Dropping The Drop Boxes: Current “Best Set” Keywords For Solid-State Terahertz DMol3 Calculations
Tuesday, October 30th, 2007Of the few discussions I’ve had with fellow theoreticians, the general consensus is that employing drop boxes for defining parameters in quantum chemistry programs is methodologically akin to using solvents and starting materials as purchased from a chemical company “as is.” Certainly good enough to get something decent to happen because, of course, the companies providing the product need to stay in business, but certainly not ideal and, by their removal of the background steps and understanding of the generation methods, somehow lacking, neglecting the efforts of researchers “behind the scenes” that made the (pardon) gross simplification of “hitting the dirt running” possible.
The subject of this post is the Materials Studio (graphical interface) implementation of DMol3 (DOS/shell) and the settings available via the MS interface (coarse, medium, fine in many cases, with the actual values associated with these general settings varying slightly with the properties of the crystal cell). Are the standard options available for setting job parameters good enough to get a workable result out? In the case of “fine”, yes, with very few instances noted here of calculations behaving badly. Can one do better? Definitely. For terahertz spectroscopy studies alone, the parameter options provided by the Materials Studio interface are good enough to get something out to aid in assignments, and certainly good enough to convince anyone that ISOLATED-MOLECULE CALCULATIONS ARE ENTIRELY INAPPROPRIATE TO USE FOR THE ASSIGNMENT OF SOLID-STATE TERAHERTZ SPECTRA. At this point in the game, however, I would not consider them rigorously publication-quality results (I would not argue if agreement was good and the spectra clean, but I would certainly check the level of theory carefully), given that agreement can be improved and a better overall assignment achieved by tweaking the keyword list.
It is with that in mind (and Matt Hudson’s reminder) that I’m posting a keyword set that, over the course of a few hundred DMol3 calculations or so, has been found to provide a decent balance of time (longer than the menu options but only by %15 – %20 or so) and interpretive power (or predictive power if no one’s taken the Terahertz spectrum). Note that the “#” is the “comment/ignore” character, so much of the content below is ignored by DMol3.
#################################################################### # # Input cheat-sheet for THz DMol3 solid-state calculations # Quests, comms, complaints, Damian Allis, damian@somewhereville.com # # Definitely works with DMol3 version 3.2 # #################################################################### ### Optimization Properties #################################### #################################################################### Calculate optimize ### Optimize the structure ### Opt_energy_convergence 5.0000e-007 ### Energy convergence (dE step to finish) ### Opt_gradient_convergence 1.0000e-004 A ### Conv when largest grad vector comp < this ### Opt_displacement_convergence 1.0000e-004 A ### Conv when largest atom disp < this ### Opt_iterations 100 ### Steps to opt (large value for tight conv) ### Opt_max_displacement 0.3000 A ### Max length geom update vector ### #################################################################### ### Electronic Structure Descriptions ########################## #################################################################### Spin_polarization restricted ### un/restricted wavefunction description ### Charge 0 ### System charge ### ### Basis set approx's are not direct comparisons Basis dnp ### Basis dnp Gaussian approx 6-31G(d,p) basis set ### Basis dnd Gaussian approx 6-31G(d) basis set ### Basis dn Gaussian approx 6-31G basis set ### Basis min Gaussian approx 3-21G basis set ### Pseudopotential none ### Consider for transition metal systems ### ### GGA Functionals - Gen Grad Approx: density and gradient Functional bp ### So far, bp is the best all-around THz freq functional ### Functional blyp ### ### Functional bop ### ### Functional gga(p91) ### ### Functional hcth407 ### ### Functional vwn-bp ### So far, vwn-bp is the 2nd best THz freq functional ### Functional rpbe ### ### Functional pbe ### ### ### LDA Functionals - Local Den Approx: density at position ### Functional pwc ### ### Functional vwn ### ### #################################################################### ### Additional Electronic Structure Parameters ################# #################################################################### ### ### Integration_grid - mesh points for numerical integr procedure ### value ipa iomax iomin thres rmaxp sp ### Integration_grid xcoarse ### 6 3 1 0.01 10.0 1.0 ### Integration_grid