Comp Notes

Lecture Notes for CBE 60547
William F. Schneider
November 20, 2019
Contents
1 Introduction 1
2 Refresher on Quantum Mechanics 5
3 Hydrogen atom: simplest chemical “thing” 10
4 (Two is too) many electrons 18
5 Practical electronic structure 27
6 Potential energy surfaces 39
7 WFT Beyond Hartree-Fock 55
8 First-principles thermodynamics 57
9 Plane waves and core potentials 69
10 Periodic electronic structure 79
11 Practical supercell calculations 82
12 Surfaces 83
13 Implicit solvation 84
14 Density functional theory 84
15 Electron correlation methods 86
1 Introduction
1.1 What do we care about?
Things chemistry/materials-related:
• What are the properties of atoms?
1
Computational Chemistry ND CBE 60547
• What molecules do they make? What other substances do they make?
• What are the shapes of those molecules? Structures of those solids? Properties of them?
• How do those substances react with each other?
• What are the energies of those reactions?
• What are the rates of those reactions?
• What is the strongest substance?
• How do we make a substance to do. . . .?
• add your own questions. . . .
Things that relate to the chemical properties of substances.
1.2 How are we going to figure these out? With only a computer?
1926 Erwin Schrödinger equation: ĤΨ = EΨ
1929 Paul Dirac, British physicist
The fundamental laws necessary for the mathematical treatment of a large part of physics
and the whole of chemistry are thus completely known, and the difficulty lies only in the
fact that application of these laws leads to equations that are too complex to be solved.
It therefore becomes desirable that approximate practical methods of applying quantum
mechanics should be developed, which can lead to an explanation of the main features
of complex atomic systems without too much computation.
1930’s-1950’s Elaboration, analytical applications
1950’s Computers start to appear for technical applications
1960’s Density functional theory emerges.
1960’s-1970’s Numerical solutions of Schrödinger equation for atoms/molecules—expert users
1980’s “Supercomputer” era—routine use of computational chemistry software becomes possible
Figure 1: Ohio State Cray Y-MP supercomputer, ca. 1989. World’s fastest computer at the time.
333 MFlop top speed, 512 Mb RAM
November 20, 2019 © 2019 W. F. Schneider 2

1990’s “Chemical accuracy” era—very precise solutions routinely available, for a cost! See Schnei-
der, JPC 1994.
1990’s Density functional theory (DFT) allows applications to solids/surfaces/liquids to become
common. See Hass, Science, 1998
1990’s Visualization moves to the desktop
2000’s Computational “screening,” materials discovery (Gurkan, J. Phys. Chem. Lett., 2010),
materials genome (https://materialsproject.org/).
Today Computational chemistry widely integrated into all aspects of chemical, materials, biological
research
Computational chemistry is now so vast it is impossible to cover everything completely. We limit
ourselves to quantum-mechanics-based calculations.
1.3 Our goals

1. Understand when it is appropriate to use quantum-mechanics methods.
2. Be able to state the basic theoretical, mathematical, and numerical concepts behind quantum
mechanical “wavefunction theory” (WFT) and “density functional theory,” (DFT) calculations.
3. Understand the terminology and practical issues associated with doing quantum chemical
simulations.
4. Get hands-on experience with these concepts using popular computational tools of today,
including GAMESS for molecular systems and Vasp for condensed phase systems.
5. Learn how to set up, execute, and analyze results in a modern, Python notebook environment.
6. Learn how to apply the results of quantum chemical simulations to calculate things you care
about.
7. Demonstrate an ability to formulate a problem and apply QM methods to it.
8. Develop the skills to understand a literature paper in the area.
1.4 Reading resources

• These notes
• Chris Cramer, Essentials of Computational Chemistry, Wiley, 2004
• Martin, Electronic Structure, Cambridge, 2004
• Sholl and Steckel, Density Functional Theory: A Practical Introduction, Wiley, 2009
• Kitchin book, http://kitchingroup.cheme.cmu.edu/dft-book/
1.5 Software tools

1.5.1 Notebooks
• jupyter/ipython
• emacs/org-mode

1.5.2 Molecular methods
• Avogadro environment http://avogadro.cc/wiki/Main_Page
• GAMESS code http://www.msg.ameslab.gov/GAMESS/GAMESS.html
1.5.3 Supercell methods

• Vasp code http://www.vasp.at/
• ASE environment https://wiki.fysik.dtu.dk/ase/
1.5.4 Great for getting started

• Webmo http://www.webmo.net/

2 Refresher on Quantum Mechanics
2.1 Why quantum mechanics?
Want to describe “mechanics” (equations of motion) of atomic-scale things, like electrons in atoms
and molecules
Why? These ultimately determine the energy, the shape, and all the properties of matter.
de Broglie wavelength (1924)
λ = h/p = h/mv (1)
h = 6.626 × 10−34 J s(Planck’s constant) (2)
Car Electron
mass m 1000 kg 9.1 × 10−31 kg
velocity v 100 km/hr 0.01 c
typical value on the highway typical value in an atom
momentum p 2.8 × 10−4 kg m/s 2.7 × 10−24 kg m/s
wavelength λ 2.4 × 10−38 m 2.4 × 10−10 m
too small to detect. Classical! Comparable to size of an atom.
Must treat with QM!
How to describe wave properties of an electron? Schrödinger equation (1926)
Kinetic energy + Potential energy = Total Energy
Expressed as differential equation (Single particle, non-relativistic):
~2 2 ∂
− ∇ Ψ(r, t) + V (r, t)Ψ(r, t) = −i~ Ψ(r, t) (3)
2m ∂t
If the potential V is time-invariant, can use separation of variables to show that the steady-state,
time-independent solutions are characterized by an energy E and described by:
~2 2
− ∇ ψ(r) + V (r)ψ(r) = Eψ(r) (4)
2m
Ψ(r, t) = ψ(r)e−iEt/~ (5)
2.2 Postulates of non-relativistic quantum mechanics

2.3 Notes on constants and units
Resource on physical constants: http://physics.nist.gov/cuu/Constants/ Resource for unit
conversions: http://www.digitaldutch.com/unitconverter/
Unit converter available in Calc mode in Gnu emacs highly recommended
Energy units 1 eV = 1.602 18 × 10−19 J = 96.485 kJ/mol = 8065.5 {cm-1 } = 11 064 K kB
2.4 Example: Energy states of a particle in a box

System defined by potential experienced by particle:
V (r) = 0, 0 < x, y, z < L
V (r) = ∞, x, y, z ≤ 0, x, y, z ≥ L

Table 1: Postulates of Non-relativistic Quantum Mechanics

Postulate 1: The physical state of a system is completely described by its wavefunction
Ψ. In general, Ψ is a complex function of the spatial coordinates and time. Ψ is required to
be:
I. Single-valued
II. continuous and twice
R differentiable
III. square-integrable ( Ψ∗ Ψdτ is defined over all finite domains)
IV. For bound systems, Ψ can always be normalized such that Ψ∗ Ψdτ = 1
R
Postulate 2: To every physical observable quantity M there corresponds a Hermitian operator M̂ .

The only observable values of M are the eignevalues of M̂ .
Physical quantity Operator Expression
Position x, y, z x̂, ŷ, ẑ x·, y·, z·
∂
Linear momentum px , . . . p̂x , . . . −i~ , . . .
∂x
∂ ∂
Angular momentum lx , . . . p̂x , . . . −i~ y −z ,...
∂z ∂y
~2 2
Kinetic energy T T̂ − ∇
2m
Potential energy V V̂ V (r, t)
~2 2
Total energy E Ĥ − ∇ + V (r, t)
2m
Postulate 3: If a particular observable M is measured many times on many identical systems is a

state Ψ, the average resuts with be the expectation value of the operator M̂ :
Z
hM i = Ψ∗ (M̂ Ψ)dτ
Postulate 4: The energy-invariant states of a system are solutions of the equation

∂
ĤΨ(r, t) = i~ Ψ(r, t)
∂t
Ĥ = T̂ + V̂
The time-independent, stationary states of the system are solutions to the equation
ĤΨ(r) = EΨ(r)
Postulate 5: (The uncertainty principle.) Operators that do not commute (Â(B̂Ψ) 6= B̂(ÂΨ))
are called conjugate. Conjugate observables cannot be determined simultaneously to arbitrary
accuracy. For example, the standard deviation in the measured positions and momenta of
particles all described by the same Ψ must satisfy ∆x∆px ≥ ~/2.

Table 2: Atomic units common for quantum mechanical calculations (see http://en.wikipedia.
org/wiki/Atomic_units)
Atomic unit SI unit Common unit

Charge e=1 1.6021 × 10−19 C
Length a0 = 1 (bohr) 5.291 77 × 10−11 m 0.529 177
Mass me = 1 9.109 38 × 10−31 kg
Angular momentum ~=1 1.054 572 × 10−34 J s
Energy Eh = 1 (hartree) 4.359 744 × 10−18 J 27.2114 eV
Electrostatic force 1/(4π0 ) = 1 8.987 552 × 10−9 N m2 /C2
Boltzmann constant 1.380 65 × 10−23 J/K 8.617 33 × 10−5 eV/K
3D box → 3 degrees of freedom/coordinates

Schrödinger equation
~2 ∂2 ∂2 ∂2

− + + ψ(x, y, z) = Eψ(x, y, z) (6)
2me ∂x2 ∂y 2 ∂z 2
ψ(x, y, z) = 0, x, y, z ≤ 0, x, y, z ≥ L (7)
A second-order, linear, partial differential equation. Boundary value problem. Solve by separation
of variables. Postulate ψ(x, y, z) = X(x)Y (y)Z(z). Substituting and rearrange to get
~2 1 ∂ 2 X(x) 1 ∂ 2 Y (y) 1 ∂ 2 Z(z)

− + + =E 0 < x, y, z < L (8)
2me X(x) ∂x2 Y (y) ∂y 2 Z(z) ∂z 2
ftn x + ftn y + ftn z = constant → each term must be constant.

Equation for each dimension
~2 ∂ 2 X(x)
− = Ex X(x), X(0) = X(L) = 0 (9)
2me ∂x2
Seek function that twice differentiated returns itself and satisfies boundary conditions.
nx πx
X(x) = sin , nx = 1, 2, 3, . . . (10)
L

n2x π 2 ~2
Enx = (11)
2me L2
Solutions called eigenfunctions (or wavefunctions) and eigenvalues. Characterized by quantum num-
bers, one for each degree of freedom. These (and all QM) solutions have certain special properties,
including that they orthonormal and form a complete set.
Normalization
Seek a constant such that the inner eigenfunction product is unity.
Z L r
2 nx πx 2
C 2
sin 2
dx = C L/2 = 1 → C = ± (12)
0 L L
r
2 nx πx
X(x) = ± sin , nx = 1, 2, 3, . . . (13)
L L
Orthonormal
hXnx |Xn0x i = δnx ,n0x Dirac notation (14)
• Energy increase with number of nodes.

• Is this real? See Ho, J. Phys. Chem. B 2005, 109, 20657. Where is the electron?
Complete set Any function on the same domain that satisfies the same boundary conditions can
be represented as a linear combination of these solutions:
X Z L X
f (x) = Xi (x)f (x)dx Xi (x) = Ci Xi (x) (15)
i 0 i
X
|f i = |Xi ihXi |f i Dirac notation (16)
i

Illustrates idea of a basis set. These functions are the basis in “plane wave” supercell methods.
Three-dimensional solutions
3/2
2 nx πx ny πy nz πz
ψ(x, y, z) = X(x)Y (y)Z(z) = sin sin sin , nx , ny , nz = 1, 2, 3, . . .
L L L L
(17)
(n2x + n2y + n2z )π 2 ~2
E = Ex + Ey + Ez = (18)
2mL2
Properties of solutions:
• Symmetry of system introduces degeneracy in solutions
• Energy depends on volume → pressure!

Figure 2: Energy sates of 3D Particle in a box
3 Hydrogen atom: simplest chemical “thing”

3.1 Schrödinger equation
Place massive nucleus at origin and describe position of electron in spherical coordinates
~2 2

− ∇ + V (r) ψ(r) = Eψ(r) (19)
2me
e2 1
V (r) = − (20)
4π0 |r|

Coulomb potential—our nemesis! Decays slowly with distance. Boundary conditions?
3.2 Analytical solutions

1. Separate: ψ(r, θ, φ) = R(r)Θ(θ, φ)
2. Angular equation L̂2 Θ(θ, φ) = EL Θ(θ, φ)
(a) Θ = Ylml (θ, φ) are “spherical harmonics”, describe angular motion
(b) Azimuthal quantum number l = 0, 1, ..., n − 1, correspond to s, p, d, . . . orbital sub-shells;
angular “shape,” number of angular nodes, angular momentum of electron
(c) Magnetic quantum number ml = −l, −l + 1, ..., l, . . . orientation of orbital
3. Radial equation
~2 d ~2 l(l + 1) e2 1

− + − rR(r) = ErR(r)
2me dr2 2me r2 4π0 r
Solutions are a polynomial * exponential decay. Exponential part called a Slater function. Larger
the exponent, faster the decay. Degree of polynomial determined by principle quantum number
n = 1, 2, . . ..
Energy expression, corresponds to our conventional H atom spectrum
2
1 e 1
En = − 2 = −13.6 eV · 2 , n = 1, 2, . . .
n 2a0 n
Questions: Ionization energy of an H atom? 1s → 2s energy? Thermal populations?

Integrate out angular components to get radial probability function Pnl (r) = r2 Rnl
2 (r)

3 2
Z
hri = rPnl (r)dr = n − l(l + 1) a0
2
Note darn electron doesn’t want to stay in the Coulomb well! Wavefunction extends beyond the
classical region defined by En = V (rclassical ). This phenomenon is called tunneling, is a purely
quantum mechanical effect, is pervasive in chemistry, leading for instance to chemical bonding.
3.3 Variational principle

What if we don’t know where to look to find the R(r)? Or an analytical solution doesn’t exist?
Solve numerically.
l = 0 case, in atomic units:
1 d2

1
− − rR(r) = ErR(r), 0 < r∞
2 dr2 r
Guess something. Must obey appropriate boundary conditions and be a well-behaved function. For
example, a Gaussian:
2
gξ (r) = e−ξr

Table 3: Hydrogen atom
e2 1
V (r) = − ,0 < r < ∞
4π0 r
( )
~2 L̂2

1 ∂ 2∂
Ĥ = − r − 2 2 + V (r)
2me r2 ∂r ∂r ~ r
1 ∂2

2 2 1 ∂ ∂
L̂ = −~ 2 ∂φ2 + sin θ ∂θ sin θ
sin θ ∂θ
ψ(r, θ, φ) = R(r)Yl,ml (θ, φ)
~2 d ~2 l(l + 1) e2 1

− + − rR(r) = ErR(r)
2me dr2 2me r2 4π0 r
2r
Rnl (r) = Nnl e−x/2 xl Lnl (x), x=
na0
Pnl (r) = r2 Rnl
2
n = 1, 2, . . . , l = 0, . . . , n − 1 ml = 0, ±1, . . . , ±l
s 3
2 (n − l − 1)!
Nnl =
na0 2n(n + l)!
L10 = L21 = L32 = . . . = 1 L20 = 2 − x L31 = 4 − x
1 ~2 1 EH 1
En = − 2 2
=−
2 me a0 n 2 n2
p
|L| = ~ l(l + 1), Lz = ml ~

3 2 1 a0
hri = n − l(l + 1)
2 2 Z

Figure 3: H atom wavefunctions
Figure 4: H atom radial probability
Figure 5: Pythonic s (l = 0), p (l = 1), and d (l = 2) spherical harmonics. Color scale from red
to white to blue corresponds to positive to zero to negative sign of wavefunction.

