E.

Kwan Lecture 9: Introduction to Computational Chemistry Chem 117

Introduction to Computational Chemistry Key Questions

(1) What kinds of questions can computational chemistry
Eugene E. Kwan help us answer?
February 22, 2010.
- Mechanistic:
What are the intermediates and transition states along the
reaction coordinate?

What factors are responsible for selectivity?

Scope of Lecture
the As molecules pass from reactants to products, do they
the Odyssey PES stay along the minimum energy path?
Cluster optimization
getting started - Physical:
molecular What is the equilibrium geometry of a molecule?
with Gaussian introduction to mechanics
DFT computational chemistry What will the spectra of a molecule look like (IR, UV-vis
conformational NMR, etc.)? What do the lines represent?
electron searching
correlation basis sets HF methods - Conceptual:
and orbitals Where are the charges in a molecule? What do its
orbitals look like? Where are the electrons?
Key References
1. Molecular Modeling Basics Jensen, J.H. CRC Press, 2009. Why are some molecules more stable than others?
Is a molecule aromatic? What are the important
2. Computational Organic Chemistry Bachrach, S.M. Wiley, hyperconjugative interactions in a molecule?
2007.
(2) How are potential energy surfaces (PESs) studied?
3. Essentials of Computational Chemistry: Theories and Models How do I locate ground states and transition states?
(2nd ed.) Cramer, C.J. Wiley, 2004.
(3) How is energy evaluated?
4. Calculation of NMR and EPR Parameters: Theory and What are the differences between HF, DFT, MP2, etc?
Applications Kaupp, M.; Buhl, M.; Malkin, V.G., eds. Wiley- When is each method applicable?
VCH, 2004.
(4) How do I perform calculations here at Harvard?
How do I use Gaussian?
I thank Professor Jensen (Copenhagen) for providing some What do I need to do to get started on the Odyssey Cluster?
useful material for this lecture.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
The Potential Energy Surface (PES) the energy is a function of all the nuclear coordinates:
In every introductory organic textbook, you see diagrams like: E  q   E  x1 , y1 , z1 , x2 , y2 , z2 
energy Now, if you're sharp, you can point out that we don't really care
transition states
where the center of mass is, but what we really care about in
0 terms of chemistry, is simply the bond length r. So in some
reactants sense, the PES is single-dimensional. The typical appearance
products of such potentials for bonds is something like:
C H + Cl -10
R C H + HCl
H
-21
R C H + HCl
R
-31
R C R + HCl
R
"reaction coordinate"
Here are some important questions that you should ask:

- What is a "reaction coordinate," exactly?

- What does a transition state look like?
- Where do the numbers for these energies come from?

I can't answer all of these questions here, but the primary Wikipedia
purpose of computational chemistry is to connect
experimental results with a theoretical potential energy Note that at the bottom of the well, the PES is described well
surface so we can understand nature. One can ask static by a harmonic oscillator--a quadratic function. This is an
questions like "What is the equilibrium geometry for this important approximation that is made in many calculations.
molecule?" or dynamic ones like "What is the mechanism of
this reaction?". But, wait! Where are the electrons? Why aren't we giving
them coordinates? It is a basic assumption of standard
What exactly is a potential energy surface (PES)? Consider computational techniques that because the nuclei are much
dihydrogen: heavier than the electrons, the nuclei are essentially "frozen"
r from the point of view of the electrons. This is the Born-
H H Oppenheimer approximation. From quantum mechanics, we
know that the electrons are described by wave functions,
How many variables do I need to describe the geometry of rather than specific position and momentum coordinates as in
dihydrogen? Six. In Cartesian coordinates, I could say that classical mechanics.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
The Potential Energy Surface (PES) solid = function; dashed = derivative
Now, a more complicated molecule will necessarily require more
coordinates. For example, you can think of water as needing 6
two O-H bond lengths and the H-O-H angle:
4
H
2
r2

O H
r1 - 1.5 - 1.0 - 0.5 0.5 1.0 1.5
-2
If you want to characterize the entire PES, and you want to do
10 points per coordinate, that would make a thousand points
-4
total. Since evaluating the energy as a function of geometry
requires minutes if not hours per point, this is clearly going to be
-6
impractical for all but the exceptionally patient (and we're not).
If we want to know whether a particular stationary point is a
To summarize, the PES is a multidimensional function. It takes local minimum or maximum, we compute the second derivative,
the molecular geometry as input, and gives the energy as which is positive for minima, 0 for asymptotes, and negative for
output. It has a lot of dimensions and is very, very complicated. maxima. For multidimensional functions, the analog of the
second derivative is called the Hessian:
Fortunately, transition state theory tells us that we only need
to characterize a few stationary points on the PES, rather than
the entire thing. A stationary point is anywhere the gradient is
zero. For dihydrogen:
E E E
    0
x1 x2 y1
This is the multidimensional analog of the derivative. For
example, consider this function:
y  x5  5 x  1
In chemistry, we are particularly interested in energy minima.
From high school calculus, you know that the derivative is: A stationary point is a local minimum if the Hessian is postive-
dy definite there. To evaluate this, one transforms from Cartesian
 5x4  5 to normal coordinates, which amounts to diagonalizing the
dx Hessian or "performing a frequency analysis." The "lack of
The function has a local maximum and a local minimum, which any imaginary frequencies" means that the matrix is positive
occur when dy/dx is zero: definite.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
The Potential Energy Surface (PES) The minimum energy path, reaction coordinate, or intrinsic
Can one have a stationary point that is neither a local maximum reaction coordinate connect the starting materials and
nor a local minimum? Certainly: saddle points. Consider y=x3: products. Thus, to understand a reaction, we will need to
2
locate a minimum of three stationary points: one each for the
starting material, transition state, and product.
1
energy
transition states

- 1.0 - 0.5
0
0.5 1.0 reactants
products
-1 C H + Cl -10
R C H + HCl
H
-2
-21
R C H + HCl
starting materials and products correspond to local minima R
(stationary points with zero imaginary frequencies) while -31
R C R + HCl
transition states correspond to first-order saddle points R
(stationary points with exactly one imaginary frequency). "reaction coordinate"
The energy difference between the reactant and transition
state will tell us about the rate of the reaction; the difference
between the reactant and product will tell us about how
exothermic the reaction is. We can also examine the geometry
of the molecules at all the stationary points, and then draw
some conclusions about reactivity, selectivity, or other trends.

Q: How do I know if the answers that the computer gives

actually mean something real?

It is very important to be considering this question at all times

when dealing with computations. To fit into the scientific
method, a computation must make a tangible prediction that
can be verified by experiment. One can now predict geometries
(X-ray), reaction barriers (kinetics), spectroscopic properties (IR
and NMR), etc. As it turns out, in many cases, the agreement
between theory and experiment is now very good.
http://www.chem.wayne.edu/~hbs/chm6440/PES.html
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Energy Minimization 1  dE  1
Q: How do we locate the stationary points? Re  Rg     Rg  F
k  dR  Rg k
The general approach is: (1) Make a guess for the geometry of
a molecule at its stationary point; (2) Use a computer program Here, Rg is some guess for the value of Req. F is the force, or
to move the atoms in such a way that the gradient drops to negative gradient. That means that if we know what the spring
zero; and (3) Perform a frequency analysis to verify the nature of constant k is, we can get to the stationary point in one step.
the stationary point is correct. However, we usually don't know this, so we take small scaled
steps in the direction of the gradient until it falls below some
For now, let us consider step (2); the optimization process. value. This is the method of steepest descent.
Imagine we have a one-dimensional, quadratic PES with
coordinate R (discussion borrowed from Jensen, pp. 8-12):
Energy

1.4

1.2

1.0

0.8
Clearly, we want to be taking as few steps as possible. For a
0.6 harmonic oscillator, k is the derivative of the gradient:
Rg  d 2E   dE 
Re  Rg   2 
0.4
 
Req 0.2  dR  Rg  dR  Rg
R
- 1.0 - 0.5 0.0 0.5 1.0 This works great if we are near a quadratic portion of the PES.
In reality, many PESs have very flat regions where this kind of
The energy E of this harmonic oscillator potential is given by:
Newton-Raphson step will be too large. (An excellent
1
E k ( R  Req )2 animation of how this works can be found on Wikipedia at
2 http://en.wikipedia.org/wiki/File:NewtonIteration_Ani.gif.) Thus,
most programs will scale back quadratic steps when their size
Taking the derivative, we have: exceeds some pre-determined threshold.
dE
 k ( R  Req ) Note that this method requires the Hessian, which is clearly
dR expensive to calculate. In many cases, one can use
Clearly, if R is not the equilibrium value, Req, then the gradient approximate methods to make a good guess at what the
is non-zero, and we are not at a stationary point. But we can Hessian looks like, and then update the guess on every
rearrange this equation to see how to get there: iteration.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Energy Minimization Gaussian plots the energy and gradient of each geometry
In practice, neither of these methods is satisfactory and various optimization step:
improvements have been made. In Gaussian 09, the Berny
algorithim using GEDIIS in internally redundant coordinates is
used by default. For details, see http://www.gaussian.com/
g_tech/g_ur/k_opt.htm. Internally redundant coordinates means
that instead of using the Cartesian coordinate system, which
has certain mathematical pathologies, a more natural set of
coordinates involving bond lengths, bond angles, and dihedral
angles is used. This has been shown to be faster for many
organic molecules.

Let's take a look at how this works in real life. My co-worker Joe
Wzorek and I are interested in the conformations of macrocycles.
One structure we need the equilibrium geometry and energy of
is a conformer of an intermediate used by Professor Shair along
his route towards longithorone A:

(1) Energy is given in hartrees. 1 hartree = 627.509 469 kcal.

(2) The energy drops quickly at first and then slowly converges.
Seeing the energy spike in the middle is common;
sometimes the optimizer will get off track. By its nature,
optimization is a chaotic phenomenon, and can show high
sensitivity to initial conditions, oscillatory behavior, or other
pathologies.

