Galerkin reduced order model for two-dimensional Rayleigh-B'enard convection

arXiv:2504.02961 · physics.flu-dyn · Submitted 2025-04-03 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "Galerkin reduced order model for two-dimensional Rayleigh-B'enard convection".

Mira: Here's the gist of the paper. Galerkin projection is used to build reduced-order models (ROM) for two-dimensional Rayleigh-Bénard (RB) convection with no-slip walls The coupled approach shows better agreement with DNS in terms of mean vertical profiles and Nusselt number scaling These ROMs do not require closure models and remain numerically stable The coupled approach shows better agreement with DNS in terms of mean vertical profiles and Nusselt number scaling Model Development Strategies The study compares an uncoupled projection approach that uses separate orthonormal bases for velocity and temperature with a coupled formalism where the equations are projected onto a single basis combining velocity and temperature components The orthogonal bases for modal projection are obtained as the eigenfunctions of the controllability Gramian of the linearized RB equations, eliminating the need for DNS snapshot databases required by traditional POD-based approaches Two projection strategies are systematically compared: an uncoupled approach with independent modal bases for temperature and velocity, and a coupled approach with one single orthonormal basis that includes velocity and temperature components Validation Against Direct Numerical Simulations ROM simulations are compared to Direct Numerical Simulations (DNS) for model validation The comparison is performed in terms of mean vertical profiles, total heat flux, flow structures, bifurcation diagrams and energy spectra For the coupled ROMs C96–C384, a very good agreement with DNS results is obtained up to R = 23.4 (Ra = 40 000) The comparison between the two projection approaches reveals that coupled models show better performance than uncoupled ROMs giving more accurate results for the same truncation level Dynamical Regime Characterization A detailed bifurcation analysis is conducted at P r = 10 using Poincar´e sections and Lyapunov exponents to precisely identify the transitions between periodic, quasiperiodic, and chaotic states with significant reductions of computational cost The transition to chaos in model C192 is shown to occur at approximately R ≃ 87 The combination of Poincar´e sections and Lyapunov exponent analysis enabled precise identification of the transition to chaos at R ≃ 87 in model C192 Spectral Characteristics Analysis The power spectra comparison between DNS and ROMs is established using mode c48 for model C192 and mode a14 for model U192 Model C192 captures the amplitude and frequency of the three dominant peaks in the DNS spectrum at f ≃ 0.05, 0.25, and 0.45 The ROM spectrum exhibits two spurious peaks that do not correspond to frequency components present in the DNS signal Overall, these results indicate that Galerkin ROMs reproduce the main features of the nonlinear system dynamics and are promising candidates to serve as surrogates for the high-dimensional system in applications where a low computational cost is required The coupled projection strategy has been found to outperform the more conventional uncoupled approach suggesting that coupled modal bases better capture the nonlinear interactions between large scale structures These ROMs have proven to be numerically stable when integrated over large characteristic times, T = 104, at high Rayleigh numbers, R = 106 The combination of Poincar´e sections and Lyapunov exponent analysis enabled precise identification of the transition to chaos at R ≃ 87 in model C192 Appendix Details The derivation of the adjoint state matrix A+ is obtained by equating corresponding terms on both sides of the equation The expression for the inner product in the state space is given by eq.

Kai: (A3) with a weight matrix for the inner product defined as M = -∇2 0 0 γ2 The adjoint control matrix B+ is obtained by arranging these results into matrix form to yield B+ =

∂y 0 −ikx 0 0 γ2: The final product BB+ gives BB+ =

1 0 0 γ2: Key Findings on Scaling The DNS show that in 2D RB convection with no-slip walls the Nusselt number varies approximately as Nu ∝ R1/4 for 1 < R < 103 Both the uncoupled and coupled ROMs capture the scaling of the Nusselt number with the reduced Rayleigh number R over two orders of magnitude This indicates that they could be reliably used to predict the total heat flux across the fluid layer in heat transfer applications involving RB convection The choice of normalization is key for the stability of coupled ROMs following Soucasse et al.

