Accuracy of order-N density-functional theory calculations on DNA systems using CONQUEST

In preparation for large-scale modelling of DNA systems using the linear-scaling density-functional theory methods implemented in the CONQUEST code, we investigate the effect of the approximations used in the code for DNA test systems. The results of CONQUEST calculations on single DNA bases and on hydrogen-bonded base pairs are compared with experimental ones and with the results from other codes in order to gauge the errors incurred by the use of pseudo-atomic orbital (PAO) basis sets and to assess the accuracy of different density functionals. We then use calculations on hydrated and unhydrated DNA systems containing up to similar to 3400 atoms to test the effect of the spatial cut-off R-L required to achieve linear-scaling operation in CONQUEST. We find that PAO basis sets of double-zeta plus polarization quality give satisfactory results, and that generalized gradient functionals reproduce well the energetics of hydrogen bonding between base pairs. The linear-scaling errors can readily be rendered negligible with moderate values of R-L.