You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
$n$ is the total number of bins being divided into (positive integer)
$a$ is the number of space groups (positive integer)
$z$ is proportional to the number of standard deviations captured on the normal distribution (practically, will control how aggressively we will select for high space groups, the larger the magnitude of $z$, the more aggressive the selection)
Note:$\Chi(x)$ is a modified form of $\Phi(x)$, the CDF of a standard normal distribution. The $\frac{1}{\sqrt{2\pi}}$ term is divided out and as such does not need to be computed.
In the non-edge-biased case
Let's say that we have a specific space group that we want to bias towards. We can then use something like this:
Let $p$ and $q$ be the lower and upper bounds of the desired range to bias towards (so the mean $\mu$ of the distribution is centered between them). Define a new inverse distribution $\Psi$:
$$\Psi(a,b) = b - a + \int_b^a{\exp\left(-\frac{t^2}{2}\right)dt}$$
Opposite to a bell curve, this density plot is 0 at input 0 and approaches 1 at large magnitude input.
We'll also define some useful terms that represent the z-values of space group 0 and space group $a$, $0^$ and $a^$ respectively:
We then have:
$$B_i = \text{round}\left[a\left(\frac{\Psi(0^, \frac{i(a^ - 0^)}{n}+0^)}{\Psi(0^, a^)}\right)\right]$$
where $a$, $i$, and $n$ mean the same the same thing, and $z$ this time refers to how far apart $p$ and $q$ are on the distribution (when $z = 1$, $p$ and $q$ are 1 standard deviation apart). Similar to before, the higher $z$ is, the more aggressively the distribution will be biased towards the range we are selecting for. Because the distribution density function is flipped, the value of $z$ is replaced by its multipliciative inverse to keep its behavior consistent. (So $z = 2$ means $p$ and $q$ are 0.5 std apart).