Ashwin Inala

NCSSM Online

North Carolina School of Science and Mathematics

Durham, North Carolina

February 25, 2017

Abstract:

Density functional theory is a computational technique of using elec-

tron density, not wavefunction, to predict properties of molecular systems. Over the

years, many scientists, such as Pierre Hohenberg, Erwin Schrodinger, and Walter

Kohn, have contributed to the fast growth of density functional theory. To find the

energy of system, scientists must approximate the exchange-correlation functional.

The various types of functionals that scientists use to predict the exchange-correlation

functional fall under three main categories: local density approximations, gradient-

corrected functionals, and hybrid functionals. From the review, it was found that

the PBE gradient-corrected functional and the PBE0 hybrid functional provide ac-

curate approximations for a variety of properties of a system. Overall, local density

approximations should not be the functional of choice to approximate the exchange-

correlation functional.

Key words:

Density Functional Theory, Local Density Approximations, Gradient-

Corrected Functionals, Hybrid Functionals, Electron Density

Introduction

The density functional theory is now used by scientists around the world to determine the character-

istics and properties of various molecules and atoms based on the electron density of the molecule.

Some of these properties include the electronic structure of orbitals, atomic reactivity, and even the

UV-Vis spectra. Unlike the wavefunction, electron density can be measured. Density functional

theory has turned into the most popular way of computationally finding molecular properties for

many reasons. The method has high structure accuracy, does an impressive job figuring out Gibbs

free energy values for certain reactions, is not as difficult to run computationally as compared to

other methods, and takes into account electron correlation, which provides an accurate represen-

tation of a system’s energy. The interaction that takes place between certain electrons in a give

electronic structure is known as the electron correlation. Electron correlation can be used to find

the discrepancies of the Hartree-Fock model. The Hartree-Fock method only finds repulsion en-

ergy as an average over the entire molecular orbital, which is not accurate. The correlation energy

is determined by finding the difference between the true energy and the Hartree-Fock energy. The

idea that electrons are avoiding one another is known as dynamical correlation. One of the issues

with density functional theory is that the scientist has to choose which DFT method to use for

certain applications.

Two scientists have had a significant influence in popularizing the density functional theory in

theoretical science: Pierre Hohenberg and Walter Kohn. They stated that the energy of the system in

question could be found using the electron density functional, which is shown below (a functional

is a function of a function).

sents the kinetic energy of the non-

interacting system, J[p] represents the electrostatic distribution of the charge, and Exc[p], known as the exchange-correlation functional, represents the information that is not known about the system.

Currently, scientists are approximating this exchange-correlation functional by trying out various functional methods, which is the topic of this paper.

Pierre C. Hohenberg, born in 1934 in France, is a world-renowned theoretical physicist. Earning

his bachelor’s degree, master’s degree, and doctorate from Harvard University, Hohenberg began

working at Bell Laboratories in Murray Hill, where from 1989 to 1995 he became the directer of

the department of theoretical physics. Hohenberg has a unique fondness for his home country, as he

was guess professor in Paris in 1963, 1964, and 1988. 2004 saw him advance to Senior Vice Provost

of Research at NYU [1], but he soon stepped down to be a professor in the Physics Department.

Hohenberg was a leading figure outside academia as well, as he was on the human rights committee

of the New York Academy of Sciences and a proud member of the National Academy of Sciences

and American Academy of Arts and Sciences. His theorems, with Walter Kohn, in a paper written

in the Physical Review would give birth to the density functional theory we have today.

Walter Kohn, born in 1923 in Austria, is one of the most important people in the area of theoretical

chemistry. After receiving a PhD from Harvard, in the field of physics, Kohn would go on to

become a physics professor at Carnegie Mellon University, University of California at San Diego,

and University of California at Santa Barbara [2]. Like Hohenberg, Kohn had a large impact outside

of academia, as he was a member of the National Academy of Sciences and American Academy

of Arts and Sciences. He is, however, best known for winning the 1998 Nobel Prize in Chemistry,

who he shared with physicist John A. Pople. He was recognized with the prestigious award for his

”development of the density functional theory”. He passed away last year due after his battle with

jaw cancer.

