In this paper, we consider retransmission-based point-to-point dual energy harvesting (EH) links, where both the transmitter and receiver are EH nodes (EHNs). The transmitter needs to periodically send a packet to the receiver and the packet is dropped if it is not delivered within a given number of slots. The goal is to find a retransmission index-based power management policy (RIP), which minimizes the packet drop probability (PDP). To this end, first, we establish the near-optimality of policies that operate in the energy unconstrained regime (EUR), i.e., the regime where the average rate of energy use at each EHN is less than the average harvesting rate. Specifically, we analytically show that for such policies, the gap between the PDP of the dual EH systems with finite and infinite capacity batteries decreases exponentially with the size of the battery at the transmitter and receiver. Next, we show that, in the EUR, the non-convex problem of designing optimal RIPs can be reformulated as a geometric program, which leads to a provably convergent and computationally efficient solution. We design the RIPs for both slow and fast fading channels, and with two different retransmission protocols, namely, the automatic repeat request (ARQ) and hybrid ARQ with chase combining. Numerical results obtained through Monte Carlo simulations show that the proposed RIPs outperform the state-of-the-art policies.