Researchers at EMPA, St. Gallen, Switzerland cultivated an in vitro neural network on one side of a multielectrode array and grew myofibrils on the other side to measure the activity between them. The neural action potential signals resulting from 60 spatially separated channels are analyzed by using correlation functions C(i,j)(τ) between channels i and j. Integrating the correlation functions up to a certain value results in the correlation matrix a(i,j) which is a measure for the strength of the correlations.
Due to the asymmetry of a(i,j) an information flow can be extracted from the network which results in a transfer matrix A(i,j) defining a Markov process. A set of correlation measures is established by taking a(i,j) to the power of [beta].
By introducing a corresponding partition function Z([beta]) phase-transition-like behavior near a critical [beta] is studied. At early days of the experiment the network is strongly synchronized in contrast to later times. The transitions are stronger for the asynchronous case where the peaks of the second derivative of the partition function are much higher than for the synchronous case.