Let’s normalize: −1/2 1/4
∞
8ξ 3
Z
N= gξ2 (r)r2 dr =2
0 π
g̃ξ (r) = N gξ (r)

Now evaluate energy, say for ξ = 1:
hEi = hg̃1 |Ĥ|g̃1 i = −0.096 Hartree
Hmmm, not very good, much higher in energy than true answer of −0.5 Hartree.
Let’s try adding two Gaussians together, with equal weight:
b(r) = Ns (g̃1 (r) + g̃0.5 (r))
Normalize:
hb(r)|b(r)i = Ns2 (hg̃1 |g̃1 i + hg̃0.5 |g̃0.5 i + 2hg̃1 |g̃0.5 i) (21)

= Ns (1 + 1 + 2S) = 1 (22)
1
Ns = p (23)
2(1 + S)
Note appearance of “overlap integral” S = hg̃1 |g̃0.5 i, shows how similar or different gi are.
Re-evaluate energy
hb(r)|Ĥ|b(r)i = −0.306 Hartree
Much closer to the truth!
Could even weight the two Gaussians differently:
c(r) = Ns0 (g̃1 (r) + 1.5g̃0.5 (r))
hc(r)|Ĥ|c(r)i = −0.333 Hartree

Better yet!
Could continue to add Gaussians of various exponents, and could vary weights, or could even add
in any other functions that we want that are “well-behaved.” Would find that no matter what we
do, the “model” energy would be greater than the “true” value. Basis of the variational principle:
For any system described by the Hamiltonian Ĥ and any satisfactory trial wavefunction
Ψ,
hΨ|Ĥ|Ψi
E= ≥ E0
hΨ|i
where E0 is the true ground state energy of the system.
Consequence of the completeness of the solutions of the Schrödinger equation. Extremely important
to us, because we can use the calculus of variations to seek energy-optimal Ψ.

Figure 6: Comparison of exact and approximations 2 H 1s radial function
3.4 Basis functions

Recognize that we are approximating
X
R10 (r) ≈ f (r) = ci φi (r)
i
φi are basis functions and ci are variational parameters, or coefficients.

If we find ci that minimize energy, then we have an optimal approximation to E0 within our basis,
and we are sure that E0 is an upper bound on the truth. Adding more basis functions /must/lower
energy.
Common trade-offs:
1. Want to choose φi that are good approximation to the “truth”
2. Want to choose φi that are mathematically convenient
Slaters Gaussians Plane waves Mixed basis
Accurate Moderate accuracy Poor accuracy
Expensive modest cost cheap!
ADF Gaussian, GAMESS, Vasp, CPMD, FLAPW, CP2k
NWChem, Qchem QuantumEspresso
Virtually all quantum codes work on this principle, and the main differences are in details of
implementation and ancillary functionality provided.
There are exceptions. GPAW for instance solves the QM equations by finite difference expressions
on a numerical grid. Lends itself to parallelization and may be the future. . . remains to be seen!
3.5 Secular equations

Apply variational principle to two basis functions for the H atom:

f (r) = c1 φ1 (r) + c2 φ2 (r)
hf (r)|Ĥ|f (r)i
hEi =
hf (r)|f (r)i
Substitute and solve ∂hEi/∂c1 = ∂hEi/∂c2 = 0. Each gives a linear secular equation in c1 and c2 :

H11 − E H12 − S12 E c1
=0
H12 − S12 E H22 c2
where Hij = hφi |Ĥ|φj i is a matrix element and Sij = hφi |iφj i is an overlap. If Sij = 0, basis is
orthogonal, problem simplifies. If Sij ≈ 1, basis is redundant, not efficient!
Evaluate secular determinant
H11 − E H12 − S12 E
=0
H12 − S12 E H22
Gives a quadratic in E. Yields two solutions, which would be approximations to the 1s and 2s
orbital energies of hydrogen. Back substitution to get coeffients:
1s: hE1s i > E1s,true 1 and c2
c1s 1s
2s: hE2s i > E2s,true c1 and c2

2s 2s
Note we always get one solution for each basis function. Secular matrix grows as the square of the
number of basis functions, gets expensive to find roots.
If basis is not orthogonal, common to orthogonalize. Find linear transformation that makes hSij =
δij . Evaluate
c0 = S 1/2 c → Ĥ 0 = S −1/2 HS 1/2
0 − E H0 c01

H11 12
0 0 =0
H12 H22 c02
H 0 c0 = Ec0
Secular equations reduce to standard linear eigenvalue problem. All the tricks of linear algebra
can be applied to find the eigenvalues (orbital energies) and eigenvectors (wavefunctions). Called
diagonalizing the matrix. Tricks can be used to find the lowest energy roots only.
Same basic idea is used in virtually all calculations on atoms, molecules, . . . . Basis of semi-empirical
and first principles methods.

3.6 Spin
Can’t leave the H atom without mentioning electron spin. Non-relativistic QM gives us three
quantum numbers. Relativity teaches that space and time are equivalent. Relativistic H atom
solutions introduce a 4th degree of freedom that, under many circumstances, decouples from other
three. Call it the electron spin, because it behaves like the electron has an intrinsic quantum angular
momentum with magnitude s = 1/2.
ms = +1/2, “spin up”, α

ms = −1/2, “spin down”, β
ms specifies z component of angular momentum, sz = ms ~.

To fully specify state of H atom, must specify all four quantum numbers.

4 (Two is too) many electrons
Helium: next (after hydrogen) simplest atom
In a sense, we “know” the answer. . . 1s2 . But is this same 1s as H? No! Different nuclear charge,
interactions between the two electrons. This is an approximation and a very convenient shorthand!
4.0.1 Schrödinger equation for He

Wavefunction Ψ(r1 , r2 ), atom energy E.
Define 1-electron operator for each electron, in atomic units. Include kinetic energy of electron and
its attraction to nucleus of charge Z = 2:
1 Z
ĥi = − ∇2i −
2 |ri |
Looks similar to hydrogen atom.
BUT, electrons also repel. Total Schrödinger equation for He:

1
ĥ1 + ĥ2 + Ψ(r1 , r2 ) = EΨ(r1 , r2 )
|r2 − r1 |
Last term accounts for electron-electron electrostatic repulsion. Makes problem non-separable and
really hard to solve. (How many solutions are there?)
Generalize to n-electron atom, in atomic units:
X X 1
Ĥ = ĥi + (24)
|rj − ri |
i j>i+1
ĤΨ(r1 , . . . , rn ) = EΨ(r1 , . . . , rn ) (25)
First summation over all electrons, second gets all electron pairs.
Solutions are many-dimensional functions of the coordinates of all the electrons. Cannot solve this
analytically, although approaches exist (eg quantum Monte Carlo) that can in principle get very
close. Thankfully, though, we can make approximations that work out really well. We’ll look at
three historically important ones.
4.1 The Hartree atom

Simplest approach is to approximate Ψ. Douglas Hartree (1897-1958) writes:
Ψ(r1 , r2 ) ≈ ψ1 (r1 ) · ψ2 (r2 )
So-called Hartree product. Can’t be right. It gives the probability of two electrons being in the
same place as some number > 0! Neglects electron correlation. How to apply?
1. Apply variational principle: What’s the best possible set of ψi ? We’ll say best are the set that
give the lowest expectation value of energy.
hEi = hΨ|Ĥ|Ψi/hΨ|Ψi (26)

δhEi
= 0, ∀i (27)
δψi

2. Lagrange multipliers to impose orthonormality constraint on ψi :
hψi |ψj i = δij (28)
X
L = hEi − ij (hψi |ψj i − δij ) (29)
i,j
δL = 0 (30)
3. Coupled, one-electron Hartree eigenvalue equations for energy-optimal ψi :

n o
ĥi + v̂iHartree ψi (r1 ) = i ψi (r1 ) (31)
XZ 1
v̂iHartree (ri ) = |ψj (r2 )|2 dr2 (32)
|r2 − r1 |
j6=i
Have to solve this for all n electrons of an atom/molecule. “Hartree potential” represents Coulomb
repulsion between electron i and all other electrons, averaged over position of those electrons.
Always positive. This is a mean field approximation. Note appearance of “one electron” energies,
i , kinetic energy plus repulsion of electron with all others. Total energy is sum of these i corrected
to avoid overcounting repulsions:
X 1X
hEi = i − hψi |v̂iHartree |ψi i
2
i i
Presents an obvious difficulty. If we don’t know ψj ahead of time, how can we even construct
Hartree equations, let alone solve them? Hartree offered a numerical solution, in the 1930’s, called
the self-consistent field (SCF) approach:
1. Guess an initial set of ψi , one for each electron (he did this on a grid, and jumping ahead a
bit, allowed each ψi to represent two electrons)
2. Construct Hartree potential for each ψi
3. Solve the n differential equations for n new ψi
4. Compare new to old ψi
5. If the same within a desired tolerance, you are done!
6. If not, return to step 2, using new ψi , and repeat.
Hartree’s father did this by hand for all the atoms of the periodic table, tabulating wavefunctions
and energies for all the electrons in each. See Hartree, Douglas R. The Calculation of Atomic
Structures (1957). For instance, for He, he’d solve one equation, self-consistently, to get one ψ1 ,
and then combine to get Ψ(1, 2) = ψ1 (1)α(1)ψ1 (2)β2 . Tedious! Qualitatively great, quantitatively
not so hot. Mean-field approximation just not so hot.
Nonetheless, basic idea of representing many-body wavefunction in terms of “orbitals,” of setting
up orbital equations, and solving using a self-consistent procedure, remain today at the heart of
virtually all electronic structure calculations. Hurrah Hartree!
Note: It would be very cool to write a simple Python code to illustrate the SCF procedure
for two electrons in an atom. Could be done on a grid or in a basis. See eg http:
// www. users. csbsju. edu/ ~frioux/ scf/ scf-he. pdf .

4.2 The Pauli principle
One big conceptual short-coming of the Hartree model is that it treats the electrons as if they
were distinguishable. QM says electrons are indistinguishable. Furthermore, they have a quantized
angular momentum, called a spin, that is either up or down, making them fermions.
Pauli principle: The wavefunction of a multi-particle fermion system must be anti-
symmetric to coordinate exchange.
Ψ(x1 , x2 ) = −Ψ(x2 , x1 )
Here the coordinate x includes both the position and the spin (up or down, α or β) of the electron.
Sorry Hartree. Can fix for He by writing
Ψ(x1 , x2 ) = ψ1 (r1 )ψ1 (r2 ) (α(1)β(2) − β(1)α(2))
Hey, gentle reader, check, does this work? Yes! Exchanging the coordinates changes the sign but
keeps everything else the same. Normalizing is easy if we take ψ1 to be normalized and recall that
spin functions are orthogonal:
1
Ψ(x1 , x2 ) = √ ψ1 (r1 )ψ1 (r2 ) (α(1)β(2) − β(1)α(2))
2
Note it is impossible to construct an antisymmetric wavefunction in which both electrons have the
same spatial function and the same spin. Two electrons cannot have the same space and spin
variables.
4.3 Slater determinants and Hartree-Fock

Slater determinant a general way to assure that a wavefunction satisfies Pauli principle:
ψ1 (1) ψ2 (1) · · · ψn (1)

1 ψ1 (2) ψ2 (2) · · · ψn (2)
Ψ= √ .. .. .. .. = |ψ1 ψ2 · · · ψn i
n! . . . .
ψ1 (n) ψ2 (n) · · · ψn (n)
Swapping rows swaps coordinates and, by rules of determinants, changes sign.

Let’s compare. Two spin-paired electrons in two different orbitals:
ψ1 (1)α(1) ψ2 (1)β(1)
= ψ1 (1)ψ2 (2)α(1)β(2) − ψ2 (1)ψ1 (2)β(1)α(2)
ψ1 (2)α(2) ψ2 (2)β(2)
Antisymmetric? What happens when the two electrons have the same spatial coordinate?

Two spin-aligned electrons in two different orbitals:
ψ1 (1)α(1) ψ2 (1)α(1)
= (ψ1 (1)ψ2 (2) − ψ2 (1)ψ1 (2)) α(1)α(2)
ψ1 (2)α(2) ψ2 (2)α(2)
What happens now?

Exchange guarantees that two electrons of same spin cannot be in the same place! No such guarantee
for electrons of opposite spin.
4.4 Hartree-Fock equation

Same song and dance:
1. Apply variational principle to Slater determinant. The best ψi are those that minimize the
expectation value of the energy.
2. Use method of Lagrange multipliers to keep ψi orthogonal.
For simplicity, restrict ourself to cases with an even number of electrons N, all spin-paired, so
two electrons in every orbital. Let index j run over all occupied orbitals. Arrive at restricted
Hartree-Fock equations:
n
ĥi + v̂iHartree + v̂ exchange }ψi (r1 ) = i ψi (r1 ) (33)
1
X Z
v̂iHartree (r1 ) = 2 |ψj (r2 )|2 dr2 (34)
|r2 − r1 |
j6=i
XZ 1
v̂iexchange (r1 )ψi (r1 ) = −ψj (r1 ) ψj (r2 ) · ψi (r2 ) dr2 (35)
|r2 − r1 |
j6=i
Yikes! Slater determinant wavefunction results in appearance of the “exchange” operator, which
turns a ψi into a ψj . Exchange operator is not a simple multiplication. Must be solved self-
consistently, and is much harder to do than the simple Hartree expression.
Slight simplification possible, noting that i = j terms cancel out and slightly redefining operators:
n
ĥi + v̂iHartree + v̂ exchange }ψi (r1 ) = i ψi (r1 ) (36)
1
X Z
v̂ Hartree (r1 ) = 2 |ψj (r2 )|2 dr2 (37)
|r2 − r1 |
j
exchange
XZ 1
v̂i (r1 )ψi (r1 ) = −ψj (r1 ) ψj (r2 ) · ψi (r2 ) dr2 (38)
|r2 − r1 |
j
Now “Hartree potential” is the same for all orbitals/electrons. We can define the “charge density”
to be X
ρ(r) = 2 |ψj (r)2 |
j
(units of charge/unit volume, multiply by e to get a charge). The Hartree potential can be written
ρ(r2 )
Z
Hartree
v̂ (r1 ) = dr2
|r2 − r1 |

This is the Coulomb repulsion of an electron with all electrons, including itself ! Called Poisson
equation, well known in classical physics. Because it involves a Coulomb repulsion, will see either
v̂ Hartree or v̂ Coulomb . I’ll often write v̂ Coulomb [ρ(r)], to emphasize that the Coulomb potential is a
functional of the charge density.
(Just to make sure we are following units around, the Coulomb potential has units of energy/charge,
eg in SI it would be J/C and would have e/4π0 in front.)
The “exchange potential” cancels out the “self-interaction” of an electron with itself, and insures that
two electrons of the same spin cannot be in the same place, ie, the wavefunction vanishes whenever
the spatial coordinates of two electrons are the same. It cannot be written simply in terms of the
charge density.
4.4.1 Basis of wavefunction theory (WFT)

Hartree-Fock model is much better than Hartree alone, widely implemented in codes. Not par-
ticularly good by today’s standards. However, it is systematic and rigorous, by requiring exact
adherence to the Pauli principle, and it can be systematically improved. It is the foundational
basis of all wavefunction theory (WFT) models, all of which are characterized by exactly treating
exchange. The only approach some people call ab initio.
4.5 Hartree-Fock-Slater
In 1951 John Slater introduced an approximation to the Hartree-Fock model that turned out to
anticipate a whole new approach to solving the electronic structure problem, called density functional
theory.
Rewrite exchange part as (and shorten “exchange” to “x”):
XZ 1
v̂ix (r1 )ψi (r1 ) = −ψj (r1 ) ψj (r2 ) · ψi (r2 ) dr2 (39)
|r2 − r1 |
j
XZ 1 ψi (r1 )ψi (r1 )
= −ψj (r1 ) ψj (r2 ) · ψi (r2 ) dr2 · (40)
|r2 − r1 | ψi (r1 )ψi (r1 )
j
ρxi (r1 ; r2 )
Z
= − dr2 ψi (r1 ) (41)
|r2 − r1 |
X ψi (r1 )ψj (r2 )ψj (r1 )ψi (r2 )
ρxi (r1 ; r2 ) = (42)
ψi (r1 )ψi (r2 )
j
This looks like the Coulomb expression, but the density thing is different for each orbital i. The
“exchange density” does have units of charge density enters in the same way, but with minus sign,
to the electron density. Suggests that exchange can be thought of as an electron “hole” around an
electron. This exchange density has some special properties:
1. Every electron at any position r1 has an exchange hole around it equal to one electron of the
same spin as itself: Z
ρxi (r1 ; r2 )dr2 = 1

