128
1 Carlo Adamo Density Functional Theory Concepts and models [email protected] Équipe de Chimie Théorique et Modélisation Institut de Recherche de Chimie Paris Ecole Nationale Supérieure de Chimie de Paris Chimie ParisTech

DFT adamo reseau - French National Centre for Scientific Research · 2017. 1. 25. · 7kh edvlf lghd ri +duwuhh )rfn fdofxodwlrqv vroyh + (Ö< < iru d v\vwhp ri 0qxfohl dqg 1hohfwurqv

  • Upload
    others

  • View
    4

  • Download
    0

Embed Size (px)

Citation preview

  • 1

    Carlo Adamo

    Density Functional Theory

    Concepts and models

    [email protected]

    Équipe de Chimie Théorique et ModélisationInstitut de Recherche de Chimie Paris

    Ecole Nationale Supérieure de Chimie de Paris Chimie ParisTech

  • 1

    K

    i ig c

    x r

    2

    Hartree-Fock Model

    The basic idea of Hartree-Fock calculations:

    • solve Ĥ E for a system of M nuclei and N electrons using:

    1. the full Hamiltonian within the Born-Oppenheimer approximation:

    1 12

    1 1 1 1 1

    1 1ˆ2

    N M N N N M MI JI

    ii I i i j i I J IIi ij IJ

    Z ZZH

    r r r

    2. a trial wavefunction consisting of one Slater determinant:

    1 2... ...SD

    i j N

    3. molecular orbitals expressed as linear combinations of basis functions:

    basis function with a fixed form

    coefficient in linear expansion (called molecular orbital coefficients)

    artificial spin function

    Hartree-Fock models is the basis for virtually all quantum chemical methods

  • 3

    The basic idea of Hartree-Fock calculations:

    • solve Ĥ E for a system of M nuclei and N electrons using:

    1. the full Hamiltonian within the Born-Oppenheimer approximation:

    1 12

    1 1 1 1 1

    1 1ˆ2

    N M N N N M MI JI

    ii I i i j i I J IIi ij IJ

    Z ZZH

    r r r

    2. a trial wavefunction consisting of one Slater determinant:

    1 2... ...SD

    i j N

    3. molecular orbitals expressed as linear combinations of basis functions:

    1

    K

    i ig c

    x r

    4. variational optimization of using the molecular orbital coefficients as variational parameters

    Hartree-Fock Model

  • 4

    Hartree-Fock Energy

    to use the variational method, we need a relationship between the energy and the MO coefficients

    * ˆE H d for any normalized wavefunction:

    in Hartree-Fock theory:

    1 2... ...SD

    i j N

    and if you work it out, the energy becomes:

    2* 1

    1 1 11 1

    * * * *1 1 2 2 1 1 2 2

    1 2 1 21 1 12 12

    2

    1

    2

    N MI

    a aa I Ii

    N Na a b b a b b a

    a b

    ZE d

    r

    d d d dr r

    x x x

    x x x x x x x xx x x x

  • 5

    Hartree-Fock Energy

    2* 1

    1 1 11 1

    * * * *1 1 2 2 1 1 2 2

    1 2 1 21 1 12 12

    2

    1

    2

    N MI

    a aa I Ii

    N Na a b b a b b a

    a b

    ZE d

    r

    d d d dr r

    x x x

    x x x x x x x xx x x x

    to use the variational method, we need a relationship between the energy and the MO coefficients

    recall that molecular orbitals are linear combinations of basis functions

    1

    K

    i ig c

    x r

    so, now we have a direct relationship between the Hartree-Fock energy and molecular orbital coefficients

  • 6

    1 1 1

    1

    2

    N N N

    aa ab aba a b

    E h J K

    Hartree-Fock Energy

    or in standard short-hand notation:

    one electron energy

    Coulomb integral

    exchange integral

    two electron energy

    2* 1

    1 1 11 1

    * * * *1 1 2 2 1 1 2 2

    1 2 1 21 1 12 12

    2

    1

    2

    N MI

    a aa I Ii

    N Na a b b a b b a

    a b

    ZE d

    r

    d d d dr r

    x x x

    x x x x x x x xx x x x

    in terms of the MOs, the Hartree-Fock energy is:

  • 7

    Electron Correlation: Summary

    • electron correlation is the electron-electron energy ‘missing’ in an Hartree-Fock calculation

    0corr HFE E E there are two types of electron correlation

    1. Dynamic correlation: • accounts for fact that electrons move such that they avoid other electrons

    2. Static correlation: • in systems with multiple resonance states, electrons can avoid each other by occupying different resonance states

    • captured by using a multi-determinant wavefunction

    • captured by using multiple Slater determinants in the wavefunction

    • each Slater determinant is made of a unique set of molecular orbitals and represents a different electronic

    configuration

    • each Slater determinant is built by ‘exciting’ electrons within the Hartree-Fock determinant

    • the molecular orbitals in each Slater determinant are not re-optimized

    density functional theory treats the electron density quantum mechanically

  • 8

    Density vs. Wavefunction

    Wavefunction:

    • contains all of the information about the system

    • not an observable quantity

    • function of 4N variables (3 spatial, 1 spin per electron)

    Density:

    • function of only 3 spatial variables

    • physically measurable quantity

    • connected to the wavefunction

    2 d r r r

    does the electron density contain the same information that is contained in the wavefunction?

    • for a one electron wavefunction:

  • The ingredients (1):

    Density : It provides us information about how something is distributed/spread on a given space (volume)Electron Density :It tells us where the electrons are likely to exist

    Electron density is an « observable »

    2* )()()()( rrrr

    bis(diiminosuccinonitrilo)nickelexp DFT

    9

  • The ingredients (2):

    1234

    14916Y=f(X)

    A function (f) maps a set of numbers to another set of

    numbers

    A functional (F) is a function of a function

    Y

    A function which maps a set of functions to a set of

    numbersEx. F[A(X),B(X),C(X),….]

    A(X)B(X)C(X)D(X)

    2013F1

    234

    X

    YX

    Y=X2

    Example : Fxc[(r),(r)]10

  • Why DFT?

    From a pragmatic point of view: E. Bright Wilson, 1965

    • If one knows the exact electron density, (r), then the cusps ofthis density would occur at the positions of the nuclei.

    • Furthermore, a knowledge of |(r) | at the nuclei would give theirnuclear charges.

    • Thus the full Schrödinger Hamiltonian was known because it iscompletely defined once the position and charge of the nucleiare given.

    • In principle, the wavefunction and E are known, and thuseverything is known.

    The knowledge of the density is all that is necessary for a complete determination of all molecular properties.

    11

  • Can the total energy be expressed as a function of the density?

    HE ˆ

    21211

    rr11 rrrrrrrrrr

    '11

    ddr

    dVdH ext ),(1

    )()(),(2

    1ˆ2

    121

    21

    N2N21N2'111 xxxxr...xxrrr dddN ...)...,(*),(...),( 1111

    NN11 xxxxxrr ddddNN

    ...)...,(...2

    )1(),( 321

    2

    222

    Critical ingredients: kinetic energy and electron-electron interaction

    12

  • ][)()(

    2

    1

    12

    Jddr

    2121 rrrr

    ][)()(][ˆ eeext EdVTHE rrr

    Early attempts: Thomas-Fermi (1927)

    1) Exact kin. en. (T[]) is substituted by the kin. en. of a homogenous electrons gas

    cF=constante de Fermi

    e-/e- (Coulomb)Nuclei-electrons interaction(Coulomb)

    ][)()(][ eeTFTF EdvTE rrr

    2) The external potential (Vext[]) is that generated by the nuclei3) The electron-electron interaction is the Coulomb repulsion

    rr

    r dZ

    A

    A AR

    rr dc f3/5

    13

  • Thomas-Fermi (1927): how does it works?

