Skip to content

How to Read the Formulas

This page does not teach new physics. Its only job is to help you read the equations that already appear elsewhere in this handbook. When an equation stops a chapter dead for you, come back here, then return to the chapter.

The core idea

An equation is compressed language. Someone wrote a sentence — sometimes a whole paragraph — and squeezed it into symbols to save space. Reading an equation means doing the reverse: unpacking it slowly, one symbol at a time, back into ordinary words. You are not expected to "see" it all at once. Nobody does. Going slowly is the correct speed.

This is a slower, friendlier companion to the terse reference material; it overlaps on purpose so you have somewhere gentle to land.

First: what is the equation for?

Before reading any symbol, ask one question: what does this equation compute? Most equations have a left-hand side and a right-hand side, and usually one side is the thing you want and the other side is the recipe for getting it.

Inputs and outputs

In \(E = T + V\) (total energy = kinetic + potential), the output is \(E\) on the left; the inputs are \(T\) and \(V\) on the right. Reading direction: "to get \(E\), add \(T\) and \(V\)." Knowing which symbols are inputs and which is the output is half the battle.

A quick test: ask "if I changed this symbol, would the answer change?" If yes, it is an input. The symbol alone on one side is usually the output.

Reading subscripts and superscripts

Small marks above and below a symbol are not decoration. They carry meaning, and the same-looking mark can mean different things.

  • Subscript as a label or index. \(n_i\) means "the \(n\) belonging to item \(i\)" — for example the occupation of orbital number \(i\). The \(i\) picks one out of many.
  • Subscript on a vector. \(\mathbf{r}_i\) is the position vector of atom \(i\). Bold says "vector"; the subscript says "which atom".
  • Subscript as a named level. \(\varepsilon_n\) is the energy of level \(n\) (the \(n\)-th eigenvalue). Here \(n\) counts states.
  • Superscript as a power. In \(n^{4/3}\) the superscript is an exponent: raise \(n\) to the power \(4/3\).
  • Superscript as an operation, not a power. \(\psi^*\) means the complex conjugate of \(\psi\), not "\(\psi\) squared". Context tells you which: a number on top is usually a power; a symbol like \(*\) or \(\dagger\) is an operation.

Warning

\(\psi^2\) and \(\psi^*\) look similar but are unrelated. When you meet a superscript, pause and decide: power, or operation?

Six kinds of object — and why it matters

Symbols stand for different kinds of thing. Telling them apart prevents most confusion. Here is each kind with one example drawn from this book.

  • Scalar — a single number (with units). Example: the total energy \(E\). One value.
  • Vector — a list of numbers with direction, written in bold. Example: a position \(\mathbf{r} = (x, y, z)\). Three numbers describing one point in space.
  • Matrix — a grid of numbers, indexed by two subscripts. Example: the Hamiltonian matrix \(H_{ij}\), where \(i\) labels the row and \(j\) the column.
  • Function — a rule that takes an input and returns a number. Example: the electron density \(n(\mathbf{r})\) takes a position \(\mathbf{r}\) and returns "how much electron is here". Give it a point, get a number.
  • Operator — a rule that takes a function and returns another function. Example: the Hamiltonian operator \(\hat{H}\) (the hat is the giveaway) acts on a wavefunction and hands you back a new function.
  • Functional — a rule that takes a whole function and returns a single number. Example: the energy functional \(E[n]\) takes the entire density function \(n(\mathbf{r})\) and returns one energy value. Note the square brackets \([\,\cdot\,]\) — a strong hint that you are looking at a functional.

Function vs functional — read this twice

A function like \(n(\mathbf{r})\) eats one input (a point) and gives one number. A functional like \(E[n]\) eats an entire function (the whole density, everywhere at once) and gives one number. The clue is the brackets: round \((\,)\) for a function, square \([\,]\) for a functional. This distinction is the whole reason "density functional theory" is named the way it is: the energy is a functional of the density function. Get this and the name stops being mysterious.

Tracking units

Every physical quantity carries units, and units obey the same algebra as the symbols. This gives you a free error-checker.

