While the Hodgkin-Huxley model captures the exact biophysical reality of squid axons, simulating millions of 4-variable non-linear differential equations across large neocortical networks is computationally prohibitive. Theoretical neuroscience solves this through Geometric Dimensionality Reduction. By projecting the high-dimensional dynamics onto lower-dimensional manifolds, models like the Leaky Integrate-and-Fire (LIF), the 2D FitzHugh-Nagumo system, and the Izhikevich model capture essential neural excitability and bifurcation classes at fraction of the computational cost.
1. The Leaky Integrate-and-Fire (LIF) Model
Originally formulated by Louis Lapicque in 1907, the Leaky Integrate-and-Fire neuron abstracts the action potential as a linear passive RC circuit combined with a discontinuous threshold reset. Below threshold, the subthreshold membrane voltage $V(t)$ obeys:
tau_m frac{dV}{dt} = -(V – V_{text{rest}}) + R_m I(t), quad tau_m equiv R_m C_m
When the voltage reaches a threshold $V_{text{th}}$, the neuron fires a formal spike $delta(t – t_{text{spike}})$, and the voltage is instantaneously reset to $V_{text{reset}}$ where it is held for an absolute refractory period $tau_{text{ref}}$:
text{if } V(t) ge V_{text{th}} implies V(t^+) = V_{text{reset}}
Under constant injected current $I_0 > (V_{text{th}} – V_{text{rest}})/R_m$, the analytic firing rate $f(I_0)$ is derived explicitly by integrating the membrane equation:
f(I_0) = left[ tau_{text{ref}} + tau_m ln left( frac{R_m I_0 – (V_{text{reset}} – V_{text{rest}})}{R_m I_0 – (V_{text{th}} – V_{text{rest}})} right) right]^{-1}
2. The FitzHugh-Nagumo 2D Phase Plane
In 1961, Richard FitzHugh observed that in the Hodgkin-Huxley equations, the sodium activation gate $m$ equilibrates almost instantaneously ($tau_m ll 1$), while sodium inactivation $h$ and potassium activation $n$ evolve on similar slow timescales ($h + n approx text{const}$).
By collapsing $(V, m)$ into a fast voltage variable $v$ and $(h, n)$ into a slow recovery variable $w$, FitzHugh and Nagumo isolated the topological essence of excitability in a 2D polynomial system:
frac{dv}{dt} = v – frac{v^3}{3} – w + I_{text{ext}}
frac{dw}{dt} = epsilon (v + a – b w)
The system is analyzed via its Nullclines ($dot{v} = 0 implies w = v – v^3/3 + I$ and $dot{w} = 0 implies w = (v+a)/b$). As input current $I$ shifts the cubic $v$-nullcline vertically, the unique intersection (equilibrium) crosses the local maximum of the cubic, triggering a subcritical or supercritical Hopf bifurcation that gives birth to macroscopic limit cycles.
3. The Izhikevich Model: Biophysical Richness at LIF Speed
In 2003, Eugene Izhikevich introduced a hybrid model that combines the continuous non-linear phase-plane dynamics of quadratic integrate-and-fire systems with discontinuous state resets:
frac{dv}{dt} = 0.04 v^2 + 5 v + 140 – u + I
frac{du}{dt} = a(b v – u)
text{if } v ge +30text{ mV}, quad text{then } begin{cases} v leftarrow c \ u leftarrow u + d end{cases}
Here, $v$ represents the membrane potential, and $u$ represents a slow membrane recovery variable accounting for $text{K}^+$ and $text{Na}^+$ channel kinetics. By tuning only four dimensionless parameters ($a, b, c, d$), the Izhikevich model reproduces all twenty biologically observed neuro-computational firing patterns—including regular spiking (pyramidal cells), fast spiking (parvalbumin interneurons), intrinsically bursting, chattering, and thalamocortical relay oscillations—at nearly the computational speed of LIF.
4. Mathematical Excitability Classification: Type I vs Type II
Dynamically, neural excitability divides into two fundamental classes based on the nature of their bifurcation:
- Type I Excitability (Saddle-Node on Invariant Circle, SNIC): The frequency of firing starts at arbitrarily low rates ($f to 0$ as $I to I_c$). The neuron behaves as a pure integrator, encoding stimulus intensity directly into spike frequency.
- Type II Excitability (Subcritical Hopf): Firing begins abruptly with a non-zero minimum frequency ($f > f_{text{min}}$). The neuron exhibits damped subthreshold oscillations and acts as a resonator, sensitive to specific input frequencies.
