Posted by: matheuscmss | June 25, 2008

Bohm and Wilking’s method of deformation of Ricci flow invariant curvature conditions

In a previous post about the proof of the differentiable sphere theorem (due to S. Brendle and R. Schoen), we saw that a fundamental ingredient was the recent technique of deformation of Ricci flow invariant cones of ”curvature conditions” introduced by C. Bohm and B. Wilking. Today we will discuss a little bit more of the details of this interesting new method of Bohm and Wilking for the analysis of the Ricci flow.

Let me start with some historical remarks: Ricci flow was introduced in 1982 by R. Hamilton in order to show that compact 3-manifolds with positive Ricci curvature admit metrics of positive constant sectional curvature (i.e., they are spherical space-forms). See a link to Hamilton’s article here. This gave a positive answer to a question of J. Bourguignon). Also, in 1986, Hamilton obtained similar conclusions in dimension 4 for positive curvature operators. Nevertheless, H. Chen generalized Hamilton’s result by showing that compact 4-manifolds with 2-positive curvature operator are spherical space-forms (recall that 2-positive curvature means that the sum of the two smallest eigenvalues of the curvature operator is positive). Finally, in view of these results, Hamilton conjectured that compact manifolds of any dimension with positive curvature operators are spherical space-forms.

In this direction, Bohm and Wilking introduced a new important tool allowing to confirm Hamilton’s conjecture. More precisely, they showed the following theorem:

Theorem (Bohm and Wilking, 2006). The normalized Ricci flow of a compact manifold with 2-positive curvature operator evolves to a limit metric with positive constant sectional curvature.

During the rest of this post, we will discuss some aspects of Bohm and Wilking’s technique and its application to the proof of the theorem above. To do so, we divide the post into four sections: in the next section we will outline Bohm and Wilking’s argument; after that, we dedicate the subsequent section to review relevant facts concerning the algebraic properties of curvature operators; the third section will contain a description of Bohm and Wilking construction of ”pinching families” of cones of curvature operators and the last section contains a brief exposition of the application of this pinching families method to the proof of the desired theorem.

Outline of Bohm and Wilking method

Recall that the Ricci flow is the evolution equation

\frac{\partial g}{\partial t} = -2 Ric(g),

where g=g_t is a curve of Riemannian metrics on a manifold M^n. Following R. Hamilton (see also this book of B. Chow and D. Knopf), one can use moving frames (a.k.a. Uhlenbeck’s trick) to obtain that the curvature operators R_t of the metrics g_t satisfies the following evolution equation:

\frac{\partial R}{\partial t} = \Delta R + 2(R^2+R^{\#})

where R^{\#}:=ad\circ (R\wedge R)\circ ad^* (here we are using the identification \Lambda^2 T_pM \simeq \mathfrak{so}(T_pM) and ad: \Lambda^2(T_pM)\to\mathfrak{so}(T_pM) the adjoint representation).

Remark 1. In the sequel, we normalize the constants so that the curvature operator of the round sphere of radius 1 is the identity.

A important tool for the study of the Ricci flow is the so-called ”Hamilton’s maximum principle”: it says that a closed convex O(n)-invariant subset C of curvature operators staying invariant under the ODE

(1) \frac{dR}{dt} = Q(R):=R^2+R^{\#}

defines a Ricci flow invariant curvature condition. In other words, if the curvature operator of the initial metric g_0 belongs to a subset C which remains invariant under (1), then the curvature operator of every g_t also belongs to C (i.e., curvature constraints of g_0 preserved by (1) are not destroyed by the Ricci flow). Of course, it is not easy to apply Hamilton’s maximum principle in higher dimensions because the ODE (1) is not well-understood (this explains why the results of Hamilton and Chen quoted in the beginning are restricted to dimensions 3 and 4).

Nevertheless, Bohm and Wilking were able to exploit Hamilton’s maximum principle in the following way: we start with C(0) given by the closure of the open cone of 2-positive curvature operators (C(0) is taken closed just to fit the assumptions of Hamilton’s maximum principle) and we would like to show that any metric whose curvature operator R belongs to the interior of C(0) evolves by the Ricci flow to a metric of positive constant sectional curvature. Assume that one can construct a family of invariant cones C(s), 0\leq s< 1, starting at C(0), such that the vector field Q(R)=R^2+R^{\#} always points towards the interior of C(s) for any s and C(s) converges to \mathbb{R}^+\cdot I when s\to 1 (with I being the identity curvature operator).