    212121 )()(

    2

    1)(][ rr

    rr

    rrr

    rr ddd

    ZTE

    A

    ATFTF

    AR

    classical Coulomb interactions

    Not very well :in TF theory no molecular system is stable relative to dissociation

    into consitutents fragments

    non bonding theorem

    (Teller (1962) / Balazs (1967) / Lieb (1973) / Simon (1977))

    What is missing?

    Exchange

    Correlation

    and furthermore … TTF is local

    14

  • Add exchange : Thomas-Fermi-Dirac model (TFD)

    rrrrrrr

    rrr

    dcJdvdcE

    KJdvTE

    xfTF

    TFTF

    3/43/5 )(][)()()(

    ][][)()(][

    cf = 3/10(3p2)2/ 3cx = 3/4(3/p)1/ 3

    Add gradient corrections to kinetic energy :Thomas-Fermi-Dirac-Weizsacker model (TFDW or TFD-lW)

    rrrrrrr

    rrr dcJdvddcE xfTF

    3/4

    2

    3/5 )(][)()()(

    )(

    8

    1)(

    l

    Everything is expressed as a function of the density (or gradient of the density)

    15

  • HF

    A different approach:

    Slater and the X model (or HFS model)

    Cx = constant 0.75 < < 1

    Exchange only; no correlation

    Works EXTREMELY well (and it is still used especially for solid state)

    )1()1()1()1(22

    1

    2/,1

    21 ii

    Naaa

    A A

    A KJRr

    Z

    )1()1()1()1(2

    1 21 iiiXHFcoulomb

    A A

    A VVRr

    Z

    SLATER

    3/1rcV xXHF

    16

  • Why HFS is better than TFD?

    17

  • X performances

    18

  • The Hohenberg and Kohn theorems (1964)

    • All properties of the many-body system are determined by the ground state density 0(r)

    • Each property is a functional of the ground state density 0(r) which is written as f [0]

    )()(

    )(

    0i

    0

    rr

    rr

    extv

    19

  • 0(r) define the external potential Vext(r) (except for constant) and thus all properties

    r

    Hohenberg and Kohn: theorem I

    20

  • 21

    First Hohenberg-Kohn Theorem

    the ground state energy of a system is uniquely determined by the ground state electron density

    12

    1 1 1 1

    1 1ˆ2

    N N M N NI

    a ai i I i j iiI ij

    ZH

    r r

    consider two different Hamiltonians:

    12

    1 1 1 1

    1 1ˆ2

    N N M N NI

    b bi i I i j iiI ij

    ZH

    r r

    0, 0, 0,ˆa a a aH E

    0, 0, 0,ˆb b b bH E

    the two Hamiltonians differ in the electron-nuclear attraction term:

    • electron-nuclear attraction is called external potential in DFT lingo

    • each Hamiltonian has a unique ground state wavefunction and ground state energy

  • 22

    First Hohenberg-Kohn Theorem

    can two different wavefunctions give us the same electron density??

    *0, 0, 0,

    ˆa b a bE H d r

    *0, 0, 0,ˆ ˆ ˆa b a b b bE H H H d r

    * *0, 0, 0, 0, 0,ˆ ˆ ˆa b a b b b b bE H H d H d r r

    we know:

    1 1 1 1

    ˆ ˆN M N M

    I Ia b a ba b

    i I i IiI iI

    Z ZH H

    r r

    *0, 0, 0,

    ˆb b b bH d E r

    0, 0 0,a bd d r r r r r

    based on the variational principle:

  • 23

    First Hohenberg-Kohn Theorem

    can two different wavefunctions give us the same electron density??

    * *0, 0, 0, 0, 0,ˆ ˆ ˆa b a b b b b bE H H d H d r r *0, 0, 0, 0,a a b b b bE d E r

    0, 0 0,a a b bE d E r r

    if we do the same for E0,b:

    0, 0 0,b b a aE d E r r

    * *0, 0, 0, 0, 0a a b bd d r r r r r r r

    in both cases we assumed that:

  • 24

    First Hohenberg-Kohn Theorem

    can two different wavefunctions give us the same electron density??

    0, 0 0,a a b bE d E r r

    0, 0 0,b b a aE d E r r+

    0, 0, 0 0, 0,a b a b b a b aE E d E E r r0

    0, 0, 0, 0,a b b aE E E E

    clearly, the sum of two energies cannot be less than the sum of the same two energies

    therefore, our assumption that two different wavefunctions with different energies can give the same density is wrong

    the ground state energy of a system is uniquely determined by the ground state electron density

  • 25

    First Hohenberg-Kohn Theorem

    the ground state energy of a system is uniquely determined by the ground state electron density

    the first Hohenberg-Kohn theorem is an existence theorem

    • it tells us that a unique relationship exists between the ground stateenergy and the ground state electron density

    • it does not tell us what that relationship is

    we do know that the energy is a functional of the density

    E E r• a functional is a function whose argument is another function

    • it can be shown that all other ground state properties of the system can be expressed as functionals of the electron density, too

  • In principle, one can find all other properties and they are functionals of 0(r).

    r

    r

    E

    E

    rrr

    rrr

    dV

    dVVTE

    extHK

    extHK

    F

    int

    26

  • 27

    Second Hohenberg-Kohn Theorem

    the ground state density is a variational quantity

    This means:

    • if we pick any arbitrary density it will give us an upper bound on the true ground state density

    • if we compare two densities, the one with the lower energy is the better one

    • we can employ linear variation techniques to optimize the density, like we did with wavefunction methods

    Proof:

    • we know there is unique correspondence between the density and the wavefunction

    • so, if we pick an arbitrary trial density, we will get an arbitrary trial wavefunction

    • the variational theory tells us that for any trial wavefunction:

    *0

    ˆtrial trial trialE H d E

    trial trial

  • 28

    Energy Functionals

    First HK Theorem: • there exists an exact functional relationship between the ground state density and the ground state energy

    Second HK Theorem: • the electron density can be optimized variationally

    to use variational optimization procedures, we must know how E and are related!!!

  • )()(

    )(

    0i

    0

    rr

    rr

    extv

    There is an infinite number of w.f. yielding 0

    DFT- HK

    QC

    • How do I know, given an arbitrary function (r), that it is a density coming from an antisymmetric N-body wavefunction Ψ(r1,...,rN)?

    All observables

    N-representability

    • How do I know, given an arbitrary function (r), that it is the ground statedensity of a local potential v(r)?

    How to get the wf from the density?

    Solved: any square-integrable nonnegative function satisfies it

    V-representability Constrained-search formalism (Levy-Lieb)Definition of the universal functional

    How to get the ground state density (r)?

    29

  • Constrained search formalism (Levy and Lieb (1977))

    eeextVVTE ˆˆˆminmin0

    drVVTE extee )(ˆˆminmin0 r

    Double minimization procedure:

    12

    eeVTF ˆˆmin

    Definition of the universal functional :

    eeVTF ˆˆmin

    Wavefunctions giving density ( r )

    30

  • Double minimisation or how to find the tallest child in a school?

