The persistence exponent θ for the global order parameter, M (t), of a system quenched from the disordered phase to its critical point describes the probability, p(t) ∼ t −θ , that M (t) does not change sign in the time interval t following the quench. We calculate θ to O(ǫ 2 ) for model A of critical dynamics (and to order ǫ for model C) and show that at this order M (t) is a non-Markov process. Consequently, θ is a new exponent. The calculation is performed by expanding around a Markov process, using a simplified version of the perturbation theory recently introduced by Sire [Phys. Rev. Lett. 77, 1420 (1996)].The 'persistence exponent', θ, which characterizes the decay of the probability that a stochastic variable exceeds a threshold value (typically its mean value) throughout a time interval, has attracted a great deal of recent interest [1][2][3][4][5][6][7][8][9][10][11]. One of the most surprising properties of this exponent is that its value is highly non-trivial even in simple systems. For example, θ is irrational for the q > 2 Potts model in one dimension [6] (where the fraction of spins that have not changed their state in the time t after a quench to T = 0 decays as t −θ ) and is apparently not a simple fraction for the diffusion equation [9,10] (where the fraction of space where the diffusion field has always exceeded its mean decays as t −θ ).A recent study of non-equilibrium model A critical dynamics, where a system coarsens at its critical point starting from a disordered initial condition, looked at the probability P (t 1 , t 2 ) that the global magnetization does not change sign during the interval t 1 < t < t 2 [11]. The persistence exponent for this system is defined by P (t 1 , t 2 ) ∼ (t 1 /t 2 ) θ in the limit t 2 /t 1 → ∞. Explicit results were obtained for the 1D Ising model, the n → ∞ limit of the O(n) model, and to order ǫ = 4 − d near dimension d = 4. For these systems it was found that the value of θ was related to the dynamic critical exponent z, the static critical exponent η, and 'nonequilibrium' exponent λ (which describes the decay of the autocorrelation with the initial condition, φ(x, t)φ(x, 0) ∼ t −λ/z ) by the scaling relation θz = λ − d + 1 − η/2. This relation may be derived from the assumption that the dynamics is Markovian, which is indeed the case for all of the cases considered in that paper.From a consideration of the structure of the diagrams which appear at order ǫ 2 (and higher order), however, it was argued that the dynamics of the global order parameter should not be Markovian to all orders, implying that the exponent θ does not obey exactly that 'Markovian scaling relation ' [11]. This means that θ is a new exponent. Monte-Carlo simulations in 2 dimensions indeed suggest weak violation of the Markov scaling relation [11].In this Rapid Communication we present an explicit calculation of the non-Markovian properties of the global order parameter. The nonequilibrium magnetizationmagnetization correlation function is calculated to order ǫ 2 , and this is then used ...