coarse ### 6 4 1 0.001 10.0 1.0 ### Integration_grid medium ### 6 6 1 0.0001 10.0 1.0 Integration_grid fine ### 6 6 1 0.00001 12.0 1.2 ### Integration_grid xfine ### 6 7 1 0.000001 15.0 1.5 ### Aux_density octupole ### Max ang momentum multipolar fit funcs ### Occupation fermi ### Converger aid. See manual for more ### Cutoff_Global 4.0000 angstrom ### Atom-centered basis set cut-off distance ### Scf_density_convergence 1.0000e-008 ### Conv when density conv < this ### Scf_charge_mixing 0.2000 ### Init charge density mix coeff restrict ### Scf_iterations 100 ### Max iters for SCF (large b/c conv’rs) ### Scf_diis 6 pulay ### Max size of DIIS subspace for SCF calc ### ### Kpoint -> defined for each a,b,c lattice vector ### General protocol (appropr of selections is theological): ### If a,b,c < 5 set kpoint to 5 ### If a,b,c 5 < X < 10 set kpoint to 4 ### If a,b,c 10 < X < 15 set kpoint to 3 ### If a,b,c 15 < X < 20 set kpoint to 2 ### If a,b,c < 20 set kpoint to 1 ### a b c Kpoints on 5 5 5 ### Max set here. Adjust as approp ### ### Print options Print vib_hess ### amount of printout. Read manual. ### #################################################################### ### Terahertz Vibrational Analysis ############################# #################################################################### ### Symmetry on ### Recalc symm after opt for use in freq ### Mulliken_analysis charge ### Mulliken charges. MS 4.2 broke this (!) ### Hirshfeld_analysis charge ### Hirshfeld charges. ### Frequency_analysis on ### Perform normal mode analysis ###
Either copy+paste the above, or download DMol3_THz_complete.input.
For those a bit overcome by what Frank Zappa would refer to as the “statistical density” of the content above, the whole file reduces down to the following (the selection of Kpoints being the only keyword one has to think about below).
Calculate optimize Opt_energy_convergence 5.0000e-007 Opt_gradient_convergence 1.0000e-004 A Opt_displacement_convergence 1.0000e-004 A Opt_iterations 100 Opt_max_displacement 0.3000 A Spin_polarization restricted Charge 0 Basis dnp Functional bp Integration_grid fine Aux_density octupole Occupation fermi Cutoff_Global 4.0000 angstrom Scf_density_convergence 1.0000e-008 Scf_charge_mixing 0.2000 Scf_iterations 100 Scf_diis 6 pulay Kpoints on 5 5 5 Print vib_hess Symmetry on Mulliken_analysis charge Hirshfeld_analysis charge Frequency_analysis on
Either copy+paste the above, or download DMol3_THz_min.input.
Yes, I am aware that I am recommending the replacement of one “standard” set of keywords (Accelrys) with another “standard” set (mine). Further, it must be noted that these above keywords are very much terahertz-specific, where the lowest-energy, largest-amplitude, most neighboring molecule-dependent solid-state properties are considered. This is NOT the “be all, end all” keyword set for every system (certainly not even for general inelastic neutron scattering spectroscopy use, where the agreement between theory and experiment is very dependent on the choice of density functional (blyp and bop seem to be the best in the 500 to 1500 cm-1 region, by the way)). And, far from trying to use this post to complain about theory, program, and parameters, I note that DMol3 has been the mainstay of my solid-state terahertz theoretical work to date and I’ve spoken quite favorably of it in the past. Any program with keywords and input parameters should be expected to have optimal settings for different tasks, a point in no way lost on researchers familiar with the variety of empirically-derived functionals in density functional theory. We don’t refer to the computational chemistry implementations of quantum theory as the “approximate methods” for nothing.
Since it is just a text file, settings above may undergo further refinement at a more rapid pace than a commercial interface. That said, if your goal is reasonable terahertz simulation with as standard a set of keywords as possible, the set above will (so far, anyway) bring you quite close to your destination.
It is probably far less interesting to list my explanations for each choice (like the Kpoints settings, functional choice, the super-tight convergence criteria) than reading any comments you (the reader) might make, so I invite/encourage questions in the comment section that will keep record of choices, my reasons, and the logic of others.
en.wikipedia.org/wiki/Quantum_chemistry
www.accelrys.com/products/mstudio
www.accelrys.com/products/mstudio/modeling/quantumandcatalysis/dmol3.html
en.wikipedia.org/wiki/Time_domain_terahertz_spectroscopy
en.wikipedia.org/wiki/Frank_Zappa
www.zappa.com
en.wikipedia.org/wiki/The_Black_Page
www.accelrys.com
en.wikipedia.org/wiki/Inelastic_neutron_scattering
www.accelrys.com/reference/cases/studies/korter.pdf
en.wikipedia.org/wiki/Density_functional_theory
