Integrate-and-fire neuron

The integrate-and-fire neuron is the simplest mathematical model of a biological neuron — a reduction that discards all biophysical detail and retains only the essential dynamical mechanism: slow accumulation of input until a threshold is reached, followed by an instantaneous reset. It is the neuroscience equivalent of a relaxation oscillator, and it demonstrates that neural computation does not require Hodgkin-Huxley complexity. The threshold-and-reset logic is sufficient.

The Model

The canonical integrate-and-fire model is a single differential equation:

τ · dV/dt = −(V − V_rest) + R · I(t)

where V is the membrane potential, V_rest is the resting potential, τ is the membrane time constant, R is the input resistance, and I(t) is the input current. The equation describes a leaky integrator: the neuron sums its inputs over time, but the sum decays exponentially with time constant τ. When V reaches a threshold V_th, the neuron "fires" — an action potential is recorded, and V is instantly reset to V_reset.

The key feature is the threshold. For subthreshold inputs, the neuron acts as a passive filter. For suprathreshold inputs, it fires periodically. The firing rate is a nonlinear function of input current: zero below threshold, then increasing with input strength above threshold. This nonlinearity — the all-or-none response — is the computational primitive from which neural networks build their functions.

From Biology to Computation

The integrate-and-fire model is a caricature. It omits the shape of the action potential, the refractory period, the dendritic computation, the synaptic dynamics, and the ion channel kinetics. But it captures the essential input-output relationship: integrate until threshold, then fire and reset. This relationship is the basis of neural coding. Neurons do not transmit analog values. They transmit spikes — discrete events whose timing carries information. The integrate-and-fire model is the minimal system that produces this discrete output from continuous input.

The model belongs to the class of relaxation oscillators: a slow phase (integration) followed by a fast phase (firing and reset). In the limit where the reset is instantaneous and the integration is slow, the dynamics maps onto the slow-fast geometry of the FitzHugh-Nagumo model. The two models are cousins: FitzHugh-Nagumo adds a recovery variable that produces a realistic action potential shape, while integrate-and-fire abstracts the shape away entirely and keeps only the threshold crossing.

The integrate-and-fire neuron is not a simplification of biology. It is a simplification of computation. It asks: what is the minimum dynamical system that can transform a continuous input into a discrete, timed output? The answer — a leaky integrator with a threshold — is so simple that it is easy to miss how profound it is. This is the mechanism that underlies every thought you have ever had.