This paper presents a new robust scheme for coupled physics nuclear reactor calculations. We focus specifically on high-fidelity whole-core transport calculations with coarse mesh finite difference (CMFD) coupled to thermal hydraulics. These simulations traditionally employ rthe Picard iteration for the coupled solution, where it has been observed that the use of CMFD (or nonlinear diffusion acceleration) is detrimental to the overall convergence of the coupled problem. Moreover, (1) if the acceleration equations are tightly converged every iteration, the overall multiphysics iteration becomes less stable and (2) properly loosening the convergence criteria of the acceleration equations at each iteration can stabilize the overall scheme. In this paper, we develop a Fourier analysis for a simplified CMFD-accelerated neutron transport problem with feedback from flux-dependent cross sections to provide a theoretical explanation for, and gain insight into, the aforementioned observations. Furthermore, we establish the theoretical relationship between relaxation and partial convergence of the low-order problem. Using this result, a relaxation-free iteration scheme is then proposed, with a formula to determine the nearly optimal partial convergence of the low-order diffusion problem. The new CMFD method is called the nearly optimally partially converged coarse mesh finite difference (NOPC-CMFD) method. It is shown theoretically that the NOPC-CMFD method in problems with feedback has stability properties comparable to CMFD in problems without feedback and requires no relaxation factor, i.e., is relaxation free. The results presented in this paper provide a theoretical foundation for the development of a robust multiphysics iteration scheme for nuclear reactor modeling. The implementation of the method and application to various test cases are presented in the companion paper.