Brain oscillations exhibit long-range temporal correlations (LRTCs), which reflect
the regularity of their fluctuations: low values representing more random (decorrelated)
while high values more persistent (correlated) dynamics. LRTCs constitute supporting
evidence that the brain operates near criticality, a state where neuronal activities
are balanced between order and randomness. Here, healthy adults used closed-loop brain
training (neurofeedback, NFB) to reduce the amplitude of alpha oscillations, producing
a significant increase in spontaneous LRTCs post-training. This effect was reproduced
in patients with post-traumatic stress disorder, where abnormally random dynamics
were reversed by NFB, correlating with significant improvements in hyperarousal. Notably,
regions manifesting abnormally low LRTCs (i.e., excessive randomness) normalized toward
healthy population levels, consistent with theoretical predictions about self-organized
criticality. Hence, when exposed to appropriate training, spontaneous cortical activity
reveals a residual capacity for "self-tuning" its own temporal complexity, despite
manifesting the abnormal dynamics seen in individuals with psychiatric disorder. Lastly,
we observed an inverse-U relationship between strength of LRTC and oscillation amplitude,
suggesting a breakdown of long-range dependence at high/low synchronization extremes,
in line with recent computational models. Together, our findings offer a broader mechanistic
framework for motivating research and clinical applications of NFB, encompassing disorders
with perturbed LRTCs.