Computational Chemistry — Lecture Notes

1 · What is computational chemistry?

Computational chemistry uses the laws of physics, solved on a computer, to predict the structure, energy, properties and behaviour of molecules and materials — often before, or instead of, ever making them in a lab. Where a bench chemist mixes reagents in a flask, a computational chemist sets up the equations that describe how a molecule's electrons and nuclei interact, and lets a computer solve them. The output is numbers and pictures: the shape a molecule adopts, how much energy it holds, what colour it absorbs, how fast a reaction goes, how tightly a drug binds its target.

You will sometimes hear two related terms. Computational chemistry is the broad practice of computing chemical results. Theoretical / quantum chemistry is the underlying theory — the equations and approximations. Molecular modelling usually means the more visual, structure-and-shape side. In practice they overlap heavily, and MoleBench touches all of them.

2 · Why it matters

For centuries, science rested on two pillars: theory (ideas and equations) and experiment (measurements). Computation is now the firmly established third pillar. Here is why it changed chemistry:

• It sees the invisible. Individual electrons, the precise geometry of a transition state (the fleeting instant a bond breaks, lasting ~10⁻¹³ s), short-lived reactive intermediates — none can be photographed, but all can be computed.
• It is cheaper, faster and safer. You can screen millions of candidate molecules in silico, test dangerous or expensive reactions virtually, and explore conditions impossible in a lab (extreme pressure, isolated single molecules).
• It explains why. Experiment tells you a reaction works; computation can tell you the mechanism — which bonds break first, why one product forms over another, what the catalyst actually does.
• It lets you design, not just analyse. Given a target property, you can propose a new molecule, catalyst, drug or battery material from scratch and predict whether it will work.

The field's importance is written into the Nobel record: 1998 (Walter Kohn & John Pople, for density functional theory and computational methods), 2013 (Karplus, Levitt & Warshel, for multiscale models of complex chemical systems), and 2024 (for computational protein-structure prediction). When the prize committee keeps returning to your field, it has become foundational.

3 · The one big idea: energy landscapes

If you remember a single concept from this lecture, make it this one. Imagine taking a molecule and placing its atoms in every possible arrangement — stretching bonds, bending angles, twisting. To each arrangement, assign one number: its energy. That mapping — geometry → energy — is the potential energy surface (PES).

Picture it as a landscape of hills and valleys:

• Valleys (minima) are stable structures — the shapes a molecule actually adopts. The deepest valley is the most stable.
• Mountain passes (saddle points) between two valleys are transition states — the highest point a molecule must climb to get from one structure (or molecule) to another.
• The height of a valley relative to others tells you which form dominates; the height of a pass tells you how fast you can get over it — i.e. the reaction rate.

4 · Describing a molecule to a computer

Before you can compute anything, you must tell the computer what the molecule is. There are two complementary descriptions:

• Connectivity — which atoms are bonded to which. The compact text format SMILES does this: O is water, CCO is ethanol, c1ccccc1 is benzene, CC(=O)Nc1ccc(O)cc1 is paracetamol. A SMILES string is a molecule written as one line.
• 3D coordinates — the (x, y, z) position of every atom in space. This is what most quantum methods actually need, because energy depends on the precise geometry.

One approximation underlies almost everything and deserves a name: the Born–Oppenheimer approximation. Nuclei are about 1,800 times heavier than electrons, so they move far more slowly. We therefore freeze the nuclei in place, solve for the fast electrons swarming around them to get the energy, then move the nuclei a little and repeat. This separation is exactly what makes the "geometry → energy" potential energy surface a well-defined object. Nearly every method below works within it.

5 · The ladder of methods

There is no single "calculate this molecule" button, because there is no single right trade-off between accuracy and cost. Instead there is a ladder, from cheap-and-approximate at the bottom to exact-and-ruinously-expensive at the top. Choosing the right rung for the question is the core skill of the field.

