How memory remembers — the physical logic that drives the brainwave

Memory is a physical attractor: a Hebbian write deepens an engram-cell bistable well, a stored pattern completes from a 10% cue, and a theta-phase clock — writing and reading on opposite phases — measurably protects old memory [verified]. This timing drives the brainwave; absolute capacity in patterns and milliseconds remains open.

Memory is a physical attractor. A Hebbian write deepens an engram cell’s bistable well until the stored pattern self-sustains and completes from a tenth of a cue. The theta brainwave is the clock: writing on one phase and retrieving on the other keeps new learning from overwriting old memory — phase separation measurably protects it.

The question the brainwave forces

A brainwave is the ringing of remembering tissue, so the engine has to answer a physical question the old paper left abstract: how does a cell actually hold a memory? Not metaphorically — what is the mechanism that lets a pattern be stored, survive, and be recalled from a fragment, and how does the emerged theta rhythm of §4 turn that mechanism into a working memory? This chapter answers it as module M2 (reproduce).

An engram cell is a bistable switch

Each hippocampal engram cell is the same R19 switch used everywhere in this program: a unit with two stable states and an energy barrier between them. A memory is a pattern of which cells are up and which are down. The reason a memory can persist at all is the barrier: once a cell is pushed into a state, the fold holds it there against noise. Memory is not stored in activity that must be refreshed; it is stored in the shape of the wells.

Writing deepens the well

Storage is Hebbian: cells that are co-active wire together, which in the network means adding an outer product of the pattern to the coupling. Physically this deepens the attractor — it tilts every cell’s landscape so the stored pattern sits at the bottom of a basin. In the engine the written pattern’s basin is about 56× as deep as a single cell’s fold threshold, so it survives a full opposite-tilt kick. Released with no cue at all, the pattern reconstructs itself perfectly (overlap = 1.0): it is a genuine self-sustaining attractor. [V]

Retrieval is pattern completion

Recall is the network falling downhill. Impose only a fraction of the stored pattern as a cue and let the rest relax: each free cell moves to the state its neighbours’ pull dictates, and the whole network slides into the nearest basin. In the engine a cue of as little as 10% of the cells completes to the full memory (overlap = 1.0 across cue fractions from 0.1 to 0.7). That is content-addressable memory — a fragment of a scene retrieves the scene — emerging from the fold geometry, with nothing stored but the wells. [V]

Capacity, honestly

A finite network cannot hold unlimited memories: stack too many and their basins overlap and corrupt recall. The engine reaches a clean capacity of 21 patterns in 120 cells before cross-talk breaks the first one — about 0.18 per cell, the same order as the classical Hopfield bound (≈ 0.14). The number is reported, not tuned. [V]

The theta brainwave is the clock that protects memory

This is where the emerged brainwave earns its place. New learning and old recall both use the same synapses, so writing all the time would steadily overwrite what is already stored — catastrophic interference. The fix is physical and it is the theta rhythm itself (Hasselmo’s mechanism): the network writes on one phase of theta (when its own recurrent pull is suppressed, so it learns the input cleanly) and retrieves on the opposite phase (when the recurrent pull is strong, so it completes from memory). Encoding and retrieval never happen at the same instant.

The engine measures the payoff directly. Interleaving new writes with old recall on a single phase corrupts the old memory (interference ≈ 0.036); separating them across theta phases drives that corruption to zero. Phase separation protects memory — a reproducible result, and the concrete reason the brainwave is not decoration: the wave is the timing on which storage physically depends. [V]

Working memory is the theta/gamma slot count

Held memory and momentary memory share the same rhythm. The number of gamma cycles that nest inside one theta cycle — the ≈ 6 from §4 — is exactly the number of items the system can keep active at once: working-memory capacity is a count of slots in the emerged brainwave, not a separate assumption. The bridge from rhythm to cognition is mechanical.

What stays open

The mechanism is verified: attractor storage, cue completion, capacity, theta-phase protection, and the slot count all reproduce from shipped code. The magnitudes are open — the depth of a real synapse in millivolts, the duration of a theta cycle in milliseconds, the micro-amp scale of the currents — [O]. Mapping the dimensionless wells onto SI needs the membrane constants this engine does not carry. The honest claim is the physics of remembering, reproducibly; its units are deferred.

What this hands forward

With a substrate that emerges, a brainwave that radiates, and a memory that physically stores and protects itself on the brainwave’s clock, the functional arc can begin: the parallel eddies are candidate memories igniting at once, selection picks one, and the stream is their serial recall — every step now standing on emerged physics rather than assumption.