(3) The gradient usually tracks with the energy, but not always.
When it reaches a certain threshold, the job is finished.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Energy Minimization atoms, and each one has a particular gradient associated
By the way, one of the nicest programs for rendering structures, with it, we need the RMS gradient. Remember, gradient is the
as shown on the previous slide is CYLView, is available from negative of force. Displacement is something that tells you how
Professor Legault (Sherbrooke) at www.cylview.org. much the atoms moved between the last iteration and this one.
Occasionally, the displacement will not converge, but the forces
We can learn more by looking at the raw output file (.out). The will go to zero. In that case, the optimization will automatically
fundamental file format in computational chemistry is text, so it is terminate.
useful to learn some techniques for handling large amounts of [ekwan@iliadaccess03 output]$ grep -B 7 Stationary
text. Virtually all scientific computing is done in the magical land longithorone_OPLS_low_4.out
of Linux, so we will start there: Item Value Threshold Converged?
Maximum Force 0.000008 0.000450 YES
[ekwan@iliadaccess03 output]$ grep -A 3 "Maximum Force"
RMS Force 0.000001 0.000300 YES
longithorone_OPLS_low_4.out | more
Maximum Displacement 0.001556 0.001800 YES
RMS Displacement 0.000316 0.001200 YES
Maximum Force 0.037142 0.000450 NO
Predicted change in Energy=-5.392921D-09
RMS Force 0.004970 0.000300 NO
Optimization completed.
Maximum Displacement 0.948905 0.001800 NO
-- Stationary point found.
RMS Displacement 0.234625 0.001200 NO
-- In this case, everything converged after 64 iterations. What is
Maximum Force 0.016448 0.000450 NO the energy? At each step, the output file contains the energy:
RMS Force 0.002426 0.000300 NO [ekwan@iliadaccess03 output]$ grep "SCF Done"
Maximum Displacement 1.085268 0.001800 NO longithorone_OPLS_low_4.out
RMS Displacement 0.284053 0.001200 NO SCF Done: E(RB3LYP) = -1532.21514319 A.U. after 12 cycles
... SCF Done: E(RB3LYP) = -1532.23006873 A.U. after 11 cycles
...
Here, I have issued a command to search for the words SCF Done: E(RB3LYP) = -1532.24125618 A.U. after 5 cycles
"Maximum Force" in the file longithorone_OPLS_low_4.out. SCF Done: E(RB3LYP) = -1532.24125618 A.U. after 1 cycles
Specifically, I want all the instances of that phrase, along with
the three lines after it. The "more" command tells it to give me Notice that the energy starts out high, and then decreases.
the output page by page (quite a few optimization steps were E(B3LYP) tells you that this a density functional method was
required, so there is a lot of output). used to get at the energy--more on this in a moment. The last
energy is the energy of the molecule in the geometry of this
The first column of numbers is the value of the parameter in the particular stationary point. Specifically, this is the electronic
current iteration; the second is the desired threshold value. As energy. Roughly speaking, this is the energy it takes to bring
you can see, the first two steps were quite far away from together all the atoms and electrons, which is why it looks like
equilibrium. a huge negative number. Every geometry step involves an
energy evaluation, which is itself an iterative process. As we
RMS means "root mean square," which is basically an average get closer and closer to equilibrium, the energy evaluation
that works on signed quantities. Because there are a lot of process speeds up and requires fewer cycles.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Of course, we need to verify this structure is a true local These corrections take into account the fact that, above absolute
minimum with a frequency analysis. (There is no way to know if zero, various vibrational and rotational energy levels get
it is the global minimum. However, there are methods I will populated, and therefore contribute to the energy. The very last
discuss shortly that can give one some confidence the global line represents the Gibbs free energy of this compound in this
minimum, or something near it, has in fact been found). geometry, at least as estimated by Gaussian using this particular
GaussView can display all the normal modes (Results... method.
Vibrations):
If there had been any imaginary frequencies, there would have
been a message like "1 imaginary frequency ignored." This
would mean a transition state.
Constrained Optimizations and Transition States
Because these are still stationary points, one might think that
the same optimization methods would apply. Unfortunately,
they seem to be much better at optimizing downwards towards
a ground state than upwards towards a transition state. As a
result, unless the starting structure is already very close to the
transition state stationary point, it will optimize away from it.

The best strategy for getting a starting structure close to a TS is

to perform a "scan" in which the forming bond distance is fixed
at certain values, while all the other internally redundant
coordinates are allowed to relax. Here is a scan I performed to
find the transition state for a Michael reaction:

The fact that all the frequencies are positive numbers indicates
that this is a true local minimum. This is also reflected in the
thermochemical energy output:
[ekwan@iliadaccess03 output]$ grep -4 Gibbs
longithorone_OPLS_low_4.out

Zero-point correction= 0.636208 (Hartree/Particle)

Thermal correction to Energy= 0.672630
Thermal correction to Enthalpy= 0.673575
Thermal correction to Gibbs Free Energy= 0.567792
Sum of electronic and zero-point Energies= -1531.605048
Sum of electronic and thermal Energies= -1531.568626
Sum of electronic and thermal Enthalpies= -1531.567682
Sum of electronic and thermal Free Energies= -1531.673465
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Constrained Optimizations and Transition States One can animate these normal modes to verify that the negative
The scan began with the two molecules quite separated, and frequency corresponds to the desired bond formation event:
then the forming bond distance was fixed at around 3.8 A and
the structure was optimized. The resulting pseudo-stationary
point was then perturbed slightly to bring the reacting partners
0.07 A closer together (the "step size"), and then a new
constrained optimization was performed. In such a procedure,
one hopes that the scan will reach an energy maximum, which
will turn out to be the transition state.

As it turned out in this case, the scan did reach a maximum, but
then the energy dropped considerably. This is a consequence
of the step size being too large--the perturbation between steps
was enough to cause the structure to "kink" into a much lower
energy conformation. This turned out to be of no consequence,
however, as a finer step size, followed by unconstrained
transition optimization eventually resulted in the transition state.
This time, the frequency analysis shows one negative frequency:
that of the forming bond: A more rigorous analysis would involve an intrinsic reaction
coordinate (IRC) scan, which would take the transition state
and follow the minimum energy path in both directions to check
that the observed transition state actually connects the desired
starting materials and products.

Other transition state search methods are available that try to

connect a starting material structure with a product structure.
These can work well in some cases, but are not as reliable.

You may notice that I say "observed" as if this sort of thing were
experimental in nature. While this is not strictly speaking, from
a philosophical perspective, true, it is true in practice. As I often
say, like anything to do with computers, computational
chemistry obeys the rule "garbage in, garbage out." One has to
ask a meaningful question to get a meaningful result. Even
then, apparently logical computations (optimizations) will often
fail to converge, or may give non-sensical results. One must
always use chemical intuition to see if the results being found
actually make sense, and then rigorously follow up the
theoretical calculations with real experiments.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117

    r1  rOH ,eq 
Molecular Mechanics 1
 kOH  k AB
2
E (3)
r  r
2 
Q: How are geometries turned into energies? 1 OH , eq
In organic chemistry, we have the functional group concept,
which says that a ketone in one molecule usually behaves a lot The term with the superscript (3) is the cubic force constant, or
like a ketone in another molecule. When this transferability "anharmonic" constant. Unfortunately, such a function would
concept is applied to computational chemistry, one has diverge to negative infinity at large separations r. Therefore, in
molecular mechanics (MM). practice, one needs to include a quartic term. The standard
organic force field MM3 uses such terms. One can also
MM is the cheapest of the computational methods and envision representing complex periodic behavior in torsional
provides reasonable geometries and energies for ground potentials by using higher order Fourier series.
states. However, it will not understand bonds that are stretched
from equilibrium, like in transition states or other "exotic" What is missing here? So far, we have only considered the
behaviors like attractive dispersion forces unless they have been bonded terms. However, real molecules also have all kinds of
explicitly programed in. In this way, MM methods are a lot like non-covalent interactions that must be represented by additional
the empirical NMR prediction methods used in ChemDraw. non-bonded terms.

Here is water again: Consider the hard sphere and Lennard-Jones potentials
H
(diagram taken from page 26 of Cramer):

r2

O H
r1
A molecular mechanic treatment of the energy of water would be
something like:
E  kOH  r1  rOH ,eq   kOH  r2  rOH ,eq   k HOH    HOH ,eq 
2 2 2

