7
$\begingroup$

I want to know the procedure to calculate the Lennard-Jones potential for a metal-halogen pair (specifically vanadium-chlorine). Is it possible to calculate using any QM packages like Mopac, NWChem, or Gaussian?

I am specifically looking for values of A and B in the 12-6 potential.

$\endgroup$
4
  • 3
    $\begingroup$ Scan the bond length around its equilibrium and fit the energies afterwards. $\endgroup$ Jun 24, 2017 at 14:39
  • 1
    $\begingroup$ @pH13-YetanotherPhilipp That won't work, since it will incorporate all interactions up to the method level (electrostatics, polarization, exchange repulsion, dispersion, and so on). The LJ potential should only be used to approximate exchange repulsion and dispersion. $\endgroup$ Jun 24, 2017 at 15:06
  • $\begingroup$ @pentavalentcarbon Do you have any link/source to provide further information on that? I never heard of such a limitation. $\endgroup$ Jun 24, 2017 at 15:12
  • 2
    $\begingroup$ @pH13-YetanotherPhilipp You can tell just based on the mathematical form. The Coulomb potential goes like $1/r$, and since the gross features of interactions between atoms/molecules tend to be dominated by charge-charge interactions rather than charge-dipole/higher-order ones, I wouldn't expect $(1/r^{12})-(1/r^{6})$ to handle this well. For anything charged a Coulomb term would be necessary. en.wikipedia.org/wiki/Interatomic_potential $\endgroup$ Jun 24, 2017 at 18:26

1 Answer 1

16
$\begingroup$

Yes, this is technically possible. A basic tutorial for this is in the excellent Psi4Numpy project, which I'll reproduce here with minor modifications. Their example fits the counterpoise-corrected MP2/aug-cc-pVDZ total interaction energy of the helium dimer.

from __future__ import print_function

import psi4

import numpy as np

import matplotlib as mpl
mpl.use('Agg')
import matplotlib.pyplot as plt


he_dimer = """
He
--
He 1 **R**
"""

distances = [2.875, 3.0, 3.125, 3.25, 3.375, 3.5, 3.75, 4.0, 4.5, 5.0, 6.0, 7.0]
energies = []
for d in distances:
    # Build a new molecule at each separation
    mol = psi4.geometry(he_dimer.replace('**R**', str(d)))

    # Compute the Counterpoise-Corrected interaction energy
    en = psi4.energy('MP2/aug-cc-pVDZ', molecule=mol, bsse_type='cp')

    # Place in a reasonable unit, Wavenumbers in this case
    en *= 219474.6

    # Append the value to our list
    energies.append(en)

print("Finished computing the potential!")

# Fit data in least-squares way to a -12, -6 polynomial
powers = [-12, -6]
x = np.power(np.array(distances).reshape(-1, 1), powers)
coeffs = np.linalg.lstsq(x, energies)[0]

# Build list of points
fpoints = np.linspace(2, 7, 50).reshape(-1, 1)
fdata = np.power(fpoints, powers)

fit_energies = np.dot(fdata, coeffs)

fig, ax = plt.subplots()
ax.set_xlim((2, 7))  # X limits
ax.set_ylim((-7, 2))  # Y limits
ax.scatter(distances, energies)  # Scatter plot of the distances/energies
ax.plot(fpoints, fit_energies)  # Fit data
ax.plot([0,10], [0,0], 'k-')  # Make a line at 0
ax.set_xlabel(r'intertomic separation ($\AA{}$)')
ax.set_ylabel(r'interaction energy ($\mathrm{cm^{-1}}$)')
fig.savefig('1b_molecule.pdf', bbox_inches='tight')

Once the set of energies is calculated, a least-squares polynomial fit is performed, giving

$$ f(x) = 6677721.45419193x^{-12} - 11394.79882998x^{-6}. $$

This isn't a good (efficient or general) way of evaluating the fit function, but illustrates the placement of correct signs in the above code:

def f(z):
    return coeffs[0]*z**powers[0] + coeffs[1]*z**powers[1]

