MolAgent
Request early access

Examples

Five studies of the kind MolAgent prepares and runs, from five application areas: the question, the system and the plan of calculations. The structures were built for this page with RDKit, ASE and Packmol and checked for composition, connectivity and geometry. They are starting points, not results.

Periodic box of a lithium-ion battery electrolyte: ethylene carbonate and dimethyl carbonate molecules with lithium and hexafluorophosphate ions, in ball-and-stick.

1 M LiPF6 in EC:DMC (1:1 by volume): 24 ion pairs, 180 EC and 142 DMC molecules, 3,696 atoms packed in a 34.2 Å periodic box.

Drag to rotate, scroll or pinch to zoom. The 3D viewer is about 1 MB.

Batteries and electrolytes

Li+ solvation and transport in a carbonate electrolyte

How are lithium ions solvated in 1 M LiPF6 in EC:DMC, a standard reference electrolyte in lithium-ion battery research, and how fast do the ions move?

Composition
24 LiPF6, 180 ethylene carbonate, 142 dimethyl carbonate (EC:DMC 1:1 by volume)
Size
3,696 atoms, packed in a cubic box of 34.2 Å
Model
OPLS-AA, ionic charges scaled by 0.8
View in 3D

Plan

  1. Build EC, DMC and PF6− and pack the box with Packmol.
  2. Assign OPLS-AA parameters, with ionic charges scaled to account for electronic polarisation.
  3. Equilibrate under NPT at 298 K and check the density against experiment.
  4. Run NVT production; compute Li–O and Li–F radial distribution functions and the coordination numbers of EC, DMC and PF6−.
  5. Compute diffusion coefficients from mean-square displacements, with a finite-size correction, and the ionic conductivity from the collective displacement of the ions.
  6. Check the preferred solvation shell against DFT binding energies of small Li+ clusters.
md/in.npt, NPT equilibration in LAMMPS
# 1 M LiPF6 in EC:DMC (1:1 v/v), OPLS-AA with scaled ionic charges
units           real
atom_style      full
boundary        p p p

pair_style      lj/cut/coul/long 12.0
pair_modify     mix geometric tail yes
kspace_style    pppm 1.0e-5
bond_style      harmonic
angle_style     harmonic
dihedral_style  opls
improper_style  cvff
special_bonds   lj/coul 0.0 0.0 0.5

read_data       ../build/system.data

thermo_style    custom step temp press density pe etotal
thermo          1000

velocity        all create 298.15 81273 dist gaussian
fix             1 all npt temp 298.15 298.15 100.0 iso 1.0 1.0 1000.0
timestep        1.0
run             5000000
write_data      npt.data

Catalysis and surfaces

Intermediates of CO2 electroreduction on Cu(111)

Where and how strongly do the intermediates of CO2 reduction bind on copper, and which proton–electron transfer step sets the limiting potential for reduction to CO and beyond?

Surface
Cu(111), p(6×6), four layers with the bottom two fixed, 20 Å of vacuum
Adsorbates
Three CO* and one COOH* on top sites, three H* in fcc hollows
Size
157 atoms
View in 3D

Plan

  1. Optimise the bulk lattice constant with the chosen functional, here PBE with D3(BJ) dispersion.
  2. Converge the slab model: number of layers, k-point mesh, plane-wave cut-off and vacuum.
  3. Screen adsorption sites for COOH*, CO*, CHO*, COH* and H*, first with a machine-learned potential, then with DFT. Check that CO prefers the top site, as observed experimentally; GGA functionals tend to favour hollow sites.
  4. Compute the vibrational frequencies of the adsorbates for zero-point energies and entropies.
  5. Build the free-energy diagram with the computational hydrogen electrode and find the limiting potential.
  6. Compute activation barriers for the key proton–electron transfer steps, with explicit water at the surface: the limiting potential is a thermodynamic quantity and does not identify the rate-determining step.
  7. Test how the result depends on coverage and on implicit solvation.
coads/INCAR, relaxation of the co-adsorbed configuration in VASP
SYSTEM  = CO*, COOH* and H* on Cu(111) p(6x6), 4 layers, bottom 2 fixed
PREC    = Accurate
ENCUT   = 450
GGA     = PE
IVDW    = 12          ! D3(BJ)
ISMEAR  = 1
SIGMA   = 0.10
EDIFF   = 1E-6
EDIFFG  = -0.02
IBRION  = 2
NSW     = 300
ISIF    = 2
LDIPOL  = .TRUE.
IDIPOL  = 3
LREAL   = Auto
ALGO    = Fast
NCORE   = 8
LWAVE   = .FALSE.
LCHARG  = .FALSE.

Semiconductors and 2D materials

Sulfur vacancies in monolayer MoS2

What does it cost to remove a sulfur atom from monolayer MoS2, which charge states of the vacancy are stable, and where do its levels lie in the band gap?