Rung	Idea	Speed / use
Molecular mechanics (force fields: MMFF, AMBER, OPLS)	Atoms are balls, bonds are springs. No electrons at all.	Blazing fast — millions of atoms. Great for shapes, conformers, proteins. Cannot break bonds or describe electronics.
Semi-empirical (GFN2-xTB, PM3, AM1)	Real quantum mechanics, but with many integrals replaced by parameters fitted to data.	Fast; handles bonding and reasonable geometries/energies. MoleBench's everyday xTB engine.
Hartree–Fock (HF)	The first true "from first principles" (ab initio) method: each electron feels the average field of all the others.	Moderate. Misses "electron correlation" (electrons dodging each other), so energies are systematically off.
Density Functional Theory (DFT)	Work with the electron density instead of the full wavefunction; fold correlation into an "exchange–correlation functional" (B3LYP, PBE, ωB97X-D…).	The workhorse of modern chemistry — the best accuracy-for-cost for most problems. MoleBench's DFT.
Post-Hartree–Fock (MP2, CCSD, CCSD(T))	Systematically add electron correlation back on top of HF.	Accurate and expensive. CCSD(T) with a big basis is the "gold standard."
Multireference (CASSCF, MRCI)	For cases where a single electronic configuration isn't enough.	The hard problems: bond-breaking, excited states, many transition-metal systems.

6 · Basis sets

To do quantum chemistry you must represent each electron's orbital as a mathematical function. We build those functions from a fixed library of simpler functions — the basis set. A bigger, more flexible basis set lets the orbitals take more realistic shapes, giving more accurate results, at higher cost.

Basis	What it is
STO-3G	Minimal — the bare minimum of functions. Crude but fast; useful for a rough first look.
3-21G, 6-31G*	Split-valence — the everyday workhorses. The * adds polarization functions, letting orbitals distort around bonds (important for accuracy).
6-31+G*	The + adds diffuse functions — fat, far-reaching orbitals needed for anions, excited states and weak interactions.
def2-TZVP, cc-pVTZ	Large, "triple-zeta" sets approaching the basis-set limit. Accurate and costly.

7 · What you actually compute

Once a molecule and a method are chosen, a handful of standard calculations answer most questions:

• Single-point energy — the energy at one fixed geometry. The most basic result; the building block for everything else.
• Geometry optimization — starting from a guess, follow the slope of the PES downhill until you reach a minimum: the molecule's true, relaxed shape.
• Vibrational frequencies (the Hessian) — the curvature of the energy landscape around a minimum gives the molecule's natural vibrations. From these you get the IR spectrum, the zero-point energy, and full thermochemistry — enthalpy (H), entropy (S) and Gibbs free energy (G) at a temperature. Frequencies also verify a structure: all real (positive) frequencies means a genuine minimum; exactly one imaginary frequency means a transition state.
• Molecular properties — dipole moment, partial atomic charges, the HOMO–LUMO gap, polarizability, and more, read off the converged electronic structure.

8 · Spectroscopy from first principles

One of the most striking powers of computation is predicting what a laboratory instrument would measure — without the instrument. Each kind of spectroscopy probes a different response of the molecule, and each can be computed:

Infrared (IR) — molecular vibrations

Bonds behave like springs; each "normal mode" of vibration has a characteristic frequency set by the bond stiffness and the masses of the atoms. The Hessian (§7) gives those modes and frequencies directly, and how much each absorbs infrared light. The result is the familiar IR spectrum — and because each mode is a specific collective motion, you can animate it to literally watch the bond stretch or the ring breathe.

NMR — magnetic shielding

In an NMR magnet, the electrons around each nucleus circulate and generate a tiny opposing field, "shielding" the nucleus from the full applied field. Electron-rich environments shield more (low chemical shift, "upfield"); deshielded ones (next to electronegative atoms, in aromatic rings) appear "downfield." Computing that shielding for every nucleus (the GIAO method) predicts the ¹H and ¹³C spectrum. A faster empirical route adds up tabulated substituent effects for an instant estimate — the same trick the famous prediction tools use.

