T he response of a material to external stimuli depends on its low-energy excitations. In conventional metals, these excitations are electrons on the Fermi surface-a contour in momentum (k) space that encloses all of the occupied states for non-interacting electrons. The pseudogap phase in the copper oxide superconductors, however, is a most unusual state of matter 1 . It is metallic, but part of its Fermi surface is 'gapped out' (refs 2,3); low-energy electronic excitations occupy disconnected segments known as Fermi arcs 4 . Two main interpretations of its origin have been proposed: either the pseudogap is a precursor to superconductivity 5 , or it arises from another order competing with superconductivity 6 . Using angle-resolved photoemission spectroscopy, we show that the anisotropy of the pseudogap in k-space and the resulting arcs depend only on the ratio T/ T * (x), where T * (x) is the temperature below which the pseudogap first develops at a given hole doping x. The arcs collapse linearly with T/ T * (x) and extrapolate to zero extent as T → 0. This suggests that the T = 0 pseudogap state is a nodal liquid-a strange metallic state whose gapless excitations exist only at points in k-space, just as in a d-wave superconducting state.In Fig. 1a,b we show data for a slightly underdoped sample of Bi 2 Sr 2 CaCu 2 O 8+δ (Bi2212) with a transition temperature T c = 90 K, for the superconducting state at 40 K, and the pseudogap phase at 140 K. The energy distribution curves (EDCs) at the Fermi momentum k F , which have been symmetrized 4 to remove the effects of the Fermi function on the spectra. k F is determined by the minimum separation between the peaks in the symmetrized spectra along each momentum cut. Fifteen momentum cuts were measured, as shown in Fig. 1e. Details of the symmetrization procedure are explained in the Methods section. The difference between the spectra in the two states is apparent: sharp spectral peaks are present in the superconducting state, indicating longlived excitations, and the superconducting gap vanishes only at points in the Brillouin zone, known as nodes; on the other hand, the spectra in the pseudogap phase are much broader, indicating short-lived excitations. Although a pseudogap is seen in cuts 1-7, substantial parts of the Fermi surface, cuts 8-15, show spectra peaked at the Fermi energy, indicating a Fermi arc of gapless excitations.The gap size can be estimated as half the peak-to-peak separation in energy. A more quantitative estimate is obtained by using a simple phenomenological function to describe the spectral lineshapes 7