From the above we can see that the first and second coefficients in the polynomial fit correspond directly to $A$ and $B$, respectively. In this example, the energies are in wavenumbers ($\pu{cm^{-1}}$). Most quantum packages print in atomic units (Hartrees) and dynamics packages print in kJ/mol, which is fine, but be aware that in the above fit the units are incorporated into $A$ and $B$. Here is the resulting fit compared against the individual interaction energy calculations:


This was not in the question, but I think it's important to ask if this is the correct approach and investigate further. Notice how the fit is qualitatively ok near the minimum, but the shape in the dissociation region is qualitatively wrong, and the potential simply isn't attractive enough in the limit of infinite separation. This could arise in three ways:

  1. The curve fitting isn't working properly, or there's a bug in the code.
  2. The functional form of the chosen interatomic potential is incorrect.
  3. The (Boys-Bernardi) counterpoise correction, which is known to overcompensate for BSSE, is giving strange results at large separation.

Assume that point 1 isn't an issue (if there's a bug anywhere, please comment). Point 2 can be investigated by also fitting against the Buckingham and shifted Morse potentials, which should display better short-range and long-range behavior:

\begin{align} V_{\text{Buckingham}}(r;A,B,C) &= A e^{-Br} - \frac{C}{r^{6}} \\ V_{\text{Morse}}(r;A,B,C,D) &= A \left( 1 - e^{-B(r-C)} \right)^{2} + D \end{align}

Note that for these energies, an unshifted Morse potential (no $D$ parameter) fails miserably. See here for more information about fitting Morse-type potentials. The Buckingham and Morse fits require a more general (arbitrary) curve fitting routine, and the script that made this plot can be found as an HTML comment in the post source code below.

I find it interesting that the Buckingham potential gives a worse fit than the Lennard-Jones one, even though it is supposed to reproduce the repulsive wall better with $e^{-r}$ rather than $\frac{1}{r^{6}}$. The Morse fit is remarkably good, which makes sense considering there are 4 free parameters rather than 3 (Buckingham) or 2 (Lennard-Jones). Before going any further, a check on point 3 by setting bsse_type=nocp:

This reveals that the counterpoise correction is definitely interfering with the quality of the fit, and the functional forms of the Lennard-Jones and Buckingham potential appear to not describe dissociation properly, though extended to infinite separation, the Morse potential is the one that is qualitatively incorrect. This can be attributed to using a single functional form to describe the total interaction or binding energy, and not the different components. I suspect that the apparent poor fit is due to a non-zero Coulomb interaction, rather than van der Waals-type interactions which at least the Lennard-Jones potential is meant to be used for.

To test this, here are interaction energy calculations at two levels of sophistication, both based on symmetry-adapted perturbation theory (SAPT). This is still the helium dimer, at 7 angstroms separation, with the aug-cc-pVDZ basis set and the monomer-centered basis approximation. All units are kcal/mol.

$$ \small \begin{array}{lrr} \hline \text{Component} & \text{SAPT0} & \text{SAPT2+3(CCD)}\delta_{\text{MP2}} \\ \hline \text{Electrostatics} & -0.01796877 & -0.01777334 \\ \text{Exchange} & 0.00000000 & 0.00000000 \\ \text{Induction} & 0.01587019 & 0.01508578 \\ \text{Dispersion} & -0.00013235 & -0.00016773 \\ \hline \text{Electrostatics + Induction} & -0.00209858 & -0.00268756 \\ \hline \text{Total} & -0.00223093 & -0.00285529 \\ \hline \end{array} $$

At both levels of SAPT, the non-dispersive part of the interaction energy accounts for 94% of the total interaction! Is this true at 3 angstroms separation?

