💻 computer science

A priori error estimates for optimal control problems governed by the transient Stokes equations and subject to state constraints pointwise in time

This paper establishes a priori error estimates and demonstrates improved regularity for the optimal control in a state-constrained transient Stokes problem discretized via inf-sup stable finite elements in space and a discontinuous Galerkin method in time, supported by numerical results.

Original authors: Dmitriy Leykekhman, Boris Vexler, Jakob Wagner

Published 2026-05-20

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Dmitriy Leykekhman, Boris Vexler, Jakob Wagner

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to steer a complex, invisible river of fluid (like water or air) through a pipe system. You have a control panel with knobs (the control) that you can turn to push the fluid in different directions. Your goal is to make the fluid flow exactly like a specific "dream pattern" (the desired state) that you have in mind.

However, you have two major rules to follow:

The Knobs Have Limits: You can't turn the knobs infinitely hard; they have minimum and maximum settings.
The River Must Stay Safe: At every single moment in time, the total "push" of the fluid in a specific direction must not exceed a certain safety limit. If it does, the system fails.

This paper is about figuring out the perfect way to turn those knobs to get the closest possible match to your dream pattern without breaking the safety rules, and then proving mathematically how close a computer simulation can get to that perfect solution.

Here is a breakdown of the paper's journey using simple analogies:

1. The Problem: Steering a Wobbly River

The fluid in this story follows the Stokes equations. Think of this as the "physics rulebook" for slow-moving, thick fluids (like honey or very slow water). It's tricky because the fluid has to be incompressible (you can't squeeze it into a smaller space) and it has to move smoothly.

The authors are dealing with a Transient problem, meaning the fluid is moving and changing over time, not just sitting still. They want to find the best control $q$ (the knobs) to minimize the difference between the actual fluid flow and the desired flow, while paying a "cost" for turning the knobs too hard (regularization).

2. The Safety Net: The "Time-Point" Constraint

The most unique part of this paper is the state constraint.

The Analogy: Imagine the fluid is a balloon. You are allowed to blow it up, but at every single instant in time, the balloon's volume (measured by a specific formula) cannot exceed a certain size $\beta$ .
The Challenge: In many math problems, constraints are checked only at the very end or on average. Here, the rule must be obeyed pointwise in time. It's like driving a car where you must never exceed the speed limit for even a split second, not just on average over the whole trip.

3. The Solution: Building a Digital Model

Since we can't solve these complex fluid equations with a pencil and paper, we use a computer. The authors build a digital model using two main tools:

Space (The Mesh): They chop the physical space (the pipe) into tiny triangles (like a mosaic). This is the Finite Element Method.
Time (The Steps): They chop the time interval into small slices. They use a method called Discontinuous Galerkin, which is like taking a series of snapshots of the fluid. Unlike a smooth movie, these snapshots can jump a little bit from one frame to the next, which gives the math more flexibility to handle sharp changes.

4. The Big Question: How Good is the Computer Model?

The core of the paper is an Error Estimate.

The Question: If the "perfect" solution is a real, continuous river, and our computer model is a pixelated, step-by-step simulation, how far apart are they?
The Result: The authors prove a specific formula for the distance between the perfect solution and the computer solution.
- The error gets smaller as you make the time steps ( $k$ ) and the space triangles ( $h$ ) smaller.
- The formula looks roughly like: Error $\approx$ (Time Step) $\times$ (Space Step) $\times$ a small logarithmic factor.
- Crucially, they show that even with the tricky "safety limit" constraint, the computer model converges to the truth at a predictable rate.

5. The "By-Product": Smoother Controls

One of the surprising findings is about the smoothness of the solution.

The Analogy: You might expect the perfect way to turn the knobs to be jerky or jagged because of the safety limits.
The Discovery: The math proves that the optimal control is actually quite smooth (it belongs to a specific class of "nice" functions). This is a hidden bonus of their analysis; by proving the error estimates, they accidentally proved that the solution is well-behaved.

6. The Proof: Numerical Experiments

Finally, the authors didn't just do the math; they ran computer simulations to test it.

Example 1 (Smooth Data): They created a scenario where everything is perfectly smooth. The computer results matched the theory perfectly, showing the expected rate of improvement as they refined the mesh.
Example 2 (Rough Data): They introduced "rough" data (jagged, less smooth inputs). As expected, the computer model improved more slowly, but it still followed the theoretical predictions.
Example 3 (Knob Limits): They added the rule that the knobs themselves couldn't turn past a certain point. The theory held up here too, proving that the safety limits on the knobs don't break the math.

Summary

In short, this paper is a rigorous mathematical guarantee. It says: "If you use this specific computer method to steer a fluid with strict, moment-by-moment safety limits, you can trust that your simulation will get closer and closer to the truth at a specific, predictable speed."

They didn't just say "it works"; they wrote down the exact formula for how well it works, even when the problem gets complicated with time-dependent safety rules.