    From Parr&Yang

    min

    min

    1. Find the tallest in each classroom

    2. Find the tallest of the tallest

    31

  • Various extensions have been proposed:

    Extensions to spin dependent systems (Barth, Hedin, 1972)

    [ , ]E n n

    Extension to relativistic systems (Vignale, Kohn, 1988)

    [ ( )]E j rExtension to finite temperatures

    [ ] [ ] [ ]F n E n TS n

    Time-Dependent DFT (Runge, Gross, 1984)

    32

  • iiieffiKS rvH

    2

    2

    Kohn and Sham ansatz (1965)

    Replace the interacting-particles hamiltonian with one that it can be solved more easily

    KS hamiltonian: an hamiltonian describing N non-interacting particlesassumed to have the same density as the true interacting system.

    NN ...det! 3212/1

    If you don’t like the answer, change the question

    33

    What have we gained so far?

    Apparently Nothing: The only result is that the density determines the potential

    We are still left with the original many-body problem

  • 34

    Kohn-Sham Energy Functional

    Kohn and Sham suggested decomposing the total energy into kinetic and potential energy contributions

    classical non classicalne ee eeE T V V V

    ?T • accounts for the kinetic energy of the electrons

    1

    MI I

    neI I

    ZV d

    R rr

    r R

    • accounts for nuclear-electron Coulombic attraction

    1 2 1 21 2

    1

    2classicaleeV d d

    r r

    r rr r

    • accounts for average electron-electron Coulombic repulsion

    ?non classicaleeV • accounts for all electron-electron interactions not in classicaleeV• includes exchange and instantaneous electron-electron

    correlation

  • 35

    Kohn-Sham Energy Functional

    Kohn and Sham recognized that it’s easier to calculate the kinetic energy if we have a wavefunction

    2

    *

    2iT d

    r

    but we don’t know the ground state wavefunction (if we did, we wouldn’t bother with DFT)

    So, they suggested:

    • instead of using the real wavefunction and density, let’s use a Slater determinant wavefunction built up from a set of one-electron orbitals (like molecular orbitals) to build up an artificial electron system that represents the ground state density

    • ‘artificial system’ is often called the non-interacting reference system

    • the ‘one-electron orbitals’ are called Kohn-Sham orbitals

    • the Kohn-Sham orbitals are orthonormal:

    * 1 if

    0 otherwisei j iji j

    d

  • 36

    Kohn-Sham Energy Functional

    the Kohn-Sham orbitals give us an easy way to construct the density and get an approximate value of the kinetic energy

    kinetic energy of N one-electron orbitals:

    2*

    1 2

    Ni

    rs i ii

    T d

    r

    density from N one-electron orbitals:

    *1

    N

    i ii

    r r r

    N d r r

    the density integrates to give the total number of electrons:

  • 37

    Kohn-Sham Energy Functional

    a Slater determinant wavefunction built up from Kohn-Sham orbitals will never give the exact ground state kinetic energy

    2*

    2i

    exactT d

    r2

    *

    1 2

    Ni

    rs i ii

    T d

    r

    (the closest approximation to the real wavefunction you can get with a single Slater determinant is the Hartree-Fock wavefunction)

    so, Kohn and Sham suggested splitting up the kinetic energy

    exact rsT T T

    exact ground state kinetic energy kinetic energy of the

    reference system

    correction accounting for kinetic energy not contained in Trs

  • 38

    Kohn-Sham Energy Functional

    classical non classicalne ee eeE T V V V

    incorporating the orbital-based expressions into the total energy functional:

    gives:

    classical non classicalrs ne ee eeE T V V V T

    terms we have explicit expressions for with everything that

    Kohn-Sham orbitals we don't know how to solve

    classicalrs ne ee xcE T V V E

    this is the Kohn-Sham total energy functional

  • 39

    Kohn-Sham Energy Functional

    classicalrs ne ee xcE T V V E

    2*

    1 2

    Ni

    rs i ii

    T d

    r

    1

    MI I

    neI I

    ZV d

    R rr

    r R

    *1

    N

    i ii

    r r r

    with Kohn-Sham orbitals, we can calculate:

    we don’t have an expression for Exc:

    • Exc is called the exchange correlation functional

    • it acts as a repository for all contributions to the energy that we do not know how to calculate exactly

    • if we had an exact expression for Exc we could calculate the energy and density exactly

    1 2 1 21 2

    1

    2classicaleeV d d

    r r

    r rr r

  • 40

    The Exchange-Correlation Functional

    the 1st Hohengberg-Kohn theorem tells us that an exact functional relationship between energy and the density

    classicalrs ne ee xcE T V V E

    this means an exact form of Exc must exist, and if we had it, the total energy functional would be exact

    unfortunately, we do not know the exact form of Exc

    • the accuracy of DFT calculations hinges on the accuracy various approximations to Exc

    • the development of exchange-correlation functionals is a major area of research in modern theoretical chemistry

    • we’ll talk about about exchange-correlation functionals in greater detail later

    • for now, let’s just treat the exchange-correlation functional in a generic sense

  • 41

    The Kohn-Sham Orbitals

    how do we solve for the Kohn-Sham orbitals?

    the second Hohenberg-Kohn theorem tells us that the ground state density minimizes the energy

    this means:

    we’ll use variational theory

    • we need connections between the energy, orbitals, and density

    • we need a way to ‘optimize’ the orbitals

    • expanding the Kohn-Sham orbitals as linear combinations of basis functions:

    a variational treatment of the orbitals will involve

    • using linear variation methods to get the set of coefficients that give us a density that minimizes the energy

    basis function with a fixed form

    coefficient in linear expansion

    1

    K

    i ig c

    x r

    artificial spin function

  • 42

    The Kohn-Sham Orbitals

    how do we solve for the Kohn-Sham orbitals?

    in terms of Kohn-Sham orbitals:

    classicalrs ne ee xcE T V V E

    2* *

    1 1

    2*1 2 1 1

    1 1

    2

    1

    2

    N Mi I

    i i i ii I I

    N

    i i xci

    ZE d d

    d d E

    r r r r r rr R

    rr r r r

    r r2

    2

    1 2*1 2

    1 1 1 1

    1

    2 2

    N M NIi

    i i xci I iI

    ZE d d d d E

    r r r

    r r r r r rr R r r2

  • 43

    The Kohn-Sham Orbitals

    how do we solve for the Kohn-Sham orbitals?

    we have connections between the orbitals, density and energy:

    2* *

    1 1

    2*1 2 1 1

    1 1

    2

    1

    2

    N Mi I

    i i i ii I I

    N

    i i xci

    ZE d d

    d d E

    2

    r r r r r rr R

    rr r r r

    r r

    *1

    N

    i ii

    r r r

    we also have a connection to the orbital coefficients:

    1

    K

    i ig c

    x r

  • 44

    The Kohn-Sham Orbitals

    we have to apply the apply the linear variational approach to the Kohn-Sham DFT energy

    with the constraint that the Kohn-Sham orbitals remain orthonormal

    classicalrs ne ee xcE T V V E

    *i j ijd

    if you do this (we won’t), you find out that the ‘best’ set of Kohn-Sham orbitals are given by:

