Dear AMBER Users,
I am performing thermodynamic integration (TI) calculations in AMBER for absolute free energy estimation using soft-core potentials, where a molecule is decoupled from the rest of the system.
I had a few questions regarding interpretation of the output and best practices for analysis:
1.
DV/DL values during the simulation
In the MD output file, DV/DL (dV/dλ) values are printed at intervals . My understanding is that these correspond to the instantaneous derivative of the potential energy with respect to λ at a given configuration.
Is this correct?
Should these instantaneous values be averaged over the trajectory to obtain ⟨dV/dλ⟩ for that λ window?
2.
logdvdl output
When logdvdl=1 is enabled, AMBER produces DV/DL values at the end of the output file.
Are these values different from those printed in the main output, or simply more frequently sampled?
For computing free energy differences, should one rely on the logdvdl file rather than the main output?
3.
Which values to use for ΔG calculation
To compute ΔG via TI, my current understanding is:
For each λ window, compute the ensemble average ⟨dV/dλ⟩
Numerically integrate ⟨dV/dλ⟩ over λ (for example, trapezoidal rule)
Could you confirm if this workflow is correct, and whether additional processing such as block averaging or decorrelation is recommended before integration?
1.
MBAR output
When the MBAR flag is activated, AMBER reports additional values.
Could you clarify what these values represent physically?
In particular, what does it indicate if one λ window shows a very large positive value? Does this suggest poor overlap, insufficient sampling, or a setup issue?
2.
Convergence assessment
I would also appreciate guidance on how to assess convergence of TI calculations. Specifically:
Is it sufficient to monitor stability of ⟨dV/dλ⟩ over time within each λ window?
Are there recommended diagnostics such as forward versus backward integration, block averaging, or overlap checks that should be routinely applied?
Any best practices for choosing simulation length per λ window in soft-core decoupling simulations?
Thank you for your time and help.
Best regards,
Rahul Singal
IIIT Hyderabad
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Mar 19 2026 - 04:00:02 PDT
