Source
- Xu, Long, Feng & Gong (2023), LIT-638, read in full in NOTE-502: Figs. 2–8, the Methods on scale selection, phase and vorticity fields, spiral detection, the null model and the phase-oscillator lattice, and the Discussion.
What was actually shown
The data. Xu et al. (LIT-638) analyse Human Connectome Project 3T fMRI: three random cohorts of 100 subjects (rest, a language task, and a replication of the language task) and 100 doing the working-memory task. The signal is BOLD, temporally band-passed to 0.01–0.1 Hz and projected onto the HCP flat map of each hemisphere. Everything below is therefore about infra-slow haemodynamic fluctuations on a flattened sheet. The paper cites, but does not show, that such fluctuations track infra-slow electrical activity.
The scale was chosen on the outcome. The spatial band-pass was set by sweeping difference-of-Gaussian filters across the wavelengths of the first twelve geometric eigenmode groups and keeping the range in which spiral-based task classification was highest: 70.17 mm < Sc < 137.95 mm, “the classification accuracy is maximized” (Methods). The authors say spirals at other scales behave similarly, without showing it in the main text. The decoding numbers below are therefore optimistic estimates; the replication cohort used the same band.
The defects. The Hilbert phase of each vertex defines a phase field, its gradient a phase vector field, and its curl a vorticity. Spirals are localised rotations centred on phase singularities, points of near-zero amplitude round which every phase occurs. A spiral counts only if the filtered field matches the unfiltered one beyond the 95th percentile of a 3-D Fourier phase-randomised null, because spatial filtering makes spirals appear in noise too. At rest the left hemisphere holds 16.33 ± 2.22 at a time (18.8 ± 2.6 during the task), of radius 10.46 ± 4.77 mm; 62% span more than one of 22 regions, and their density is highest on boundaries between functional regions and networks, not on sulci, gyri or inflection points (Fig. 2, Extended Data Fig. 2).
Their dynamics. Centres drift at 2.19 ± 1.80 mm/s, superdiffusively (mean-squared-displacement exponent between 1 and 1.9, mean about 1.5, for 96% of trajectories). Of pairwise interactions, 51.04% are full annihilation of two spirals of opposite rotation, 46.40% partial annihilation, and 2.55% repulsion between spirals of the same rotation (Fig. 4). A full annihilation leaves a plane wave travelling the other way.
Their bearing on the task. Spiral positions and rotation senses decode four language conditions on single trials at 48.33% (replication 50.35%; chance 25%; phase-randomised null 24.32%), against 30.2% for a decoder given raw fMRI amplitude at matched input size, and working-memory stimulus type, load and accuracy above chance. Trial-averaged unfiltered task fMRI shows spirals at the same places and with the same senses. The authors then read routing rules off phase-vector streamlines: a pair of opposite spirals forms an S-shaped “gate” that passes flow, a like pair a “wall” that diverts it, and during math listening and answering spiral clusters route flow bottom-up and top-down respectively between auditory and lateral temporal cortex (Fig. 8). In a 30 × 30 lattice of coupled phase oscillators, annihilation is again the commonest interaction, and reversing selected spirals reverses the flow between them.
What could have come out otherwise: spirals could have vanished against the null, or sat on sulcal folds, or interacted without the charge-dependent asymmetry, or decoded no better than amplitude. None did.
Why “a pattern-forming medium”, and not more
The claim this account files is that the brain’s large-scale slow activity has the organisation that non-equilibrium pattern-forming media have: a phase field whose topological defects, of charge ±1, move, annihilate in opposite pairs and govern the flow around them. The record holds that organisation in chemistry and physics. Zhabotinsky’s review of the Belousov–Zhabotinsky reaction (LIT-531, read in NOTE-414) reports that in the unstirred, excitable medium colliding waves annihilate and broken fronts curl into spirals. Cross and Hohenberg (LIT-527) hold that phase dynamics and topological defects of a broken continuous symmetry are established in non-equilibrium laboratory systems. Turing’s morphogenesis paper (LIT-535, read in NOTE-427) shows that a reaction–diffusion medium can produce travelling waves, his case (e), which needs at least three morphogens.
That is an analogy, and the record draws it, not the paper. Xu et al.
cite Kuramoto’s oscillator lattices and defect-mediated turbulence, liken
their spirals to vortices in turbulence, active matter and the heart, and
note that brain spirals differ from fluid defects in diffusing
superdiffusively. They do not use the word “excitable”, and their model is
a lattice of phase oscillators, which is an oscillatory medium and not an
excitable one. So the title says “pattern-forming medium” rather than
“excitable medium”: the evidence fixes the defect phenomenology, not which
class of medium produces it. No extends is declared. The record holds no
theory of pattern formation that this one carries further, and LIT-531,
LIT-527 and LIT-535 are works, not accounts.
The rival description
Vohryzek et al.’s whole-brain-modelling review (LIT-080, read in NOTE-054)
describes non-stationary brain dynamics as visits to a repertoire of
discrete substates with dwell times and transition probabilities on a
metastable attractor landscape. Xu et al. argue the reverse for the same
phenomenon: spiral-based coordination means transitions between functional
networks “evolve continuously across space and time, rather than discrete
or sudden changes between different metastable states as previously
suggested” (Discussion). LIT-638 already declares rivals against
LIT-080. No theory in the record states the discrete-substate view (a
grep of record/theory.d for metastability, substates and LIT-080 finds
none), so no rivals is declared here. That claim in the paper is also
the weakest part of it: an interpretation of the routing result, not
tested against a state-switching model. This account takes the defect
organisation as its claim and leaves the continuous-versus-discrete
question open; if a discrete-substate theory is filed, it and this one
should be weighed then, and they may turn out compatible descriptions at
different grains rather than rivals.
Added 2026-10-16. The discrete-substate view is now filed as THEORY-571, from LIT-080. It is a framing from a review, not a tested account, and it does not claim that a continuous description fails. So it is not declared a rival here either; the comparison this account’s promote_when names is what would decide between them, and QUESTION-079 holds the open choice.
What this does not say
- It does not say the brain is an excitable medium. See above: the defect phenomenology is shared by oscillatory and excitable media, and the paper models only the former.
- It does not say the spirals are electrical. The evidence is BOLD at 0.01–0.1 Hz. Muller et al.’s review of travelling waves (LIT-647) is the record’s entry for electrophysiological waves, and it is Deferred and unread, so nothing here compares the two scales.
- It does not say the spirals route activity. “Flow” is the phase gradient, not measured transfer of activity or information, and the gate and wall rules come from streamlines and a phenomenological lattice. Nothing in the brain was perturbed.
- It does not give a decoding accuracy to quote. The scale was chosen where decoding peaked.
- It does not exclude the flat map. Flattening distorts distances, and a wave reverberating round a sulcus could appear as a spiral; the authors argue against this with a sulcal-depth test only.
- It does not say what makes cortical defects superdiffusive, which is where the analogy with fluids breaks.
Connections
- Coordination by synchrony. Verzhbinsky et al. (LIT-633) place long-range coordination in brief co-occurring ripples near 90 Hz, with no measured direction of travel; this account places it in propagating phase structure at 0.01–0.1 Hz. They work three orders of magnitude apart in time and neither tests the other.
- THEORY-069. The record’s other account of coordination as phase dynamics, the HKB transition in relative phase. It concerns two coupled components, not a field of defects; no relation is declared.