Remark 2. At this point I strongly recommend the reader to make some qualitative pictures of an ODE associated to a vector field with these properties (since I don’t know how to insert xfig files in WordPress posts ­čśŽ ).

As the reader can from the pictures and Hamilton’s maximum principle, the existence of such a family of cones of curvature operators guarantees (intuitively) that the flow generated by the ODE (1) evolves any metric inside C(0) to a metric with curvature operator equals to a constant multiple of the identity, i.e., a metric of constant sectional curvature. Another way of thinking about this argument passes by a continuity argument: we begin with the cone C(0) and we run the Ricci flow. Since the vector field points towards the interior of C(0), we evolve to a curvature operator inside a cone C(\varepsilon_0) for some \varepsilon_0>0. Moreover, from Hamilton’s maximum principle, each time we attain a given cone C(\varepsilon_0), our future evolution is faded to stay inside the cone C(\varepsilon_0) forever. Nevertheless, we know that the vector field points towards the interior of C(\varepsilon_0), so that we eventually reach a smaller cone C(\varepsilon_1) for some 0<\varepsilon_0<\varepsilon_1. Reproducing this argument ad infinitum (and crossing fingers ­čÖé ) we ”conclude” that the Ricci flow will send our initial metric (whose operator belongs to C(0)) to a metric inside the limit cone \lim\limits_{s\to 1}C(s) = \mathbb{R}^+\cdot I, so that the proof would be complete.

Thus, it remains ”only” to justify (heuristically, at least) why one should expect such a family of cones to exist. A (very) naive idea consists into directly connect a nonnegative curvature operator R to a multiple of identity part, say l_{c}(R):=R+c R_I, where R_I = \frac{tr(Ric)}{n(n-1)} I, Ric is the Ricci tensor of R, and consider the family of cones l_c(C(0)). Observe that l_c(C(0)) converges to \mathbb{R}^+\cdot I when c\to\infty (since \frac{1}{c}l_c(R)\to R_I when c\to\infty). However, it is not obvious that this family is invariant by the ODE (1) and the vector field Q(R) is always pointing inward. In order to check these two conditions, one can take the pull-back l_c^{-1}(l_c(R)^2+l_c(R)^{\#}) of the vector-field Q(R)=R^2+R^{\#} by l_c. It follows that l_c(C(0)) is invariant by the ODE (1) if and only if l_c^{-1}(l_c(R)^2+l_c(R)^{\#}) belongs to the tangent cone T_R C(0) of C(0) at R. On the other hand, since we are assuming that Q(R)\in T_RC(0), it suffices to get that the difference

l_c^{-1}(l_c(R)^2+l_c(R)^{\#}) - R^2-R^{\#}

lies inside T_RC(0). It turns out that this difference can be explicitly computed. Moreover, this difference depends only on the Ricci tensor of R (but not on its Weyl tensor) making the computations quite pleasant. However, a quick inspection of the formulas shows that this family is not invariant by the ODE (1) essentially because this difference has (very) negative scalar curvature for large c (so that it can’t stay inside the cone of 2-positive curvature operators). To overcome this technical difficulty, Bohm and Wilking introduce a more general family of linear operators

l_{a,b}(R) := R+2(n-1)aR_I+(n-2)bR_{Ric_0},

where R_{Ric_0} is the traceless part of the Ricci tensor of R. Here we use 2(n-1)a and (n-2)b instead of a and b just for normalization purposes. Again, we look at the associated family of cones l_{a,b}(C(0)). It is not hard to see that, for each fixed a, the family l_{a,b}(C(0)) converges to \mathbb{R}^+\cdot I. Taking the pull-back l_{a,b}^{-1}(l_{a,b}(R)^2+l_{a,b}(R)^{\#}) of the vector field Q(R) under l_{a,b} and looking at the corresponding cone l_{a,b}(C(0)), we see again that this family of cones is invariant by the ODE (1) if and only if the difference

D_{a,b}:=l_{a,b}^{-1}(l_{a,b}(R)^2+l_{a,b}(R)^{\#}) - R^2-R^{\#}

lies on T_R C(0). It turns out that D_{a,b} can be computed explicitly (essentially because it depends only on the Ricci tensor) and we can adjust the values of a, b so that D_{a,b} belongs to T_R C(0) (i.e., this family of cones is invariant by (1)) and l_{a,b}(C(0)) join C(0) to \mathbb{R}^+\cdot I (this is a specially tricky part of the proof, but it is not terrible difficult once one get an explicit formula for D_{a,b}). In particular, this allows to get the desired family of cones, so that our sketch of the Bohm and Wilking arguments is complete.

