ k ( 0 )) , 2 c ≤ ℓ k ≤ 2 d .
After passing to a subsequence, ℓ k \ell_k ℓ k converges to ℓ ∞ ∈ [ 2 c , 2 d ] \ell_\infty\in[2c,2d] ℓ ∞ ∈ [ 2 c , 2 d ] . Take τ k \tau_k τ k to be the deck generators of M ~ k → M k \widetilde{M}_k\to M_k M k → M k such that in (42) coordinates
τ k ( s , n ) = ( s + ℓ k , φ k − 1 ( n ) ) , h k ( s + ℓ k ) = φ k ∗ h k ( s ) \tau_k(s,n)=(s+\ell_k,\varphi_k^{-1}(n)),\qquad h_k(s+\ell_k)=\varphi_k^*h_k(s) τ k ( s , n ) = ( s + ℓ k , φ k − 1 ( n )) , h k ( s + ℓ k ) = φ k ∗ h k ( s )
where φ k ∈ Aut ( N k ) × ( N k ) L \varphi_k\in\operatorname{Aut}(N_k)\times(N_k)_L φ k ∈ Aut ( N k ) × ( N k ) L is the holonomy. Set h k = φ k ( e ) h_k=\varphi_k(e) h k = φ k ( e ) . Then the maps
σ k : = τ k ∘ R h k \sigma_k:=\tau_k\circ R_{h_k} σ k := τ k ∘ R h k
are isometries of g ~ k ( t ) \widetilde{g}_k(t) g k ( t ) and at t = 0 t=0 t = 0
σ k ( 0 , e ) = ( ℓ k , e ) , dist g ~ k ( ( 0 , e ) , σ k ( 0 , e ) ) = ℓ k ≤ 2 d . \sigma_k(0,e)=(\ell_k,e),\qquad\operatorname{dist}_{\widetilde{g}_k}((0,e),\sigma_k(0,e))=\ell_k\leq2d. σ k ( 0 , e ) = ( ℓ k , e ) , dist g k (( 0 , e ) , σ k ( 0 , e )) = ℓ k ≤ 2 d .
Compactness of isometries with bounded basepoint displacement (cf. [31 Lem. B.3] ) therefore gives, on a further subsequence
Φ k − 1 ∘ σ k ∘ Φ k → σ ∞ \Phi_k^{-1}\circ\sigma_k\circ\Phi_k\to\sigma_\infty Φ k − 1 ∘ σ k ∘ Φ k → σ ∞
smoothly on compact subsets. As s ∞ ∘ σ ∞ = s ∞ + ℓ ∞ s_\infty\circ\sigma_\infty=s_\infty+\ell_\infty s ∞ ∘ σ ∞ = s ∞ + ℓ ∞ and ℓ ∞ > 0 \ell_\infty>0 ℓ ∞ > 0 , the action of ⟨ σ ∞ ⟩ \langle\sigma_\infty\rangle ⟨ σ ∞ ⟩ is free and properly discontinuous. Also, σ ∞ \sigma_\infty σ ∞ commutes with the gradient flow of s ∞ s_\infty s ∞ , so it is of the form
σ ∞ ( s , n ) = ( s + ℓ ∞ , φ ∞ ( n ) ) (54) \sigma_\infty(s,n)=(s+\ell_\infty,\varphi_\infty(n)) \tag*{(54)} σ ∞ ( s , n ) = ( s + ℓ ∞ , φ ∞ ( n )) (54)
where φ ∞ : ( N ∞ , h ∞ ( 0 ) ) → ( N ∞ , h ∞ ( ℓ ∞ ) ) \varphi_{\infty} : (N_{\infty}, h_{\infty}(0)) \to(N_{\infty}, h_{\infty}(\ell_{\infty})) φ ∞ : ( N ∞ , h ∞ ( 0 )) → ( N ∞ , h ∞ ( ℓ ∞ )) is an isometry. Isometries between right-invariant metrics on connected nilpotent Lie groups are affine [17 Thm. 1.2] . Then, φ ∞ ∈ Aut ( N ∞ ) ⋉ ( N ∞ ) L \varphi_{\infty} \in\operatorname{Aut}(N_{\infty}) \ltimes(N_{\infty})_{L} φ ∞ ∈ Aut ( N ∞ ) ⋉ ( N ∞ ) L and
( R × N ∞ ) / ⟨ σ ∞ ⟩ → R / ℓ ∞ Z (\mathbb{R} \times N_{\infty})/\langle\sigma_{\infty}\rangle\to\mathbb{R}/\ell_{\infty}\mathbb{Z} ( R × N ∞ ) / ⟨ σ ∞ ⟩ → R / ℓ ∞ Z
is a twisted principal N ∞ N_{\infty} N ∞ -bundle over S 1 S^{1} S 1 . The metrics g ~ ∞ ( t ) \widetilde{g}_{\infty}(t) g ∞ ( t ) descend to invariant metrics solving the normalised Ricci flow.
Part (c). It remains to show convergence of ( M k , g k ( t ) , p k ) (M_k,g_k(t),p_k) ( M k , g k ( t ) , p k ) under the assumptions
0 < c ≤ diam ( S 1 , g ˇ k ( t ) ) , dist g ~ k ( 0 ) ( p ~ k , τ k ( p ~ k ) ) ≤ l , 0 < c \le\operatorname{diam}(S^{1},\check{g}_k(t)), \qquad\operatorname{dist}_{\widetilde{g}_k(0)}(\widetilde{p}_k,\tau_k(\widetilde{p}_k)) \le l, 0 < c ≤ diam ( S 1 , g ˇ k ( t )) , dist g k ( 0 ) ( p k , τ k ( p k )) ≤ l ,
for some c , l > 0 c,l > 0 c , l > 0 . First, we prove an injectivity radius estimate. Set g k ( 0 ) = g k g_k(0)=g_k g k ( 0 ) = g k and
r 1 = min { c , π 4 K } . r_1=\min\left\{c,\frac{\pi}{4\sqrt{K}}\right\}. r 1 = min { c , 4 K π } .
Take horizontal unit-speed geodesics γ k : ( − r 1 2 , r 1 2 ) → M k \gamma_k:\left(-\frac{r_1}{2},\frac{r_1}{2}\right)\to M_k γ k : ( − 2 r 1 , 2 r 1 ) → M k with γ k ( 0 ) = p k \gamma_k(0)=p_k γ k ( 0 ) = p k . Define the regions
Ω k = { γ k ( s ) ⋅ exp X : ∣ s ∣ < r 1 / 2 , ∣ X ∣ h k ( s ) e < r 1 / 2 } . \Omega_k=\{\gamma_k(s)\cdot\exp X: |s|<r_1/2,\ |X|_{h_k(s)_e}<r_1/2\}. Ω k = { γ k ( s ) ⋅ exp X : ∣ s ∣ < r 1 /2 , ∣ X ∣ h k ( s ) e < r 1 /2 } .
As before, Ω k ⊆ B g k ( p k , r 1 ) \Omega_k\subseteq B_{g_k}(p_k,r_1) Ω k ⊆ B g k ( p k , r 1 ) , and so
Vol g k B g k ( p k , r 1 ) ≥ Vol g k Ω k = ∫ − r 1 / 2 r 1 / 2 Vol h k ( s ) exp ( B h k ( s ) ( 0 , r 1 / 2 ) ) d s = ω n ( r 1 2 ) n r 1 . \operatorname{Vol}_{g_k} B_{g_k}(p_k,r_1)\ge\operatorname{Vol}_{g_k}\Omega_k=\int_{-r_1/2}^{r_1/2}\operatorname{Vol}_{h_k(s)}\exp\left(B_{h_k(s)}(0,r_1/2)\right)\,ds=\omega_n\left(\frac{r_1}{2}\right)^n r_1. Vol g k B g k ( p k , r 1 ) ≥ Vol g k Ω k = ∫ − r 1 /2 r 1 /2 Vol h k ( s ) exp ( B h k ( s ) ( 0 , r 1 /2 ) ) d s = ω n ( 2 r 1 ) n r 1 .
The Cheeger–Gromov–Taylor injectivity radius estimate [8 Thm. 4.7] and the curvature bound yield
inj ( p k , g k ) ≥ r 1 2 1 1 + ( V 2 r 1 − K / ω n ( r 1 2 ) n r 1 ) = : i ( n , c , K ) > 0. \operatorname{inj}(p_k,g_k)\ge\frac{r_1}{2}\frac{1}{1+\left(V_{2r_1}^{-K}/\omega_n\left(\frac{r_1}{2}\right)^n r_1\right)}=:i(n,c,K)>0. inj ( p k , g k ) ≥ 2 r 1 1 + ( V 2 r 1 − K / ω n ( 2 r 1 ) n r 1 ) 1 =: i ( n , c , K ) > 0.
By Hamilton’s compactness Theorem [14 ] , after passing to a subsequence
( M k , g k ( t ) , p k ) → ( M ∞ , g ∞ ( t ) , p ∞ ) (55) (M_k,g_k(t),p_k)\to(M_{\infty},g_{\infty}(t),p_{\infty}) \tag*{(55)} ( M k , g k ( t ) , p k ) → ( M ∞ , g ∞ ( t ) , p ∞ ) (55)
in the pointed smooth Cheeger–Gromov topology. Let ( Ψ k ) k ∈ N (\Psi_k)_{k\in\mathbb{N}} ( Ψ k ) k ∈ N be the exhausting pointed embeddings of this Cheeger–Gromov convergence.
