We revisit the problem of the molecular mechanism of the chain diffusion between crystalline and amorphous fractions in semicrystalline polyethylene (PE). There exists a long-standing controversy on the nature of the topological point defects which diffuse along the chain stems in crystallites and shift the stems. Namely, the conformational (including gauche conformations) twist-compression (interstitial-like) and the smooth (soliton-like) twist-tension (vacancy-like) localized defects were offered for this role. However, none of the proposed models for the process could explain all the experimental facts which seemed unclear and contradictory. Moreover, it was discovered recently that in PE samples of uncommon morphology (electron beam irradiated samples, fibers and single crystals) the diffusion process has the activation energy about 3 times less than that in common melt-crystallized samples. No explanation ever followed. We have carried out molecular dynamics (MD) simulation of both the defects in a realistic model of PE crystal and obtained estimates for their formation energies and diffusion coefficients. These estimates together with analysis of available experimental data allow to solve both the problems and to propose models for molecular mechanisms of the observed diffusion processes. The agents of the 'old' diffusion process are the smooth twist-tension defects. Shifts in a chain stem of a crystallite in a common sample are initiated at the interface to an amorphous region through extended thermal motion of the chain stem in the amorphous region. If the motion causes a strong pull (with a twist) at the chain stem in the crystallite, such motion produces a smooth defect of twist-tension on this stem. The proposed molecular model conforms with available mechanical experiments if one accepts that the process corresponds to the most low temperature (alpha(1)) from the alpha-peaks observed. The 'new' diffusion process results from diffusion of the conformational twist-compression defects in crystallites. The needed sequence of conformations appears near a crystallite as a result of a quick gamma process. Because the state of the semicrystalline polymer is unstable, the position of the boundary between the crystalline and disordered regions fluctuates so that segments of chains pass from disordered to crystalline state (and vice versa). The conformational defects in disordered region are captured through expansion of the crystalline region where they become stable and diffuse along the chains. Our MD estimate for the activation energy of the process E-act <= 8.65 kcal/mol is in a good agreement with the experimental value 7 kcal/mol. The diffusion coefficients of both the defects are too high to have effect on the statistics of both of these very slow processes. Therefore the statistics of the 'old' process is the statistics of strong thermal pulls at chain stems in crystallites, and the statistics of the 'new' process is related to the statistics of fluctuations of the position of the boundaries between crystalline and disordered fractions. (c) 2007 Elsevier Ltd. All rights reserved.