Introduction to Computational Chemistry

By Benjamin Lynch and Patton Fast June 14, 2005

A few important areas

Ab initio calculations Quantum dynamics Nanoparticles Biochemistry

Ab initio calculations
ab initio is Latin for from the beginning We begin with fundamental physical properties, and we calculate how electrons and nuclei interact. Most often this requires solving approximations to the time-independent Schdinger equation. Occasionally we need to solve the time-dependent Schdinger equation (this will be covered later when we get to quantum dynamics)

 What can we predict with modern Ab Initio methods?

Geometry of a molecule Dipole moment Energy of reaction Reaction barrier height Vibrational frequencies IR spectra NMR spectra Reaction rate Partition function Free energy Any physical observable of a small molecule

Ab initio calculations
The underlying physical laws necessary for the mathematical theory of a large part of physics and the whole of chemistry are thus completely known, and the difficulty is only that that the exact application of these laws leads to equations much too complicated to be solvable.
P.A.M. Dirac

The challenge in computational chemistry is to simplify the calculation enough to be solvable, but still accurate enough to predict the desired physical quantity.

 How accurately can we predict those quantities? Thats where you come in
The best compromise of accuracy and computational cost requires a skilled computational chemist at the helm. There is an enormous toolbox of theoretical methods available, and it will take skill and creativity to solve real-world problems.

Electronic Structure Theory

This is where we describe the motion of the electrons
First we decide on a mathematical basis to describe our electronic wavefunction.

A sum of atom-centered Gaussian functions is the most common approach.

$ = ! ci e
i

"# i r 2

Electronic Structure Theory

A plane-wave basis set is a common choice for predicting properties of a crystal Other functions can be used to describe the wavefunction, however most other forms have severe limitations.

$ = ! ci sin( 2#i" )
i

$ = ! ci e
i

"# i r

Choosing a basis set

Large basis sets have more freedom to properly describe the electronic state of the system. Large basis sets can drastically increase the cost of a calculation, and occasionally they introduce new numerical challenges.

Electronic Structure Theory

Level of theory
The level of theory is the set of approximations used to simplify the complex motion of the electrons.

Common methods
HF, MP2, MP4, CISD, CCSD, CCSD(T)

Electronic Structure Theory

Density functional theory (DFT)
DFT is an alternative to wavefunction-based or explicitly correlated methods. The correlation energy and the exchange energy is expressed as a functional of the electron density. The fitting of the functional is often done with empirical data. Some density functionals can be considered ab initio because they do not fit to empirical data.

Common methods
B3LYP, MPW1K, PBE0

How calculations are run

Input files describe the system, basis set, and the level of theory System requirements will also sometimes appear in the input files
%NPROC=2 %MEM=64MB #MP2/6-31G* This is a Gaussian Input File 0 1 O H 1 r1 H 1 r1 2 a1 r1=0.90 a1=115.0

How calculations are run

Command line execution

blynch % blynch %

g03 myinput.inp myoutput.out vi myoutput.out

How calculations are run

GUI Graphical User Interfaces are sometimes useful.

After we have an electronic wavefunction

After we have solved for a wavefunction for our system, we can extract and display many properties. To the right is a plot of a molecular orbital. These can be used to better understand the nature of chemical bonding in various systems.

Geometry optimization
Once we have a theory and a basis set, we can start predicting energies. We can calculate the first derivative of the energy with respect to the nuclear coordinates to determine if the system is at a local minimum. We can continue to change the position of the nuclei to minimize the energy. Local minima and saddle points often correspond to important structures.

Geometry optimization
After optimizing a geometry, we can predict experimental structures.

Potential Energy Surfaces

This is a picture of the potential energy as a function of 2 coordinates

Born-Oppenheimer Approximation
The potential surface is a Born-Oppenheimer potentials surface, where the potential energy is a function of geometry. Motion of the nuclei is assumed to be independent of the motion of the electrons

These are all Stationary points

Transition state theory can be used to calculate reaction rates

Cost of calculations
The cost of a calculation will determine how large of a system can be simulated. The scaling properties of a method are often expressed as NX where N is the number of atoms in the system (of the same type). Hartree-Fock theory (HF) scales as N4. Doubling the size will increase the cost by a factor of 16. CCSDTQ scales as N10. Doubling the size will increase the cost by a factor of 1024.