As we have assumed that the displacement of p ~ k \widetilde{p}_k p k by the deck generators τ k \tau_k τ k is bounded, compactness of isometries (cf. [31 Lem. B.3] ) gives
Ψ k − 1 ∘ τ k ∘ Ψ k → τ ∞ \Psi_k^{-1}\circ\tau_k\circ\Psi_k\to\tau_{\infty} Ψ k − 1 ∘ τ k ∘ Ψ k → τ ∞
smoothly on compact subsets. By the same argument as for σ ∞ \sigma_{\infty} σ ∞ , the group ⟨ τ ∞ ⟩ \langle\tau_{\infty}\rangle ⟨ τ ∞ ⟩ acts freely and properly discontinuously, and
τ ∞ ( s , n ) = ( s + ℓ ∞ , φ ∞ ( n ) ) \tau_{\infty}(s,n)=(s+\ell_{\infty},\varphi_{\infty}(n)) τ ∞ ( s , n ) = ( s + ℓ ∞ , φ ∞ ( n ))
for some affine φ ∞ ∈ Aut ( N ∞ ) ⋉ ( N ∞ ) L \varphi_{\infty}\in\operatorname{Aut}(N_{\infty})\ltimes(N_{\infty})_{L} φ ∞ ∈ Aut ( N ∞ ) ⋉ ( N ∞ ) L . To identify the limit in (55) with this quotient, let q k : M ~ k → M k q_k:\widetilde{M}_k\to M_k q k : M k → M k be the covering maps. Then the maps
Ψ k − 1 ∘ q k ∘ Φ k \Psi_k^{-1}\circ q_k\circ\Phi_k Ψ k − 1 ∘ q k ∘ Φ k
are defined on exhausting subsets of M ~ ∞ \widetilde{M}_{\infty} M ∞ and send p ~ ∞ \widetilde{p}_{\infty} p ∞ to p ∞ p_{\infty} p ∞ . Their images of compact subsets remain in bounded metric balls since q k q_k q k does not increase distances. As these maps are local isometries for the pulled-back metrics, their derivatives are controlled by the smoothly converging metrics. Passing to a further subsequence gives a smooth limit
q ∞ : M ~ ∞ → M ∞ , q ∞ ∗ g ∞ ( t ) = g ~ ∞ ( t ) q_{\infty}:\widetilde{M}_{\infty}\to M_{\infty},\qquad q_{\infty}^{*}g_{\infty}(t)=\widetilde{g}_{\infty}(t) q ∞ : M ∞ → M ∞ , q ∞ ∗ g ∞ ( t ) = g ∞ ( t )
for compact time intervals. Indeed, q ∞ q_{\infty} q ∞ is a covering and q ∞ ∘ τ ∞ = q ∞ q_{\infty}\circ\tau_{\infty}=q_{\infty} q ∞ ∘ τ ∞ = q ∞ . Finally, we show that the fibres of q ∞ q_{\infty} q ∞ are the orbits of τ ∞ \tau_{\infty} τ ∞ . Suppose that q ∞ ( x ) = q ∞ ( y ) q_{\infty}(x)=q_{\infty}(y) q ∞ ( x ) = q ∞ ( y ) . Choose sufficiently small neighbourhoods of x x x and y y y such that q ∞ q_{\infty} q ∞ is a diffeomorphism. By smooth convergence and the inverse function theorem, there exist sequences of points x k , y k ∈ M ~ k x_k,y_k\in\widetilde{M}_k x k , y k ∈ M k such that
Φ k − 1 ( x k ) → x , Φ k − 1 ( y k ) → y , q k ( x k ) = q k ( y k ) . \Phi_k^{-1}(x_k)\to x,\qquad\Phi_k^{-1}(y_k)\to y,\qquad q_k(x_k)=q_k(y_k). Φ k − 1 ( x k ) → x , Φ k − 1 ( y k ) → y , q k ( x k ) = q k ( y k ) .
Indeed, y k = τ k m k ( x k ) y_k=\tau_k^{m_k}(x_k) y k = τ k m k ( x k ) for some m k ∈ Z m_k\in\mathbb{Z} m k ∈ Z . These points lie in B g ~ k ( p ~ k , R ) B_{\widetilde{g}_k}(\widetilde{p}_k,R) B g k ( p k , R ) for fixed R R R . Since the fibres are simply connected, the bundle projections M k → S 1 M_k\to S^1 M k → S 1 induce an isomorphism of fundamental groups. Every curve joining x k x_k x k to τ m k ( x k ) \tau^{m_k}(x_k) τ m k ( x k ) projects to a loop in S 1 S^1 S 1 with winding number m k m_k m k . Indeed,
∣ m k ∣ ℓ k ≤ dist g ~ k ( x k , y k ) ≤ 2 R , |m_k|\ell_k \le\operatorname{dist}_{\widetilde{g}_k}(x_k,y_k) \le2R, ∣ m k ∣ ℓ k ≤ dist g k ( x k , y k ) ≤ 2 R ,
and since ℓ k ≥ 2 c \ell_k \ge2c ℓ k ≥ 2 c , we have ∣ m k ∣ ≤ R / c |m_k| \le R/c ∣ m k ∣ ≤ R / c . After passing to a subsequence, m k = m m_k=m m k = m and y = τ ∞ m ( x ) y=\tau_\infty^m(x) y = τ ∞ m ( x ) , so
( M ~ ∞ , g ~ ∞ ( t ) ) / ⟨ τ ∞ ⟩ ≅ ( M ∞ , g ∞ ( t ) ) . (\widetilde{M}_\infty,\widetilde{g}_\infty(t))/\langle\tau_\infty\rangle\cong(M_\infty,g_\infty(t)). ( M ∞ , g ∞ ( t )) / ⟨ τ ∞ ⟩ ≅ ( M ∞ , g ∞ ( t )) .
Proposition 5.2. Fix n ∈ N n \in\mathbb{N} n ∈ N and K , c , d > 0 K,c,d>0 K , c , d > 0 . Let I ⊆ R I \subseteq\mathbb{R} I ⊆ R be an open interval containing 0 0 0 . Let π k : M k → S 1 \pi_k:M_k\to S^1 π k : M k → S 1 be a sequence of twisted principal N k N_k N k -bundles where N k N_k N k is nilpotent, simply-connected and dim N k = n \dim N_k=n dim N k = n . For every k ∈ N k\in\mathbb{N} k ∈ N , let g k ( t ) g_k(t) g k ( t ) be a normalised Ricci flow of invariant metrics on M k M_k M k defined for t ∈ I t\in I t ∈ I such that
∣ Rm g k ( t ) ∣ g k ( t ) ≤ K , 0 < c ≤ diam ( S 1 , g ˇ k ( t ) ) ≤ d . |\operatorname{Rm}_{g_k(t)}|_{g_k(t)}\le K,\qquad0<c\le\operatorname{diam}(S^1,\check{g}_k(t))\le d. ∣ Rm g k ( t ) ∣ g k ( t ) ≤ K , 0 < c ≤ diam ( S 1 , g ˇ k ( t )) ≤ d . Take a sequence of points p k ∈ M k p_k\in M_k p k ∈ M k and their lifts to the universal covers p ~ k ∈ M ~ k \widetilde{p}_k\in\widetilde{M}_k p k ∈ M k . At t = 0 t=0 t = 0 , let P k ∈ Aut ( n k ) P_k\in\operatorname{Aut}(\mathfrak{n}_k) P k ∈ Aut ( n k ) be the adjoint holonomy of ∇ k ad \nabla^{\operatorname{ad}}_k ∇ k ad based at π k ( p k ) \pi_k(p_k) π k ( p k ) . Let ( M ~ ∞ , g ~ ∞ ( t ) , p ~ ∞ ) (\widetilde{M}_\infty,\widetilde{g}_\infty(t),\widetilde{p}_\infty) ( M ∞ , g ∞ ( t ) , p ∞ ) be a subsequential limit that descends to π ∞ : M ∞ → S 1 \pi_\infty:M_\infty\to S^1 π ∞ : M ∞ → S 1 as in Theorem 5.1 . Then, there exist linear isometries between the fibres of the adjoint bundle ρ k : n ∞ → n k \rho_k:\mathfrak{n}_\infty\to\mathfrak{n}_k ρ k : n ∞ → n k such that
ρ k − 1 P k ρ k → P ∞ \rho_k^{-1}P_k\rho_k\to P_\infty ρ k − 1 P k ρ k → P ∞ where P ∞ P_\infty P ∞ is the holonomy of ∇ ∞ ad \nabla^{\operatorname{ad}}_\infty ∇ ∞ ad based at π ∞ ( p ∞ ) \pi_\infty(p_\infty) π ∞ ( p ∞ ) at time t = 0 t=0 t = 0 .
Proof. Let { V i , k # } \{V^{\#}_{i,k}\} { V i , k # } be the converging orthonormal bases (44) in the proof of Theorem 5.1 . Define ρ k \rho_k ρ k by ρ k V i , ∞ # = V i , k # \rho_kV^{\#}_{i,\infty}=V^{\#}_{i,k} ρ k V i , ∞ # = V i , k # . In the horizontal trivialisation based at p ~ k \widetilde{p}_k p k , write
τ k ( s , n ) = ( s + ℓ k , φ k − 1 ( n ) ) . \tau_k(s,n)=(s+\ell_k,\varphi_k^{-1}(n)). τ k ( s , n ) = ( s + ℓ k , φ k − 1 ( n )) .
where ℓ k \ell_k ℓ k is the length of ( S 1 , g ˇ k ( 0 ) ) (S^1,\check{g}_k(0)) ( S 1 , g ˇ k ( 0 )) and φ k \varphi_k φ k is the twisted principal bundle holonomy. Define ψ k = R φ k ( e ) − 1 ∘ φ k \psi_k=R_{\varphi_k(e)}^{-1}\circ\varphi_k ψ k = R φ k ( e ) − 1 ∘ φ k and σ k = τ k ∘ R φ k ( e ) \sigma_k=\tau_k\circ R_{\varphi_k(e)} σ k = τ k ∘ R φ k ( e ) . The adjoint holonomy is P k = d ( ψ k ) e P_k=d(\psi_k)_e P k = d ( ψ k ) e and
σ k ( s , n ) = ( s + ℓ k , ψ k − 1 ( n ) ) . \sigma_k(s,n)=(s+\ell_k,\psi_k^{-1}(n)). σ k ( s , n ) = ( s + ℓ k , ψ k − 1 ( n )) .
