Let $C\subset\mathcal{H}$ be a convex subset of a real Hilbert space $\mathcal{H}$ and let $\alpha,\beta$ be two nonnegative real numbers. I am asked to show that $\alpha C + \beta C = (\alpha + \beta) C$ where we have defined $\alpha C = \{\alpha x : x\in C\}$ and $C + D = \{x+y : x\in C, y\in D\}$.

It's clear to me that this property does not hold for nonconvex sets $C$, for example for the subset of $\mathbb{R}^2$ given by $C=\{(-1,-1),(1,1)\}$ it is trivial to show that $C+C \neq 2C$ by direct calculation.

So far I have tried to directly show subset inclusion in both directions.

Let $x\in (\alpha + \beta) C$. Then, $x=(\alpha + \beta)y$ for some vector $y\in C$. However, $(\alpha + \beta)y = \alpha y + \beta y \in \alpha C + \beta C\implies (\alpha + \beta)C \subset \alpha C + \beta C$.

In the other direction, I have the following, let $x\in \alpha C + \beta C$. Then, $x = \alpha y + \beta z$ with $y,z\in C$. I cannot see how convexity comes into play here? A hint is preferred to an actual answer.


Given $x\in\alpha C+\beta C$, the only points of $C$ you can immediately put your fingers on are $y$ and $z$ with $x=\alpha y+\beta z$.

Then the only additional points that convexity gives you are $ty+(1-t)z$ for $0\le t\le 1$.

The only points in $(\alpha+\beta)C$ that such a point gives you, are $(\alpha+\beta)(ty+(1-t)z)$ for $0\le t\le 1$.

Alas, you'd rather have $x$ ...

  • $\begingroup$ Ok I see now, we know that $\frac{\alpha}{\alpha+\beta}\in]0,1[$ and $\frac{\beta}{\alpha+\beta}\in]0,1[$ as well. Furthermore we know $\frac{\alpha}{\alpha+\beta} + \frac{\beta}{\alpha+\beta} =1$ so we can write $\alpha y + \beta z = (\alpha + \beta)(\frac{\alpha}{\alpha+\beta}y + \frac{\beta}{\alpha+\beta}z)\in (\alpha+\beta)C$ $\endgroup$ – Tony Oct 13 '17 at 16:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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