Programs
GAUSSIAN GAMESS Molpro NWCHEM Columbus Many others

 Sometimes we cannot accurately describe a system with the Born-Oppenheimer approximation. The motion of the nuclei may be coupled to the motion of the electrons The following techniques are commonly applied to photochemical reactions.

Quantum Dynamics
Some methods propagate wave packets and swarms of trajectories are used to gain insight into reaction rates. These methods are incredibly expensive, and are limited to ~3-6 atoms.

Quantum Dynamics
Some less-expensive alternatives have also been developed. One technique that can be used is called surfacehopping A potential surface is mapped out for 2 (or more) surfaces which have a small separation in energy.

 These problems cannot be solved with black box software. Most software is developed by the research groups who use it. Fitting the surfaces requires optimizing hundreds of parameters, sometimes with genetic algorithms.

Nanoparticles
Nanotechnology is of increasing interest Properties of nanoparticles can be expensive to determine experimentally Applications include electronics, catalysis, propellants, explosives, medicine, photonic materials,

What is a nanoparticle?
A particle with dimensions between 109 m and 106 m A cluster of 43 aluminum atoms is ~1 nm in diameter. Nanoparticles have properties that range from atom-like to bulk-like

How are such large systems handled?

A different class of methods are required to handle such large systems. Very simple theories are used to describe the electronic structure.
Usually semi-empirical methods

Nanoparticles
Calculations on an infinite crystal can be used to validate a theory The calculated properties can be compared to macroscopic physical properties

Nanoparticles
Periodic boundary conditions and planewave basis sets can be used for an infinite crystal. Simple, parameterized electronic structure methods can be used for more irregular structures.

Nanoparticles
Localized electronic structure methods are very useful.
Interactions beyond a certain distance can be ignored. These methods scale better than more complete methods.

More challenges
Many nanomaterials have a high fraction of metals. Metals, in general, are quite difficult to simulate.

How are such large systems handled?

This is a hot area of research Many new methods are required
Methods to determine reaction paths
Many similar reaction channels may exists

Methods that can handle main-group elements as well as metals Methods that can predict electronic properties of metal nanoparticles Methods that can accurately predict thermochemical properties

Challenges in simulating large systems

Many similar reaction channels may exist for a single reaction Long-range electrostatics can have a high computational cost. Surrounding medium may play an important role (is this nanoparticle in water?) All of these challenges are also present in biochemistry problems

Programs
VASP Crystal Gaussian and other ab initio packages TB and other tight binding codes

Biochemistry

Molecular Mechanics
Ball and Spring Modeling Useful for Molecular systems containing thousands of atoms Organics, oligonucleotides, peptides, etc. Vacuum, implicit solvent, explicit solvent environments Ground state Thermodynamic and kinetic properties (molecular dynamics)

Molecular Mechanics
Ball and Spring Modeling Principle assumptions Nuclei and electrons are lumped into balls (atom-like particles) Balls are spherical and have a net charge Interactions are based on springs and classical potentials Interactions are preassigned to specific sets of atoms Interactions determine the spatial distribution of the balls and their energy

Molecular Mechanics
Ball and Spring Modeling

from the NIH Guide to Molecular Modeling

Energy = Stretching + Bending + Torsion + Non-Bonded

Molecular Mechanics
Ball and Spring Modeling

E = E bond + E angle + E dihedral + E vdw + E elec

2 1 = # K b (b " b0 ) 2 bonds 2 1 + # K$ ($ " $ 0 ) 2 angles 2 1 + # K% [1+ cos( n% ")] 2 dihedrals

4 .( ' +12 ( ' + 6 18 6 4&0* - " * - 36 6 /) r , ) r , 26 + # 5 9 6 non"bonds 6 q q 1 2 6+ 6 7 Dr :

(short range)

(long range)

QM/MM

E = QM + QM/MM + Polarization + Boundary

QM/MM
NADPH QM/MM boundaries

N5-preprotonated DHF Pu, Ma, Gao, Truhlar, J. Phys. Chem. B, 109(18), 2005.

Structure Determination
X-ray Crystallography

Protein preparation and analysis

Protein Crystallization

Diffraction Data

Model Building and Structure Refinement

Phase Estimation

 Ben Lynch
(612) 624-4122

Patton Fast
(612) 625-6573

MSI Helpline
(612) 626-0802