JOVANA
Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
All guides

Analytic Functions & Cauchy–Riemann

When you let x become a complex number z = x + iy, asking for a derivative suddenly demands enormous rigidity — and out of that one demand fall the Cauchy–Riemann equations and the harmonic functions that make the whole complex plane sing.

A derivative, but in two directions at once

Back in Volume I a derivative was a limit of a difference quotient: f'(x) is the limit of (f(x+h) - f(x))/h as the step h shrinks to zero. On the real line there are only two ways to approach a point — from the left or from the right — and we insisted both give the same slope. Now we play the same game with a complex variable z = x + iy, where f(z) is itself a complex number. The definition looks identical: f'(z) is the limit of (f(z+h) - f(z))/h as h goes to zero. The twist is that h is now a complex number, so it can approach zero from infinitely many directions in the plane — along the real axis, along the imaginary axis, spiralling in at any angle.

For the limit to exist as a single complex number, every one of those infinitely many directions must hand back the same answer. That is a staggeringly strong requirement. On the real line, being differentiable is a mild condition — most functions you meet are. In the complex plane, demanding one consistent derivative in all directions is so restrictive that the functions which pass are a rare and beautifully behaved elite. We call such a function analytic (or holomorphic) at a point when the complex derivative exists there and throughout a small disk around it.

Two directions, one equation: deriving Cauchy–Riemann

Let us cash out "the same answer from every direction" into a usable rule. Split the function into its real and imaginary parts: write f(z) = u(x, y) + i v(x, y), where u and v are ordinary real-valued functions of the two real coordinates x and y. The complex derivative, if it exists, must come out the same whether we let the step h run along the real axis or along the imaginary axis. Comparing just those two approach directions is enough to pin everything down.

Approach along the real axis (take h = a tiny real number). Then the difference quotient only changes x, so the derivative reads as a partial derivative in x: f'(z) = du/dx + i dv/dx. Now approach along the imaginary axis instead (take h = i times a tiny real number). Dividing by that imaginary step rotates everything by a factor of 1/i = -i, and the derivative now reads f'(z) = dv/dy - i du/dy. These two expressions for the very same number f'(z) must agree. Matching real parts and imaginary parts separately gives the two famous equations.

Approach along real axis:        f' = u_x + i v_x
Approach along imaginary axis:   f' = v_y - i u_y

The two must be equal, so:
      u_x = v_y          (real parts match)
      u_y = -v_x         (imaginary parts match)

   <-- these are the Cauchy-Riemann equations -->
( subscripts mean partial derivatives: u_x = du/dx )
Forcing the real-axis and imaginary-axis derivatives to agree yields the Cauchy–Riemann equations.

What the equations are really saying

So the Cauchy–Riemann equations are du/dx = dv/dy and du/dy = -dv/dx. They are not arbitrary bookkeeping — they are the precise price of admission for being analytic. Read geometrically, they say the real and imaginary parts are locked together as two faces of a single rigid motion. At each point an analytic map acts, to first order, like a rotation combined with a uniform scaling — it can stretch and spin a tiny neighbourhood, but it can never squash a circle into an ellipse or flip orientation. That rigidity is exactly why analytic maps are conformal: wherever the derivative is nonzero, they preserve angles between curves.

Let us test the equations on the friendliest example, the complex exponential f(z) = e^z. Euler's formula expands it as e^z = e^x (cos y + i sin y), so u = e^x cos y and v = e^x sin y. Differentiate: du/dx = e^x cos y and dv/dy = e^x cos y — they match. And du/dy = -e^x sin y while -dv/dx = -e^x sin y — they match too. Both Cauchy–Riemann equations hold everywhere, so e^z is analytic on the whole plane, and indeed its derivative is e^z again, just as you would hope.

Harmonic functions fall out for free

Now comes the payoff that makes complex analysis indispensable in physics. Suppose f = u + iv is analytic, and (a fact we will take on faith here, and prove later) that u and v are then automatically smooth enough to have second partials. Differentiate the first Cauchy–Riemann equation du/dx = dv/dy with respect to x, and the second du/dy = -dv/dx with respect to y. The mixed partials of v are equal — order of differentiation does not matter for smooth functions, a Volume I fact about partial derivatives — so when you add the two results, the v-terms cancel exactly and you are left with d^2u/dx^2 + d^2u/dy^2 = 0.

That equation, d^2u/dx^2 + d^2u/dy^2 = 0, written compactly as nabla^2 u = 0, is Laplace's equation, and any function satisfying it is called harmonic. Run the same argument the other way and you find v is harmonic too. So the real and imaginary parts of every analytic function are a matched pair of harmonic functions — and harmonic functions are exactly the steady-state temperature distributions, the electrostatic potentials, the incompressible ideal flows of physics. This is the deep reason engineers reach for the complex plane: every analytic function quietly carries two solutions of Laplace's equation inside it.

The pairing is not symmetric in a subtle and useful way: given a harmonic u, the Cauchy–Riemann equations tell you the gradient of v (you know dv/dx and dv/dy from u), so you can integrate to recover v up to a constant. The v built this way is called the harmonic conjugate of u, and u + iv is then analytic. There is a small honest caveat — this reconstruction is guaranteed on a simply connected region (one with no holes); on a domain with a hole, like an annulus, a harmonic conjugate may fail to exist globally. The logarithm's potential is the classic place this subtlety bites.

Using the criterion, and where it stops

Here is the practical checklist that turns all of this into a tool you can actually use on a function f(z).

  1. Write f(z) = u(x, y) + i v(x, y) by separating the real and imaginary parts explicitly, using z = x + iy and Euler's formula wherever an exponential or trig appears.
  2. Compute all four first partials: u_x, u_y, v_x, v_y.
  3. Check the two Cauchy–Riemann equations u_x = v_y and u_y = -v_x. Wherever both hold (and the partials are continuous), f is analytic; wherever either fails, f is not.
  4. Where it is analytic, read off the derivative for free from f'(z) = u_x + i v_x, and rest assured u and v each solve Laplace's equation.

Two honest boundaries before you go. First, the equations alone are necessary but not quite sufficient: a function can satisfy Cauchy–Riemann at an isolated point yet still fail to be differentiable there. The clean theorem is that if u and v have continuous first partials and satisfy the equations on an open set, then f is analytic on that set — continuity of the partials is the hypothesis that closes the gap. Second, analyticity is local: most interesting functions are analytic on most of the plane but not all of it. The complex logarithm needs a branch cut and 1/z blows up at the origin — these failures of analyticity, called singularities, are not bugs but the very features the contour integration and residue machinery of the next guides will exploit.