34: With this scaling, the magnitude of dimensionless velocities remains nearly constant with Ra and of the same order as the temperature perturbations, u ∼ θ′ The DNS show that in 2D RB convection with no-slip walls the Nusselt number varies approximately as Nu ∝ R1/4 for 1 < R < 103 Numerical Stability and Robustness Both coupled and uncoupled ROMs maintain numerical stability up to R = Ra/Rac = 106 for integration times of T = 104 This numerical robustness is achieved without any explicit closure model, in contrast to previous POD-Galerkin studies for buoyancy-driven convection in cavities that required additional dissipation terms to prevent numerical blow-up The Galerkin ROMs are numerically stable and do not require closure models The system is integrated using an Adams/BDF method with automatic stiffness detection

15: Application Potential The capabilities of these models are exploited to conduct a detailed bifurcation analysis at P r = 10 using Poincar´e sections and Lyapunov exponents, precisely identifying the transitions between periodic, quasiperiodic, and chaotic states with significant reductions of computational cost These ROMs can serve as DNS surrogates over a wide range of Rayleigh numbers setting the stage for future applications in state observation, control and optimization The detailed bifurcation analysis presented below demonstrates that these ROMs reproduce the same route to chaos observed in DNS—periodic, quasiperiodic, and chaotic dynamics—as a function of the Rayleigh number These ROMs are promising candidates to serve as surrogates for the high-dimensional system in applications where a low computational cost is required The detailed bifurcation analysis presented below demonstrates that these ROMs reproduce the same route to chaos observed in DNS—periodic, quasiperiodic, and chaotic dynamics—as a function of the Rayleigh number References

1: E.

Kai: Bodenschatz, W.

Paper summary: Kai: So we just covered how the paper sets up this comparison between uncoupled and coupled projections for building these ROMs. The core thesis here is that using a single combined basis actually gives better results when you look at the mean vertical profiles and how the Nusselt number scales.

Mira: Exactly, it claims that this coupled approach is superior because it captures those large-scale flow structures more accurately, which is what matters for real applications.

Lev: What's interesting about the method they used to find those bases is that instead of needing huge databases from direct numerical simulations, they get the orthonormal bases from the eigenfunctions of the controllability Gramian of the linearized equations.

Kai: So they bypassed needing tons of DNS snapshot data, which is a big win for making these models practical for things like real-time control where you can't run a full simulation every time.

Mira: And this method works for both approaches, uncoupled and coupled, generating ROMs with different numbers of modes to check their domain of validity across a wide range of Rayleigh numbers.

Lev: The paper tests these against DNS over a wide range, and they show that the coupled models are better performing than the uncoupled ones for the same truncation level.

Kai: It really emphasizes that these ROMs can serve as good surrogates for full-order models because they accurately reproduce key physics like dominant flow structures and integral quantities, which is what you need for heat transfer applications.

Conclusion: Mira: Thinking about the title of "Galerkin reduced order model for two-dimensional Rayleigh-B'enard convection", this paper is essentially showing that you can build a manageable mathematical description of complex fluid motion without needing the full computational power.

Kai: It’s about taking a high-dimensional system, like this Rayleigh-Bénard convection, and reducing it to something computationally cheap, but the key finding is that coupling those variables helps you keep the physics right.

Lev: The implication for us in error correction research is that if we can build models like these that accurately capture dynamics across different states—periodic, quasiperiodic, chaotic—it opens up possibilities for state observers and real-time control of physical systems.

Mira: Right. And they show this model reproduces the same route to chaos you see in direct simulations as the Rayleigh number changes, which means we can use this model to predict how a system will behave before we actually run a massive simulation on it.

Kai: So, for anyone who just listens, the point is that these coupled ROMs are numerically stable over long integration times at high Rayleigh numbers, and they provide a way to get reliable predictions for heat flux scaling across two orders of magnitude in Rayleigh number.