Now let us turn to the details.


Denote by S^2_B(\mathfrak{so}(n)) the vector-space of curvature operators, i.e., symmetric endomorphisms of \mathbb{R}^n verifying the first Bianchi identity (so that in our notation S^2_B(\mathfrak{so}(n)), S stands for ”symmetric” and B stands for ”Bianchi”). Consider the linear transformation l_{a,b}: S^2_B(\mathfrak{so}(n))\to S^2_B(\mathfrak{so}(n))(R) introduced in the previous section:

l_{a,b}(R):= R+2(n-1)aR_I+(n-2)bR_{Ric_0}

and the difference

D_{a,b}:=l_{a,b}^{-1}(l_{a,b}(R)^2+l_{a,b}(R)^{\#}) - R^2-R^{\#}

between the pull-back l_{a,b}^*Q(R) and Q(R). As we discussed above, one relevant step in Bohm and Wilking argument consists into deriving an explicit formula for D_{a,b}:

Theorem 1 (theorem 2 of Bohm and Wilking). \begin{array}{l} D_{a,b}:= \frac{tr(Ric_0^2)}{n+2n(n-1)a}(nb^2(1-2b)-2(a-b)(1-2b+nb^2))I \\ ((n-2)b^2-2(a-b))Ric_0\wedge Ric_0 + 2a Ric\wedge Ric+ 2b^2Ric_0^2\wedge id \end{array}.

Remark 3. Putting a=0 and b=c, we see that the difference D_{0,c} is exactly l_c^{*}Q(R)-Q(R). Using this theorem, we see that the scalar curvature of D_{0,c} (i.e., the quantity in front of the operator I) becomes negative when b=c\to\infty, so that Hamilton’s maximum principle can’t be applied to the related family of cones (in particular this family is not invariant by the ODE (1), as we announced in the previous section).

The proof of this theorem is not hard but it involves a certain amount of calculation. Thus, we will just briefly indicate the main steps.

Proof of theorem 1 (sketch). One begin by calculating the eigenvalues of the linearization R+R\# I of the vector-field Q(R)=R^2+R^{\#} around the identity I. It is not hard to see that

R+R\# I = (n-1)R_I+\frac{n-2}{2}R_{Ric_0}=Ric\wedge I.

See lemma 2.1 of Bohm and Wilking. After that, one applies this formula for an explicit calculation of the vector-field itself when the curvature operator R is a Ricci operator i.e. R=R_I+R_{Ric_0}. The resulting formula is

\begin{array}{l}R^2+R^{\#} = \frac{1}{n-2}Ric_0\wedge Ric_0 + \frac{2\lambda}{(n-1)}Ric_0\wedge id - \frac{2}{(n-2)^2}(Ric_0^2)_0\wedge id + \\ +\left(\frac{\lambda}{(n-1)}+\frac{\sigma}{(n-2)}\right)I \end{array},

where \lambda = tr(Ric)/n and \sigma = \|Ric_0\|^2/n. See lemma 2.2 of Bohm and Wilking. Finally, combining some linear algebra arguments (based on the canonical decomposition of a curvature operator R into a multiple of the identity R_I=\frac{\lambda}{n-1}I, its traceless Ricci tensor R_{Ric_0} and its Weyl tensor R_W) with this formula for Q(R) (when R is Ricci), one deduces the desired expression for the difference D_{a,b}. See the proof of theorem 2 of Bohm and Wilking. This completes our sketch of the proof of theorem 1. \square

An interesting corollary of this theorem is:

Corollary 1 (corollary 2.3 of Bohm and Wilking). Let e_1,\dots,e_n be an orthonormal basis of eigenvectors associated to the eigenvalues \lambda_1,\dots,\lambda_n of Ric_0. Then, e_i\wedge e_j is an eigenvector of D_{a,b} associated to the eigenvalue

\begin{array}{l} d_{ij} = \left((n-2)b^2- 2(a-b)\right)\lambda_i\lambda_j + 2a (\lambda+\lambda_i)(\lambda+\lambda_j) + b^2(\lambda_i^2+\lambda_j^2) \\ +\frac{\sigma}{1+2(n-1)a}(nb^2(1-2b)-2(a-b)(1-2b+nb^2)) \end{array}

where \lambda = tr(Ric)/n and \sigma = \|Ric_0\|^2/n. Also, e_i is an eigenvector of the Ricci tensor Ric(D_{a,b}) of D_{a,b} associated to the eigenvalue

\begin{array}{l} r_i = -2b\lambda_i^2+2a\lambda(n-2)\lambda_i+ 2a(n-1)\lambda^2\\ +\frac{\sigma}{1+2(n-1)a}(n^2b^2(1-2b)-2(n-1)(a-b)(1-2b)) \end{array}.

The relevance of this corollary is that the knowledge of the eigenvalues of D_{a,b} will guide our choices of parameters a,b during the construction of the desired family of cones.

Construction of pinching families of cones of curvature operators

Before starting the discussion of this section, let us recall (in a more precise fashion) the properties of the desired family of cones:

Definition 1. A continuous family \{C(s): 0\leq s<1\}\subset S_B^2(\mathfrak{so}(n)) of closed convex O(n)-invariant cones (of full dimension) is called a pinching family whenever

  • each R\in C(s)-\{0\} has positive scalar curvature,
  • Q(R)=R^2+R^{\#} stays inside the interior of the tangent cone T_R C(s) of C(s) at R for any R\in C(s)-\{0\} and 0<s<1,
  • C(s) converges (in the Hausdorff topology) to the 1-dimensional cone \mathbb{R}^+\cdot I when s\to 1.

Example 1. The family of cones of 3-dimensional curvature operators

C(s):=\left\{R\in S_B^2(\mathfrak{so}(3)): Ric\geq s\cdot\frac{tr(Ric)}{3}id\right\}, 0\leq s<1

is a pinching family starting at the cone C(0) of curvature operators (in dimension 3) with nonnegative Ricci curvature. This family of cones was used by R. Hamilton in order to show that the Ricci flow on 3-manifolds evolves any metric with positive Ricci curvature to a limit metric with positive constant sectional curvature.

Roughly speaking, the goal of this section is to show how one can use theorem 1 (and corollary 1) to generalize the previous example to the higher dimensional case.

Theorem 2 (theorem 3.1 of Bohm and Wilking). There exists a pinching family C(s), 0\leq s<1 of closed convex cones starting at the cone C(0) of 2-nonnegative curvature operators.

This pinching family will be defined piecewise by 3 subfamilies. For the construction of the first subfamily, we need the following proposition:

Proposition 1 (proposition 3.2 of Bohm and Wilking). Take n\geq 3 and let C be a closed convex O(n)-invariant cone of curvature operators which is invariant by the ODE (1). Assume that C contains all nonnegative curvature operators of rank 1, any R\in C has nonnegative Ricci curvature and C-\{0\} is contained in the half-space of positive scalar curvature operators. Then, the cone l_{a,b}(C) is invariant by the vector-field Q(R)=R^2+R^{\#} associated to the ODE (1) for

0<b\leq \frac{\sqrt{2n(n-2)+4}-2}{n(n-2)} and 2a:= 2b+(n-2)b^2.

Proof. It suffices to show that X_{a,b} = l_{a,b}^*Q(R) = l_{a,b}^{-1}(l_{a,b}(R)^2 + l_{a,b}(R)^{\#}) belongs to the interior of the tangent cone T_R C. Since Q(R)=R^2+R^{\#}\in T_R C (because C is invariant by the ODE), our task is reduce to prove that D_{a,b} = X_{a,b}-Q(R)\in int(T_RC). Since C contains all nonnegative curvature operators of rank 1 (and consequently its linear combinations because C is convex), we are done if one can show that D_{a,b} is positive for b>0. Looking at the eigenvalues of D_{a,b} computed in corollary 1 and using the definition of a in terms of b, we see that D_{a,b} is positive if and only if 0\geq b^2(n(1-2b)-(n-2)(1-2b+nb^2)), which holds in the range 0<b\leq \frac{\sqrt{2n(n-2)+4}-2}{n(n-2)} as the reader can directly check. \square

