This question is inspired by On moments of inertia of planar and 3D convex bodies. Let $f:{\mathbb R}^3\setminus\{0\}\to{\mathbb R}$ be an even homogeneous ($f(kx)=f(x)$ for all real $k\neq 0$) continuous function. Can one always find an orthogonal basis $(x_1,x_2,x_3)$ such that $f(x_1)=f(x_2)=f(x_3)$ ?

Remark. Even without the condition that $f$ is even, I do not know a counterexample. This condition comes from the application that I mention in the first sentence.

Remark 2. Thanks to @Aleksei Kulikov, who found the reference: this was a problem posed by Rademacher and solved by Kakutani in 1942 (Annals of math., 43 (1942) 739-741). Most interestingly, the reviewer in MR0007267 wrote that the corresponding problem in $R^n, n\geq 4$ was unsolved in 1942. I found that the $n$-diensonal generalization was obtained by H. Yamabe and Z. Yujobô, Osaka Math. J. 2 (1950), 19–22.

  • $\begingroup$ Since $f$ is even, it's tempting to identify antipodal points. For functions on $S^1$, this shows that the corresponding statement is correct and in fact the same as the (trivial, in $d=1$) Borsuk-Ulam theorem. In your actual setting, for $d=2$, we obtain the projective plane instead of the sphere, and the orthogonality condition does not seem to have a good description there. $\endgroup$ Dec 4, 2022 at 18:19
  • 2
    $\begingroup$ Is there some easy counterexample if we don't require $f$ to be even? $\endgroup$
    – Saúl RM
    Dec 4, 2022 at 19:19
  • $\begingroup$ @Saul RM: no even if one drops the evenness condition, I do not know a counterexample. Evenness condition come from the application that I had in mind (see the link in the question). $\endgroup$ Dec 4, 2022 at 20:39
  • 1
    $\begingroup$ @Christian Remling: you are right; this is about continuous functions on the real projective plane. Standard spherical Riemannian metric (in which the diameter of the sphere is $\pi$) descends to the projective plane, and the diameter becomes $\pi/2$. So the question is: can you find three "diametrally opposite" points on the projective plane, such that... $\endgroup$ Dec 4, 2022 at 20:54
  • 5
    $\begingroup$ This is actually a famous result of Kakutani (A Proof That there Exists a Circumscribing Cube Around Any Bounded Closed Convex Set in $\mathbb{R}^3$) and the proof is more or less the same as in the accepted answer. $\endgroup$ Dec 5, 2022 at 14:37

1 Answer 1


It seems there aren't any counterexamples, even if $f$ is homogeneous but not even.

If there is some counterexample $f$ to the question, and letting $X=\mathbb{R}^3\setminus\{x_1=x_2=x_3\}$, we can consider the map $F:SO(3)\to X$, given by $F(u,v,w)=(f(u),f(v),f(w))$ (we will sometimes represent elements $r\in SO(3)$ by $(r(e_1),r(e_2),r(e_3))$, $(e_i)_{i=1}^3$ being the usual basis of $\mathbb{R}^3$).

As $\pi_1(SO(3))\cong\mathbb{Z}_2$ and $\pi_1(X)\cong\mathbb{Z}$, we will have $F_*(\gamma)=0\;\forall\gamma\in\pi_1(SO(3))$.

Now consider the loop $\gamma:t\mapsto\gamma_t=r_{2\pi t}(v)$ given by rotations of angle $2\pi t$ around $v:=(1,1,1)$. Note that $\gamma_{\frac{1}{3}}(e_1)=e_2,\gamma_{\frac{1}{3}}(e_2)=e_3$ and $\gamma_{\frac{1}{3}}(e_3)=e_1$. So if $\gamma_t=(u,v,w)$, then $\gamma_{t+\frac{1}{3}}=(\gamma_{t+\frac{1}{3}}(e_1),\gamma_{t+\frac{1}{3}}(e_2),\gamma_{t+\frac{1}{3}}(e_3))=(\gamma_{t}(e_2),\gamma_{t}(e_3),\gamma_{t}(e_1))=(v,w,u)$.

So the path $\alpha=F_*\gamma\in\pi_1(X)$ satisfies that if $\alpha(t)=(x_1,x_2,x_3)$, then $\alpha(t+\frac{1}{3})=(x_2,x_3,x_1)$. From this we will deduce that $\alpha$ is not trivial, which is a contradiction. To deduce that $\alpha$ is not trivial, we can first deformation retract $X$ to a circumference perpendicular to the line $\{x_1=x_2=x_3\}$, so that identifying this circumference with $\mathbb{S}^1$, the rotation $(x_1,x_2,x_3)\mapsto(x_2,x_3,x_1)$ becomes the rotation $r_{\frac{2\pi}{3}}$ of $\mathbb{S}^1$.

So we have to prove that any loop $\beta:[0,1]\to\mathbb{S}^1$ satisfying $\beta(t+\frac{1}{3})=e^{\frac{2\pi i}{3}}\beta(t)$ is not trivial. This is true because as $\beta(\frac{1}{3})=e^{\frac{2\pi i}{3}}$, the path $\beta|_{[0,\frac{1}{3}]}$ rotates a total angle of $2\pi(n+\frac{1}{3})$ for some $n\in\mathbb{Z}$, so the winding number of $\beta$ is $3(n+\frac{1}{3})\neq0$.


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.