Lev: The limitation they flag is that while two dimensions capture the essential dynamics of pattern formation, three dimensional effects can still be important if you need that extra detail.

Mira: That's the caveat—the model is excellent for capturing the core 2D physics, but if your application demands full three dee fidelity, you're still looking at a bigger challenge <ref:2504.02961#pg3>.

Enrique Flores-Montoya, André V. G. Cavalieri

Aerothermals department, ITP Aero · Divisγao de Engenharia Aeroespacial, Instituto Tecnológico de Aeronática

physics.flu-dyn

Submitted: 2025-04-03

Updated: 2026-11-09

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 79/100

The gist: Galerkin projection is used to build reduced-order models (ROM) for two-dimensional Rayleigh-Bénard (RB) convection with no-slip walls The coupled approach shows better agreement with DNS in terms…

Key concepts

Galerkin Projection
This technique is used to create simplified mathematical models (ROMs) by projecting the full, complex governing equations onto a smaller set of chosen basis functions. In this study, it was applied using a coupled approach combining velocity and temperature components into a single basis.
Coupled ROM Approach
This strategy involves projecting the velocity and temperature equations onto one single orthonormal basis that includes both components. This method was found to be superior to uncoupled approaches because it better captures the nonlinear interactions between large-scale structures in the convection system.
Direct Numerical Simulation (DNS)
DNS is a high-fidelity computational method used as a benchmark for validation. It solves the full, complex equations of 2D Rayleigh-Bénard convection without any model reduction. The ROMs were compared against DNS results for mean profiles and scaling to prove their accuracy.
Transition to Chaos
This refers to the point where the fluid flow dynamics shift from predictable, repeating patterns (periodic) or slightly irregular patterns (quasiperiodic) into a highly complex, unpredictable state (chaotic). The study precisely identified this transition point in model C192 using tools like Poincaré sections and Lyapunov exponents.

Terminology

Summary

Galerkin projection is used to build reduced-order models (ROM) for two-dimensional Rayleigh-Bénard (RB) convection with no-slip walls The coupled approach shows better agreement with DNS in terms of mean vertical profiles and Nusselt number scaling These ROMs do not require closure models and remain numerically stable The coupled approach shows better agreement with DNS in terms of mean vertical profiles and Nusselt number scaling

Model Development Strategies

The study compares an uncoupled projection approach that uses separate orthonormal bases for velocity and temperature with a coupled formalism where the equations are projected onto a single basis combining velocity and temperature components The orthogonal bases for modal projection are obtained as the eigenfunctions of the controllability Gramian of the linearized RB equations, eliminating the need for DNS snapshot databases required by traditional POD-based approaches Two projection strategies are systematically compared: an uncoupled approach with independent modal bases for temperature and velocity, and a coupled approach with one single orthonormal basis that includes velocity and temperature components

Validation Against Direct Numerical Simulations

ROM simulations are compared to Direct Numerical Simulations (DNS) for model validation The comparison is performed in terms of mean vertical profiles, total heat flux, flow structures, bifurcation diagrams and energy spectra For the coupled ROMs C96–C384, a very good agreement with DNS results is obtained up to R = 23.4 (Ra = 40 000) The comparison between the two projection approaches reveals that coupled models show better performance than uncoupled ROMs giving more accurate results for the same truncation level

Dynamical Regime Characterization

A detailed bifurcation analysis is conducted at P r = 10 using Poincar´e sections and Lyapunov exponents to precisely identify the transitions between periodic, quasiperiodic, and chaotic states with significant reductions of computational cost The transition to chaos in model C192 is shown to occur at approximately R ≃ 87 The combination of Poincar´e sections and Lyapunov exponent analysis enabled precise identification of the transition to chaos at R ≃ 87 in model C192

Spectral Characteristics Analysis

The power spectra comparison between DNS and ROMs is established using mode c48 for model C192 and mode a14 for model U192 Model C192 captures the amplitude and frequency of the three dominant peaks in the DNS spectrum at f ≃ 0.05, 0.25, and 0.45 The ROM spectrum exhibits two spurious peaks that do not correspond to frequency components present in the DNS signal