2. The exchange hole exactly cancels out all electrons of the same spin at the electron location.
Ie, it exactly fixes self-interaction:
X
ρxi (r1 ; r1 ) = |ψj (r)2 |
j
Thus, the Coulomb repulsion felt by an electron is diminished by an exchange hole that follows
the electron around, exactly canceling out the charge at its current location. It’s not necessarily
spherical and is not the same for all orbitals, but the fact that it has these general properties gives
hope that it can be approximated somehow.
Hey, I have an idea! (Actually, Slater had an idea.) What if we had a homogeneous (density
the same everywhere) gas of electrons, like electrons of a given density ρ in an infinite box? By
symmetry the exchange hole would be spherical, and if it must integrate to 1, then it must have a
radius (factor of 2 comes from fact we are only including electrons of the same spin):
1/3
4πρ/2
Rhole =
3
The potential felt by an electron due to this spherical hole is
9πρ 1/3

1
Z
ρ
v̂ x = − dr = −
2 sphere r 4
Now, let’s assume that an electron in a real system experiences an exchange hole potential at any
point exactly like that of a homogeneous electron gas of the same density at that point. This is the
basis of the Hartree-Fock-Slater model:

3 3ρ(r1 )
v̂ x,HF S (r1 ) = − − Cρ(r1 )1/3
2 π
Some ambiguity as to the right value of the constant C, so sometimes just taken as a parameter.
Can now write the Hartree-Fock-Slater equation:
n o
ĥ + v̂ Coulomb [ρ] + v̂ x,HFS [ρ] ψi (r) = i ψi (r)
This is much simpler to solve than Hartree-Fock equation, because the left hand side is the same
for all electrons given a total density ρ(r). Still must be solved iteratively, using the self-consistent
field.
4.5.1 Notes
1. Exchange potential scales with the total number of electrons around: more electrons (like near
a nucleus) means a more compact, denser exchange hole, more electron exchange “screening,”
a greater decrease in potential energy. Further from nucleus, more diffuse exchange hole, less
screening.
2. Screening is not exact, though; does not exactly cancel self-interaction. Clearest in one-
electron case: Coulomb and exchange potentials should then exactly cancel, which they evi-
dently do not! HFS energy of an H atom is not exactly 0.5 au!

3. From a computational point of view, the exchange potential goes from being the hard thing
to evaluate to being the easy thing. The Coulomb potential takes more effort to evaluate, and
tricks are often implemented to simplify that, like fitting the density to an auxiliary basis set.
On the other hand, the 1/3 power makes the exchange potential non-analytic, and solution of
the HFS equation (and all DFT methods) involves some form of numerical quadrature.
4. How does the HFS model do? Pretty darn well, in particular for calculating the structures of
things, and it works nicely for things like metals. Not a bad deal! Way to go, Slater!
5. Another aside: back in the day, the numerical implementations of Xα were very crude and
sometimes gave unreasonable results (like linear water!). Slater still sold it very hard, which did
not enamor him or all of DFT to the chemical community, although the physics community was
far more accepting. For many years DFT was unaccepted by chemists, until solid numerical
implementations in codes like Gaussian brought it to the mainstream.
4.5.2 Basis of density functional theory (DFT)

Slater’s arguments are not rigorous. However, as we will see later, they can be made rigorous. HFS
is the very simplest example of a density functional theory model, because it is a model built entirely
on charge density. Such approach is justifiable.
4.6 Implementations
4.6.1 GAMESS
Hartree-Fock method always paired with basis set methods and implemented in the codes available
at http://webmo.net. Example GAMESS input for Hartree-Fock Ar:
$CONTRL SCFTYP=RHF RUNTYP=ENERGY ISPHER=1
ICHARG=0 MULT=1 COORD=CART $END
$BASIS GBASIS=CCT $END
$DATA
Ar
OH 1
Ar 18 0.00000000 0.00000000 0.00000000

$END
And for Hartree-Fock-Slater Ar:
$CONTRL SCFTYP= RHF RUNTYP=ENERGY DFTTYP=Slater ISPHER=1
ICHARG=0 MULT=1 COORD=CART $END
$BASIS GBASIS=CCT $END
$DATA
Ar
OH 1
Ar 18 0.00000000 0.00000000 0.00000000

$END

4.6.2 FDA
Hartree-Fock-Slater is intrinsically numerical. Historically interesting is the Herman-Skillman code,
that solves the problem numerically on a grid. Available to us as the fda code, see https://www.
chemsoft.ch/qc/fda.htm and ./Resources/00READ.ME.
Ar input:
300 0.0001 30.0
50 0.00001 0.10 0.50 0.682 0.0042
18.0 5
1 0 1.0 1.0
2 0 1.0 1.0
2 1 3.0 3.0
3 0 1.0 1.0
3 1 3.0 3.0
Output at ./Resources/Ar.out.
4.7 Performance
One metric is the ability to predict ionization energies.
Koopman’s theorem: The negative of the energy of an occupied orbital (−i )) approxi-
mates the energy to extract an electron from that orbital, ie to ionize the system. The
energy of a virtual orbital approximates the energy to add an additional electron to a
system, i.e. the electron affinity. Assumes no relaxation of orbitals.

Figure 7: HFS vs exact ionization energies
4.8 Correlation
If solved to reasonable precision, both the Hartree-Fock and Hartree-Fock-Slater models work pretty
darn well for things like the shapes of molecules, structures of solids, charge distributions, vibrational
frequencies, . . . . Don’t work so well for computing things that involve making and breaking bonds,
like a reaction energy or an activation energy.
Why? All the models discussed here neglect electron correlation, the fact that the potential felt by
an electron is a function of the instantaneous positions of all the other electrons. The contribution of
correlation to absolute energies is not big by proportion, but it is very imporant to energy differences.
Any “orbital” model cannot capture correlation. It can be introduced systematically and exactly
into H-F models (at great computational expense) and systematically and approximately into DFT
models (at much more modest expense). Hence the popularity of DFT!

5 Practical electronic structure
5.1 Born-Oppenheimer approximation
In principle all nuclei and electrons should be described quantum mechanically. For H2 , for instance,
true wavefunction would be a function of the positions of nuclei and electrons, Υ(r1 , r2 , Rα , Rβ ).
Nuclei much heavier than electrons and move much more slowly. Assume nuclei are fixed in space
(“clamped”) and electrons move in static field of those electrons. Equivalent to assuming that nuclear
kinetic energy is decoupled from electron dynamics. Only change is that
1 X Zα
ĥ = − ∇2 −
2 α
|r − Rα |
Schrödinger equation becomes parameteric in nuclear positions; solutions E(Rα , Rβ ) define a po-
tential energy surface (PES).
1 X Zα Zβ
EPES (Rα , Rβ ) = ESchr +
2 |Rβ − Rα |
α,β
5.2 Model chemistry

Essentially always start with
n o
ĥ + vCoulomb [ρ] + vexchange [ψi ] + vcorrelation [ψi ] ψi (r) = i ψi (r) (43)

Standard models of today all treat the one-electron and Coulomb pieces exactly and treat the
electron-electron interactions at various levels of approximation.
vexchange vcorrelation
Wave function theory (WFT)
Hartree self-interaction neglect historic
Hartree-Fock exact neglect superceded
MPn, CC exact perturbative state-of-the-art
CI exact variational specialized
Density functional theory (DFT)
Hartree-Fock-Slater [ρ4/3 ] neglect historic
Local density approximation [ρ4/3 ] [ρ] general purpose solids
(LDA)
Generalized gradient approximation [ρ, ∇ρ] [ρ, ∇ρ] general purpose solids/surfaces
(GGA)
“Improved” GGA [ρ, ∇ρ] [ρ, ∇ρ] general purpose
(RPBE, BEEF, Mxx)
Hybrid ≈ exact [ρ, ∇ρ] general purpose molecules
(B3LYP, PBE0, HSE06) specialty solids/surfaces
Meta GGA [ρ, ∇ρ, ∇2 ρ] [ρ, ∇ρ, ∇2 ρ] developing
The choice of the electronic structure model is the most fundamental approximation in applying
these methods. Determined from experience and need.
Specification in GAMESS (https://www.msg.chem.iastate.edu/GAMESS/GAMESS.html) is a bit ar-
cane. Default is Hartree-Fock. To specify DFT model, use
$CONTRL DFTTYP = Slater (HFS), SVWN (LDA), PBE (GGA), B3LYP (Hybrid), M06 (Minnesota optimize
5.2.1 Beyond Hartree-Fock

Many methods available. See manual for full description. Most common is second-order perturba-
tion theory, “MP2”:
$CONTRL MPLEVL=2 $END
If you want a very high quality number, have a big computer and time to wait, try “coupled cluster”:
$CONTRL CCTYP=CCSD(T) $END
5.3 Bring back the basis sets

The one-electron equations eq 5.2 give us defining expressions for the energy-optimal orbitals, but
they aren’t convenient to solve for anything more complicated than an atom. Expand solutions in
a basis set: X
ψi (r) = Cµi φν (r)
ν
Often atom-centered. You’ll see the term “linear combination of atomic orbitals,” LCAO.

ˆ
Abbreviate f ψi = i ψi . Substitute in ψi , mulitple through by a basis function φµ :
X X
Fµν Cνi = i Sµν Cνi , F C = SC
ν ν
where
Fµν = hψµ |fˆ|φν i Sµν = hψµ |φν i
Matrix equation to solve.
Historically interesting, “semi-empirical” methods (MNDO, . . . ) worked by parameterizing the
matrix elements against atom properties.
Recall fˆ depends on the density, which can be written
X X
ρ(r) = Pµν φµ (r)φν (r), Pµν = 2 Cµi Cνi
µν i
Depending on implementation, pieces of fˆ can often be computed just once and reused, eg one-
electron integrals hφµ |ĥ|φν i.
Algorithm:
1. Put your atoms somewhere in space
2. Select a basis
3. Pre-compute what you can
4. Guess some coefficients/density/density matrix
5. Construct secular matrix elements
6. Solve secular matrix equation for C and
7. Construct and compare new density to old
8. Update density and repeat, or . . .
9. . . . if less than tolerance, all done!
ALWAYS check to be sure result has converged, to the state you want!
5.4 H2O Energy Example

Hartree-Fock calculation on H2 O, minimal (STO-3G) basis set.
! File created by the GAMESS Input Deck Generator Plugin for Avogadro
$BASIS GBASIS=STO NGAUSS=3 $END
$CONTRL SCFTYP=RHF RUNTYP=ENERGY COORD=CART $END
$DATA
Title: H2O energy evaluation
C1
O 8.0 -0.89600 3.13196 0.00000
H 1.0 0.07400 3.13196 0.00000
H 1.0 -1.21933 3.71670 0.70316
$END
See ./Resources/H2O-STO3G.gamout.

5.5 Symmetry
Often problem can be simplified by taking advantage of symmetry of the system.
! File created by the GAMESS Input Deck Generator Plugin for Avogadro
$CONTRL SCFTYP=RHF RUNTYP=ENERGY COORD=CART $END
$DATA
Title: H2O energy evaluation
CNV 2
O 8.0 -0.89600 3.13196 0.00000

H 1.0 0.07400 3.13196 0.00000
H 1.0 -1.21933 3.71670 0.70316
$END
Results get labeled by symmetry labels.
See ./Resources/H2O-C2V.gamout.
5.6 Examples
5.6.1 Dissociating H2+ example
Compute energy vs distance. Should dissociate to H atom and H+ ion.
Oops, come on, LDA! Illustrates self-interaction problem in LDA. Electron is too eager to be diffuse,
spreads out over both atoms when it should localize on one.

5.6.2 Dissociating HHe+ example
Compare Hartree-Fock and LDA for H−He+ vs distance. (Isoelectronic to H2 , but avoids any
problems with symmetry. Should dissociate to H+ and He. Does it?
$CONTRL SCFTYP=RHF RUNTYP=ENERGY ICHARG=1 MULT=1 $END
$DATA
Title
C1
H 1.0 0. 0. 0.
He 2.0 0. 0. XXX
$END
Equilibrium distance? How’s the dissociation state? Bond energy? Truth is about −0.075 Hartree.
LDA has advantage of cancellation of errors between exchange and correlation errors. A good thing!

5.6.3 Dissociated He2 example
5.7 Open-shell systems

First, some jargon related to unpaired electrons:
# unpaired electrons S 2S + 1 name
0 0 1 singlet
1 1/2 2 doublet
2 1 3 triplet
3 3/2 4 quartet
Model has to be generalized somewhat to deal with systems with unpaired electrons. One approach
is to construct wavefunctions that are exactly spin-adapted (eigenfunctions of the Ŝ operator).
Possible in the Hartree-Fock world, but messy. More common is to relax that constraint a bit,
define different orbital wavefunctions for spin-up and spin-down electrons, called unrestricted or
spin-polarized (opposite of non-spin-polarized !). Means that electron density has different spin-up
and spin-down parts.

Controlled in GAMESS using the $CONTRL group:

$CONTRL SCFTYP = RHF non-spin-polarized, default
SCFTYP = UHF spin-polarized
MULT = 1 (default), 2,... spin multiplicity = 1 + number of unpaired electrons
ICHARG = 0 (default), 1,... net charge
$END
5.8 Gaussian basis sets

2
Gaussian functions (e−ζ|r| ) are the most popular choice for atom-centered basis sets. They do not
efficiently represent molecular wavefunctions, but one- and two-electron integrals in WFT can be
solved analytically over Gaussians.
Other choices, like Slater functions (e−ζ|r| ) are possible but require numerical quadrature.
Gaussian basis sets have to be created for any given atom and must be used consistently within a
set of calculations.
• Primitive is a single Gaussian function, possibly multipled by a polynomial to look like s,
p, . . . . Defined by an exponent ζ that determines how extensive (small ζ) or compact the
function is.
• Contraction is a pre-set linear combination of several primitive Gaussians.
• Basis set is a predefined set of exponents and contraction coefficients appropriate for some
specific atom.
5.8.1 Gaussian basis set nomenclature

• Minimal basis contains one contracted function for every atomic orbital. STO-3G is the poster
child.
• Double zeta contains two contracted functions for every atomic orbital
• Split valence is more common, single zeta in core, double zete in valence, typical of Pople basis
sets, eg “6-31G”
• Triple-split valence would be “6-311G”

• Spherical vs Cartesian determines whether a d function have 6 or 5 parts (the sixth of which
is an s).
• Polarization functions are functions of one angular momentum greater than the highest angular
momentum occupied states, eg p function for H, or d function for C. Important to capture
the polarization of charge when atoms make molecules. Arcane nomenclature, eg 6-31G(d,p)
or 6-31G**.
• Diffuse functions are small exponent functions to describe anions or loosely bound electrons.
Again argane nomenclature, eg 6-31+G(d,p). Yech.
• Correlation consistent and atomic natural orbital are series of basis sets that are constructured
to be efficient and to improve systematically.
• Complete basis set (CBS) limit is notion of extrapolating energies from a series of systemati-
cally improving basis sets. Very common in high accuracy calculations.
5.8.2 Standard basis sets in GAMESS

Specified in $BASIS group. Some common choices, in increasing level of sophistication:
Name Type Flags

Pople type The most venerable and widely used
STO-3G Minimal GBASIS=STO NGAUSS = 3
3-21G Split valence GBASIS=N21 NGAUSS=3
6-31G(d) Split valence polarized GBASIS=N31 NGAUSS =6 NDFUNC=1
6-311+G(d,p) Triple-split valence GBASIS=N311 NGAUSS=6 NDFUNC=1 NPFUNC=1
polarized and augmented DIFFSP=1
Polarization-consistent Good for DFT

PC0 Split valence GBASIS=PCseg-0 ISPHER=1
PC1 Split valence polarized GBASIS=PCseg-1 ISPHER=1
PC2 Triple split double polarized GBASIS=PCseg-22 ISPHER=1
Correlation-consistent Good for MP2 and beyond

cc-pVDZ Split valence polarized GBASIS=CC2
cc-pVTZ Triple split double polarized GBASIS=CC3
aug-cc-pvDZ augmented with diffuse functions GBASIS=ACC2
Effective core potentials Good for treating heavy atoms

SBKJC Split valence + core potential GBASIS=SBKJC
Hay-Wadt Split valence + core potential GBASIS=HW
Complicated field, which is why the old standards live on in routine calculations. Optimal approach
is to employ a composite model, calibrated by someone else, with well defined set of basis functions
and treatments of exchance and correlation. A composite model pieces together results from a
number of different calculations to estimate a higher accuracy model.

