Spectral tilt underlies mathematical problem solving

Neural activity associated with successful cognition appears as a tilt in the power spectrum of the local field potential, wherein increases in high-frequency power accompany decreases in low frequency power. Whereas this pattern has been shown in a wide range of memory tasks, it is unknown whether this increased spectral tilt reflects underlying memory-specific processes or rather a domain-general index of task engagement. To address the question of whether increased spectral tilt reflects increased attention to a cognitive task, we collected intracranial recordings from three hundred thirty neurosurgical patients as they performed a mathematical problem solving task. We used a mathematical problem solving task, because it allowed us to decouple task-specific processes with domain-general attention in a novel way. Using a statistical model to control for inherent problem complexity, we classified individual math problems based on whether a subject performed faster than predicted (high-attention or fast) or slower than predicted (low-attention, or slow) based on residual response times. In contrast to the domain-general attentional account, problems that took longer than predicted produced stronger evidence for the spectral tilt: widespread increases in high frequency (31–180 Hz) power and decreases in low frequency (3–17 Hz) power across frontal, temporal, and parietal cortices. The pattern emerged early within each trial and was sustained throughout the response period but was not observed in the medial temporal lobe. The data show that engaging in mathematical problem solving leads to a distributed spectral tilt pattern, even when accounting for variability in performance driven by the arithmetic demands of the problems themselves, and suggest that broadband changes in the power spectrum reflect an index of information processing in the brain beyond simple attention to the cognitive task.