Conclusion on Model Utility

Overall, these results indicate that Galerkin ROMs reproduce the main features of the nonlinear system dynamics and are promising candidates to serve as surrogates for the high-dimensional system in applications where a low computational cost is required The coupled projection strategy has been found to outperform the more conventional uncoupled approach suggesting that coupled modal bases better capture the nonlinear interactions between large scale structures These ROMs have proven to be numerically stable when integrated over large characteristic times, T = 104, at high Rayleigh numbers, R = 106 The combination of Poincar´e sections and Lyapunov exponent analysis enabled precise identification of the transition to chaos at R ≃ 87 in model C192

Appendix Details

The derivation of the adjoint state matrix A+ is obtained by equating corresponding terms on both sides of the equation The expression for the inner product in the state space is given by eq. (A3) with a weight matrix for the inner product defined as M = -∇2 0 0 γ2 The adjoint control matrix B+ is obtained by arranging these results into matrix form to yield B+ = [∂y 0 −ikx 0 0 γ2] The final product BB+ gives BB+ = [1 0 0 γ2]

Key Findings on Scaling

The DNS show that in 2D RB convection with no-slip walls the Nusselt number varies approximately as Nu ∝ R1/4 for 1 < R < 103 Both the uncoupled and coupled ROMs capture the scaling of the Nusselt number with the reduced Rayleigh number R over two orders of magnitude This indicates that they could be reliably used to predict the total heat flux across the fluid layer in heat transfer applications involving RB convection The choice of normalization is key for the stability of coupled ROMs following Soucasse et al. [34] With this scaling, the magnitude of dimensionless velocities remains nearly constant with Ra and of the same order as the temperature perturbations, u ∼ θ′ The DNS show that in 2D RB convection with no-slip walls the Nusselt number varies approximately as Nu ∝ R1/4 for 1 < R < 103

Numerical Stability and Robustness

Both coupled and uncoupled ROMs maintain numerical stability up to R = Ra/Rac = 106 for integration times of T = 104 This numerical robustness is achieved without any explicit closure model, in contrast to previous POD-Galerkin studies for buoyancy-driven convection in cavities that required additional dissipation terms to prevent numerical blow-up The Galerkin ROMs are numerically stable and do not require closure models The system is integrated using an Adams/BDF method with automatic stiffness detection [15]

Application Potential

The capabilities of these models are exploited to conduct a detailed bifurcation analysis at P r = 10 using Poincar´e sections and Lyapunov exponents, precisely identifying the transitions between periodic, quasiperiodic, and chaotic states with significant reductions of computational cost These ROMs can serve as DNS surrogates over a wide range of Rayleigh numbers setting the stage for future applications in state observation, control and optimization The detailed bifurcation analysis presented below demonstrates that these ROMs reproduce the same route to chaos observed in DNS—periodic, quasiperiodic, and chaotic dynamics—as a function of the Rayleigh number These ROMs are promising candidates to serve as surrogates for the high-dimensional system in applications where a low computational cost is required The detailed bifurcation analysis presented below demonstrates that these ROMs reproduce the same route to chaos observed in DNS—periodic, quasiperiodic, and chaotic dynamics—as a function of the Rayleigh number

References

[1] E. Bodenschatz, W. Pesch, and G. Ahlers. Recent developments in Rayleigh-B´enard convection

[2] P. Manneville. Rayleigh-B´enard convection: thirty years of experimental, theoretical, and modeling work

[3] F. H. Busse. The stability of finite amplitude cellular convection and its relation to an extremum principle

[4] F. H. Busse. Non-linear properties of thermal convection

[5] H. B´enard. Les tourbillons cellulaires dans une nappe liquide propageant de la chaleur par convection, en r´egime permanent

[6] L. Rayleigh. LIX. On convection currents in a horizontal layer of fluid, when the higher temperature is on the under side

