🔢 mathematics

Approximate controllability of a bilinear wave equation and minimum time

This paper establishes the global approximate controllability and determines the minimum control time for a bilinear Klein-Gordon wave equation on a $d$ -dimensional torus with controls on the first $(2d+1)$ Fourier modes, proving that the minimum time equals the maximum radius of the initial state's essential zero set in dimensions one and two, while showing zero minimum time is achievable in higher dimensions if the zero set has zero Lebesgue measure.

Original authors: Karine Beauchard, Thomas Perrin, Eugenio Pozzoli

Published 2026-01-28

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Karine Beauchard, Thomas Perrin, Eugenio Pozzoli

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 giant, invisible drum skin (the "wave") that is stretched over a multi-dimensional donut shape (the "torus"). You can't touch the drum skin directly. Instead, you have a set of "magic wands" (controls) that can gently tap or press on specific, very limited spots on the drum. Your goal is to make the drum vibrate in any pattern you want, no matter how complex.

This paper is a mathematical investigation into how fast you can do this steering, and what obstacles might stop you.

Here is the breakdown of their findings using simple analogies:

1. The Setup: The Drum and the Magic Wands

The System: The drum skin follows the laws of physics (specifically, the Klein-Gordon wave equation). It has a natural way of vibrating.
The Controls: You don't have infinite wands. You only have $2d + 1$ wands (where $d$ is the number of dimensions). These wands can only tap on the "lowest notes" of the drum (the first few Fourier modes).
The Goal: "Approximate Controllability." This means you don't need to hit the target pattern perfectly (like hitting a bullseye with a dart), but you need to get arbitrarily close to it. If you can get within a hair's breadth of the desired vibration, you win.

2. The Big Obstacle: The "Silent Zone"

Imagine the drum skin is currently vibrating. There might be a patch of the drum that is perfectly still (not moving up or down). The authors call this the Zero Set ( $Z(W_0)$ ).

The Problem: If you have a large, solid patch of the drum that is completely silent, your magic wands (which only tap on specific low notes) cannot instantly "wake up" that silent patch.
The Speed Limit: Waves travel at a finite speed. If you have a silent island in the middle of the drum, it takes time for the "wake-up signal" to travel from the edge of the island to its center.
The Minimum Time: The authors define a number called $r(W_0)$ $r (W_{0})$ . Think of this as the radius of the largest silent circle on your drum.
- The Rule: You cannot control the drum faster than the time it takes for a wave to cross that silent circle. If your silent circle has a radius of 5 meters, and waves travel at 1 meter per second, you simply cannot control the drum in less than 5 seconds.

3. The Main Discoveries: How Fast Can We Go?

The authors found that the answer depends heavily on the dimension of the space (how many directions the drum can vibrate in).

Case A: Low Dimensions (1D and 2D)

The Scenario: Think of a 1D string (like a guitar string) or a 2D drum skin.
The Result: The minimum time you need is exactly the time it takes for a wave to cross the largest silent circle ( $r(W_0)$ ).
The Metaphor: In a 1D or 2D world, the "wake-up signal" spreads out perfectly efficiently. As soon as the wave has had enough time to physically reach the center of the silent zone, you can take over the whole drum. You can't do it any faster, but you don't have to wait any longer either.

Case B: High Dimensions (3D and up)

The Scenario: Think of a 3D volume of air or higher-dimensional space.
The Result: It gets tricky.
- If the silent zone is "thin" (has zero volume): If the silent part is just a few scattered points or a thin line (like a needle), you can control the drum instantly (time = 0). The wave can "squeeze" through these tiny gaps immediately.
- If the silent zone is "thick" (has volume): If there is a solid block of silence, you might need a long time. However, the authors prove that if you wait long enough, you can always eventually control the drum, no matter how big the silent block is.
- The Catch: In 3D and higher, the wave doesn't spread out as neatly as in 2D. Sometimes, even after the wave reaches the center of a silent zone, it might not be strong enough to take control immediately. You might have to wait for the wave to bounce around and build up enough energy.

4. How Did They Prove This? (The "Lie Bracket" Trick)

The authors used a clever mathematical technique called Lie Bracket techniques (inspired by Agrachev and Sarychev).

The Analogy: Imagine you are trying to drive a car that can only move forward or turn left. You can't move sideways. But, if you move forward, turn left, move backward, and turn right, you can actually end up moving sideways.
The Math: They showed that by rapidly switching their "magic wands" on and off in specific patterns, they could create "virtual forces." These virtual forces allowed them to manipulate the drum in ways the individual wands couldn't do alone.
The Strategy:
1. Wake up the velocity: First, they used these tricks to make the drum skin start moving (change its speed) without changing its shape too much.
2. Transfer the motion: Then, they used the movement to change the shape of the drum.
3. Wait for the wave: If there was a big silent zone, they let the natural physics of the wave do the work of spreading the energy until the whole drum was "awake," and then they took control.

Summary

The Question: How fast can we steer a wave equation using limited controls?
The Answer: It depends on the size of the "silent" areas on the wave.
- In 1D and 2D, the answer is exactly the time it takes for a wave to cross that silent area.
- In 3D and higher, if the silent area is tiny, you can do it instantly. If it's big, you can do it eventually, but you might have to wait longer than the simple "crossing time" because the wave behaves differently in higher dimensions.

