Higher-Order Normality and No-Gap Conditions in Impulsive Control with $L^1$-Control Topology

This paper establishes that a notion of higher-order normality, derived from iterated Lie brackets, is sufficient to prevent infimum gaps in impulsive extensions of control-affine systems under an $L^1$-control topology, thereby extending existing no-gap results beyond the more common $L^\infty$-trajectory framework.

The Gist

Here is an explanation of the paper, translated into everyday language with some creative analogies.

The Big Picture: The "Gap" Problem

Imagine you are trying to find the absolute best route to drive from your house to a friend's party. You want to minimize the time and fuel (this is your Optimal Control Problem).

However, the map you have is tricky. The roads are narrow, the traffic rules are strict, and maybe you can't even drive fast enough to get there in time. You try every possible path, but you can't find a single "best" route that actually exists. It's like chasing a horizon that keeps moving away.

To fix this, mathematicians often use a trick called Extension. They imagine a "super-car" that can drive faster, jump over obstacles, or even teleport for a split second. This "super-car" represents an Extended Control System. Because this super-car has more freedom, it can find a perfect, optimal route.

The Danger Zone (The Infimum Gap):
Here is the catch. Just because the super-car found a perfect route, doesn't mean the regular car can actually do it.

• The Gap: If the super-car's best time is 10 minutes, but the regular car's best possible time is 15 minutes, there is a Gap. The extension failed to help the original problem.
• The Goal: We want to know: When can we be sure there is NO gap? When can we trust that the super-car's solution is actually achievable by the regular car?

The Old Way vs. The New Way

For decades, mathematicians had a rule of thumb (called Normality) to predict if there would be a gap.

• The Rule: "If the math equations describing the super-car's best route look 'normal' (a specific technical condition), then there is no gap. The regular car can do it."

However, this rule was tricky. It depended heavily on how you measured "closeness."

• The Old Ruler ($L^\infty$): Imagine measuring the route by looking at the worst single moment. Did the car ever swerve wildly? If yes, the routes are "far apart." This is a very strict ruler.
• The New Ruler ($L^1$): Imagine measuring the route by looking at the total fuel used or the total distance traveled. A few small swerves don't matter as much as the total effort. This is a more flexible ruler.

The Problem: A famous mathematician (R.B. Vinter) showed a counter-example. He found a situation where the super-car's route looked "normal" under the strict ruler, but there was still a gap. This made people worry: "Does the 'Normality' rule even work if we use the more flexible, realistic ruler?"

What This Paper Does

This paper says: "Yes, the rule still works, even with the flexible ruler, but we need to look deeper."

The authors prove that if you check for "Normality" using Higher-Order Conditions (looking at complex interactions between the car's movements, like how turning the wheel affects the speed and the direction simultaneously), you can guarantee there is no gap.

The Analogy: The Tightrope Walker

Imagine a tightrope walker (the Regular Car) trying to cross a canyon.

1. The Extension: We imagine a safety net (the Super-Car) that allows the walker to bounce or fly slightly.
2. The Gap: If the safety net allows the walker to land on the other side, but the real walker falls, there is a gap.
3. The Measurement:
   • Old Way ($L^\infty$): We check if the walker ever wobbled more than 1 inch.
   • New Way ($L^1$): We check the total amount of wobbling over the whole trip.
4. The Discovery: The authors found that if you only check the total wobbling, you might miss a subtle, dangerous twist in the rope. But, if you look at Higher-Order Normality—checking how the twists interact with each other (like a knot in the rope)—you can predict with certainty that the walker will make it across without falling.

The Secret Weapon: "Set Separation"

How did they prove this? They used a technique called Set Separation.

Imagine two groups of people:

• Group A: People who can actually walk the path (Real solutions).
• Group B: People who can fly or teleport (Extended solutions).

The authors used a mathematical "fence" (based on Lie Brackets, which are fancy ways of describing how different movements combine) to separate these two groups.

• If the "fence" is strong enough (based on their new Higher-Order Normality check), it proves that Group A and Group B are actually touching. There is no gap between them.
• If the fence is weak, the groups might be far apart, and a gap exists.

Why This Matters

1. Realism: The "New Ruler" ($L^1$) is often more realistic for engineering problems where we care about total energy or total time, rather than just the worst single second.
2. Safety: It gives engineers and mathematicians a reliable checklist. If their system passes this "Higher-Order Normality" test, they can stop worrying about gaps and trust their extended models.
3. Mathematical Bridge: It connects the strict, old-school math with the more flexible, modern approach, showing that the fundamental rules of control theory hold up even when we change how we measure things.

Summary in One Sentence

This paper proves that by looking at the complex, "higher-order" interactions of a control system (like how different forces twist and turn together), we can guarantee that a simplified, idealized model of a system will always match the reality of the real system, even when we measure success by total effort rather than worst-case moments.

Technical Summary

Here is a detailed technical summary of the paper "Higher-Order Normality and No-Gap Conditions in Impulsive Control with L1-Control Topology".

1. Problem Statement

The paper addresses a fundamental issue in optimal control theory: the infimum gap.

• Context: When an optimal control problem lacks a minimizer within the original class of admissible controls (strict-sense controls), it is common to embed the problem into a larger "extended" problem (e.g., via relaxation or impulsive extensions) that guarantees the existence of a minimizer.
• The Gap: An "infimum gap" occurs if the minimum value of the extended problem is strictly lower than the infimum value of the original problem. This renders the extension useless for approximating the original solution.
• Specific Challenge: The authors focus on impulsive extensions of control-affine systems with unbounded controls. The novelty lies in analyzing this gap under the $L^1$-control topology (distance between control functions) rather than the standard $L^\infty$-trajectory topology.
• Motivation: Previous results established that "normality" of necessary conditions (where the cost multiplier is non-zero) prevents gaps in $L^\infty$ settings. However, a counterexample by R. B. Vinter showed that for certain extensions, normality with respect to the $L^1$-norm of controls is insufficient to guarantee no-gap. This paper seeks to identify conditions under which normality does suffice in the $L^1$ setting.