UV-Vis — electronic excitations & colour

When a molecule absorbs visible or ultraviolet light, an electron jumps from an occupied orbital to an empty one. Time-dependent DFT (TD-DFT) predicts the energies (wavelengths) and intensities of those jumps. This is, quite literally, the computational origin of colour: extend a molecule's conjugation and the HOMO–LUMO gap shrinks, the absorption red-shifts, and a colourless compound becomes yellow, orange, red.

9 · Seeing electrons: orbitals & ESP

Because computation works from the electronic structure, it can show you the electrons — the part of chemistry you can never directly see:

• Molecular orbitals (HOMO & LUMO). The orbitals are the "rooms" electrons occupy. The HOMO (highest occupied) holds the most available electrons — where the molecule donates them. The LUMO (lowest unoccupied) is where it accepts them. The gap between them governs reactivity, hardness and colour. Frontier-orbital thinking explains a huge fraction of organic reactivity.
• Electron density. The cloud of negative charge — the "real," fuzzy shape of a molecule, the thing X-ray crystallography actually measures.
• Electrostatic potential (ESP) map. Colour the molecule's surface by what a tiny positive test charge would feel: red = electron-rich (negative potential, where electrophiles attack), blue = electron-poor (positive potential, where nucleophiles attack). One glance at an ESP map tells you where a molecule is reactive and how it will dock into another.

10 · Reactions & transition states

A chemical reaction is a journey across the energy landscape from a reactant valley, over a pass, to a product valley. The summit of that pass — the transition state — is the crux of chemistry. It exists for only ~10⁻¹³ seconds and can never be bottled, yet it can be computed, and its height above the reactants is the activation barrier.

That single number is enormously powerful. Through the Arrhenius and Eyring equations, the barrier sets the reaction rate. Computing barriers lets you:

• Predict selectivity — if two products are possible, the one with the lower barrier usually wins (kinetic control).
• Understand and design catalysts — a catalyst works by lowering the barrier; computation shows exactly how, and lets you engineer a better one.
• Map mechanisms — trace a multi-step reaction intermediate by intermediate.

11 · Accuracy: how to trust a number

A computed number is only as good as the model behind it. Being a good computational chemist is largely about knowing how much to trust your own results.

• "Chemical accuracy" ≈ 1 kcal/mol. Because reaction rates depend exponentially on energy, you typically need energies good to about 1 kcal/mol to reliably predict outcomes. That is a demanding target.
• Errors have three sources: the method (how correlation is treated), the basis set (how flexible the orbitals are), and the model (gas phase vs. solvent, a single conformer vs. many, 0 K vs. room temperature).
• Trends beat absolutes. Comparing two similar molecules with the same method cancels much of the systematic error, so relative answers ("which is more stable?") are far more reliable than absolute ones.
• Calibration helps. Many practical predictions (pKa, NMR shifts) apply an empirical scaling fitted against experiment to correct a method's systematic bias — turning a consistent error into a useful number.
• Know the regime. Semi-empirical and empirical methods are for trends and teaching, not publication; transition metals, radicals and excited states need careful, higher-level treatment.

12 · Bigger & more real

Real chemistry happens in messy, crowded environments and over time. Several techniques bridge from a single gas-phase molecule to that reality:

• Solvent. An implicit model treats the solvent as a smooth polarizable continuum (fast, captures the bulk effect); an explicit model surrounds the solute with real solvent molecules (slower, captures specific hydrogen bonds).
• Molecular dynamics (MD). Instead of a single static structure, let every atom move under the forces, step by tiny step, to simulate real motion: flexibility, diffusion, folding, and how a drug actually wriggles into a binding site. Classically driven by force fields; increasingly by machine-learned potentials.
• QM/MM. Treat the chemically interesting part (say, an enzyme's active site) with quantum mechanics and the vast surroundings with cheap molecular mechanics. This multiscale idea — letting different parts of one system be modelled at different levels — is what the 2013 Nobel honoured.
• Biomolecules. Docking ranks how molecules fit together; free-energy methods estimate binding strength; and AI structure prediction now gives protein shapes that once took years of crystallography.

