Find the error in this proof: \begin{lemma} \label{lem:probabilitylimits} Let…
- User
Find the error in this proof: \begin{lemma} \label{lem:probability_limits} Let be a sequence of random probabilities on given by the empirical measures of i.i.d. samples from the uniform measure. Then satisfies: \begin{itemize} \item , \end{itemize} with both of the limits in probability. \end{lemma} \begin{proof} First of all notice that, fixing an orthonormal basis of , and : \begin{equation*} \begin{split} \mathbb{E}[(N^{\frac{1}{2}-\beta}|\rho_0^N|{H^{-s}})^2]=&N^{-2\beta}\mathbb{E}\left[\sum{k} N k^{-2s}\langle \rho_0^N, e_k\rangle^2\right]\leq\ &\leq N^{-2\beta}\sum_k Nk^{-2s}\mathbb{E}\left[\left(\frac{1}{N}\sum_{i=1}^N e_k(x_i)\right)^2\right]\leq\ &\leq N^{-2\beta}\sum_k k^{-2s}\to 0, \end{split} \end{equation*} where we used Beppo Levi and the indipendence of samples. In particular by Chebychev's lemma this implies in probability. Taking and we get the first limit.
- Assistant
The primary error in the proof lies in the estimation of the expectation ( \mathbb{E}[(N^{\frac{1}{2}-\beta}|\rho_0^N|_{H^{-s}})^2] ). Specifically, the proof mistakenly omits the ( \frac{1}{N} ) factor that arises from the variance of the empirical measure's projection onto each basis function.
Let's break down where the mistake occurs:
-
Correct Calculation of the Expectation:
For each orthonormal basis function ( e_k ), the inner product ( \langle \rho_0^N, e_k \rangle ) is given by: [ \langle \rho_0^N, e_k \rangle = \frac{1}{N} \sum_{i=1}^N e_k(x_i) ] Since the ( x_i ) are i.i.d. samples from the uniform measure, the variance of this inner product is: [ \text{Var}\left(\langle \rho_0^N, e_k \rangle\right) = \frac{1}{N} \text{Var}(e_k(x_i)) = \frac{1}{N} ] Therefore, the expectation is: [ \mathbb{E}\left[\langle \rho_0^N, e_k \rangle^2\right] = \frac{1}{N} ]
-
Expectation of the ( H^{-s} ) Norm:
Using the correct variance, the expectation becomes: [ \mathbb{E}\left[(N^{\frac{1}{2}-\beta}|\rho_0^N|_{H^{-s}})^2\right] = N^{-2\beta} \sum_k k^{-2s} \mathbb{E}\left[\langle \rho_0^N, e_k \rangle^2\right] = N^{-2\beta} \sum_k \frac{k^{-2s}}{N} = N^{-2\beta -1} \sum_k k^{-2s} ]
-
Error in the Proof:
The proof incorrectly estimates the expectation as: [ \mathbb{E}\left[(N^{\frac{1}{2}-\beta}|\rho_0^N|_{H^{-s}})^2\right] \leq N^{-2\beta} \sum_k k^{-2s} ] This upper bound ignores the additional ( \frac{1}{N} ) factor, leading to an incorrect conclusion.
-
Implications of the Error:
While the final conclusion ( N^{\frac{1}{4}}|\rho_0^N|_{H^{-1}} \to 0 ) in probability remains correct when ( \beta = \frac{1}{4} ) and ( s = 1 ), the intermediary steps in the proof are flawed due to the oversight of the ( \frac{1}{N} ) term.
Corrected Argument:
To fix the proof, one should acknowledge the ( \frac{1}{N} ) factor:
[ \mathbb{E}\left[(N^{\frac{1}{2}-\beta}|\rho_0^N|_{H^{-s}})^2\right] = N^{-2\beta -1} \sum_k k^{-2s} \to 0 \quad \text{as } N \to \infty ]
This properly shows that ( N^{\frac{1}{2}-\beta}|\rho_0^N|_{H^{-s}} \to 0 ) in probability, ensuring the lemma holds as stated.
Conclusion:
The proof erroneously omits the ( \frac{1}{N} ) factor when estimating the expectation of the squared ( H^{-s} ) norm, leading to an incorrect intermediate bound.
Answer:
After expanding, the expectation should include a 1⁄N factor, but the proof mistakenly drops this term. It incorrectly bounds E[(N¹⁄²–β‖ρ₀ᴺ‖_{H⁻ˢ})²] by N^–2β Σk^–2s instead of N^–2β–1 Σk^–2s. Thus, the proof erroneously omits the 1/N in the expectation calculation.
-