Thus, we are treating the bonds and bond angles as harmonic

potentials: balls and springs. In more complicated molecules,
dihedral angles will also appear with similar potentials. However,
since torsional potentials are necessarily periodic, they will often
contain trigonometric terms. Now, in this approximation, the
energies are expanded as second-order terms. But as bonds
get stretched farther from equilbrium, the validity of such In the hard-sphere potential, the two balls don't interact until
potentials decreases. The answer is to use higher-order terms some critical distance, the sum of the radii of the two balls.
in the Taylor expansion. Inclusion of a cubic term would lead to This is AB. In the Lennard-Jones potential, there is a highly
an expression like: repulse 1/r12 term at close distances and a weakly attractive
1/r6 term at long distances.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Molecular Mechanics attenuate interactions mediated by a torsional angle. Finally,
As Cramer puts it, "one of the more profound manifestations of hydrogen-bonding can also be described and is usually
quantum mechanics is that [the hard sphere potential] does not parametrized by a rapidly decaying "10-12" potential.
accurately represent reality. Instead, because the 'motions' of
electrons are correlated...the two atoms simultaneously develop Obviously, one needs a lot of parameters to have a good force
electrical moments so as to be mutually attractive. The force field. These parameters are generally fitted to reproduce
associated with this interaction is referred to variously as experimental data and perform well in "normal" situations. For
'dispersion,' the 'London' force, or the 'attractive van der Waals' the standard atoms in the organic chemist's toolkit, there are
force." In fact, the helium dimer is bound with one vibrational good quality parameters for most functional groups.
state with an equilibrium bond length of 55 A! The proper
description of dispersive interactions is currently a major Note that there is no quantum mechanical justification for
research topic in computational chemistry. In general, most partitioning the energy into a sum of pairwise-additive
higher level methods such as HF, MP2, or DFT have some contributions! Nonetheless, MM is a viable way to model
problems describing things like benzene dimer, although some large systems or deal with smaller systems which have a lot of
progress has been made (see Chem 106, Lecture 30). conformational flexibility because the computational resources
it demands are tiny.
Of course, there are other non-bonded terms other than steric
repulsion and dispersion. In particular, one must take electro- There are many flavors of MM force fields, which are sets of
static interactions into account. For an intermolecular complex parameters designed for different systems. They are separated
between two molecules A and B, one can consider the classical into "all atom" (aa) and "united atom" (ua) types. In the ua
energy of interaction to be the interaction of the multipole approximation, some groups like methyl groups are treated as
moments M of A (zeroth-order: charge; first-order: the x, y, and z a bigger "ball" to save computational time. A good summary
components of the dipole moment; second-order: the nine appears in Chapter 2 of Cramer. Here is my summary of his
quadrupole moments; etc.) and the electrical potentials from summary:
the moments of B. However, it's not clear how this would work
for intramolecular interactions. CHARMM/m: Developed by Karplus and co-workers for
biomolecules
Instead, it is easier to assign every atom a partial charge q
and calculate the electrostatic interaction energy UAB as: MM2/MM3/...: Developed by Allinger and co-workers for
organic molecules. MM2 has been superseded by MM3.
q A qB
U AB 
 AB rAB MMFF: Halgren and co-workers. For organics and
biomolecules.
where  is a constant that describes how well the charges
interact through space. One can assign the charge simply OPLS: Jorgensen and co-workers for organics and
based on the atom type, or on a more complex scheme biomolecules with an emphasis on parameters that fit the
involving the kinds of atoms near the atom in question. , too, experimental properties of liquids.
can vary and is generally parameterized to neglect electrostatic
interactions between atoms that are directly connected and The choice of force field is a matter of taste and circumstance.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Molecular Mechanics Here is how you can visualize this process. My co-worker Joe
Q: How do we find the global minimum? and I are interested in macrocyclic conformations:
A frequency analysis can only tell us if a structure is a local
cut open Monte Carlo
minimum. You have a few options.
macrocycle searching
(1) Guess Yourself. This means drawing all the likely looking
structures, minimizing them, and hoping you are thorough starting structure randomly vary
enough to find the global minimum. For complicated bond torsions
structures, particularly with more expensive quantum-
mechanical methods, this is essentially the only option. E if ring can be
re-connected,
(2) Have the Computer Search Everything. If you are willing
to accept a lower cost method (molecular mechanics) or accept structure
E and minimize
perhaps wait a long time, then you can have the computer
E
search the entire phase space of the system. It will try every
combination of dihedral angles, bond angles, etc. Although remove
this guarantees you will find the global minimum, it amounts duplicates
variety of candidate structures
to characterizing the entire PES, and is usually impractical.
Here are the lowest 20 structures for epothilone
(3) Have the Computer Guess. Here, you have the computer (as far as MM knows from this search):
try and search part of the phase space at random. There are
a number of methods for doing this, but a very common one
is the Monte Carlo method.

1) Start with some geometry.

2) Randomly change the bond angles, dihedrals, etc.
3) Calculate the new energy.
4) If the new energy is lower than the old energy, then
"accept" this structure and use this structure in step 2.
5) If the new energy is higher, then maybe reject it. Reject
structures that are much higher in energy more frequently
than structures that are a bit higher in energy. More
precisely, reject if exp[-(E2-E1)/RT] is less than some
random number Z.

The fact that there is a chance of accepting higher energy

structures means that the optimization algorithm can climb
out of local minima.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Quantum Mechanical Methods energy. The rotational energies are also quantized:
For many problems, the quality of molecular mechanics energies h
is simply insufficient. For better results, we turn to quantum Erot  2 J J  1  
J = 0, 1, 2...
mechanics. As you know, in non-relativistic quantum mechanics, 8 cI
absolutely everything is described by Schrodinger's equation,
shown here in its one-dimensional, time-independent form: where I is the moment of inertia and J is the rotational quantum
number. The ground rotational state is exclusively occupied at
2 d 2 ( x) absolute zero, so this, too, makes no contribution to the energy.
 2
 V ( x) ( x)  E ( x) However, the vibrational energy levels are:
2m dx
 1
 
Hˆ  Tˆ  Vˆ   E Evib     h
 2
The Hamiltonian operator H is composed of a kinetic energy This means that the minimum vibrational energy, or zero point
part T and a potential part V. The potential energy operator V energy (ZPE) is nonzero (EZPE = h/2). Thus, we find that at
essential defines the nature of the question; the energy E is the absolute zero, the energy of the molecule is:
answer.
E  Eelec  ZPE
Most of the time, we are interested in molecules. The simplest
molecule is the hydrogen ion H2+: Q: What is the electronic energy?
e-
E  Enuc  Eelec This is the quantity that is generally delivered to you by a
quantum chemistry program like Gaussian. For the hydrogen
ion, the Schrodinger equation is:
R Hˆ nuc  Hˆ trans  Hˆ rot  Hˆ vib
H H
 2 2 1 1 1 
      (r ; R )  E ( r ; R )
If we make the Born-Oppenheimer approximation and assume
that rotations and vibrations can be separated (the rigid rotor  2 m ra rb R 
approximation), then we can come up with some equations for r is the position of the electron (variable) while R is the
the translational, rotational, and vibrational energy of the internuclear separation (parameter). Because we are making
nuclei. The translational energy depends on the volume of the the Born-Oppenheimer approximation, we solve the equation
cube V in which the molecule can move: for a fixed value of R. This is what I mean when I say that we
h2 turn a geometry into an energy. (The position of the electrons
2/3  x
Etrans  n 2  ny2  nz2  n = 1, 2, 3, ... is not fixed and is not part of the geometry.)
8mV
After quite a bit of math, one finds that the solution looks like...
h is Planck's constant while the n are quantum numbers. At
absolute zero, we can assume the molecule is not moving,
so translational energy makes no contribution to the total
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Quantum Mechanical Methods a "guess" for the wavefunction; has to be
 "reasonable" (satisfy the boundary conditions
of the problem being considered)
* the complex conjugate of  ( need not be real)

H the Hamiltonian operator, which you can think of

as a widget which gives the energy of the guess 

d a notation that means "integrate over all space"

W the variational energy

(Note that  has nothing to do with the spherical coordinate.)

So, for any guess , the ground state energy of the system E1
is no higher than W. This suggests we should try to vary , and
seek a form of it that minimizes W:
http://farside.ph.utexas.edu/teaching/qmech/lectures/node129.html
Translation:
The solid line is the bound state; the dashed line is the unbound
state. Exact solution Make a reasonable
unknown. guess, 
Unfortunately, for anything more complicated, an analytical or
exact solution is not possible. One has to resort to making
guesses. Fortunately, our guesses are very good. Just how Compute the The real ground state energy
good are they? This is answered by the Variational Theorem. integral W. is below the value of the integral.
The Variational Theorem

If a system has a ground state energy of E1, and  is a Refine  and

normalized, well-behaved function of the system's, then try again.
E1 is no higher than the variational integral W:
Q: How do we systematically improve the guess ?
W =   * H  d  E 1
Proof: Levine, I. Quantum Chemistry (5th ed.) pg 208-209.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
The Linear Combination of Atomic Orbitals (LCAO) Note: Sij  S ji (overlap of i with j = overlap of j with i)
Instead of varying the function , make  a linear combination of
some other functions, and vary the coefficients to minimize the Hij  H *ji (the Hamiltonian is Hermitian)
variational integral.
For details, please see Levine, 220-223. If we look at the
Let's try this for the hydrogen molecule. A reasonable guess solutions, we find something like this:
for  might be 1s orbitals of the hydrogen atom. Thus, we form
a linear combination: energy

= c11 + c22 first excited state,
H H 1 = 1s orbital on hydrogen A anti-bonding
A B 2 = 1s orbital on hydrogen B W2

Thus, we seek values of c1 and c2 that minimize W. More

precisely:
W W 1 2
0 0 H11
c1 c2
This is advantageous because it's hard to vary a function, W1
but it's easy to vary the coefficients c1 and c2 in a linear ground state,
combination. bonding
The Secular Equations 1s 
orbital
As it turns out, if you vary c1 and c2 to minimize W, then c1 and
c2 have to satisfy the secular equations: (The factors N are to normalize the wavefunctions. This
amounts to saying that the electron has a 100% chance of
c1  H11  S11W   c2  H12  S12W   0 being somewhere.)
Basis Sets
c1  H 21  S 21W   c2  H 22  S 22W   0 Q: What form do these orbitals take?