As in the proof of Theorem 5.1 , after passing to a subsequence, ℓ k → ℓ ∞ > 0 \ell_k\to\ell_\infty>0 ℓ k → ℓ ∞ > 0 and, in the Cheeger–Gromov maps, σ k → σ ∞ \sigma_k\to\sigma_\infty σ k → σ ∞ where
σ ∞ ( s , n ) = ( s + ℓ ∞ , ψ ∞ − 1 ( n ) ) \sigma_\infty(s,n)=(s+\ell_\infty,\psi_\infty^{-1}(n)) σ ∞ ( s , n ) = ( s + ℓ ∞ , ψ ∞ − 1 ( n ))
for some ψ ∞ ∈ Aut ( N ∞ ) \psi_\infty\in\operatorname{Aut}(N_\infty) ψ ∞ ∈ Aut ( N ∞ ) . In the Killing basis
( σ k ) ∗ V i , k # = ∑ j = 1 n ( P k − 1 ) i j V j , k # . (\sigma_k)_*V^{\#}_{i,k}=\sum_{j=1}^{n}(P_k^{-1})^j_iV^{\#}_{j,k}. ( σ k ) ∗ V i , k # = j = 1 ∑ n ( P k − 1 ) i j V j , k # .
By the convergence of σ k \sigma_k σ k and the vertical Killing fields, we have
ρ k − 1 P k − 1 ρ k → d ( ψ ∞ − 1 ) e . \rho_k^{-1}P_k^{-1}\rho_k\to d(\psi_\infty^{-1})_e. ρ k − 1 P k − 1 ρ k → d ( ψ ∞ − 1 ) e .
The limit flow descends to M ∞ = ( R × N ∞ ) / ⟨ σ ∞ ⟩ M_\infty=(\mathbb{R}\times N_\infty)/\langle\sigma_\infty\rangle M ∞ = ( R × N ∞ ) / ⟨ σ ∞ ⟩ that has bundle holonomy ψ ∞ \psi_\infty ψ ∞ and adjoint holonomy P ∞ = d ( ψ ∞ ) e P_\infty=d(\psi_\infty)_e P ∞ = d ( ψ ∞ ) e .
5.2. Subconvergence to solvsolitons. Recall that a solvsoliton is a left-invariant Ricci soliton metric on a solvable Lie group S S S such that when evaluated at the identity, the inner product on the Lie algebra s \mathfrak{s} s satisfies
Ric g = c Id + D , (56) \operatorname{Ric}_g=c\operatorname{Id}+D, \tag*{(56)} Ric g = c Id + D , (56)
for some c ∈ R c\in\mathbb{R} c ∈ R and some derivation of the Lie algebra D D D . Similarly, a nilsoliton is a left-invariant Ricci soliton metric on a nilpotent Lie group N N N that satisfies (55). We call right-invariant metrics solvsolitons or nilsolitons too if they satisfy (55), since inversion identifies these metrics with left-invariant metrics with the same inner product at the identity.
Let N N N be a simply-connected nilpotent Lie group. On twisted principal N N N -bundles M → S 1 M\to S^1 M → S 1 , consider the normalised Ricci flow of invariant metrics in the following gauge
∂ g ∂ t = − 2 ric g + L N g , g ( 0 ) = g 0 . (57) \frac{\partial g}{\partial t}=-2\operatorname{ric}_g+\mathcal{L}_{N}g,\qquad g(0)=g_0. \tag*{(57)} ∂ t ∂ g = − 2 ric g + L N g , g ( 0 ) = g 0 . (57)
where N = − ( tr L ) X ‾ N=-(\operatorname{tr}L)\overline{X} N = − ( tr L ) X is the mean curvature vector of the fibres. Given a sequence t k → ∞ t_k\to\infty t k → ∞ , fix a point p ∈ M p\in M p ∈ M and take the sequence of time-translated flows g k ( t ) : = g ( t k + t ) g_k(t):=g(t_k+t) g k ( t ) := g ( t k + t ) . By Theorem 5.1 , the flows lifted to the universal cover subconverge to an invariant solution to (57) on a principal N ∞ N_\infty N ∞ -bundle M ~ ∞ → R \widetilde{M}_\infty\to\mathbb{R} M ∞ → R for a simply-connected nilpotent Lie group N ∞ N_\infty N ∞ . Under the additional assumption that the diameter of ( S 1 , g ˇ k ( t ) ) (S^1,\check{g}_k(t)) ( S 1 , g ˇ k ( t )) is bounded below, we extract properties of the limit ( M ~ ∞ , g ~ ∞ ( t ) ) (\widetilde{M}_\infty,\widetilde{g}_\infty(t)) ( M ∞ , g ∞ ( t )) using the Gindi–Streets functional [11 Def. 4.1]
I ( g ) = ∫ S 1 ( ∣ L ∣ 2 + 1 2 ) d s . \mathcal{I}(g)=\int_{S^1}\left(\lvert L\rvert^2+\frac{1}{2}\right)\,ds. I ( g ) = ∫ S 1 ( ∣ L ∣ 2 + 2 1 ) d s .
This functional is invariant under twisted principal bundle isomorphisms.
Lemma 5.3. Let M → S 1 M\to S^1 M → S 1 be a twisted principal N-bundle where N is a nilpotent Lie group. If g ( t ) g(t) g ( t ) is an invariant normalised Ricci flow on M M M , then
d d t I ( g ( t ) ) = − ∫ S 1 ( ∣ δ [ ⋅ , ⋅ ] ( L ) ∣ 2 + ( ∣ L ∣ 2 − 1 2 ) 2 + 4 ∣ ∇ ^ ∗ L ∣ 2 + 2 ∣ ∇ L ∣ 2 ) d s , \frac{d}{dt}\mathcal{I}(g(t))=-\int_{S^1}\left(\lvert\delta_{[\,\cdot,\cdot\,]}(L)\rvert^2+\left(\lvert L\rvert^2-\frac{1}{2}\right)^2+4\lvert\hat{\nabla}^{*}L\rvert^2+2\lvert\nabla L\rvert^2\right)\,ds, d t d I ( g ( t )) = − ∫ S 1 ( ∣ δ [ ⋅ , ⋅ ] ( L ) ∣ 2 + ( ∣ L ∣ 2 − 2 1 ) 2 + 4 ∣ ∇ ^ ∗ L ∣ 2 + 2 ∣ ∇ L ∣ 2 ) d s , where [ ⋅ , ⋅ ] [\,\cdot,\cdot\,] [ ⋅ , ⋅ ] is the Lie bracket on the adjoint bundle.
Proof. By Theorem 2.6, under the normalised flow in the mean curvature gauge (57) the volume form of the base evolves by
∂ t d s = ( ∣ L ∣ 2 − 1 2 ) d s . \partial_t ds=\left(\lvert L\rvert^2-\frac{1}{2}\right)ds. ∂ t d s = ( ∣ L ∣ 2 − 2 1 ) d s .
By equation (41) in the proof of Theorem 4.4 ,
( ∂ ∂ t − Δ S 1 ) ∣ L ∣ 2 = − 2 h ( ∇ Ric g ^ , L ) − 2 ∣ L ∣ 4 + ∣ L ∣ 2 − 4 h ( ∇ ^ ∗ L − L H , ∇ ^ ∗ L ) − 2 ∣ ∇ L ∣ 2 = − ∣ δ [ ⋅ , ⋅ ] ( L ) ∣ 2 − 2 ∣ L ∣ 2 ( ∣ L ∣ 2 − 1 2 ) − 4 ∣ ∇ ^ ∗ L ∣ 2 − 2 ∣ ∇ L ∣ 2 , \begin{aligned}
\left(\frac{\partial}{\partial t}-\Delta_{S^1}\right)\lvert L\rvert^2=-2h(\nabla\operatorname{Ric}_{\hat{g}},L)-2\lvert L\rvert^4+\lvert L\rvert^2-4h(\hat{\nabla}^{*}L-LH,\hat{\nabla}^{*}L)-2\lvert\nabla L\rvert^2 \\
&=-\lvert\delta_{[\,\cdot,\cdot\,]}(L)\rvert^2-2\lvert L\rvert^2\left(\lvert L\rvert^2-\frac{1}{2}\right)-4\lvert\hat{\nabla}^{*}L\rvert^2-2\lvert\nabla L\rvert^2,
\end{aligned} ( ∂ t ∂ − Δ S 1 ) ∣ L ∣ 2 = − 2 h ( ∇ Ric g ^ , L ) − 2 ∣ L ∣ 4 + ∣ L ∣ 2 − 4 h ( ∇ ^ ∗ L − L H , ∇ ^ ∗ L ) − 2 ∣ ∇ L ∣ 2 = − ∣ δ [ ⋅ , ⋅ ] ( L ) ∣ 2 − 2 ∣ L ∣ 2 ( ∣ L ∣ 2 − 2 1 ) − 4 ∣ ∇ ^ ∗ L ∣ 2 − 2 ∣ ∇ L ∣ 2 ,
where the second line follows from nilpotence and Theorem 3.5. Combining these, we get
d d t I ( g ( t ) ) = − ∫ S 1 ( ∣ δ [ ⋅ , ⋅ ] ( L ) ∣ 2 + ( ∣ L ∣ 2 − 1 2 ) 2 + 4 ∣ ∇ ^ ∗ L ∣ 2 + 2 ∣ ∇ L ∣ 2 ) d s . \frac{d}{dt}\mathcal{I}(g(t))=-\int_{S^1}\left(\lvert\delta_{[\,\cdot,\cdot\,]}(L)\rvert^2+\left(\lvert L\rvert^2-\frac{1}{2}\right)^2+4\lvert\hat{\nabla}^{*}L\rvert^2+2\lvert\nabla L\rvert^2\right)\,ds. d t d I ( g ( t )) = − ∫ S 1 ( ∣ δ [ ⋅ , ⋅ ] ( L ) ∣ 2 + ( ∣ L ∣ 2 − 2 1 ) 2 + 4 ∣ ∇ ^ ∗ L ∣ 2 + 2 ∣ ∇ L ∣ 2 ) d s .