$$ \small \begin{array}{lrr} \hline \text{Component} & \text{SAPT0} & \text{SAPT2+3(CCD)}\delta_{\text{MP2}} \\ \hline \text{Electrostatics} & -0.04852183 & -0.04836334 \\ \text{Exchange} & 0.02045965 & 0.02219227 \\ \text{Induction} & 0.04202349 & 0.04081412 \\ \text{Dispersion} & -0.02228767 & -0.02843936 \\ \hline \text{Electrostatics + Induction} & -0.00649834 & -0.00754922 \\ \hline \text{Total} & -0.00832637 & -0.01379630 \\ \hline \end{array} $$

The non-vdW interactions now account for 78% and 55% of the total interaction energy for each SAPT flavor, respectively. Although induction (also called polarization) is repulsive, I am including it in the Coulomb-type interaction since it is not purely quantum mechanical in nature like the exchange term. It is the exchange term, not induction, that is closer to the charge cloud penetration picture. Even so, the interaction between two helium atoms is not purely based on dispersion, and requires fitting another nonbonded term for electrostatics, the Coulomb term:

$$ V_{\text{Coulomb}}(\vec{r}_{i},\vec{r}_{j},q_{i},q_{j}) = \frac{1}{4\pi\epsilon_0} \frac{q_{i}q_{j}}{|\vec{r}_{i} - \vec{r}_{j}|} $$

Our goal is to separate out the Coulomb-like terms from the interaction energy, so that $A$ will describe only the exchange contribution and $B$ will describe only the dispersion contribution; electrostatics and induction will be handled separately in the Coulomb term. This will be done by fitting only to the combination of exchange and dispersion, which are taken from SAPT calculations. MP2 binding energies are not CP-corrected.

As you can see, using SAPT energies provides a much better fit than naively using the total interaction or binding energies. I'm not sure why the total SAPT energy works so well, but it may be due to the absolute magnitudes at each point being so small.

The importance of only fitting exchange and dispersion is obvious for $\ce{Na^+---Cl^-}$ (using def2-SVP for the basis).

This is an unfair example, since the dispersion interaction is dwarfed by exchange, and it is an ionically-bound molecule. As a final example, consider $\ce{V(III)---Cl^-}$ (using def2-SV(P) for the basis, vanadium as a quintet).

I had some trouble converging many of the calculations, presumably due to a non-optimal spin state at long range. For this reason (spin-state crossing), Lennard-Jones type models are poor for metal-x interactions. The problem appears to be similar to the sodium chloride, but the exchange plus dispersion fit looks good.

If you want to reproduce any of the work, rather than fish through the source for inputs and scripts, everything is in a GitHub repository. More references for SAPT can be found here, along with the Psi4 manual. Note that I didn't do any literature searching here, but I expect people have used energy decomposition approaches in the past as part of force field design. Hopefully this serves as a good starting point.

$\endgroup$
6
  • $\begingroup$ This is just awesome! $\endgroup$ Jun 28, 2017 at 13:40
  • 2
    $\begingroup$ All of the credit should go to the Psi4 developers; if you're a Python hacker, they have the best quantum package out there. $\endgroup$ Jun 28, 2017 at 15:02
  • $\begingroup$ On second thought, I suspect that the real reason the LJ fits for most of the curves didn't work is due to a bad guess in scipy.optimize.curve_fit. This is worth investigating and would invalidate some of my argument. $\endgroup$ Jun 28, 2017 at 15:05
  • $\begingroup$ Thank you soo much for this answer but I really couldn't understand how to use it. May be because I don't know how to use phyton. Is there any video tutorial on a molecule? $\endgroup$ May 28, 2018 at 18:56
  • 2
    $\begingroup$ @BarisVvolf I really suggest learning Python. For intermolecular potentials, I recommend searching for that phrase or "Lennard-Jones"; the Wikipedia article and related ones are very good. Increasing the level of detail, Anthony Stone's "The Theory of Intermolecular Forces" is excellent. For curve fitting, YouTube has plenty. For running Psi4 from Python, YouTube! For this specific example, I could be convinced to make a video... $\endgroup$ May 28, 2018 at 19:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.