The integrate-and-fire (IF) neuron model, first proposed by Louis Lapicque in 1907, is the simplest mathematical model that captures the essential logic of neural computation: a neuron integrates its synaptic inputs, and when the resulting membrane potential reaches a threshold, it fires an action potential and resets. Despite its simplicity, the IF model and its variants form the basis of most large-scale computational neuroscience models and provide the neural substrate for mathematical psychology models that bridge between cognitive and neural levels of description.
The Leaky Integrate-and-Fire Model
τₘ = membrane time constant (10–30 ms)
V_rest = resting potential (≈ −70 mV)
Rₘ = membrane resistance
I(t) = total synaptic input current
When V(t) ≥ V_th: fire spike, reset V → V_reset
Refractory period: V held at V_reset for τ_ref
The leaky IF (LIF) model adds passive membrane leak: in the absence of input, the membrane potential decays exponentially back to the resting potential with time constant τₘ. This leak means that the neuron acts as a temporal integrator with a finite memory — recent inputs contribute more to the membrane potential than older ones. The time constant τₘ sets the integration window: with a typical τₘ of 20 ms, the neuron effectively integrates inputs over about 20–40 ms. The LIF model is an RC circuit, making it amenable to exact analytical solutions for constant or piecewise-constant inputs.
Variants and Extensions
Several important variants extend the LIF model's realism. The quadratic IF (Latham, Richmond, Nelson, & Nirenberg, 2000) replaces the linear leak with a quadratic voltage dependence, producing a more realistic approach to threshold. The exponential IF (Fourcaud-Trocmé, Hansel, van Vreeswijk, & Brunel, 2003) adds an exponential term that models the rapid upswing of the action potential, closely matching the behavior of detailed Hodgkin-Huxley models while remaining analytically tractable. The adaptive exponential IF (AdEx) adds a slow adaptation variable that captures spike-frequency adaptation and bursting behavior.
In mathematical psychology, networks of IF neurons provide the neural substrate for accumulator models of decision making. Wang's (2002) model of perceptual decision making, for instance, is built from networks of LIF neurons with realistic synaptic dynamics. The firing-rate output of these neural populations approximates the evidence accumulation described by the drift-diffusion model. IF neurons are also used in models of working memory, attention, and learning, where the spiking dynamics add realistic temporal structure to the computation while remaining computationally efficient enough for large-scale simulation.
The IF model's main limitation is that it does not model the biophysics of spike generation — the action potential is simply declared when threshold is reached, without modeling the ionic currents that produce it. For this, the Hodgkin-Huxley model is needed. However, for the purpose of modeling network-level computation — where the key variables are firing rates, synaptic interactions, and population dynamics rather than the detailed waveform of individual spikes — the IF model provides an excellent balance of biological realism and mathematical tractability.