[7] F. H. Busse. Transition to turbulence in Rayleigh-Be´enard convection

[8] S. Chandrasekhar. Hydrodynamic and hydromagnetic stability

[9] D. R. Moore and N. O. Weiss. Two-dimensional Rayleigh-B´enard convection

[10] H. Yahata. Transition to turbulence in the Rayleigh-B´enard convection

[11] H. Yahata. Period-doubling cascade in the Rayleigh-B´enard convection

[12] J. Pallares, I. Cuesta, F. X. Grau, and F. Giralt. Natural convection in a cubical cavity heated from below at low Rayleigh numbers

[13] E. Zienicke, N. Seehafer, and F. Feudel. Bifurcations in two-dimensional Rayleigh-B´enard convection

[14] J. Pallares, F. X. Grau, and F. Giralt. Flow transitions in laminar Rayleigh–B´enard convection in a cubical cavity at moderate Rayleigh numbers

[15] J. B. McLaughlin and S. A. Orszag. Transition from periodic to chaotic thermal convection

[16] J. H. Curry, J. R. Herring, J. Loncaric, and S. A. Orszag. Order and disorder in two-and three-dimensional B´enard convection

[17] S. Paul, M. K. Verma, P Wahi, S K Reddy, and K Kumar. Bifurcation analysis of the flow patterns in two-dimensional Rayleigh–B´enard convection

[18] F. Waleffe. On a self-sustaining process in shear flows

[19] B. R. Noack, K. Afanasiev, M Morzy´nski, G Tadmor, and F Thiele. A hierarchy of low-dimensional models for the transient and post-transient cylinder wake

[20] C. W. Rowley and S T M Dawson. p. 6

Where the paper says it

(56) Here, A+, B+ and C + denote the adjoint state, control and observation matrices. This procedure is similar to the one described by Jovanovi´c and Bamieh [51] and Ilak and Rowley [52] for a linearized three-dimensional channel flow.

Model reduction for flow analysis and control

[21] S L Brunton and J N Kutz. Data-driven science and engineering: Machine learning, dynamical systems, and control

[22] A V G Cavalieri. Structure interactions in a reduced-order model for wall-bounded turbulence

[23] S.-C. Liao, A Leonid Heide, M. S. Hemati, and P J Seiler. A convex optimization approach to compute trapping regions for lossless quadratic systems

[24] B Saltzman. Finite amplitude free convection as an initial value problem—I J Atmos Sci

[25] E. N. Lorenz. Deterministic nonperiodic flow

[26] P Manneville. A two-dimensional model for three-dimensional convective patterns in wide containers

[27] S Paul, P Wahi, and M K Verma. Bifurcations and chaos in large-Prandtl number Rayleigh–B´enard convection

[28] D Puigjaner, J Herrero, F Giralt, and C Sim´o. Stability analysis of the flow in a cubical cavity heated from below at low Rayleigh numbers

[29] D Puigjaner, J Herrero, F Giralt, and C Sim´o. Bifurcation analysis of multiple steady flow patterns for Rayleigh-B´enard convection in a cubical cavity at Pr= 130

[30] D Puigjaner, J Herrero, C Simo, and F Giralt. Bifurcation analysis of steady Rayleigh–B´enard convection in a cubical cavity with conducting sidewalls

[31] D Puigjaner, J Herrero, C Simo, and F Giralt. From steady solutions to chaotic flows in a Rayleigh–B´enard problem at moderate Rayleigh numbers

[32] B Podvin and P Le Qu´er´e. Low-order models for the flow in a differentially heated cavity

[33] B Podvin and A Sergent. Proper orthogonal decomposition investigation of turbulent Rayleigh-B´enard convection in a rectangular cavity

[34] L Soucasse, B Podvin, P Rivi´ere, and A Soufiani. Proper orthogonal decomposition analysis and modelling of large-scale flow reorientations in a cubic Rayleigh–B´enard cell