Technical Summary: A Priori Error Estimates for Optimal Control Problems Governed by Transient Stokes Equations with Pointwise-in-Time State Constraints

Problem Formulation
This paper investigates a state-constrained optimal control problem governed by the transient Stokes equations. The objective is to minimize a cost functional comprising the $L^2$ -distance between the state $u$ and a desired state $u_d$ , plus a regularization term for the control $q$ . The system is subject to:

State Equation: The transient Stokes equations with homogeneous initial data and zero Dirichlet boundary conditions.
Control Constraints: Pointwise bounds on the control $q$ (i.e., $q_a \leq q \leq q_b$ ).
State Constraints: A constraint on an $L^2(\Omega)$ functional of the state, required to satisfy a pointwise bound in time. Specifically, $\int_\Omega u(t,x) \cdot w(x) \, dx \leq \beta$ for all $t \in [0, T]$ .

The domain $\Omega \subset \mathbb{R}^d$ ( $d=2,3$ ) is assumed to be a convex polygonal or polyhedral domain. The authors note that while the analysis focuses on homogeneous initial data, the results extend to the inhomogeneous case.

Methodology
The analysis proceeds through the following steps:

Continuous Analysis:
- The authors establish the well-posedness of the state equation and the existence of a unique optimal control under a Slater condition.
- They derive the first-order necessary optimality conditions, involving a state equation, an adjoint equation, and a Lagrange multiplier $\bar{\mu} \in (C(\bar{I}))^*$ associated with the state constraint.
- A significant portion of the continuous analysis is dedicated to deriving regularity results for the optimal control $\bar{q}$ and the adjoint state $\bar{z}$ . By analyzing the structure of the adjoint equation with a measure-valued right-hand side (due to the Lagrange multiplier), the authors prove that $\bar{q}$ possesses improved regularity, specifically $\bar{q} \in L^2(I; H^1(\Omega)^d) \cap C(\bar{I}; L^2(\Omega)^d)$ , and under certain conditions on the weight function $w$ , $\bar{q}$ is continuous in time.
Discretization Scheme:
- Spatial Discretization: The state equation is discretized using inf-sup stable finite elements (e.g., Taylor-Hood or MINI elements) in space. The analysis utilizes a discretely divergence-free formulation.
- Temporal Discretization: A discontinuous Galerkin (dG) method of order 0 (piecewise constant in time) is employed for time discretization.
- Control Discretization: The paper analyzes two scenarios:
  - Variational Discretization: The control space is not discretized a priori; the control is determined implicitly via the optimality conditions.
  - Fully Discrete: The control is discretized using piecewise constant functions in both space and time.
Error Analysis:
- The authors leverage previously established best-approximation type error estimates for the transient Stokes equations in the $L^\infty(I; L^2(\Omega))$ norm.
- They derive error estimates for the optimal control and state by comparing the continuous problem with the variationally discretized problem, and subsequently with the fully discrete problem.
- A key technical challenge addressed is the coupling between the spatial mesh size $h$ and the time step $k$ . The authors impose a mild coupling condition (Assumption 3) to ensure the discrete Slater condition holds, which is necessary for the existence of discrete solutions and the validity of the error analysis.

Key Contributions and Results
The primary result of the paper is the derivation of a priori error estimates for the optimal control $\bar{q}$ and the state $\bar{u}$ . Specifically, Theorem 6.6 establishes that the error between the continuous optimal control $\bar{q}$ and the fully discrete optimal control $\bar{q}_\sigma$ satisfies:
$\|\bar{q} - \bar{q}_\sigma\|_{L^2(I \times \Omega)} \leq C \frac{\ln(T/k)}{\sqrt{\alpha}} (k^{1/2} + h)$
where $C$ is a constant independent of $k$ and $h$ , and $\alpha$ is the regularization parameter.

Key aspects of these results include:

Convergence Rates: The estimate indicates a convergence rate of order $O(h)$ in space and $O(k^{1/2})$ in time (up to a logarithmic factor).
Regularity Independence: The error estimates are derived without assuming additional a priori smoothness on the optimal control beyond what is guaranteed by the first-order optimality conditions.
Improved Regularity: As a by-product of the error analysis, the authors show that the optimal control $\bar{q}$ enjoys improved regularity, specifically $\bar{q} \in L^2(I; H^1(\Omega)^d) \cap H^s(I; L^2(\Omega)^d)$ for $s < 1/2$ .
Slater Condition: The paper provides a rigorous proof that the Slater condition for the continuous problem implies a discrete Slater condition under a mild coupling of $h$ and $k$ , ensuring the feasibility of the discrete problem.

Significance and Claims
The authors position this work as a step toward analyzing transient flow control problems with drag or lift constraints at every point in time. While previous literature has addressed optimal control of Navier-Stokes equations with general state constraints or stationary Stokes problems with pointwise constraints, this paper provides the first rigorous a priori error estimates for transient Stokes equations with pointwise-in-time state constraints on an $L^2$ functional.

The paper claims that its techniques allow for the derivation of error estimates without strong coupling conditions between $h$ and $k$ , requiring only that the product of the logarithmic time factor and the mesh size converges to zero. The theoretical findings are complemented by numerical experiments in Section 8, which confirm the predicted convergence orders for both smooth and non-smooth data scenarios, including cases with active control constraints. The authors note that while the derived regularity results are interesting, their direct application to deriving even sharper error estimates remains an open question.