    221 2 1 11 1 1 2

    1

    2

    MI

    xc i i iI I

    Zd V

    r

    r r rr R r r

    Kohn-Sham operator Kohn-Sham orbital i

    energy of Kohn-Sham orbital i

  • 45

    The Kohn-Sham Operator

    the Kohn-Sham orbitals are eigenfunctions of the Kohn-Sham operator

    221 2 1 11 1 1 2

    1

    2

    MI

    xc i i iI I

    Zd V

    r

    r r rr R r r

    Kohn-Sham operator Kohn-Sham orbital i

    energy of Kohn-Sham orbital i

    the first three terms account for: • the kinetic energy of the electron in i• the Coulombic attraction between the

    nuclei and the electron in i• the static Coulombic repulsion between

    the electron in i and the total electron density

    the last term accounts for: • exchange repulsion, electron correlation, self-interaction energy and kinetic energy

    correction associated with the electron in i

  • 46

    Comparison with Hartree-Fock

    the Kohn-Sham orbitals are eigenfunctions of the Kohn-Sham operator

    221 2 1 11 1 1 2

    1

    2

    MI

    xc i i iI I

    Zd V

    r

    r r rr R r r

    2 *2

    2 2 2 2 211 1 1 1

    1 1 12 12

    ( )

    2

    Mb b aI

    a a b a aI b a b aI

    d dZ

    r r r

    x x x x x

    x x x x

    the Hartree-Fock molecular orbitals are eigenfunctions of the Fock operator

    these two expressions are very similar:

    • the kinetic energy, nuclear-electron, and static Coulomb electron-electron repulsion terms in each operator are identical (but they may be written slightly differently)

    • the last terms differ:

    • in the Fock operator, only exchange interactions are considered

    • in the Kohn-Sham operator, all remaining interactions are considered

    can never give exact results

    can give exact results, if Vxc is exact

  • 47

    Solving for the Kohn-Sham Orbitals

    we want to solve

    1 1 1ˆ ( )KS i i if r r r

    1 1 11 1

    ˆ ( )K K

    KSi i if c c

    r r r

    multiply on the left by * and integrate:

    * *1 1 1 1 1 1 11 1

    ˆ

    KS

    K KKS

    i i i

    SF

    c d f c d

    r r r r r r r

    FKS is an element of the Kohn-Sham matrix

    S is an element of the overlap matrix

    Note: there will be K of these equations because there are K basis functions

    using a linear combination of basis functions

    this is completely analogous to what we do in the Hartree-Fock method

    1

    K

    i ig c

    x r

  • 48

    Solving for the Kohn-Sham Orbitals

    by converting to a matrix form, we can solve for the orbitals with a self-consistent procedure like we did with Hartree-Fock

    * *1 1 1 1 1 1 11 1

    ˆ

    KS

    K KKS

    i i i

    SF

    c d f c d

    r r r r r r r

    FKSC = SC

    C is a K x K matrix whose columns define the coefficients, ci:

    is a diagonal matrix of the orbital energies, i:

    1

    2

    K

    0

    11 12 1

    21 22 2

    1 2

    K

    K

    K K KK

    c c c

    c c c

    c c c

    C

    1 2 K

  • 49

    221 2 1 11 1 1 2

    1

    2

    MI

    xc i i iI I

    Zd V

    r

    r r rr R r r

    Solving for the Kohn-Sham Orbitals

    once again, the Kohn-Sham matrix elements depend on the Kohn-Sham orbitals coefficients

    *1

    N

    i ii

    r r r

    * *1 1 1

    K K N

    i i i ii

    P

    c c

    r r r

    P is the density matrix, it quantifies the amount of electron density ‘shared’ between basis functions and

  • 50

    once again, the Kohn-Sham matrix elements depend on the Kohn-Sham orbitals coefficients

    Solving for the Kohn-Sham Orbitals

    Solution: • solve with iterative techniques

    • make an initial guess of C1• build FKS1 and solve for C2

    • use C2 to build FKS2 and solve for C3...

    • use CN to build FKSN and solve for CN+1

    • stop when CN and CN+1 are the same (or very similar)

    • this self-consistent field (SCF) approach was also used in Hartree-Fock

    • final matrix C defines the Kohn-Sham orbitals that give the density that minimizes the energy

  • 51

    Kohn-Sham Density Functional Theory

    isn’t this just Hartree-Fock?

    No, in Hartree-Fock we try to solve the Schrödinger equation using a trial wavefunction consisting of a single Slater determinant

    1 1 1ˆ a a af x x x

    1 2... ...HF

    i j N

    * ˆHF HF HFE H d since the Hartree-Fock wavefunction is necessarily approximate, this can

    never give the exact ground state energy

  • 52

    in Kohn-Sham DFT we only use the Kohn-Sham orbitals

    to give us the density:

    *1

    N

    i ii

    r r r

    that minimizes the Kohn-Sham energy functional:

    2

    1 2*1 2

    1 1 1 1 2

    1

    2 2

    N M NIi

    i i xci I iI

    ZE d d d d E

    r r r

    r r r r r rr R r r

    Kohn-Sham Density Functional Theory

    isn’t this just Hartree-Fock?

    1 1 1ˆ ( )KS i i if r r r

    if we have an exact form of Exc, the total energy functional is exact and we can get the exact ground state energy and density with DFT

  • 53

    Meaning of KS Eigenvalues

    The ionization potential is

    i

    iif2

    )()( rr f= occupation number (could be fractional)

    kkkkk

    vNk

    Nkk fdffdf

    EEEI

    1

    0

    1

    0

    1

    Using the mean-value approximation 5.0kkI

    This is the so-called Janak-(Slater) transition state theorem

    It could be considered as the equivalent of the Koopman’s theorem

  • 54

    Exchange Correlation Functionals

    classicalrs ne ee xcE T V V E

    the accuracy of Kohn-Sham DFT rests on the availability of accurate exchange-correlation functionals

    although the 1st Hohenberg-Kohn theorem tells that an exact form of Excmust exist, we don’t know what it is

    The exchange-correlation functional should describe:

    1. exchange (Pauli) repulsion between electrons of the same spin

    2. electron correlations from instantaneous electron-electron interactions

    3. kinetic energy corrections

    in modern Kohn-Sham DFT, we use approximate exchange-correlation functionals that account for a portion of these interactions

    The problem: Exc exact is unknown

    The theory is exact, the functionals are approximated

  • 55

    Self-Interaction Energy

    1 2 1 21 2

    1

    2classicaleeV d d

    r r

    r rr r

    electron density of the hydrogen atom

    For the Hydrogen atom:

    • there is only one electron

    • we represent the electron with a probability distribution and average it over all space

    • so the electron makes a contribution to the density at r1 and r2

    • leads to a spurious electron repulsion

    • the situation extends to multi-electron systems

    • the self-interaction error must be cancelled out

    • in Hartree-Fock, exchange cancels out the self-interaction

    • in DFT, we use the exchange-correlation functional to cancel out the self-interaction

  • rrr dVEJTE extXCSHK )()(

    JVTT

    JTEE

    eeS

    SHKXC

    EHF Hi 12i1,N

    J ij Kij i, j1,N

    Consequences of the use of approximate vxc : self interaction error (SIE)

    In exact KS as in HF: there is no Coulomb interaction of one electron with itself

    SIE(N) J i Exc i ,0

    i

    But in approximate DFT this is not the case:

    56

  • Effects of self interaction error

    He2+

    He2+

    CCSD(T)

    BLYP

    LDA

    PW91

    (He++He)

    2(He+0.5)

    57

  • How to get an approximate xc functional?

