We present measurements of the near-field heat transfer between the tip of a thermal profiler and planar material surfaces under ultrahigh vacuum conditions. For tip-sample distances below 10 −8 m our results differ markedly from the prediction of fluctuating electrodynamics. We argue that these differences are due to the existence of a material-dependent small length scale below which the macroscopic description of the dielectric properties fails, and discuss a corresponding model which yields fair agreement with the available data. These results are of importance for the quantitative interpretation of signals obtained by scanning thermal microscopes capable of detecting local temperature variations on surfaces.PACS numbers: 44.40.+a, 03.50.De, 78.20.Ci Radiative heat transfer between macroscopic bodies increases strongly when their spacing is made smaller than the dominant wavelength λ th of thermal radiation. This effect, caused by evanescent electromagnetic fields existing close to the surface of the bodies, has been studied theoretically already in 1971 by Polder and van Hove for the model of two infinitely extended, planar surfaces separated by a vacuum gap [1], and re-investigated later by Loomis and Maris [2] and Volokitin and Persson [3,4]. While early pioneering measurements with flat chromium bodies had to remain restricted to gap widths above 1 µm [5], and later studies employing an indium needle in close proximity to a planar thermocouple remained inconclusive [6], an unambiguous demonstration of near-field heat transfer under ultrahigh vacuum conditions and, thus, in the absence of disturbing moisture films covering the surfaces, could be given in Ref. [7].The theoretical treatment of radiative near-field heat transfer is based on fluctuating electrodynamics [8]. Within this framework, the macroscopic Maxwell equations are augmented by fluctuating currents inside each body, constituting stochastic sources of the electric and magnetic fields E and H. The individual frequency components j(r, ω) of these currents are considered as Gaussian stochastic variables. According to the fluctuationdissipation theorem, their correlation function reads [9]where E(ω, β) = ω/ exp(β ω) − 1 , with the usual inverse temperature variable β = 1/(k B T ); the angular brackets indicate an ensemble average. Moreover, ǫ ′′ (ω) denotes the imaginary part of the complex dielectric function ǫ(ω) = ǫ ′ (ω) + iǫ ′′ (ω). It describes the dissipative properties of the material under consideration, which is assumed to be homogeneous and non-magnetic. Thus, Eq. (1) contains the idealization that stochastic sources residing at different points r, r ′ are uncorrelated, no matter how small their distance may be. Applied to a material occupying the half-space z < 0, facing the vacuum in the complementary half-space z > 0, these propositions can be evaluated to yield the electromagnetic energy density in the distance z above the surface, giving [10]dκ ρ E (ω, κ, β, z) + ρ H (ω, κ, β, z)