Clearly, we need some functions if we want to form a linear

where H12  1 H 2   1* H 2 d Hamiltonian
combination and variationally minimize the coefficients. A
matrix element/
resonance integral basis set is a collection of such functions. The shape of these
functions has been optimized for all the different elements in
S12  1 2   1*2 d
overlap the periodic table for various considerations: computational
integral efficiency, quality of results, etc.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Basis Sets What does the 6-31G* nomenclature mean? It is equivalent to
The "Gaussian-type orbitals" were popularized by Pople and the 6-31g(d) designation:
Boys and are particularly numerically efficient: d-type orbitals
are added to
gijk  Nxi y j z k exp   r 2 
i + j + k = 0: s-type Gaussian the core is described by
i + j + k = 1: p-type Gaussian
i + j + k = 2: d-type Gaussian
one CGTO composed of 6-31g(d) heavy atoms
six primitive GTOs

normalization this is a split-valence these are Gaussian-

a Gaussian basis set
constant indicates type orbitals (GTOs)
angular
momentum two CGTOs are used to describe
every atomic orbital: one with three
Each gijk is a primitive Gaussian. Special linear combinations
CGTOs and one with one CGTO
of primitives are selected to look like hydrogenlike orbitals.
These are called contracted Gaussian-type orbitals (CGTO). This generalizes to bigger basis sets:
These orbitals are used because integrating Gaussians is
very easy computationally. 6-31g(3d2f,2p)
The most common basis set of this type is called 6-31G*.
For a hydrogen atom: heavy atoms have three hydrogen atoms have
hydrogens: 2x1s sets of d-orbitals and two two p-orbitals added
sets of f-orbitals added to them
Carbons, however, have a lot more electrons, and so need Since J-couplings in NMR are transmitted through protons, the
many more basis orbitals: addition of p-orbitals to hydrogens is important for the quality of
the results of NMR predictions.
CORE VALENCE
The orbitals of anions, excited electornic states, and loosely
associated supramolecular complexes require diffuse
carbons: functions, which are designated by "+" as in 6-31+g(d,p).
These are an additional set of orbitals for the heavy atoms
2x1s 2x2s 6x2p 6x3d with small orbital exponents.
A basis set that is divided into core and valence orbitals is Q: Where do orbital exponents come from?
called "split valence." The emphasis on valence orbitals is a
reflection of the fact that more mathematical flexibility is required
to describe the behavior of the bonding electrons, which differs
between molecules, than that of the core electrons, which is
very similar between molecules.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Basis Sets However, with split-valence basis sets, one can fit one set of the
Let me remind you that we want to construct a variational guess basis functions to the STOs, but it is unclear what to fit the
for the wavefunction that has coefficients that can be optimized remaining basis functions to. By "fit," I mean adjust the orbital
to minimize its variational energy. The number of two-electron exponents of the CGTOs and the linear combination coefficients
integrals increases as N4 where N is the number of basis of the CGTO expansion to best fit the STO. The answer is to
functions. Therefore, we want to minimize not only the number use the variational principle. Pople and co-workers came up
of basis functions but the computational cost involved in with a test set of molecules, and optimized the parameters to
evaluating any particular integral. Ideal orbitals should also give the best results for a given method.
have a lot of density where there is electron density in real life.
There are many other basis sets as well. The most popular
If you recall the hydrogen molecule example, we guessed that competitor are the Dunning correlation-consistent basis sets.
the orbitals in the hydrogen molecule (which we don't know) One can imagine that as the basis set gets larger, the
were similar to the orbitals in the hydrogen atom (which we do mathematical flexibility (the "span") gets larger and larger until
know). The generalization of this is the Slater-type orbitals it becomes infinitely flexible. Thus, the variational energy should
(STOs): drop as the basis set decreases in size:
energy

basis set size

For reasons that will be obvius soon, the infinite basis set is
called the "Hartree-Fock (HF) limit." The question is: how
smoothly will the energy go down as the basis set size goes
up? As it turns out, the quality of a calculation depends a lot
on how well electron correlation is taken into account. The
Dunning basis sets are variationally optimized to add constant
increments of electron correlation energy as the basis set size
The STOs are advantageous from a chemical standpoint increases. The nomenclature is "cc-pVNZ," for correlation
because "real" orbitals have cusps (whereas GTO primitives consistent polarized valence (double (D), triplet (T), ...) zeta.
do not). However, integrals of STOs are very annoying from a Diffuse functions are designated by the "aug" prefix, as in:
computational standpoint, so they are not generally used. aug-cc-pVDZ for a double-zeta basis set.

GTOs are also "bad" chemically because they have no radial A new trend is the use of density fitting or resolution of
nodes. The solution to both these problems is to combine a the identity (RI) methods. These expand the basis set with
number of GTO primitives into a CGTO. some auxiliary basis set to speed up calculations.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
The Orbital Approximation and the Pauli Principle
A helium atom has two protons and two electrons:
1  r1  2  r2  3  r3   1  r2  2  r1  3  r3 
r12 All I have done here is to make 1 a function of the coordinates
e- e- of electron 2 instead of electron 1 and the reverse for 2. So
what would satisfy the Pauli principle? The Slater
RA1 RA2 Determinant.
H
As you know, if you shoot some lithium atoms through a
RA magnetic field, some of them will curve to the left and some
The Hamiltonian now includes an electron-electron repulsion will curve to the right (whereas helium atoms won't curve):
term:

  2r1  r22 2 2 1 
        r1 , r2   E   r1 , r2  S N S N

 2 2 R A1 R A 2 r12 
Li
Nobody knows how to deal with this exactly, so one makes
the orbital approximation that the overall wavefunction is the This called spin because the lithium atoms behave like
product of a number of one-electron wavefunctions, or orbitals. clasically spinning charged spheres (even though they're not
The composite wavefunction is called a Hartree product: really spheres or spinning). For every spatial wavefunction,
we can put an  or  electron in it to make a spin-orbital.
  r1 , r2   1  r1  2  r2  For example, putting an  electron in orbital 1 gives:
For a lithium atom, you might think that you can just write a  1 (1)  1 (1) (1)
Hartree product like this:
The Slater determinant gives us a way to construct anti-
  r1 , r2 , r3   1  r1  2  r2  3  r3  symmetric wavefunctions. For helium, we need to construct
one for two electrons:
However, you will find that if you use hydrogen-like orbitals for
the , you will find that the answer violates the variational 1 1 (1) (1) 2 (1) (1)
principle! That is, the variational energy of the wavefunction  He 
will actually be lower than the experimental wavefunction. 2 1 (2) (2) 2 (1) (1)
The problem is that this wavefunction violates the Pauli
1
exclusion Principle, which not only prevents two electrons
from having the same quantum numbers, but also requires
 1 (1) (1)2 (1) (1)  2 (1) (1)1 (2) (2)
that the wavefunction be antisymmetric with respect to
2
electron interchange. If I interchange the indices in the trial Now, interchanging the electrons does result in an anti-
function for lithium above, I do not get the negative of the symmetric wavefunction. Note that every electron appears in
trial function: every spin orbital somewhere; the electrons are identical.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Electron Correlation - Processes which break bonds cannot be described well.
I already mentioned that electron correlation is essential to For example, pericylic reactions can involve a number of
non-covalent interactions like -stacking. The crosses in the stretched bonds with some diradical character. HF cannot
graph below show the potential energy curve for benzene predict the barrier heights of these correctly.
dimer:
- Other measures of energy differences like isomerization
energy (e.g., CO + HO radical vs. H radical + CO2) don't
come out right either.

In general, molecular geometries are more accurate than

energies, regardless of the method. HF is not exception.
Note that geometry with the HF method (or any other method)
requires analytic gradients to be fast enough for practical use.
This means we must know the derivatives of the density
matrix elements with respect to the coordinates of the system,
which is not really obvious. Fortunately, Pulay discovered an
efficient way to do this in 1969, and fast HF methods are now
widely available.

The important point is that despite all this computations

accurately describe reality:

Accuracy of HF Results (David Sherill)

Notice that HF thinks that the benzene dimer is not bound at bond lengths: to within 0.02 A
all! Clearly, this is wrong. Interestingly, some of the other bond angles: to within 2 degrees
methods there do account for some electron correlation, but vibrational frequencies: to within 10%
clearly don't do it quite correctly. For example, BLYP is a dipole moments: to within 0.3 D
common kind of density functional (to be discussed shortly) bond dissociation energies: 25-40 kcal/mol off
and accounts for some electron correlation, but still doesn't
understand the stacking interaction. Fancier methods are better. For coupled-cluster:

The neglect of electron correlation has other serious bond lengths: to within 0.004 A
consequences: bond angles: to within 0.03 degrees
vibrational frequencies: to within 2%
- Heats of formation are very inaccurate. For a bunch of dipole moments: to within 0.05 D
small molecules calculated at HF/aug-cc-pVQZ, a very bond dissociation energies: 1-2 kcal/mol off
large mean unsigned error of 62 kcal/mol was found!
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Post-HF Methods Hartree-Fock limit: HF with an infinite basis set
Q: How can we take electron correlation into account?
If we variationally minimize the Slater determinants resulting
There are many methods available for this, but in general, the from all these excitations and have an infinite basis set, then
more accurate the method, the more costly it is. The "scaling" we have reached an exact numerical solution for the non-
of the method tells you how much more time you need as you relativistic Schrdoinger equation--the configuration interaction
increase the number of basis functions N. One must also limit.
consider a "pre-factor" which tells you how much "up front" cost
is required. Of course, all these excitations are very costly. In CISD, we
just make single and double excitations. In CCSD, we are
Hartree-Fock ~ N4 trying to do the same thing, but in a different way. We apply
DFT (density functional theory) ~ N3 (but larger pre-factor an operator exp[T] to the HF wavefunction:
than HF)
  eT  HF
ˆ
MP2 (Moller-Plesset perturbation theory) ~ N5
CISD (configuration interaction singles and doubles) ~ N6
CCSD(T) (coupled cluster singles and doubles with Tˆ  Tˆ1  Tˆ2  Tˆ3 
perturbative estimates of triples) ~ N7
where the indices indicate the level of excitation the operator
The last one, CCSD(T), is the "gold standard" computational
represent. This has certain advantages, among them size
method, but is staggeringly expensive in CPU and memory
requirements. Just calculating the benzene dimer with CCSD(T) consistency. If you calculate the energy of A separately and B
separately, the sum should be about the same as calculating the
with it was a research-grade project worthy of a JACS article
energy of A and B together, but separated by quite a large
just five years ago!
distance. If that's true, the method is called "size consistent."
The idea behind the MP2, CISD, and CCSD(T) methods is
In MP2, we apply perturbation theory to correct the energy of
relatively simple to understand. The Hartree-Fock forms its
the Hartree-Fock wavefunction. Thus, the orbitals are the same,
guess from a single Slater determinant. If we want to improve
but the energies are different. (This means the optimized
the guess, we can include determinants involving "excited
geometries are different, too.) As organic chemists, we work
states":
with molecules containing dozens of heavy atoms, if not
hundreds. As such, we can only afford HF and MP2, and even
then MP2 is problematic for larger atoms.
CI A huge breakthrough in quantum chemistry arrive in the 1990s,
+ + many more
when density functional theory (DFT) became widely
available. Although it has a larger pre-factor than HF, its
scaling with N is considerably less than the other post-HF
methods, making it an attractive, practical method.
"single" "double"
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Density Functional Theory (DFT) In Thomas-Fermi DFT (1927), one tries to calculate the
Q: How can we take electron correlation into account in energy in a classical way. The attraction between the density
a practical way? and the nuclei is:
(This discussion is taken from Cramer, Chapter 8.) nuclei
Z
Ab initio methods like Hartree-Fock or MP2 are ways to arrive
Vne    r      r  r   r  dr
k
k