    Contains information on the many-body system of interacting electrons

    The easiest way: Local Density Approximation – LDA

    • Assume the functional is the same as a model problem –the homogeneous electron gas

    • Separate the exchange and correlation contributions: Exc = Ex + Ec

    • Exc can be calculated as a function of the density only

    58

  • LDA

    rrrrrrr ddE CXXCLDAXC ))(())(()())(()(][ homogeneous electron gas exchange-

    correlation energy per particle(at the point r)

    probability of finding the particle at r

    model problem : the homogeneous electron gas

    The value of the xc energy depends only on the local density.

    The e- density () may vary as a function of r, but is single-valued, and the fluctuations in away from r do not

    affect the value of Excat r.

    59

  • • around each electron other electrons tend to be excluded

    Definition of “x-c hole”:

    • Excis the interaction of the electron with the “hole” : it involves only a spherical average

    Exchange hole in Ne atomGunnarsson, et. al.

    Very non-spherical!

    Spherical average very closeto the hole in a homogeneouselectron gas!

    nucleus electron

    Spherical averagearound electron

    Is the Local Density Approximation physically sounding?

    )( r'r,xc

    r'r

    r'r,rr dd][

    r'-

    )( xcxcE

    60

  • Exchange-correlation (x-c) hole in silicon

    • Calculated by Monte Carlo methods

    Hood et al

    Exchange Correlation

    Hole is reasonably well localized near the electronSupports a local approximation

    61

  • LDA

    rrrrrrr ddE CXXCLDAXC ))(())(()())(()(][ model problem : the homogeneous electron gas

    Get an expression for them

    62

  • 3/1)(r xLDAX C

    LDA : Exchange partDerived by Bloch et Dirac (1929/1930) for

    homogeneous electron gas

    rrr dE XLDAX ))(()(][

    (per particle)

    Usually called Slater exchange functional

    Functional form identical to that of Slater (HFS)

    LDA : Correlation part• No explicit formulation• Approximate analytical expression to reproduce accurate quantum Monte-Carlo (Ceperly & Alder, 1980)results for a homogeneous electron gas• Most used LDA approximation for correlation Volsko, Wilk et Nusair (1980): VWN.

    bx

    Q

    Q

    xbxx

    xX

    bx

    bx

    Q

    Q

    b

    xX

    xALDAC 2

    arctan)2(2

    )ln()(2

    arctan2

    )(ln

    202

    00

    02

    63

  • Local Spin Density Correlation Functional

    • Not for the faint of heart:

    709921.1

    5198421.0

    /2)1()1('

    ),49671.0,88026.0,6231.3,357.10,11125.0,0168869.0(

    ),62517.0,3662.3,1977.6,1189.14,20548.0,01554535.0(

    ),4294.0,6382.1,5876.3,5957.7,21370.0,0310907.0(

    )1(')'1()'1(],[

    3/43/4

    2/1

    2/1

    2/1

    444

    zz

    scorrm

    scorrp

    scorru

    zz

    mpus

    LSD

    C

    f

    rG

    rGe

    rGe

    feer

    )(((2

    11log)1(2),,,,,,(

    4321

    2

    143211 rbbrbrbarraarbbbbaaGcorr

  • LDA: how it works

    From A. V. Morozov

    65

    Discontinuty of thepotential for the fillingof a electronic shell

    rr

    xcxc

    xcxc

    Ev

  • • A: High density, large kinetic energy, LDA approximation unimportant

    • B: Small density gradient, LDA is good

    • C:large gradient, LDA fails

    66

    LDA: how it works

  • Exchange-Correlation Hole

    • Due to phenomena of exchange there is a depletion of density (of the same spin) around each electron.

    • Mathematically described as

    • The exchange correlation energy written as

    rr ,xc

    rr

    rr

    rrrddxcExc

    ,

  • Properties of the hole

    • Subject of much research.

    0'',

    1'',

    0',

    ',',',

    rrr

    rrr

    rr

    rrrrrr

    d

    d

    c

    x

    x

    cxxc

    The LDA must obey these.

    Negative quantity determining thedecrease in the probability of finding anelectron of the same spin at position r’when one electron is known to be atreference position r

  • Why is this important?

    • Huge error made to the integral would occur if the hole is not normalised correctly.

    • The LDA has this correct – it is the correct expression for a proper physical system.

    • In fact, only need the spherical average of the hole is needed.

  • • Different densities for different spins : split the total density

    Open shell systems and Local Spin Density Approximation(LSDA)

    rrrr dE XCLSDAXC ))(),(()(],[

    )()()( rrr

    Measure of spin polarization :

    )(

    )()(

    r

    rr

    Remark : in principle, since the external potential is spin independent there is no need to split the different spin densities

    70

  • How to ameliorate the LDA?

    Atoms, molecules or solids are not a homogeneous electron gas :

    Include non local effects

    GEA (Gradient Expansion Approximation): F(r), (r)

    ',

    3/23/2, ...),(),()(],[

    rrr dCdE XCXC

    GEAXC

    Taylor expansion

    Note: mathematically speaking GEAs are still local

    How do they work? Not a great improvementReason : Exchange correlation hole properties not satisfied

    71

  • GGA idea

    • A brute force fix.

    • If x(r,r’)>0, set it to zero.• If sum rule violated, truncate the hole.

    • Resulting expressions look like:

    rr

    r

    3/4

    3/4

    n

    ns

    dnsFLSDAEGGAE xx

  • Impose the fulfillment of the properties of the exchange-correlation hole

    rdfEGGAXC ),,,(],[ GGAC

    GGAX

    GGAXC EEE

    GGA (Generalized Gradient Approximation)

    rr dsFEELDAX

    GGAX )()(

    3/4

    )(

    )()(

    3/4 r

    rr

    swhere s is the reduced density

    S high for high gradient or small density regions (far from nuclei)S small for small gradients (bonding region)S intermediate for high gradient and small density (near the nuclei)

    Measure local inhomogenity

    73

  • ss

    sF B

    1

    2

    sinh61

    Becke, 1988 (B ou B88)

    Example of commonly used exchange functionals

    74

    • Chemistry stable– Empirical

    – =0.0042, fitted to exchange energies of He ... Rn.– Gives correct asymptotic form in exponential tails.

  • Example of commonly used exchange functionals

    Problem: How to get a GGA?

    Functional form can get extremely complex (especially for correlation functionals)

    75

    • The physics stable:– Principled, parameter free

    – Numerous analytic properties

    – Slow varying limit should give LDA response. This requires Fx →s2 , =0.21951

    – Density scaling, n(r)→l3n(lr), Ex→l Ex

    804.0,/1 2

    s

    sF

  • • Parametrized functionals –semi-empirical-use adjusted parameters to reproduced exact or experimental data (ex. atomicenergies)

    • Non parametrized functionalsimpose physical constrains (uniform electron gas limit, asymptotic behavior)

    Fxauthors Fc

    authors

    B

    PW91

    PBE

    mPW

    HCTH

    B97

    Becke (1988)

    Perdew et Wang (1991)

    Perdew, Burke, Ernzerhof (1996)

    Adamo, Barone (1997)

    Handy et al. (1999)

    Becke

    P86

    LYP

    PW91

    PBE

    Perdew (1986)

    Parr et al. (1988)

    Perdew et Wang (1991)

    Perdew, Burke, Ernzerhof (1996)

