Integral Formulation and the Brézis-Ekeland-Nayroles-Type Principle for Prox-Regular Sweeping Processes

This paper establishes a unified bounded-variation solution framework for prox-regular sweeping processes by proving the equivalence between a new integral formulation with a quadratic correction term and the standard differential-measure formulation, while also deriving a Brézis-Ekeland-Nayroles-type variational characterization that ensures stability under uniform limits.

The Gist

Imagine you are walking through a crowded room where the walls are constantly moving, shifting, and sometimes even jumping to new locations. This is the world of Sweeping Processes.

In mathematics, a "sweeping process" describes how an object (like a particle or a robot) moves while being forced to stay inside a moving container (the "constraint"). If the container is a convex shape (like a smooth ball or a box), the rules are well-known. But what if the container is weirdly shaped (non-convex), has sharp corners, or suddenly jumps to a new spot? That's where this paper comes in.

Here is the story of what the authors, Juan Guillermo Garrido and Emilio Vilches, have discovered, explained through simple analogies.

1. The Problem: The "Jumping" Room

In the classic version of this problem (invented by Jean-Jacques Moreau in the 1970s), the walls of the room move smoothly. If the wall moves a tiny bit, the person inside moves a tiny bit.

But in the real world (like in car crashes, metal bending, or robot collisions), things don't always move smoothly. A wall might suddenly snap to a new position. The person inside might get "kicked" by the wall.

• The Challenge: How do you mathematically describe a path when the rules change abruptly?
• The Setting: The authors look at "Prox-Regular" sets. Think of these as shapes that aren't perfectly smooth (like a sphere) but aren't jagged nightmares either. They have "rounded" corners or smooth curves, but they aren't necessarily convex (they could be shaped like a crescent moon or a donut).

2. Two Ways to Describe the Walk

The paper tackles a fundamental question: How do we define a "solution" (a valid path) when the walls jump?

The authors compare two different ways of describing the walker's journey:

Method A: The "Local Police Officer" (Differential-Measure Formulation)

Imagine a police officer standing right next to the walker at every single instant.

• The officer checks: "Are you touching the wall? If yes, are you trying to walk into the wall?"
• If you are touching the wall, the officer says, "You must stop or slide along the wall. You cannot push through it."
• The Catch: This method is very local. It looks at one tiny split-second at a time. It uses complex math (differential measures) to handle the "kicks" when the wall jumps.

Method B: The "Global Coach" (Integral Formulation)

Now, imagine a coach looking at the entire walk from start to finish.

• The coach doesn't care about split-seconds. Instead, the coach compares your actual path to every other possible path you could have taken that stayed inside the room.
• The coach says: "Your path is the best one if, when you compare it to any other valid path, you don't lose too much 'energy' or 'distance'."
• The Twist: Because the walls are "Prox-Regular" (not perfectly smooth), the coach has to add a correction term. It's like a penalty for taking a sharp turn. If the wall is curved, the math needs a little extra "safety buffer" (a quadratic term) to account for the fact that the wall isn't a flat mirror.

3. The Big Discovery: They Are the Same!

For a long time, mathematicians wondered if these two methods (the Local Police and the Global Coach) agreed on who was a "valid walker."

The Paper's Main Result:
Under reasonable conditions (the walls don't disappear, they are connected, and you can always find a way to extend a path), the Local Police and the Global Coach agree perfectly.

• If the Local Police says you are a valid walker, the Global Coach agrees.
• If the Global Coach says you are a valid walker, the Local Police agrees.

This is huge because the "Global Coach" method (Integral Formulation) is much easier to work with for computers and simulations. It turns a messy, split-second problem into a smooth, big-picture inequality.

4. The "Brezis-Ekeland-Nayroles" Principle: The Perfect Score

The authors introduce a fancy-sounding concept called a Variational Residual. Let's call it the "Error Score."

• Imagine you are playing a video game where you try to walk through the moving room.
• Every time you take a step, the game calculates an Error Score.
• The Rule: If you are following the rules perfectly, your Error Score is Zero.
• If you are doing something wrong (like walking into a wall or taking a weird shortcut), your Error Score is Negative (or positive, depending on how you count, but the point is: it's not zero).