[35] A V G Cavalieri and P A S Nogueira. Reduced-order Galerkin models of plane Couette flow

[36] A V G Cavalieri. Non-linear Galerkin reduced-order models of a mixing layer

[37] M McCormack, A V G Cavalieri, and Y Hwang. Multi-scale invariant solutions in plane Couette flow: a reduced-order model approach

[38] S E Ahmed, S Pawar, O San, A Rasheed, T Iliescu, and B R Noack. On closures for reduced order models—A spectrum of first-principle to machine-learned avenues

[39] I A Maia and A Cavalieri. Turbulence suppression in plane Couette flow using reduced-order models

[40] G Muller. Convection and inhomogeneities in crystal growth from the melt

[41] M Gunzburger, E Ozugurlu, J Turner, and H Zhang. Controlling transport phenomena in the Czochralski crystal growth process

[42] Z Alloui, Y Alloui, and P Vasseur. Control of Rayleigh-B´enard convection in a fluid layer with internal heat generation

[43] A M¨unch and B Wagner. Galerkin method for feedback controlled Rayleigh–B´enard convection

[44] E Flores-Montoya, A F C da Silva, and A V G Cavalieri. State estimation of Rayleigh–B´enard convection with reduced-order models

[45] E P Van Der Poel, R J A M Stevens, and D Lohse. Comparison between two-and three-dimensional Rayleigh–B´enard convection

[46] M A Ahmed, H J Bae, A F Thompson, and B J McKeon. Resolvent analysis of stratification effects on wall-bounded shear flows

[47] E. N. Lorenz. Available potential energy and the maintenance of the general circulation

[48] K. B Winters, P N Lombard, J J Riley, and E A D’Asaro. Available potential energy and mixing in density-stratified fluids

[49] G O Hughes, B Gayen, and R W Griffiths. Available potential energy in Rayleigh–B´enard convection

[50] S. Bagheri, D S Henningson, J Hœpffner, and P J Schmid. Input-output analysis and control design applied to a linear model of spatially developing flows

[51] M R Jovanovi´c and B Bamieh. Componentwise energy amplification in channel flows

[52] M Ilak and C W Rowley. Modeling of transitional channel flow using balanced proper orthogonal decomposition

[53] J Jim´enez and P Moin. The minimal flow unit in near-wall turbulence

[54] Q Wang, K L Chong, R J A M Stevens, R Verzicco, and D Lohse. From zonal flow to convection rolls in Rayleigh–B´enard convection with free-slip plates

[55] A V Getling. Rayleigh-B´enard convection: structures and dynamics, volume 11

[56] L N Trefethen. Spectral methods in MATLAB

[57] E Anderson, Z Bai, C Bischof, L S Blackford, J Demmel, J Dongarra, J Du Croz, A Greenbaum, S Hammarling, A McKenney et al. LAPACK users’ guide

[58] F Waleffe. The nature of triad interactions in homogeneous turbulence

[59] A C Hindmarsh. ODEPACK, a systemized collection of ODE solvers

[60] L Petzold. Automatic selection of methods for solving stiff and nonstiff systems of ordinary differential equations

[61] K J Burns, G M Vasil, J S Oishi, D Lecoanet, and B P Brown. Dedalus: A flexible framework for numerical simulations with spectral methods

[62] L Howle. A comparison of the reduced Galerkin and pseudo-spectral methods for simulation of steady Rayleigh-B´enard convection

[63] P Winchester, P D Howell, and V Dallas. The onset of zonal modes in two-dimensional Rayleigh–B´enard convection

[64] H Johnston and C R Doering. Comparison of turbulent thermal convection between conditions of constant temperature and constant flux

[65] C Grebogi, E Ott, and J A Yorke. Crises, sudden changes in chaotic attractors, and transient chaos

[66] K Kashinath, I C Waugh, and M P Juniper. Nonlinear self-excited thermoacoustic oscillations of a ducted premixed flame: bifurcations and routes to chaos