Before the discovery of the density functional theory, scientists worked to find the wavefunction

of a system, as with it they could completely describe the properties of that molecular system.

Every single particle of an atom is given a specific wavefunction, which gives all the calculable

information about the particle. The Schrodinger equation, in layman’s terms, predicts the behavior

of a dynamic system in the future. It represents how a wavefunction of a system changes and

evolves over time. In order to find the wavefunction in the Schrodinger equation, the computer

needs initial numbers, which are found using basis set and basis functions. In the Schrodinger

equation, which is shown below, the H stands for the Hamiltonian.

the Hamiltonian acts as an operator, just like how

the plus sign and minus sign work. The energy (E)

of the system can be found by performing the Hamil-

tonian on the wave-function (also called psi and symbolized as). The Schrodinger equation is known as an eigenfunction. In an eigenfunction, the operation that takes place on the left-hand side of the equation results in the same operation times some constant value on the other side of the equation. In the Schrodinger equation, the constant value is the energy of the system.

Of course, there were issues surrounding solely using the wavefunction to provide information

about a certain system. First of all, the wavefunction is not something that can be physically

analyzed or quantified in a laboratory setting. It is also a function that is confusing, due it having

multi-dimensions. In addition, only the square of the wavefunction has any physical value, as it

gives information about the probability density. The probability density is made up of numerous

variables, adding to the complexity of this technique. There are other ways of making predictions in quantum mechanics, including path integral formulation and matrix mechanics.

Erwin Schrodinger, born in 1887 in Austria, made significant contributions to quantum mechanics.

Earning a doctorate from the University of Vienna in 1910, he later went on to serve in the military

during World War I. At the University of Zurich in 1926, Schrodinger published his work that

would set the stage for the rise of quantum wave mechanics. He stated that matter particles, in

some circumstances, behave like waves. His equation, the Schrodinger equation, outlines how a

wave equation can predict the behaviour of a certain system. Schrodinger argued that his equation

could also determine measurable energies of a system. Not surprisingly, many physicists around

the world did not believe Schrodinger’s work and were very reluctant to accept his theory. His

most well known objection has become known as Schrodinger’s cat, which states a that poisonous

flask, a cat, and a radioactive source are placed in a box. In the event of a decaying atom, the

cat is killed due to the release of the poison. When someone opens the box, the cat will be either

alive or dead, not both dead and alive, supporting Schrodinger’s claim that questions the validity

of quantum superposition.

Schrodinger would go on to win the 1933 Nobel Prize in Physics, along with Paul Adrien Maurice

Dirac, ”for the discovery of new productive forms of atomic theory” [3]. His novel

What is Life? was one of the earliest attempts to explain genetic structure and added new interest into the field.

The last true Renaissance man, Schrodinger, throughout his illustrious life, changed science into

what it is today.

William Rowan Hamilton, born in 1805 in Ireland, is well known for his work that paved the road

for the rise of quantum mechanics and quantum relativity theory. At Trinity College, he was Profes-

sor of Astronomy, the last position he had before he went to Dunsink Observatory [4]. Hamilton’s

tweaking of classical mechanics gave scientists new ways to use motion equations.

In this paper, I take a look at the various functionals used to approximate the exchange-correlation

functional, which finishes the electron density functional that provides us with the energy of the

system in question.

**Local Density Approximations**

Focused only on the electron density at a certain point in space, local density approximations are

a simple type of exchange-correlation functional. Hohenberg and Kohn first came up with local

density approximations in their density functional theory paper, making this functional type one of

the oldest out there. To make use of local density approximations, the exchange-correlation energy

of a electron gas at a certain density must be found [5]. These approximations tend to work, as small

errors in correlation and exchange energy densities cross each other out. In Marcel Swart’s 2016

density functional theory poll, local density approximations was ranked the eighth most popular

functional in the first division. Local density approximations are regarded as the most important

functional in chemistry.

The Exc(p(r)) part inside the integral is the exchange-correlation energy of a single particle with elec-

tron gas density of p(r). The exchange-correlation

energy of this single particle is then weighted with the

probability p(r), which takes into account that an electron exists in this place in a system. The exchange-correlation energy can be further broken down into exchange and correlation terms, as shown below.