k
at the wavefunction, which is itself a proxy for electron density
Electron-electron repulsions are given by:
(or other observables via the action of operators). In DFT, we
try to compute the electron density  directly. The total number 1   r1    r2 
2   r1  r2
of electrons N is conserved over all space: Vee  dr1dr2
N     r  dr What about the potential energy? In this crude treatment, one
envisions an infinite number of electrons moving through an
Is the electron density enough information to reconstruct the infinite volume of space containing a uniformly distributed
information we want--the energies and wave functions? Yes! positive charge. This is the "uniform electron gas." Thomas
The Hamiltonian requires the positions and atomic numbers of and Fermi showed that:
the nuclei and the total number of electrons. Clearly, having
3
 3 2    5/3  r  dr
2/3
the density can give us the total number of electrons. Tueg 
10
What about the nuclei? They are point positive charges, so the
electron density reaches local maxima at the nuclear positions. No two electrons can be in the same place, so one introduces a
So we can reconstruct the positions of the nuclei from the "hole function" to account for this:
maxima in electron density. Their atomic numbers can be
extracted from: electrons
1 1   r1    r2  1   r1  h  r1 ; r2 
    
r1  r2
dr1dr2   
r1  r2
dr1dr2
  rA  i j rij 2 2
 2 Z A   rA 
rA rA  0
LHS: exact quantum-mechanical interelectronic repulsion
RHS, first term: classical interelectronic repulsion
where rA is the position of an electron density maximum,  bar RHS, second term: corrects the density with a hole function h
is the spherically averaged density, and Z is the atomic number
of nucleus A. The hole function for a multielectron system is not obvious and
must often be approximated. Not having correct hole functions
So we have shown that if we had the electron density, we would has serious consequences--electrons can interact with
(at least in principle) be able to write down the Schrodinger themselves! In HF, there is no self-interaction error, but
equation, solve it, and get the energies and wavefunctions. all DFT methods suffer from this to some extent. This is why
But how do we actually go about doing that? Specifically, many DFT methods are combined with some degree of HF
how does one turn the density into energy? exchange energy.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Density Functional Theory (DFT) So far, it looks like we have to guess a density, convert it into a
The fact that electrons are indistinguishable and must be be candidate Hamiltonian and wavefunction, evaluate the energy
antisymmetric with respect to interchange leads to a purely by solving the Schrodinger equation, and keep guessing until
quantum mechanical effect known as exchange and has no the variational energy is minimized. But this is obviously rather
classical analog. The presence of exchange keeps electrons unsatisfactory! How do we go about finding better densities?
farther apart than they would be predicted to be classically and How do we convert these densities into energies without solving
is responsible for steric repulsion. the Schrodinger equation? Remember, the whole point of using
the electron density is to avoid solving the Schrodinger equation
As it turns out, the contribution of exchange to the classical in the first place!
interaction energy is much larger (1-2 orders of magnitude) than
the electron correlation energy. Slater proposed that the The answer is the Kohn-Sham self-consistent field method,
"exchange hole" can be approximated by a sphere of constant which is analogous to the SCF method used in Hartree-Fock.
potential with a radius depending on the magnitude of the The idea is to consider a fictitious system of non-interacting
density at that position (Cramer, pg 252): electrons that have the same electron density as the real system
1/3
where the electrons do interact. Then, we divide the energy
9 3  functional into different components:
Ex       r  dr
 4/3

8  E    r    Tni    r    Vne    r    Vee    r  

T    r    Vee    r  
This is "Slater exchange." Unfortunately, Thomas-Dirac DFT is
entirely inadequate to describe anything of chemical interest. In where Tni is the kinetic energy of the non-interacting electrons,
fact, it erroneously predicts that all molecules are unbound Vne is the classical nuclear-electron attraction, Vee is the
states. Additionally, there was no clear analog to the variational classical electron-electron repulsion, T is the quantum-
principle, so the theory made little impact on chemistry. mechanical correction to the kinetic energy, and V is the
quantum-mechanical correction to the electron-electron
In 1964, Hohenberg and Kohn proved two theorems that proved repulsion. (A functional is a function that takes functions, rather
crucial to establishing DFT as a useful tool in quantum chemistry. than numbers, as arguments.)
This led to the awarding of the Nobel Prize to Kohn in 1998
(along with Pople for developing other computational methods). What is the nature of T and V? This is called the "exchange
correlation energy (Exc)" and accounts for self-interaction
Hohenberg-Kohn Existence Theorem energy and the difference in kinetic energy between the
fictitious and real systems, in addition to exchange and
There is a one-to-one correspondence between the ground correlation. Nobody knows what the exact functional is. (If
state wavefunction and the ground state electron density. you could figure this out, you'd win a Nobel Prize.) In DFT
methods, one makes a guess for what the functional is. That's
Hohenberg-Kohn Variational Theorem where all the different flavors of DFT come from (B3LYP, PW91,
M05-2X, etc.)--they are different approximations to Exc. Note
The electron density that minimizes the total energy is the that while exact DFT is variationally correct, approximate DFT
exact ground state density. is not variationally correct. However, both are size-consistent.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Density Functional Theory (DFT) In adiabatic correction methods, one computes some of the
So Hartree-Fock is an exact solution to an approximate exchange correction exactly with HF, and then includes some
theory, while DFT is an approximate solution to an exact terms from the other expressions. For example, in the popular
theory. It just so turns out that DFT gives much better results B3LYP functional, one uses the three-parameter scheme:
than HF. Purists will complain that since approximate DFT is
not variationally correct, there is no systematic way to improve
B 3 LYP
E XC  (1  a) E XLSD  aE XC
HF
 bE XB88  cECLYP  (1  c) ECLSD
DFT results, and therefore, wavefunction methods are
preferable. They are correct about the first part, but wrong This is a "hyper GGA" functional, since it includes a GGA part
about the second part--it is now well established that DFT gives and a Hartree-Fock exchange part. This gives rise to a ladder
chemically meaningful information about a wide variety of of functionals:
interesting systems.
unocc.
Q: What are the various flavors of DFT? orbitals fully non-local (not available)
Eex hybrid/hybrid meta/hyper GGA
There are a lot of choices here, so I will just summarize the key (B3LYP, B3PW91, B1B95...)
points. 


meta GGA (BB95, MPW1K, TPSS...)

accuracy,
The local density approximation (LDA) assumes that the 

exchange-correlation functional only depends on the electron GGA (BLYP, PBE, BP86...)

cost
density at the point where the functional is being evaluated. 
Systems involving unpaired electrons use the local spin LSDA (useful for solid state)
density approximation (LSDA). The functionals developed by
Vosko, Wilk, and Nusair (VWN) are an example of this. The There are numerous benchmarks which show that the accuracy
details are quite abtruse--it's not clear where all the terms in goes up as you go up the ladder (with a lot of nuances). For
the expression come from, physically speaking. They were this course, we will just use B3LYP, which is known to work
obtained by empirically fitting parameters to some known well for many systems. However, it does not describe medium-
results. In this sense, DFT methods can be considered truly to long-range electron correlation very well like that involved in
semi-empirical. -stacking. Functionals like DFT-D (Grimme) and M06-2X
(Truhlar) have been specifically parametrized to work well for
Of course, in a real system, the electron density is not uniform. these cases.
The first-order correction to this is to account for not only the
electron density at a particular point, but also the rate at which DFT scales as N3, whereas HF scales as N4, where N is the
the density is changing at that point--the gradient. This leads to numebr of basis functions, so there is a clear advantage for
the generalized gradient approximation (GGA). One popular larger systems. However, note that the pre-factor for DFT is
kind of GGA DFT is called PW91. larger, particularly if HF exchange is needed.

Going futher, we can have meta-GGA methods which include Further Information: Nick Mosey has prepared a detailed set of