5.9 Electron cores
Low energy “core” electrons typically don’t participate in chemical bonding but can add substantially
to computational cost. Generally seek approximations, especially for heavy elements/metals.
Heart of approach is to partition an atom into core and valence parts. Seek ways to express the
influence of the core on the valence without actually having to compute the core. Essentially seek
to write
v̂ ee ≈ v̂ ee,core + v̂ ee,val
where the core potential is some simpler, composite expression of the influence of the core on the
valence. Typically take the electron cores as “frozen” in the pure atomic states, and express influence
on valence through angular-momentum-dependent operators parameterized against accurate atomic
calculations. Goal is to recover valence wavefunctions with less effort.
5.9.1 Relativistic effects

Relativistic kinetic energy is relativistic total energy minus the rest energy:
q
T = p2 c2 + m20 c4 − m0 c2
Taylor expanding about p2 = 0 gives the first-order mass-velocity correction:
p2 p4
T ≈ −
2m0 8m30 c2
Reduces to non-relativistic result when c → ∞. Electrons near core move at speeds close to c,
second term becomes non-negligible and diminishes their energy. Most important for s states that
penetrate closest to nucleus; they shield nucleus better and other valence states rise up in energy.
Electron spin and orbital magnetic moments also couple when l > 0, leads to spin-orbit coupling
that splits p, d, . . . states into j = l ± s states.
Darwin correction corrects s orbitals for electron and nucleus being at the same point; comes from
solution of full Dirac relativistic equation for the atom.
Relativistic effects typically incorporated implicitly, by including in model for core electrons and
thus capturing their effect on the valence. Spin-orbit, if necessary, added after the fact.
5.9.2 Implementations
Non-relativistic and relativistic effective core potentials (ECPs) available for many elements. These
specify the potential felt by the valence electrons due to the core in terms of radial potential functions
and angular projection operators. Typically these have to be combined with basis functions designed
to work with them.
Most common are Hay-Wadt (LANL) and Stevens-Basch-Krause (SBK). Other more modern ones
also available, like Stuttgart.
Essential to all plane-wave codes, like Vasp, but implemented differently. Will touch on later in
class.

Figure 8: Comparison of Non-Relativistic and Relativistic Atomic States.

5.10 Population analysis
The molecular orbitals contain information that can be helpful in understanding structure and
bonding:
• Charge density most direct representation of electron distribution
X X
ρ(r) = |ψi (r)|2 = Pµ,ν φµ (r)φν (r)
occupied µ,ν
• Moments of charge density (dipole, quadrupole), useful for thinking about molecule-molecule
interactions. (Only exactly defined for neutrals!)
• Electrostatic potential, or Coulomb potential created by electrons and nuclei. More refined
way of thinking about “how spots” on a molecule. Commonly used to parameterize classical
forcefields, by seeking set of atom-centered charges that reproduce calculated electrostatic
potential, as is done with CHELPG. Not uniquely defined.
• Population analyses, which attempt to distribute electrons to individual atoms and possibly
bonds based on decomposition of molecular orbitals. Chemically it is intuitively nice to assign
charge to individual atoms. There is no single “right” way to do this. . . the “charge” on an
atom in a molecule in not uniquely defined! Consider an occupied molecular orbital ψ made
up of two basis functions on two different atoms, α and β:
ψ = cα φα + cβ φβ (44)
hψ|ψi = c2α + c2β + 2cα cβ hφα |φβ i (45)
In Mulliken analysis, c2α is fraction of ψ assignable to the atom of α, c2β fraction assigned to atom
of β. Remainder is the “overlap” population, which is split evenly between the two. Summing over
all occupied orbitals and subtracting nuclear charges gives gross atomic charges.
In Löwdin analysis, basis functions are pre-orthogonalized, so last term vanishes.
Both approaches very sensitive to choice of basis set. Only sensible to compare within a common
model type across molecules.
• Localized orbitals is notion of creating linear combinations of ψ that satisfy some constraint
for being compact. Leads to orbitals that are more naturally “bonding.”
• Natural orbitals a rigorous scheme for orthogonalizing and assigning charge. Based on recog-
nition that there is a set of orthogonal orbitals that optimally describe the density. Localizing
these give natural bonding orbitals. See 06Weinhold.pdf.
• Bader analysis another method, based on a geometric analysis of the total charge density.
Define See Bader, R. F. W. Atoms in Molecules: A Quantum Theory; Oxford University
Press: Oxford, 1990.
5.11 Molecular orbital (MO) diagrams

Correlate molecular orbitals with their parent fragments. Use to be the thing. Seldom now

5.12 Implementation details of SCF methods
Basis is often orthonormalized to eliminate overlap from H-F-R equation; allows equations to be
solved by matrix diagonalization.
Initial density matrix P are obtained by solving an approximate Hamiltonian (like extended Hückel).
Always beware! Initial guess can influence final converged state.
Because the number of 2-electron integrals grows as N 4 , they are sometimes calculated as needed
“on-the-fly”, so-called direct SCF.
Hartree-Fock integrals can be computed analytically in a Gaussian basis. Any other choice of basis,
or any DFT functional, requires integrals to be computed by quadrature. Used to be a real hang-up.
Today, algorithms are very robust to establish grids and do quadrature.
5.13 SCF updating

The SCF procedure is an optimization problem: find set of coefficients that minimizes the total
energy. As discussed above, success depends on a reasonable initial guess for density matrix and
judicious updating. Various strategies can be used to speed and stabilize convergence, like damped
mixing of previous cycles.
Second-order SCF is a convergence acceleration method that requires calculation or estimation of
the first- and second-derivatives of the energy with respect to the orbital coefficients. See e.g.
Chaban et al., Theor. Chem. Accts. 1997, 97, 88-95.
Pulay’s “direct inversion in the iterative subspace,” or “DIIS,” is a popular and powerful acceleration
procedure that extrapolates from several previous Fock matrices to predict optimal next Fock to
diagonalize. An opportunity for machine learning?
Controlled in GAMESS using the $SCF group.
$SCF DIRSCF= .T./.F. controls direct scf
SOSCF= .T./.F. second-order scf
DIIS= .F./.T. direct inversion in the iterative subspace
DAMP= .T./.F. damping, on for initial iterations
$END
Can also control initial guess orbitals. Particularly powerful feature is to restart from converged
orbitals from a previous calculations (.dat file), controlled with $GUESS group.
$GUESS GUESS = HUCKEL construct intial guess from a simple Hamiltonian
= MOREAD read in orbitals from $VEC group.

6 Potential energy surfaces
The potential energy surface (“PES”) is the sum of the repulsive energy of the nuclei and the kinetic
and potential energies of all the electrons:
N N
X X Zα Zβ e 2
EPES (Rα , Rβ , . . .) = Eelec + (46)
Rαβ
α=1 β=α+1
6.1 Specifying atomic positions
F1#
H2# F1#
r3# H4#
r2# a1# r3# H2# H4#
r1#
C1# C2# a2#
d1#
r3# a2#
H3# r3# r2#
H3# H1#
H1# F2#
F2#
6.1.1 Cartesian
Computationally straightforward but don’t correspond with our physical notion of bonds, bends,
etc. Easiest to get out of a piece of software. A molecule has 3N − 6 internal degrees of freedom
(3N − 5 if linear), but Cartesians specify 3N . The extra values correspond to the location of the
center of mass and molecular oriendation. Codes will typically center and reorient the Cartesians.
In GAMESS, would specify Cartesian coordinates for FCH2 CH2 F like this:
$CONTRL COORD=CART $END
$DATA
FCH2CH2F drag calculation
C1
C 6.0 -3.76764 0.33879 0.03727
C 6.0 -2.35246 0.34495 0.03689
F 9.0 -4.72277 0.58147 -1.18012
F 9.0 -1.59909 -0.68487 -0.83662
H 1.0 -4.04387 1.08375 0.75395
H 1.0 -3.92958 -0.71060 0.16941
H 1.0 -2.03786 0.18875 1.04760
H 1.0 -2.09983 1.28759 -0.40187
$END
6.1.2 Internal coordinates

These provide a more intuitive representation and can be convenient when building molecules by
hand. In codes like GAMESS, most commonly defined using “z-matrix” notation. Specify each atom

in terms of its distance, angle, and dihedral angle with three previous atoms.
In Gamess, would specify z-matrix for FCH2 CH2 F like this:
$CONTRL SCFTYP=RHF RUNTYP=ENERGY COORD=ZMT $END
$DATA
FCH2CH2F drag calculation
C1
C
C 1 r1
F 2 r2 1 A1
H 2 r3 1 A2 3 D1
H 2 r4 1 A3 3 D2
F 1 r2 2 A1 3 D3
H 1 r3 2 A2 6 D1
H 1 r4 2 A3 6 D2
r1=1.5386
r2=1.39462
r3=1.11456
r4=1.12
A1=109.54214
A2=111.
A3=110.
D1=120.
D2=-120.5
D3=50.
$END
Particularly convenient when you’d like to “scan” over the value of some coordinate. Variable can
be applied to more than one independent coordinate, if the molecule has symmetry.

6.2 Features of potential energy surfaces
6.2.1 One-dimensional example
F1$ F1$ F1$

d1$
H4$ F2$ H2$ F2$
H4$ H2$
d1$
d1$
H3$ H1$ H3$ H1$ H3$ H1$

18 H2$ F2$ H4$
16
14
12
Saddle&point&
EPES$(kJ/mol)"
Transi4on&state &
10
dE d 2E
g= = 0, H = 2 < 0
8 dq dq
6
dE d 2E
g= < 0, H = 2 < 0
4 dq dq
0 Local&
0 50
Global& 100 150Minimum&
250 200 300 350 400
2
Minimum& g = dE = 0, H = d E > 0
2
dq dq
Note 3-fold periodicity as expected for rotation about a CC single bond. Note too there are some
special points:
• Minima are places where energy bottoms out. More formally, first derivative of energy, or
slope, or “gradient” g = 0, and second derivative, or curvature, or “Hessian” H > 0. These are
the locally stable conformations of the molecule. Note that lowest energy in this case is not
trans, but rather gauche conformations. Are you surprised?
• Saddle points are places where energy is maximized. Physicially, corresponds to “transition
states” connecting low-energy conformations. Gradient g = 0, but curvature H < 0.
6.2.2 Many-dimensional PES features

Gradient becomes vector and Hessian a matrix
∂2E ∂2E
 
∂E ···
 
∂q1 ∂q12 ∂q1 ∂q3N
..  .. .. .. 
g= . H= . . .
  
  
∂E ∂2E ∂2E
∂q3N ∂q1 ∂q3N ··· 2
∂q3N
• gradient is vector tangent to PES. The force on an object is F = −g, so the gradients are
often called the forces. Where gradient (slope) is negative, force is positive, and vice versa.

Figure 9: From Schlegel, J. Comp. Chem. 2003, 24, 1514-1527.
Force always pushes system toward nearest minimum. If the potential is harmonic, then the
force constant k = H, so the Hessian is also called the “force constant.”
• Hessian matrix is real and symmetric. Diagonalization gives eignevalues and eigenvectors.
Eigenvectors give “natural” directions along PES (physically, the harmonic vibrational modes),
and eigenvalues indicate curvature in that direction.
• Minimum on multidimensional PES has gradient vector g = 0 and all positive Hessian eigen-
values.
• First-order saddle point, or transition state, has g = 0 and one and only one negative Hessian
eigenvalues. (Physically, one unique direction that leads downhill in energy.) Must correspond
to lowest-energy point connecting two minima.
• Minimum energy pathway (MEP) or intrinsic reaction coordinate (IRC) is steepest descent
pathway (in mass-weighted coordinates) from saddle point to nearby minima. Path a marble
with infinite inertia would follow.
• Higher order saddle points have g = 0 and more than one negative Hessian eigenvalue. Can
always lead to lower energy first order saddle point. These generally do not have chemical
significance.
In computational chemistry/materials science, it is frequently our job to identify the critical points
(minima and transition states on a PES). In liquids, PES much more flat and lightly corrugated.
Statistical mechanics becomes mores important.
Each distinct electronic state defines its own PES. Remember that there are multiple PES’s for any
atom configuration, corresponding to different electronic states. Sometimes these states can interact,
intersect, giving avoided crossings, conical intersections. Lead to more complicated dynamical
behavior.

6.3 Energy gradients and second derivatives
6.3.1 Gradients
∂Eelec ∂ ∂Ψ ∂ Ĥ ∂Ψ
= hΨ|Ĥ|Ψi = h |Ĥ|Ψi + hΨ| |Ψi + hΨ|Ĥ| Ψi
∂qi ∂qi ∂qi ∂qi ∂qi
• Hellman-Feynmann theorem says that sum of first and last terms vanish, so in principle only
need to compute middle, which involves derivative of Coulomb potential and can be evaluated.
• Pulay forces are forces from first and last terms that appear when basis functions are centered
on atoms. An advantage of plane-wave basis sets, for which these terms vanish.
6.3.2 Hessian
In some electronic structure models can be computed analytically. More commonly, determined
from numerical differentiation of gradients. Implementations typically assume that system is at
minimum.
h must be small enough to stay in the harmonic region, but big enough to avoid numerical noise
swamping the gradients.
For a molecule with N atoms, to construct complete 3N × 3N Hessian, have to evaluate gradients
6N times for two-sided differencing. Each pair of displacements completes one row of Hessian.
Obviously tends to be quite expensive.
To get better precision and accuracy, could calculate more than two displacements, and could fit to
a more complicated function than a harmonic potential.
6.4 Geometry optimization algorithms

6.4.1 Energy-only
• Trudge infers gradient and locates minimum by wandering around
6.4.2 Energy + gradient

• Steepest descent, just march down hill. R0 = R − λg. Safe, very inefficient near minimum.
May do line search to adapt λ.

