Re: [AMBER] TI simulation blowing up with pmemd.CUDA

From: Debarati DasGupta <>
Date: Mon, 11 Jan 2021 13:52:01 +0000

Hello Senal,

I hope you are doing good.
Welcome to the “tricky-TI” project.
Well, I have been doing some TI calculations in AMBER20 since last 1 year, and I have faced many issues and a lot of help/directions were given by Prof Case
( I genuinely thank him for that).
What are your system details? Total number of atoms? Are you doing RBFE or ABFE? What’s the process for which you are trying to calculate the FE?
I have always done the concerted protocol, where you decharge and remove the LJ interactions in 1 step.. Yes it can be error prone but using the smoothstep softcore (SSC P=2) potentials ( search Darrin York latest TI papers published in June/July 2020 JCIM) , a lot of those issues got resolved. I would also recommend doing a non-TI, simple conventional MD equilibration and production to help alleviate unfavourable interactions, high energies, steric clashes, prior to setting up the TI calculations. I can provide you some more tips I used, depending on what system and what FE you are trying to calculate. If possible switch to AMBER20, just switching from AMBER18 to20 solved some of my TI based problems.
Best wishes

Sent: 10 January 2021 15:08
To: AMBER Mailing List<>
Subject: Re: [AMBER] TI simulation blowing up with pmemd.CUDA

Hello Dr. Case,

The unit cell dimensions are as 52.981 x 42.118 x 52.650 angstroms and the amber 18 version have been updated to the latest. There wasn't any issue during the decharging step throughout the lambda values. So, I'm unsure as to why a sudden energy burst should happen at the second lambda point simulation for vdW removal.

Following is the terminal output I obtained from running my script on the CPU code.

simulating vdw_bonded
changing directory to lambda = 0.00922
equilibration for 350 ps
equilibration for 1 ns
sampling for 10 ns
changing directory to lambda = 0.04794
equilibration for 350 ps
equilibration for 1 ns
sampling for 10 ns
Error: an illegal memory access was encountered launching kernel kNLSkinTest

Thank you

Senal Liyanage
From: David A Case <>
Sent: Sunday, January 10, 2021 1:28 PM
To: AMBER Mailing List <>
Subject: Re: [AMBER] TI simulation blowing up with pmemd.CUDA

On Sun, Jan 10, 2021, Delwakkada Liyanage, Senal Dinuka wrote:
>I am trying to do TI in Amber18 with pmemd.cuda. The system is a small
>porphyrin ligand ( C44H38N8), prepared in a 16 A TIP3P water box.

What are the unit cell dimensions for your system? Small systems can be
unstable with with our GPU code, but I'm not sure when the relevant error
message was inserted into the code.

For sure, be certain that you have applied all updates to the code. If it's
not something obvious, you will probably need to upgrade to Amber20: there
have been significant changes in how softcore TI is done, and not all of the
fixes have been back-ported to Amber18.


AMBER mailing list

AMBER mailing list

AMBER mailing list
Received on Mon Jan 11 2021 - 06:00:07 PST
Custom Search