The same limit, written for the plane
You already know the derivative as a limit of slopes: the rate at which output changes per unit of input, found by shrinking a small step to nothing. The complex derivative is written with the very same formula. We form the difference quotient (f(z_0 + h) - f(z_0)) / h, where now z_0 and the step h are complex numbers, and we let h shrink toward 0. If that ratio settles on a single complex number, we call it f'(z_0) and say f is differentiable at z_0. On paper, nothing has changed from calculus.
f'(z_0) = lim ( f(z_0 + h) - f(z_0) ) / h
h -> 0
here z_0 and h are COMPLEX,
so h -> 0 means h approaches 0 from
any direction in the plane.But hidden in those three innocent symbols, h -> 0, is the whole story. Recall from the previous rung that a complex limit only exists when approach from every direction agrees on one answer. Here h is a complex step, so h -> 0 means h can shrink to zero along the real axis, along the imaginary axis, along any slanted line, even spiralling in. For f'(z_0) to exist, the difference quotient must converge to the same complex number no matter which of those infinitely many roads h takes home.
Why one direction is not enough
This is exactly where complex differentiability parts ways with the real derivative and becomes a much stricter demand. On the real line a function could only be approached from left and right, two roads. In the plane there are infinitely many, and they must all agree — so directional independence is no longer a mild courtesy but a heavy constraint baked into the very definition.
Let's watch a function fail. Take f(z) = z-bar, the conjugate, which reflects the plane across the real axis. Its difference quotient is (z-bar_0 + h-bar - z-bar_0) / h = h-bar / h. Now send h in along the real axis, h = t with t real: then h-bar = t, so the ratio is t/t = 1. Now send h in along the imaginary axis, h = i t: then h-bar = -i t, so the ratio is (-i t)/(i t) = -1. Two directions, two different answers — so the limit does not exist, and z-bar is differentiable nowhere, even though as a real map it is as smooth and harmless as can be.
What the demand secretly forces
Why does insisting that all directions agree pin a function down so tightly? Compare just two roads. Let h come in horizontally, h = t real: the quotient tends to the rate of change in the x-direction, which works out to u_x + i v_x. Let h come in vertically, h = i s: dividing by i s, the quotient tends to (u_y + i v_y) / i = v_y - i u_y. The derivative exists only if these two are equal, and matching real parts to real parts and imaginary to imaginary gives u_x = v_y and u_y = -v_x.
Setting those two limits equal, real part to real part and imaginary part to imaginary part, gives the pair u_x = v_y and u_y = -v_x. Those two relations are the celebrated Cauchy-Riemann equations, and we have just seen they are a necessary condition for the complex derivative to exist. They are the heart of the next guide, so we will not mine them fully here; the point for now is that the seemingly modest request 'let every direction agree' instantly couples the real part u and the imaginary part v to each other. You can no longer choose u and v independently — fix one and the other is hugely constrained. That coupling is the source of every miracle to come.
Two faces of the same number: amplitwist
When the derivative does exist, it is a single complex number f'(z_0), and a complex number always carries a geometric meaning: multiplying by it rotates and scales. So a complex-differentiable map, looked at very closely near z_0, behaves like multiplication by f'(z_0) — it spins every tiny arrow leaving z_0 by the angle arg f'(z_0) and stretches it by the factor |f'(z_0)|, the same for all directions at once. This is exactly the rotation-and-scaling map you met as a z, now arising as the local picture of any differentiable f.
Tristan Needham gave this local action a memorable name: the amplitwist — an amplification (the stretch by |f'|) twinned with a twist (the turn by arg f'). The deep reason directional independence matters is now visible: the only way a tiny piece of the plane can be sent so that every direction is amplified and twisted by the same amount is precisely when one complex number f'(z_0) governs the whole neighbourhood. This is why we will be able to call such maps conformal — angle-preserving — wherever f'(z_0) is non-zero. We will draw the amplitwist in full in guide 4.
Holomorphic, and the reward for being so
A function that is complex-differentiable not merely at one point but at every point of an open region is called holomorphic there. The 'open region' part matters: holomorphy is about being differentiable in a whole little neighbourhood, with room to wiggle in all directions, never just on a curve or at an isolated point. (When this holds on the entire plane, the function earns the further title entire — polynomials and e^z are the headline examples.)
Now the payoff that makes the steep entry fee worth it. In real calculus, a function can be differentiable once and then have a jagged, non-differentiable derivative — one derivative buys you almost nothing about the next. In the complex world the opposite is true, spectacularly so: if f is holomorphic on a region then it is automatically differentiable infinitely many times there, and more, it equals its own convergent power series sum a_n (z - z_0)^n on any disk inside the region. One complex derivative quietly hands you all of them, plus a Taylor series, for free.
A reliable routine for testing a function
Putting it together, here is the workflow you will reach for whenever a new function lands on your desk and you want to know whether it is differentiable. It turns an abstract limit into concrete partial-derivative arithmetic — the same split into u and v that you practised on the previous rung, now doing real work.
- Write z = x + i y and expand f into its real and imaginary parts, f = u(x, y) + i v(x, y).
- Compute the four partial derivatives u_x, u_y, v_x, v_y.
- Check the Cauchy-Riemann equations u_x = v_y and u_y = -v_x, and note exactly where they hold.
- Confirm the four partials are continuous near those points — then f is complex-differentiable there, with f'(z) = u_x + i v_x.
Run this on f(z) = z^2, whose parts you found before to be u = x^2 - y^2 and v = 2 x y. Then u_x = 2x = v_y and u_y = -2y = -v_x: Cauchy-Riemann holds everywhere, the partials are polynomials and so continuous, and we read off f'(z) = u_x + i v_x = 2x + i (2y) = 2z. The familiar power rule survives, and z^2 is differentiable across the whole plane — entire, in fact. Contrast z-bar from earlier: there u = x, v = -y, giving u_x = 1 but v_y = -1, so Cauchy-Riemann fails at every point, confirming z-bar is holomorphic nowhere. With this routine in hand, you are ready to meet the Cauchy-Riemann equations head-on in the next guide.