High Accuracy Numerical Methods for Thermally Perfect Gas Flows with Chemistry

Ronald Fedkiw, Barry Merriman, Stanley OsherJournal of Computational Physics 1997 · DOI

TL;DR

This paper presents a robust numerical framework for simulating chemically reacting gas flows, extending modern high-accuracy computational fluid dynamics (CFD) methods to handle the complexities of multi-species reactions. The authors demonstrate that by carefully formulating the governing equations and employing precise numerical techniques, accurate and stable simulations of phenomena like combustion can be achieved without resorting to simplifying approximations.

What Problem Does This Paper Address?

Simulating chemically reacting gas flows, such as those found in combustion engines or atmospheric phenomena, is a significant challenge in computational physics. These flows are governed by the compressible Navier-Stokes equations extended to include multiple chemical species and their reactions. This leads to a system of equations with two major difficulties:

  1. Stiff Source Terms: Chemical reactions can occur extremely rapidly, leading to "stiff" source terms in the equations that require specialized, computationally expensive time integration methods.
  2. Steep Spatial Gradients: The interplay of energetic reactions and compressible flow can create sharp fronts (like shocks or reaction zones) that are difficult for numerical methods to resolve accurately without generating spurious oscillations.

Traditional approaches often simplify these complexities, for instance, by approximating reaction rates or using simplified models for gas properties. These approximations can lead to inaccurate results or numerical instabilities. This paper aims to develop a framework that allows modern, high-accuracy numerical methods, typically used for non-reacting flows, to be applied effectively to these complex reacting systems.

Methods

The authors propose a multi-pronged approach:

  • Enthalpy-Based Formulation: They reformulate the governing equations using enthalpy as the primary thermodynamic variable. This form is advantageous for physical modeling and, crucially, allows for the analytical derivation of the system's characteristics (eigenvalues and eigenvectors), which are essential for many modern CFD schemes.
  • Time Splitting for Chemistry: To handle the stiff reaction terms efficiently, they employ a Strang time splitting scheme. This method separates the fluid dynamics (convection and diffusion) from the chemical reactions, allowing each part to be treated with the most appropriate numerical technique. A key contribution here is showing how to perform this split correctly, avoiding common but inaccurate practices like "freezing" the temperature during the reaction step. They demonstrate that temperature is an implicit function of other conserved variables during the reaction step, solvable via a simple root-finding procedure.
  • High-Accuracy Numerical Schemes:
    • For the convective terms (fluid transport), they use the Essentially Non-Oscillatory (ENO) finite difference method, known for its ability to capture steep gradients without oscillations.
  • For the diffusive terms (viscosity, heat conduction), they use standard central differencing.
  • For time integration, they combine a 3rd-order TVD Runge-Kutta scheme for the fluid dynamics and a 2nd-order accurate Strang splitting for the chemistry.
  • Accurate Characteristic Data: A significant part of the paper is dedicated to deriving exact analytical expressions for the eigenvalues and eigenvectors of the linearized convective flux Jacobian. They emphasize that using approximate expressions can lead to significant numerical errors and oscillations, as demonstrated in their examples.

Datasets

The paper does not use traditional "datasets" for training. Instead, it relies on:

  • CHEMKIN: A widely used chemical kinetics package is employed to provide thermodynamic and chemistry data (e.g., specific heats, heats of formation) for the gas species. This data is used to construct tables of thermodynamic properties as functions of temperature.
  • JANAF Thermochemical Tables: Referenced for standard thermodynamic constants.

The paper introduces its own benchmark problems for validation:

  • 1-D Reacting Shock Tube: A standard test case for reacting flows, modified to include combustion.
  • 2-D Combustor Simulation: A more complex problem simulating the injection and combustion of hydrogen in an oxygen-rich environment.

Key Insights and Ideas

  • The Importance of Exact Characteristics: The paper strongly argues that using approximate analytical expressions for the eigenvalues and eigenvectors of the flux Jacobian can lead to significant numerical errors and oscillations, even for seemingly small perturbations. This highlights the need for precise mathematical formulation in CFD.
  • Correct Time Splitting for Chemistry: The authors clarify that a proper time split for stiff reaction terms requires holding certain thermodynamic quantities constant (but not necessarily temperature) and that temperature is an implicit function of these quantities. This avoids common pitfalls of freezing temperature or introducing unnecessary ODEs for temperature.
  • Enthalpy Formulation Advantage: The enthalpy-based formulation of the energy equation simplifies the derivation of characteristics and is convenient for physical modeling, especially when dealing with gases that can be thermally perfect over certain temperature ranges.
  • Extensibility of Modern CFD: The work demonstrates that advanced, high-accuracy numerical methods developed for non-reacting flows can be successfully extended to handle the added complexity of chemically reacting flows, provided the underlying equations are carefully formulated and treated.

Key Results

The paper presents numerical results for both 1-D and 2-D problems, demonstrating the effectiveness of their approach:

  • Accurate Shock Tube Simulation: They show a 1-D reacting shock tube simulation where the reaction wave and shock interact, leading to a complex wave structure.
  • 2-D Combustor Simulation: The 2-D combustor simulation successfully captures phenomena like vortex shedding, ignition, mixing of reactants, and the propagation of combustion fronts. They visualize the flow field, reactant concentrations, and product formation.
  • Comparison of Assumptions: They show comparative results for "thermally perfect" versus "calorically perfect" gas assumptions, highlighting how the latter can lead to premature reactions and non-physical results, underscoring the importance of accurate thermodynamic modeling.
  • Robustness to Perturbations: Their "separating box" example illustrates that even small perturbations to the Jacobian matrix (mimicking approximations) can lead to significant oscillations when using ENO and TVD Runge-Kutta schemes, reinforcing the need for exact characteristic data.

Limitations and Future Work

  • Computational Cost: While their method is accurate, the paper acknowledges that handling stiff source terms and using high-order schemes can be computationally expensive. The time splitting helps, but more advanced techniques for stiff ODEs might be needed for even stiffer problems.
  • Specific Chemical Mechanisms: The paper uses a specific chemical mechanism for its examples. Extending the framework to arbitrary, highly complex chemical kinetics would require careful implementation and validation.
  • Future Work: The authors suggest that their framework is applicable to a wide range of modern numerical schemes and can be extended to more complex geometries and flow regimes. The successful application to 2-D problems suggests potential for 3-D simulations. The ability to model both thermally and calorically perfect gases within the same framework opens avenues for studying the impact of these assumptions on combustion phenomena.

Summary updated

BibTeX
@article{fedkiw1997high,
  author = {Ronald Fedkiw and Barry Merriman and Stanley Osher},
  title = {{High Accuracy Numerical Methods for Thermally Perfect Gas Flows with Chemistry}},
  journal = {Journal of Computational Physics},
  year = {1997},
  doi = {10.1006/jcph.1996.5622},
}
Figure