• Conjugate gradient is steepest descent plus orthogonalization to previous step. Safe, less very
inefficient, common choice when far from minima.
6.4.3 Energy + gradient + higher order

• Quasi-Newton Raphson takes advantages of both first and second derivative information:
R0 = R − H −1 (R)g(R)
Typically do not know Hessian and it is expensive to calculate. Make an initial guess, then update
Hessian with gradient information from each geometry step. “Learning” PES as we go. Generally
converges very rapidly near minima, where surface is not too anharmonic.
• Rational function optimization is similar in spirit, also constructs Hessian, but uses more
sophisticated (than quadratic) guess form of PES to update positions.
• Direct inversion in the iterative subspace (DIIS) uses sizes of QNR steps as estimates of error
and constructs new step from linear combination of previous that minimizes error inferred
from previous steps: X
err(R) = ci Hi−1 gi
i
Generally very efficient in region of minimum. Algorithm can misbehave away from minima, possibly
even converging to nearby saddle points, so often started with conjugate gradient steps.
6.4.4 Machine learning

All of these based on some form of assumed model of underlying PES (Taylor expand to second
order around minimum). Emerging are methods that “fingerprint” structure and construct and
improve energy model with each step. Remains to be seen if a new “standard” emerges.
6.4.5 Global optimizations

Simulated annealling, genetic algorithms, . . . , more on the exotic side.
6.4.6 Convergence criteria

Typically determine convergence by enforcing maximum on each individual force component
GAMESS offers a limited set of algorithms.
$STATPT METHOD = NR ! Quasi-Newton Raphson
= RFO ! rational function optimization
OPTTOL = 0.0001 ! convergence criterion in au/bohr.
HESS = GUESS ! guess an initial Hessian
= READ ! read from $VIB group
$END

6.5 Specify GAMESS calculation type
Specified in $CONTRL group by RUNTYP flag:
Calculation RUNTYP=
Single-point energy Energy
Single-point energy + force Gradient
Geometry optimization Optimize
Linear scan over PES Scan
Energy + second derivative Hessian
Transition state search Sadpoint
Intrinisc reaction coordinate IRC
Note too that specifying EXETYP=CHECK will check your input without actually running the
job.
Hessian (force) calculation can be done analytically or by numerical differentiation of forces, de-
pending on electronic structure method:
$FORCE METHOD = ANALYTICAL
= SEMINUM
VIBSIZ = 0.01 ! step size (bohr)
$END
6.6 Efficient coordinate systems

Optimizations are most efficient in coordinate system that diagonalizes Hessian, so that optimiza-
tions along each direction are (nearly) independent. Large off-diagonal Hessian terms imply strong
coupling. Cartesian coordinates do not reflect physical forces in system and are generally poor
choice/slow convergence for optimizations. Can choose other coordinate systems. Forces/Hessian
always computed in cartesians, so any other coordinate system requires transformations back and
forth.
• Cartesians simplest to implement, consistent performance. Typically only choice in supercell
calculations.
• Z-matrix are easy to define and use for organic molecules with no rings. For 3N − 6 degrees
of freedom, must specify N − 1 distances, N − 2 angles, and N − 3 dihedrals. Typically will
converge much faster than Cartesians for small molecules. Implemented in most molecular
codes, including GAMESS.
$CONTRL COORD=ZMAT NZVAR = 3N-6 $END
• Natural internals are generalizations of z-matrix that are coordinates that approximately
diagonalize Hessian. Hard to generate a priori or automatically. Specification in GAMESS
is arcane, done in $ZMAT.
• Redundant internals are an over-determined set of internal coordinates, like all bond distance,
all angles between bonded atoms, all dihedrals. Easily constructed automatically. Mapping
between Cartesians and redundant internals becomes more complicated; transform from re-
dundant to Cartesians is overdetermined and has to be solved iteratively. Works very efficiently
for molecules though. Implemented (in spirit) in GAMESS.

$CONTRL NZVAR = 3N-6 $END
$ZMAT DLC=.TRUE. AUTO=.TRUE. $END
FCH2CH2F C5H10
Cartesian 13 30
Z-matrix 11 failed
Redundant/DLC 16 failed
6.7 Performance of models for geometries

Generally pretty good! Gross geometries of molecules can be computed with good reliability using
common approximations (H-F, LDA, GGA). Subtler things (does FCH2 CH2 F really prefer trans or
gauche?) can take more care in choice of electronic structure model. https://cccbdb.nist.gov/
is a great place to look for benchmarks.
6.8 Vibrational frequencies

Suppose we are at a minimum on a PES. Near that minimum, we can Taylor expand the PES
and truncate at second order (“harmonic”, or quadratic, approximation). The Schrödinder equation
describing the dynamical motion of the nuclei can be written in terms of displacements from the
equilibrium position, qi = xi − xeq
i , i = 1, . . . , 3N and the Hessian H:
~2 X 1 ∂ 2 1X
Ĥ = − + Hij qi qj
2 mi dqi2 2
i i,j
√
or in mass-weighted coordinates, ξi = mi qi :
~2 X ∂ 2 1X 1
Ĥ = − + H̃ij ξi ξj , H̃ij = √ Hij
2
i
dξi2 2 i,j mi mj
From eigenvalues κi and eignevectors si (“normal modes”) of mass-weighted Hessian, can transform
into 3N one-dimensional problems:
~2 d2
Ĥi = − + κi s2i
2 ds2i
This is one-dimensional harmonic oscillator Hamiltonian, solutions well known.

Table 4: Harmonic oscillator model
1
V (x) = κx2 , −∞ < x < ∞
2
2 /2α2
ψv (x) = Nv Hv (x/α)e−x , v = 0, 1, 2, . . .
√
α = (~2 /κ)1/4 , Nv = (2v v!α π)−1/2
Hermite polynomials
H0 (y) = 1
H1 (y) = 2y
H2 (y) = 4y 2 − 2
Hn+1 (y) = 2yHn (y) − 2nHn−1 (y)
1 √
ν= κ
2π
1
Ev = v + hν, v = 0, 1, 2, ...
2
Do this in 3N Cartesian space, so 6 (or 5) of the normal modes correspond to translations and
rotations of the molecule. If calculation is exact, these will have κi = 0. Numerical errors may
make them somewhat non-zero. If necessary, these can be projected out by transforming Hessian
to internal and back to Cartesian coordinates.
Note it is impossible for molecule to just sit at q = 0. Nuclei are always vibrating about xeq . Gives
zero point energy
1X
ZP E = hνi
2
i
For a chemical bond, κ ≈ 500 N m−1 , hν ≈ 0.8 eV.
6.8.1 Absorption intensities

Vibrational modes can be probed/observed spectroscopically. Intensity of stimulated absorption
from vibrational state i to f given by Einstein coefficient of stimulated absorption:
|µif |2
Bif =
60 ~2
Arises from coupling of electric field of light with dipole of system. Transition dipole moment,
µif ) givenby : µif = hψi |µ̂|ψf iwhere µ̂ is dipole operator,
X
µ̂ = qi r i
where sum runs over all charged particles and qi is each charge. Dipole moment changes as molecule
vibrates. In direction ξ, can write

dµ
µ̂(ξ(t)) = µ(0) + ξ(t) + ...
dξ ξ=0

Use harmonic oscillator wavefunctions for ψi and ψf :

dµ
µif = hψi |µ̂|ψf i = hψi |µ(0)|ψf i + hψi |ξ|ψf i + . . .
dξ ξ=0
First integral vanishes for i 6= f . Second integral provides gross selection rule that intensity of tran-
sition proportional to the dynamic dipole moment along vibrational normal mode ξ and particular
selection rule that intensity of transition is zero unless f = i ± 1. Latter comes from nature of Her-
mite polynomials. At normal temperatures, i = 0, and the only observable vibrational transitions
are 1 → 2.
6.8.2 Performance of harmonic approximation for vibrational spectrum

Harmonic vibrational frequency systematically overestimate experiment. Convolution of harmonic
approximation error: actual PES is not exactly harmonic, and errors intrinsic to electronic structure
model. Errors are typically systematic (see eg https://doi.org/10.1021/jp960976r).
scale factor
HF/3-21G 0.9085
HF/6-31G(d) 0.8929
MP2/6-31G(d) 0.9434
B3LYP/6-31G(d) 0.9613
Relative intensities are generally predicted with good reliability. Model predicts absorption peaks
to be delta functions. Peaks always broadened due to a variety of fundamental and instrumental
considerations. Common in displaying spectra to arbitrarily broaden using a Lorentzian function.
Figure 10: Scaled HF/6-31G(d) vibrational spectrum vs experiment

6.9 Transition states
6.9.1 Symmetry
Can exploit symmetry to force calculation to converge to a transition state. Eclipsed form of
FCH2 −CH2 F has mirror symmetry (Cs ), if initialized in that configuration, optimization must
preserve symmetry.
$DATA
FCH2CH2F eclipsed z-matrix (Cs, or mirror, symmetry)
CS
C
C 1 1.5000000
F 2 1.4000000 1 109.5421400
H 2 1.1000000 1 109.5421400 3 120.0000000 0
H 2 1.1000000 1 109.5421400 3 -120.0000000 0
F 1 1.5000000 2 109.5421400 3 0.0000000 0
H 1 1.1000000 2 109.5421400 6 120.0000000 0
H 1 1.1000000 2 109.5421400 6 -120.0000000 0
$END
See results in ../Labs/Gamess/FCH2CH2F/SYMMETRY.
−→
−277.937 777 074 9 au −277.925 347 844 1 au
∆E = 0.34 eV
6.9.2 Coordinate drag

If reaction state coordinate maps closely onto some internal coordinate, then can do a series of
“constrained” optimizations, fixing the coordinate of interest to a series of values and relaxing all
other coordinates. Use IFREEZ within GAMESS. For example, define FCH2 CH2 F using z-matrix and
freeze F−C−C−F dihedral angle at a series of values.
$CONTRL SCFTYP=RHF DFTTYP=PBE RUNTYP=OPTIMIZE COORD=ZMT NZVAR=18 ISPHER=1 $END
$BASIS GBASIS=PCseg-1 $END
$STATPT IFREEZ(1)=12 $END
$DATA
FCH2CH2F eclipsed TS
C1

C
C 1 r1
F 2 r2 1 A1
H 2 r3 1 A2 3 D1
H 2 r4 1 A3 3 D2
F 1 r5 2 A4 3 D3
H 1 r6 2 A5 6 D4
H 1 r7 2 A6 6 D5
...
D4=120.
...
$END
See results in ../Labs/Gamess/FCH2CH2F/SCAN.
Figure 11: GGA/PCseg-1 rotational scan
A quasi-NR optimization started at one of the approximate TS’s will usually converge to the exact
TS.
$CONTRL SCFTYP=RHF DFTTYP=PBE RUNTYP=OPTIMIZE COORD=ZMT NZVAR=18 ISPHER=1 $END
$STATPT METHOD=NR $END
$DATA
FCH2CH2F zmatrix optimization near saddle point, no hessian
C1
C
C 1 1.5256029

F 2 1.4018052 1 110.3919125
H 2 1.1066523 1 112.5759655 3 119.5903058 0
H 2 1.1070172 1 108.5521191 3 -119.1579857 0
F 1 1.4018888 2 110.3079213 3 120.0000000 0
H 1 1.1065092 2 112.5372024 6 119.5247291 0
H 1 1.1069911 2 108.5976341 6 -119.1645584 0
$END
See results in ../Labs/Gamess/FCH2CH2F/QNR-TS. Always good practice to follow up with a
frequency calculation to confirm.
−→
−277.937 777 074 9 au −277.933 307 329 4 au
∆E ‡ = 0.12 eV
ν = 154i cm−1
Coordinate dragging can fail when the reaction coordinate is non-linearly related to the multiple
internal coordinates. Plus, it is relatively expensive, as it involves a lot of optimizations.
6.9.3 1st order methods (NEB, . . . )

A scan is a first order method that freezes the gradient along a search direction. A better algorithm
would use the information along the search direction to find the TS and the entire MEP. That’s the
spirit of a nudged elastic band calculation, for instance. Also of newer methods to generate more
sophisticated approximations. Touch on that later. . . .
6.9.4 2nd order methods

Hessian-based methods (like quasi-NR and DIIS) generally work more efficiently, assuming you can
find a region reasonably close to the TS and can get a good guess of the Hessian with the appropriate
(1!) number of negative eigenvalues. These work like regular optimization methods, but search uphill
along the negative eigenvector and downhill along the other directions. The Hessian update scheme
has to be appropriately modified to accommodate the negative eigenvalue. The choice of coordinate
system can become even more crucial.
Algorithm:
1. Identify and optimize structures of reactant and product states
2. Guess a structure for the TS, interpolated either mathematically or empirically between re-
actant and product

3. Compute initial TS Hessian (RUNTYP = HESSIAN). Only needs to be approximately correct, so
appropriate to use a less-expensive method.
4. Confirm Hessian has desired number of imaginary modes, or at least an imaginary mode that
maps well onto desired reaction coordinate.
5. Do saddle-point search with guessed Hessian (RUNTYP=SADPOINT, HESS=READ).
6. Check result with final Hessian calculation
$CONTRL SCFTYP=RHF DFTTYP=PBE RUNTYP=SADPOINT COORD=CART ISPHER=1 $END
$FORCE HESS=READ $END
$DATA
H2ON2
C1 1
N 7.0 0.9405942384 -0.3241591975 -0.0156283505
N 7.0 0.0107640781 0.6540868381 0.0152029180
O 8.0 -0.9317086136 -0.2406652788 -0.0189127039
H 1.0 -0.2853795803 -0.9336844866 0.3015543918
H 1.0 1.8537652825 0.1690870887 0.0045149723
$END
$HESS
ENERGY IS -184.6737814637 E(NUC) IS 74.1740447013
1 1 8.31017204E-01 1.46125961E-01 7.06555414E-04-1.42105871E-01 2.88851646E-02
1 2-1.06097190E-03-3.64083936E-01 2.22079857E-02-2.60487424E-03-3.19838288E-02
.....
$END
See Results in ../Labs/Gamess/H2NNO-TS.
−→ −→
−185.630 793 7 au −185.585 420 9 au −185.625 487 8 au
∆E ‡ = ∆E =
1.23 eV 0.14 eV
6.9.5 Well climbing

Algorithms exist that will climb out of basins and seek transition states. . . .
6.10 Intrinsic reaction coordinates

In more complicated systems, it can often be difficult to know exactly what minima a particular
transition state corresponds to. The intrinsic reaction coordinate (IRC) or minimum energy path
(MEP) is steepestdescent path from TS towards both basins. Starts from TS, steps forward in

direction given by gradient, using second order method. Have to use care in selecting algorithm and
step sizes to stay on the path. Can be useful for locating variational transition state. . . configuration
where free energy (rather than energy) is maximized.
$CONTRL SCFTYP=RHF DFTTYP=PBE RUNTYP=IRC COORD=UNIQUE ISPHER=1 $END
$IRC PACE=LINEAR STABLZ=.TRUE. NPOINT=10 SADDLE=.T. FORWRD=.T. $END
$FORCE HESS=READ $END
$DATA
H2ON2
C1 1
N 7.0 0.9233485039 -0.2399318160 0.0736256617
N 7.0 -0.0272197420 0.6088503884 -0.0717581187
O 8.0 -1.0933769256 -0.1081391759 0.0571959479
H 1.0 -0.1059622027 -1.0519964547 0.2191639975
H 1.0 1.8912457716 0.1158820221 0.0085037392
$END
$HESS
ENERGY IS -185.5854194050 E(NUC) IS 73.0033764178
1 1 8.43287379E-01-4.86363743E-02 1.07610516E-04-3.07227749E-01 1.52208588E-01
.....

6.11 Molecular dynamics
Basic idea is to propogate atoms forward in time using some model to compute the forces on atoms
and Newton’s laws to describe kinetic energy of atoms. May be done at constant energy (NVE )
or, by coupling kinetic energy to an appropriate reservoir, at constant temperature (NVT ). Details
beyond this course (perhaps beyond this instructor!). Primary point for us is that cost of electronic
structure calculations is great enough that typically some adjustments must be made to make force
calculations cheap and fast enough to support number of evaluations necessary to do meaningful
dynamics.
6.11.1 Cheap parameters

Simplest trick is to back off on the precision and accuracy of calculations to reduce force evaluation
cost.
6.11.2 Car-Parrinello dynamics

Treat wavefunction itself as a dynamical variable and propogate forward in time along with the nu-
clei. Ideally parallels but does not exactly follow Born-Oppenheimer energy surface. Huge advance
when originally introduced, has fallen out of favor as more conventional B-O approaches compete
more effectively.
6.12 Biased sampling

Ask Whitmer. See https://doi.org/10.1021/acs.jctc.8b00192.

