# ASE for QM/MM Simulations¶

QM/MM Simulations couple two (or, in principle, more) descriptions to get total energy
and forces for the entire system in an efficiant manner.
ASE has a native Explicit Interaction calculator, `EIQMMM`

, that uses an electrostatic embedding
model to couple the subsystems explicitly. See
the method paper for more info.,

Examples of what this code has been used for can be seen here, and here.

This section will show you how to setup up various QM/MM simulations. We will be using GPAW for the QM part. Other QM calculators should be straightforwardly compatible with the subtractive-scheme SimpleQMMM calculator, but for the Excplicit Interaction EIQMMM calculator, you would need to be able to put an electrostatic external potential into the calculator for the QM subsystem.

You might also be interested in the solvent MM potentials included in ASE. The tutorial on Equilibrating A TIPnP Water Box could be relevant to have a look at.

## Electrostatic Embedding QM/MM¶

The total energy expression for the full QM/MM system is:

The MM region is modelled using point charge force fields, with charges \(q_i\) and \(\tau_i\) denoting their spatial coordinates, so the QM/MM coupling term \(E_\mathrm{I}\) will be

where \(n({\bf r})\) is the spatial electronic density of the quantum region, \(Z_\alpha\) and \({\bf R}_\alpha\) are the charge and coordinates of the nuclei in the QM region, respectively, and \(E_\mathrm{RD}\) is the term describing the remaining, non-Coulomb interactions between the two subsystems.

For the MM point-charge external potential in GPAW, we use the total pseudo- charge density \(\tilde{\rho}({\bf r})\) for the coupling, and since the Coloumb integral is evaluated numerically on the real space grid, thus the coupling term ends up like this:

Currently, the term for \(E_{\mathrm{RD}}\) implemented is a Lennard- Jones-type potential:

Let’s first do a very simple electrostatic embedding QM/MM single point
energy calculation on the water dimer. The necessary inputs are described in
the `ase.calculators.qmmm.EIQMMM`

class.

The following script will calculate the QM/MM single point energy of the water dimer from the S22 database of weakly interacting dimers and complexes, using LDA and TIP3P, for illustration purposes.

```
from __future__ import print_function
from ase.data import s22
from ase.calculators.tip3p import TIP3P, epsilon0, sigma0
from ase.calculators.qmmm import EIQMMM, LJInteractions, Embedding
from gpaw import GPAW
# Create system
atoms = s22.create_s22_system('Water_dimer')
atoms.center(vacuum=4.0)
# Make QM atoms selection of first water molecule:
qm_idx = range(3)
# Set up interaction & embedding object
interaction = LJInteractions({('O', 'O'): (epsilon0, sigma0)})
embedding = Embedding(rc=0.02) # Short range analytical potential cutoff
# Set up calculator
atoms.calc = EIQMMM(qm_idx,
GPAW(txt='qm.out'),
TIP3P(),
interaction,
embedding=embedding,
vacuum=None, # if None, QM cell = MM cell
output='qmmm.log')
print(atoms.get_potential_energy())
```

Here, we have just used the TIP3P LJ parameters for the QM part as well. If this is a good idea or not isn’t trivial. The LJInteractions module needs combined parameters for all possible permutations of atom types in your system, that have LJ parameters. A list of combination rules can be found here. Here’s a code snippet of how to combine LJ parameters of atom types A and B via the Lorentz-Berthelot rules:

```
import itertools as it
parameters = {'A': (epsAA, sigAA),
'B': (epsBB, sigBB)}
def lorenz_berthelot(p):
combined = {}
for comb in it.product(p.keys(), repeat=2):
combined[comb] = ((p[comb[0]][0] * p[comb[1]][0])**0.5,
(p[comb[0]][1] + p[comb[1]][1])/2)
return combined
combined = lorenz_berthelot(parameters)
interaction = LJInteractions(combined)
```

This will (somewhat redundantly) yield:

```
>>>combined
{('A', 'A'): (epsAA, sigAA),
('A', 'B'): (epsAB, sigAB),
('B', 'A'): (epsAB, sigAB),
('B', 'B'): (epsBB, sigBB)}
```

It is also possible to run structural relaxations and molecular dynamics using the electrostatic embedding scheme:

```
from ase.constraints import FixBondLengths
from ase.optimize import LBFGS
mm_bonds = [(3, 4), (4, 5), (5, 3)]
atoms.constraints = FixBondLengths(mm_bonds)
dyn = LBFGS(atoms=atoms, trajectory='dimer.traj')
dyn.run(fmax=0.05)
```

Since TIP3P is a rigid potential, we constrain all interatomic distances. QM bond lengths can be constrained too, in the same manner.

The implementation was developed with the focus of modelling ions and complexes in solutions, we’re working on expanding its functionality to encompass surfaces.

In broad strokes, the steps to performing QM/MM MD simulations for thermal sampling or dynamics studies, these are the steps:

QM/MM MD General Strategy for A QM complex in an MM solvent:

- Equillibrate an MM solvent box using one of the MM potentials built into ASE (see Equilibrating A TIPnP Water Box for water potentials), one of the compatible external MM codes, or write your own potential (see Adding new calculators)
- Optimize the gas-phase structure of your QM complex in GPAW, analyze what level of accuracy you will need for your task.
- Place the relaxed structure of the QM molecule in your MM solvent box, deleting overlapping MM molecules.
- Re-equillibrate the QM/MM system.
- Run production runs.

For these types of simulations, you’d probably want two cells: a QM (non- periodic) and and MM cell (periodic):

```
atoms.set_pbc(True)
# Set up calculator
atoms.calc = EIQMMM(
qm_idx,
GPAW(txt='qm.out'),
TIP3P(),
interaction,
embedding=embedding,
vacuum=4., # Now QM cell has walls min. 4 Å from QM atoms
output='qmmm.log')
```

This will center the QM subsystem in the MM cell.

Current limitations:

- No QM/MM border over bonds
- No QM PBCs