The GFDE are used in many particle-based chemical simulations codes. The sampling of the Brownian propagators is therefore of great interest for such codes. The GFDE for the ABCD reactions are quite complex; however, it is possible to use the relatively simple algorithms presented in this paper to simulate the evolution in time and space of pairs of particles. In general, the performance of our algorithm is similar or better than that of table methods. Table methods usually necessitate an extensive set-up to calculate and hold the pre-calculated data, which may be very time and memory consuming, especially when the table has more than one dimension [33] and, as it will be the case in our future work, when large systems comprising many types of particles with different reaction channels are simulated.
This method might be useful for particle-based event-driven simulation schemes of the FPKMC (first-passage kinetic Monte Carlo) [13] and GFRD (Green's functions reaction dynamics) [8]. For example, in the GFRD method, simple geometric domains such as spheres are put around at most two particles to shield them from the influence of other particles. The pairs of particles are then propagated locally in an exact manner by basically sampling times and associated positions up to the time up to which any other particle outside the domain is guaranteed not to enter it. By propagating the domains subsequently, an event-driven particle-based simulation can be set up that is both exact and efficient, using exact Green's functions in order to skip other diffusion events. Such a scheme naturally gets rid of the problems associated with the possible interference of other particles at long times. One of the still pending problems of GFRD-like methods is the lack of Green's functions that correctly incorporate the back-reaction; for that reason, product particles that dissociate have to be put in contact (from where they reacted), which diminishes the performance of the scheme. This scheme would highly benefit from having exact Green's functions and associated sampling prescriptions that incorporate both the forward and backward reaction, such as the one described in the paper.
In radiation chemistry, the IRT method has been used to calculate the yields of the radiolytic species in solutions [25], [26], [27], [28] and [29] and also in chemical dosimeters [30] and [31]. The IRT method is based on the survival probability of the pairs of particles in the system, and the competition between reactions are taken into consideration by sorting the sampled reaction time. Although the purpose of this article is not the validation of the IRT method, it is interesting to note that our simulation results are in excellent agreement with those predicted by IRT. In fact, to our knowledge, the comparison of the predictions of the analytical Green's functions with the IRT results for 2-particle systems with reversible diffusion-influenced reactions has not been done before. Even if the simulation results obtained by the IRT method fail to yield the proper asymptotic long time dependence, the agreement with the GFDE is excellent at short times, which is of interest for radiation chemistry codes.
Because the algorithms are simple and use only a few kilobytes of memory, they can probably be used on a general-purpose graphic processing unit (GPGPU). A GPGPU is a computing device operating as a co-processor to the main central processing unit (CPU). GPGPUs comprise up to several hundred cores and have their own memory. They are used to compute functions which are executed a large number of times, but independently on different data. Therefore, the algorithms could be implemented on a GPGPU to simulate a chemical system comprising different types of particles. This work should be useful for chemistry codes that are based on this approach to study biochemical interactions occurring in cells, which may eventually be included in event-based models of space radiation risks.