Monsoon rainfall anomalies are often organized into dynamical modes that are more predictable than rainfall at specific locations. Here, a linear response function (LRF) is constructed to extract the neutral modes (NM) of the dynamical atmospheric system and their optimal forcing that govern precipitation variability, providing a dynamical underpinning for the empirically derived, recurrent rainfall patterns. The leading mode (NM1) features a shift in the Atlantic ITCZ, a drying over the Indian Ocean and Indian subcontinent, and a “South-Flood-North Drought (SFND)” pattern over Eastern China, a pattern inferable from changes in the atmospheric energy flow forced by the corresponding optimal forcing and further amplified by the water vapor and cloud feedbacks. Moreover, the slow evolving component of the NM1 in response to CO2 forcing can be skillfully predicted by running the ocean heat uptake through the LRF, implicating a deterministic ocean dynamic source for the NM1-like trend under climate warming.