Fitting equation of state parameters in parallel computers

Abstract

-This work compares two strategies to fit parameters of equations of state in parallel computers, emphasizing solutions that require few changes to existing sequential programs. One strategy uses the conventional Nelder-Mead algorithm coupled with parallel objective function evaluation (SSPO). The other strategy uses a parallel Nelder-Mead algorithm coupled with sequential objective function evaluation (PSSO). The PSSO strategy, which executes parallel one-dimensional searches during each iteration, is simpler to implement and converged to parameter sets with objective functions smaller than those obtained by the SSPO strategy. The SSPO strategy produced speedups consistent with the number of processes used and is more suitable when many processors are available. Both strategies are potentially useful and choosing between them is a matter of convenience, depending on the problem at hand. With parallel computers increasingly available, the easy implementation and convenience of these two strategies should appeal to developers and users of thermodynamic models.

Modeling of Gas Solubility in Aqueous Electrolyte Solutions with the eSAFT-VR Mie Equation of State

“…The optimum parameters, presented in Table 8, are determined in the complete database using an ionic parametrization framework according to the step-wise methodology introduced by Castier et al in the framework of Q-Electrolattice EoS. 87,88 The objective function of Eq. 29, has been fitted by a least squares minimization tool:…”

“…We initially used the step-wise methodology of Castier at al. 87,88 to fit the parameters between methane and ions. Afterwards, we allowed for all ionic parameters to be fitted simultaneously.…”

Utilization of parallel computing in chemical engineering

“…Nonetheless, many legacy sequential codes exist in languages such as Fortran and C#. Their adaptation for maximum performance in parallel computers may require extensive reprogramming but substantial performance gains are possible in certain applications with only a few changes in the existing codes (Castier et al, 2014). This case study shows that EOS parameter fitting is one of such applications comparing two types of software utilizing the Simplex algorithm in parallel computers, applicable both in single multiprocessor PCs and in cluster supercomputers.…”

“…Data parallelism is a form of computing parallelization across multiple processors in parallel computing environments, focused on data distribution across different parallel computing nodes. Not long ago, parallel computers were associated with fairly large machines, most of them running Linux, designed for special computations requiring extreme performance, such as quantum mechanical calculations, computational fluid dynamics, molecular simulations, and chemical process optimization, among other applications relevant to chemical industry (Castier et al, 2014). The first case study presented refers on a global optimization method exploiting parallel computers, used to fit the equation of state (EOS) para meters.…”