Closed yanze039 closed 11 months ago
Thanks for reporting this. For u_nk, it is usually recommended to use dE
for decorrelation. I wonder why do you want to use all
?
I think in some sense, this is a more statistics problem than a coding problem. How do one compute the self-correlation for a timeseries including NaN.
Ohh, yeah, I didn't notice that I used method='all'
. After changing to dE
, it looks good now.
Can I generally ask why dE
is better?
Thanks!
Let's say for this specific example, the lambda=1 is infinite, so if you sum all the lambdas up, you would get an infinite number. So for this timeseries, you would just get a long timeseries of infinite number.
@Dead-fisher Are you happy with the solution that I provided so I could close this ticket?
Yeah, absolutely. Thanks!
Hi,
I am processing output files from
Amber22
TI simulations with ACES functionality (interaction scaling + HREMD). My trajectory at $\lambda=0$ has some frames which have very high energy (even Inf) evaluated at $\lambda=1$An output example is like below, a simulation under $\lambda=0.0$
In this case,
decorrelate_u_nk
doen't work properly, since it can't read a nan.I have already used softcore potentials to wrap the changing atoms up here, so I have no idea why amber still outputs
******
here.I guess we can allow
decorrelate_u_nk
to read this case and fill nan with a very large value.