Reduced Order Methods for Parametrized Non-linear and Time Dependent Optimal Flow Control Problems, Towards Applications in Biomedical and Environmental Sciences

Maria Strazzullo, Zakia Zainib, Francesco Ballarin, Gianluigi Rozza

TL;DR

This paper introduces an efficient framework using Reduced Order Methods (ROMs) to solve complex optimal flow control problems governed by non-linear and time-dependent partial differential equations. By significantly reducing computational cost, this approach makes previously intractable simulations feasible for applications in environmental and biomedical fields.

What Problem Does This Paper Address?

Solving optimal control problems constrained by partial differential equations (PDEs) is computationally demanding, especially when these PDEs are non-linear, time-dependent, or involve geometric and physical parameters. This computational burden often makes it impossible to perform real-time simulations or explore a wide range of parameter variations. The authors aim to bridge this gap by developing a method that can rapidly and accurately solve these problems, opening doors for applications in areas like environmental monitoring and patient-specific cardiovascular modeling.

Methods

The core of the paper lies in the application of Reduced Order Methods (ROMs), specifically a POD-Galerkin approach, to tackle these complex optimal flow control problems. The process involves two main phases:

  • Offline Phase: This is a computationally intensive, one-time setup. It involves generating "snapshots" of the system's behavior across various parameter values. These snapshots are then used to construct low-dimensional "reduced order spaces" using Proper Orthogonal Decomposition (POD). This phase also includes adding "supremizers" to ensure the stability and uniqueness of the reduced-order solution, particularly for pressure.
  • Online Phase: Once the reduced order spaces are built, solving the problem for a new parameter value becomes extremely fast. The original, high-dimensional problem is projected onto these low-dimensional spaces, resulting in a significantly smaller system that can be solved in near real-time.

The authors formulate the optimal control problem using the Karush-Kuhn-Tucker (KKT) conditions, creating an "optimality system" that is then reduced.

Datasets

The paper does not explicitly mention or use external datasets for training. Instead, it generates its own "snapshots" of the system's behavior by solving the full PDE-constrained optimal control problem for a set of parameter values. These snapshots are then used to build the ROMs. The paper presents two test cases:

  1. Time-dependent Stokes equations: This case uses a bifurcation geometry relevant to environmental applications (e.g., riverbeds).
  2. Steady non-linear Navier-Stokes equations: This case also uses a bifurcation geometry, but it's tailored for biomedical applications (e.g., arterial bypass grafts).

Key Insights and Ideas

  • ROMs as a Game Changer for Optimal Control: The paper convincingly demonstrates that ROMs can transform computationally prohibitive optimal control problems into tractable ones, enabling rapid exploration of parameter spaces.
  • Bridging Environmental and Biomedical Sciences: The framework is shown to be versatile, successfully applied to two distinct problem types (Stokes and Navier-Stokes) with relevance to both environmental fluid dynamics and cardiovascular modeling.
  • The Power of POD-Galerkin: The POD-Galerkin approach, combined with supremizers for stability, proves to be an effective strategy for reducing the dimensionality of the optimality system while maintaining accuracy.
  • Offline/Online Paradigm Efficiency: The clear separation into an offline (computationally expensive, one-time) and online (very fast, repeatable) phase is crucial for achieving real-time performance for varying parameters.

Key Results

The paper presents compelling evidence of the effectiveness of their ROM approach:

  • Significant Speedup: For the time-dependent Stokes problem, a speedup of up to 17,420 times is reported with a reduced order basis of size N=35, compared to the full-order model.
  • High Accuracy: The relative error between the full-order model (FOM) and the ROM solutions is consistently low, often reaching 10⁻⁸ to 10⁻¹⁰ for the cost functional and solution variables. Figure 1 visually confirms this low relative error across various parameters.
  • Visual Validation: Figure 3 shows excellent qualitative agreement between the FOM and ROM solutions for state velocity and pressure in the Stokes case. Figure 8 provides similar visual validation for the Navier-Stokes case.
  • Computational Efficiency for Navier-Stokes: For the Navier-Stokes problem, the online phase of the ROM is significantly faster (9 x 10³ seconds) than the full mesh size computation (4.9 x 10³ seconds), indicating a substantial speedup, although the exact speedup factor is not explicitly stated in the table.

Limitations and Future Work

  • Affine Assumption: The efficiency of the POD-Galerkin approach relies on the "affine assumption," meaning that the forms can be expressed as a linear combination of parameter-dependent functions and parameter-independent quantities. While this is a standard assumption for many ROMs, it might limit applicability to problems where this assumption doesn't hold.
  • Generalizability: While the paper demonstrates success on two specific problems, further work could explore the application of this framework to a wider range of PDE-constrained optimal control problems and different geometries.
  • Non-linearities and Time-Dependence: The paper tackles both non-linear and time-dependent aspects. Future research could focus on pushing the boundaries of complexity, perhaps by exploring more intricate non-linearities or more challenging time-dependent scenarios.
  • Patient-Specific Applications: The authors highlight the potential for patient-specific applications. Further development could involve integrating this framework with medical imaging data and clinical workflows.

Summary updated

Abstract

We introduce reduced order methods as an efficient strategy to solve parametrized non-linear and time dependent optimal flow control problems governed by partial differential equations. Indeed, the optimal control problems require a huge computational effort in order to be solved, most of all in physical and/or geometrical parametrized settings. Reduced order methods are a reliable and suitable approach, increasingly gaining popularity, to achieve rapid and accurate optimal solutions in several fields, such as in biomedical and environmental sciences. In this work, we employ a POD-Galerkin reduction approach over a parametrized optimality system, derived from the Karush-Kuhn-Tucker conditions. The methodology presented is tested on two boundary control problems, governed respectively by (i) time dependent Stokes equations and (ii) steady non-linear Navier-Stokes equations.
BibTeX
@inproceedings{strazzullo2019reduced,
  author = {Maria Strazzullo and Zakia Zainib and Francesco Ballarin and Gianluigi Rozza},
  title = {{Reduced Order Methods for Parametrized Non-linear and Time Dependent Optimal Flow Control Problems, Towards Applications in Biomedical and Environmental Sciences}},
  booktitle = {Enumath},
  year = {2019},
  doi = {10.1007/978-3-030-55874-1\_83},
}
Figure