This leaves us with the modified barrier as the only fitting parameter. In the numerical calculations we varied the modified barrier around the standard diffusion barrier of a vacancy in Cu(001) (0.426 eV [15]), plus an additional 25 meV repulsive interaction. The jump length distributions that we obtained for each of the modified barriers were compared with the measured jump length distribution. The results of this comparison are given in Fig. 7. It shows the normalized χ2 values that were obtained for each of the calculated distributions. The amount of data in the measured jump length distribution did limit to some extent the χ2 test, however, guiding values were obtained. From Fig. 7 we observe that the minimum χ2 values this to lower values of Eeff/kBT with decreasing terrace size. This is not unexpected since a narrow terrace will automatically shorten the random walk of the vacancy and lead to fewer exchanges of the vacancy and the impurity atom. Or, in other words, to obtain an equally good fit on a larger terrace, the repulsive interaction between the vacancy and the impurity atom needs to be increased to reduce the number of exchanges between them. For the experimentally observed terrace width of ≈1000 Å, we find a best fit repulsive