Minimizers that are not Impulsive Minimizers and Higher Order Abnormality

This paper resolves compatibility issues between set-separation and penalization approaches in optimal control by establishing conditions under which Clarke tangent cones are Quasi Differential Quotient approximating cones, and subsequently applies this result to prove that infimum gaps for strict-sense minimizers correspond to higher-order abnormality in the Maximum Principle.

The Gist

Imagine you are trying to navigate a ship to a specific island (the Target) while trying to spend the least amount of fuel possible (the Cost). This is the essence of an Optimal Control Problem.

However, the ocean is tricky. Sometimes, the "perfect" route requires the ship to move so fast or change direction so violently that it breaks the laws of physics as we usually understand them (like having infinite speed). In math, we call these "unbounded controls."

This paper tackles two big puzzles about how we find these perfect routes and what happens when the "perfect" route doesn't actually exist in the real world.

Puzzle 1: Two Different Maps, One Island

Mathematicians have two main ways to figure out the rules for the best route. Let's call them Team A and Team B.

• Team A (The Set-Separation Team): They look at the island and draw a fence around it. They ask, "Can we draw a line that separates our ship from the island without touching it?" They use a very strict, geometric definition of "touching" (called the Clarke Tangent Cone).
• Team B (The Penalization Team): They pretend the island is a giant, sticky wall. If you get too close, you get a huge fine. They ask, "If we make the fine infinitely big, where does the ship stop?" They use a slightly different, more flexible definition of "touching" (called QDQ Approximating Cones).

The Problem: Usually, Team A and Team B give you different answers. They are looking at the island through different lenses, so their "rules for the best path" don't match up. It's like one person saying, "You can't enter the park," and the other saying, "You can't step on the grass," but they are describing different boundaries.

The Solution: The authors found a special condition where these two teams finally agree. They discovered that if the island's edge is "smooth enough" (mathematically, if it's Quasi Prox-Regular or looks like a nice, round ball locally), then Team A's strict fence and Team B's sticky wall actually describe the exact same boundary.

Why this matters: It allows mathematicians to combine the best tools from both teams. They can use the powerful, flexible math of Team B to prove things, and then translate those results into the strict, rigorous language of Team A.

---

Puzzle 2: The "Gap" in the Road

Now, let's talk about the Infimum Gap.

Imagine you are looking for the cheapest way to get to the island.

• Scenario 1: You find a route that costs $100. You check all other routes, and none are cheaper. Great!
• Scenario 2 (The Gap): You find a route that costs $100. But then you realize, "Wait, if I allow my ship to do something impossible—like teleport or move at infinite speed—I could get there for $50."

In math, this is called an Infimum Gap. The "real" best price is $100, but the "theoretical" best price (allowing impossible moves) is $50. The gap is the difference.

Usually, if a gap exists, it means the "impossible" route is the only way to get the lowest cost. But here is the scary part: If a gap exists, the math that tells us we are on the best path breaks down.

The paper proves a very specific and surprising connection:

1. If a gap exists: The mathematical "compass" (called the Maximum Principle) that guides the ship stops working normally. It becomes Abnormal.
   • Analogy: Imagine your GPS says, "Turn left," but the compass needle is spinning wildly and pointing nowhere. The system is "abnormal" because the cost of the trip has become irrelevant to the direction; the system is so desperate to find a path that it ignores the fuel cost entirely.
2. The Twist: Before this paper, we knew this "Abnormal" behavior happened for the impossible routes (the ones with infinite speed). But we didn't know if it happened for the real routes (the ones with normal speed) if they were stuck with a gap.

The Big Discovery: The authors used their "Team A + Team B" compatibility trick to prove that even for real, normal ships, if there is a gap between the real world and the theoretical "impossible" world, the ship's guidance system will go "Abnormal."

The "Higher-Order" Secret Sauce

The paper goes one step further. It's not just about the basic direction (First Order). It's about the curvature and the twists of the path (Higher Order).

Think of driving a car.

• First Order: "Turn the wheel left."
• Higher Order: "Turn the wheel left, but also account for how the road curves and how the car's suspension bounces."