higher-order terms like the Laplacian 2. course notes on DFT which are available on the web here:
www.chem.queensu.ca/people/faculty/Mosey/chem938.htm.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Jumping the Gun To get started, you will need several things:
Clearly, computational chemistry is immensely complicated,
and you are not going to learn it all from me in an hour. I don't (a) GaussView 5.0 installed on your computer (freely available
really even understand a lot of it. But don't feel bad--there is for PC and Mac from the chemistry library)
absolutely no need to understand everything before getting (b) a secure FTP client (Mac: Fetch; PC: WinSCP)
started, just as you don't need to know exactly how a car's (c) a secure telnet client that supports keychain authorization
engine works before driving. Professor Jacobsen reminds me (PC: PuTTY; Mac: terminal)
on occasion that as organic chemists, our talent is for making (d) an account on the Odyssey Computing Cluster (http://
very simple models of very complicated systems, and getting rc.fas.harvard.edu; [email protected]).
out some results that make a lot of sense. My philosophy is that
knowing a little about how things work will enable us to choose Computations themselves do not cost any money, but buying
computational methods appropriately, design computational computers to which you have priority access on the cluster does.
studies that answer real chemical questions, and troubleshoot They're not cheap (starting price: $10 000). But the cluster has
calculations when they invariably crash. However, it is easy to computers everyone can use for free (but you have to wait).
get bogged down in the details of, say, how to compute the The Odyssey Cluster
Hartree-Fock density matrix. Although learning these details old days: buy one really fast, expensive computer that can do
takes a copious amount of time, they don't really help us get one calculation at a time
any of the answers we need in organic chemists. So I suggest
we all try to get a general sense for all of the different methods, these days: buy a lot of decent computers, and then use a lot
and open up dialogs with physical chemists to collaborate when of them at once
we need to do something a bit more complicated.
The Odyssey Cluster lets us do the latter. The magic of
In this tutorial, which I strongly suggest you go through at home, computer science means that computations can be done in
we will do two things: parallel, which means that every calculation is broken up into
many pieces, and different computers work on different parts
(1) Compute the energy difference between axial and of the calculation simultaneously. This results in drastic
equatorial 2-chlorotetrahydropyran. speedups in computational efficiency, without the enormous
O O cost of supercomputers.
Cl
Cl Anyone with an FAS affiliation can have an account on
(As you know, the stability difference is anomeric in nature.) Odyssey and run calculations. If you belong to a particular
research group, you may have access to certain computer
(2) Compute the rotational barrier in ethane. resources that other people don't. For example, the Evans
group has 8 "nodes." Every node contains 8 CPUs, or cores.
H H H
H H H H H H At the moment, "cross-node parallelism" is theoretically
H possible, but unnecessary for our purposes. Thus, we are
H H H H generally limited to 8 cores/job. (The Jacobsen group has
H HH H recently purchased some 12-core nodes.)
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Computational Workflow Q: How do I access the cluster?
No matter what the project you're working on is, everyone The first step is to login to the cluster via SSH (odyssey.fas.
follows the same workflow. We'll consider each step in turn: harvard.edu). The first time you login, you may be asked
whether to accept a "fingerprint;" say yes. For security
This is the hardest part, but reasons, you are asked to type in your login name, password,
think of a computation unfortunately, there's no room to
that you want to run talk much about this. and RSA security token passcode. The idea is that nobody can
login as you, unless they have both your password and
passcode. The passcode changes with time in a pre-defined
You have to tell Gaussian, a way; you read it off a small plastic key fob ("hard token") or a
create a Gaussian
quantum chemistry software readout on your screen ("soft token"). Most accounts are now
input file (*.gjf)
package what you want it to do. setup with soft tokens.

Computing resources are precious, When I login, it looks like this:

submit job login as: ekwan
to Odyssey and we all share them using the Using keyboard-interactive authentication.
LSF queueing system. Password:
Using keyboard-interactive authentication.
Enter PASSCODE:
retrieve results (*.out) Output comes back in a text file, Last login: Fri Feb 4 09:35:40 2011 from c-66-30-9-
8.hsd1.ma.comcast.net
from Odyssey but you can use GaussView to
look at the results graphically. Java Runtime Environment 1.6 module
****************************************************

analyze results This is where you have to use This module sets up the following environment
your chemical intuition. variables for Java Runtime Environment 1.6:
PATH /n/sw/jdk1.6.0_12/bin

Idea: Not everyone is using their computers all the time, so ****************************************************
let's share the unused resources in a fair way.
Loading module devel/intel-11.1.046.
Loading module hpc/gaussian-09_ekwan.
Implementation: Different clusters use different programs, [ekwan@iliadaccess01 ~]$
but here at Harvard, we use the LSF program. Here is some When I typed my password and passcode, no text was mirrored
terminology: to the screen (so it's harder to steal my password if you happen
to be looking over my shoulder). The other information tells me
job: a computation you want to run on the cluster when I logged in last and from where, as well as what modules
core: an individual CPU have been loaded. The cluster is not just for Gaussian; many
node: a bunch of CPUs put together into one computer other scientists are using other programs. Loading modules
prepares the cluster to use Gaussian by setting certain
queue: a group of jobs waiting to run on the cluster; different environment variables, aliases, etc.
queues target different nodes with differing priorities depending
on the user who submitted the jobs Now, we're ready to take a look around.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
The Queue System immediately. My job will run, and then your job will resume.
Q: What computing resources are available? The "parallel" queues are also open access, but are not subject
[ekwan@iliadaccess01 ~]$ bqueues to suspension.
QUEUE_NAME PRIO STATUS NJOBS PEND RUN SUSP
dae 200 Open:Active 64 0 64 0 Both the parallel and serial queues are subject to time limits.
enj 200 Open:Active 84 0 84 0 Your job will terminate automatically if it exceeds its allotted
karplus 196 Open:Active 24 0 24 0 time, which is calculated from when it starts (not when it's
lsdiv 55 Open:Active 143 0 143 0
betley 45 Open:Active 4 0 4 0 submitted).
short_parallel 30 Open:Active 2 0 2 0
normal_parallel 25 Open:Active 2561 2284 265 0 short: one hour limit
long_parallel 20 Open:Active 2969 1304 1665 0
long_serial 15 Open:Active 150 0 150 0 normal: one day limit
short_serial 10 Open:Active 8472 7907 532 33 long: one week limit
normal_serial 5 Open:Active 3296 1906 1150 240 unrestricted: no limit
unrestricted_pa 3 Open:Active 159 0 159 0
unrestricted_se 1 Open:Active 161 97 64 0
number of cores pending, The longer the time limit, the longer it will take for your jobs to
(This is a truncated listing.) running, or suspended start running. When the cluster is busy (about three quarters of
the time), you will find that jobs submitted to normal_parallel
The LSF system tries to allocate the available resources fairly. will run within a day. (You will get a sense of how long your
Jobs are submitted to queues, each of which targets certain jobs will take as you go along. The molecules in this tutorial
nodes. However, unlike lines in real life, these are not first-in, will take less than an hour.)
first-out. Priority within a queue is assigned based on how
much computing power you have been using within the last Both the parallel and serial queues target eight-core nodes.
few days. For example, suppose my co-worker Joe submits One consideration is that parallel queues run jobs "exclusively,"
100 eight-core jobs to dae. dae only has 8 eight-core nodes, which means that only one job can run on them at a time.
so supposing they're all free at the moment, all 8 start running Therefore, you should always submit 8-core jobs to the parallel
immediately. Then, suppose I decide to submit a job, and I nodes. The serial nodes do not have this restriction, so the
haven't been doing any computations lately. My job will run fewer nodes you request, the higher the chances are that there
before any of Joe's remaining 92 jobs, but not before one of the will be a node somewhere that can accomodate your request.
8 jobs currently on dae finish.
What is the total activity on the cluster?
Not everyone has access to all the queues. For example, only [ekwan@iliadaccess01 ~]$ busers allusers
people in the Evans group have access to the dae queue
USER/GROUP NJOBS PEND RUN SSUSP USUSP RSV
(since Professor Evans paid for the computers). However, allusers 23416 14599 8398 249 0 170
you can still run jobs on our nodes if you submit your jobs jwzorek 1072 768 304 0 0 0
to the common queues, which end with "serial." These target
a mixture of FAS-owned computers and computers owned by There are about 15 000 cores worth of jobs pending right now,
individual research groups. However, these jobs are subject which means the cluster is busy (but not super busy). Joe and
to "suspension," which means that if you are running a job on I are working on a big project right now, so we have quite a few
a dae node, and I submit a job, then your job will be be paused jobs waiting.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Managing Your Jobs
Q: How do I see what jobs I'm running right now?
[ekwan@iliadaccess01 ~]$ bjobs -u jwzorek -w -a

JOBID USER STAT QUEUE FROM_HOST EXEC_HOST JOB_NAME SUBMIT_TIME

24467474 jwzorek RUN dae iliadaccess01 8*dae022 monorhizopodin_OPLS_low_39 Jan 28 14:59
24467480 jwzorek RUN dae iliadaccess01 8*dae012 monorhizopodin_OPLS_low_4 Jan 28 14:59
24467481 jwzorek RUN dae iliadaccess01 8*dae013 monorhizopodin_OPLS_low_4_unsolv Jan 28 14:59
24467486 jwzorek RUN long_parallel iliadaccess01 8*hero0408 monorhizopodin_OPLS_low_7 Jan 28 14:59
24467487 jwzorek RUN long_parallel iliadaccess01 8*hero2409 monorhizopodin_OPLS_low_7_unsolv Jan 28 14:59
25611512 jwzorek PEND normal_parallel iliadaccess02 - still_38_OPLS_1_prod Feb 7 20:04
25611515 jwzorek PEND normal_parallel iliadaccess02 - still_38_OPLS_low_11_prod Feb 7 20:04
25611516 jwzorek EXIT normal_parallel iliadaccess02 8*hero1916 still_39_OPLS_low_1_prod Feb 7 20:04
24467468 jwzorek DONE long_parallel iliadaccess01 8*hero1405 monorhizopodin_OPLS_low_36 Jan 28 14:59

(This is also a partial listing.)

(1) I needed the "-u jwzorek" flag because I'm not running any jobs at the moment. That tells it to display only the jobs being run
by jwzorek. The -w flag requests a "wide" listing, which helps me see the long names of the jobs. Long, descriptive names,
preferably connected with a painfully detailed and rigorous filing system are a very good idea, particularly if you are going to run
a lot of jobs at once. The -a flag will show "all" the jobs, including the ones that have recently crashed or completed.

(2) The status column tells you if the jobs are running (RUN), waiting in a queue (PEND), have recently completed successfully
(DONE), or have recently crashed (EXIT). Every time a job terminates, you get an email from LSF.

(3) EXEC_HOST tells you where the job is running right now. For example, the first job is running on eight cores on dae022.

