[Abridged] n-Particle Irreducible Effective Actions (nPIEA) are a powerful tool for extracting non-perturbative and non-equilibrium physics from quantum field theories. Unfortunately, practical truncations of nPIEA can unphysically violate symmetries. Pilaftsis and Teresi (PT) addressed this by introducing a "symmetry improvement" scheme in the context of the 2PIEA for an O(2) scalar theory, ensuring that the Goldstone boson is massless in the broken symmetry phase [A. Pilaftsis and D. Teresi,

more »

... uc.Phys. B 874, 2 (2013), pp. 594--619]. We extend this by introducing a symmetry improved 3PIEA for O(N) theories, for which the basic variables are the 1-, 2- and 3-point correlation functions. This requires the imposition of a Ward identity involving the 3-point function. The method leads to an infinity of physically distinct schemes, though an analogue of d'Alembert's principle is used to single out a unique scheme. The standard equivalence hierarchy of nPIEA no longer holds with symmetry improvement and we investigate the difference between the symmetry improved 3PIEA and 2PIEA. We present renormalized equations of motion and counter-terms for 2 and 3 loop truncations of the effective action, leaving their numerical solution to future work. We solve the Hartree-Fock approximation and find that our method achieves a middle ground between the unimproved 2PIEA and PT methods. The phase transition predicted by our method is weakly first order and the Goldstone theorem is satisfied. We also show that, in contrast to PT, the symmetry improved 3PIEA at 2 loops does not predict the correct Higgs decay rate, but does at 3 loops. These results suggest that symmetry improvement should not be applied to nPIEA truncated to <n loops. We also show that symmetry improvement is compatible with the Coleman-Mermin-Wagner theorem, a check on the consistency of the formalism.