Commit e83a5d18 authored by Michael Wimmer's avatar Michael Wimmer
Browse files

shorter text in math

parent fbf2fa89
Pipeline #58156 passed with stages
in 1 minute and 41 seconds
......@@ -81,17 +81,17 @@ In this way we can make sure to approximately focus on the physically relevant c
Doing this requires us to rewrite the integral as
$$\begin{eqnarray}
\int p_\text{real}(R) A(R) dR = &
\int p_\text{sampling}(R) \underbrace{\frac{p_\text{real}(R)}{p_\text{sampling}(R)}}_{=w(R)} A(R) dR\\
= & \int p_\text{sampling}(R) w(R) A(R) dR\\
\int p_\text{sample}(R) \underbrace{\frac{p_\text{real}(R)}{p_\text{sample}(R)}}_{=w(R)} A(R) dR\\
= & \int p_\text{sample}(R) w(R) A(R) dR\\
\approx & \frac{1}{N} \sum_{i=1}^N w(R_i) A(R_i) \tag{3}
\end{eqnarray}$$
where the configurations $R_i$ are now sampled from $p_\text{sampling}(R)$. When using this approximate probability distribution
where the configurations $R_i$ are now sampled from $p_\text{sample}(R)$. When using this approximate probability distribution
we thus have to introduce *weights* $w(R)$ into the average.
### Why approximate importance sampling eventually fails
Approximate importance sampling is an attractive way of sampling: if we have a convenient and computationally efficient
$p_\text{sampling}$ we can apply the Monte Carlo integration and seemingly focus on the relevant part of the configuration space.
$p_\text{sample}$ we can apply the Monte Carlo integration and seemingly focus on the relevant part of the configuration space.
Unfortunatley, this approach becomes increasingly worse as the dimension of the configuration space increases. This is related to the
the very conter-intuitive fact thatin high-dimensional space "all the volume is near the surface". This defies our intuition that
......@@ -115,10 +115,10 @@ This effect directly shows in the weights. Let us demonstrate this using a simpl
where
$$p_\text{real}(x_1, \dots, x_d) = (2 \pi \sigma_\text{real})^{d/2} e^{-\frac{\sum_{k=1}^d x_k^2}{2\sigma_\text{real}}}$$
is a normal distribution with standard deviation $\sigma_\text{real}$. For the sampling distribution we use
also a normal distribution, but with a slightly differen standard deviation $\sigma_\text{sampling}$:
$$p_\text{sampling}(x_1, \dots, x_d) = (2 \pi \sigma_\text{sampling})^{d/2} e^{-\frac{\sum_{k=1}^d x_k^2}{2\sigma_\text{sampling}}}\,.$$
We will now compute how the weights $p_\text{real}/p_\text{sampling}$ are distributed for different dimensionality
$d$. In the example below we have chosen $\sigma_\text{real} = 1$ and $\sigma_\text{sampling} = 0.9$ and sampling over $N=10000$
also a normal distribution, but with a slightly differen standard deviation $\sigma_\text{sample}$:
$$p_\text{sample}(x_1, \dots, x_d) = (2 \pi \sigma_\text{sample})^{d/2} e^{-\frac{\sum_{k=1}^d x_k^2}{2\sigma_\text{sample}}}\,.$$
We will now compute how the weights $p_\text{real}/p_\text{sample}$ are distributed for different dimensionality
$d$. In the example below we have chosen $\sigma_\text{real} = 1$ and $\sigma_\text{sample} = 0.9$ and sampling over $N=10000$
configurations:
![Vanishing weights in high-dimensional space](figures/weights.svg)
We see that as dimensionality increses, the distribution of weights gets more and more skewed towards 0. For a large dimensionality,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment