Exercise 5.5 from Capinski's and Kopp's book "Measure, Integral and Probability" asks to show that it is impossible to define an inner product on the space $L^1([0,1])$. In order to get this result we need to show that parallelogram law

$$\Vert f+g \Vert^2 + \Vert f-g \Vert^2 = 2( \Vert f\Vert^2 + \Vert g \Vert^2 ) $$

does not hold for $\Vert \cdot \Vert_1$ where $$\Vert f \Vert_1=\int_0^1|f(x)|\mathsf dx.$$ The hint in the book suggests to consider two following functions:

\begin{align} f(x)&=\frac12-x\\ g(x)&=x-\frac12. \end{align}

However, $g(x)=-f(x)$ and thus $\Vert g \Vert_1 = \Vert f \Vert_1 $ . Moreover, $\Vert f+g \Vert_1 = 0$ and $$\Vert f-g \Vert_1 = \Vert 2f \Vert_1 = 2\Vert f \Vert_1.$$

Finally I get:

\begin{align} \Vert f+g \Vert_1^2 + \Vert f-g \Vert_1^2 &= 4\Vert f \Vert_1^2\\ &= 2(2\Vert f \Vert_1^2)\\&= 2( \Vert f\Vert_1^2 + \Vert g \Vert_1^2 ) \end{align}

which in turn shows that actually Parallelogram law holds.

What did I miss?

May it be related to the fact that $g$ is a scaled version of $f$?


There is nothing wrong about your argumentation.

Here is a "real" counterexample: Set

$$f(x) := 1_{[0,1/2]}(x) \qquad \text{and} \qquad g(x) := 1_{[1/2,1]}(x).$$


$$\|f\|_1 = \|g\|_1 = \frac{1}{2}$$

and therefore

$$2 (\|f\|_1^2+ \|g\|_1^2) = 1.$$

On the other hand, it is not difficult to see that $$\|f+g\|_1 = \|f-g\|_1 = 1.$$


$$\|f+g\|_1^2 + \|f-g\|_1^2 = 2 \neq 1 = 2 (\|f\|_1^2+ \|g\|_1^2).$$

  • $\begingroup$ thanks. It is obvious that that parallelogram law holds for your $f$ and $g$. Could you comment on the case when $g=-f$, because I thought that the law should NOT hold for all functions in $L^1([0,1])$. $\endgroup$ – Ruth90 Jul 30 '15 at 15:28
  • $\begingroup$ @Ruth90 Actually, the parallelogram law does not hold for the $f$ and $g$ from my answer. You wanted to find $f$,$g$ such that the parallelogram law doesn't hold, right? If $g=-f$, then the parallelogram law does hold true; that's exactly what you proved. $\endgroup$ – saz Jul 30 '15 at 15:31
  • $\begingroup$ sorry you are right - in your case the law does NOT hold, which is fine since, if I am correct, the law holds just in $L^2$ space (I am working with $L^p$ spaces). My question would be why does parallelogram does hold in the case of $g=-f$? Which in turn means that the $\Vert ^. \Vert_1$ can be induced by inner product. $\endgroup$ – Ruth90 Jul 30 '15 at 15:35
  • $\begingroup$ @Ruth90 Well, why not? That the parallelogram law doesn't hold true in $L^1$ doesn't mean that the identity fails for all $f,g \in L^1$; it simply means that there are $f,g$ such that the identity doesn't hold. $\endgroup$ – saz Jul 30 '15 at 15:39
  • $\begingroup$ but if parallelogram law holds then the norm can be induced by inner product and thus the space being considered is also the inner product space. In my case $L^1$ IS the inner product space and in yours - it is not. $\endgroup$ – Ruth90 Jul 30 '15 at 15:44

By linearity one has: $$f=g:\quad\left(\int|f+g|\mathrm{d}\mu\right)^2+\left(\int|f-g|\mathrm{d}\mu\right)^2=2\left(\int|f|\mathrm{d}\mu\right)^2+2\left(\int|g|\mathrm{d}\mu\right)^2$$ $$\mathrm{supp}f\cap\mathrm{supp}g=\varnothing:\quad\int|f\pm g|\mathrm{d}\lambda=\int|f|\mathrm{d}\lambda+\int|g|\mathrm{d}\lambda$$

So shift a block: $$f:=\chi_{[0,a]}:\quad f_\varepsilon(x):=f(x-\varepsilon)$$

Then one obtains: $$\|f+f_\varepsilon\|_1^2+\|f-f_\varepsilon\|_1^2\neq2\|f\|_1^2+2\|f_\varepsilon\|_1^2$$

See the thread: Parallelogram Law


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.