Shi's Global Derivative Estimates

Shi's Derivative Estimates (1989)

A fundamental difficulty in Ricci flow is that pointwise curvature bounds do not automatically control the derivatives of curvature. Shi's estimates resolve this by showing that an initial curvature bound propagates to bounds on all covariant derivatives, with a time-dependent constant that blows up only as $t \to 0$.

Theorem (Shi, 1989)

Let $(M^n, g(t))$ be a complete solution of Ricci flow on $[0, T]$ with

$$|\mathrm{Rm}| \leq K \quad \text{on } M \times [0, T].$$

Then for each $k \geq 1$ there exists a constant $C(n, k)$ depending only on the dimension $n$ and the order $k$ such that

$$\boxed{|\nabla^k \mathrm{Rm}|^2 \leq \frac{C(n,k)\, K^2}{t^k}}$$

for all $t \in (0, T]$.

The factor $t^{-k}$ in the denominator reflects the parabolic scaling of Ricci flow: derivatives of order $k$ in space correspond to $k/2$ derivatives in time. Note that as $t \to 0^+$ the estimate degenerates, which is expected since the initial metric may have no derivative bounds.

Proof for k = 1: The Key Case

The proof for $k = 1$ illustrates the essential idea. We construct a barrier function that combines $|\nabla \mathrm{Rm}|^2$ with $|\mathrm{Rm}|^2$, weighted by a factor of $t$ to handle the initial degeneracy.

Step 1: Define the Barrier Function

Set

$$F = t\,|\nabla \mathrm{Rm}|^2 + A\,|\mathrm{Rm}|^2$$

where $A > 0$ is a constant to be determined. The factor of $t$ ensures that $F \big|_{t=0} = A\,|\mathrm{Rm}|^2 \leq A K^2$, giving a controlled initial value.

Step 2: Evolution Equations

Under Ricci flow, the evolution of $|\mathrm{Rm}|^2$ satisfies

$$\frac{\partial}{\partial t}|\mathrm{Rm}|^2 \leq \Delta|\mathrm{Rm}|^2 - 2|\nabla \mathrm{Rm}|^2 + C_0\,|\mathrm{Rm}|^3$$

and the evolution of $|\nabla \mathrm{Rm}|^2$ satisfies

$$\frac{\partial}{\partial t}|\nabla \mathrm{Rm}|^2 \leq \Delta|\nabla \mathrm{Rm}|^2 - 2|\nabla^2 \mathrm{Rm}|^2 + C_1\,|\mathrm{Rm}|\,|\nabla \mathrm{Rm}|^2$$

where $C_0$ and $C_1$ are dimensional constants.

Step 3: Evolution of F

Combining the two evolution equations, we compute

$$\frac{\partial F}{\partial t} = |\nabla \mathrm{Rm}|^2 + t\,\frac{\partial}{\partial t}|\nabla \mathrm{Rm}|^2 + A\,\frac{\partial}{\partial t}|\mathrm{Rm}|^2$$

Using the bounds above and $|\mathrm{Rm}| \leq K$:

$$\frac{\partial F}{\partial t} \leq \Delta F + (1 + C_1 K t - 2A)\,|\nabla \mathrm{Rm}|^2 + C_0 A\, K^3$$

Step 4: Choose A and Apply Maximum Principle

Choose

$$A = C_1 K T + 1.$$

Then for all $t \in [0, T]$ we have $1 + C_1 K t - 2A \leq -1 < 0$, so the coefficient of $|\nabla \mathrm{Rm}|^2$ is negative. The evolution inequality becomes

$$\frac{\partial F}{\partial t} \leq \Delta F + C_0 A\, K^3.$$

By the scalar maximum principle on a complete manifold:

$$F(t) \leq F(0) + C_0 A K^3 t \leq A K^2 + C_0 A K^3 T.$$

Since $F = t\,|\nabla \mathrm{Rm}|^2 + A\,|\mathrm{Rm}|^2 \geq t\,|\nabla \mathrm{Rm}|^2$, we conclude

$$|\nabla \mathrm{Rm}|^2 \leq \frac{C(n)\, K^2}{t}$$

which is Shi's estimate for $k = 1$.

Induction for General k

The general case proceeds by induction on $k$. Assuming the estimate holds for all orders up to $k - 1$, one defines the barrier function

$$F_k = t^k\,|\nabla^k \mathrm{Rm}|^2 + B_k\, t^{k-1}\,|\nabla^{k-1} \mathrm{Rm}|^2$$

where $B_k$ is chosen large enough so that the cross terms in the evolution of $F_k$are absorbed. The inductive hypothesis provides control of $|\nabla^{k-1} \mathrm{Rm}|^2$, while the factor $t^k$ ensures the correct initial data. The same maximum principle argument then yields

$$|\nabla^k \mathrm{Rm}|^2 \leq \frac{C(n,k)\, K^2}{t^k}.$$

The key structural feature is that each step of the induction uses a barrier that couples order $k$ with order $k - 1$, and the power of $t$ increases by one at each stage, matching the parabolic scaling.

← Evolution Equations Next: Hamilton-Ivey Pinching →