Context. The elliptical galaxy NGC 3923 is surrounded by numerous stellar shells that are concentric arcs centered on the galactic core. They are very likely a result of a minor merger and they consist of stars in nearly radial orbits. For a given potential, the shell radii at a given time after the merger can be calculated and compared to observations. The Modified Newtonian Dynamics (MOND) is a theory that aims to solve the missing mass problem by modifying the laws of classical dynamics in the limit of small accelerations. Hernquist & Quinn(1987) claimed that the shell distribution of NGC 3923 contradicted MOND, but Milgrom(1988) found several substantial insufficiencies in their work. Aims. We test whether the observed shell distribution in NGC 3923 is consistent with MOND using the current observational knowledge of the shell number and positions and of the host galaxy surface brightness profile, which supersede the data available in the 1980s when the last (and negative) tests of MOND viability were performed on NGC 3923. Methods. Using the 3.6 um bandpass image of NGC 3923 from the Spitzer space telescope we construct the mass profile of the galaxy. The evolution of shell radii in MOND is then computed using analytical formulae. We use 27 currently observed shells and allow for their multi-generation formation, unlike the Hernquist & Quinn one-generation model that used the 18 shells known at the time. Results. Our model reproduces the observed shell radii with a maximum deviation of 5% for 25 out of 27 known shells while keeping a reasonable formation scenario. A multi-generation nature of the shell system, resulting from successive passages of the surviving core of the tidally disrupted dwarf galaxy, is one of key ingredients of our scenario supported by the extreme shell radial range. The 25 reproduced shells are interpreted as belonging to three generations.

Stellar shells observed in many giant elliptical and lenticular as well as a few spiral and dwarf galaxies, presumably result from galaxy mergers. Line-of-sight velocity distributions of the shells could, in principle, if measured with a sufficiently high S/N, constitute one of methods to constrain the gravitational potential of the host galaxy. Merrifield & Kuijken (1998) predicted a double-peaked line profile for stationary shells resulting from a nearly radial minor merger. In this paper, we aim at extending their analysis to a more realistic case of expanding shells, inherent to the merging process, whereas we assume the same type of merger and the same orbital geometry. We use analytical approach as well as test particle simulations to predict the line-of-sight velocity profile across the shell structure. Simulated line profiles are convolved with spectral PSFs to estimate the peak detectability. The resulting line-of-sight velocity distributions are more complex than previously predicted due to non-zero phase velocity of the shells. In principle, each of the Merrifield & Kuijken (1998) peaks splits into two, giving a quadruple-peaked line profile, which allows more precise determination of the potential of the host galaxy and, moreover, contains additional information. We find simple analytical expressions that connect the positions of the four peaks of the line profile and the mass distribution of the galaxy, namely the circular velocity at the given shell radius and the propagation velocity of the shell. The analytical expressions were applied to a test-particle simulation of a radial minor merger and the potential of the simulated host galaxy was successfully recovered. The shell kinematics can thus become an independent tool to determine the content and distribution of the dark matter in shell galaxies, up to ~100 kpc from the center of the host galaxy.