(4) If you are impatient and want to see why your job is still pending:
[ekwan@iliadaccess01 ~]$ bjobs -l 25611512 Among a lot of other omitted output, this tells
us that the job isn't running mostly because
Job <25611512>, Job Name <still_38_OPLS_1_prod>, User <jwzorek>, Project <defau
lt>, Status <PEND>, Queue <normal_parallel>, Command <#!/b there aren't enough free nodes right now.
in/bash; #BSUB -q normal_parallel...
Mon Feb 7 20:04:44: Submitted from host <iliadaccess02>, CWD <$HOME/jobs1/stil
l_38_OPLS_1_prod>, Exclusive Execution, 8 Processors Reque
sted, Requested Resources <span[ptile=8]>;
PENDING REASONS:
Not specified in job submission: 902 hosts;
Job's requirement for exclusive execution not satisfied: 422 hosts;
Not enough hosts to meet the job's spanning requirement: 7 hosts;
Job slot limit reached: 7 hosts;
Closed by LSF administrator: 16 hosts;
Unable to reach slave batch server: 2 hosts;
Load information unavailable: 11 hosts;
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Managing Your Jobs This confirms that a Gaussian subroutine, called a "link," is
Occasionally, you will want to shuffle your jobs around. Here currently active and using all eight cores. Every link is
are some useful commands, each of which requires a jobid to associated with a particular part of Gaussian. In this case,
tell it which job to work on. If you put in "0," it tries to apply the link 701 is "one electon integrals first or second derivatives."
command to all your jobs. A full listing can be found here:

bkill [jobid] - abort a certain job http://www.gaussian.com/g_tech/g_ur/m_linklist.htm

Sometimes, a job will fail to terminate, even if it is done or you Finally, you can look at what's going on with "Ganglia," which is
issue the bkill command manually. In that case, use the "-r" flag available at http://software.rc.fas.harvard.edu/ganglia/:
to kill the zombie job.

bstop [jobid] - pause a certain job; status changes to USUSP

bresume [jobid] - resume said job; status reverts to PEND or
RUN, depending on what's going on

btop [jobid] - move the job to the top of the queue (at least, the
top of the jobs from you in that queue)

You cannot alter anyone else's jobs.

To monitor the activity on a certain node that's running your job: The above shows aggregate CPU usage; the below shows
[ekwan@iliadaccess01 ~]$ ssh dae022
CPU usage per node. There are a lot of other options.

Last login: Mon Nov 15 01:05:53 2010 from iliadaccess02.rc.fas

harvard.edu

...

[ekwan@dae022 ~]$ top

...

PID USER VIRT RES SHR S %CPU %MEM TIME+ COMMAND

1546 jwzorek 4468m 420m 3124 R 798.1 1.2 30642:03 l701.exe
13838 ekwan 10988 1136 680 R 3.9 0.0 0:00.03 top
24309 root 21760 5792 1524 S 1.9 0.0 132:22.94 lim

(There's not enough room here to see everything. The other

columns are about other resource usage. For a full explanation,
try "man top" to pull up the manual page for top.)
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Test Case #1
I am now ready to take you through submitting your first job.
Our first task is to draw axial 2-chlorotetrahydropyran in
Gaussview. Click on the leftmost button in the toolbox that
looks like "6C" and then on "carbon tetrahedral."

You have various options for manipulating the whole screen.

If you are not clicking on any part of a bond or molecule:

roll mouse wheel: zoom in or out

click and drag: rotate entire view
control-click and drag: rotate entire view in plane
shift-click and drag: move entire view

If you are clicking on part of the molecule:

alt-drag: rotate that molecular fragment

shift-alt-drag: move that molecular fragment

To delete atoms, click on the small X tool button (not the big
X button, which deletes the entire molecule!). Click on
individual atoms to delete them.
delete whole molecule use this

A periodic table comes up. You can select different valences

for the various elements. Click somewhere on the blank
window behind the periodic table. A methane molecule appears.
Clicking somewhere else produces a new methane molecule:
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117

You can connect two atoms with a bond using the single bond The two adjacent tools adjust the bond angle (select three
tool. However, keep in mind that the bonds you draw are a atoms) and dihedral angle (select four atoms). The tool with a
convenient tool for visualizing the connectivity in the molecule, question mark in it will show you the relevant bond distance,
and not relevant to the energy calculation (unless you’re doing angle, or dihedral in the status bar if you select the relevant
molecular mechanics). Click on two atoms, and then on the atoms.
tool, which pulls up this dialog box:
single bond tool Clearly, this is a cumbersome way to go about your business.
For the more common fragments, a template tool is available
by clicking on the benzene ring next to the “6C” button:

Clicking in the blue window will paste in a cyclohexane ring, or

whatever other template you desire. Use these tools to draw
axial 2-chlorotetrahydropyran. (Clicking on an existing atom
with a different atom type will convert it to the new atom type.
Use this to construct the ether.)
adjust bond distance here
(angstroms by default)
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117

My result looks like this: H -0.32653189 0.84083827 2.12419100

H -1.71108389 1.38162127 1.09393800
O 1.47538911 2.75652027 1.16066100
Cl 1.60357089 2.74535830 -1.51344298

1 2 1.0 5 1.0 10 1.0 9 1.0

2 3 1.0 7 1.0 8 1.0 This is the “connectivity table.”
3 6 1.0 15 1.0 16 1.0 It tells Gaussian about where
4 5 1.0 11 1.0 12 1.0 15 1.0 all the bonds are. This can be
5 14 1.0 13 1.0
6 useful for molecular mechanics
7 or in some rare cases for
8 troubleshooting a problem with
9
10 the coordinate system, but we
11 don’t need it here, so delete it.
12
13
14
15
16
Next, we need to save the file. Save it as a Gaussian input
file (*.gjf) and check “Write Cartesians.” This generates a text The coordinates will be different for the molecule you draw,
file that you can open in Wordpad or some other text editor: due to translation and rotation of the whole molecule.
However, Gaussian will rotate the entire specification to the
%chk=C:\Users\Eugene Kwan\Desktop\test.chk “standard basis” before running the calculation, so this is of
# hf/3-21g geom=connectivity no consequence.
Title Card Required
The header tells Gaussian what to do. The default header
0 1 is not what we want, so change it to this:
C -0.04198489 0.54090527 0.00000000
C 1.47312111 0.54090527 0.00000000 %chk=checkpoint.chk
C 2.02505211 1.95198327 0.00000000 %mem=3GB
C -0.03973589 2.75718127 1.16017200 %nprocshared=8
C -0.59253589 1.34656027 1.15887600 #t opt freq=noraman b3lyp/6-31g(d) pop=nbo
H 3.14365111 1.91799227 0.06271400
H 1.84567011 -0.00472873 0.90656200
title
H 1.84841511 -0.00895373 -0.90191000
H -0.41467289 0.97393427 -0.96538500
H -0.41758189 -0.51331473 0.06350200 0 1
H -0.41505589 3.30660927 2.06228600
H -0.41159089 3.30378227 0.25384900
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
%chk=checkpoint.chk (4) Route Card: This is the line that starts with a pound (#)
%mem=3GB
sign. This is the crucial line that tells Gaussian what to
%nprocshared=8
#t opt freq=noraman b3lyp/6-31g(d) pop=nbo do with the geometry you have given it. If you need to
write more, you can continue it onto the next line.
title
#: Indicates start of route card
0 1 t: Requests “terse” output. Nothing gives a standard
(1) Checkpoints: Gaussian stores results every so often in amount of output while p gives a verbose level of
a checkpoint file so that the calculation can be restarted output.
if it crashes. The first line tells it to store the checkpoint opt: Requests an optimization to a ground state
for this job in checkpoint.chk. freq: Requests a frequency job. The sub-keyword
“noraman” tells it not to compute the Raman transitions.
(2) Memory: This is the amount of RAM that the job will b3lyp/6-31g(d): method and basis set
use. This is not to be confused with the amount of swap pop=nbo: ask for a standard orbital analysis
space, which will be dynamically allocated. Volatile
RAM is 10-100 times faster than more permanent hard I don’t have room to explain all the Gaussian keywords
disk space. Most nodes have at least 24 GB available, here. However, a detailed list is available online here:
but 3 GB is perfectly sufficient for most jobs. Allocating
too much memory can slow a calculation down. In www.gaussian.com/g_tech/g_ur/l_keywords09.htm
general, the number of integrals a calculation will require
is far greater than the amount that can be stored in
RAM. Thus, a “direct” algorithm is used in which (5) Title Card: This is the title of the job. You don’t have to
integrals are calculated, and then forgotten, on the fly. change this if you don’t want to. I keep track of jobs by
This turns out to be faster than swapping them to disk. their filename instead.

(3) Shared Processors: This tells Gaussian that you want (6) Charge and Multiplicity: This tells Gaussian how many
it to use up to 8 processors. Occasionally, you will want electrons to use. 1 1 would indicate a positively charged
to cut this number. Because there are some overhead singlet. -1 2 would indicate a negatively charged
costs associated with parallel computing, an eight-core doublet. Calculations involving open shell species are
job is not eight times faster than a one-core job. significantly more complicated, and should only be
However, the parallel code in Gaussian is very good and undertaken if you know what you’re doing. It’s not
the extra speedup is worth it, particularly for calculations simply a question of plugging in the geometry and
where you want a result quickly. getting an optimized ground state result.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Now, we are ready to submit the job. You will need to use your Troubleshooting Note: Occasionally, there is still a problem with
secure FTP client to do this. Use odyssey.fas.harvard.edu as the format of the carriage returns and line breaks when you
the domain name and enter in the same password. You will transfer a file from your computer to the cluster. In that case, try
need a new passcode, however. Note that the same passcode the dos2unix command (use “man dos2unix” for more details).
cannot be used for two logins; if you use one for logging into the
terminal, then you need to wait for a new one to start an SFTP Now that the job is copied onto the cluster, we can take a look at
session. On my screen (different if you use a different client): it:
[ekwan@iliadaccess01 chem106]$ ls -l
total 160
-rw-r--r-- 1 ekwan evans_lab 1044 Oct 19 11:00 2-
chloroTHP_ax.gjf
-rwxr--r-- 1 ekwan evans_lab 757 Feb 25 2010 analyze.sh
-rwxr--r-- 1 ekwan evans_lab 1436 Feb 25 2010 eek.sh
drwxr-xr-x 2 ekwan evans_lab 2048 Nov 24 18:02 jobs
drwxr-xr-x 2 ekwan evans_lab 7168 Jan 15 19:15 output
-rwxr--r-- 1 ekwan evans_lab 347 Jan 16 2010 submit.sh
-rwxr--r-- 1 ekwan evans_lab 521 Jan 13 07:50 template.sh