13 · The frontier: machine learning, AI & quantum computing

Computational chemistry is in the middle of its biggest shift since DFT. What it is becoming:

• Machine-learning potentials. Neural networks trained on millions of quantum calculations now reproduce quantum-level forces at force-field speed — promising to make accurate dynamics of large systems routine.
• Generative AI for molecules. Models that, given a desired property profile, invent new drug candidates or materials — inverting the usual workflow from "analyse what we have" to "design what we want."
• AI structure & property prediction. The success of protein-structure prediction (a 2024 Nobel) showed that learned models can predict chemistry and biology directly, sometimes faster and better than first-principles simulation.
• Quantum computers. Simulating molecules is a natural fit for quantum hardware — arguably its flagship application — with the long-term promise of solving electronic-structure problems that are intractable for any classical computer.

14 · Where it is used

This is not an academic curiosity — it underpins industries:

• Drug discovery & pharma. Virtual screening of huge libraries, protein–ligand docking, free-energy binding predictions, ADMET (absorption/toxicity) modelling, and lead optimization — cutting years and millions from a drug program.
• Materials & energy. Designing battery electrolytes and cathodes, solar-cell and OLED molecules, catalysts for cleaner industrial reactions, polymers, and porous frameworks (MOFs) for gas capture.
• Environment & sustainability. CO₂ capture materials, hydrogen storage, atmospheric and combustion reaction networks, green-chemistry catalyst design.
• Biochemistry. Enzyme mechanisms, protein dynamics, and molecular recognition.
• And beyond — astrochemistry (molecules in space), geochemistry, flavour and fragrance, forensics, and fundamental discovery.

15 · How to get started

You do not need a supercomputer or a PhD to begin — you need the right mental model and some hands-on play.

1. Internalise the three big ideas from this lecture: everything is governed by quantum mechanics (§1); everything is exploring an energy landscape (§3); and there is a ladder of methods trading cost for accuracy (§5).
2. Get hands-on immediately. In the MoleBench Studio — entirely in your browser, much of it free — build a molecule, optimize it, read its properties, predict its spectra, and look at its orbitals and ESP. Seeing the ideas move turns abstractions into intuition.
3. Follow a path. A good first sequence: Properties → xTB optimization → DFT energy → IR/NMR/UV-Vis spectra → orbitals & ESP → conformers → a reaction barrier. The guided Tutorial walks several of these, and the Labs turn them into structured exercises.
4. Go deeper. Strengthen a little linear algebra and introductory quantum mechanics; meet the production tools (Psi4, ORCA, NWChem, Gaussian); and read a standard text — Jensen's Introduction to Computational Chemistry or Cramer's Essentials of Computational Chemistry are excellent next steps.

Open the Studio →   Start the guided tutorial

16 · Glossary

Term	Plain meaning
Ab initio	"From first principles" — solving the quantum equations without empirical parameters (HF, MP2, CC).
Basis set	The library of functions used to build orbitals; bigger = more accurate & costly.
Born–Oppenheimer	Approximation that freezes the heavy nuclei while solving for the fast electrons.
DFT	Density Functional Theory — the workhorse method, working from electron density.
Electron correlation	Electrons avoiding each other; the part Hartree–Fock misses and DFT/post-HF recover.
ESP	Electrostatic potential — a surface map of where a molecule is electron-rich (red) vs electron-poor (blue).
Force field	A molecular-mechanics model: balls and springs, no electrons.
Hartree–Fock (HF)	The simplest ab initio method; each electron sees the average of the rest.
HOMO / LUMO	Highest occupied / lowest unoccupied molecular orbital — the frontier orbitals governing reactivity.
PES	Potential energy surface — the geometry → energy landscape of valleys and passes.
SMILES	A text notation that writes a molecule's connectivity as one line.
Transition state	The energy summit between reactants and products; its height is the activation barrier.
Zero-point energy	The residual vibrational energy a molecule keeps even at absolute zero.

These notes are an introduction, not a textbook — every section here is a doorway into a deeper subject. The fastest way to make the ideas stick is to open the Studio and try them on a molecule you care about.