Turning One Neuron Into Math

The membrane is already a circuit

You already know the physical neuron: a thin, oily membrane separating salty water inside from salty water outside, with tiny gates — ion channels — punched through it. To turn that picture into math, notice that this setup is literally an electrical circuit. The membrane holds charge apart, so it acts like a tiny battery and a tiny capacitor. Each channel lets charge leak through, so it acts like a resistor. Math does not replace the neuron here — it just relabels the parts you can already see.

  outside  ───────────────────────
            │      │      │        
          [Na+]  [K+]  [leak]   gates = resistors
            │      │      │        
  ══════════╪══════╪══════╪═══════  membrane = capacitor
            │      │      │        
  inside   ───────────────────────
         charge piles up here → voltage V

Hodgkin and Huxley: the full portrait

In the 1950s, Alan Hodgkin and Andrew Huxley measured a giant squid axon and wrote the first equation for the action potential. Their core rule is just bookkeeping for charge: how fast the voltage changes equals the total current flowing in and out. The currents are simply the gates from the picture above — a sodium current, a potassium current, and a small leak — each one a voltage-gated channel whose openness depends on the voltage itself.

  C · dV/dt  =  I_input  −  I_Na  −  I_K  −  I_leak
  └─ how fast    └─ what    └──── the gates from
     V changes      you         the picture, each
                     inject       opening with voltage

The magic is the feedback loop. Push the voltage up a little and sodium gates fly open, letting more positive charge rush in, which pushes the voltage up further, which opens more gates — an explosion. Then slower potassium gates open and yank the voltage back down. That self-igniting, self-quenching loop is the spike. The Hodgkin-Huxley model won a Nobel Prize because it predicted the exact shape of a real action potential from gate behavior alone.

Integrate-and-fire: the leaky bucket

The full portrait is gorgeous but heavy — four coupled equations per neuron, too slow to simulate a million of them. So modelers asked a blunt question: if all we care about is *when* a neuron fires, do we need the whole spike shape? The integrate-and-fire model says no. Picture the membrane as a leaky bucket. Input current pours water in; the leak drains some out. The water level is the voltage.

1. Integrate: add up incoming current, so the voltage (water level) climbs.
2. Leak: a little charge always drains away, so with no input the level slides back toward rest.
3. Fire: the moment the level hits the threshold, stamp down a spike, dump the bucket back to empty, and start again.

Cable theory: down the wire

Both models so far treat the neuron as a single point. But a real signal has to travel — from the dendrites, across the cell body, and down the long axon. How far does a nudge of voltage spread before it fades? That is exactly the question engineers asked about undersea telegraph cables a century earlier, so neuroscientists borrowed their answer: cable theory.

The idea is a tug-of-war along the cable. Charge wants to flow lengthwise down the inside (easy if the fiber is fat), but at every step some leaks out sideways through the membrane (faster if the membrane is leaky). The balance gives a single number, the length constant — the distance over which a passive signal shrinks to about a third of its size. Fatter, better-insulated fibers reach farther, which is exactly why myelin wrapping lets signals jump cleanly between gaps.

  inject here
      │
      v
  ════█═══════════════════════════ axon
      ██▓▓▒▒░░ ........              
      └─ voltage fades with distance →
         shrinks to ~1/3 over one
         "length constant" λ

From one neuron to the frontier

These three equations are the launchpad for all of computational neuroscience. Stitch millions of integrate-and-fire units together and you get a spiking neural network that can be modeled — or built directly in silicon, the goal of neuromorphic computing. Strip the spike away entirely and keep only 'weighted inputs cross a threshold,' and you have the artificial neuron behind today's brain-inspired AI. The same leaky-bucket logic, in disguise, now runs the chatbots in your pocket.