The Magic:
The authors prove that the only way to get a score of Zero is to be a true solution.

• If you are a valid path, your score is 0.
• If your score is 0, you are a valid path.
• If you are close to a valid path, your score is close to 0.

This is like a "Goldilocks" principle: The solution is the "just right" path that minimizes the error to exactly zero.

5. Why Does This Matter? (Stability)

Why do we care about this "Error Score"? Because it helps us predict what happens when things go wrong or when we approximate the problem.

• Scenario: Imagine you are simulating a car crash on a computer. You can't calculate every tiny fraction of a second perfectly, so you use a "step-by-step" approximation.
• The Result: If your approximation steps are getting better and better, the "Error Score" of your simulation will get closer and closer to Zero.
• The Guarantee: The authors prove that if your Error Score goes to Zero, your simulation is guaranteed to converge to the real physical solution.

It's like saying: "If your GPS route keeps getting closer to the 'perfect' route (zero error), you can trust that you will eventually arrive at the correct destination, even if the road conditions (the moving walls) are chaotic."

Summary in a Nutshell

1. The Problem: How to mathematically describe an object moving in a room with weird, jumping walls.
2. The Solution: The authors proved that a "local" rule (checking every second) and a "global" rule (checking the whole trip) are actually the same thing.
3. The Tool: They created a "Scorecard" (Variational Residual). If the score is zero, you are doing it right.
4. The Benefit: This scorecard makes it much easier to build computer simulations and prove that those simulations will work, even when the walls are jumping around and the shapes are non-convex.

This paper provides a robust, unified language for understanding how things move when the rules of the game change abruptly, bridging the gap between complex local physics and global mathematical stability.

Technical Summary

Here is a detailed technical summary of the paper "Integral Formulation and the Br´ezis-Ekeland-Nayroles-Type Principle for Prox-Regular Sweeping Processes" by Juan Guillermo Garrido and Emilio Vilches.

1. Problem Statement

The paper addresses the mathematical modeling of sweeping processes in a Hilbert space $H$. A sweeping process is a first-order dynamical system where a trajectory $x(t)$ is constrained to remain within a moving closed set $C(t)$, governed by the normal cone to the set.

Key Challenges:

• Nonconvexity: The moving sets $C(t)$ are uniformly prox-regular, a class that strictly extends convexity to include nonconvex sets with sufficiently smooth geometry (e.g., $C^2$ submanifolds).
• Discontinuities: The moving constraint $C(t)$ is allowed to exhibit discontinuities of bounded variation (BV) in time, rather than requiring Lipschitz continuity. This is crucial for applications in contact mechanics involving collisions and jumps.
• Formulation Gap: While differential-measure formulations exist for such problems, a global integral variational formulation (tested against continuous trajectories) had not been established for the prox-regular, nonconvex, discontinuous setting.

2. Methodology and Framework

The authors develop a unified framework combining tools from nonsmooth analysis, measure theory, and set-valued analysis.

A. Geometric and Topological Assumptions

The moving set $C: [0, T] \rightrightarrows H$ satisfies three core hypotheses:

1. (H1) Uniform Prox-regularity: For all $t$, $C(t)$ is nonempty, closed, connected, and $\rho$-uniformly prox-regular. This ensures the existence of a unique metric projection in a neighborhood and controls the "curvature" of the boundary.
2. (H2) Lower Semicontinuity (LSC): The set-valued map is lower semicontinuous in time, allowing for the existence of continuous selections.
3. (H3) Bounded Selection-Extension Property: A critical technical condition ensuring that any continuous selection defined on a compact subset of time (staying within a tube around a bounded trajectory) can be extended to a global continuous selection with a uniform bound. This guarantees the availability of a rich class of admissible test trajectories.