As you can see, the .gjf file is now in the directory. The
filenames appear in the rightmost column. Their permissions
appear in the leftmost column. The dates the files were last
The left-hand panel shows the directory structure of the remote
modified are also shown, along with their sizes in bytes.
session. The right-hand panel shows the contents of the active
directory. The top bar shows the current location. Note that the
To submit the job, you can invoke the submission script I have
current location in your SFTP session is not the same as in your
written. (A script is a small program written, in this case, in
current SSH session.
bash, which is a high-level, interpreted language.) It performs a
number of repetitive housekeeping duties that you will not want
Copy the input file (.gjf extension) into a directory of your choice.
to be bothered with. In particular, it runs your job in a separate
Be sure to transfer it as text not binary or you will encounter
directory with the same name as the input file, tells LSF how
some weird errors. Typically, you will not want to copy it into
many cores you want and which queue to submit the job to,
your root. Here are some useful file management commands
creates a scratch directory on the local node, etc.
you can use in your SSH session:
To use the scripts, you will need to transfer them onto the
pwd – ask for the current directory
cluster as well. They are available on the course website. You
mkdir [dirname] – create directory dirname
will need to execute the command “chmod u+rwx *.sh” before
rm –rf [dirname] – delete directory dirname and anything in it
you can run the scripts. This will give the computer permission
rm [filename] – delete filename
to execute the programs (something of a safety feature).
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
To submit the job, do this: not a good place to put things.) Let’s go there now:
[ekwan@iliadaccess01 chem106]$ ./submit.sh 8 short_parallel [ekwan@iliadaccess01 chem106]$ pwd
/n/home11/ekwan/chem106
Beginning batch submission process... [ekwan@iliadaccess01 chem106]$ cd output
Each job is being submitted to 8 processors in the queue [ekwan@iliadaccess01 output]$ ls
short_parallel. output.txt 2-chloroTHP_ax.out

Submitting job file 2-chloroTHP_ax.gjf... As you can see, the output file is there. The script I wrote
Job <13724321> is submitted to queue <short_parallel> automatically extracts the energies. To save time, I ran the jobs
Job submission complete.
for both the axial and equatorial:
[ekwan@iliadaccess01 chem106]$ more output.txt
This will submit every .gjf file in the directory into the queue *************************************
short_parallel, requesting 8 cores for every job. (Caution! Job 2-chloroTHP_ax started at...
Tue Oct 19 12:01:27 EDT 2010 electronic energy
Double check your job file before you submit the job. Gaussian
will be very unhappy if you make any syntax errors, and your job Job finished at:
will crash or do something you don’t expect it to.) Tue Oct 19 12:04:11 EDT 2010
free energy
[ekwan@iliadaccess01 chem106]$ bjobs -w Job terminated normally...

13724321 ekwan RUN short_parallel iliadaccess Final energy:

8*hero1807 2-chloroTHP_ax Jan 28 14:59 SCF Done: E(RB3LYP) = -731.374863805 A.U. after 1 cycles

As you can see, the job is now running in short_parallel. You No imaginary frequencies found.
Sum of electronic and thermal Free Energies= -731.267453
will see its status as PEND if it’s waiting. If you suddenly realize *************************************
you made a mistake, use “bkill [jobid]” to cancel the job.
*************************************
Job 2-chloroTHP_eq started at... this is a true
Eventually, your job will run and complete and you will get a long Tue Oct 19 12:04:38 EDT 2010
email from LSF full of gibberish. I like these notifications, but local minimum
filter them away from my inbox. At that point, an output file (in Job finished at:
Tue Oct 19 12:06:40 EDT 2010
this case, 2-chloroTHP_ax.out) will be copied to the output
directory. Note that you will run into a problem if there is no Job terminated normally...
output directory present; use “mkdir output” to make such a
Final energy:
directory if it does not already exist. You only have to do this SCF Done: E(RB3LYP) = -731.368902477 A.U. after 1 cycles
once per directory containing the various scripts.
No imaginary frequencies found.
Sum of electronic and thermal Free Energies= -731.262071
In this case, my output directory is ~/chem106/output. ~ *************************************
represents your home directory or root. (The actual root, /, is
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
For reasons that are not clear to me, all the energies are
reported in hartree. To find the energy difference in kcal/mol,
you will need to subtract the two numbers and multiply by
627.509469. In this case, one finds that the axial conformer is
more stable by 3.4 kcal/mol.

You can also have a look at the optimized geometries in

GaussView. To open the file, make sure you select “Gaussian
Output Files (*.out *.log)” and “Read Intermediate Geometries
(Gaussian Optimizations Only).”

As before, there aren’t any negative numbers in the frequency

column, so this is a real local minimum. This information is
also captured by the output.txt file when it says “No imaginary
frequencies found.”

Notice that there’s no bond drawn between the carbon and the
chlorine. This is because Gaussian doesn’t know anything
To see all the intermediate geometries, you can click through about where the bonds are. GaussView simply draws in bonds
the circled box. As you can see, this optimization went when the distances between atoms are within certain
smoothly, converging nicely to a stationary point. You can also thresholds. The C-Cl bond is abnormally long due to the
check to make sure that no imaginary frequencies were found anomeric effect, so it is not recognized as a bond. However, the
by going Results…Vibrations: lack of a line connecting the C and Cl has no significance.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
Finally, we want to find a transition state for the rotational degrees. If you scan too coarsely, you can have “kinking”
barrier in ethane. As I mentioned already, the most robust way problems where the energy profile you get is not smooth. Here
to find transition states is to perform a series of constrained is my input file:
optimization where the reacting coordinate is frozen in %chk=checkpoint.chk
%mem=3GB
increments. For many reactions, this is a forming bond %nprocshared=8
distance. In this case, it is a bond torsion. Draw ethane and #t opt=modredundant b3lyp/6-31g(d) pop=none
use the question mark tool to identify the atom numbers of a H-
title
C-C-H torsion:
0 1
C -1.27672721 1.36139405 0.37637410
H -0.91105109 1.81418310 -0.52149173
H -2.00301536 0.61545024 0.12944624
H -1.72899504 2.10885135 0.99416159
C -0.10678799 0.70753803 1.13486901
H 0.34547984 -0.03991928 0.51708152
H 0.61950015 1.45348183 1.38179688
H -0.47246412 0.25474897 2.03273484

D 8 5 1 2 S 18 10.000000
This results in a nice energy profile (Results…Scan):

Now, we need to tell Gaussian to perform a constrained

optimization. In the world of Gaussian, this is called a
“modredundant optimization.” The notation

D 8 5 1 2 S 18 10.000000

means to scan the dihedral (D; use B for bond and A for angle)
angle between atoms 8-5-1-2 eighteen times in steps of 10
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
However, we are not done yet. Just because we have found a Take note of the sub-keywords under the main opt keyword.
maximum in the energy profile is no guarantee it is a true
transition state. For that, we need the gradient to be zero (or at ts: Request optimization to a first-order saddle point (i.e., a
least below the convergence threshold) and there to be exactly transition state)
one imaginary frequency. To do this, save the geometry
corresponding to an energy maximum as a new .gjf file: calcfc: Perform a frequency job at the beginning of the
calculation (i.e., calculate the Hessian). An accurate Hessian is
required for any TS search. Occasionally, you can read the
frequencies in from the checkpoint file; in that case, you should
use the readfc sub-keyword. So you must either have calcfc or
readfc for any transition state search.

noeigentest: Suppress curvature testing. You should turn this

on to avoid crashes. (The manual recommends curvature
testing in most cases, but this seems to be an anachronism.)

We also need freq=noraman, which will perform a second

frequency calculation at the end of the job to verify that there is,
in fact, only one negative frequency.

Let me re-iterate that:

(1) You will only find the transition state if you are very, very
Notice that the energy maximum, is, in fact, an eclipsed close to it already.
conformation. This is a good sign. We will need a new header
to request an unconstrained transition state search: (2) Even if you’re very close to the correct geometry, transition
%chk=checkpoint.chk state optimizations often fail.
%mem=3GB
%nprocshared=8
#t opt=(calcfc,ts,noeigentest) freq=noraman b3lyp/6-31g(d) (3) Check your syntax! Check your syntax!
pop=nbo
Gaussian also has the ability to try and connect a starting
title
material and product with a transition state directly; see the
0 1 QST2 and QST3 methods under the “opt” keyword. I have
never used these, so I can’t offer any advice on them.
E. Kwan Lecture 9: Introduction to Computational Chemistry Chem 117
In my run, the transition state optimization converged nicely: state along the reaction coordinate. In general, if your
optimization does not converge to a transition state within 30
steps or so, you are in trouble.

Ambiguous:
Sometimes, the transition state energy will go down, but so
will the gradient. This may result in a successful optimization.
However, this often later results in…

Bad Signs:
The transition states “collapses” to starting material or product.
In the case below, it opened up to starting material (use
Plots…Molecular Property to get the third plot):

You should check on your optimizations as they progress to

make sure they’re not going “off the reservation.” That just
wastes valuable computer time. (You can just download the
output file from the directory the job is running in, and ignore
the error GaussView gives you when it opens the file.)

Good Signs:

In an ideal world, every TS optimization will look like this one:

the geometry starts off correct, and is minimized to a
stationary point of correct curvature in one step. This gives
rise to the flat energy optimization curve shown above. In
other cases, the energy will go up and settle down into a
This concludes the tutorial. You should work through these
transition state, while the RMS gradient will drop down and go
examples yourself. Let me close by saying: computations
to zero as the stationary point is reached. That is good
are hard work! Don’t think that it’s easy just because you’re
news—the optimization algorithm is approaching the transition
sitting at your desk…