[67] A V G Cavalieri, E L Rempel, and P A S Nogueira. Transition to chaos in a reduced-order model of a shear layer

[68] K Geist, U Parlitz, and W Lauterborn. Comparison of different methods for computing Lyapunov exponents

[69] M S Jolly and E S Van Vleck. Numerical Techniques for Approximating Lyapunov Exponents and Their Implementation

Appendix A Details

The derivation of the expression for the inner product in the state space begins with eq. (51) The decomposition into Fourier modes in the x direction allows us to reduce the inner product to an integral in the y direction only Applying integration by parts to the first term of the integrand and invoking the boundary conditions yields Z 1 0 h Ψ˜⋆1∂yΨ˜2i10 + Z 1 0 −Ψ˜⋆1∂2yΨ˜2 + k 2xΨ˜⋆1Ψ˜2 + γ squared θ⋆1θ2dy = Z 1 0 -∇2 Ψtilde 5. p. 6

Where the paper says it

It is possible to derive the expression of the inner product in the state space from the inner product definition for the coupled modal basis given in eq. The decomposition into Fourier modes in the x direction allows reducing the inner product to an integral in the y direction only, ⟨y1 , Wy2 ⟩ = 1 Ly Z Ly 0 y H 1 Wy2dy = Z 1 0 (Cz1) H W Cz2dy = Z 1 0 z H 1 Mz2dy, (51) where superscript −H represents the conjugate transpose operator.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Coupled versus uncoupled ROM comparison: The improved AI system can perform more accurate surrogate modeling for Rayleigh-Bénard convection by using coupled modal bases, as coupled models show better agreement with DNS profiles, particularly for the convective heat flux and the normal Reynolds stresses compared to uncoupled approaches.

  2. Dynamical regime identification: The system can precisely identify transitions between periodic, quasiperiodic, and chaotic states by combining Poincar´e sections with Lyapunov exponents to precisely identify the transitions between different dynamical regimes.

  3. Bifurcation analysis at low cost: The AI system can conduct a "detailed bifurcation analysis at P r = 10 using Poincar´e sections and Lyapunov exponents, precisely identifying the transitions between periodic, quasiperiodic, and chaotic states with significant reductions of computational cost."

  4. Prediction of heat flux scaling: The system can reliably predict total heat flux across the fluid layer by capturing the scaling behavior observed in DNS results up to R = 105–106 using coupled and uncoupled-basis ROMs.

  5. Flow control application: The improved AI system can serve as a promising platform for developing active control strategies by exploiting the ROM's ability to reproduce dominant flow structures, which is valuable for applications such as design of control strategies that laminarize turbulent flow.

Abstract

In this work, Galerkin projection is used to build reduced-order models (ROM) for two-dimensional Rayleigh-Bénard (RB) convection with no-slip walls. We compare an uncoupled projection approach that uses separate orthonormal bases for velocity and temperature with a coupled formalism where the equations are projected onto a single basis combining velocity and temperature components. Orthonormal bases for modal projection are obtained as the eigenfunctions of the controllability Gramian of the linearized RB equations, eliminating the need for DNS snapshot databases required by traditional POD-based approaches. Various coupled and uncoupled ROMs with different numbers of modes are generated and validated against direct numerical simulations (DNS) over a wide range of Rayleigh numbers. One of the objectives is to determine their domain of validity as a function of the system dimension and the Rayleigh number. DNS and ROM results are compared in terms of mean vertical profiles, heat flux, flow structures, dynamical regimes and energy spectra. Crucially, unlike previous POD-based Galerkin models for thermal convection, these ROMs do not require closure models and remain numerically stable. The coupled approach shows better agreement with DNS in terms of mean vertical profiles and Nusselt number scaling. The capabilities of these models are exploited to conduct a detailed bifurcation analysis at Pr = 10 using Poincaré sections and Lyapunov exponents, precisely identifying the transitions between periodic, quasiperiodic, and chaotic states with significant reductions of computational cost.

Sources

Related papers