Exc is the exchange- correlation energy, while Ex is the electron exchange term and Ec is the electron correlation term. The electron correlation term deals with how one electron in an atom interacts with another one. The electron exchange term deals with the exchange of bosonic and fermionic elec-

trons.

**Gradient-Corrected Functionals**

Local density approximations, and their low to average accuracy, are not useful for many scien-

tific applications. Therefore, for many years, the field of computational chemistry was never im-

pacted by density functional theory. Soon gradient-corrected functionals came along, also known

as generalized gradient approximation functionals (GGA). These functionals are sometimes called

non-local functionals, which is not true since gradient and electron density provide only local in-

formation about a system. A gradient-corrected approximation solely relies on local density and

the gradient of it [6]. The gradient-corrected approximation is an improvement over local density

approximations since information of the gradient of the charge density is known. A gradient, in

mathematics, calculates the rate of change of a property in question. Gradient-corrected function-

als aim take into consideration the inconsistencies of electron density. These type of functionals

also have lower error in energy through bond dissociation (the change in enthalpy when a bond

is broken through homolysis) and have improvements in transition-state barriers. Functionals that

take into account kinetic energy density and second-order gradients, fall under meta-generalized

gradient approximation functionals.

**BLYP**

A type of gradient-corrected functional and popular with chemists, BLYP has a Becke88 exchange

functional and a Lee, Yang, Parr correlation functional, as shown below.

accurate energy values. However, when dealing with organic systems, BLYP is not very accurate.

BLYP should especially not be used to find these molecular properties: excitation energy, spin state

splittings, and chiroptical properties.

**BP86**

A type of gradient-corrected functional, BP86 has a Becke88 exchange functional and a Perdew

correlation functional. In Marcel Swart’s 2016 density functional theory poll, BP86 was ranked the

seventh most popular functional in the first division. BP86 accurately predicts molecular geome-

tries, transition metals, and relativistic elements. Scientists don’t use it when trying to find spin

state splittings or anything related to nuclear magnetic spectroscopy.

**PBE**

Another type of gradient-corrected functional, PBE has a Perdew, Burke, Ernzerhof exchange func-

tional and a Perdew, Burke, Ernzerhof correlation functional.

In Marcel Swart’s 2016 density functional theory poll, PBE was ranked the most popular functional

in the first division. There is a reason why this functional is so popular, as it very accurate in predicting reaction barriers, transition elements, main group elements, relativistic elements, and properties

related to nuclear magnetic spectroscopy. However, PBE should not be used to find excitation

energies.

**PW91**

A type of gradient-corrected functional, PW91 has a Perdew-Wang exchange functional and a

Perdew-Wang correlation functional. With various functionals in his name, it is easy to see the

massive role Dr. Yang has played in the development of the density functional theory (picture

provided below). In Marcel Swart’s 2016 density functional theory poll, PW91 was ranked the

eleventh most popular functional in the first division. To form the hybrid functional B3LYP, the

PW91 correlation functional gets replaced by the Lee-Yang-Parr functional.

**Hybrid Functionals**

For those in the computational chemistry field, hybrid functionals are the most popular and com-

mon type of DFT method. They are even sometimes known as ACM functionals, which stands

for adiabatic connection method. Hybrid functionals are very accurate at predicting the change

of Gibb’s free energy in a system and at finding a geometry optimization but are more computa-

tionally costly than local density approximations and gradient-corrected functionals. These type

of functionals bring together ab initio methods (majority of the time Hartree-Fock methods) and

density functional theory mathematics. In other words, a hybrid functional is a functional that

has electron correlation, exchange energy with a density functional theory approximation, and ex-

change energy with a Hartree-Fock approximation all combined together [7]. Ab initio methods

are those in which Schrodinger equation is used to create the entire model mathematically. More

specifically, Hartree-Fock methods take into consideration the average effect of correlation of one

electron on another electron in a system. Hybrid functionals usually overpredict bond lengths of

molecules.

