Measuring Fermi velocities with ARPES in narrow band systems: The case of layered cobaltates

► Fermi velocities are measured in various cobaltates compounds, thanks to Angle-Resolved Photoemission. ► The dependence of this value on the method of analysis is emphasized and attributed to a k-dependence of the self-energy. ► These results should be useful for extracting correct Fermi velocities in any strongly correlated materials, with small Fermi velocities. ARPES is a priori a technique of choice to measure the Fermi velocities vF in metals. In correlated systems, it is interesting to compare this experimental value to that obtained in band structure calculations, as deviations are usually taken as a good indicator of the presence of strong electronic correlations. Nevertheless, it is not always straightforward to extract vF from ARPES spectra. We study here the case of layered cobaltates, an interesting family of correlated metals. We compare the results obtained by standard methods, namely the fit of spectra at constant momentum k (energy distribution curve, EDC) or constant binding energy ω (momentum distribution curve, MDC). We find that the difference of vF between the two methods can be as large as a factor 2. The reliability of the 2 methods is intimately linked to the degree of k- and ω-dependence of the electronic self-energy. As the k-dependence is usually much smaller than the ω dependence for a correlated system, the MDC analysis is generally expected to give more reliable results. However, we review here several examples within cobaltates, where the MDC analysis apparently leads to unphysical results, while the EDC analysis appears coherent. We attribute the difference between the EDC and MDC analysis to a strong variation of the photoemission intensity with the momentum k. This distorts the MDC lineshapes but does not affect the EDC ones. Simulations including a k dependence of the intensity allow to reproduce the difference between MDC and EDC analysis very well. This momentum dependence could be of extrinsic or intrinsic. We argue that the latter is the most likely and actually contains valuable information on the nature of the correlations that would be interesting to extract further.