Introduction
Rayleigh–Bénard convection (RBC)—the buoyancy-driven flow of fluid heated from below—serves as a fundamental model for diverse phenomena, from atmospheric circulation to mantle convection. A central challenge in studying RBC is characterizing its "solution landscape": the set of all possible steady-state flow patterns for a given set of parameters. As the Rayleigh number ($Ra$), representing the thermal driving force, increases, the system undergoes a series of bifurcations, transitioning from a quiescent state to complex, multi-cellular rolls and eventually to turbulence.
Traditional numerical continuation methods are effective at tracking a single solution branch as parameters vary. However, they often fail to detect "disconnected branches"—solution states that exist in isolation and cannot be reached through continuous deformation of a known state. This paper utilizes a novel technique, deflated continuation, to systematically uncover these hidden states in 2D RBC.
Numerical Methodology: Deflated Continuation
The authors solve the steady-state Oberbeck–Boussinesq equations in a 2D square domain with no-slip boundaries using the Firedrake finite element library. To overcome the limitations of standard solvers, they employ deflated continuation. This algorithm works by finding a solution using Newton's method and then applying a "deflation operator" to the governing equations. This operator effectively removes the discovered solution from the search space, forcing the solver to converge to a different, previously unknown steady state.
By iteratively deflating known solutions and using the eigenmodes of the trivial (no-flow) state as initial guesses, the researchers can map out a much larger portion of the bifurcation diagram than previously possible.
Mapping the Solution Landscape
The study identifies the first ten primary bifurcations where the quiescent fluid state becomes unstable and gives way to convective rolls. These critical Rayleigh numbers range from $Ra_c^{(1)} \approx 2,586$ to $Ra_c^{(10)} \approx 94,543$.
At a higher Rayleigh number of $Ra = 10^5$, the algorithm discovered 43 distinct solution branches. When accounting for the physical symmetries of the system (mirror and Boussinesq symmetries), this number expands to 129 possible steady states. Many of these branches are disconnected from the primary branch, forming an "archipelago" of solutions that would be invisible to standard methods.
Heat Transfer Efficiency and Flow Reversals
A key metric for evaluating these convective states is the Nusselt number ($Nu$), which measures the efficiency of heat transfer relative to pure conduction. The bifurcation diagram illustrates how $Nu$ evolves with $Ra$ for various discovered branches.
One of the most significant findings is the discovery of Branch 11. This branch exhibits an S-shaped curve characterized by hysteresis, meaning the system can support multiple stable states at the same $Ra$ and may "jump" between them. The flow patterns associated with Branch 11 closely resemble the large-scale circulations observed in turbulent flow reversal experiments. These disconnected branches likely serve as the "skeleton" of phase space, guiding the system's behavior even when the steady states themselves are unstable.
Conclusion and Future Outlook
By successfully applying deflated continuation to 2D Rayleigh–Bénard convection, the authors have revealed a far more complex solution landscape than previously understood. The discovery of numerous disconnected branches provides a structural explanation for complex phenomena like flow reversals in turbulent systems.
Future work will involve extending these methods to three-dimensional geometries and higher Rayleigh numbers. While 2D models provide valuable insights, 3D convection introduces additional degrees of freedom and complex structures, such as twisting rolls, that are essential for a complete understanding of high-$Ra$ turbulence.
Summary updated