Does the sign of curvature remain constant under the sum of their metrics?

265 Views Asked by At

Does the sign of curvature remain constant under the sum of their metrics over a fixed manifold? e.g. If $(M,g_1)$ and $(M,g_2)$ are closed Riemannian manifolds of positive curvature then $(M,g_1+g_2)$ is also of positive curvature?

Update: Q2: What if we weaken the problem by allowing manifolds having boundary?

1

There are 1 best solutions below

23
On BEST ANSWER

Edit As C.F.G. noted, I had originally answered different but related question. So, let me answer the actual question asked with a No: the sum of two positively curved metrics can have negative curvature. I'll construct an example below.

To do so, consider $S^3$ with a metric $g_\lambda$ making it a Berger Sphere. Recall that this is obtained by beginning with the usual round metric $g_0$ and scaling it in the direction of the Hopf fibers. More precisely, given any vectors $v,w\in T_p S^3$, we define $g_\lambda(v,w) = g_0(v,w) + \lambda g_0(v',w')$ where $v'$ is the $g_0$-projection of $v$ to the tangent space of the Hopf circle through $p$, and likewise for $w'$.

It is known that the resulting metric $g_\lambda$ is positive curved iff $-1<\lambda < 1/3$. (If $\lambda \leq -1$, then $g_\lambda(v,v) \leq 0$ for all $v$ tangent to a Hopf circle, so we don't even get a metric in this case).

Pick any $\lambda$ strictly between $0$ and $1/3$.

Now, consider the metric $g:=g_{\lambda} + t g_0$ on $S^3$. For any $t > 0$, this is a sum of positively curved metrics. However, I claim that for any $t$ with $0< t < 3 \lambda -1$, that $g$ has negative curvature planes.

To see this, note that \begin{align*} g(v,w) &= g_\lambda(v,w) + t g_0(v,w)\\ &= (1+t) g_0(v,w) + \lambda g_0(v',w'),\end{align*} so $$\frac{1}{1+t} g(v,w) = g_0(v,w) + \frac{\lambda}{1+t} g_0(v',w')$$ is a Berger metric $g_{\lambda/(1+t)}$. Since scaling doesn't change the sign of curvature, it follows that if we can show that $\lambda/(1+t) > \frac{1}{3}$, we are done. Since $t< 3 \lambda - 1$, we find $\lambda/(1+t) > \lambda/(1 + 3\lambda - 1) = \frac{1}{3}.$

End edit - the old answer is left below

Proposition: There is a positive curved metric $g_b$

No, the sum of metrics of fixed signed curvature is, in general, not of the same sign.

Proposition: If the sum of any two positively curved metrics is positively curved, then every closed manifold of dimension $2$ or more admits a metric of positive curvature.

Proof: First, if $g$ is a metric with positive curvature, then $\lambda g$ is for any positive $\lambda$. Also, an easy induction argument shows that the hypothesis imply that any finite sum of positively curved metrics is positively curved.

Now, given any closed manifold $M^n$, compactness allows us to cover $M$ by a finite collection of open sets $U_1,..., U_k$, each diffeomorphic to $\mathbb{R}^n$ with the property that their closures $\overline{U_i}$ are contained in open sets $V_i$ which are also diffeomorphic to $\mathbb{R}^n$. Let $\lambda_i$ be a partition of unity supported on $\overline{U}_i$.

Since $n\geq 2$, each $V_i$ admits a metric $g_i$ which is positively curved. For example view $\mathbb{R}^n$ as the complement of a point in $S^n$ then pull the induced metric back to $g_i$.

Now the metic $g = \sum \lambda_i g_i|_{U_i}$ is positively curved by the first paragraph. $\square$

Of course, there are obstructions to admitting a positively curved metric. So, a consequence of this is that the sum of positively curved metrics need not be positively curved.