From stellar spectra, a variety of physical properties of stars can be derived. In particular, the chemical composition of stellar atmospheres can be inferred from absorption line analyses. These provide key information on large scales, such as the formation of our Galaxy, down to the small-scale nucleosynthesis processes that take place in stars and supernovae. By extending the observed wavelength range toward bluer wavelengths, we optimize such studies to also include critical absorption lines in metal-poor stars, and allow for studies of heavy elements (Z = 38) whose formation processes remain poorly constrained. In this context, spectrographs optimized for observing blue wavelength ranges are essential, since many absorption lines at redder wavelengths are too weak to be detected in metal-poor stars. This means that some elements cannot be studied in the visual-redder regions, and important scientific tracers and science cases are lost. The present era of large public surveys will target millions of stars. It is therefore important that the next generation of spectrographs are designed such that they cover a wide wavelength range and can observe a large number of stars simultaneously. Only then, we can gain the full information from stellar spectra, from both metal-poor to metal-rich ones, that will allow us to understand the aforementioned formation scenarios in greater detail. Here we describe the requirements driving the design of the forthcoming survey instrument 4MOST, a multi-object spectrograph commissioned for the ESO VISTA 4 m-telescope. While 4MOST is also intended for studies of active galactic nuclei, baryonic acoustic oscillations, weak lensing, cosmological constants, supernovae and other transients, we focus here on high-density, wide-area survey of stars and the science that can be achieved with high-resolution stellar spectroscopy. Scientific and technical requirements that governed the design are described along with a thorough line blending analysis. For the high-resolution spectrograph, we find that a sampling of >= 2.5 (pixels per resolving element), spectral resolution of 18 000 or higher, and a wavelength range covering 393-436 nm, is the most well-balanced solution for the instrument. A spectrograph with these characteristics will enable accurate abundance analysis (+/- 0.1 dex) in the blue and allow us to confront the outlined scientific questions. (C) 2015 WILEY-VCH Verlag GmbH& Co. KGaA, Weinheim