TL;DR
This paper introduces a novel adaptive two-grid preconditioner for simulating gas transport in fractured shale formations. By combining an efficient implicit-explicit time-stepping scheme with a spectral coarse-grid approximation, the method achieves robust and contrast-independent convergence, crucial for tackling the complex multiscale physics of these reservoirs.
What Problem Does This Paper Address?
Simulating gas transport in shale formations is a significant challenge due to the inherent multiscale nature of the problem, involving both the tight shale matrix and highly conductive fractures. These formations are also highly heterogeneous, leading to extreme contrasts in material properties (like permeability) between the matrix and fractures. Traditional numerical methods struggle with such problems, often requiring prohibitively fine grids or suffering from slow convergence that depends heavily on the contrast in properties. This paper aims to develop an efficient and robust numerical method that can accurately capture these complex transport phenomena, particularly focusing on developing a fast iterative solver through an advanced preconditioning technique.
Methods
The authors propose a two-pronged approach:
- Implicit-Explicit Time-Stepping Scheme: They employ a linearly stable implicit-explicit (ImEx) scheme for time integration. This scheme separates the stiff, linear components of the problem (treated implicitly for stability) from the nonlinear components (treated explicitly). A key innovation is using a fixed linear operator for the implicit part throughout the simulation, which significantly reduces computational cost by allowing the preconditioner to be constructed only once.
- Adaptive Two-Grid Preconditioner with Spectral Coarse Space: For solving the resulting linear systems at each time step, they develop a two-grid preconditioner. This preconditioner utilizes a standard pointwise smoother on the fine grid and, crucially, an adaptive multiscale spectral coarse space. This coarse space is constructed using local generalized eigenvalue problems, where the number of basis functions for each local subdomain is determined adaptively based on a threshold of eigenvalues. This approach aims to capture the essential physics of the fractured media efficiently, leading to contrast-independent convergence. The method is inspired by Generalized Multiscale Finite Element Methods (GMsFEM).
Datasets
Not discussed in the paper. The authors present numerical experiments on synthetic geometries with varying numbers of fractures and different matrix properties.
Key Insights and Ideas
- Fixed Implicit Operator for Efficiency: The decision to use a fixed linear operator for the implicit part of the ImEx scheme is a significant practical improvement. It decouples the preconditioner construction from the time-stepping, allowing for offline computation and reuse, which is critical for complex multiscale problems.
- Adaptive Spectral Coarse Space is Key: The core innovation lies in the adaptive construction of the coarse space. By solving local spectral problems and adaptively selecting basis functions based on eigenvalues, the method can efficiently capture the complex interactions between fractures and the porous matrix, even with high permeability contrasts.
- Contrast-Independent Convergence: The primary goal and achievement is demonstrating that the proposed preconditioner leads to iterative solvers whose convergence rate is largely independent of the permeability contrast between fractures and the matrix. This is a major hurdle in simulating fractured porous media.
- Multiscale Nature Captured by Basis Functions: The multiscale basis functions derived from local spectral problems can be interpreted as representing different "continua" within each local subdomain, effectively capturing the multiscale physics without explicit coupling terms between them.
Key Results
- The proposed method demonstrates robust convergence for shale gas transport simulations in fractured porous media.
- Numerical results for test cases with 30 and 160 fractures, and varying contrasts (10³, 10⁶, 10⁹), show that the iterative solver converges in a small, fixed number of iterations, largely independent of the contrast.
- The adaptive approach successfully identifies the necessary number of local basis functions, leading to efficient coarse-grid representations and fast convergence.
- The linearly implicit scheme shows good stability and achieves first-order convergence in time.
Limitations and Future Work
- Parameter Tuning: The adaptive threshold parameter ($\delta\lambda$) for selecting basis functions requires tuning, although the authors suggest that a single choice can be robust across various problems.
- Online Adaptivity: The current adaptive approach is performed in an "offline" stage. Future work could explore "online" adaptivity, where the coarse space is enriched dynamically based on a-posteriori error estimates during the simulation.
- Higher-Order Schemes: The current time-stepping scheme is first-order. Exploring higher-order schemes for the time derivative is mentioned as a topic for future work.
- Dynamic Transfer Operators: For more complex nonlinear problems where the physics evolve dynamically, the transfer operators might need to be updated dynamically.
Summary updated