I have $n$ successive observation $A_\mu $ of a quantity $A$ and I need to understand how the expectation values of the square of the statistical error depends from the autocorrelation time but a single step in the demonstration is giving me trouble. the author of the book (Monte Carlo simulations in statistical physics -Binder) wrote :

\begin{align*} \langle {(\delta A)^2} \rangle &= \bigg\langle \bigg[\frac{1}{n} \sum_{\mu =1}^n (A_\mu - \langle A \rangle) \bigg]^2 \bigg\rangle \\ &= \frac{1}{n^2} \sum_{\mu=1}^n \langle(A_\mu -\langle A \rangle)^2 \rangle + \frac{2}{n^2}\sum_{ \mu_1= 1}^n \sum_{\mu_2 = \mu_1 +1}^n \bigg(\langle A_{\mu_1} A_{\mu_2} \rangle- \langle A\rangle \bigg) \end{align*}

changing the summation index $\mu_2 $ to $\mu_1 +\mu $ with $ \mu = \mu_2 - \mu_1$this equation can be rewritten as :

$$ \langle {(\delta A)^2} \rangle = \frac{1}{n} \bigg[\langle A^2\rangle - \langle A \rangle ^2 + 2 \sum_{\mu =1}^n \bigg( 1-\frac{\mu}{n} \bigg) \bigg(\langle A_0 A_\mu \rangle - \langle A \rangle^2 \bigg) \bigg] $$

where <> mean the expectation value.

how does this change in the summation index works? i originally posted this question on mathematics.stackexchange but users there advised me to try here.


Stationarity (time origin invariance) of the correlation function means that $$ \langle A_{\mu_1} A_{\mu_2}\rangle = \langle A_{0} A_{\mu}\rangle $$ and the double sum of interest may be expressed as $$ \sum_{\mu_1=1}^n \sum_{\mu=1}^{n-\mu_1} \left( \langle A_{0} A_{\mu}\rangle - \langle A\rangle^2 \right) . $$ By the way, I think there's a typo in your question, the very last term should be $\langle A\rangle^2$ not $\langle A\rangle$, so that's what I'm writing.

Now the summand is independent of $\mu_1$, and the idea is to exchange the order of the two sums. This can be illustrated by drawing a 2D grid, with $\mu$ and $\mu_1$ as the axes, and shading in the relevant terms. For any given value of $\mu$, the sum over $\mu_1$ ranges from $1$ to $n-\mu$. $$ \sum_{\mu_1=1}^n \sum_{\mu=1}^{n-\mu_1} = \sum_{\mu=1}^{n}\sum_{\mu_1=1}^{n-\mu} $$ But since the summand is independent of $\mu_1$, there are $(n-\mu)$ identical terms in the inner sum and we can do it immediately. We are left with a single sum, over $\mu$, and a factor $(n-\mu)$ inside. In your answer, a factor $1/n$ is taken inside too, producing the overall $(1-\frac{\mu}{n})$ factor.

| cite | improve this answer | |

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.