7 WFT Beyond Hartree-Fock
DFT or WFT calculations of PES’s are expensive. Not uncommon to use an heirarcy of methods:
1. Characterize PES using some low-cost model (LDA)
2. Improve structures using some more reliable and expensive method
3. Improve final energies using some even more reliable and expensive method
WFT has the advantage of offering a series of systematically improving models. Starts with Hartree-
Fock ground-state wavefunction and build in “many-body” features/electron correlation by combin-
ing with “excited” Hartree-Fock wavefunctions:
Electron Correlation Summary

Hartree-Fock single determinant
One (spin)-orbital one electron
No electron correlation
Introduce correlation with many-body

wavefunction
Add excited state determinants to H-F
ground state
Add a handful of excited
Add configurations of a
Add all configurations
of a given excitation
given perturbation
configurations
order
type
MCSCF CIS MPn/MBPTn

CASSCF CISD Coupled-cluster (CC..)
Full CI Quadratic CI (QCI..)
Variational Variational Size consistent
Not size-consistent Not size-consistent Not variational
Good for describing static Good for describing excited
correlation—bond breaking states and spectroscopy Good general purpose methods
Not so good for quantitative Not so good for quantitative Excellent structures/energies for
energies energies “normal” things
• static correlation are electron correlation effects that arise from the restrictive form of the H-F
wavefunction
• dynamic correlation is like it sounds, the “dance” of all electrons about one another
• configuration interaction is a variational approach to adding in dynamical correlation by com-
bining H-F determinants. Difficult to apply, rarely used any more.
• size consistency is the property that the energy model scales properly with the system size.
CI lacks this.
• perturbation theory (MPn) is non-variational, size-consistent, and user-friendly approach

• coupled-cluster (CCxxx) is a systematic, size-consistent approach to introducing H-F correla-
tion. Improvement on perturbation theory.
• quadratic configuration interaction (QCIxxx) is an approximate coupled cluster model.
Heirarcy of models:
HF < MP2 ~ MP3 ~ CCD < CCSD < MP4 < CCSD(T) ~
Reliability also tied to basis set completeness.
“Model chemistry” some linear combination of these, often calibrated against experimental data.
“G2” most venerable:
Hartree-Fock MP2 MP4 QCISD(T)
6-31G(d) ZPE Structure
6-311G(d) 1 2 3
6-311+G(d,p) 4 5
6-311G(2df,p) 6 7
6-311+G(3df,p) 8
QCISD(T)/6-311+G(3df,p) ≈ 2 + (3 − 2) + (5 − 2) + (7 − 2) + (8 − 1) − (4 − 1) − (6 − 1)
∆E(HLC) = −2.50 mHa ∗ ( electron pairs) − 0.19 mHa ∗ ( unpaired electrons)

E(G2) = QCISD(T)/6-311+G(3df,p) + 0.8929 ∗ ZPE + ∆E(HLC)
Quality assessed using, eg, mean absolute deviation from some reliable data set. Many elaborations
on this same idea in the literature.

8 First-principles thermodynamics
8.1 Connection Between QM and Thermodynamics
8.1.1 Internal energy
The internal electronic energy of a single molecule from a WFT/DFT calculation is the energy
associated with taking infinitely separated constituent nuclei and electrons at rest and forming a
molecule:
2 H+ + 8 O8+ + 10 e− → H2 O E elec (47)
Calculate E elec within the Born-Oppenheimer approximation, so the nuclei are fixed in space at the
minimum energy configuration.
8.1.2 Zero-point vibrational energy

E 0 = E elec + ZP V E (48)
ZPVE can be calculated reliably within the harmonic approximation, according to
3n−6
1 X
ZPVE = h νi (49)
2
i=1
where νi are the harmonic vibrational frequencies, obtained from a vibrational frequency analysis.
E 0 is the minimum physically meaninfful energy of the molecule.
8.1.3 Finite T energy

Energy can be deposited in a syste as translational and rotational kinetic energy, in excited vibra-
tional modes, in the interaction of a molecule with an external electric or magnetic or gravitational
field, or . . . . If we assume that the energy in these various degrees of freedom are separable, we can
write:
Ei = E 0 + E trans + E rot + E vib + E elec∗ + E ext (50)
To fully describe microscopic energetic state of a system, would have to specify all of these.
Typically want collective properties at equilibrium, like the internal energy U or enthalpy H or Gibbs
energy G, under some external constraints like temperature T or volume V . These thermodynamic
quantities are averages over the energy states of an ensemble of molecules. The way this averaging
is performed is the realm of statistical thermodynamics.
8.1.4 Canonical ensemble

Free variables are the number of molecules N , the total volume V , and the temperature T . Proba-
bility for a molecule to be in some energy state Ei above E 0 is given by the Boltzmann factor,
P (Ei ) ∝ e−Ei β = e−Ei /kB T , β = 1/kB T (51)
Defines an exponentially decaying probability function for a state to be occupied at some tempera-
ture. Temperature is the characteristic of a system following this distribution.

Probability
30
20
10
Energy eV
0.02 0.04 0.06 0.08 0.10
8.1.5 Averages and partition functions

Let’s use this to calculate the internal energy U of a molecule at some temperature.
P
i Ei P (Ei )
U (T ) = P (52)
i P (Ei )
where the denominator ensures that the probability is normalized.

−Ei β
P
i Ei e
U (T ) = P −Ei β
(53)
ie
∂ −Ei β
P
∂β ie
= P −E β (54)
ie
i
P −Ei β
∂ ln i e
= − (55)
∂β
The sum over energy states is evidently a special quantity, called the partition function:
X
Q= e−Ei β (56)
i
All thermodynamic quantities can be written in terms of the partition function.

Table 5: Equations of the Canoncial (N V T ) Ensemble
β = 1/kB T Full Ensemble Distinguishable particles Indistinguishable particles

(e.g. atoms in a lattice) (e.g. molecules in a fluid)
Single particle X X
partition function q(V, T ) = e−i β q(V, T ) = e−i β
i i
Full partition X
function Q(N, V, T ) = e−Uj β Q = q(V, T )N Q = q(V, T )N /N !
j
Log partition ln Q N log q N ln q − ln N !
function ≈ N (ln Q − ln N + 1)
ln Q N ln q N q
Helmholtz energy − − − ln +1
β β β N
(A = U − T S)

∂ ln Q ∂ ln q ∂ ln q
Internal energy (U ) − −N −N
∂β NV ∂β V ∂β V

1 ∂ ln Q N ∂ ln q N ∂ ln q
Pressure (P )
β ∂V Nβ β ∂V β β ∂V β
Entropy (S/kB ) βU + ln Q βU + N ln q βU + N (ln(q/N ) + 1)

1 ∂ ln Q ln q ln(q/N )
Chemical potential (µ) − − −
β ∂N VT β β
NOTE! All energies are referenced to their values at 0 K. Enthalpy H = U + P V , Gibb’s Energy
G = A + PV .

The good news: if we know Q we can find any
thermodynamic quantity.
The bad
Computational news:
Chemistry can be difficult to calculate for some ND CBE 60547
8.2 Stat Mech applied to stuff
systems
Ideal Gas Bulk Liquid Macroscopic Solid
H
H
H
N N O
O
O
H H
H
H O
O
H
H
No intermolecular interactions Need all inter/intramolecular No rotation or translation.
Only sum over states states, MC/MD to get thermo- Sum over vibrational modes
of individual molecule dynamic quantities
8.2.1 Separability
in principle need to sum over all the types of energy states (translational, rotational, vibrational,
. . . ) of every molecule. Seemingly impossible task. One simplification is if we can write energy as
sum of energies of individual elements (molecules) of system:
Paolucci Thermodynamics and Kinetics March 18, 2014 5 / 14
Ej = j (1) + j (2) + ... + j (N ) (57)
X
Q(N, V, T ) = e−Ej β (58)
j
X
= e−(j (1)+j (2)+...+j (N ))β (59)
j
If molecules/elements of system can be distinguished from each other (like atoms in a fixed lattice),
expression can be factored:
   
X X
Q(N, V, T ) =  e−j (1)β  · · ·  e−j (N )β  (60)
j j
= q(1) · · · q(N ) (61)

Assuming all the elements are the same: (62)
= qN (63)
If not distinguishable (like molecules in a liquid or gas, or electrons in a solid), problem is dif-
ficult, because identical arrangements of energy amongst elements should only be counted once.
Approximate solution, good almost all the time:
Q(N, V, T ) = q N /N ! (64)
Sidebar: “Correct” factoring depends on whether individual elements are fermions or bosons, leads
to funny things like superconductivity and superfluidity.
This q(V, T ) is the molecular partition function, and is calculated by summing over the individual
energy states of a single molecule (starting at E0 ).

Further simplified by factoring into contributions from various (3N ) molecular degrees of freedom:
! ! ! !
X X X X
q(V, T ) = e−etrans β e−erot β e−evib β e−eelec β (65)
trans rot vib elec
= qtrans qrot qvib qelec (66)
U = E0 + Utrans + Urot + Uvib + Uelec (67)
Similarly for other thermodynamic quantities, for example,

∂U
Cv = = Cv,trans + Cv,rot + Cv,vib + Cv,elec (68)
∂T V
Thermodynamic quantities are sums of contributions from indvidual degrees of freedom.
Have to somehow model these motions and have to use our quantum mechanical results to param-
eterize the models.
8.3 Ideal gas

8.3.1 Ideal gas of molecules
Assume molecules are indistinguishable and that internal energy is seperable
(qtrans qrot qvib qelec )N
Qig (N, V, T ) =
N!
F (N, V, T ) = Ftrans + Frot + Fvib + Felec
for any thermodynamic function F .
8.3.2 Electronic partition functions → spin multiplicity

Governed by Fermi-Dirac distribution. Electronic degeneracy at normal T.
8.3.3 Vibrational thermodynamics: harmonic oscillator

Energy spectrum above ZPE given
Ev = hvν, v = 0, 1, 2, ... (69)
Define characteristic temperature Θ = hν/kB .

∞
X
q(T ) = e−v(Θ/T ) (70)
v=0
1
= (71)
1 − e−Θ/T
(geometric series). Partition function increasing function of T. Look at limits.
Internal energy:
∂ ln q
U (T ) = − (72)
∂β
0
= (73)
e0 β − 1

Need vibrational spectrum to get these contributions.
Results only as good as H-O model! For low frequency modes, errors can be substantial, esp for
entropy. Can apply more sophisticated models.
8.3.4 Rotational thermodynamics: rigid rotor

Characteristic temperature Θrot = ~2 /2IkB . Moment of intertia I only depends on shape or
molecule—geometry optimzation.
“High” T qrot (T ) ≈ σΘrot /T , most often true
8.3.5 Translational states: particle-in-a-box

n2 π 2 ~2
En = = 0 n2 , n = 1, 2, 3, . . .
2mL2
Θtrans = 0 /kB is typically tiny, allows partition function sum to be approximated by integral:
Z ∞
2
qtrans,1D ≈ e−x β0 dx = L/Λ
0
h
Λ= √ thermal wavelength
2πmkB T
qtrans,3D = V /Λ3
Thermal wavelength Λ depends only a molecule mass and is of the order the box dimensions at
which quantization is evident. Typically a tiny number (eg 1.7 × 10−11 m for Ar in a 1 L volume at
298 K. qtrans is thus enormous: lots of translational freedom. q depends on volume, introduces vol-
ume/concentration/pressure dependence into thermo functions. Conventional to define a standard
state V ◦ volume, or corresponding pressure.
8.4 Chemical reactions and equilibria

8.4.1 Chemical reaction
1. General chemical reaction = 0, νi stoichiometric coefficients
P
i νi Ai
2. Thermodynamic change ∆W ◦ (T ) = i νi Wi◦ (T ), where W = A, U, S, G, . . .

P
3. “Standard state” derives from concentration dependence of entropy

4. “Standard state” corresponds to some standard choice, (N/V )◦ = c◦ , e.g. 1 mol/l (T-independent),
or (N/V )◦ = P ◦ /RT , e.g. 1 bar (T-dependent)
5. Permits functions to be easily computed at other concentrations, e.g.
A(T, N/V ) = A◦ (T ) + kT ln ((N/V )/(N/V )◦ ) = A◦ (T ) + kT ln (c/c◦ )
6. Example: ethane dehydrogenation, C2 H6 −−→ C2 H4 + H2 , 1 bar standard state

7. Reaction entropy captures contributions of all degrees of freedom
8. Reaction energy (internal, Helmholtz, . . . ) must also capture difference in 0 K electronic
energy
∆U ◦ (T ) = UB◦ (T ) − UA◦ (T ) + ∆E(0)

Table 6: Statistical Thermodynamics of an Ideal Gas

Translational DOFs 3-D particle in a box model
1/2
π 2 ~2

β
θtrans = ,Λ=h
2mL2 kB 2πm
For T >> Θtrans , Λ << L, qtrans = V /Λ3 (essentially always
! true) !
3 3 5/2
e V ◦ e5/2 kB T
◦
Utrans = RT Cv,trans = R Strans = R ln = R ln
2 2 N ◦ Λ3 P ◦ Λ3
Rotational DOFs Rigid rotor model
Linear molecule θrot = hcB/kB
∞
1, unsymmetric

1X 1 T
qrot = (2l + 1)e−l(l+1)θrot /T , ≈ , T >> θrot σ=
σ σ θrot 2, symmetric
l=0
◦
Urot = RT Cv,rot = R Srot = R(1 − ln(σθrot /T ))
Non-linear molecule θrot,α = hcBα /kB
1/2
πT 3

1
qrot ≈ , T >> θrot,α,β,γ σ = rotational symmetry number
σ θrot,α θrot,β θrot,γ

3 3 ◦ R σθrot,α θrot,β θrot,γ
Urot = RT Cv,rot = R Srot = 3 − ln
2 2 2 πT 3
Vibrational DOFs Harmonic oscillator model
Single harmonic mode θvib = hν/kB
1 T
qvib = ≈ , T >> θvib
1− e−θvib /T θvib
Uvib = Cv,vib = ◦
Svib,i =
!2
θvib eθvib /2T

θvib θvib /T
R θ /T R R − ln(1 − e−θvib /T )
e vib − 1 T eθvib /T − 1 e /T − 1
θ vib
Multiple harmonic modes θvib,i = hνi /kB

Y 1
qvib =
i
1− e−θvib,i /T
Uvib = Cv,vib = ◦
Svib,i =
!2
X θvib,i X θvib,i eθvib,i /2T
θvib,i /T

−θvib,i /T
R R R − ln(1 − e )
i
eθvib,i /T − 1 i
T eθvib,i /T − 1 eθvib,i /T − 1
Electronic DOFs qelec = spin multiplicity

Table 7: Contributions to ideal gas thermodynamics
Characteristic Characteristic States @ RT

Energy (cm-1 ) Temperature (K)
translational ~2 /2mL2 ≈ 10−21 10−21 1030 classical limit
rotational ≈1 ≈1 100’s semi-classical
vibrational ≈ 1000 ≈ 1000 1 non-classical
electronic ≈ 10, 000 ≈ 10, 000 1 non-classical
Table 8: Ethane thermodynamics

Table 9: Ethane to ethylene plus hydrogen standard state (1 bar) thermodynamcs

8.4.2 Chemical equilibrium
1. At chemical equilbrium, total free energy minimized with respect to reaction advancement ξ
X
G(T, ξ) = ξ(∆G◦ + kT νi ln Pi /P ◦ )
i
2. Equilibrium condition—equate chemical potentials
µA (N, V, T ) = µB (N, V, T )
EA (0) − kT ln(qA /NA ) = EB (0) − kT ln(qB /NB )
NB NB /V qB (T, V )/V −∆E(0)/kT
= = e
NA NA /V qA (T, V )/V
3. q/V = 1/Λ3 has units of number/volume, or concentration

