Let $X$ be a centered continuous stochastic process which is square integrable on $[0,2]\times \Omega$ and the basis of $L^2(0,2)$ is $\{e_i\}$. By using Karhunen-Leove Theorem one can write for all $t\in[0,2]$, $$X_t=\sum_{i\geq 1}\left(\int_0^2X_te_i(t)dt\right) e_i(t), in\; L^2(\Omega).$$

Let us define another stochastic process $Y$ on $[0,1]\times \Omega'$, which is defined as $$Y_t=X_t+X_{t+1},\; \; \forall t\in [0,1].$$

Then using the above expansion of $X$, we can easily write for all $t\in[0,1], $$$Y_t=\sum_{i\geq 1}\left(\int_0^2X_te_i(t)dt\right) \left(e_i(t)+e_i(t+1)\right).$$

This expression have a similar type of expression as that of Karhunen-Leove expansion of $Y$. But to show that, this is actually K-L expansion, we need to show that $\left\{e_i(t)+e_i(t+1)\right\}$ is a basis for $L^2[0,1]$.

Here is I am getting stuck: Could anyone please guide me: how to check whether $\left\{e_i(t)+e_i(t+1)\right\}$ is a basis for $L^2[0,1] or not. Thanks.


1 Answer 1


Consider the Fourier basis of $L^2(0,2)$: $a_n(t)=cos(n\pi t/2)$ ($n\in \mathbb N$) and $b_n(t)=sin(n\pi t/2)$ ($n\in \mathbb N, n\geq 1$).

If you compute $a_n(t)+a_n(t+1)$ and $b_n(t)+b_n(t+1)$, depending on whether $n$ is even or odd, some of them vanish. So no, they're not a basis.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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