The Density Matrix in Your Brainwaves
What a borrowed mathematical formalism reveals about inference under bounded computational precision
The headline, in three lines.
- A pre-specified confirmatory test on the textbook frontal theta/beta-ratio biomarker failed FDR correction at \(N = 28\) paediatric resting-state EEG, Indonesia.
- An exploratory density-matrix feature space, built directly from the multichannel analytic signal, reached mean balanced accuracy 0.657 and ROC AUC 0.780 under matched 10-fold × 10 cross-validation; the best classical-QEEG classifier on the same children reached 0.618 / 0.615.
- Single-cohort pilot. Replication queued on the OpenNeuro ds004284 paediatric cohort. The full AUC comparison sits at Figure 5.
In an Indonesian primary school, a child sits in front of a laptop with fifteen wet electrodes on her scalp. She closes her eyes during a resting-state recording. Somewhere on the recording laptop, a Python script is logging her EEG at 250 Hz. Another twenty-seven children, ages six to twelve, contributed to the same dataset.
The data she produced has been analysed under the same protocol that defines paediatric-EEG biomarker work. The pre-specified confirmatory test on her cohort, between the textbook frontal theta/beta ratio and an executive-function score, did not survive correction.
The literature is not wrong. Four decades of replication established that the ratio carries real signal. What the same four decades also did was collapse a rich multidimensional signal into a single ratio, and then teach the field that the ratio was the signal. It is not. The ratio is one projection of a much larger object, and the rest of that object carries information the textbook menu has been systematically discarding.
The methodological move that recovered the signal comes from a corner of physics where a system in a mixed state is described by a matrix that retains everything you can know about it, including the parts a power spectrum would average over. Section 3 builds that matrix directly from the EEG, with no quantum hardware in the loop, and uses its entries as features.
What follows is twelve minutes of the pilot: the textbook biomarker, the hypothesis-test failure, the methodological move that recovered the signal, and a like-for-like comparison on the same twenty-eight children. The result is exploratory, on a single cohort, and points in a direction worth replicating.
Figure 0. A 4-second loop showing fifteen 10-20-system electrodes appearing on a child’s scalp during the impedance check. Cohort: N = 28 children, ages 6-12, recorded at Islamic Green School, Indonesia. Pipeline: HAPPE → ICA → average reference. Sampling rate 250 Hz.