The diffusion of heat can also be imagined as the random motion of particles, agitated by heat [9, 10], and random walks and Brownian motion are closely related to the heat equation, [11, 12, 13, 14]. In addition, due to the fact that parabolic differential equations can be solved by using stochastic processes [15] and together with the idea of diffusion approximation through Markov processes [16, 17, 18] and the ideas in [19, 20], we can modell heat diffusion as particles which are driven by a stochastic process, specifically an continuous-time infinite-state Markov chain. Each particle randomly jumps between nodes of a lattice we impose on a given object. Thinking of three dimensional objects, it is more natural to use a lattice of unequally spaced nodes on an arbitrary object, due to curvature. When simulating heat particles, we may also want to place more nodes in a specific region, which also leads to irregularity. Hence, our framework focusses on such irregular lattices.
The rate at which a heat particle jumps to neighboring nodes, the transition rate, is determined by the distances to all neighboring nodes and the distances between neighboring nodes themselves. Reference [21] used this idea to approximate general jump-diffusion processes. These transition rates allow us to completely characterize the infinitesimal generator Q of the Markov chain. Q acts as an operator which drives the diffusion of particles.
In continuous space, where no lattice is used, path integrals describe the idea of jumping from one point to another with a specific probability, based on all possible paths between those points [22, 23]. To find the right transition rates, such that our framework approximates the heat diffusion, we evaluate the variance and mean of the distribution of heat particles. The desired distribution in one dimension is a Gaussian distribution with variance