2. Methodology

The authors employ a rigorous mathematical framework combining geometric control theory and set-separation techniques.

• System Formulation:

  • Original System: Control-affine dynamics $\dot{x} = f(x) + \sum g_i(x)u_i$ with unbounded controls $u \in L^1$.
  • Extended System: Utilizes a time-reparameterization approach (graph-completion) where the state evolves in an extended space $(y^0, y, \beta)$. The control becomes $(w^0, w)$, where $w^0$ acts as a time-scaling factor. Impulsive behavior corresponds to intervals where $w^0 = 0$.
  • Topology: The analysis is conducted using the $L^1$-distance between control pairs $(w^0, w)$, defined as $d(z_1, z_2) = |S_1 - S_2| + \|(w^0_1, w_1) - (w^0_2, w_2)\|_{L^1}$.

• Key Analytical Tools:

  • Higher-Order Necessary Conditions: The paper utilizes conditions based on iterated Lie brackets of the system's vector fields. This goes beyond first-order Pontryagin Maximum Principle (PMP) conditions.
  • QDQ Approximating Cones: The authors use Quasi Differential Quotient (QDQ) approximating cones. This concept, introduced in prior works by the authors, generalizes classical tangent cones and allows for a robust "Open Mapping Theorem" essential for set-separation arguments.
  • Set-Separation Argument: The core proof strategy relies on showing that if a gap exists, the reachable set of the original (strict-sense) system and the reachable set of the extended system can be separated by a hyperplane defined by the adjoint variables.

• Technical Adaptation ($L^1$ vs. $L^\infty$):

  • The authors adapt the proof techniques from their previous work (which used $L^\infty$-trajectory distance).
  • They introduce Lemmas 4.1, 4.2, and 4.3 to handle the specific challenges of the $L^1$-control topology.
  • Rate Independence: They exploit the rate-independence of the extended system to show that the existence of a gap is invariant under time-reparameterization (equivalence classes of processes).
  • Canonical Processes: They demonstrate that it suffices to analyze "canonical" processes (where $w^0 + |w| = 1$ a.e.) to determine the existence of a gap, bridging the gap between general extended processes and embedded strict-sense processes in the $L^1$ metric.

3. Key Contributions

1. Extension to $L^1$-Topology: The paper successfully extends the "no-gap" results from the strong $L^\infty$-trajectory topology to the weaker $L^1$-control topology. This is significant because $L^1$ is the natural topology for unbounded controls and impulsive systems.
2. Higher-Order Normality as a Sufficient Condition: The authors prove that higher-order normality (based on iterated Lie brackets) is a sufficient condition to prevent an infimum gap in this specific setting.
   • Specifically, if an extended minimizer is a "normal higher-order extremal" (meaning the cost multiplier $\lambda \neq 0$ for all multipliers satisfying the higher-order conditions), then no gap exists.
3. Resolution of a Specific Counterexample Context: The work addresses the limitations highlighted by Vinter's counterexample. While Vinter showed that first-order normality in $L^1$ is insufficient for convexified extensions, this paper shows that for impulsive extensions, incorporating higher-order conditions (Lie brackets) restores the sufficiency of normality to prevent gaps.
4. Refined Technical Lemmas: The paper provides rigorous proofs (Lemmas 4.1–4.3) establishing the equivalence of gap definitions across different process classes (canonical vs. non-canonical) under the $L^1$ metric, a non-trivial technical hurdle.

4. Main Results

• Theorem 3.1 (Gap and Higher-Order Abnormality): If a feasible extended process exhibits a local infimum gap with respect to the $L^1$-control distance, then it must be an abnormal higher-order extremal with respect to any QDQ approximating cone of the target. In other words, if a gap exists, the cost multiplier $\lambda$ must be zero.
• Theorem 3.2 (Higher-Order Normality and No-Gap): Conversely, if a feasible extended process is a normal higher-order extremal (i.e., $\lambda \neq 0$), then no local infimum gap occurs at that process.
  • This implies that the infimum of the original problem coincides with the minimum of the extended problem.

5. Significance

• Theoretical Advancement: This work bridges a gap in the theoretical understanding of impulsive control. It clarifies that while first-order conditions are insufficient in the $L^1$ setting for certain extensions, the inclusion of higher-order geometric information (Lie brackets) restores the validity of the "normality implies no-gap" principle.
• Practical Implications: For engineers and applied mathematicians dealing with systems requiring impulsive controls (e.g., robotics, aerospace maneuvers, financial trading with transaction costs), this result provides a rigorous justification for using impulsive extensions. It assures that if the resulting optimal solution satisfies higher-order normality conditions, the solution is a valid approximation of the original constrained problem without a loss of optimality value.
• Methodological Robustness: The use of QDQ cones and set-separation techniques offers a powerful alternative to the traditional Ekeland variational principle, particularly for problems involving non-smooth dynamics or unbounded controls where standard perturbation arguments may fail.

In summary, the paper establishes that higher-order normality is the key to ensuring the absence of infimum gaps in impulsive control problems when analyzed under the natural $L^1$-control topology, thereby validating the use of impulsive extensions for solving difficult optimal control problems.