4. Equilibrium constant—convert units to some standard concentration c◦ or pressure P ◦
◦
qA (T ) = (qA (T, V )/V )(1/c◦ )
◦
qA (T ) = (qA (T, V )/V )(RT /P ◦ )
q ◦ (T ) −∆E(0)/kT ◦
Keq (T ) = B ◦ e = e−∆G (T )/kT
qA (T )
8.5 Reaction rates

Rate: number per unit time per unit something (volume, area, . . . ). Reactions hypothesized to
occur through sequence of elementary steps.
For example, ozone decomposition, rate second-order at high PO2 , first-order at low PO2
2O3 −−→ 3O2
k
O3 −−1→ O2 + O
k− 1
O2 + O −−→ O3
k
O + O3 −−2→ 2O2
See two compute rates of individual steps.
8.5.1 Transition state theory (TST)

1. Assumptions
(a) Existence of reaction coordinate (PES)
(b) Existence of dividing surface
(c) Equilibrium between reactants and “transition state”
(d) Harmonic approximation for transition state

2. rate proportional to concentration of “activated complex” over reactants times crossing fre-
quency
r = kCA CB
‡
= k ‡ CAB
= ν ‡ K ‡ CA CB
kB T
= ν ‡ ‡ K̄ ‡ (T )CA CB
hν
kB T q ‡ (T )
= e−∆E(0)/kB T CA CB
h qA (T )qB (T )
3. application to two molecules - vinyl alcohol to acetaldehyde

4. microscopic reversibility
5. equilibrium requirement Keq (T ) = kf (T )/kr (T )
8.5.2 Thermodynamic connection

1. Relate activated complex equilibrium constant to activation free energy
◦‡ (T )/kT ◦‡ (T )/k T ◦‡ (T )/k
K̄ ‡ (T ) = e−∆G = e−∆H B
e∆S B
2. Compare to Arrhenius expression

kB T 1 ∆S ◦‡ (T )/kB
Ea = ∆H ◦‡ (T ) + kT, A = e e
h

Table 10: Vinyl alcohol to acetaldehyde

9 Plane waves and core potentials
9.1 Periodic boundary conditions
Free particle moving in one dimension
φG (x) = eiGx
G can take any value. Eigenfunction of momentum and kinetic energy operators. In particular,
~2 2
EG = G
2m
Suppose we impose periodic boundary conditions, so that φG (x) = φG (x + na), n = ±1, . . .. a is a
lattice vector.
eiGx = eiG(x+a) → eiGa = 1
Places constraints on G:
2π
G= n, n = ±1, . . .
a
2π/a is a reciprocal lattice vector.
Normalized on domain a,
1
φG (x) = √ eiGx
a
Properties of these plane wave functions:
1. Periodic, λG = a/n
2. Orthonormal, hφG |φG0 i = δG,G0
3. Definite kinetic energy
4. Complete
X
ψ(x) = φG (x)hφG |ψi (74)
G
X
= φG (x)ψ(G) (75)
G
ψ(G) is Fourier transform of ψ(x). Both contain exactly the same information. Can take advantage
of Fourier transform machinery to transform between two representations. For general function
ψ(x), represent on some grid of size N. Size of grid determined by maximum frequency/minimum
wavelength included. N log(N ) operations to transform back and forth.
Integrals can be evaluated in either x or G space:
Z
I = A∗ (x)B(x)dx (76)
a Z
0
X
= A∗ (G)B(G) e−iGx eiG x dx (77)
G,G0
X
= A∗ (G)B(G)δG,G0 (78)
G,G0
X
= a A∗ (G)B(G) (79)
G

Can leverage to simplify solving any problems that are periodic.
9.2 Example
9.3 Hydrogen atom in a box

Suppose we place an atom at some point R in each box. In principle, can represent solutions as
linear combinations of plane waves. Plane waves become basis functions.
Does where we put the atom matter to the answer? No. Alway related by structure factor.
In principle, would have to let G run to infinity to have complete basis. In practice, would have to
cut off at some point, defined by
~2 2 ~2 2
G < G = Ecutoff
2me 2me cut
Sets minimum wavelength retained in basis, and thus size of basis. For fixed Ecutoff , larger a implies
larger basis. Basis size has to be a whole number, so changes discontinuously with Ecutoff .
Recall our basic equation
n o
ĥ + vCoulomb [ρ] + vexchange [ψi ] + vcorrelation [ψi ] ψi (r) = i ψi (r) (80)

Contains kinetic and potential energy terms.
Kinetic energy is diagonal in plane wave basis. . . easy!
1 d2 1
hφG | − |φG0 i = G2 δG,G0
2 dx2 2
Potential energy terms can take advantage of Fourier transforms.

00
X
v(x) = v(G00 )eiG x (81)
G00
X
hφG |v(x)|φG0 i = v(G00 )hφG |φG00 |φG0 i (82)
G00
= v(G0 − G) (83)
How many Fourier components to include in the sums? Turns out for a basis of dimension m you
need 2m components to specify the potential exactly, but you can generally get away with smaller.
The cost and accuracy of the calculation scale with this choice.
Solutions given by ψi (x) = G ci (G)φG (x).
P
One key thing we would need to evaluate potentials is electron density, ρ(x). Let fi be occupancy
of each orbital.
X
ρ(x) = fi |ψi (x)|2 (84)
i
1X X ∗ 0
= fi ci (G)ci (G0 )ei(G−G )x (85)
a 0
i G,G
2G
X max
= n(G)eiGx (86)
−2Gmax
Electron density exactly represented in plane wave basis with cutoff four times that of the basis set
cutoff.
9.4 Three-dimensional periodicity

Same ideas generalize to three dimensional periodicity.
Lattice constants of periodic cell become three vectors, a1 , a2 , a3 . As we’ll discuss a little bit later,
these vectors will be from one of the Bravais lattices. Every point RisequivalenttoanyotherpointR0 =
R + n1 a1 + n2 a2 + n3 a3 .
Periodic box matrix h = [a1 , a2 , a3 ]. Periodic box volume Ω = deth.
Reciprocal lattice defined such that bi · aj = 2πδij . Given by
2π(ht )−1 = [b1 , b2 , b3 ]
Reciprocal lattice vectors become G = ib1 + jb2 + kb3 . Basis functions become φ(r) = 1 iG·r
Ωe .
Locations of atoms Rα within this periodic cell can be specified in Cartesians or as fractional
coordinates of the lattice vectors. Related by Rα = hf .

In general, can use this periodic representation to describe atoms, molecules, whatever. If we want
to describe isolated things, box dimensions must be large enough to avoid artificial interactions
between periodic images.
In particular, Coulomb potentials decay like 1/r and are thus long-ranged. Have to use special
tricks (Ewald sums) to evaluate these sums, and have to carefully group electrostatic terms to avoid
non-convergent sums.
Key limitations:
1. Periodic cell must be charge-neutral. The electrostatic energy of an infinite, charged system
diverges.
2. Supercell must not have a net electric field.
3. The absolute electrostatic potential is not well-defined (there is no “vacuum” to reference the
energy of an electron to).
These limitations can be overcome to some extent by applying artificial backgrounds.
9.5 GAMESS vs. Vasp

GAMESS Vasp
System in infinite vacuum System periodically replicated to ∞
Any charge Charge-neutral
Cartesian or internals atom positions relative to supercell
Gaussian basis set Plane-wave basis set
Hartree units eV units
Energy referenced to isolated charge particles Energy referenced to potential model
1 input file POSCAR: structure input
INCAR: program options
POTCAR: identity of atoms
KPOINTS: k-point sampling
2 output files OUTCAR: main output
OSZICAR: summary output
CONTCAR: final geometry
CHGCAR: charge density on 3D grid
WAVECAR: final wavefunctions
DOSCAR: summary densities of states
Vasp documentation available at https://cms.mpi.univie.ac.at/wiki/index.php/The_VASP_Manual.
9.6 Vasp structure definition

POSCAR file specifies lattice vectors, numbers and types of atoms, and their positions in either
Cartesian or fractional coordinates. Vasp uses convention in POSCAR that atoms be specified in
groups of like type, and that the order in the POSCAR corresponds to the order of atoms in a
composite POTCAR. Here’s a POSCAR for an N2 in a 10 cubic box.
10 Ang. cell, diagonal orientation
10.
1.0 0.0 0.0

0.0 1.0 0.0
0.0 0.0 1.0
2
cartesian
0. 0. 0.
0.0692820323028 0.0692820323028 0.0692820323028
Vasp generates a CONTCAR at the end of any job of identical format to a POSCAR. XDATCAR contains
a trajectory of the same format.
9.7 Vasp model specification

9.7.1 Electronic structure
ISPIN = 1, 2 non-spin-polarized or spin-polarized
NUPDOWN = xx set number of up - down electrons
MAGMOM = xx specifies initial magnetic moment of each atom
NELECT specifies total number of electrons (seldom used)
FERWE specifies occupancy of spin-up levels (seldom used)
FERDO ditto for spin-down
ISMEAR specifies how electrons are distributed, or “smeared out,” amongst orbitals/bands near Fermi
level
ISMEAR = 0 “Gaussian” smearing, a safe options for molecules and most solids
ISMEAR = 1 “Methfessel and Paxton,” useful for geometry optimizations for metals
ISMEAR = -5 tetrahedron method with Blöchl corrections, useful for accurate single-point energies
SIGMA = 0.05 smearing parameter for ISMEAR. For molecules and insulators, 0.05 eV is a sensible
value. 0.2 eV for metals. Larger values can help convergence but too large gives unphysical
results. Aim to keep the electronic “entropy” (difference between free and total electronic
energies) <1 meV/atom.
9.7.2 Electron-electron interaction model

GGA specifies exchange-correlation model. Defaults to POTCAR value, typically PBE GGA. Many
options possible. It is only meaningful to compare energies between calculations computed
with exactly the same exchange-correlation model.
GGA=RPBE “revised” PBE
GGA=OR “optimized” PBE
GGA=BF “Baysian-error estimated functional”
LHFCALC turns on exact H-F exchange. Typically quite expensive and used sparingly.
HFSCREEN=0.6 turns on HSE06 “screened” exchange, which interpolates smoothly between exact
exchange at short distance and GGA exchange at long range

9.7.3 Dispersion corrections
Generally methods that include some wavefunction-based electron correlation are needed to capture
dispersion (non-bonded) interactions well. A pragmatic approach in plane-wave calculations is to
add-on a dispersion correction, so that Edisp-corrected = EDFT + Edisp .
IVDW turns on approximate corrections to introduce a dispersion correction to energy/forces
IVDW=10 “DFT-D2” method of Grimme, adds on Lennard-Jones-type dispersion correction. Func-
tion only of atom positions.
IVDW=11 “DFT-D3” method of Grimme, corrects for differences in local coordination.
IVDW=20 Tkatchenko-Scheffler method, uses converged charge density to parameterize D2 model
9.7.4 Precision parameters

Precision controlled by size of basis and grids used to transform wavefunctions and density back
and forth between real and reciprocal space. It is only meaningful to compare energies between
calculations computed with exactly the same precision parameters.
ENCUT sets plane-wave cutoff (basis set size), in eV. Defaults to largest value in supplied POTCARs.
PREC general precision parameter to set ENCUT and corresponding FFT grid and PAW parameters
PREC=NORMAL for routine calculations
PREC=ACCURATE for higher precision, eg for forces for a Hessian calculation
LREAL controls evaluation of parts of PAW potential
LREAL=.FALSE. (default) evaluates exactly in reciprocal space. More computationally costly, esp
for larger systems.
LREAL=Auto generates optimized real-space representation of PAWs. Typically cheaper, esp for
larger systems.
ROPT accuracy of LREAL real-space method
IDIPOL=1-3 turns on dipole corrections for dipolar supercell
9.7.5 SCF
Vasp uses same essential idea as GAMESS. Guess initial charge density and set of orbital coefficients.
Vasp initial guesses are not internally consistent: charge density typically a superposition of atoms,
but wavefunctions are random. Density and wavefuctions are optimized iteratively and in parallel
until internally consistent. A mixture of algorithms are used, in particular conjugate-gradient,
Davidson, and DIIS.
ALGO=Normal (default) block Davidson optimization, very safe
ALGO=Fast starts with block Davidson, then switches to DIIS. Good compromise method.
ALGO=VeryFast only DIIS. More aggressive.

NBANDS how many orbitals (bands) to calculate. You always want to calculate all the occupied ones
and, for numerical reasons, at least some of the unoccupied ones. In general for performance
reasons you don’t need or want to calculate all of the empty ones. The default is safe.
NLEM maximum number of SCF cycles
EDIFF=1e-5 precision criterion to stop SCF, in eV. 1e-4 is generous, 1e-5 normal, 1e-6 more accurate
9.8 Core electron treatment

Wavefunctions oscillate too rapidly near core to capture reasonably using plane-waves; cut-off would
be too high. Always combine plane waves with some model for core states of the system. Typically
assume core electrons are “frozen,” and incorporate relativistic effects. Analogous to core potentials
in GAMESS.
9.8.1 Orthogonalized plane-wave (OPW)

If we know the core electron wavefuctions, and recast DFT model to calculate difference between
any valence orbital and the core orbitals. Create new OPW basis:
X
OPW
G (r) = φG (r) + bmG χm (r)
m
where χm are core orbitals that are fixed. Equivalent to modifying the electron-electron interaction
by X
v̂ OPW = v̂ee + ( − m )|χm ihχm |
m
The last part is expensive to evaluate, but can be approximated.
9.8.2 Pseudopotential
Replace last part with a potential designed to replicate the influence of the core electrons (and
nuclei) on the valence. Loose any description of core, but (hopefully) valence is retained.
• Cutoff radius - boundary between core and valence regions

• Transferability - ability of PP to be applied in different chemical environment
• Local vs non-local - spherically symmetric vs angular-momentum-dependent

• Norm-conserving - preserves |ψ|2 of wave-functions
• Soft vs hard - location of boundary. Closer in, “harder” the potential, higher the plane-wave
cutoff needed. Further out, “softer” the potential, lower the plane-wave cutoff.
• Ultra-soft pseudo-potential - (Vanderbilt) relax norm-conserving requirement and fix up with
augmentation charges. First potentials that enabled large-scale calculations
9.8.3 Augemented plane-wave (APW)

Partition space into a spherical core and plane-wave valence and patch up at the boundary
9.8.4 Projector-augmented plane-wave (PAW)

Most modern approach, combines advantages of APW and of ultrasoft PPs. Not strictly a PP
approach: constructs full wavefunction and density as combination of valence (plane wave) part
and precomputed core parts from atomic calculations using same XC potential. Retains full nodal
structure of valence wavefunctions. Generation of these potentials is not for the casual user (just as
the case for making Gaussian basis sets). The great strength of Vasp is a very complete and reliable
library of PAW models for the whole periodic table and for several exchange-correlation functional.
Stored in POTCAR.
9.9 Wavefunctions and charge densities

Primary outputs of Vasp calculation are wavefunction, density, energy, and forces.
LCHARGE=.TRUE. turns on creation of the CHGCAR file, which contains the total charge and spin
(up-down) densities, evaluated on N GX × N GY × N GZ grid. Not so useful by itself, but can
be used to construct charge density differences.
LWAVE=.TRUE. turns on creation of the WAVECAR file, which contains the final wavefunctions, useful
for restarts
LORBIT controls output of DOSCAR and PROCAR, which contain analysis of total density of states and
of each band. Controlled with RWIGS, EMIN, and EMAX
Bader charges can be extracted from CHGCAR.
9.10 Exploring potential energy surfaces