Structure
MoS2 monolayer, 9×9 supercell, one vacancy in the upper sulfur plane; in the viewer, its three Mo neighbours are shown in red
Size
242 atoms, 18 Å of vacuum
View in 3D

Plan

  1. Converge the pristine monolayer: lattice constant, cut-off, k-points and vacuum.
  2. Relax the vacancy in its neutral and charged states, in supercells of increasing size.
  3. Compute formation energies as a function of the Fermi level and the sulfur chemical potential, with corrections for charged defects in two dimensions.
  4. Take the charge transition levels from the formation energies, with hybrid-functional band edges, and compare them with the defect states in the projected density of states.
  5. Extend to vacancy migration barriers and to passivation by oxygen.
vS/pw.relax.in, relaxation with Quantum ESPRESSO (cell and atomic positions omitted)
&control
  calculation = 'relax'
  prefix = 'mos2_vS'
  pseudo_dir = './pseudo'
  forc_conv_thr = 1.0d-4
/
&system
  ibrav = 0
  nat = 242
  ntyp = 2
  ecutwfc = 50
  ecutrho = 400
  occupations = 'smearing'
  smearing = 'mv'
  degauss = 0.005
  assume_isolated = '2D'
/
&electrons
  conv_thr = 1.0d-8
  mixing_beta = 0.3
/
&ions
/
ATOMIC_SPECIES
  Mo  95.95  Mo.pbe-spn-rrkjus_psl.1.0.0.UPF
  S   32.06  S.pbe-n-rrkjus_psl.1.0.0.UPF
K_POINTS automatic
  3 3 1 0 0 0

Pharmaceuticals and formulation

Ibuprofen binding to β-cyclodextrin

How strongly does β-cyclodextrin bind ibuprofen in water, in which orientation, and how does the answer change with the protonation state of the drug?

Host
β-cyclodextrin, seven glucose units
Guest
(S)-ibuprofen, inside the cavity; in the viewer, its carbon atoms are shown in orange
Size
180 atoms before solvation
View in 3D

β-cyclodextrin coordinates adapted from the MobleyLab benchmark sets (BSD licence).

Plan

  1. Set up the neutral and the anionic form of ibuprofen; the carboxylic acid is deprotonated at physiological pH.
  2. Parameterise the guest with GAFF2 and RESP charges. For the host, compare GAFF2 with a carbohydrate force field (GLYCAM06 or q4md-CD): the choice of host parameters changes computed binding free energies. Solvate in explicit water.
  3. Place the guest in both orientations, with the carboxyl group at the wide or at the narrow rim, and equilibrate each complex.
  4. Compute the binding free energy by alchemical double decoupling, with Boresch restraints and their standard-state correction and, for the anion, a net-charge correction; and independently by umbrella sampling along the dissociation path.
  5. Compare with experimental binding constants and report the populations of the two orientations.

Polymers, resins and composites

Curing and glass transition of an epoxy network

How do the glass transition temperature and the elastic modulus of an amine-cured epoxy depend on the degree of cure?

Resin
70 DGEBA (bisphenol A diglycidyl ether), RR, RS and SS stereoisomers in about 1:2:1
Hardener
28 DETA (diethylenetriamine)
Ratio
Stoichiometric: 140 epoxide groups, 140 amine hydrogens
Size
3,990 atoms, 40 Å box before compression
View in 3D

Plan

  1. Pack the monomers with Packmol and compress the mixture to liquid density under NPT.
  2. Parameterise the monomers, and the reaction templates before and after each bond forms, with a general force field.
  3. Crosslink with REACTER (LAMMPS fix bond/react): when an amine nitrogen that still carries a hydrogen comes within a cut-off distance of a terminal epoxide carbon, the N–C bond forms, the epoxide C–O bond at that carbon breaks and the hydrogen moves from nitrogen to oxygen, giving a hydroxyl group. Primary and secondary amines react through separate templates, and the network is relaxed as it forms, up to conversions between 70 and 95 %.
  4. Cool each network stepwise and take the glass transition temperature from the change of slope of density against temperature.
  5. Deform the networks uniaxially for Young's modulus and Poisson's ratio, and compare with experiment, accounting for the cooling and strain rates of simulation.
build/packmol.inp, packing of the resin and hardener
tolerance 2.0
filetype xyz
output mix_packmol.xyz
seed 4321
structure dgeba_rr.xyz
  number 18
  inside box 1.0 1.0 1.0 39.0 39.0 39.0
end structure
structure dgeba_rs.xyz
  number 34
  inside box 1.0 1.0 1.0 39.0 39.0 39.0
end structure
structure dgeba_ss.xyz
  number 18
  inside box 1.0 1.0 1.0 39.0 39.0 39.0
end structure
structure deta.xyz
  number 28
  inside box 1.0 1.0 1.0 39.0 39.0 39.0
end structure

Early access

MolAgent is in alpha. We are installing it with a small number of research groups now; a private beta opens in 2027. Write to us with the systems you study, the codes you use and the machines you compute on.

Request early access How setup works

Or write to contact@molagent.ai.