Computational chemistry, and its one trade-off

One trade-off

Accuracy, size, time.Pick two

Every method in computational chemistry is a position on that triangle. Knowing where a method sits tells you what it can be asked.

Accuracy

How close to the real quantum answer. Improving it costs steeply — often as the fourth power of system size.

System size

Tens of atoms for high-level quantum work, millions for classical force fields. Four orders of magnitude apart.

Timescale

Picoseconds where electrons are explicit, microseconds where they are not. Also four orders of magnitude.

The ladder

From electronsto billiard balls

Four broad families, each about an order of magnitude cheaper and an order of magnitude less fundamental than the one above.

Wavefunction methodsHartree-Fock and beyond — MP2, coupled cluster. The most accurate, and the most brutally expensive.

Density functional theoryApproximate the electron density rather than the wavefunction. The workhorse for most systems.

Semi-empirical and QM/MMParameterised shortcuts, or quantum treatment of an active site inside a classical protein.

Classical force fieldsNo electrons at all. Millions of atoms and long trajectories, and no bond ever breaks.

The jump that matters most is between quantum and classical. Above it, electrons exist and bonds can break, so you can study a reaction — at the cost of being limited to hundreds of atoms and picoseconds. Below it, atoms are spheres joined by springs, so you get millions of atoms and microseconds, and no chemistry whatsoever.

What you can ask

Four questionswith settled methods

The field is large, but most projects reduce to one of these, and each has an established route.

What shape is it

Geometry optimisation — walk downhill on the potential energy surface until the forces vanish.

How fast does it react

Find the transition state, get the barrier, and a rate follows from transition state theory.

What spectrum does it give

Vibrational frequencies for IR and Raman; excited states for UV-visible. Directly comparable to experiment.

How tightly does it bind

Docking for a ranking, free energy methods for a number. The second is far more expensive and far more trusted.

Reading a method

The basis setis half the claim

A quantum result is meaningless without the method and basis set attached, and the notation is less forbidding than it looks.

B3LYPMethodthe functional, or the theory level

6-31G(d)Basis sethow flexibly orbitals are described

Or neitherBotha good method with a poor basis is not good

A label like B3LYP/6-31G(d) names the functional and the basis set — how the electron density is approximated, and how flexibly each orbital is described. Both matter, and a good functional with a small basis set is not a good calculation. When comparing published numbers, check that the level of theory matches before concluding anything from the difference.

Converged is not correctAn optimisation converging says the algorithm stopped, not that it found the structure you wanted.

Check the frequenciesA true minimum has no imaginary frequencies; a transition state has exactly one. Skipping this is common and wrong.

Cost is superlinearDoubling the atoms rarely doubles the time. Test on a small analogue before committing machine hours.

On a machine

GROMACS isalready built

Nothing here needs compiling. Pick a machine, and these two commands put it on there.

hub://gromacs

$ dxflow workflow create --identity gromacs hub://gromacs

$ dxflow workflow start gromacs

Start a machine What GROMACS needs

Read it, then run itPrepaid and by the hour, on a machine that is yours about a minute after you ask.

Start a machine All notes