Chapter 4. Reduced Cell Multipole Method

4.1. Introduction

Extending the CMM to handle systems with periodic boundary conditions substantially expands the range of problems that can be attacked. PBC is especially important for the simulation of bulk materials.

The Reduced Cell Multipole Method (RCMM) [1] is a relatively simple way of extending the CMM to periodic systems that maintains the overall scaling and memory usage advantages of the CMM. It builds on the idea of representing distant atoms with multipole expansions.

4.2. Enhanced Multipoles

First, an enlarged set of multipoles (through at least the hexadecapole moments) is computed from the atoms in the unit cell. This is an extended version of the standard CMM multipole computation and uses the same octree structure to ensure that it is O(N) in the number of atoms.

Next, a random set of points within the unit cell is chosen. Charges are placed on these points so as to reproduce the enlarged multipole set determined above. The number of random points is equal to the number of multipole coefficients, so this amounts to merely solving a set of simultaneous linear equations. This "reduced set" of random points and charges then can be used to substitute for the unit cell at large enough distances without substantially decreasing accuracy.

4.3. Ewald Summation

In order to handle the infinite periodic lattice, a standard Ewald summation method is used [2,3]. In this method, we divide the effective field of the infinite system at a given point into two pieces: a sum over atoms close to the unit cell in real space and a sum over terms in reciprocal space.

Computing an Ewald sum over all of the atoms in the unit cell would be computationally infeasible, so we only compute the sum over the reduced set, which is much smaller. This still reproduces all of the effects of the infinite system.

The real space part of the effective field is composed of the fields due to each of the nearby (reduced set) atoms, modified by a screening function.

V_real = \sum_atoms {q_atom/R erfc(R/\eta)} (1)

where R = |R_atom - R_eval| and \eta is a parameter that determines the cutoff range for the real space sum.

If there is an atom that is very close to the point at which the field is being evaluated, a different expression is used for the term due to that atom to account for the singularity in the field.

(-2 + 2/3 R^2/\eta^2 - R^4/\eta^4) q_atom / (\eta sqrt(\pi)) (2)

The sum of these screened terms converges very rapidly as the distance from the evaluation point increases.

To correct for the screening functions introduced in real space, we must add terms that represent their complement. This sum now converges slowly in real space, but its Fourier transform converges rapidly in reciprocal space.

V_recip = \sum_h {e^{-\pi^2 \eta^2 h^2} / (\pi \Omega h^2) e^{ih.R_eval} exp(i \sum_atoms {q_atom h.R_atom})} (3)

where h is the reciprocal lattice vector and \Omega is the volume of the unit cell.

Finally, a correction is added to handle the case of charged unit cells.

V_charge = - \pi \eta^2 q_cell / \Omega (4)

V = V_real + V_recip + V_charge (5)

4.4. Exclusion of Neighbors

At shorter ranges, the reduced set may not be an adequate representation of the atoms in the unit cell. Therefore, we will use a different technique (an extended CMM) to compute interactions with the nearest-neighbor unit cells. The Ewald sum over the reduced set already attempts to include these interactions, however, so we subtract out the field contribution from the nearest-neighbor unit cell images. Since this is within the range of the real-space portion of the Ewald sum, this subtraction is quite easy; the screening function in Equation 1 above is simply modified by subtracting 1.

4.5. Combination with CMM

The Ewald sum is evaluated at a specific set of points within the unit cell. Taylor series coefficients representing the resulting field values are then computed by interpolation. These coefficients compose the farfield due to the infinite array of atoms, except for those in the nearest-neighbor unit cells. This is exactly the correct farfield to use for the (level 0) unit cell at the beginning of the induction step of the standard CMM.

To compute the interactions with the neighboring unit cells, we can extend the CMM into those cells. We use the multipole expansions and fields calculated for the unit cell, but translated to the neighboring images. This adds extra calculations for CMM cells that used to be at the edge of the bounding box but now have image cells as neighbors, as well as additional calculations at level 1 in the octree, since the cells at that level now have PNCs in the image unit cells.

4.6. Noncubic Unit Cells

If the unit cell is not cubic, adjustments have to be made to the CMM. The method chosen is to compute a transformation matrix that converts the unit cell to a unit cube. The octree decomposition is then performed on the cube, with the cell coordinates being mapped back to real-space Cartesian coordinates by the inverse of the transformation matrix. The points at which the Ewald sum is computed are expressed in terms of transformed (unit cube) coordinates; the resulting interpolated Taylor series coefficients are then transformed back to real-space coordinates.
Next / Previous / Table of Contents
Kian-Tat Lim