7
Part Zero · Foundations

Complex numbers and complex analysis

No ruler has ever measured a length of −1\sqrt{-1} metres, and no scale has ever read ii kilograms. Yet the central equation of quantum mechanics has an ii written into it, and every wavefunction is a complex number at every point. Why would nature need a number whose square is negative?

The short answer is that ii is not really a number for counting. It is a number for turning. Once you see that, the strangeness drains away, and you get a tool that makes waves, circuits and integrals far easier.

Multiplying by i is a quarter turn

Draw the ordinary numbers along a line. Multiplying by −1 flips a number to the other side: a half turn. So what multiplied twice gives a half turn? A quarter turn. Put a second axis at right angles and call one step along it ii: then i×ii \times i is two quarter turns, which is −1-1.

Every complex number z=x+iyz = x + iy is a point in this plane. Multiplying two complex numbers multiplies their lengths and adds their angles. Turn the arrow to an angle of 40 ° and watch its shadows on the two axes.

Euler’s formula

The arrow of length one at angle θ\theta has shadows cos⁡θ\cos\theta and sin⁡θ\sin\theta. Leonhard Euler found the compact way to write it:

eiθ=cos⁡θ+isin⁡θe^{i\theta} = \cos\theta + i\sin\theta

Put θ=π\theta = \pi and you get eiπ=−1e^{i\pi} = -1: half a turn. Let θ=ωt\theta = \omega t grow with time and the arrow spins steadily while its shadow traces a cosine wave. That spinning arrow is a phasor. Adding two waves of the same frequency becomes adding two arrows, which is far easier than adding two trigonometric functions. Electrical engineers design every circuit this way.

DeeperCompare Taylor series: ex=∑xn/n!e^{x} = \sum x^n/n!. Put x=iθx = i\theta; the even powers carry i2k=(−1)ki^{2k} = (-1)^k and give the series for cos⁡θ\cos\theta, the odd powers give isin⁡θi\sin\theta.

Functions that bend the plane

A complex function moves every point of the plane somewhere else. The plate draws a phase portrait: thin amber lines where the output points in the same direction, and pale rings where it has the same size. Look at , and .

Where the phase lines all crowd into a point, the function is zero or infinite. Lines meeting and running the same way round mark a zero; the opposite way round, a pole, where the function blows up.

Functions that can be differentiated in the complex sense are called analytic, and they are astonishingly rigid: the real and imaginary parts must satisfy the Cauchy–Riemann equations ∂u/∂x=∂v/∂y\partial u/\partial x = \partial v/\partial y and ∂u/∂y=−∂v/∂x\partial u/\partial y = -\partial v/\partial x. Know an analytic function along a tiny curve and you know it everywhere.

Integrals that only see the poles

Integrate an analytic function around a closed loop and you get zero, unless the loop encloses a pole. Then you get 2πi2\pi i times a number called the residue, which measures how strongly the function blows up there:

∮f(z) dz=2πi∑residues inside\oint f(z)\,dz = 2\pi i \sum \text{residues inside}

The function 1/(1+z2)1/(1+z^2) has poles at +i+i and −i-i. Move the loop’s centre to height 0.0. The integral is 0: zero when the loop holds no pole, π\pi when it holds the pole at +i+i, and −π-\pi when it holds the one at −i-i. Nothing else about the loop matters.

A real integral, done the easy way

What is ∫−∞∞dx/(1+x2)\int_{-\infty}^{\infty} dx/(1+x^2)? Close the real line with a huge half-circle in the upper half of the plane. The half-circle contributes nothing as it grows, and the loop holds only the pole at +i+i, whose residue is 1/(2i)1/(2i). So the answer is 2πi⋅1/(2i)=π2\pi i \cdot 1/(2i) = \pi.

You can check it the ordinary way, with arctan⁡\arctan, but contour integration does the same job for integrals that have no ordinary antiderivative. It will compute propagators in quantum field theory for us later.

Going round twice

Some functions cannot make up their minds. Take the square root. Start at z=1z = 1, where z=1\sqrt z = 1, and , following the root continuously. You arrive back at z=1z = 1 with z=−1\sqrt z = -1. Go round again to get back to +1+1.

To keep such a function single-valued we draw a branch cut, a line we agree not to cross. The cut is a bookkeeping choice; the doubling is real, and it will reappear when an electron needs a full 720° turn to return to itself.

Does quantum mechanics really need i?

In 2021 Marc-Olivier Renou and colleagues proposed an experiment, with three parties and two independent sources of entangled pairs, whose results no theory using only real numbers could reproduce, if systems are combined in the standard way. Experiments in 2022 agreed with the complex version. Established

In 2025 Timothée Hoffreumon and Mischa Woods argued that a real-number theory with a different rule for combining systems reproduces every prediction of ordinary quantum mechanics, so no experiment can tell them apart. Whether ii is a fact about nature or just the most convenient bookkeeping is now an open argument. Debated

Either way, the complex form is the natural language: the real versions still carry the fingerprints of complex arithmetic. This book will use ii freely.

What you now have

Complex numbers as arrows that multiply by turning, Euler’s formula, phasors, analytic functions, residues and branch cuts.

· · ·

Three questions to carry with you.

What does multiplying by i do geometrically?tap
It turns the arrow a quarter turn anticlockwise without changing its length.
Why is a loop integral of an analytic function zero unless it encloses a pole?tap
Cauchy’s theorem: analytic functions have no “sources” inside the loop; only the residues at poles contribute 2πi each.
What happens to √z as you walk once around zero?tap
It comes back with the opposite sign; it takes two turns to return to the start.

And one to work out. Using i=eiπ/2i = e^{i\pi/2}, the number iii^i is real. What is it?

iⁱ =

Sources: T. Needham, Visual Complex Analysis. M.-O. Renou et al., Nature 600, 625 (2021). T. Hoffreumon and M. P. Woods, arXiv:2504.02808 (2025). ∫dx/(1+x²) = π and iⁱ = e^(−π/2) checked by code.

Next · Chapter 8
Linear algebra I: vectors, matrices, eigenvalues

Every quantum measurement outcome is an eigenvalue.

Turn the page