B. Two Solution Concepts

The paper compares two definitions of solutions for trajectories of bounded variation ($BV$):

1. Differential-Measure Formulation (Local): The trajectory $x$ satisfies $dx \in -N^P(C(t); x(t))$ in the sense of measures, where $N^P$ is the proximal normal cone.
2. Integral Formulation (Global): A global variational inequality tested against continuous admissible trajectories $y(t) \in C(t)$. Crucially, due to the nonconvexity of prox-regular sets, this inequality includes a quadratic correction term:
   $$\int_0^T \left( \langle v(t), y(t) - x(t) \rangle + \frac{\|v(t)\|}{2\rho} \|y(t) - x(t)\|^2 \right) d\nu(t) \geq 0$$
   where $v = \frac{dx}{d\nu}$ is the density of the differential measure. The quadratic term compensates for the hypomonotonicity of the proximal normal cone.

C. Variational Residual

The authors define a prox-regular variational residual $E_\nu(x)$, which measures the "defect" of a trajectory in satisfying the integral inequality.
$$E_\nu(x) = \inf_{y \in \mathcal{A}_\nu} \int_0^T \left( \langle v(t), y(t) - x(t) \rangle + \frac{\|v(t)\|}{2\rho} \|y(t) - x(t)\|^2 \right) d\nu(t)$$
where $\mathcal{A}_\nu$ is the class of continuous admissible test trajectories.

3. Key Contributions and Results

1. Equivalence of Formulations (Theorem 4.6)

The primary theoretical result is the proof that the Integral Formulation and the Differential-Measure Formulation are equivalent under assumptions (H1)–(H3).

• Significance: This unifies the local measure-theoretic approach with a global variational approach for nonconvex, discontinuous sweeping processes.
• Mechanism: The proof relies heavily on the selection-extension property (H3) to construct test trajectories that approximate the trajectory $x$ itself, allowing the recovery of the local normal cone inclusion from the global integral inequality.

2. Br´ezis-Ekeland-Nayroles (BEN) Type Principle (Theorem 5.1)

The paper establishes a variational characterization of solutions analogous to the classical BEN principle for convex systems.

• Result: A trajectory $x$ is a solution if and only if its variational residual vanishes: $E_\nu(x) = 0$.
• Implication: Solutions are exactly those trajectories that maximize the residual (since the residual is non-positive for admissible trajectories, the maximum is 0). This provides a global-in-time characterization of the solution.

3. Stability of Approximations (Theorem 6.1)

Using the residual framework, the authors prove a robust stability result.

• Setup: Consider a sequence of approximating moving sets $C_n$ and admissible trajectories $x_n$ with residuals $E_\nu^n(x_n) \to 0$.
• Result: If $x_n$ converges uniformly to $x$, and the sequence of differential measures satisfies appropriate boundedness and weak convergence conditions, then the limit $x$ is a solution to the limit sweeping process driven by $C$.
• Conditions: The proof requires an Outer Semicontinuity (OSC) condition for the sets and a Selection Approximation Property (SAP) to ensure test trajectories for the limit can be approximated by those for the perturbations.

4. Significance and Impact

• Unification of Nonconvex Analysis: The work bridges the gap between local measure differential inclusions and global variational principles in the challenging setting of nonconvex, prox-regular sets.
• Handling Discontinuities: By working in the $BV$ framework with Radon measures, the theory accommodates jumps in the constraint set, which are physically realistic in contact mechanics but mathematically difficult.
• Numerical and Approximation Implications: The introduction of the variational residual provides a quantitative tool for:
  • A posteriori error estimation: The magnitude of the residual indicates how far a numerical approximation (e.g., from a catching-up scheme) is from the true solution.
  • Stability Analysis: It offers a rigorous way to prove convergence of discretization schemes (time-stepping) and geometric regularizations without relying solely on compactness arguments.
• Extension of Classical Theory: It generalizes the classical Br´ezis-Ekeland-Nayroles principle (originally for convex/maximal monotone operators) to a broader class of non-monotone, nonconvex operators arising from prox-regular geometry.

In summary, this paper provides a robust, unified variational framework for sweeping processes with discontinuous, nonconvex constraints, offering new tools for stability analysis and numerical approximation in nonsmooth mechanics.