DMol



3       General Methodology

This chapter explains the general methodology for performing DMol calculations, whether you are using DMol in the Insight environment or the standalone mode. Specific instructions for using DMol in either of these modes are contained in Chapters 5 and 8, respectively.

A thorough understanding of the DFT method is necessary for an appreciation of many of the input options for DMol. Chapter 2, Theory and Implementation, presents a brief overview of the DFT method as implemented in DMol and relates aspects of DFT theory to some of the input options. An understanding of the computational aspects of DFT theory enables you to use the DMol program more effectively.

If you are using the Insight program, you can still run DMol in the standalone mode, so you can, for example, set up a run with the Insight interface, run the calculation in standalone mode, and then analyze some of the results back in the Insight interface. The input and output files are all compatible between both modes. You could, for example, use the Insight modeling facilities to construct a molecule and the DMol module to determine the symmetry, set some constraints, and write out the command input file. You could then exit the Insight environment to modify the input file with a text editor or through the dmol_master interface and then submit the DMol job directly. After the run finishes, you can read the output back into the Insight program to, for example, examine the orbital density-of-states or animate the normal modes of vibration.


Choosing a job type

DMol calculations can be divided into three main types:

  1. Energy calculation.

  2. Geometry optimization.

  3. Calculation of frequency and thermodynamic properties.

Optional program functions include calculations for:

The input specifications needed for the three main job types are summarized briefly below.

Energy calculation

Energy calculations evaluate the following molecular properties at a single (user-specified) geometry:

Whenever you run a calculation, you always obtain the energy of the system. You can specify the number of iterations and the convergence criterion for the wavefunction. The accuracy required depends on the type of properties that you want to calculate. For example, dipole moments require a higher degree of convergence than density plots, which are less sensitive to inaccuracies.

You can, optionally, calculate the other properties listed above. How to do this is discussed in detail in the online help and Appendix E (Commands--Standalone Mode). Dipole moments, charges, and the properties to be graphed are each selected by setting a single parameter or keyword, either indirectly from within the Insight program or directly if you are using DMol in the standalone mode. However, calculating the optical polarizability requires four separate calculations with electric fields imposed on the molecule.

Geometry optimization

Geometry optimization determines the minimum-energy structure of a molecule, starting from the initial geometry entered. After the geometry is minimized, any other requested properties are computed. You can control the following aspects of the geometry optimization process:

During minimization, the program updates the .car file with the most recent geometry and saves the current Hessian in the .hessian file. This allows you to restart an interrupted minimization, starting with the most recent geometry and gradient information. The DMol restart capability does not allow for extrapolation of the input density at the old mesh to the current numerical integration mesh. As a result, the old input .tpotl file has to have the same mesh and basis set as the current input.

After the geometry is successfully minimized, the program then computes any other properties requested.

DMol may exhibit slow SCF convergence when there is a small HOMO-LUMO gap. Using smearing and/or decreasing the mixing parameters are recommended for solving this problem.

The DIIS option should always be used, to improve SCF convergence.

Optimal strategy

The optimal strategy for geometry optimization may be as follows:

  1. Perform an LSD calculation using a Medium grid and relaxed SCF and geometry optimization criteria.

For example, use SCF_Density_Convergence = 0.001; Opt_Energy_Convergence = 0.0001; Gradient_Convergence = 0.01 when running DMol in standalone mode (Appendix E). The equivalent parameter names for the DMol module in Insight are SCF_Dens, Ener_Conv, and Grad_Conv (Chapter 5).

  1. After the first calculation is finished, run the calculation again with much better convergence criteria.

Start from the geometry obtained at the end of the first run and increase the grid to Fine.

Some molecules that have a flat potential energy surface may require an Xfine grid in order to meet the default convergence criteria.

Vibrational frequencies

DMol can evaluate frequencies by finite differences of analytic gradients. Specifying a vibrational calculation results in a geometry optimization followed by computation of the frequencies. You can control several aspects of how frequencies are computed:


Setting up the computation

The input to DMol is designed to offer maximum flexibility in specifying options for solving the DFT equations. This flexibility enables you to balance the computational cost of a problem against the desired accuracy of the results.

To perform any calculation, you must specify at least the geometry of the molecule. You can also specify the following input parameters, to restrict or refine your calculations:

The program is designed to automatically select reasonable default values whenever possible and to still preform reliable calculations.

Using the frozen core option together with symmetry and the minimal basis set is not recommended.

Symmetry can be used together with the Point_Charges option. However, you must be sure to check whether the input symmetry is a subgroup of both the molecule and the point-charge system.


Molecule size and hardware requirements

Please see Memory allocation in DMol if you receive messages referring to errors in memory requirements.


Outline of basic steps for a DMol calculation

Four basic steps are involved in a DMol calculation:

  1. Create and modify the input molecule.

  2. Set up calculation parameters.

  3. Run the calculation.

  4. Analyze the results.

How to implement these steps differs, depending on whether you are using DMol in the Insight environment or the standalone mode. Specific instructions are therefore contained in Chapter 5 (Insight environment) and Chapter 8 (standalone mode).

You need not remain in one mode as you proceed through all four steps. For example, you could (1) set up a molecular system using one or more modules within the Insight program; (2) set up the calculation parameters with the DMol module of the Insight program, then perhaps refine the input files after quitting the Insight environment, by editing them with a text editor; (3) start the run from the Insight interface or from the operating system, whichever is more convenient or appropriate; and finally (4) read the results into the Insight program for analysis in the Analysis, DMol, and other modules. Alternatively, you could perform Steps 1 and 2 both with the dialog interface to the standalone mode of DMol, start the run from the operating system using dmol_master, and use the Insight interface only for some analysis of the results.

For additional information on constructing and modifying molecules, you can refer to the documentation for the Insight program. Several computational products are available that you may want to use to refine a molecule before submitting it to a DMol run, including several packaged as modules or pulldowns within the Insight program, as well as separately licensed products such as the Discover and Search_Compare programs.




Last updated September 25, 1997 at 03:13PM PDT.
Copyright © 1997, Molecular Simulations, Inc. All rights reserved.