**B3LYP**

In Marcel Swart’s 2016 density functional theory poll, B3LYP was ranked the third most popular

functional in the first division. For the majority of systems, the B3LYP/6-31G model chemistry

will get the best approximations of a certain system, especially those involving organics. This

functional approximates properties related to geometries really well [8]. B3LYP should not be

chosen as a functional if one is trying to approximate the excitation energies, reaction barriers,

transition metals, and relativistic elements of a system. The B3LYP exchange-correlation func-

tional, a mixing scheme of 3 parameters, is ”made” from pieces of the Hartree-Fock exchange

functional, local spin density approximation exchange functional, Becke88 exchange functional,

Lee-Yang-Parr correlation functional, and Vosko-Wilk-Nusair correlation functional.

**PBE0**

In Marcel Swart’s 2016 density functional theory poll, PBE0 was ranked the second most popu-

lar functional in the first division. Like the PBE functional, the PBE0 functional does a great job

approximating numerous properties of a system, such as reaction barriers, excitation energies, ge-

ometries, transition metals, chiroptical properties, and much more. The functional combines parts

of the Hartree-Fock exchange functional, PBE exchange functional, and modified PW91 correla-

tion functional.

**BHandH**

In Marcel Swart’s 2016 density functional theory poll, BHandH was ranked the fifth most popular

functional in the second division. Pieces of the Hartree-Fock exchange functional, local density

spin approximation exchange functional, and Lee-Yang-Parr correlation functional are needed to

create the BHandH functional. Scientists veer away from this functional when approximating

properties such as hydrogen bonds, main group elements, transition metals, and relativistic ele-

ments.

**Conclusions**

As the search for the best functional to approximate the exchange-correlation functional progresses,

it can be said that the easiest way to get a PhD in chemistry is to create a better functional than

those that exist today. From the information in the previous section, it can be seen that the two best

functionals are the PBE gradient-corrected functional and PBE0 hybrid functional, due to their

versatility to find numerous properties of a system. Whenever in doubt about what functional to

use for a given system, the B3LYP hybrid functional, as well as the two functionals stated above,

should be chosen. Keeping in mind that each functional class has certain applications and scenarios

it works best in, in general, scientists should stick away from local density approximations, due to

their primitiveness and by the availability of more useful functionals. As density functional theory

continues to grow and become more and more popular, the field of computational chemistry will

soon be pushed into the spotlight.

**Acknowledgements**

The author thanks Mr. Robert Gotwals for his assistance with this paper.

**References**

[1] ”Prize Recipient.” Prize Recipient. Web. 10 Jan. 2017.

https://www.aps.org/

programs/honors/prizes/prizerecipient.cfm?first_nm=Pierre&

last_nm=Hohenberg&year=2003

[2] ”Walter Kohn - Facts.” Walter Kohn - Facts. Web. 11 Jan. 2017.

https:

//www.nobelprize.org/nobel_prizes/chemistry/laureates/1998/

kohn-facts.html

[3] ”Erwin Schr

?

odinger - Biographical.” Erwin Schr

?

odinger - Biographical. Web. 11 Jan. 2017.

http://www.nobelprize.org/nobel_prizes/physics/laureates/1933/

schrodinger-bio.html

[4] Wilkins, David. ”Sir William Rowan Hamilton.” Encyclopædia Britannica. Encyclopædia

Britannica, Inc., 02 Aug. 2002. Web. 15 Jan. 2017.

https://www.britannica.com/

biography/William-Rowan-Hamilton

[5] ”Density Functional Theory for Beginners.” Density Functional Theory for Beginners. Web. 15

Jan. 2017.

http://newton.ex.ac.uk/research/qsystems/people/coomer/

dft_intro.html

[6] Koch, Wolfram, and Max C. Holthausen. A Chemist’s Guide to Density Functional Theory.

Weinheim: Wiley-VCH, 2001. Print.

[7] Gotwals, Robert R., Jr., and Shawn Sendlinger. A Beginner’s Guide to Computational Chem-

istry. N.p.: n.p., n.d. Print.

[8] Lee, Chengteh, Weitao Yang, and Robert G. Parr. ”Development of the Colle-Salvetti

correlation-energy formula into a functional of the electron density.” Physical Review B 37.2

(1988): 785-89. Web.

[9] ”Marcel Swart’s Website.” Marcel Swart’s Website. Web. 15 Jan. 2017.

http://www.

marcelswart.eu/

.