The authors show that if a gap exists, the "Abnormal" signal isn't just a simple "stop." It's a complex, high-level signal involving Lie Brackets.

• Analogy: A Lie Bracket is like asking, "If I go forward, then turn, then go backward, then turn again, where do I end up?" It measures the hidden complexity of the movement.
• The paper proves that if a gap exists, these complex, hidden movements (Lie Brackets) must also be "Abnormal" (zeroed out or broken).

Summary in Plain English

1. Two Methods, One Truth: The authors fixed a disagreement between two mathematical methods for finding the best path. They found that if the destination is "smooth enough," both methods agree.
2. The Gap Warning: They proved that if there is a "gap" between what is possible in the real world and what is possible in a theoretical "super-world" (with infinite speed), the mathematical rules for the best path break down.
3. The Abnormal Signal: This breakdown isn't just a simple error; it's a complex, high-level failure (involving Lie Brackets) that affects even the normal, real-world paths.
4. Why it helps: This helps engineers and scientists know when a solution is unstable. If they see this "Abnormal" signal, they know their "best" solution is actually a mirage, and they need to rethink the problem or the constraints.

In short: If the math says "Go!" but the rules of the game are broken (a gap), the compass spins wild (Abnormality), and this happens even for the best real-world routes.

Technical Summary

Here is a detailed technical summary of the paper "Minimizers that are not Impulsive Minimizers and Higher Order Abnormality" by Motta, Palladino, and Rampazzo.

1. Problem Statement

The paper addresses two interconnected challenges in the theory of optimal control, specifically concerning problems with unbounded controls and final state constraints (targets):

1. Compatibility of Necessary Conditions: There are two classical approaches to deriving necessary conditions for optimality: the set-separation approach (based on geometric separation of reachable sets) and the penalization approach (based on Ekeland's variational principle). These methods often yield non-equivalent conditions because they rely on different notions of tangency to the target set $S$. The set-separation approach typically uses "approximating cones" (like the Boltyanski cone), while penalization techniques often utilize the Clarke tangent cone. The authors investigate under what conditions the Clarke tangent cone is also a valid "approximating cone" (specifically a Quasi Differential Quotient (QDQ) approximating cone), thereby unifying the two approaches.
2. Infimum Gap Phenomena: In optimal control with unbounded controls, the infimum of the cost functional may not be attained within the class of "strict-sense" (standard) controls. This leads to an impulsive extension of the problem. A critical issue is the infimum gap: a situation where the infimum of the extended problem is strictly lower than that of the original problem.
   • The paper distinguishes between two types of gaps:
     • Type 1: An extended minimizer has a strictly lower cost than any nearby strict-sense process.
     • Type 2: A local strict-sense minimizer fails to be a local minimizer for the extended problem (i.e., there exist extended processes arbitrarily close to it with lower costs).
   • While the link between Type 1 gaps and "abnormality" (where the cost multiplier is zero) in higher-order Maximum Principles is known, the link for Type 2 gaps involving strict-sense minimizers and higher-order conditions (involving Lie brackets) was previously unestablished.

2. Methodology

The authors employ a hybrid methodology combining nonsmooth analysis, geometric control theory, and topological arguments:

• QDQ Approximating Cones: They utilize the concept of Quasi Differential Quotient (QDQ) approximating cones. These are generalized differential structures that encompass various types of approximating cones used in the literature.
• Quasi Prox-Regular Sets: To bridge the gap between the Clarke tangent cone and QDQ cones, the authors introduce the notion of Quasi Prox-Regular sets. A set is Quasi Prox-Regular at a point if it either locally contains the cone generated by its Clarke tangent cone or coincides locally with an $r$-prox-regular set (a set with positive reach, generalizing convex and $C^2$ sets).
• Impulsive Extension via Graph Completion: The unbounded control problem is embedded into an extended problem using a time-reparameterization technique (graph completion). This transforms unbounded controls into bounded controls on a variable time interval, allowing for impulsive trajectories (where the state jumps instantaneously).
• Topological Limit Arguments: To connect strict-sense minimizers to extended processes, the authors prove that a strict-sense minimizer exhibiting an infimum gap is the limit (in the $L^1$-control topology) of a sequence of isolated extended processes (processes that exhibit a local infimum gap).
• Higher-Order Maximum Principle: They apply a Maximum Principle involving iterated Lie brackets of the system vector fields. This allows for the detection of abnormality even when first-order conditions are normal.

