Compute with physical units

brainevent is fully compatible with BrainUnit, so synaptic weights and currents can carry physical units (e.g. millisiemens, millivolts) and be dimensionally checked throughout a simulation. This guide shows how units flow through event-driven operations.

Attach units to weights

Give a connectivity weight a unit and the result of the matrix product inherits it:

import brainevent
import brainunit as u
import jax.numpy as jnp

spikes = brainevent.BinaryArray(jnp.array([1, 0, 1, 0, 1], dtype=jnp.float32))

conn = brainevent.JITCScalarR(
    num_pre=5, num_post=3, prob=0.5,
    weight=1.62 * u.mS,        # weight carries a unit
    seed=0,
)

current = spikes @ conn        # current is a unit-aware quantity (mS-scaled)

Why this matters

• Dimensional safety — multiplying a conductance by a voltage and adding it to a current raises an error if the dimensions do not line up, catching modeling bugs early.

• Readable models — parameters read as 20. * u.ms or -50. * u.mV instead of bare floats whose units you must remember.

• No performance cost — units are tracked at trace time and erased before the kernel runs, so event-driven kernels stay just as fast.

Mixing units and plain arrays

Unit-aware quantities and plain JAX arrays interoperate: a BinaryArray of spikes is dimensionless, and it picks up the weight’s unit through the product. Everything continues to work inside jax.jit and jax.grad.

See also

The BrainUnit documentation covers the full unit system. Most brainevent constructors that accept a weight accept a unit-aware weight in its place.