The goal of this thesis is to identify the physical mechanisms that limit the performance of an atom Michelson interferometer using Bose-Einstein condensates propagating in a magnetic guide. This work builds on the previous work by Ying-ju Wang and expands the propagation time of the condensates in the guide from 4 ms to about 200 ms. Whereas the previous work demonstrated the coherence of our atom interferometer sequence, this work focuses on the evolution of the relative phase as atoms spend more time in the guide.

We observe a coherent behavior of the condensate after separating them spatially by more than 700 microns. After such a long propagation the cloud shape does not match conventional fitting models so we implement data processing methods, called principal component analysis (PCA) and independent component analysis (ICA), that allow for model-free analysis.

In order to observe any interference effect, the timing of our interferometer needs to be very accurate so that the condensates overlap appropriately during recombination. We explain the different factors that control this overlap and present experimental methods to maximize the interference contrast.

Although the dephasing of our interferometer is sufficiently small to observe fringes, external perturbations modify the relative phase of the interferometer in a way that is unknown, leading to apparently random fluctuations of the interference pattern. We identify two of the major external perturbations, vibrations and magnetic fields, and we explain how they couple to the atoms. We measure the vibrations of our interferometer and include its contribution in the overall phase shift calculation. We also attenuate them in order to study the effect of a controllable magnetic phase shift on our interferometer. Beyond these external perturbations, we study the effects of more intrinsic phase-noise inducing causes leading to a loss of fringe visibility. The role of quantum phase diffusion is also discussed and is shown to contribute significantly to the loss of fringe visibility under the conditions of our experiment.