Topological quantum error correction codes are known to be able to tolerate arbitrary local errors given sufficient qubits. This includes correlated errors involving many local qubits. In this work, we quantify this level of tolerance, numerically studying the effects of many-qubit errors on the performance of the surface code. We find that if increasingly large area errors are at least moderately exponentially suppressed, arbitrarily reliable quantum computation can still be achieved with practical overhead. We furthermore quantify the effect of non-local two-qubit correlated errors, which would be expected in arrays of qubits coupled by a polynomially decaying interaction, and when using many-qubit coupling devices. We surprisingly find that the surface code is very robust to this class of errors, despite a provable lack of a threshold error rate when such errors are present.Many different approaches to achieving reliable quantum computation are under investigation [1][2][3][4][5]. The current most practical known approach, the Kitaev surface code [6,7], calls for a 2-D array of qubits with nearest neighbor interactions and a universal set of quantum gates with error rates below an approximate threshold of 1% [8][9][10]. Superconducting qubits with error rates at the surface code threshold now exist [11].There is extensive prior work showing the existence of a threshold error rate when arbitrary quantum error correction codes are subjected to a wide variety of noise models, including algebraically decaying two-body correlated noise [12], Gaussian non-Markovian noise [13], and arbitrarily many-body correlated noise [14]. In this work we focus on the simulated performance of the surface code below threshold.To date, when the surface code has been simulated, quantum gates have only had the potential to introduce errors on the qubits they manipulated directly. In reality, manipulating any given qubit may disturb the state of a large number of surrounding qubits. Not all types of disturbance are particularly dangerous. Small random or systematic rotations of surrounding qubits lead only to independent random errors. Only correlated manyqubit errors deserve specific attention. This distinction is discussed in detail in Section I. In this work we present a detailed study of precisely how well the surface code can handle this class of correlated errors.Another important class of errors that has not received attention to date is those that would arise in an array of qubits interacting directly with one another via a polynomially decaying interaction such as the Coulomb or magnetic dipole interaction, or via a device coupling to many qubits. Pairs of qubits initially antiparallel can both flip without changing the energy of the total system. Two-qubit errors can therefore appear on widely separated qubits. We also present a detailed study of this class of correlated errors.The discussion is organized as follows. In Section I, the meaning of independent and correlated errors is dis-cussed in detail. In Section II, the sur...