We report on the first direct measurement of the change of the surface stress in the reconstruction of the Au(111) and the Au(100) surfaces. For both surfaces the reconstruction relaxes the intrinsic tensile stress, by 22% and 5%, respectively. A discussion of the data on the Au(111) surface in the FrenkelKontorova model shows that the energy gain due to the surface stress is not quite large enough to make the reconstructed phase energetically favored without the formation of the secondary herringbone structure of the solitons. On the Au(100) surface, the gain in elastic strain energy is clearly insufficient to cause the surface to reconstruct. [S0031-9007(97) The reconstruction has been discussed frequently using the Frenkel-Kontorova (FK) model [7][8][9][10]: In the onedimensional (1D) version, a chain of atoms (representing the top layer) linked by springs with nearest-neighbor force constant w 00 and "natural" spacing b is placed in a sinusoidal potential with amplitude W ͞2 representing a rigid substrate with periodicity a. The prime feature of the reconstruction, the soliton domain wall, results from energy minimization. Further minimization of a (very small) strain energy associated with the anisotropy of the stress relief in the soliton reconstruction leads to a secondary ("herringbone") structure of the solitons [11]. Despite the general consensus that the soliton reconstruction is driven by the large tensile stress [12,13] on the Pt(111) [4][5][6]14] and the Au(111) surfaces, there is no direct experimental evidence, e.g., by measurements on the stress relief in the reconstruction.On the reconstructed (100) surfaces of Ir, Pt, and Au, the surface atoms form quasi-hexagonal (hex) commensurate and incommensurate overlayers. The atom density in the surface layer is higher by (20-25)%. From firstprinciples calculations, Fiorentini et al. [15] find the surface stress for the unreconstructed (100) surfaces of Ir, Pt, and Au to be even larger than for the (111) surfaces. Furthermore, the stress on the (100) surfaces of the 5d fcc metals Ir, Pt, and Au is also significantly larger than for the (100) surfaces of the corresponding 4d metals. They [15] identify the tensile stress as driving the reconstruction of the (100) surfaces on the 5d metals. Indirect experimental evidence for stress relief in the reconstruction was obtained for the Ir(100) surface [16]. However, neither theory nor experiment could determine the change in the surface stress quantitatively. Without direct information about the actual amount of stress energy relieved in the reconstruction of the (100) surfaces, this interpretation of the origin of the reconstruction is questionable. We dispute it below.In this Letter we report the first measurements of the change in surface stress which accompanies the reconstruction of both the Au(111) and the Au(100) surfaces. The stress measurements were performed in an electrochemical cell, with the crystals immersed in a 0.1 M HClO 4 solution, using the cantilever bending method [17][18][19]; the bendin...