    Brief (and extremely non exhaustive) list of commonly used GGA functionals

    Over parametrisation Semi-empirical Good properties

    « universal » (worse) chemical properties

    76

  • Exchange, Ex

    LSDA GGA Meta-GGA

    X1951

    Dirac1930

    G96

    B86 B88

    PW91PBE1996

    RPBE1999

    revPBE1998

    xPBE2004

    PW86

    mPW

    TPSS2003

    BR89

    PKZB1999

  • CS1975

    Correlation, Ec

    LSDA GGA Meta-GGA

    W38

    xPBE2004

    PW86

    PBE1996PW91

    LYP1988

    B95

    TPSS2003

    PKZB1999

    B88VWN1980

    PZ81

    PW92CA Data1980

  • 79

    Some theoretical contraints

    Size consistency : E(AB)=E(A)+E(B)Virial TheoremSelf-interaction errorJanak Theorem

    Lieb-Oxfod boundCoordinate scaling

    Hirao JCP 1999

  • 80

    Lieb-Oxford Limit

    Bonding region

    Adamo JCP 2002

  • • Exchange-correlation functionals must be numerically integrated– not as robust as analytic methods

    • Energies and gradients are 1-3 times the cost of Hartree-Fock

    • Frequencies are 2-4 times the cost of HF• Some of this computational cost can be

    recuperated for pure density functionals by employing the density fitting approximation for the Coulomb interaction.

    Calculating Exc Terms

  • HF LDA GGA Exp.

    H2 3.64 4.90 4.55 4.73

    H2O 6.72 11.58 10.15 10.06

    HF 4.21 7.03 6.16 6.11

    O2 1.43 7.59 6.24 5.25

    F2 -1.60 3.34 2.30 1.69

    CH4 14.22 20.03 18.21 18.17

    Atomisation energies (in eV) of several molecules: theory vs experiment.

    Property HF LDA GGA Exp.

    a0, Å 3.58 3.53 3.57 3.567

    Ea, eV -5.2 -8.87 -7.72 -7.55

    K0, GPa 471 455 438 442

    Solids : diamond

    82

  • fully nonlocal+ explicit dependence on unoccupied orbitals

    rung 5

    hybrid functionals

    + explicit dependence on occupied orbitals

    rung 4

    meta-GGAs + explicit dependence on kinetic energy density

    rung 3

    GGAs +explicit dependence on gradients of the density

    rung 2

    LDAlocal density onlyrung 1

    John Perdew Jacob Ladder*

    *DFT conference Menton 2000

    Philippe Ratner

    EARTH (Hartree theory)

    HEAVEN (chemical accuracy)

    How to improve GGA results?

    83Update 2007:

    Rung 4bis: Hyper GGAs

  • meta-GGA or dependent

    Fxc, , occ,1i2

    i kinetic energy density

    Introduce quasi-local information

    84

  • How to improve the exchange energy : Hybrid Functionals

    How does it work? Very bad!Mean average error 32 kcal/mole for the G2 ensemble (50 molecules) while aMAE of 5-7 kcal/mol is obtained with standard GGA

    KSC

    exactXXC EEE

    Idea: HF exchange is exact. Therefore we could think to combine exact (HF) exchange with a (GGA ) correlation functional (Lie & Clementi 1974)

    Actually combining a percentage of HF exchange with a GGA exchange works much better!!!

    Theoretical justification : adiabatic connection (Becke 1993 – Half and Half)

    In practise : HF% between 20 and 30%

    85

  • The Adiabatic Connection

    What do we know about ?

    86

  • What does Wl look like?From Yang

    87

  • Use of Adiabatic Connection

    None of these functionals is self-interaction free,

    even with the use

    88

  • 9110

    3 )1( PWccLSDc

    Bxx

    HFxxo

    LSDxx

    Bxc EaEEaEaEaE

    PBE0

    PBEc

    PBEx

    HFx

    PBExc EEEE 4

    3

    4

    10

    B3LYP

    (Becke, JCP 1993)3 parameters fitted on G2 (ionization and atomization energies)

    (Adamo, Scuseria JCP 1999)No fitted parameters

    Normally hybrid functionals outperform all GGA and meta-GGA functionals and they are the reference for chemical applications

    89

  • Method Distance(Å)

    D0(kcal/mol)

    Dipole moment (D)

    Harmonic freq. (cm-1)

    HF & post-HF

    HF 0.022 82.0 0.29 144

    MP2 0.014 23.7 0.28 99

    CCSD[T] 0.005 11.5 0.10 31

    LDA & GGA

    LSDA 0.017 43.5 0.25 75

    BPW 0.014 6.0 0.11 69

    BLYP 0.014 9.6 0.10 59

    LGLYP 0.013 7.2 0.10 60

    PWPW 0.012 8.6 0.12 66

    mPWPW 6.7 0.11 65

    Hybrid 3 parameters

    B3LYP 0.004 2.4 0.08 31

    B3PW 0.008 4.8 0.08 45

    mPW3PW 0.008 2.6 0.08 37

    Hybrid ab-initio

    B1LYP 0.005 3.1 0.08 33

    B1PW 0.010 5.4 0.10 48

    LG1LYP 0.005 4.0 0.10 45

    Performance of selected functionals

    diatomic molecules

    90

  • MAE for harmonic frequencies(cm-1)

    (G2 set,>50 organic molecules)

    Level of thoery Errorr

    HF and post-HF

    HF/6-311G(3df,2p) 144

    MP2/6-31G(d,p) 99

    CCSD/6-311G(3df,2p) 31

    LSDA

    SVWN/6-31G(d,p) 75

    GGA

    BLYP/6-311G(d,p) 59

    BPW91/6-311G(d,p) 69

    PWPW91/6-311G(d,p) 66

    mPWPW91/6-311G(d,p) 66

    Hybrid functionals

    B1LYP/6-311G(d,p) 33

    B1PW91/6-311G(d,p) 48

    mPW1PW91/6-311G(d,p) 39

    B3LYP/6-311G(d,p) 31

    B3PW91/6-311G(d,p) 45

    mPW3PW91/6-311G(d,p) 37

    Performance of selected functionals

    91

  • 92

    Rutile - GTO

  • 93

  • TS 13 P14

    BHTTS 15

    G. Talarico, P. H. M. Budzelaar, V. Barone and C. Adamo Chem. Phys. Lett. 329, 99, (2000).G. Talarico, V. Barone, P. H. M. Budzelaar and C. Adamo, J. Phys. Chem. A, 105 (2001) 9014G. Talarico, V. Barone, L. Joubert and C. Adamo Int. J. Quantum Chem. 91 (2003) 474

    The XC functional can be crucial for chemical understanding

    R1 = H, R2 = iso-propyl ;R1 = tert-butyl, R2 = iso-propR=growing chain

    N

    Al

    N

    R1

    R2

    R

    R2

    The catalyst

    Homogeneous catalysis of ethylenegrowing chain

    termination

    94

  • Different functionals different CHEMICAL

    answers.

    insertion channel

    BLYP

    BP86

    B3LYP

    B3PW91

    B1LYP

    mPW0

    B98

    PBE0

    B1Bc95

    VSXC

    MP2

    best est.

    15 20 25 30 35

    E#ins

    E(kcal/mol)

    fun

    ctio

    na

    l

    BLYP

    BP86

    B3LYP

    B3PW91

    B1LYP

    mPW0

    B98

    PBE0

    B1Bc95

    VSXC

    MP2

    best est.

    5 10 15 20 25

    E#

    BHT

    E(kcal/mol)

    func

    tion

    al

    termination channel

    BP86 termination most probable(low m.w.)VSXC insertion most probable(higher m.w.)

