One neuron, four equations
Hodgkin–Huxley, integrated live. Sodium in, potassium out, all-or-none. The sliders are the two conductances every sodium- and potassium-channel drug moves.
Take sodium to 80%. The threshold drive climbs steeply. Below about 70% the cell cannot fire at any drive — that is a sodium-channel blocker at a high dose.
Take potassium to 60%. Threshold falls, the cell fires at almost any drive, and each spike gets wider. Push it to 150% and it goes quiet.
Raise drive slowly through the threshold. Firing starts abruptly at about 50 Hz, not from zero.
Under the hood
Hodgkin and Huxley 1952, squid giant axon, integrated by Runge–Kutta at 0.01 ms. The conductance sliders scale ḡNa and ḡK directly.
The firing-rate curve is measured by simulating the model at each drive, not fitted. It starts at a finite rate rather than at zero because this system becomes rhythmic through a Hopf bifurcation. Real cortical pyramidal cells mostly do not behave this way, which is one reason the network tab uses a different neuron model.
A patch of cortex, wired to itself
Poke one spot and watch what the patch does with it. The excitability slider decides whether the poke fades, travels, or never stops. That last one is a seizure, and nothing is driving it.
Quiet
Background firing only.
Under the hood
2500 Izhikevich neurons, 80% excitatory, on a 50 × 50 sheet. Each cell connects mostly to neighbours within a few cells, plus a small fraction of long-range axons. That local wiring is what makes activity propagate rather than appear everywhere at once.
The self-sustaining state has a mechanism rather than a tuning. Hard firing loads intracellular chloride faster than KCC2 clears it, the GABAA reversal drifts depolarised, and inhibition loses its grip — so more firing weakens inhibition, which permits more firing. That is the documented route to ictal inhibitory failure, and it is what gives the patch a second state it can stay in.
Parameters are chosen so the three regimes are far apart on screen. Rates and conductances here are not calibrated to any measurement.
A field far too weak to fire a cell, and it stops the seizure
Current through the scalp polarises every neuron a little, in the same direction, at the same moment. It cannot make a cell fire. It does not need to.
Quiet
Press Trigger seizure.
It is shared. The field pushes every cell the same way at the same instant. Synaptic noise is independent between cells, so averaging over a million neurons keeps the signal and cancels the noise.
The tissue is unstable. Close to the tipping point, a tiny nudge moves the network a long way. That is the whole reason this works, and it predicts tDCS should do nothing to healthy cortex.
It runs for half an hour. Sustained polarisation also drives LTD-like synaptic depression, which lowers excitability itself rather than just biasing it — a lasting change, not a nudge.
| Parameter | Typical |
|---|---|
| Current | 1–2 mA |
| Duration | 20–30 min |
| Sessions | 5–14 daily |
| Cathode | over the onset zone |
| Anode | contralateral supraorbital |
| Field at cortex | 0.4–0.8 V/m |
| Somatic shift | ≈ 0.05–0.1 mV |
Under the hood, and what this leaves out
The head cross-section is a real solve, not a drawing. Scalp, skull, CSF and brain get their own conductivities and Laplace's equation is relaxed on a grid, then differentiated to get the field. The skull is roughly eighty times more resistive than the tissue either side of it, which is why the field spreads out.
Polarisation goes as the field strength times the cosine of the angle between the field and the cell's long axis, about 0.12 mV per V/m at the soma (Bikson 2004). A cell becomes a dipole: one end depolarises, the other hyperpolarises.
Depth. The commonest focal epilepsy is mesial temporal, and the hippocampus is about as far from a scalp electrode as brain tissue gets. Geometry. The folding problem above never goes away. Variation. Skull and CSF differences move the delivered field about twofold between people at identical dose. Evidence. Cathodal tDCS over the onset zone reduces seizure frequency in drug-resistant focal epilepsy — real, modest, heterogeneous, with small trials and short follow-up. Interictal discharge rates fall more reliably than seizures do.
The stimulator slider is labelled in millivolts, but its strength is set so the collapse lands mid-travel on this model. It is not a claim that 0.15 mV aborts a human seizure.