In dimension $4$, it is known that the curvature operator $\operatorname{Rm} : \Lambda^2(M) \to \Lambda^2(M)$ admits a block decomposition of the form $$\operatorname{Rm} = \begin{pmatrix} A & B \\ B^T & C\end{pmatrix}$$

Choosing convenient orthonormal basis, the matrices above can be written in diagonal form (i.e $A_{ij} = a_i \delta_{ij}$, $B_{ij} = b_i \delta_{ij}$, $C_{ij} = c_i \delta_{ij}$).

In his "Four-manifolds with positive curvature operator" paper, Hamilton proved the following estimates:

If we choose successively positive constants $G$ large enough, $H$ large enough, $\delta$ small enough, $J$ large enough, $\varepsilon$ small enough, $K$ large enough, $\theta$ small enough, and $L$ large enough, with each depending on those chosen before, then the closed convex subset $X$ of $\{M_{\alpha \beta} \geq 0 \}$ defined by the inequalities

  • $(b_2 + b_3)^2 \leq G a_1 c_1$
  • $a_3 \leq H a_1$ and $c_3 \leq H c_1$
  • $(b_2 + b_3)^{2 + \delta} \leq J a_1 c_1 (a - 2b + c)^{\delta}$
  • $(b_2 + b_3)^{2 + \varepsilon} \leq K a_1 c_1$
  • $a_3 \leq a_1 + L a_1^{1 - \theta}$ and $c_3 \leq c_1 + L c_1^{1 - \theta}$

is preserved under the Ricci flow for a suitable ODE.

For further context, one can see page $165$ of this book, where it is then claimed without any proof that an immediate consequence of the above estimates is the following estimate on the traceless Riemannian curvature operator:

$$\| \mathring{\mathrm{Rm}} \| \leq \varepsilon R + C_{\varepsilon}$$

where $\varepsilon$ can be arbitrarily small and $C_{\varepsilon} < \infty$ is a constant. How does this estimate follow from the previous ones? I've tried to express $\| \mathring{\mathrm{Rm}} \|$ in terms of the $a_i$'s, $b_i$'s and $c_i$'s in order to use the estimates proved for them, but that led me nowhere. How can one conclude this estimate from the previous ones? I can't fill this hole in the paper. I'd appreciate any help!

  • $\begingroup$ Could you clarify what book you are referring to? When I click on the link, I get an article that starts on page 167. $\endgroup$
    – Deane Yang
    Jun 24, 2022 at 2:12
  • 1
    $\begingroup$ @DeaneYang I guess it was a mistake on my part to call it a book. The link is correct. The page I referred to is page $165$ of the PDF and page $329$ of the article (where the first estimates I cited were proved, which the authors then claim in the corollary $5.2.7$ (located at the end of page $168/332$) implies the estimate I'm having trouble with $\endgroup$ Jun 24, 2022 at 2:17

1 Answer 1


After a lot more time thinking about it, I think I've figured it out. Let $\Gamma$ be a constant such that $ \Gamma \|g \odot g \| = 1$ (where $\odot$ denotes the Kulkarni Nomizu product, and the only reason I don't make $\Gamma$ explicit here is because it depends on the conventions for the definitions of $g \odot g$ and the inner product, but they all agree up to a factor). Clearly, we can control $\| \mathrm{Rm} \|^2$ once we control $A, B, C$, i.e

$$\| \mathrm{Rm} \|^2 \leq\beta( \|A \|^2 + 2\|B \|^2 + \|C \|^2)$$

where $\beta$ is some constant. For convenience, I'll commit the (harmless) abuse of denoting both the operators on $\Lambda^{2}_{\pm}(M)$ and its associated $(0, 4)$ tensors by the same letters. Recalling that $R = \mathrm{tr}(A) = \mathrm{tr}(C)$, by a very similar argument to the one seen in the end of Chapter 9 of Peter Topping's lectures on Ricci flow , we see that the estimates provided by Hamilton imply that for all $\varepsilon > 0$ (however small) there exists a constant $C_{\varepsilon}$ such that $$2 \max\left\{\left\|A - \frac{2}{3}\Gamma R (g \odot g)\right\| , \left\|C - \frac{2}{3} \Gamma R (g \odot g)\right\| \right\} \leq \varepsilon R+C_{\varepsilon}$$ The norm of $B$ is controlled by $b_3^2$ which is controlled by $(b_3 + b_2)^{2 + \delta}$ which in turn is controlled by $K a_1 c_1$. The latter is controlled by the estimate above. Putting all this together we get almost exactly the desired following inequality (it's worth noting this may not exactly be $\mathring{\mathrm{Rm}}$ and some undesirable $\varepsilon^2 R^2$ terms may appear, but since we'll just use the fact that $\varepsilon$ can be made arbitrarily small in the end, this will not matter).

$$\left\|\mathrm{Rm} - \frac{2}{3}\Gamma R (g \odot g)\right\| \leq \varepsilon R+C_{\varepsilon}$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.