The paper essentially draws a map of the "speed limits" for controlling waves, showing that the geometry of the starting state and the number of dimensions are the most critical factors.

Technical Summary: Approximate Controllability of a Bilinear Wave Equation and Minimum Time

Problem Statement
This paper investigates the global approximate controllability (GAC) of a bilinear Klein-Gordon wave equation posed on a $d$ -dimensional torus $\mathbb{T}^d$ . The system is controlled by a potential supported on the first $(2d+1)$ Fourier modes (specifically, the constant mode and the first sine/cosine modes for each dimension). The state space is $H^1 \times L^2(\mathbb{T}^d)$ , and the control $u(t)$ is a piecewise constant function taking values in $\mathbb{R}^{2d+1}$ .

A central focus is the determination of the minimum time $T(W_0)$ required to steer an initial state $W_0$ to any target state in the space, up to an arbitrary precision. Due to the finite speed of propagation, a lower bound for this time is established as $r(W_0)$ , defined as the maximum radius of a ball contained within the set of essential zeroes $Z(W_0)$ of the initial profile. The paper addresses two primary questions:

Is the condition $\text{Int}(Z(W_0)) = \emptyset$ (no interior zeroes) sufficient for small-time approximate controllability ( $T(W_0)=0$ )?
Is the minimum time exactly equal to $r(W_0)$ for all initial conditions?

Methodology
The proof strategy combines two distinct technical approaches:

Lie Bracket Techniques (Agrachev-Sarychev): The authors utilize commutator methods to approximate flows generated by operators not directly present in the control system. By computing commutators of the drift operator $A + VB$ and the control potentials $V_j B$ , they generate new operators. Specifically, they show that the system can approximate flows of the form $e^{\phi B}$ (changing velocity while keeping the profile) and $e^{B^*}$ (changing the profile while keeping the velocity) for a wide class of functions $\phi$ . This relies on strong convergence criteria for operator groups in infinite-dimensional spaces.
Propagation of Well-Prepared States: The authors analyze the free evolution of the wave equation (drift only) to identify "well-prepared" states. They demonstrate that for specific initial conditions (e.g., non-negative velocity), the free evolution produces a strictly positive profile in finite time. Once a state with a strictly positive profile (and thus an empty zero set) is reached, the Lie bracket techniques allow for instantaneous control to any target.

Key Contributions and Results

Small-Time Controllability for Zero-Measure Sets:
The paper proves that if the set of zeroes of the initial profile $w_0$ has zero Lebesgue measure ( $|Z(w_0)| = 0$ ), the system is small-time approximately controllable ( $T(W_0) = 0$ ). This provides a partial affirmative answer to the question of whether $\text{Int}(Z(W_0)) = \emptyset$ implies small-time controllability, as the zero-measure condition is strictly stronger.
Exact Minimum Time in Low Dimensions ( $d \in \{1, 2\}$ ):
For dimensions $d=1$ and $d=2$ with zero potential ( $V=0$ ), the authors prove that the minimum time for global approximate controllability is exactly $r(W_0)$ . This is the first instance where the minimal time for bilinear control of the wave equation is determined precisely. The proof utilizes explicit representation formulas (D'Alembert for $d=1$ , Poisson for $d=2$ ) to show that the free evolution eliminates zeroes within time $r(W_0)$ .
Large-Time Controllability in Higher Dimensions ( $d \ge 3$ ):
- If $|Z(W_0)| = 0$ , the system is small-time controllable ( $T(W_0)=0$ ) regardless of dimension.
- If $|Z(W_0)| > 0$ , the system is controllable in sufficiently large time ( $T(W_0) < \infty$ ) for any non-zero initial condition, provided the set of zeroes is closed (for $d \ge 4$ ) or under specific regularity assumptions. In dimension $d=3$ , the closedness assumption is not required.
Velocity and Profile Transfer:
The paper establishes two fundamental "instantaneous" transitions (in the limit of small time):
1. Velocity Control: Changing the velocity component $\dot{w}$ to any target $\dot{w}_f$ while maintaining the profile $w$ , provided the change vanishes on the zeroes of $w$ .
2. Profile Transfer: Transferring the velocity into the profile (changing $w$ to $w + \dot{w}$ ) while maintaining the velocity.
  These transitions form the building blocks for the controllability proofs.

Significance and Claims
The paper claims to provide the first example of bilinear control for the wave equation where the minimal time for global approximate controllability can be explicitly determined (specifically for $d \in \{1, 2\}$ ). It extends the Agrachev-Sarychev method, originally developed for Navier-Stokes and Schrödinger equations, to the context of wave equations with degenerate (finite-mode) controls.

The authors note that the regularity assumption $W_0 \in H^1 \times H^1$ (stronger than the natural $H^1 \times L^2$ ) is a technical necessity for the profile transfer step, as the operation requires the velocity to be at least as regular as the profile.

Limitations and Open Problems
The paper explicitly acknowledges limitations regarding the closedness of the zero set $Z(W_0)$ in higher dimensions. It notes that for $d \ge 4$ , the assumption that $Z(W_0)$ is closed is necessary for the large-time controllability result, and it remains an open question whether this assumption can be removed. Furthermore, the paper does not claim to solve the exact controllability problem (which is known to be impossible in these spaces) nor does it provide results for the general case of $V \neq 0$ in determining the exact minimal time, focusing instead on the $V=0$ case for the sharp time results.