3. Key Contributions

A. Theoretical Compatibility Result (Result 1)

The authors establish sufficient conditions under which the Clarke tangent cone $T^C_S(\bar{x})$ is a QDQ approximating cone.

• Theorem 3.1: If the target set $S$ is Quasi Prox-Regular at the target point $\bar{x}$ (satisfying either a local invariance condition or coinciding locally with an $r$-prox-regular set), then $T^C_S(\bar{x})$ is a QDQ approximating cone.
• Significance: This result validates the use of the Clarke tangent cone within the set-separation framework. It implies that necessary conditions derived via set-separation (using QDQ cones) can be reformulated using the Clarke normal cone, facilitating the application of these conditions to problems where the target is non-smooth but prox-regular.

B. Connection Between Strict-Sense Gaps and Higher-Order Abnormality (Result 2)

The paper provides a novel characterization of infimum gaps for strict-sense minimizers.

• Theorem 5.3: If a local strict-sense $L^1$-minimizer exhibits a local infimum gap with respect to the impulsive extension, and the target set satisfies the Quasi Prox-Regularity hypothesis, then this minimizer is abnormal for a Higher-Order Maximum Principle (involving Lie brackets) when the Clarke tangent cone is used as the approximating cone.
• Mechanism: The proof relies on showing that the strict-sense minimizer is the limit of a sequence of isolated extended processes. Since isolated extended processes are known to be abnormal for higher-order principles (Fact 1), and the adjoint variables converge, the limit (the strict-sense minimizer) inherits this abnormality.

4. Main Results

1. Compatibility of Cones: The Clarke tangent cone serves as a QDQ approximating cone for Quasi Prox-Regular sets. This unifies the transversality conditions of the set-separation and penalization approaches.
2. Fact 1 (Extended Processes): A local extended minimizer with an infimum gap is necessarily abnormal for a higher-order Maximum Principle (Theorem 5.1).
3. Theorem 5.3 (Strict-Sense Minimizers): A local strict-sense minimizer with an infimum gap is abnormal for a higher-order Maximum Principle, provided the target is Quasi Prox-Regular. This extends the "abnormality-gap" correspondence from extended processes to strict-sense processes and from first-order to higher-order conditions.
4. Counter-Example: The paper includes an illustrative example (Section 4.4) demonstrating a system where a strict-sense minimizer exists but is not a local minimizer for the extended problem, exhibiting an infimum gap.

5. Significance

• Bridging Theoretical Gaps: The paper resolves a long-standing issue regarding the non-equivalence of necessary conditions derived from different variational approaches. By identifying the class of sets (Quasi Prox-Regular) where these approaches coincide, it strengthens the theoretical foundation of optimal control with state constraints.
• Higher-Order Analysis: It demonstrates that first-order necessary conditions (standard Maximum Principle) are insufficient to detect infimum gaps in all cases. Specifically, a process might be "normal" at the first order but "abnormal" at a higher order (involving Lie brackets), which is the true indicator of an infimum gap.
• Strict-Sense vs. Extended: The results clarify the stability of optimal solutions. If a strict-sense minimizer is abnormal for a higher-order principle, it suggests that the solution is unstable under perturbations that allow for impulsive controls (the extended problem), explaining why the infimum might not be attained in the strict-sense class.
• Applicability: The findings are particularly relevant for systems with unbounded controls (common in engineering and economics) where impulsive controls are physically or mathematically necessary to model rapid changes, and where the target sets may have corners or non-smooth boundaries.

In summary, the paper provides a rigorous framework to detect when optimal solutions to unbounded control problems are unstable (infimum gaps) by linking the topological properties of the target set to the algebraic structure of higher-order necessary conditions (Lie brackets) via the unifying concept of QDQ approximating cones.