The rule: both sides of an equation must have the same units, and you can only add quantities with the same units. You cannot add an energy to a length.

Units as a safety net

Suppose you derive an "energy" and the units come out as Å (a length). You have made a mistake somewhere — no calculator needed to know that. Conversely, if both sides come out in eV, that is a good sign (though not a proof) that the algebra is right. In this book energies live in eV or hartree (\(1\,\mathrm{Ha} = 27.2114\) eV) and lengths in ångström or bohr (\(1\,a_0 = 0.529177\) Å); keeping track of which is which prevents silent errors.

Reading sums and integrals as "add up contributions"

A summation and an integral say the same thing in plain words: add up many small contributions.

  • \(\displaystyle\sum_i x_i\) reads as "add up \(x_i\) for every \(i\)" — first \(x_1\), then \(x_2\), and so on, totalling them. The \(i\) under the \(\sum\) is the counter.
  • \(\displaystyle\int f(\mathbf{r})\,\mathrm{d}\mathbf{r}\) reads as "add up \(f\) over all of space". An integral is a sum where the pieces are infinitely small. Think of chopping space into tiny boxes, evaluating \(f\) in each box, and totalling — the \(\mathrm{d}\mathbf{r}\) is the size of one tiny box.

If you can read \(\sum\) and \(\int\) both as "total it up", a large fraction of intimidating equations becomes approachable.

Reading derivatives and gradients as rates of change

Derivatives measure how fast something changes.

  • \(\dfrac{\mathrm{d}f}{\mathrm{d}x}\) is the slope: how much \(f\) changes when \(x\) changes a little. Positive means rising, negative means falling, zero means flat (a peak, valley, or plateau).
  • \(\nabla f\) (the gradient) is the slope generalised to several directions at once. It is a vector pointing "uphill" — the direction in which \(f\) increases fastest. A force is often \(-\nabla V\): push downhill in energy.
  • \(\nabla^2 f\) (the Laplacian) measures curvature — how much \(f\) bends. It appears in the kinetic-energy term of quantum mechanics.

A simple mantra: first derivative = slope, second derivative = curvature.

Reading an eigenvalue equation

The pattern \(\hat{A}\psi = a\psi\) appears constantly. In words:

"When the operator \(\hat{A}\) acts on the special function \(\psi\), the result is just \(\psi\) again, scaled by the number \(a\)."

The function \(\psi\) is called an eigenfunction and the number \(a\) its eigenvalue. The key idea is that most functions get reshaped by an operator, but eigenfunctions are special: they come back unchanged in shape, only stretched by a factor \(a\).

Reading probability expressions

In quantum mechanics, \(|\psi(\mathbf{r})|^2\) reads as a probability density: "how likely you are to find the particle near point \(\mathbf{r}\)". It is a density, so to get an actual probability you integrate it over a region — adding up contributions, exactly as above. The conjugate appears because \(|\psi|^2 = \psi^*\psi\), which is always real and non-negative, as a probability must be.

Reading a machine-learning loss

A loss function measures how wrong a model is, as a single number, so smaller is better. The mean-squared error,

\[ L = \frac{1}{N}\sum_{i=1}^{N}\bigl(y_i - \hat{y}_i\bigr)^2, \]

reads as: "for each of the \(N\) examples, take the gap between the true value \(y_i\) and the prediction \(\hat{y}_i\), square it (so over- and under-shooting both count as a penalty), then average the penalties." Squaring punishes large mistakes more than small ones. Training a model means adjusting it to make this average penalty as small as possible.

Four worked "read-this-equation" examples

For each, read the equation left to right, naming every symbol, then say what the whole thing means.

1. A simple energy expression

\[ E = T + V \]
  1. \(E\) — a scalar, the total energy (output), in eV or hartree.
  2. \(=\) — "is equal to / is computed as".
  3. \(T\) — the kinetic energy (energy of motion), an input.
  4. \(+\) — ordinary addition; allowed because \(T\) and \(V\) share units.
  5. \(V\) — the potential energy (energy of position/interaction), an input.