A nice feature of plane-wave methods is that the forces on the atoms are gotten essentially for free.
No Pulay forces, because ions and basis functions are decoupled.
Generally limited to Cartesian coordinates in PBC calculations. No z-matrix, internals, or redun-
dant coordinates. The cartesian locations of atoms can be fixed. Perhaps ASE offers more options.
9.10.1 Geometry optimizations

Again, same idea as in GAMESS. Use some algorithm to move atoms in response to forces.
EDIFFG=-0.03 convergence criterion, in eV/Å. −0.05 eV/Å for routine, −0.01 eV/Å for more pre-
cise.
IBRION=2 conjugate-gradient optimizaiton

IBRION=1 Quasi-Newton-Raphson with DIIS algorithm. Uses a digonal Hessian guess that gets
updated. Works well when initial geometry is pretty good, poorly otherwise
NFREE=xx number of previous steps to include in the DIIS update.
IBRION=3 damped molecular dynamics optimization
NSW maximum number of geometry steps
POTIM=0.5 tunes scaling of forces to optimize optimization algorithm.
ISW=1 only move the atoms in the optimization
9.10.2 Harmonic frequency calculation

IBRION=5 numerical Hessian and frequency evaluation, moving all atoms
IBRION=6 take advantage of symmetry to reduce number of unique geometries to evaluate forces
NFREE=2 two-sided differentiation
POTIM=0.01 specifies step-size for finite difference calculation
IBRION=7,8 alternative, analytical perturbation-theory-based evaluation of Hessian matrix
9.10.3 Transition states

Commonly found using either a “nudged elastic band” (NEB) method or “dimer” method. NEB
requires definition of a “string” of structures, will address later. “Dimer” needs an initial guess of
transition state and a (good) guess of the reaction coordinate.
IBRION=44 specifies dimer calculation. See sample input below. Get trial direction from prior
Hessian calculation.
ammonia flipping
1.
6. 0. 0.
0. 7. 0.
0. 0. 8.
H N
3 1
cart
-0.872954 0.000000 -0.504000 ! coordinates for atom 1
0.000000 0.000000 1.008000
0.872954 0.000000 -0.504000
0.000000 0.000000 0.000000 ! coordinates for atom N
! here we define trial unstable direction:
0.000001 0.522103 -0.000009 ! components for atom 1
-0.000006 0.530068 0.000000
-0.000005 0.522067 -0.000007
0.000001 -0.111442 0.000001 ! components for atom N
IMAGES turns on nudged elastic band

9.10.4 Molecular dynamics
IBRION=0 molecular dynamics, propogating atomic positions forward in time according to forces.
Typically combine with PREC=LOW.
NSW number of MD steps
POTIM=0.5 step size, in fs
SMASS=-3 NVE dynamics, which (should) conserve energy
SMASS=-1 scaled velocity MD, eg for simulated annealling
SMASS>0 NVT dynamics, SMASS specifies the thermostat mass
TEBEG initial temperature, in K
TEEND final temperature, if temperature is to be scaled

10 Periodic electronic structure
10.1 Isolated vs. periodic systems
Power of periodic boundary condition model is ability to describe things that are extensive: solids,
surfaces, even liquids.
(peroxyacetyl nitrate) (MgO bulk)

(MgO slab)
c
c
b b
a
a
cluster models supercell models

Isolated molecule in vacuum 3-D periodic boundary conditions applied to atomic
(or dielectric continuum) configuration
Gas-phase Bulk solids

Solution-phase Bulk liquids
Localized chemistry Surfaces
spaced polymer, then seeing that it’s subject and Interfaces
to an instabil- priate symmetry-adapted linear combinations cy (remem-
Delocalized
ity, called a Peierls distortion. Other words around that chemistry
ber translation is just as good a symmetry operation as any
Amenable to highly characterization would be strong electron-phonon cou- other one we know) are given in 6 . Here a is the lattice
accurate calculations pling, a pairing distortion, or Amenable
a 2 k F instability. efficientspacing
And the
to highly (the unit cell being in one dimension) and k is an
calculations
physicist would come to the conclusion that the initially index which labels which irreducible representation of the
equally spaced H polymer would form a chain of hydrogen translation group ty transforms as. We will see in a mo-
molecules. I mention this thought process here to make the ment that k is much more, but for now, k is just an index
Key difference is that atoms communicate across the boundary in the periodic model. Results in
point, which I will d o again and again, that the chemist’s
intuition is really excellent. But we must bring the lan-
for an irreducible representation, just like a, e,, and e2 in
C, are labels.
discrete states → continuum of electronic states.
guages of our sister sciences into correspondence. Inciden- The process of symmetry adaptation is called in the sol-
tally, whether distortion 4 will take place at 2 Mbar is not id-state physics trade “forming Bloch f ~ n c t i o n s . ” ‘ To
~~~~~.~~
-
obvious, a n open question. reassure a chemist that one is getting what one expects
Let’s return to our chain of equally spaced H atoms. It from 5, let’s see what combinations are generated for two
10.2 Bloch’s theorem and qualitative band structure
turns out to be computationally convenient to think of that specific values of k, k = O and k = n / a (see 7). Referring
chain as an imperceptibly bent segment of a large ring (this
Images borrowed from https://doi.org/10.1002/anie.198708461.
is called applying cyclic boundary conditions). The orbi-
k=O e0 X, = 5 X, = X o + X I + X z + X J + ...
tals of medium-sized rings on the way to that very large $o=
one are quite well known. They are shown in 5.

Cyclic chain of H atoms of increasing length (Born-van Karman boundary condition):
-
ex+&cM%
--
8,-0 -
a --- --- back to 5 , we see that the wave function corresponding to
v-- -- --
8--
k=O is the most bonding one, the one for k = n / a the top
m1
v -- -- of the band. For other values of k we get a neat description
8- - a=-
of the other levels in the band. So k counts nodes as well.
80
-- -- The larger the absolute value of k, the more nodes one has
-- --
9 89- -
- --
-
in the wave function. But one has to be careful-there is a
range of k and if one goes outside of it, one doesn’t get a
---- new wave function, but repeats an old one. The unique val-
v = w= Q- 0- ues of k are in the interval - n / a I k < n / a o r I k l 1 d a .
This is called the first Brillouin zone, the range of unique
5
k.
How many values of k are there? As many as the num-
Energy levels increasing with chainFor alength. Wavefunctions
hydrogen molecule (or ethylene) there grow in a regular way, consistent with
is a bond- ber of translations in the crystal, or, alternatively, as many
ing o,(n) below a n antibonding o,*(n*). For cyclic H3 or as there are microscopic unit cells in the macroscopic crys-
(and identifiable by) underlying cyclopropenyl
symmetry. Nodes increase systematically.
we have one orbital below two degenerate
In limit, get an infinite
tal. So let us say Avogadro’s number (N,,), give or take a
band of levels. ones; for cyclobutadiene the familiar one below two below few. There is an energy level for each value of k (actually a
degenerate pair of levels for each pair of positive and ne-
one, and so on. Except for the lowest (and occasionally the
highest) level, the orbitals come in degenerate pairs. The gative k values). There is an easily proved theorem that
Infinite chain of H atoms separated by a lattice constant a. Electrons move in that periodic potential.
number of nodes increases as one rises in energy. We’d ex- E(k)=E(-k). Most representations of E(k) d o not give
the redundant E( - k), but plot E(lk1) and label it as E(k)).
Expand wavefunctions in atom-centered 1s orbitals.
pect the same for a n infinite polymer-the lowest level
Also, the allowed values of k are equally spaced in the
nodeless, the highest with the maximum number of nodes.
In between, the levels should come in pairs, with a growing space of k, which is called reciprocal or momentum space.
number of nodes. The chemist’s representation of that The relationship between k = 1//2 and momentum derives
band for the polymer is given at right in 5 . from the d e Broglie relationship il=h/’. Remarkably, k is
not only a symmetry label and a node counter, but it is also
a wave vector, and so measures momentum.
Bloch Functions, k, Band Structures
November 20, 2019 © 2019 W. F. Schneider

There i s a better way to write out all these orbitals, mak- 79
ing use of the translational symmetry. If we have a lattice
whose points are labeled by an index n=O, 1, 2, 3, 4,etc.,
as shown in 6,and if on each lattice point there is a basis
a
Atoms: H H H H H H H H H H PBC
e- e- e- V ( x) = V ( x + na)
Potential:
"n Î integers
Basis:
c 01s c11s c 21s c 31s c 41s c 51s c 61s c 71s c81s c 91s
Wavefunctions:
k = 0, l ® ¥ y 0 = å e0 cn1s = å cn1s
. n n
.
p .
k= , l = 4a y p 2a = å eip n 2 cn1s
2a . n
.
.
p y p a = å eip n cn1s = å (-1)n c n1s
k= , l = 2a
a n n
p
Bloch s Theorem: y k = å eikna c n , k £ (recall eikr is like a cosine function)
n a
Bloch’s theorem states that the wavefunctions can be written as the product of a cell-invariant part
(the H 1s functions here) and a cell-periodic part. The periodic part is indexed by the wavevector
k. k takes as many values as there are periodic units N :
π n N −1 N −1
k= , n=− , . . . , 0, . . . ,
aN 2 2
If N is infinite, then k is a continuous and real variable. The space of unique values of k is called
the first Brillouin zone. The periodic phases of the basis functions correspond to an underlying
wavelength associated with k :
λk = 2π/k
By the de Broglie relationship, then, k relates to the momentum of an electron in that energy level.
Each k is degenerate such that E(k) = E(−k) (moving with the same momentum to the left and
right gives the same energy).
The continuum of energy levels in k is called the band structure. Conventionally plotted vs |k|. The
width of the band is called the dispersion:
.
.
energy
energy
H H 0 π/a H H H H H
k
Molecular orbital diagram Band structure diagram
The highest filled energy state, in this case (k=π/2a\), is the /Fermi energy. Band structure on
• Discrete orbital energies become continuous band of energies, indexed by k
• Width
November ( dispersion ) of band©
20, 2019 determined by overlaps
2019 W. F. Schneiderbetween neighboring 80
cells
– Larger supercells à smaller dispersion
the right would correspond to a metal, as the energy difference between different momentum states
is zero.
Band dispersion depends on cell overlap, dispersion depends on type and orientation of orbitals:
10.3 Band folding

10.4 Multi-dimensional periodicity
10.5 Density of statues
10.6 Bravais lattices
10.7 Quantitative supercell calculations
10.8 Brillouin zone integration
10.8.1 k-point sampling
10.8.2 Fermi smearing

11 Practical supercell calculations

12 Surfaces
12.1 Surface planes
12.2 Slab models
12.3 Surface energy
12.4 Surface potentials and Fermi energies
12.5 Surface adsorption
12.6 Coverage-dependent adsorption
12.7 Reaction barriers

13 Implicit solvation
14 Density functional theory

14.1 Electron density ρ as fundamental quantity
14.2 Thomas-Fermi-Dirac model
14.3 Hartree-Fock-Slater model
14.4 Hohenberg-Kohn theorems
14.5 Kohn-Sham construction
14.6 Exchange-correlation functionals
So Kohn et al. showed that the DFT approach is theoretically well-grounded and provided one way
to practically apply it. Promise is that if we can find an approximation to the (unknown) true vxc
with the right balance of simplicity and accuracy, we will have one sweet theory. Has to incorporate
both exchange, like Slater tried to do, and correlation.
How to proceed? Lots of approaches, and jargon here is at least as bad as in wavefunction-based
methods. Perdew 2006 describes the “Jacob’s ladder” of approximations:
14.6.1 LDA
One well-defined limit is the homogeneous electron gas, and this is the usual starting point for
modern approximate DFT methods. Assume exchange and correlation potentials at any given
point depend only on the value of ρ there (or spin-up and spin-down ρ, if spin-polarized). We know
from Slater and Dirac’s work what the exchange potential is for this system.
It is possible to determine numerically the correlation energy for a given density from quantum
Monte Carlo calculations. Ceperley and Alder (PRL 1980, 45, 566) did this to very high accuracy,
and others (Vosko, Wilk, and Nusair, “VWN”, and Perdew and Wang, “PW”) fit these numerical
results to analytical models in ρ. This combination of local exchange and correlation defines the
LDA model.
LDA offers modest improvement over HFS for molecules. “Homogeneous” approximation pretty
severe for an atom or molecule. Nonetheless, works surprisingly well for structures and charge
distributions, but has problems in calculating accurate bond energies, typically overbinding. Also
tends to underestimate the HOMO-LUMO gap in molecules and analogous band gap in solids.
14.6.2 GGA
14.6.3 Meta-GGA
14.6.4 Hyper GGA and hybrid functionals
• “Screened” exchange

14.6.5 Beyond hyper GGA
14.7 Implementations
14.8 Performance

15 Electron correlation methods

Comp Notes

Uploaded by

Copyright:

Available Formats

Comp Notes

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Comp Notes

Uploaded by

Copyright:

Available Formats

Lecture Notes for CBE 60547

November 20, 2019

2 Refresher on Quantum Mechanics 5

3 Hydrogen atom: simplest chemical “thing” 10

4 (Two is too) many electrons 18

5 Practical electronic structure 27

6 Potential energy surfaces 39

7 WFT Beyond Hartree-Fock 55

9 Plane waves and core potentials 69

10 Periodic electronic structure 79

11 Practical supercell calculations 82

14 Density functional theory 84

15 Electron correlation methods 86

November 20, 2019 © 2019 W. F. Schneider 2

1.3 Our goals

1.4 Reading resources

1.5 Software tools

November 20, 2019 © 2019 W. F. Schneider 3

1.5.3 Supercell methods

1.5.4 Great for getting started

November 20, 2019 © 2019 W. F. Schneider 4

2.2 Postulates of non-relativistic quantum mechanics

2.4 Example: Energy states of a particle in a box

November 20, 2019 © 2019 W. F. Schneider 5

Table 1: Postulates of Non-relativistic Quantum Mechanics

Postulate 2: To every physical observable quantity M there corresponds a Hermitian operator M̂ .

Postulate 3: If a particular observable M is measured many times on many identical systems is a

Postulate 4: The energy-invariant states of a system are solutions of the equation

November 20, 2019 © 2019 W. F. Schneider 6

Atomic unit SI unit Common unit

3D box → 3 degrees of freedom/coordinates

~2 1 ∂ 2 X(x) 1 ∂ 2 Y (y) 1 ∂ 2 Z(z)

ftn x + ftn y + ftn z = constant → each term must be constant.

November 20, 2019 © 2019 W. F. Schneider 7

• Energy increase with number of nodes.

November 20, 2019 © 2019 W. F. Schneider 8

November 20, 2019 © 2019 W. F. Schneider 9

Figure 2: Energy sates of 3D Particle in a box

3 Hydrogen atom: simplest chemical “thing”

November 20, 2019 © 2019 W. F. Schneider 10

3.2 Analytical solutions

Questions: Ionization energy of an H atom? 1s → 2s energy? Thermal populations?

3.3 Variational principle

November 20, 2019 © 2019 W. F. Schneider 11

Table 3: Hydrogen atom

ψ(r, θ, φ) = R(r)Yl,ml (θ, φ)

L10 = L21 = L32 = . . . = 1 L20 = 2 − x L31 = 4 − x

November 20, 2019 © 2019 W. F. Schneider 12

Figure 3: H atom wavefunctions

Figure 4: H atom radial probability

November 20, 2019 © 2019 W. F. Schneider 13

g̃ξ (r) = N gξ (r)

hEi = hg̃1 |Ĥ|g̃1 i = −0.096 Hartree

b(r) = Ns (g̃1 (r) + g̃0.5 (r))

hb(r)|b(r)i = Ns2 (hg̃1 |g̃1 i + hg̃0.5 |g̃0.5 i + 2hg̃1 |g̃0.5 i) (21)

c(r) = Ns0 (g̃1 (r) + 1.5g̃0.5 (r))

hc(r)|Ĥ|c(r)i = −0.333 Hartree

November 20, 2019 © 2019 W. F. Schneider 14