    95

  • Known breakdowns of DFT

    LocalizationSelf Interaction Error (SIE)

    Non-dynamic correlation effects

    reaction barriers, CT complexes, Rydberg excitations, IP from Koopmans, band gap, bond length alternation, magnetic properties, vdW interactions,…

    (Un)Known causes?

    Many efforts to assess the reliability of DFT by a trial-and-error approach

    All problems coming from the approximate

    nature of the XC contribution

    among others

    The theory is exact, the functionals are approximate

    96

  • Challenges in DFT

    Better functionals (e.g. CR, MR-05)Error corrections (e.g. SIC, ad-hoc parametrization)Static correlation (e.g. MC-DFT, RSH)Localization vs delocalization

    97

    In chemical language

  • 98

    Many functionals on the market

    adapted from K. Burke

  • Known breakdowns of DFA

    LocalizationSelf Interaction Error (SIE)

    Non-dynamic correlation effectsPoor performing functionals

    reaction barriers, CT complexes, Rydberg excitations, IP from Koopmans, band gap, bond length alternation, magnetic properties,

    vdW interactions,…

    (Un)Known causes?

    Many efforts to assess the reliability of DFA by a trial-and-error approach

    All problems coming from the approximate nature of the XC

    contribution

    among others

  • From DFT to DFA

    fully nonlocal+ explicit dependence on unoccupied orbitals

    rung 5

    hybrid functionals + explicit dependence on occupied orbitals

    rung 4

    meta-GGAs + explicit dependence on kinetic energy density

    rung 3

    GGAs +explicit dependence on gradients of the density

    rung 2

    LDAlocal density onlyrung 1

    EARTH (Hartree theory)

    HEAVEN (chemical accuracy)

    The Jacob’s Ladder of John Perdew

    Update 2007:Rung 4bis: Hyper GGAs

    No simple rules for a priori reliabilityNo systematic route to improvementToo many functionals to choose fromStatistical approaches

  • It’s tail must decay like -1/r

    It must have sharp steps for stretched

    bonds

    It keeps H2 in singlet state as

    R→∞

    Adapted from a poem by John Godfrey Saxe (1816–1887)

    What we know about the « true » functional is only a part

    Exploring DFT as blind men?

  • 102

    Some theoretical contraints

    Lieb-Oxfod boundCoordinate scaling

    Hirao JCP 1999Using (some) of the known theoretical constraints todefine the exchange-correlation functional

  • The starting point: the PBE functional

    Focusing on the exchange

    Analytical formula of Becke86

    1) Recovering the uniform gas limit

    2/22, xxx EEE2) The exact exchange energy obeys

    the spin-scaling relationship

    3) Recovering the LSD linear response(small density variations)

    21)( ssFx

    1)0( xF

    Perdew-Burke-Ernzerhof PRL 1996

    2

    2

    11

    s

    sF PBEx

    rdrFE unifxGGAxGGAx 33/4)()(],[

    =0.21951,k=0.804 and s=||/2r(3p2)1/3

    3 more theoretical conditions for correlation

    4) Lieb-Oxford limit rdEx 33/4679.1,

  • HOF IP EA PA BDE NHTBH HTBH NCIE All0

    5

    10

    15

    100

    120

    MA

    D (

    kca

    l/mol

    BENCHMARK SET

    LDA BLYP PBE

    First-principles vs semiempirical

    PBE vs BLYP

    BLYP 4 parameters

    HOF = heat of formation;IP = ionization potential,

    EA = electron affinityPA = proton affinity,

    BDE = bond dissociation energy, NHTBH, HTBH = barrier heights for reactions,

    NCIE = the binding in molecular clusters data from Goddard III

  • Alternative to PBE ex

    PBE Perdew PRL 1996 k= 0.804 =0.21951revPBE Yang PRL 1996 k=1.245, =0.21951RPBE Hammer PRB 1999 k= 0.804 =0.21951mPBE Adamo JCP 2002 C1=0.804, C2=-0.015, a=0.157 PBEsol Perdew PRL 2008 k=0.804 =0.1235SOGGA Truhlar JCP 2008 k=0.552, =0.1235

    2exp11 sF RPBEx

    2

    2

    2

    22

    2

    1 111

    as

    sC

    as

    sCFmPBEx

    2

    2,

    11

    s

    sF PBEsolrevPBEx

    Bonding region

    Lieb-Oxford Limit

    RPBEx

    PBEx

    SOGGAx FFF 2

    111

  • 106

    JCP 2002

    Alternative to PBE exch

  • How to improve the exchange : Hybrid Functionals

    How does it work? Very bad!Mean average error 32 kcal/mole for the G2 ensemble (50 molecules) while aMAE of 5-7 kcal/mol is obtained with standard GGA

    KSC

    exactXXC EEE

    Idea: HF exchange is exact. Therefore we could think to combine exact (HF) exchange with a (GGA ) correlation functional (Lie & Clementi 1974)

    Actually combining a percentage of HF exchange with a GGA exchange works much better!!!

    Theoretical justification : adiabatic connection (Becke 1993 – Half and Half)

    In practise : HF% between 20 and 30%

    107

  • 108

    From Yang

    The Adiabatic Connection

  • 109

    Adiabatic Connection

    9110

    3 )1( PWccLSDc

    Bxx

    HFxxo

    LSDxx

    Bxc EaEEaEaEaE

    3 parameters fitted on G2 (atomization energies) with the PW91 correlation

    Becke, JCP 1993

  • First-principles hybrids: the PBE0 model

    Perdew’s Hypothesis(JCP 1996)

    Our working-horse functional

    1,, 1 nDFAxxDFAxchybxc EEEE lll

    )10(.... 11102, llll nnxc ccceE

    DFAxxDFAxchybxchybxc EEnEdEE 11

    0 ,ll

    1n

    The optimum integer n is derived from a pertubation theory on l-dependence of Exc,l

    The lowest order n=4

    DFAxxDFAxchybxc EEEE 41

    DFA=BLYP then B1LYP (CPL 1997)DFA=PBE then PBE0 (JCP 1999)

  • PBE0: a fair play

    PBE0 = about 3000 citations

    Called PBE1PBE, later PBEh

    Called PBE0

  • LYPcc

    LSDc

    Bxx

    HFxxo

    LSDxx

    LYPBxc EaEEaEaEaE 10

    3 )1(

    PBE0

    PBEc

    PBEx

    HFx

    PBExc EEEE 4

    3

    4

    10

    B3LYP

    3 additional parameters fitted on G2 (ionization and atomization energies)

    No fitted parameters

    Normally hybrid functionals outperform all GGA and meta-GGA functionals and they are the reference for chemical applications

    112

    PBE0 vs B3LYP

  • HOF IP EA PA BDE NHTBH HTBH NCIE All0

    4

    8

    12

    20

    24

    BENCHMARK SET

    MA

    D (

    kca

    l/mo

    l

    BLYP PBE B3LYP PBE0

    First-principles vs semiempirical

    PBE0 vs B3LYP

    B3LYP 7 parameters

    HOF = heat of formation;IP = ionization potential,

    EA = electron affinityPA = proton affinity,

    BDE = bond dissociation energy, NHTBH, HTBH = barrier heights for reactions,

    NCIE = the binding in molecular clusters

    B3LYP & PBE0 closer than BLYP & PBE

    data from Goddard III

  • Illas et al J. Chem. Phys. 129, 184110 (2008)

