The significant discrepancy between first-principles calculations and experimental analyses for the relaxation of the (001) surface of rhodium has been a puzzle for some years. In this paper we present density functional theory calculations using the local-density approximation and the generalized gradient approximation of the exchange-correlation functional. We investigate the thermal expansion of the surface and the possibility of surface magnetism. The results throw light on several, hitherto overlooked, aspects of metal surfaces. We find, that, when the free energy is considered, density-functional theory provides results in good agreement with experiments.PACS numbers: 68.35. Bs, 75.30.Pd, 68.35.Ja, 63.20.Ry The significant discrepancy between first-principles calculations [1,2,3,4,5] and low-energy electron diffraction (LEED) analyses [6,7,8] for the relaxation of the (001) surface of rhodium has been a puzzle for some years. The earlier LEED studies [6,7] concluded that the interlayer spacing of the surface layer (d 12 ) is nearly identical to that in the bulk (d 0 ), i.e., the top-layer relaxation was determined to be ∆d 12 /d 0 = +0.5 ± 1.0 %. A recent LEED study [8] found ∆d 12 /d 0 = −1.16 ± 1.6 %. On the other hand, first-principles calculations showed a large top-layer relaxation ranging from −3.2 % to −5.1 %, depending on the calculational scheme and/or the employed numerical accuracy [1,2,3,4,5]. Inward relaxations are indeed the expected behavior of transition metals surfaces (see e.g. Ref.[2] ), and the practical zero relaxation determined by LEED is at least unexpected.In order to reconcile this disagreement between their calculations and experiment, Feibelman and Hamann [1] proposed that in the experimental study the metal surface may be contaminated by residual hydrogen adsorption (see also Ref. [9]). Indeed, hydrogen is not easy to detect and quite soluble in transition metals, such as Ru, Rh, and Pd. Furthermore it is known that adsorbed hydrogen significantly reduces the inward relaxations at metal surfaces as it increases the bond coordination of the surface atoms, making them, to some extent, more bulk like. However, the possibility of hydrogen contamination was strongly rejected by later experimental papers (e.g. [8,10]).Morrison et al.[3] investigated an alternative possibility [11], namely that the presence of surface magnetism could increase the first interlayer spacing, i.e., reducing the large inward relaxation they had obtained in their non-magnetic calculation by "magnetic pressure". In fact, bulk Rh is already close to fulfilling the Stoner criterion of ferromagnetism, and the narrower density of d-states at the surface might stabilize a magnetic state at the surface. Density-functional theory (DFT) together with the local-density-approximation (LDA) gives a nonmagnetic ground state for Rh (001), but this might be due to the LDA. For example, for bulk iron, which is studied in greater detail, the LDA falsely puts the bcc magnetic ground state at a higher energy than the nonma...