In words: "The total energy is the kinetic energy plus the potential energy." A closely related density-functional form is the LDA exchange energy, \(E_x \propto \int n(\mathbf{r})^{4/3}\,\mathrm{d}\mathbf{r}\), which reads "add up \(n^{4/3}\) over all space" — a single number (a functional of the density), built by integrating a function of \(n\) at every point.

2. A Schrödinger-style eigenvalue equation

\[ \hat{H}\psi = E\psi \]
  1. \(\hat{H}\) — an operator (the hat says so); the Hamiltonian, encoding total energy.
  2. \(\psi\) — a function, the wavefunction; the eigenfunction we are looking for.
  3. \(\hat{H}\psi\) — "let \(\hat{H}\) act on \(\psi\)", producing a new function.
  4. \(=\) — the two sides are the same function.
  5. \(E\) — a scalar eigenvalue, the energy of this state.
  6. \(E\psi\) — the original \(\psi\) simply scaled by the number \(E\).

In words: "Acting on \(\psi\) with the energy operator gives back the same \(\psi\), scaled by the energy \(E\)." The special functions \(\psi\) that satisfy this are the allowed states, and the matching \(E\) values are their allowed energies.

3. A self-consistent (fixed-point) idea

\[ n_{\text{out}} = F[n_{\text{in}}], \qquad \text{converged when } n_{\text{out}} \approx n_{\text{in}} \]
  1. \(n_{\text{in}}\) — a function, the density you guess going in.
  2. \(F[\,\cdot\,]\) — a functional / procedure (square brackets): "do the whole calculation on this density". Here it stands for one full pass of the machinery.
  3. \(n_{\text{out}} = F[n_{\text{in}}]\) — feed the guess in, run the procedure, get an updated density out.
  4. \(n_{\text{out}} \approx n_{\text{in}}\) — the stopping rule: the density you get out matches the density you put in.

In words: "Put a density in, the procedure gives an updated density; repeat, feeding each output back as the next input, until output and input agree." When they agree, the density is self-consistent — it reproduces itself. This is exactly the self-consistent field (SCF) loop used in Kohn–Sham DFT (Section 5.5): guess a density, build the potential, solve, get a new density, and iterate to a fixed point. See the chapter overview, Chapter 5 (DFT), for the full picture.

4. A mean-squared-error ML loss

\[ L = \frac{1}{N}\sum_{i=1}^{N}\bigl(y_i - \hat{y}_i\bigr)^2 \]
  1. \(L\) — a scalar, the loss (output); smaller is better.
  2. \(\dfrac{1}{N}\) — divide by the number of examples \(N\): this makes it an average.
  3. \(\displaystyle\sum_{i=1}^{N}\) — "add up over every example \(i\) from \(1\) to \(N\)".
  4. \(y_i\) — the true value for example \(i\) (a label/index subscript).
  5. \(\hat{y}_i\) — the model's prediction for example \(i\) (the hat here means "estimate").
  6. \((y_i - \hat{y}_i)\) — the error: how far the prediction is from the truth.
  7. \((\,\cdots)^2\) — square it, so the penalty is always positive and big errors count extra.

In words: "For each example, square the gap between truth and prediction, then average those squared gaps." It is the average penalty for being wrong; training shrinks it.

When you meet a new equation, ask…

Run through this checklist, slowly, every time:

  1. What does it compute? Which side is the output, which side is the recipe?
  2. What kind of object is each symbol — scalar, vector, matrix, function, operator, or functional? (Watch the hats and the bracket shapes.)
  3. What do the sub/superscripts mean — an index, a label, a power, or an operation like \(*\)?
  4. Do the units match on both sides? If not, something is wrong.
  5. Is there a \(\sum\) or \(\int\)? If so, read it as "add up contributions".
  6. Is there a derivative or \(\nabla\)? Read it as a slope, a rate of change, or a curvature.
  7. Is it an eigenvalue equation \(\hat{A}\psi = a\psi\)? Then \(\psi\) comes back scaled by \(a\).
  8. Can I say the whole thing in one plain English sentence? If yes, you have read it.