Lemma 5.4. Let M → S 1 M\to S^1 M → S 1 be a twisted principal N-bundle where N is a nilpotent Lie group. Let g ( t ) g(t) g ( t ) be an invariant normalised Ricci flow on M M M . Then, the length of the base ℓ ( t ) : = 2 diam ( S 1 , g ˇ ( t ) ) \ell(t):=2\operatorname{diam}(S^1,\check{g}(t)) ℓ ( t ) := 2 diam ( S 1 , g ˇ ( t )) converges to ℓ ∞ ≥ 0 \ell_\infty\geq0 ℓ ∞ ≥ 0 as t → ∞ t\to\infty t → ∞ where
ℓ ∞ = lim t → ∞ I ( g ( t ) ) . \ell_\infty=\lim_{t\to\infty}\mathcal{I}(g(t)). ℓ ∞ = t → ∞ lim I ( g ( t )) . Furthermore, if ℓ ∞ > 0 \ell_\infty>0 ℓ ∞ > 0 , then every subsequential limit of time-translated flows obtained in Theorem 5.1 satisfies
δ [ ⋅ , ⋅ ] ( L ∞ ) = 0 , ∣ L ∞ ∣ 2 = 1 2 , ∇ ^ ∗ L ∞ = 0 , ∇ L ∞ = 0. \delta_{[\,\cdot,\cdot\,]}(L_\infty)=0,\qquad\lvert L_\infty\rvert^2=\frac{1}{2},\qquad\hat{\nabla}^{*}L_\infty=0,\qquad\nabla L_\infty=0. δ [ ⋅ , ⋅ ] ( L ∞ ) = 0 , ∣ L ∞ ∣ 2 = 2 1 , ∇ ^ ∗ L ∞ = 0 , ∇ L ∞ = 0. along the limit flow.
Proof. By Theorem 5.3, the Gindi–Streets functional I ( g ( t ) ) \mathcal{I}(g(t)) I ( g ( t )) is monotone and nonnegative, and therefore converges to some I ∞ ≥ 0 \mathcal{I}_\infty\geq0 I ∞ ≥ 0 . Since
d ℓ d t = ∫ S 1 ( ∣ L ∣ 2 − 1 2 ) d s = I ( g ( t ) ) − ℓ ( t ) , \frac{d\ell}{dt}=\int_{S^1}\left(\lvert L\rvert^2-\frac{1}{2}\right)ds=\mathcal{I}(g(t))-\ell(t), d t d ℓ = ∫ S 1 ( ∣ L ∣ 2 − 2 1 ) d s = I ( g ( t )) − ℓ ( t ) ,
we get
ℓ ( t ) = e − t ℓ ( 0 ) + ∫ 0 t e s − t I ( g ( s ) ) d s . \ell(t)=e^{-t}\ell(0)+\int_0^t e^{s-t}\mathcal{I}(g(s))\,ds. ℓ ( t ) = e − t ℓ ( 0 ) + ∫ 0 t e s − t I ( g ( s )) d s .
Hence ℓ ( t ) → I ∞ \ell(t)\to\mathcal{I}_\infty ℓ ( t ) → I ∞ as t → ∞ t\to\infty t → ∞ . Now assume that ℓ ∞ > 0 \ell_\infty>0 ℓ ∞ > 0 . Take a sequence of time-translated normalised Ricci flows of invariant metrics g ( t + t k ) g(t+t_k) g ( t + t k ) that converge as in Theorem 5.1 . For every fixed a < b a<b a < b ,
I ( g ( t k + a ) ) − I ( g ( t k + b ) ) = ∫ t k + a t k + b ∫ S 1 ( ∣ δ [ ⋅ , ⋅ ] ( L ) ∣ 2 + ( ∣ L ∣ 2 − 1 2 ) 2 + 4 ∣ ∇ ^ ∗ L ∣ 2 + 2 ∣ ∇ L ∣ 2 ) d s d t (58) \mathcal{I}(g(t_k+a))-\mathcal{I}(g(t_k+b))=\int_{t_k+a}^{t_k+b}\int_{S^1}\left(\lvert\delta_{[\,\cdot,\cdot\,]}(L)\rvert^2+\left(\lvert L\rvert^2-\frac{1}{2}\right)^2+4\lvert\hat{\nabla}^{*}L\rvert^2+2\lvert\nabla L\rvert^2\right)\,ds\,dt \tag*{(58)} I ( g ( t k + a )) − I ( g ( t k + b )) = ∫ t k + a t k + b ∫ S 1 ( ∣ δ [ ⋅ , ⋅ ] ( L ) ∣ 2 + ( ∣ L ∣ 2 − 2 1 ) 2 + 4 ∣ ∇ ^ ∗ L ∣ 2 + 2 ∣ ∇ L ∣ 2 ) d s d t (58)
converges to zero as k → ∞ k \to\infty k → ∞ . Assume that one of the terms
δ [ ⋅ , ⋅ ] ( L ∞ ) , ∣ L ∞ ∣ 2 − 1 2 , ∇ ^ ∗ L ∞ , ∇ L ∞ \delta_{[\,\cdot,\cdot\,]}(L_\infty), \qquad|L_\infty|^2-\frac{1}{2}, \qquad\widehat{\nabla}^{*}L_\infty, \qquad\nabla L_\infty δ [ ⋅ , ⋅ ] ( L ∞ ) , ∣ L ∞ ∣ 2 − 2 1 , ∇ ∗ L ∞ , ∇ L ∞
is non-zero at some point and time of the limit flow. Then there is a neighbourhood and a time interval for which the corresponding term in the integral (57) is bounded below by a positive constant for large k k k . This is a contradiction. □
After we take a subsequential limit ( M ~ ∞ , g ~ ∞ ( t ) ) (\widetilde{M}_\infty,\widetilde{g}_\infty(t)) ( M ∞ , g ∞ ( t )) , we run the normalised bracket flow. For the normalised Ricci flow of invariant metrics with gauge (56), the corresponding bracket flow solves the following system:
∂ [ ⋅ , ⋅ ] μ ∂ t = δ [ ⋅ , ⋅ ] μ ( Ric ^ μ − ( ∇ L ) + 1 2 Id ) (59) \frac{\partial[\cdot,\cdot]_\mu}{\partial t}
=
\delta_{[\cdot,\cdot]_\mu}\left(\widehat{\operatorname{Ric}}_\mu-(\nabla L)+\frac{1}{2}\operatorname{Id}\right) \tag*{(59)} ∂ t ∂ [ ⋅ , ⋅ ] μ = δ [ ⋅ , ⋅ ] μ ( Ric μ − ( ∇ L ) + 2 1 Id ) (59)
∂ ω μ ( X ‾ ) ∂ t = 2 ω μ ( ∇ ^ ∗ L ) (60) \frac{\partial\omega_\mu(\overline{X})}{\partial t}
=
2\omega_\mu(\widehat{\nabla}^{*}L) \tag*{(60)} ∂ t ∂ ω μ ( X ) = 2 ω μ ( ∇ ∗ L ) (60)
∂ g ~ μ ∂ t = ( 2 ∣ L ∣ 2 − 1 ) g ~ μ (61) \frac{\partial\widetilde{g}_\mu}{\partial t}
=
(2|L|^2-1)\widetilde{g}_\mu \tag*{(61)} ∂ t ∂ g μ = ( 2∣ L ∣ 2 − 1 ) g μ (61)
Just as in the proof of Theorem B, one obtains an equivalence of the normalised Ricci flow and normalised bracket flow by solving the fibrewise ODEs
(i) ∂ ∂ t Φ = − Φ L ( Ric ^ g − ∇ L + 1 2 Id ) ω , Φ ( 0 ) = Id , \frac{\partial}{\partial t}\Phi=-\Phi_{\!L}\left(\widehat{\operatorname{Ric}}_g-\nabla L+\frac{1}{2}\operatorname{Id}\right)\omega,\qquad\Phi(0)=\operatorname{Id}, ∂ t ∂ Φ = − Φ L ( Ric g − ∇ L + 2 1 Id ) ω , Φ ( 0 ) = Id ,
(ii) ∂ ∂ t Φ = − ι ( Ric ^ μ − ∇ L + 1 2 Id ) ω μ Φ , Φ ( 0 ) = Id \frac{\partial}{\partial t}\Phi=-\iota\left(\widehat{\operatorname{Ric}}_\mu-\nabla L+\frac{1}{2}\operatorname{Id}\right)\omega_\mu\Phi,\qquad\Phi(0)=\operatorname{Id} ∂ t ∂ Φ = − ι ( Ric μ − ∇ L + 2 1 Id ) ω μ Φ , Φ ( 0 ) = Id .
