Charge Analysis: Atoms in Molecules

Size: px

Start display at page:

Download "Charge Analysis: Atoms in Molecules"

Shana Anderson
6 years ago
Views:

1 Daubechies Wavelets in Electronic Structure Calculation: BigDFT Code Tutorial CECAM - GRENOBLE : Atoms in Molecules Ali Sadeghi Basel University 21 November 2011

2 An output of electronic structure calculations like DFT is the electronic density n(r) = ψ(r) 2 which is a continuous quantity. partitioning between atoms Q: where are atoms in molecules/bulks? Main-product: charge exchange between bonded atoms ideal ionic bond: 1e transfered (q = ±1) ideal covalent bond: 2e shared (q = 0) otherwise? Side-product: decomposed quantities atoms A = A

3 Atomic Domains Atomic domains? No real & clear boundaries between atoms! Many possible definitions & methods Still a useful tool for: bonding analysis partial charges partial multipoles atomic DOS... s-owl.cengage.com

4 No quantum-mechanical considerations Not using electronic structure Using atomic wave-functions ψ (r) Using the real-space electronic density n(r) Approaches Partitioning the space q = Z V n(r)d 3 r Partitioning the charge density q = Z n (r)d 3 r

5 Non QM-based methods: Using experimental data charges, dipoles,... Voronoi polyhedra assign nearest atom to each point q = Z V n(r)d 3 r solely mathematical! atomic types not considered possible unreasonable results!

6 Non electronic structure based methods ESP fitting method Point charges as effective atomic charges q Fitting q to reproduce electrostatic potential (ESP) Least-squares minimization with constraints total charge total dipole... M sampling i ( 2 Natom V q ( r i ) V o ( r i )) + λ( N atom q q tot ) Sampling points r i only out of atomic regions e.g. r vdw Small RMS is not guaranteed! D. Chen et al JPC. A, 114, 37 (2010)

7 Electronic structure based methods Using atomic wave-functions ψ (r) easy for LCAO basis :) dependency on basis set :( unreasonable q without orthogonal basis (Lowdin) The most common method: Mulliken population analysis projecting ψ(r) on atomic basis n (r) = Ψ 2 q = Z n (r)d 3 r orbitals population analysis old and simple

8 Electronic structure based methods Voronoi Deformation Density Space is partitioned by Voronoi polyhedra Use electronic density n(r) WRT density of non-interacting atoms n o n o i.e. the deformation density n = n q = V nd 3 r In addition to n, it uses both: geometry atomic orbitals

9 Electronic structure based methods Weighted distributing Distribute n(r) according to atomic weights w (r) n (r) = w (r)n(r) weights are normalized: w (r) = 1 fuzzy boundaries (if w are not truncated) q = Z w (r)n(r)d 3 r Simple Gaussian w (r) e r r 2 /r 2,o r,o = covalence, ionic, vdw,... radius Hirshfeld w (r) n o (r) q = (n o w n)d 3 r = n o (1 n(r) n o )d 3 r Non-interacting atoms are important: either atomic radius or electronic density

10 Electronic structure based methods Bader (AIM) Only use real-space density n(r) sharp boundaries! q = Z V nd 3 r Use n-topology to find V : n is maximized on nuclei curvatures=eigenvalues of 2 n critical points: n = 0 zero-flux surfaces are boundaries: n u = 0

nuclei positions the space is partitioned to some basins: ideally, one

11 Bader Topology of electronic density n(r) helps: Tracking Gradient of n n shows the steepest ascent direction at any point maxima in n coincide with nuclei positions the space is partitioned to some basins: ideally, one Bader volume for each atom

12 Bader Grid-based tracking of n A fast, robust and linear scaled numerical algorithm: Follow steepest ascent path along n to next grid point repeat until an n maximum is reached (a nuclei is found) assign all points in the path to this maximum repeat for any unassigned grid point if previously assigned points are reached, terminate the path and group them together (linearity) refinement needed at boundaries might fail for smooth pseudopotential (shallow maxima)

Bader The Henkelman s group implementation is already used as a post-processing tool in different packages including BigDFT can be used for periodic or

13 Bader The Henkelman s group implementation is already used as a post-processing tool in different packages including BigDFT can be used for periodic or molecular systems: B atoms in Si crystal: Water molecule: q B1 = 0.9e q B3 = 1.5e q H = q O =

14 Multipole Moments charge: q = dipole: P = q = (Z ( P chg quadrupoles : Q i,j = nd 3 r) V + P ) = (Z r Q i,j q = 0 P is origin-independent. If P 0 Q i,j are origin-dependent! q 0 P depends on origin! chg P + P is origin-independent P : from intra-atomic polarization chg P : from cores & charge-transfer V rnd 3 r) N atoms Could atomic contributions be defined? A = A Yes, but with special care!

15 with BigDFT analysis with BigDFT is possible in two ways: 1 On-the-fly: Mulliken population analysis set InputPsiId=10 in input.dft Mulliken Population Center No. Shell Rad (AU) Chg (up) Chg (down) Net Pol Gross Chg 1 s Center Quantities : s px py pz Center Quantities : Post-processing: Bader or Veronoi charge analysis set Output_Grid=1 in input.dft to get charge density use bader tool with the charge density bader [-c voronoi] [electronic_density.cube] atom# CHARGE: core electronic net

Bader Mulliken Gaussian Voronoi O -1.2-0.8-1.3 +0.7 H +.0.6 +0.4 +0.65-0.

16 Comparison Water molecule with BigDFT: Comparing to method (AIM), Mulliken underestimates Gaussian overestimates Voronoi is not trustable! Bader Mulliken Gaussian Voronoi O H For a series of organic molecules: q Hirshfeld < q Mullekin < q Bader (D. Proft et al. J. Comp. Chem. 23, 12 (2002) )

17 Several existing definitions/methods: either partition space or charge between atoms use molecule geometry, electrostatic potential, atomic orbitals/radii, electronic density (n),... are not completely consistent Atoms in a molecule can not (always) be clearly defined! method, based on n-topology, seems more reliable Grid method, a computational method, is efficient, robust, and linearly scaled with grid size; the code is freely available Care should be taken in defining atomic contributions to an observable A = N A one can do Mulliken, Bader, Voronoi and Gaussian charge analysis - Thank you for your attention!

Poisson Solver, Pseudopotentials, Atomic Forces in the BigDFT code

Poisson Solver, Pseudopotentials, Atomic Forces in the BigDFT code CECAM Tutorial on Wavelets in DFT, CECAM - LYON,, in the BigDFT code Kernel Luigi Genovese L_Sim - CEA Grenoble 28 November 2007 Outline, Kernel 1 The with Interpolating Scaling Functions in DFT for Interpolating