We consider subspaces of $\mathbb{R}^N$. Suppose that we have a property called $\mbox{Prop}$ that apply to subspaces of $\mathbb{R}^N$. That is to say a function from the set of subspaces of $\mathbb{R}^N$ into $\{0,1\}$.

The variables $X_1,\dots,X_N$ are gaussian taking their values in $\mathbb{R}^N$. They are supposed i.i.d. Deriving from an isotropic distribution (The covariance matrix is identity).

Prove that ($0\leqslant n\leqslant N$) $$ P(\mbox{Prop}(Span(X_1,\dots,X_n))=1)= P(\mbox{Prop}(Span(X_1,\dots,X_{N-n})^\bot)=1) $$ Without referring to the definition of $\mbox{Prop}$ other than the obvious that is $$ \mbox{Prop}(Span(X_1,\dots,X_k)) $$ is a measurable function on $\Omega$, the sample set.

In other words, the probability for some property to hold on a subspace spanned by $n$ variables chosen at random (gaussian isotropic) is equal to the probability for it to hold on the orthogonal space of the space spanned by $N-n$ variables.

This may translate to this "funny" question: $X_1,\dots,X_6$ are six real-valued gaussian iid variables. Prove that $$ P\left(\sin\left(\frac{|X_1|}{|X_1|+|X_2|+|X_3|}\right)>0.2\right)= P\left(\sin\left(\frac{|Y_1|}{|Y_1|+|Y_2|+|Y_3|}\right)>0.2\right) $$

where $Y_1=X_2X_6-X_3X_5$, $Y_2=-X_1X_6+X_3X_4$ and $Y_3=X_1X_5-X_2X_4$. (The vector $Y$ (in $\mathbb{R}^3$) is orthogonal to the subspace spanned by $(X_1,X_2,X_3)$ and $(X_4,X_5,X_6)$).

Thank you for your help, Saïd.


The Grassmannian $\mathrm{Gr}(n,N)$ is a compact manifold with a well defined canonical probability measure, its "uniform measure", characterized by the fact that it is invariant by the action of the orthogonal group $O(N)$. Starting from those $(X_i)_ i$, both $\mathrm{Span}(X_1,\dots,X_n)$ and $\mathrm{Span}(X_1,\dots,X_{N-n})^{\perp}$ define a uniformly distributed random element of $\mathrm{Gr}(n,N)$ (just because their probability law is $O(N)$-invariant).


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.