Along the homogeneous normalised bracket flow, if μ ( t ) \mu(t) μ ( t ) is nilpotent, then the following quantity is monotone [21 Prop 3.11]
R μ + ∣ Ric μ ∣ 2 . R_\mu+|\operatorname{Ric}_\mu|^2. R μ + ∣ Ric μ ∣ 2 .
Inspired by this, we integrate this over the base to get the following functional
F ( μ , h , g ~ ) = ∫ S 1 R ^ μ + ∣ Ric ^ μ ∣ 2 d s . \mathcal{F}(\mu,h,\widetilde{g})=\int_{S^1}\widehat{R}_\mu+|\widehat{\operatorname{Ric}}_\mu|^2\,ds. F ( μ , h , g ) = ∫ S 1 R μ + ∣ Ric μ ∣ 2 d s .
This functional is invariant under principal bundle isomorphisms.
Lemma 5.5. Let M → S 1 M\to S^1 M → S 1 be a twisted principal N-bundle where N is a nilpotent Lie group. Fix a vector bundle metric h h h on ad M \operatorname{ad}M ad M and then let ( [ ⋅ , ⋅ ] μ , ω μ , g ~ μ ) ( t ) ([\cdot,\cdot]_\mu,\omega_\mu,\widetilde{g}_\mu)(t) ([ ⋅ , ⋅ ] μ , ω μ , g μ ) ( t ) be a normalised bracket flow on M M M solving (58–60). Suppose that the initial condition has the following properties
δ μ ( L ) = 0 , ∣ L ∣ 2 = 1 2 , ∇ ^ ∗ L = 0 , ∇ L = 0. \delta_\mu(L)=0,\qquad|L|^2=\frac{1}{2},\qquad\widehat{\nabla}^{*}L=0,\qquad\nabla L=0. δ μ ( L ) = 0 , ∣ L ∣ 2 = 2 1 , ∇ ∗ L = 0 , ∇ L = 0. Along this flow, the functional
F ( μ ( t ) , h , g ~ ( t ) ) = ∫ S 1 R ^ μ + ∣ Ric ^ μ ∣ 2 d s \mathcal{F}(\mu(t),h,\widetilde{g}(t))=\int_{S^1}\widehat{R}_\mu+|\widehat{\operatorname{Ric}}_\mu|^2\,ds F ( μ ( t ) , h , g ( t )) = ∫ S 1 R μ + ∣ Ric μ ∣ 2 d s is monotonically decreasing, and
lim t → ∞ ∫ S 1 ∣ δ μ ( Ric ^ μ + 1 2 Id ) ∣ 2 d s = 0. \lim_{t\to\infty}\int_{S^1}\left|\delta_\mu\left(\widehat{\operatorname{Ric}}_\mu+\frac{1}{2}\operatorname{Id}\right)\right|^2\,ds=0. t → ∞ lim ∫ S 1 δ μ ( Ric μ + 2 1 Id ) 2 d s = 0.
Proof. Given the initial condition, the normalised bracket flow equations in the mean curvature gauge (58–60) are
∂ [ ⋅ , ⋅ ] μ ∂ t = δ [ ⋅ , ⋅ ] μ ( Ric ^ μ + 1 2 Id ) , ∂ ω μ ∂ t = 0 , ∂ g ~ μ ∂ t = 0. \frac{\partial[\cdot,\cdot]_\mu}{\partial t}
=
\delta_{[\cdot,\cdot]_\mu}\left(\widehat{\operatorname{Ric}}_\mu+\frac{1}{2}\operatorname{Id}\right),
\qquad
\frac{\partial\omega_\mu}{\partial t}=0,
\qquad
\frac{\partial\widetilde{g}_\mu}{\partial t}=0. ∂ t ∂ [ ⋅ , ⋅ ] μ = δ [ ⋅ , ⋅ ] μ ( Ric μ + 2 1 Id ) , ∂ t ∂ ω μ = 0 , ∂ t ∂ g μ = 0.
As the metric g ~ \widetilde{g} g on the base and the metric on the adjoint bundle h h h are both constant, we write F ( μ ) = F ( μ , h 0 , g ~ ) \mathcal{F}(\mu)=\mathcal{F}(\mu,h_0,\widetilde{g}) F ( μ ) = F ( μ , h 0 , g ) . Since N is nilpotent, the Ricci curvature of the fibres is Ric ^ μ = M μ \widehat{\operatorname{Ric}}_\mu=M_\mu Ric μ = M μ and scalar curvature is R ^ μ = − 1 4 ∣ μ ∣ 2 \widehat{R}_\mu=-\frac{1}{4}|\mu|^2 R μ = − 4 1 ∣ μ ∣ 2 . By [19 Lem. 6.2] ,
∂ t Ric ^ μ = − 1 2 Δ μ ( Ric ^ μ + 1 2 Id ) , \partial_t\widehat{\operatorname{Ric}}_\mu
=
-\frac{1}{2}\Delta_\mu\left(\widehat{\operatorname{Ric}}_\mu+\frac{1}{2}\operatorname{Id}\right), ∂ t Ric μ = − 2 1 Δ μ ( Ric μ + 2 1 Id ) ,
where Δ μ = S ∘ δ μ ∗ δ μ \Delta_\mu= S \circ\delta_\mu^* \delta_\mu Δ μ = S ∘ δ μ ∗ δ μ , and S ( α ) = 1 2 ( α + α ∗ ) S(\alpha) = \frac{1}{2}(\alpha+ \alpha^*) S ( α ) = 2 1 ( α + α ∗ ) , and δ μ ∗ \delta_\mu^* δ μ ∗ is the adjoint of δ μ \delta_\mu δ μ . The time derivative of the functional is
d d t F ( μ ( t ) ) = ∫ S 1 ⟨ ∂ t Ric ^ μ , Id + 2 Ric ^ μ ⟩ d s = − ∫ S 1 ∣ δ μ ( Ric ^ μ + 1 2 Id ) ∣ 2 d s . \frac{d}{dt}\mathcal{F}(\mu(t)) = \int_{S^1} \left\langle\partial_t\widehat{\operatorname{Ric}}_\mu,\operatorname{Id}+2\widehat{\operatorname{Ric}}_\mu\right\rangle\,ds = -\int_{S^1}\left|\delta_\mu\left(\widehat{\operatorname{Ric}}_\mu+\frac{1}{2}\operatorname{Id}\right)\right|^2\,ds. d t d F ( μ ( t )) = ∫ S 1 ⟨ ∂ t Ric μ , Id + 2 Ric μ ⟩ d s = − ∫ S 1 δ μ ( Ric μ + 2 1 Id ) 2 d s .
Suppose for contradiction that there exists a sequence t k → ∞ t_k \to\infty t k → ∞ and an ε > 0 \varepsilon> 0 ε > 0 such that
− ∫ S 1 ∣ δ μ ( t k ) ( Ric ^ μ ( t k ) + 1 2 Id ) ∣ 2 d s = d d t ∣ t = t k F ( μ ( t ) ) ≤ − ε < 0. -\int_{S^1}\left|\delta_{\mu(t_k)}\left(\widehat{\operatorname{Ric}}_{\mu(t_k)}+\frac{1}{2}\operatorname{Id}\right)\right|^2\,ds = \left.\frac{d}{dt}\right|_{t=t_k}\mathcal{F}(\mu(t)) \leq-\varepsilon< 0. − ∫ S 1 δ μ ( t k ) ( Ric μ ( t k ) + 2 1 Id ) 2 d s = d t d t = t k F ( μ ( t )) ≤ − ε < 0.
