Surrogate normal-forms for the numerical bifurcation and stability analysis of navier-stokes flows via machine learning
Authors:
Alessandro Della Pia,
Dimitrios G. Patsatzis,
Gianluigi Rozza,
Lucia Russo,
Constantinos Siettos
Abstract:
Inspired by the Equation-Free multiscale modeling approach, we demonstrate how the embed-learn-lift framework enables the construction of surrogate normal-forms, namely minimal-dimensional reduced-order models (ROMs), from high-fidelity Navier-Stokes simulations. These surrogate models are then used for efficient and accurate bifurcation and stability analysis. The framework proceeds in four steps…
▽ More
Inspired by the Equation-Free multiscale modeling approach, we demonstrate how the embed-learn-lift framework enables the construction of surrogate normal-forms, namely minimal-dimensional reduced-order models (ROMs), from high-fidelity Navier-Stokes simulations. These surrogate models are then used for efficient and accurate bifurcation and stability analysis. The framework proceeds in four steps. First, manifold learning reveals the intrinsic latent dimension of the high-dimensional spatio-temporal Navier-Stokes dynamics across parameter space. Second, we construct low-dimensional "normal-form" like ROMs on this latent space using Gaussian Process Regression (GPR), capturing the emergent dynamics. Third, using these models, we apply numerical bifurcation tools to compute bifurcation diagrams and perform stability analysis in the latent space. This includes tracing branches of limit cycles arising from Andronov-Hopf bifurcations - tasks intractable in full space due to computational cost. Finally, solving the pre-image problem allows reconstruction of the bifurcation structure in the original high-dimensional space. We demonstrate the methodology on two canonical flows: wake flow past an infinite circular cylinder and planar sudden-expansion channel flow. These exhibit Andronov-Hopf and pitchfork bifurcations, respectively, as Reynolds number increases. Our method identifies the latent dimensionality and constructs GPR-based surrogate normal-forms that enable the tracing and stability analysis of bifurcating solutions, including limit cycles, their period, and stability via Floquet multipliers.
△ Less
Submitted 26 June, 2025;
originally announced June 2025.
Learning the Latent dynamics of Fluid flows from High-Fidelity Numerical Simulations using Parsimonious Diffusion Maps
Authors:
Alessandro Della Pia,
Dimitris Patsatzis,
Lucia Russo,
Constantinos Siettos
Abstract:
We use parsimonious diffusion maps (PDMs) to discover the latent dynamics of high-fidelity Navier-Stokes simulations with a focus on the 2D fluidic pinball problem. By varying the Reynolds number, different flow regimes emerge, ranging from steady symmetric flows to quasi-periodic asymmetric and turbulence. We show, that the proposed non-linear manifold learning scheme, identifies in a crisp manne…
▽ More
We use parsimonious diffusion maps (PDMs) to discover the latent dynamics of high-fidelity Navier-Stokes simulations with a focus on the 2D fluidic pinball problem. By varying the Reynolds number, different flow regimes emerge, ranging from steady symmetric flows to quasi-periodic asymmetric and turbulence. We show, that the proposed non-linear manifold learning scheme, identifies in a crisp manner the expected intrinsic dimension of the underlying emerging dynamics over the parameter space. In particular, PDMs, estimate that the emergent dynamics in the oscillatory regime can be captured by just two variables, while in the chaotic regime, the dominant modes are three as anticipated by the normal form theory. On the other hand, proper orthogonal decomposition (POD)/PCA, most commonly used for dimensionality reduction in fluid mechanics, does not provide such a crisp separation between the dominant modes. To validate the performance of PDMs, we also computed the reconstruction error, by constructing a decoder using Geometric Harmonics. We show that the proposed scheme outperforms the POD/PCA over the whole Reynolds number range. Thus, we believe that the proposed scheme will allow for the development of more accurate reduced order models for high-fidelity fluid dynamics simulators, thus relaxing the curse of dimensionality in numerical analysis tasks such as bifurcation analysis, optimization and control.
△ Less
Submitted 5 August, 2024;
originally announced August 2024.
