Place Cells and Grid Cells
The mammalian brain runs a literal coordinate system, not a metaphorical one. In 1971 John O'Keefe pushed a microelectrode into a rat's hippocampus and found neurons that fired only when the animal stood at a particular spot in its enclosure. Move the rat one body-length away, the cell went silent. Move it back, the cell fired again. The cell was not "thinking about" the location. It was the location, in the neural sense.
Place cells
A place cell in CA1 or CA3 of the hippocampus has a "place field" — typically a 20–50 cm patch of an environment where it fires at high rate, and silence almost everywhere else. Take the rat to a new room and the same cell remaps: a different field, or no field at all. The map is built per-environment, not stamped on the world.
Roughly 30–50% of pyramidal cells in rat CA1 show clear place fields in a given small environment. The fields are not arranged on the cortical sheet in any spatial order — neighboring cells code distant places. The brain stores location as a population code over thousands of arbitrary lookups, not as a topographic map.
Grid cells
In 2005 May-Britt and Edvard Moser, working one synapse upstream in the medial entorhinal cortex, found something stranger. A single neuron fires at multiple locations, and those locations form a near-perfect hexagonal lattice tiling the entire enclosure. Run the rat through a 2-meter box. Plot every spike. The result is a honeycomb.
Grid cells come in modules. Cells in one module share a lattice spacing (say 40 cm vertex-to-vertex) and orientation but differ in phase. Move dorsal to ventral along the entorhinal cortex and the spacing jumps in discrete steps, each module roughly 1.4× larger than the last. Four to ten modules cover the resolvable range. This is a metric. A physical, biological, spatial frequency decomposition.
Why hexagons
Hexagonal tiling is the densest packing of circles in a plane. If the brain's job is to represent 2D position with the fewest cells at a given resolution, hexagons win on information-theoretic grounds. The 1.4× scaling between modules is close to the optimum (√e ≈ 1.65) for combining modules into a high-capacity place code via remainder arithmetic. The brain appears to have found the mathematically best solution and then committed to it across rodents, bats, monkeys, and (via fMRI grid-like signals) humans.
What's contested
The function beyond navigation is the open question. Grid-like signals show up in human fMRI when subjects move through conceptual spaces — bird-shape morphs, social hierarchies, odor gradients. Is the entorhinal cortex a general-purpose relational engine that navigation just happens to use, or is "cognitive space" a metaphor that exploits a navigation substrate? The Buzsáki and Moser camps differ on whether the hippocampal system is fundamentally about memory (with space as a special case) or fundamentally about space (with memory piggybacking). The 2014 Nobel committee carefully said "positioning system" and let the wider claim hang.
Also unsettled: how grid cells get their grid. Continuous attractor models, oscillatory interference models, and self-organizing models all reproduce hexagons; the experiments haven't decisively chosen.
Why this has to do with other realms
The grid-cell discovery is one of the cleanest cases of a Platonic mathematical object — a regular hexagonal lattice — turning up unannounced inside biology. It connects to concept emergence (a tiling no single neuron knows about, produced by network dynamics) and to the broader question in concept fermi paradox-adjacent thinking about whether intelligence converges on similar solutions: hexagonal grids work the same for bees building combs, for cell-tower coverage planners, and for rat entorhinal cortex. Three independent optimizers, one answer.
An open question
If the entorhinal grid is a general relational metric, what does a grid cell look like when a human is reasoning about a family tree, a chess position, or a sentence? The honeycomb in physical space is well-documented. The honeycomb in idea-space is the next page worth writing.
Key sources
- O'Keefe & Dostrovsky, "The hippocampus as a spatial map" (1971, Brain Research) — the original recording paper.
- Hafting, Fyhn, Molden, Moser & Moser, "Microstructure of a spatial map in the entorhinal cortex" (2005, Nature) — the grid-cell discovery.
- O'Keefe & Nadel, The Hippocampus as a Cognitive Map (1978) — the book that turned a finding into a research program.
- Moser, Moser & McNaughton, "Spatial representation in the hippocampal formation: a history" (2017, Nature Neuroscience) — the field's own retrospective.
- to verify: Constantinescu, O'Reilly & Behrens (2016, Science) on grid-like signals in human conceptual space.
Further reading
- The Cognitive Map lectures by Edvard Moser (Nobel lecture, 2014) — the discoverer telling the story straight.
- György Buzsáki, The Brain from Inside Out (2019) — the strongest counter-framing, arguing the hippocampus is about sequences and memory first, space second.
- Sanjoy Mahajan's writing on hexagonal packing — the math half of why the brain landed here.
- DeepMind's 2018 Nature paper on grid-like representations emerging in artificial agents trained to navigate — the "convergent solution" claim, tested in silicon.
See Also
- concept emergence (a hexagonal tiling no single neuron knows about — the canonical neuroscience case)
- concept memory consolidation
- person john okeefe
- concept cognitive map
- concept information theory (why hexagons are the optimal answer, not just a pretty one)
- concept fermi paradox (convergent optimization across independent substrates)