Compact quantum electrodynamics in 2 + 1 dimensions often arises as an effective theory for a Mott insulator, with the Dirac fermions representing the low-energy spinons. An important and controversial issue in this context is whether a deconfinement transition takes place. We perform a renormalization group analysis to show that deconfinement occurs when N > Nc = 36/π 3 ≈ 1.161, where N is the number of fermion replica. For N < Nc, however, there are two stable fixed points separated by a line containing a unstable non-trivial fixed point: a fixed point corresponding to the scaling limit of the non-compact theory, and another one governing the scaling behavior of the compact theory. The string tension associated to the confining interspinon potential is shown to exhibit a universal jump as N → N − c . Our results imply the stability of a spin liquid at the physical value N = 2 for Mott insulators.PACS numbers: 11.10. Kk, 71.10.Hf, 11.15.Ha An important topic currently under discussion in condensed matter physics community is the emergence of deconfined quantum critical points in gauge theories of Mott insulators in 2 + 1 dimensions [1,2]. A closely related problem concerns the stability of U (1) spin liquids in 2 + 1 dimensions [3,4]. In either case, models which are often considered as toy models in the high-energy physics literature are supposed to describe the low-energy properties of real systems in condensed matter physics. For instance, a model that frequently appears in the condensed matter literature is the (2 + 1)-dimensional quantum electrodynamics (QED3) [5,6]. It emerges, for instance, as an effective theory for Mott insulators [7,8,9]. Let us briefly recall how it arises in this context. The Hamiltonian of a SU (N ) Heisenberg antiferromagnet is written in a slave-fermion representation as. The resulting effective theory can be treated as a lattice gauge theory, where the gauge field A ij emerges as the phase of χ ij , i.e., χ ij = χ 0 e iAij , where χ 0 is determined from mean-field theory. The (2 + 1)-dimensional lowenergy effective Lagrangian in imaginary time has the form [4,7,8,9]where each ψ a is a four-component Dirac spinor and F µν = ∂ µ A ν − ∂ ν A µ is the usual field strength tensor. A rough estimate of the bare gauge coupling is given by e 2 0 ∼ χ 4 0 a 3 , where a is the lattice spacing. An anisotropic version of QED3 has also been studied in the context of phase fluctuations in d-wave superconductors [10,11]. A key feature of the QED3 theory of Mott insulators is its parity conservation. In fact, it is possible to introduce two different QED3s, one which conserves parity and one which does not. The latter theory involves two-component spinors, and allows for a chirally-invariant mass term which is not parityinvariant. In such a QED3 theory a Chern-Simons term [12] is generated by fluctuations [13]. The QED3 theory relevant to Mott insulators and d-wave superconductors involves four-component spinors, and does possess chiral symmetry [5,6]. In such a model, the chiral symmetry can ...