Since it is known that the cone of 2-positive curvature operators is invariant by the ODE (1) (see the paper of Hamilton), we can apply the previous proposition to get the following consequence:

Lemma 1 (corollary 3.3 of Bohm and Wilking). Assume that there exists a pinching family C(s) (0\leq s<1) starting at the cone of nonnegative curvature operators. Then, the statement of theorem 1 is true, i.e., there exists a pinching family starting at the cone C of 2-nonnegative curvature operators.

Proof. Note that the theorem 1 is well-known in dimension 3 (see the example 1). Therefore, we can suppose that n\geq 4.

Since the property of a cone being closed convex O(n)-invariant and invariant by the ODE (1) is closed by intersections (exercise), given a pinching family C(s), 0\leq s<1 starting at the cone C(0) of nonnegative curvature operators, it suffices to show that l_{a_0,b_0}(C)\subset C(0) where b_0=\frac{\sqrt{2n(n-2)+4}-2}{n(n-2)}, 2a_0:= 2b_0+(n-2)b_0^2 and C is the cone of 2-nonnegative curvature operators. Indeed, in this situation, we can extend the family l_{a,b}(C) (0<b\leq b_0 and 2a=2b+(n-2)b^2) to a pinching family starting at C by defining it on the second portion b_0\leq s<1 of the ”time” interval 0\leq s<1 as a reparametrization of the pinching family C(s)\cap l_{a_0,b_0}(C) (here we are using the closedness by intersection of the ODE-invariance property quoted in the beginning of the proof).

On the other hand, to see that l_{a_0,b_0}(C)\subset C(0), we recall the definition of l_{a,b} is

l_{a,b}(R) = R+2(n-1)aR_I+(n-2)bR_{Ric_0} = R+ hR_I+2bRic\wedge id

where h=2(n-1)(a-b). Fix R\in C-\{0\} a 2-nonnegative curvature operator. We know from standard estimates that the smallest eigenvalue \kappa of R verifies \kappa\geq-\frac{2tr(R)}{n(n-1)-4} and, since R is 2-nonnegative, the smallest eigenvalue of Ric is \geq (n-3)|\kappa|. Hence, in order to show that l_{a,b}(R)\in C(0) (i.e., l_{a_0,b_0}(R) is a nonnegative curvature operator), it suffices to see that

h=2(n-1)(a_0-b_0)=(n-2)b_0^2\geq (1-2b)\frac{n(n-1)}{n(n-1)-4}.

By the definition of b_0, we obtain (n-2)b_0^2 = 2(1-2b)/n, so that the previous inequality is valid (for n\geq 4). This completes the proof of the lemma. \square

At this stage, it remains to exhibit a pinching family starting at the cone of nonnegative curvature operators. To do so, we will use the following two subfamilies described in the next two lemmas:

Lemma 2 (lemma 3.4 of Bohm and Wilking). Given 0\leq b\leq 1/2, define

a=a(b):=\frac{(n-2)b^2+2b}{2+2(n-2)b^2} and p=\frac{(n-2)b^2}{1+(n-2)b^2}.

Then, the cone l_{a,b}\left(\{R\in S_B^2(\mathfrak{so}(n)): R\geq 0, Ric\geq p\frac{tr(Ric)}{n}\}\right) is invariant by the vector-field Q(R)=R^2+R^{\#}.

Lemma 3 (lemma 3.5 of Bohm and Wilking). Fix b=1/2 and for a given s\geq 0 put

a=a(s):=(1+s)/2 and p=1-\frac{4}{n+2+4s}.

Then, the cone l_{a,1/2}\left(\{R\in S_B^2(\mathfrak{so}(n)): R\geq 0, Ric\geq p\frac{tr(Ric)}{n}\}\right) is invariant by the vector-field Q(R)=R^2+R^{\#}.

