Stochastic algorithms for the analysis of numerical flame simulations (Q703754)

The basic idea for the diagnostic algorithm is to monitor the path of a tagged atom through the flow domain, including the history of which molecules transport them. The data to be analysed includes the distribution of all species in the continuum solution, as well the temperature and velocity fields. This data is obtained through numerical simulation of the reacting Navier-Stokes equations. They analysis is independent of whether the solution is obtained which compressible or low MACH number formulation but does depend on the details of the transport and kinetics models employed. They assumed that the numerical simulations that they wish to analyse are based on a mixture-averaged model for transport and a collection of fundamental reactions specified as a CHEMKIN compatible kinetic database. With these assumptions the \(k\)-th species satisfies the equation: \[ \frac{\partial\rho Y_k }{\partial t}+\nabla \cdot (u\rho Y)=\rho \frac{DY_k }{Dt}=\nabla \cdot \rho D_k \nabla Y_k +\rho \omega _k,\tag{1} \] where: \(Y_k \) is the mass fraction, \(\rho \) is fluid density, \(u\) is advection velocity, and \(D_k \)and \(\omega _k \) are the mixture-averaged diffusion coefficient and chemical production rate, respectively. They are denote the atom which they are tracking by \(A.\) For the chemical system being modelled there is a subset of the molecules: \(M_1 ,M_2 ,\dots,M_k \) that contain atoms of the type \(A\) as one of their constituents. At each time the atoms of the type \(A\) as one of their constituents. At each time, the atom is thus in one of the element \(M_k \) which we denote by \(A\in M_k \). They interpreted the species equation from the perspective of an atom \(A\in M_k \), they obtained the stochastic differential equation for the fate in the flow field: \[ dx_A =u(x{ }_A,t)dt+dW_{k(t)} (x{ }_A,t)+dR_{k:{k}'} (x{ }_A,t),\tag{2} \] where the terms on the right hand side represent advection, diffusion and reaction; \(u(x{ }_A,t)\) - is the velocity obtained from the continuum solution, \(dW_{k(t)} \) - represents a generalized Brownian motion with properties chosen to model the diffusion of the \(M_k \) molecules, \(dR_{k:{k}'} \) - denotes a ``scattering'' of \(A\) from \(M_k \) into the set \(M_{{k}'} \) as the result of reaction. After \(M_k \) reacts, the dynamics of the particle are determined by the motion of the \({k}'\) species. They integrate Eg. (2) over a time interval \(\Delta t\) using a time explicit split approach that treats each term independently. They approach to modelling the chemical reaction in Eg. (2) is based on a similar type of discrete time approximation to the underlying stochastic process. They introduced a time interval \(\Delta t_c \) over which they wish to model the chemistry. They noted that the chemical time scale depends on the particular molecule which they considered and its local environment. Thus, the computation of \(\Delta t_c \) is based on local considerations. To analyse a particular feature of a simulation they identified the type of atom they want to follow the initial molecule of which it is a constituent and the region in space and time when they wanted to begin the trajectories. This data is used to initialize an ensemble of particles that are evolved by the stochastic algorithm described above. At first they identified \(\Delta t\) for the transport calculation. Then they computed the deterministic advection of the particle followed by the random walk to represent diffusion. Then they choose \(\Delta t_c \) and advance the chemistry, subcycling as necessary until they reach \(\Delta t\). For the time dependent flows, they have approximated the data for a given time step as piecewise constant over the length of the time step of the fluid dynamics algorithm. They assumed that the data obtained from the fluid solution at time \(t^n\) is valid until time \(t^{n+1}\) which is the time at which the next fluid time step is available. They adjust \(\Delta t\) for the stochastic particle algorithm so that the diagnostic algorithm reaches \(t^{n+1}\) exactly and they computed new properties to continue the evolution. Then they interngate the collected trajectories to analyse the behaviour of the flow. This method allow them to follow a several thousand trajectories in just a few hour. They demonstrated in computational results the behaviour of the above method and illustrated the relationship between the particle trajectories and the continuum solution. They considered some examples: (1) premixed flames (2) laminar diffusion flames. Strictly speaking they considered the modulation of carbon chemistry in premixed flame interacting with a vortex and the role of cyano chemistry on NO\(_{x}\) formation in a methane diffusion flame. In both eases they demonstrated that the stochastic particle diagnostics allow their to pose complex analysis equations in a straightforward manner and provided interesting insights into the flow. Roughly speaking they demonstrated that their method can recover properties of the continuum solution and provide a mechanism for diagnosing the behaviour of complex reacting flows.