The second time derivative of the bracket on ad M \operatorname{ad}M ad M is
∂ 2 μ ∂ t 2 = 1 2 ∂ μ ∂ t + p ( μ ) , \frac{\partial^2\mu}{\partial t^2}=\frac{1}{2}\frac{\partial\mu}{\partial t}+p(\mu), ∂ t 2 ∂ 2 μ = 2 1 ∂ t ∂ μ + p ( μ ) ,
where p ( μ ) p(\mu) p ( μ ) is some polynomial of μ \mu μ . By Theorem 4.3 , under the normalised bracket flow ∣ μ ∣ ≤ C |\mu| \leq C ∣ μ ∣ ≤ C for some constant C > 0 C > 0 C > 0 . Then, for some constant C > 0 C > 0 C > 0 ,
∣ ∂ 2 μ ∂ t 2 ∣ ≤ 1 2 ∣ ∂ μ ∂ t ∣ + C , \left|\frac{\partial^2\mu}{\partial t^2}\right| \leq\frac{1}{2}\left|\frac{\partial\mu}{\partial t}\right|+C, ∂ t 2 ∂ 2 μ ≤ 2 1 ∂ t ∂ μ + C ,
and so
d 2 d t 2 F ( μ ( t ) ) = − 2 ∫ S 1 ⟨ ∂ 2 μ ∂ t 2 , ∂ μ ∂ t ⟩ d s ≤ d d t F ( μ ( t ) ) + C ℓ , \frac{d^2}{dt^2}\mathcal{F}(\mu(t))=-2\int_{S^1}\left\langle\frac{\partial^2\mu}{\partial t^2},\frac{\partial\mu}{\partial t}\right\rangle\,ds\leq\frac{d}{dt}\mathcal{F}(\mu(t))+C\ell, d t 2 d 2 F ( μ ( t )) = − 2 ∫ S 1 ⟨ ∂ t 2 ∂ 2 μ , ∂ t ∂ μ ⟩ d s ≤ d t d F ( μ ( t )) + C ℓ ,
where ℓ \ell ℓ is the length of ( S 1 , g ˇ ) (S^1,\check{g}) ( S 1 , g ˇ ) . By the maximum principle,
d d t F ( μ ( t ) ) ≤ e t − t k ( C ℓ + d d t ∣ t = t k F ( μ ( t ) ) ) − C ℓ ≤ e t − t k ( C ℓ − ε ) − C ℓ \frac{d}{dt}\mathcal{F}(\mu(t))\leq e^{t-t_k}\left(C\ell+\left.\frac{d}{dt}\right|_{t=t_k}\mathcal{F}(\mu(t))\right)-C\ell\leq e^{t-t_k}(C\ell-\varepsilon)-C\ell d t d F ( μ ( t )) ≤ e t − t k ( C ℓ + d t d t = t k F ( μ ( t )) ) − C ℓ ≤ e t − t k ( C ℓ − ε ) − C ℓ
for t ≥ t k t\geq t_k t ≥ t k . There exists a δ > 0 \delta>0 δ > 0 such that
d d t F ( μ ( t ) ) < − ε 2 \frac{d}{dt}\mathcal{F}(\mu(t))<-\frac{\varepsilon}{2} d t d F ( μ ( t )) < − 2 ε
for t − t k < δ t-t_k<\delta t − t k < δ . Pass to a subsequence with t k + 1 > t k + δ t_{k+1}>t_k+\delta t k + 1 > t k + δ for all k k k . By Cauchy–Schwarz
1 n ( R ^ μ + n 2 ) 2 − n 4 = R ^ μ + 1 n R ^ μ 2 ≤ R ^ μ + ∣ Ric ^ μ ∣ 2 , \frac{1}{n}\left(\widehat{R}_\mu+\frac{n}{2}\right)^2-\frac{n}{4}=\widehat{R}_\mu+\frac{1}{n}\widehat{R}_\mu^2\leq\widehat{R}_\mu+\left|\widehat{\operatorname{Ric}}_\mu\right|^2, n 1 ( R μ + 2 n ) 2 − 4 n = R μ + n 1 R μ 2 ≤ R μ + Ric μ 2 ,
so − n 4 ℓ ≤ F ( μ ( t ) ) -\frac{n}{4}\ell\leq\mathcal{F}(\mu(t)) − 4 n ℓ ≤ F ( μ ( t )) . This provides the following contradiction
∑ k = 1 ∞ δ ε 2 ≤ − ∑ k = 1 ∞ ∫ t k t k + δ d d t F ( μ ( t ) ) d t ≤ − ∫ 0 ∞ d d t F ( μ ( t ) ) d t ≤ F ( μ ( 0 ) ) + n 4 ℓ < ∞ . \sum_{k=1}^{\infty}\frac{\delta\varepsilon}{2}\leq-\sum_{k=1}^{\infty}\int_{t_k}^{t_k+\delta}\frac{d}{dt}\mathcal{F}(\mu(t))\,dt\leq-\int_0^\infty\frac{d}{dt}\mathcal{F}(\mu(t))\,dt\leq\mathcal{F}(\mu(0))+\frac{n}{4}\ell<\infty. k = 1 ∑ ∞ 2 δ ε ≤ − k = 1 ∑ ∞ ∫ t k t k + δ d t d F ( μ ( t )) d t ≤ − ∫ 0 ∞ d t d F ( μ ( t )) d t ≤ F ( μ ( 0 )) + 4 n ℓ < ∞.
□ \square □
Proof of Theorem C. By Theorem 5.1 , the time-translated normalised Ricci flows g k ( t ) : = g ( t + t k ) g_k(t):=g(t+t_k) g k ( t ) := g ( t + t k ) subconverge to an invariant normalised Ricci flow on a principal N ∞ N_\infty N ∞ -bundle over R \mathbb{R} R ,
( M ~ ∞ , g ~ ∞ ( t ) , p ~ ∞ ) , (\widetilde{M}_\infty,\widetilde{g}_\infty(t),\widetilde{p}_\infty), ( M ∞ , g ∞ ( t ) , p ∞ ) ,
where N ∞ N_\infty N ∞ is a simply connected nilpotent Lie group. Let ℓ ( t ) = 2 diam ( S 1 , g ˇ ( t ) ) \ell(t)=2\operatorname{diam}(S^1,\check{g}(t)) ℓ ( t ) = 2 diam ( S 1 , g ˇ ( t )) . By Theorem 5.4, ℓ ( t ) → ℓ ∞ ≥ 0 \ell(t)\to\ell_\infty\geq0 ℓ ( t ) → ℓ ∞ ≥ 0 . First, if ℓ ∞ = 0 \ell_\infty=0 ℓ ∞ = 0 , then ( M ~ ∞ , g ~ ∞ ( t ) ) (\widetilde{M}_\infty,\widetilde{g}_\infty(t)) ( M ∞ , g ∞ ( t )) is homogeneous, and by [BL18, Thm. A] and [BL23, Cor. B], we get our desired conclusion.
Next, if ℓ ∞ > 0 \ell_\infty>0 ℓ ∞ > 0 , then by Theorem 5.4, there exists a constant d > 0 d>0 d > 0 such that
0 < d ≤ diam ( S 1 , g ~ ( t ) ) 0<d\leq\operatorname{diam}(S^1,\widetilde{g}(t)) 0 < d ≤ diam ( S 1 , g ( t ))
for all $t\geq 0. By the same lemma, ( M ~ ∞ , g ~ ∞ ( t ) ) (\widetilde{M}_\infty,\widetilde{g}_\infty(t)) ( M ∞ , g ∞ ( t )) exhibits the following properties
δ [ ⋅ , ⋅ ] ( L ) = 0 , ∣ L ∣ 2 = 1 2 , ∇ ^ ∗ L = 0 , ∇ L = 0 , \delta_{[\cdot,\cdot]}(L)=0,\qquad|L|^2=\frac{1}{2},\qquad\widehat{\nabla}^*L=0,\qquad\nabla L=0, δ [ ⋅ , ⋅ ] ( L ) = 0 , ∣ L ∣ 2 = 2 1 , ∇ ∗ L = 0 , ∇ L = 0 ,
where [ ⋅ , ⋅ ] [\cdot,\cdot] [ ⋅ , ⋅ ] is the Lie bracket on the adjoint bundle. Let ( [ ⋅ , ⋅ ] μ , ω μ , g ˇ μ ) ( t ) ([\cdot,\cdot]_\mu,\omega_\mu,\check{g}_\mu)(t) ([ ⋅ , ⋅ ] μ , ω μ , g ˇ μ ) ( t ) be the bracket flow corresponding to the normalised Ricci flow g ~ ∞ ( t ) \widetilde{g}_\infty(t) g ∞ ( t ) . By the above properties, the normalised bracket flow equations in the mean curvature gauge (58–60) are
∂ [ ⋅ , ⋅ ] μ ∂ t = δ [ ⋅ , ⋅ ] μ ( Ric ^ μ + 1 2 Id ) , ∂ ω μ ∂ t = 0 , ∂ g ˇ μ ∂ t = 0. \frac{\partial[\cdot,\cdot]_\mu}{\partial t}=\delta_{[\cdot,\cdot]_\mu}\left(\widehat{\operatorname{Ric}}_\mu+\frac{1}{2}\operatorname{Id}\right),\qquad\frac{\partial\omega_\mu}{\partial t}=0,\qquad\frac{\partial\check{g}_\mu}{\partial t}=0. ∂ t ∂ [ ⋅ , ⋅ ] μ = δ [ ⋅ , ⋅ ] μ ( Ric μ + 2 1 Id ) , ∂ t ∂ ω μ = 0 , ∂ t ∂ g ˇ μ = 0.
