[Paper Review] Data Assimilation In Large Prandtl Rayleigh-benard Convection From Thermal Measurements

Aseel Farhat, Nathan Glatt-Holtz+3 more|arXiv (Cornell University)|Mar 4, 2019

Fluid Dynamics and Turbulent Flows59 references38 citations

TL;DR

This paper proposes a continuous data assimilation scheme for large-Prandtl Rayleigh-Bénard convection using only temperature measurements, employing a nudging algorithm to synchronize the model with observations. Rigorous conditions guarantee synchronization, but numerical results show the theoretical bounds are overly conservative, with successful synchronization often occurring well below required thresholds for the relaxation parameter and observation resolution.

ABSTRACT

This work applies a continuous data assimilation scheme-a framework for reconciling sparse and potentially noisy observations to a mathematical model-to Rayleigh-Benard convection at infinite or large Prandtl numbers using only the temperature field as observables. These Prandtl numbers are applicable to the earth's mantle and to gases under high pressure. We rigorously identify conditions that guarantee synchronization between the observed system and the model, then confirm the applicability of these results via numerical simulations. Our numerical experiments show that the analytically derived conditions for synchronization are far from sharp; that is, synchronization often occurs even when sufficient conditions of our theorems are not met. We also develop estimates on the convergence of an infinite Prandtl model to a large (but finite) Prandtl number generated set of observations. Numerical simulations in this hybrid setting indicate that the mathematically rigorous results are accurate, but of practical interest only for extremely large Prandtl numbers.

Motivation & Objective

• To develop a mathematically rigorous data assimilation framework for Rayleigh-Bénard convection at infinite or large Prandtl numbers using only temperature observations.
• To identify minimal conditions on the nudging parameter and number of observation modes that guarantee synchronization between the model and observed system.
• To assess the practical applicability of theoretical synchronization bounds by comparing them with numerical simulations in both infinite-Prandtl and finite-Prandtl regimes.
• To investigate the impact of model mismatch—specifically, when observations come from a finite-Prandtl system but the assimilating model assumes infinite Prandtl number—on synchronization accuracy.
• To evaluate the convergence of the infinite-Prandtl model to finite-Prandtl reality and quantify the resulting error in the presence of model-observation mismatch.

Proposed method

• Formulates a continuous data assimilation scheme via a nudging algorithm that modifies the model equations by injecting observational data through a relaxation term.
• Applies the nudging scheme to the 2D Rayleigh-Bénard system with infinite Prandtl number, using only temperature field measurements as observables.
• Derives rigorous synchronization conditions based on the relaxation parameter µ and the number of projected observation modes N, ensuring convergence of the model state to the true state as time increases.
• Performs numerical simulations to test the theoretical bounds, using high-precision solvers and adaptive time-stepping to resolve moderately turbulent flows.
• Constructs a hybrid setting where observations are generated from a finite-Prandtl-number simulation but the assimilating model assumes infinite Prandtl number, to study model-observation mismatch.
• Quantifies error decay rates and compares theoretical predictions with numerical outcomes, particularly analyzing the O(Pr⁻¹) error bound in the hybrid setting.

Experimental results

Research questions

• RQ1 Under what conditions on the relaxation parameter µ and number of observation modes N does the data assimilation scheme guarantee synchronization between the model and the true system in the infinite-Prandtl-number limit?
• RQ2 How sharp are the theoretically derived conditions for synchronization, and to what extent do numerical experiments confirm or exceed them?
• RQ3 Can synchronization be achieved in the finite-Prandtl-number regime using only temperature measurements, and how does the convergence rate compare to the infinite-Prandtl case?
• RQ4 What is the impact of model-observation mismatch—specifically, when the assimilating model assumes infinite Prandtl number but observations come from a finite-Prandtl system—on synchronization accuracy and error dynamics?
• RQ5 How does the error between the infinite-Prandtl model and the finite-Prandtl reality scale with Prandtl number, and is this error dominant over dynamical synchronization error?

Key findings

• Theoretical conditions for synchronization in the infinite-Prandtl case require a relatively large relaxation parameter µ and sufficient observation modes N, but numerical experiments show synchronization often occurs even when these conditions are not met.
• The required values of µ for theoretical synchronization are at least an order of magnitude larger than those observed to produce successful synchronization in numerical simulations, indicating the theoretical bounds are overly conservative.
• In the finite-Prandtl-number setting, synchronization is still achievable with temperature-only observations, though convergence is slower and the required parameter relationships are less clearly defined than in the infinite-Prandtl case.
• In the hybrid setting—where observations are from a finite-Prandtl simulation but the model assumes infinite Prandtl number—synchronization error is dominated by the Prandtl number mismatch, with error decaying as O(Pr⁻¹) as Pr increases.
• Numerical results confirm the theoretical O(Pr⁻¹) error bound in the hybrid setting, showing that the model error from assuming infinite Prandtl number is the primary source of discrepancy, not synchronization failure.
• Despite the theoretical rigor, the practical applicability of the data assimilation scheme is limited in realistic settings where the true model is not known exactly, especially when model mismatch (e.g., in Prandtl number or boundary conditions) is present.