I am trying to run a QM/MM MD calculation on an enzyme substrate system using AMBER after successful completion of MD production run in an Isobaric-Isothermal (NPT) ensemble without errors.
I am using the .rst file from the MD production run to perform a 100ps equilibration (to relax the system) with QM/MM using SCC-DFTB Hamiltonian but it keeps giving the below error:
QMMM SCC-DFTB: !!!! ============= WARNING ============= !!!!
QMMM SCC-DFTB: Convergence could not be achieved in this step.
QMMM SCC-DFTB: The calculation will continue, but energies and forces for this step will not be accurate.
Can anyone help me with useful hints to solve this problem?
Thanks!
Hi,
From your sample output, I assume that you are using the SCC-DFTB implementation available in AmberTools (sqm/sander). The answer of B. B. Ivanova is likely to be correct, but seems specific to the use of Gaussian as an external QM program.
Convergence issues in the QM part of a QM/MM calculation can come from many factors among which:
1) the electronic structure of your QM part can simply be very hard to solve (even in the gas phase) and requires special treatment (e.g., transition metal complex).
If that is the case, you could:
a) increase the number of SCF cycles
b) trick the method into converging better (this is quite common in DFT calculations and related approximate SCC-DFTB; see bellow)
c) increase the level of theory and the basis set if you can afford it (e.g., DFT, MP2, CCSD, ...)
If you cannot afford to increase the level of theory and want to stick with SCC-DFTB (not a bad choice), you have only a few tricks that you can apply. Increase the maximum number of SCF cycles (in Amber/sander you can change the value of itrmax in the &qmmm section of your intput) and/or set the dftb_telec keyword to a reasonable value (e.g., 100K as suggested in the manual of Amber). Quoting the manual: "The electronic temperature affects the Fermi distribution promoting some HOMO/LUMO mixing, which can accelerate the convergence in difficult cases." In short, it's a good trick to make your calculation converge.
I would recommend to set those two keywords (i.e., itrmax and dftb_telec).
2) the convergence of a QM calculation is quite sensitive to the geometry of the molecular system. Even in gas phase, it can sometimes be very hard if not impossible to converge the SCF of a geometry that is very far from a minimum of the method that you use, or far from any reasonable reaction path. When extracting structures from MM-MD, it is very likely that this particular geometry of the QM part will have some bonds, angles ... that are not reasonable from a QM point of view. For example, if a C-H bond is compressed to say 0.5 A for whatever reason in this particular snapshot of your MD, your SCF is likely to have troubles to converge. All that to say that MM geometries, especially at non-zero temperature, do not necessarily match what a QM calculation would have given. Both the QM and MM potential energy surfaces might differ significantly, especially when exploring regions far from a minimum.
Depending on your purpose, I would recommend to have a carful look at the geometry of your QM part with particular attention to chemically unreasonable interatomic distances. If your purpose is to minimize that structure, then a preliminary MM minimization would yield a geometry that would match better the QM potential energy surface.
3) the success of QM/MM calculations often depends on the way in which you treat the QM/MM interface. The defaults of Amber are usually fine but you could also explore that route. Possible issues can come from where you chose to place link atoms (if any). Other aspects of the QM/MM boundary might have significant impact on your calculation. As a start about good practice in QM/MM, I suggest you the following publication: Article Determination of enzymatic reaction pathways using QM/MM met...
If your QM part is not covalently bound to your MM part, there might still be a problem with the QM/MM interface. For example, the concentration of many negative or positive charges close to your QM part would strongly polarize your wave function (or density in your case) and give a hard time to the QM algorithm to converge the SCF.
Again, I would first look carefully at the QM part and its environment with the critical eye of a chemist and spot unreasonable situations. Also make sure that you set the correct charge and multiplicity for your QM part. In Amber, the geometry of the QM part including the link atoms (if any) is printed somewhere at the beginning of the output. Just visualize it and make sure that this is what you want and that it matches your settings. If this is not what you expected then you probably made a mistake in defining you QM mask.
Overall, if you feel like your geometry is reasonable, that your QM/MM settings are alright and if your QM part is mostly organic, then setting dftb_telec=100 should do the trick.
Hope it helps,
Good luck.
. . . or try Empirical Valence Bond (EVB) approach.
Article Empirical valence bond calculations of enzyme catalysis
- Badges
- Science topic
- Similar topics
- Bioinformatics
- Bioinformatics and Computational Biology