Discrepancy in ACSF radial function type

[mamunm]

Hi, Thanks for developing and maintaining dscribe.

According to the RuNNer documentation, the definition of g3 and g4 are the opposite of what is implemented in dscribe.

For example, the cpp code for g3 and g4 computation in dscribe is shown below:

But in the RuNNer document, g3 and g4 functions are defined like below:

Even though they are defined arbitrarily to capture different parts of the atomic environment, using the same notion as the original implementation would still be beneficial.