Assuming (momentarily) these two lemmas, we can show the theorem 2:

Proof of theorem 2 (modulo lemmas 2 and 3). Firstly we apply lemma 1 to reduce the proof of theorem 2 to the existence of a pinching family of cones starting at the cone C(0) of nonnegative curvature operators. Secondly, we note that the concatenation of the families of cones provided by lemmas 2 and 3 is a pinching family starting at the cone C(0) of nonnegative curvature operators (after a reparametrization of s, of course). In fact, it is easy to see that the family of cones of lemma 2 starts at the C(0) (b=0). Also, these two families of cones coincide at the parameters b=1/2 and s=0 (so that it is possible to make the concatenation). Finally, for the curvature operators l_{a,b}(R) belonging to the cones provided by lemma 3, we have \lim\limits_{s\to\infty}\frac{1}{a(s)}l_{a(s),1/2}(R)=2(n-1)R_I (so that these cones converges to \mathbb{R}^+\cdot I). Combining these three facts, we obtain our desired pinching family (starting at C(0)), so that the proof of theorem 2 is complete. \square

Remark 4. The construction of the pinching family using both lemmas 2 and 3 is necessary because the family of cones provided by lemma 2 is not a pinching family. In fact, although this family is invariant by the ODE (1) for all b>0, it converges to the cone of Einstein curvature operators (instead of \mathbb{R}^+\cdot I).

The proofs of the lemmas 2 and 3 have a very similar flavor. In particular, we will not repeat ourselves so that we will discuss only the proof of lemma 3 (for the details about the lemma 2 see the proof of lemma 3.4 of Bohm and Wilking).

Proof of lemma 3. It suffices to show that the pull-back X_{a,1/2}:=l_{a,1/2}^*Q(R) of the vector-field Q(R) lies on the interior of the tangent cone T_RC(p) where a=(1+s)/2, p=1-\frac{4}{n+2+4s} and

C(p):=\{R\in S_B^2(\mathfrak{so}(n)): R\geq 0, Ric\geq p\frac{tr(Ric)}{n}\}.

To do so, we begin by showing that X_{a,1/2} is positive definite. Since we are assuming that R^2+R^{\#} is positive semi-definite, it is sufficient to prove that D_{a,1/2}>0. In this direction, we look at the explicit formula of the eigenvalues d_{ij} of D_{a,1/2} from corollary 1. Using the definition of a, we see that the formula of d_{ij} simplifies to:

\begin{array}{l} d_{ij}=(\frac{n-2}{4}-s)\lambda_i\lambda_j+(s+1)(\lambda+\lambda_i)(\lambda+\lambda_j) + \frac{1}{4}(\lambda_i^2+\lambda_j^2) \\ - \frac{\sigma n s}{4n+4(n-1)s}\end{array}.

From the assumption R\in C(p), we have that \lambda_i\geq -(1-p)\lambda. In particular, \sigma\leq (n-1)(1-p)^2\lambda^2 = \frac{16(n-1)\lambda^2}{(n+2+4s)^2}. Plugging this information into the previous formula, we get (for n\geq 3):

\begin{array}{l}d_{ij} = \frac{n+2}{4}(\lambda_i+\frac{4\lambda}{n+2})(\lambda_j+\frac{4\lambda}{n+2}) + s\lambda(\lambda_i+\lambda_j+\frac{8\lambda}{n+2+4s}) \\ +\frac{\lambda_i^2+\lambda_j^2}{4}+ \frac{n-2}{n+2}\lambda^2+s\frac{n-6+4s}{n+2+4s}\lambda^2 - \frac{\sigma n s}{4n+4(n-1)s} \\ \geq \left(\frac{n-2}{n+2}+s\frac{n-6+4s}{n+2+4s} - \frac{16(n-1)ns}{4n+4(n-1)s)(n+2+4s)}\right)\lambda^2\\ > (1+s(n-6)+4s^2-s)\frac{\lambda^2}{n+2+4s}\geq 0\end{array}.

Hence, D_{a,1/2} is positive. Next, we complete the proof of lemma 3 by showing that X_{a,1/2} preserves the Ricci pinching condition (appearing in the definition of C(p)). Firstly we note that the definition of a combined with the formula of corollary 1 for the eigenvalues r_i of the traceless Ricci tensor of D_{a,b} gives

r_i = -\lambda_i^2+(s+1)\lambda(n-2)\lambda_i+(s+1)(n-1)\lambda^2+\frac{\sigma n^2}{4n+4(n-1)s}.

