Background: Current classification systems ignore the family histories of patients and psychiatric and medical comorbidity. Methods: We study a new approach of applying spectral clustering to determine distinct bipolar disorder subtypes, which is data whose clusters are of various sizes and densities. We discovered clusters by processing a SRB (Sinai-Ruelle-Bowen) similarity matrix that reflects the proximity of Von Bertalanffy’s functions fitted to phase growth dynamics of EEG (electroencephalography) within a new pipeline architecture. For this purpose, 109 patients diagnosed with bipolar disorder according to DSM-V (Diagnostic and Statistical Manual of Mental Disorders, fifth edition) were evaluated in remission period cross-sectionally. Results: We found three distinct bipolar disorder subtypes with the p-values < 0.001. We exhibit mixing sub-shifts of EEG phase gradients such that there are chaotic phase transitions but higher order phase gradients in a cone basin is always strictly convex. More surprisingly, we show that the SRB entropy measures on some time interval although there exist several equilibrium states each corresponds to equilibrium state. Conclusion: It seems subtypes of the bipolar spectrum were shaped according to seasonality, comorbidity for anxiety disorder and presence of psychotic symptom.