    JPBE0-6971-762-77

    2339-1313

    -49170

    3-35-16

    -346-492

    Illas et al. J. Chem. Phys. 129, 114103 (2008)

    J=2(E(BS)-E(T))

    PBE0 vs B3LYP

  • EPR parameters: g-tensor (Dipeptide analogue of glycyl radical, HCO–GlyR–NH2 )

    JCP 2004

    Critical GS Properties: EPR parameters

  • Excited state properties: vertical absorptions

    A large series of organic dyes

    JCTC 2008, ACR 2009, JCTC 2010

  • HF PBE PBE0 LC-PBE LC-wPBECAM-B3LYP0

    20

    40

    60

    80

    100

    120

    140

    160

    180

    MA

    E (

    nm

    )

    All Aq AB IG

    Large benchmark 483 molecules, 614 valence exc. 28 functionals: JCTC 2009

    Valence excitations TD-DFT/6-31+G(d,p)/PCM

    Excited state properties: vertical absorptions

  • 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5

    5

    6

    7

    8

    9

    10

    11

    HOMOC

    alc

    . (e

    V)

    Exp. (eV)

    6-31G 6-31G(d) 6-31G(d,p) 6-31+G(d,p) 6-31+G(2d,p) 6-31+G(2d,2p) 6-31++G(2df,2pd) 6-31++G(3df,3pd) Valence Rydberg

    Our suggestion PBE0 up to e(HOMO) + 1eVOur suggestion PBE0 up to e(HOMO) + 1eV

    BENZENE

    PBE0

    JPC A 2007

    Excited state properties: vertical absorptions

  • 300 350 400 450 500300

    350

    400

    450

    500

    l the

    or(n

    m)

    lexp

    (nm)

    PBE0 SLR-PBE0

    Coumarins

    JCP 2006

    TD-PBE0/6-31+G(d,p)/PCM// TD-PBE0/6-31+G(d,p)/PCM

    Analytical TD-DFT first derivatives

    Excited state properties: vertical emission

  • Solids

    Kresse JCP 2007Kresse JPCM 2008

    Bulk moduli & atomization energy

    B3LYP>PBE>PBE0≈HSE

    GAP (semiconductor, insulators)

    HSE

  • Perovskite KNiF3

    KNiF3 : [(Ni2F11K12Ni10)25+ + 430 point charges] 0

    K2NiF4 : [(Ni2F11K16Ni6)21+ + 1154 point charges] 0

    La2CuO4: [(Cu2O11La12Cu6)30+ + 1005 point charges] 0

    All electron

    ECP

    Critical GS Properties: magnetic coupling

    Proportional to the difference in energy between different spin states

    JCP 2004

  • 1.62 1.64 1.66 1.68 1.70 1.72 1.74 1.76 1.78 1.80 1.82 1.840

    5

    10

    15

    20

    25

    30

    35

    40

    45

    J (m

    eV

    )

    Pd

    Hybrid functionals:GGA+ HF exchange

    exchange functionals

    LDA, GGA and correlation functionals

    Exp.

    KNiF3

    JCP 2004

    PBE0

    Critical GS Properties: magnetic coupling

  • 123

    Why does PBE0 work?

    In short: PBE0 as good as B3LYP

  • One more rung: double hybrids

    Grimme’s ideas: add some PT2 contribution in a B3LYP style(JCP 2006)

    222 )1( PTcPLYPB

    cHFxx

    BLYPxx

    PLYPBxc cEbEEaEaE

    MP2 contribution computed with B2LYP orbitalsSame number of parameters as B3LYP

    Savin’s contributions: rigourous derivation + from 3 to 1 parameter(JCP 2011)

    22/1

    2,1 ][][][)1( MPcccxHFx

    DHDSxc EEEEEE llll l

    l

    l cc EE /1If the correlation does not scale with l

    2221 ][)1(][)1( MPccxHFx

    DHDSxc EEEEE llll l

  • A further step PBE0-DH

    22/1

    2,1 ][][][)1( MPcccxHFx

    DHDSxc EEEEEE llll l

    l

    Imposing the Levy-Görling limit

    2331 ][)1(][)1( MPccxHFx

    DHLSxc EEEEE llll l

    2/1

    0][lim GLcc EE ll

    i i

    xi

    ij ji

    eejiHFc

    MPc

    GLc

    fvvEEE

    ˆˆˆ

    4

    122

    22 MPc

    GLc EE 0HFcE

    Linear scaled one-parameter hybrid ][)1(][ 2/1 ll l cMPcc EEE

    JCP 2011a, JCP2011b

  • 0.0 0.2 0.4 0.6 0.8 1.04.0

    6.0

    8.0

    10.0

    12.0

    14.0

    16.0

    18.0

    MA

    E (

    kca

    l/mol

    )

    l

    PBE

    MP2

    PBE0-DH

    JCP 2011a

    20

    4

    1

    4

    7

    2

    1

    2

    1 MPccx

    HFx

    DHPBExc EEEEE

    Using the simplest approx to the adiabatic integral(as in the Becke Half&Half)

    Theoretical l almost the optimal value for atomization energies

    A further step PBE0-DH

  • CRISMAT 27/10/2011

    G2 (148 molecules)

    MAE RMS MAX

    B3LYP 3.0 4.4 -20.0 (SiF4)

    B2PLYP 2.6 4.5 -12.3 (SiF4)

    PBE0 5.3 6.7 -21.6 (SiF4)

    PBE0-DH 5.0 6.7 -24.8 (O3)

    NCB31 MAE RMS MAX

    B3LYP 0.93 1.44 4.48 (par C6H6)

    B2PLYP 0.53 0.77 1.89 (sand. C6H6)

    PBE0 0.70 1.08 2.69 (par C6H6)

    PBE0-DH 0.49 0.74 1.94 (NH3F2)

    HT HAT NS UA DBH24/total

    B3LYP 5.1 (5.1) 7.6 (7.6) 3.8 (3.8 2.6 (1.2) 4.8 (4.4)

    B2PLYP 2.1 (1.3) 2.5 (2.1) 2.4 (2.4) 1.3 (-0.6) 2.1 (1.3)

    PBE0 4.7 (4.7) 6.4 (6.4) 2.3 (2.3) 2.7 (0.4) 4.0 (3.5)

    PBE0-DH 1.5 (1.5) 1.6 (1.2) 1.2 (0.3) 2.3 (-1.2) 1.6 (0.4)

    Some results

    JCP 2011data in kcal/mol

    G2=Atomization EnergiesNCB=Non-Covalent BondingHT= Hydrogen TransferHAT=Heavy Atom TransferNS=Nucleofilic SubstitutionsUA=Unimolecular Reactions

  • CRISMAT 27/10/2011

    set PBE0 PBE0-DH B2PLYPBH76 4.12 1.68 2.24S22 2.37 1.65 1.79RG6 0.42 0.37 0.47HEAVY28 0.65 0.52 0.69ACONF 0.64 0.49 0.50SCONF 0.53 0.47 0.59CYCONF 0.58 0.41 0.22

    Some preliminary results

    data in kcal/mol

    Testing on the large GMTKN30 set (2059 energy data)

    S22 noncovalently bound dimersBH76 barrier heightsRG6 rare gas dimersHEAVY28 heavy element hydridesACONF alkane conformersSCONF sugar conformersCYCONF cysteine conformes

    References data: exp, CCSD(T)/CBS or W1-W4