As R \mathbb{R} R is contractible, M ~ ∞ \widetilde{M}_{\infty} M ∞ is a trivial principal bundle. By Theorem 2.12, the bracket flow ( [ ⋅ , ⋅ ] μ , ω μ , g ~ μ ) ( t ) ([\cdot,\cdot]_{\mu},\omega_{\mu},\widetilde{g}_{\mu})(t) ([ ⋅ , ⋅ ] μ , ω μ , g μ ) ( t ) gives us a well-defined flow of Riemannian metrics g ~ μ ( t ) \widetilde{g}_{\mu}(t) g μ ( t ) on a flow of trivial principal bundles M ~ μ ( t ) → R \widetilde{M}_{\mu}(t)\to\mathbb{R} M μ ( t ) → R . As δ μ ( L ) = 0 \delta_{\mu}(L)=0 δ μ ( L ) = 0 , the bracket on the adjoint bundle is parallel with respect to the metric connection, that is, in a parallel frame, the bracket is constant. Because of this, we can write the bracket flow as
( M ~ ∞ , g ~ μ ( t ) , p ~ ) ≅ ( R × N μ ( t ) , d s 2 + h μ ( t ) ( s ) , ( 0 , e ) ) . (\widetilde{M}_{\infty},\widetilde{g}_{\mu}(t),\widetilde{p}) \cong(\mathbb{R}\times N_{\mu}(t),ds^{2}+h_{\mu}(t)(s),(0,e)). ( M ∞ , g μ ( t ) , p ) ≅ ( R × N μ ( t ) , d s 2 + h μ ( t ) ( s ) , ( 0 , e )) .
where h μ ( t ) ( s ) h_{\mu}(t)(s) h μ ( t ) ( s ) is a flow of one-parameter families of right-invariant metrics on the flow of Lie groups N μ ( t ) N_{\mu}(t) N μ ( t ) . Let t k → ∞ t_k\to\infty t k → ∞ be some given sequence and take μ k : = μ ( t k ) \mu_k:=\mu(t_k) μ k := μ ( t k ) . By Theorem 4.3 , this sequence lies within a bounded set in a finite-dimensional space. After passing to a subsequence, μ k \mu_k μ k converges to a nilpotent bracket μ ∞ \mu_{\infty} μ ∞ . The Lie exponential maps exp μ k : ( n ∞ , μ k ) → N μ k \exp_{\mu_k}:(\mathfrak{n}_{\infty},\mu_k)\to N_{\mu_k} exp μ k : ( n ∞ , μ k ) → N μ k give us the Cheeger–Gromov diffeomorphisms:
Φ k : R × N μ ∞ → R × N μ k , Φ k ( s , exp μ ∞ ( X ) ) = ( s , exp μ k ( X ) ) . \Phi_k:\mathbb{R}\times N_{\mu_{\infty}}\to\mathbb{R}\times N_{\mu_k},\qquad\Phi_k(s,\exp_{\mu_{\infty}}(X))=(s,\exp_{\mu_k}(X)). Φ k : R × N μ ∞ → R × N μ k , Φ k ( s , exp μ ∞ ( X )) = ( s , exp μ k ( X )) .
Indeed, Φ k ( 0 , e ) = ( 0 , e ) \Phi_k(0,e)=(0,e) Φ k ( 0 , e ) = ( 0 , e ) and
Φ k ∗ ( d s 2 + h μ k ( s ) ) → d s 2 + h μ ∞ ( s ) \Phi_k^*(ds^2+h_{\mu_k}(s))\to ds^2+h_{\mu_{\infty}}(s) Φ k ∗ ( d s 2 + h μ k ( s )) → d s 2 + h μ ∞ ( s )
smoothly on compact sets, where h μ ∞ ( s ) h_{\mu_{\infty}}(s) h μ ∞ ( s ) is the right-invariant metric on N μ ∞ N_{\mu_{\infty}} N μ ∞ which agrees with h μ k ( s ) h_{\mu_k}(s) h μ k ( s ) along the identities. By Theorem 5.5, the fibres ( N μ ∞ , h μ ∞ ( s ) ) (N_{\mu_{\infty}},h_{\mu_{\infty}}(s)) ( N μ ∞ , h μ ∞ ( s )) are nilsolitons with soliton constant − 1 2 -\frac{1}{2} − 2 1 . Call the soliton derivations D ∞ ( s ) ∈ Der ( n ∞ , μ ∞ ) D_{\infty}(s)\in\operatorname{Der}(\mathfrak{n}_{\infty},\mu_{\infty}) D ∞ ( s ) ∈ Der ( n ∞ , μ ∞ ) . We retain the properties
δ μ ∞ ( L ∞ ) = 0 , ∣ L ∞ ∣ 2 = 1 2 , ∇ ^ ∗ L ∞ = 0 , ∇ L ∞ = 0. \delta_{\mu_{\infty}}(L_{\infty})=0,\qquad|L_{\infty}|^{2}=\frac{1}{2},\qquad\widehat{\nabla}^{*}L_{\infty}=0,\qquad\nabla L_{\infty}=0. δ μ ∞ ( L ∞ ) = 0 , ∣ L ∞ ∣ 2 = 2 1 , ∇ ∗ L ∞ = 0 , ∇ L ∞ = 0.
From the first property, L ∞ ∈ Der ( n ∞ , μ ∞ ) L_{\infty}\in\operatorname{Der}(\mathfrak{n}_{\infty},\mu_{\infty}) L ∞ ∈ Der ( n ∞ , μ ∞ ) . In the horizontal trivialisation R × N μ ∞ \mathbb{R}\times N_{\mu_{\infty}} R × N μ ∞ , the shape operator L ∞ L_{\infty} L ∞ is constant, hence
h μ ∞ ( s ) = ( e s L ∞ ) ∗ h μ ∞ ( 0 ) . h_{\mu_{\infty}}(s)=(e^{sL_{\infty}})^{*}h_{\mu_{\infty}}(0). h μ ∞ ( s ) = ( e s L ∞ ) ∗ h μ ∞ ( 0 ) .
Indeed, g μ ∞ = d s 2 + h μ ∞ ( s ) g_{\mu_{\infty}}=ds^{2}+h_{\mu_{\infty}}(s) g μ ∞ = d s 2 + h μ ∞ ( s ) is a homogeneous metric that is invariant under right multiplication by the solvable group
S ∞ = R ⋉ L ∞ N μ ∞ , [ a , U ] = L ∞ U . S_{\infty}=\mathbb{R}\ltimes_{L_{\infty}}N_{\mu_{\infty}},\qquad[a,U]=L_{\infty}U. S ∞ = R ⋉ L ∞ N μ ∞ , [ a , U ] = L ∞ U .
Here the multiplication is ( s , n ) ( r , m ) = ( s + r , ϕ r ( n ) m ) (s,n)(r,m)=(s+r,\phi_r(n)m) ( s , n ) ( r , m ) = ( s + r , ϕ r ( n ) m ) , where d ϕ r = e − r L ∞ d\phi_r=e^{-rL_{\infty}} d ϕ r = e − r L ∞ . It remains to show that g μ ∞ g_{\mu_{\infty}} g μ ∞ is a solvsoliton. Consider O’Neill’s formulas for the projection S ∞ → R S_{\infty}\to\mathbb{R} S ∞ → R (Theorem 2.6):
Ric g μ ∞ a d ( U ) = − 1 2 U + D ∞ ( s ) U − ( tr L ∞ ) L ∞ U , \operatorname{Ric}^{\mathrm{ad}}_{g_{\mu_{\infty}}}(U)=-\frac{1}{2}U+D_{\infty}(s)U-(\operatorname{tr}L_{\infty})L_{\infty}U, Ric g μ ∞ ad ( U ) = − 2 1 U + D ∞ ( s ) U − ( tr L ∞ ) L ∞ U ,
Ric g μ ∞ a d , T B ( X ) = 0 , \operatorname{Ric}^{\mathrm{ad},TB}_{g_{\mu_{\infty}}}(X)=0, Ric g μ ∞ ad , T B ( X ) = 0 ,
Ric g μ ∞ T B ( X ) = − 1 2 X . \operatorname{Ric}^{TB}_{g_{\mu_{\infty}}}(X)=-\frac{1}{2}X. Ric g μ ∞ T B ( X ) = − 2 1 X .
The nilsoliton derivation is parallel since, by Theorem 3.5,
h μ ∞ ( ∇ D ∞ , α ) = h μ ∞ ( ∇ Ric ^ μ ∞ , α ) = 1 2 h μ ∞ ( δ μ ∞ ( α ) , δ μ ∞ ( L ∞ ) ) = 0 h_{\mu_{\infty}}(\nabla D_{\infty},\alpha)=h_{\mu_{\infty}}(\nabla\widehat{\operatorname{Ric}}_{\mu_{\infty}},\alpha)=\frac{1}{2}h_{\mu_{\infty}}(\delta_{\mu_{\infty}}(\alpha),\delta_{\mu_{\infty}}(L_{\infty}))=0 h μ ∞ ( ∇ D ∞ , α ) = h μ ∞ ( ∇ Ric μ ∞ , α ) = 2 1 h μ ∞ ( δ μ ∞ ( α ) , δ μ ∞ ( L ∞ )) = 0
for all α ∈ End ( n ∞ ) \alpha\in\operatorname{End}(\mathfrak{n}_{\infty}) α ∈ End ( n ∞ ) . As L ∞ L_{\infty} L ∞ is a symmetric derivation, it commutes with Ric ^ μ ∞ \widehat{\operatorname{Ric}}_{\mu_{\infty}} Ric μ ∞ , and thus [ D ∞ , L ∞ ] = 0 [D_{\infty},L_{\infty}]=0 [ D ∞ , L ∞ ] = 0 [15 Lem. 2.2] . Indeed, g μ ∞ g_{\mu_{\infty}} g μ ∞ is a solvsoliton with constant − 1 2 -\frac{1}{2} − 2 1 and solvsoliton derivation equal to
D ∞ − ( tr L ∞ ) L ∞ D_{\infty}-(\operatorname{tr}L_{\infty})L_{\infty} D ∞ − ( tr L ∞ ) L ∞
on n ∞ \mathfrak{n}_{\infty} n ∞ and zero on R a \mathbb{R}a R a . □
Base collapse
Corollary 5.6 (Theorem D). Let N N N be a simply-connected nilpotent Lie group. Let M → S 1 M\to S^{1} M → S 1 be a twisted principal N N N -bundle with unipotent monodromy. If g ( t ) g(t) g ( t ) is a normalised Ricci flow of invariant metrics on M M M , then
lim t → ∞ diam ( S 1 , g ~ ( t ) ) = 0 , \lim_{t\to\infty}\operatorname{diam}(S^{1},\widetilde{g}(t))=0, t → ∞ lim diam ( S 1 , g ( t )) = 0 , where g ~ ( t ) \widetilde{g}(t) g ( t ) is the quotient metric.