Putting together this expression with the previous formula for d_{ij} and the eigenvalue estimate \lambda_i\geq -(1-p)\lambda (coming from the fact R\in C(p)), it is not hard to see that our task is to prove that

\begin{array}{l} 0\leq p^2\lambda^2-(1-p)^2\lambda^2 - (s+1)\lambda^2(n-2)(1-p)+(s+1)(n-1)\lambda^2 \\ + \frac{\sigma n^2}{4n+4(n-1)s} - p\left(\frac{n^2\sigma}{4n+4(n-1)s}+(n+(n-1)s)\lambda^2\right)\end{array}.

Since \sigma\geq 0, we can neglet the terms containing \sigma. Dividing by \lambda^2 the resulting expression, we obtain

\begin{array}{l}0\leq p^2-(1-p)^2+(s+1)+(s+1)p(n-2)-p(n+(n-1)s)\\ =s(1-p)\end{array}.

Since this last inequality is obvious, the proof of the lemma 3 is complete. \square

Now we will briefly indicate in the next section how to apply the theorem 2 in order to conclude Bohm and Wilking theorem.

Application of the pinching families to the proof of Bohm and Wilking theorem

Recall that the basic idea of the proof of Bohm and Wilking theorem is that the existence of a pinching family starting at the cone of 2-nonnegative curvature operators forces the ODE (1) associated to the Ricci flow to evolve the initial metric to a limit metric of constant sectional curvature (i.e., a ”round” metric). However, this is not quite immediate from the definition of pinching family essentially because the convergence of the initial metric to a round metric depends also on good bounds for some geometric quantity (such as the scalar curvature) in order to avoid a possible scenario where the (curvature operator of the) metric approaches the cone \mathbb{R}^+\cdot I and (at the same time) it blows up. Fortunately, the exclusion of this bad scenario can be done by standard ODE arguments which can be roughly described as follows. Let C(s) be a pinching family starting at the cone of 2-nonnegative curvature operators. By compactness of the manifold M, the (2-positive) curvature operator R(0) of the initial metric satisfies R(0)\in \{R: scal\leq h_0\}\cap C(\varepsilon) for a large constant h_0>0 and a small constant \varepsilon>0. It turns out that elementary arguments from the theory of ODE shows that this last property of R(0) implies the existence of an invariant set F containing \{R: scal\leq h_0\}\cap C(\varepsilon) such that F is essentially contained in C(s) for all \varepsilon\leq s<1 (the word essentially here means that F-C(s) is relatively compact). For the details of the construction of F, see the section 4 of the paper of Bohm and Wilking. Once we have the set F with the previous properties, we can use Shi’s derivative estimates to get a priori bounds for the curvature operators of the metrics g_t evolving by the Ricci flow. Moreover, one can show finite time extinction (at time t_0<\infty say) of the Ricci flow. Since the Ricci flow exists as long as the curvature stays bounded, if we rescale g_t by a factor \eta_t so that the maximum of the sectional curvatures of h_t=\eta_t g_t is 1, we get \eta_t\to\infty as t\to t_0. This implies that the limit metric of the Ricci flow has curvature operator inside the set

\bigcap\frac{1}{\eta_t^2}F=\mathbb{R}^+\cdot I.

In other words, the limit metric has constant sectional curvature, so that the proof of the theorem of Bohm and Wilking is complete.

Of course, there are some minor details left behind this argument but it is morally correct. Nevertheless, the reader can see that the previous argument has nothing to do with a pinching family starting at the cone of 2-positive operators but it works ipsis-literis for the case of an arbitrary pinching family. In other words, the previous paragraph provides a sketch of proof of the following result:

Theorem 3 (theorem 5.1 of Bohm and Wilking). Let C(s) be a pinching family and (M,g_0) be a Riemannian manifold whose curvature operator lies inside the interior of C(0). Then, the (normalized) Ricci flow evolves g_0 to a limit metric of constant sectional curvature.

For a more complete discussion of this theorem, see the original article of Bohm and Wilking.

At this point, we end our discussion of Bohm and Wilking method of deformation of curvature conditions which are invariant by the ODE (1) (and a fortiori by the Ricci flow).

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.


%d bloggers like this: