Abstract
Using the MRPD based time-dependent wave packet approach, we calculate the first fully converged state-to-state differential cross sections for the H2(v=1)+OD→H+HOD reaction on a highly accurate neural network PES. Two distinctive peaks are found in the J=0 reaction probabilities and the backward scattering differential cross sections. Detailed analysis reveals that these peaks originate from the Feshbach resonance states trapped in the peculiar well on the HOD(vOH=3) vibrationally adiabatic potential caused by chemical bond softening, and produces mainly HOD(vOH=2) product.