Proof. The length of the base is ℓ ( t ) = 2 diam ( S 1 , g ˇ ( t ) ) \ell(t)=2\operatorname{diam}(S^1,\check{g}(t)) ℓ ( t ) = 2 diam ( S 1 , g ˇ ( t )) . By Theorem 5.4, ℓ ( t ) → ℓ ∞ ≥ 0 \ell(t)\to\ell_\infty\geq0 ℓ ( t ) → ℓ ∞ ≥ 0 . Suppose for contradiction that ℓ ∞ > 0 \ell_\infty>0 ℓ ∞ > 0 . Then, by Theorem 5.4, there exists a constant d > 0 d>0 d > 0 such that
0 < d ≤ diam ( S 1 , g ˇ ( t ) ) 0<d\leq\operatorname{diam}(S^1,\check{g}(t)) 0 < d ≤ diam ( S 1 , g ˇ ( t ))
for all t ≥ 0 t\geq0 t ≥ 0 . By Theorem 5.1 , the time-translated normalised Ricci flows g ( t + t k ) g(t+t_k) g ( t + t k ) subconverge to an invariant normalised Ricci flow on a principal N ∞ N_\infty N ∞ -bundle over R \mathbb{R} R ,
( M ~ ∞ , g ~ ∞ ( t ) , p ~ ∞ ) , (\widetilde{M}_\infty,\widetilde{g}_\infty(t),\widetilde{p}_\infty), ( M ∞ , g ∞ ( t ) , p ∞ ) ,
where N ∞ N_\infty N ∞ is a simply connected nilpotent Lie group. From the diameter bound, ( M ~ ∞ , g ~ ∞ ( t ) ) (\widetilde{M}_\infty,\widetilde{g}_\infty(t)) ( M ∞ , g ∞ ( t )) descends to an invariant normalised Ricci flow on a twisted principal N ∞ N_\infty N ∞ -bundle M ∞ → S 1 M_\infty\to S^1 M ∞ → S 1 . By Theorem 5.4, ( M ~ ∞ , g ~ ∞ ( t ) ) (\widetilde{M}_\infty,\widetilde{g}_\infty(t)) ( M ∞ , g ∞ ( t )) has the following properties:
δ [ ⋅ , ⋅ ] ( L ∞ ) = 0 , ∣ L ∞ ∣ 2 = 1 2 , ∇ ^ ∗ L ∞ = 0 , ∇ L ∞ = 0. (62) \delta_{[\cdot,\cdot]}(L_\infty)=0,\qquad\lvert L_\infty\rvert^2=\frac{1}{2},\qquad\widehat{\nabla}^{*}L_\infty=0,\qquad\nabla L_\infty=0. \tag*{(62)} δ [ ⋅ , ⋅ ] ( L ∞ ) = 0 , ∣ L ∞ ∣ 2 = 2 1 , ∇ ∗ L ∞ = 0 , ∇ L ∞ = 0. (62)
From the last property, L ∞ L_\infty L ∞ is ∇ a d \nabla^{\mathrm{ad}} ∇ ad -parallel. In a horizontal trivialisation, L ∞ L_\infty L ∞ is constant, so if h ∞ : R → Sym + 2 ( n ∞ ∗ ) h_\infty:\mathbb{R}\to\operatorname{Sym}^{2}_{+}(\mathfrak{n}^{*}_\infty) h ∞ : R → Sym + 2 ( n ∞ ∗ ) is the vector bundle metric on the adjoint bundle, then
h ∞ ( s ) = ( e s L ∞ ) ∗ h ∞ ( 0 ) . h_\infty(s)=(e^{sL_\infty})^{*}h_\infty(0). h ∞ ( s ) = ( e s L ∞ ) ∗ h ∞ ( 0 ) .
If P = d ϕ e ∘ Ad ( n ) P=d\phi_e\circ\operatorname{Ad}(n) P = d ϕ e ∘ Ad ( n ) is the adjoint holonomy of ∇ a d \nabla^{\mathrm{ad}} ∇ ad at time t = 0 t=0 t = 0 , then
P ∗ P = e 2 ℓ ∞ L ∞ , (63) P^{*}P=e^{2\ell_\infty L_\infty}, \tag*{(63)} P ∗ P = e 2 ℓ ∞ L ∞ , (63)
where P ∗ P^{*} P ∗ is the h ∞ ( 0 ) h_\infty(0) h ∞ ( 0 ) -adjoint of P P P . Since L ∞ L_\infty L ∞ is ∇ a d \nabla^{\mathrm{ad}} ∇ ad -parallel, it commutes with P P P . It follows by (63) that P P P commutes with P ∗ P P^{*}P P ∗ P , hence
P ∗ P = P P ∗ , P^{*}P=PP^{*}, P ∗ P = P P ∗ ,
that is, P P P is normal.
Let P k ∈ Aut ( n ) P_k\in\operatorname{Aut}(\mathfrak{n}) P k ∈ Aut ( n ) be the adjoint holonomy of ∇ k a d \nabla^{\mathrm{ad}}_k ∇ k ad at time t = 0 t=0 t = 0 . Write P k = d ( ϕ k ) e ∘ Ad ( n k ) P_k=d(\phi_k)_e\circ\operatorname{Ad}(n_k) P k = d ( ϕ k ) e ∘ Ad ( n k ) , where ϕ k \phi_k ϕ k is the monodromy and n k ∈ N n_k\in N n k ∈ N . Let { n j } \{\mathfrak{n}^{j}\} { n j } denote the lower central series of n \mathfrak{n} n . As the exponential map is surjective, there exists X k ∈ n X_k\in\mathfrak{n} X k ∈ n such that Ad ( n k ) = exp ( ad X k ) \operatorname{Ad}(n_k)=\exp(\operatorname{ad}X_k) Ad ( n k ) = exp ( ad X k ) . Thus,
( Id − Ad ( n k ) ) ( n j ) ⊆ n j + 1 , (\operatorname{Id}-\operatorname{Ad}(n_k))(\mathfrak{n}^{j})\subseteq\mathfrak{n}^{j+1}, ( Id − Ad ( n k )) ( n j ) ⊆ n j + 1 ,
and as automorphisms preserve the lower central series, P k = d ( ϕ k ) e ∘ Ad ( n k ) = d ( ϕ k ) e P_k=d(\phi_k)_e\circ\operatorname{Ad}(n_k)=d(\phi_k)_e P k = d ( ϕ k ) e ∘ Ad ( n k ) = d ( ϕ k ) e on the quotients n j / n j + 1 \mathfrak{n}^{j}/\mathfrak{n}^{j+1} n j / n j + 1 . By assumption, d ( ϕ k ) e d(\phi_k)_e d ( ϕ k ) e is unipotent, so ( Id − d ( ϕ k ) e ) n = 0 (\operatorname{Id}-d(\phi_k)_e)^n=0 ( Id − d ( ϕ k ) e ) n = 0 where n = dim n n=\dim\mathfrak{n} n = dim n . Hence,
( Id − P k ) n ( n j ) = ( Id − d ( ϕ k ) e ∘ Ad ( n k ) ) n ( n j ) ⊆ n j + 1 , (\operatorname{Id}-P_k)^n(\mathfrak{n}^{j})=(\operatorname{Id}-d(\phi_k)_e\circ\operatorname{Ad}(n_k))^n(\mathfrak{n}^{j})\subseteq\mathfrak{n}^{j+1}, ( Id − P k ) n ( n j ) = ( Id − d ( ϕ k ) e ∘ Ad ( n k ) ) n ( n j ) ⊆ n j + 1 ,
so if n \mathfrak{n} n is m m m -step nilpotent, then
( Id − P k ) n m ( n ) = ( Id − d ( ϕ k ) e ∘ Ad ( n k ) ) n m ( n ) ⊆ n m + 1 = { 0 } . (\operatorname{Id}-P_k)^{nm}(\mathfrak{n})=(\operatorname{Id}-d(\phi_k)_e\circ\operatorname{Ad}(n_k))^{nm}(\mathfrak{n})\subseteq\mathfrak{n}^{m+1}=\{0\}. ( Id − P k ) nm ( n ) = ( Id − d ( ϕ k ) e ∘ Ad ( n k ) ) nm ( n ) ⊆ n m + 1 = { 0 } .
Indeed, P k P_k P k is unipotent, and by Theorem 5.2, after passing to a subsequence, there exist linear isometries ρ k : n ∞ → n \rho_k:\mathfrak{n}_\infty\to\mathfrak{n} ρ k : n ∞ → n such that ρ k − 1 P k ρ k → P \rho_k^{-1}P_k\rho_k\to P ρ k − 1 P k ρ k → P . It follows that
( Id − P ) n = lim k → ∞ ρ k − 1 ( Id − P k ) n ρ k = 0 , (\operatorname{Id}-P)^n=\lim_{k\to\infty}\rho_k^{-1}(\operatorname{Id}-P_k)^n\rho_k=0, ( Id − P ) n = k → ∞ lim ρ k − 1 ( Id − P k ) n ρ k = 0 ,
and therefore, P P P is unipotent. However, a unipotent normal matrix is the identity, so (63) becomes
Id = e 2 ℓ ∞ L ∞ . \operatorname{Id}=e^{2\ell_\infty L_\infty}. Id = e 2 ℓ ∞ L ∞ .
This means L ∞ = 0 L_\infty=0 L ∞ = 0 , which contradicts ∣ L ∞ ∣ 2 = 1 2 \lvert L_\infty\rvert^2=\frac{1}{2} ∣ L ∞ ∣ 2 = 2 1 in (62) . □
References