Frenet turns

This paper resolves a problem posed by A. Agrachev regarding the minimum number of traversals required for a circle in $\mathbb{R}^n$ to admit a deformation with a nondegenerate Frenet frame, revealing that the answer depends on the chosen topology and introducing decorated turn data to characterize accessibility via constant or time-dependent controls across various dimensions.

The Gist

Imagine you are drawing a perfect circle on a piece of paper. Now, imagine you want to lift that circle off the paper and wiggle it slightly in 3D space (or even higher dimensions) so that it never "flattens out" or loses its twist. The question mathematician Boris Shapiro is answering is: How many times do you need to draw that circle before you can wiggle it without it ever becoming flat?

The paper explores this question through three different "lenses" or ways of looking at the problem. Here is the breakdown using simple analogies.

1. The "Rough Sketch" View (The Literal Topology)

The Question: If I draw a circle $k$ times on top of itself, can I wiggle it just a tiny bit so it becomes a "perfect" 3D (or $n$-dimensional) curve that never flattens?

The Answer:

• In 2D (on paper): You only need to draw it once. A single circle is already "perfect" in 2D.
• In 3D: You need to draw it twice. If you try to wiggle a single circle in 3D, it inevitably gets "flat" at some point (like a pancake). But if you draw it twice (a double loop), you can wiggle it into a shape that twists everywhere. This is a famous result known as the Fenchel-Milnor phenomenon.
• In 4D and higher: Surprisingly, you only need to draw it once. Even though it seems like higher dimensions are harder, the extra space actually makes it easier to wiggle a single circle into a non-flat shape.

The Catch: This answer is based on a very specific, "rough" definition of "wiggling." It allows the curve to change shape drastically in terms of its internal "twistiness" (curvature), as long as the final shape looks very similar to the circle.

2. The "Strict Driver" View (The Control Problem)

The Question: What if we demand that the "steering wheel" (the mathematical controls that define the curve's twist) stays small and smooth? Can we still wiggle the circle?

The Problem:
In dimensions 4 and higher, if you try to keep the "normal" part of the steering wheel fixed (like keeping the wheels of a car pointed in one specific direction while driving), it is impossible.

• The Analogy: Imagine trying to drive a car in a circle while keeping the rear wheels locked in a straight line. In 4D space, the laws of geometry (specifically a "spherical obstruction") say you simply cannot do this without the car crashing or the steering wheel spinning infinitely.
• The Result: If you insist on this strict "fixed steering" rule, the answer is: You can never do it, no matter how many times you loop the circle. The number of turns required is infinite.

3. The "Decorated" View (The New Solution)

The Solution: Since the "Strict Driver" view leads to a dead end in higher dimensions, Shapiro suggests we change the rules slightly. Instead of locking the steering wheel, we allow the "normal" part of the steering to rotate, but we must count how many times it rotates.

The New Rule:
We describe the curve not just by how many times the main circle loops ($p$), but also by how many times the "side" of the curve rotates ($q$). We call this a "Decorated Turn Vector" $(p, q)$.

• In 4D: You need a pair of numbers, like $(1, 2)$. This means the main circle loops once, but the "side" rotates twice.
  • The Discovery: If the two numbers are different (non-resonant), you can wiggle the curve successfully.
  • The Winner: The simplest successful shape isn't a simple circle $(1, 0)$, but a shape that loops once while twisting twice $(1, 2)$.
• In Higher Even Dimensions (6D, 8D, etc.): You need a list of numbers $(p_1, p_2, \dots)$. As long as all the numbers in the list are different, you can wiggle the curve.
• In Odd Dimensions (5D, 7D, etc.): It's trickier. You can't just use a constant "steering" setting; you have to constantly adjust the steering wheel over time to cancel out a natural "drift" that happens in odd dimensions.

Summary of the Three Takeaways

1. If you just want the shape to look like a circle: In high dimensions, 1 loop is enough.
2. If you demand the steering be perfectly rigid: In high dimensions, it's impossible (infinite loops needed).
3. If you allow the steering to rotate but count the rotations: In high dimensions, you need a specific mix of rotations (like 1 main loop + 2 side twists). This is the "sweet spot" where the problem becomes solvable and interesting again.

The Big Picture:
The paper teaches us that the answer to "how many turns?" depends entirely on how strictly you define the rules. By relaxing the rules just enough to allow the "side" of the curve to rotate (but counting those rotations), we find a beautiful, solvable mathematical world where specific combinations of twists create perfect, non-flat loops.

Technical Summary

Technical Summary of "Frenet Turns" by Boris Shapiro

Problem Statement
This paper addresses a problem posed by A. Agrachev regarding the minimal number of turns a plane circle must undergo to admit a deformation into a nondegenerate curve in $\mathbb{R}^n$ with a nowhere-vanishing Frenet frame. The central difficulty lies in the interpretation of "small perturbation." The author distinguishes between two topologies:

1. The literal $C^n$ curve topology, where the curve itself is close to the original.
2. The Frenet control topology, where the Frenet curvatures (controls) are close to the degenerate controls of the plane circle.

The paper investigates the minimal number of turns, denoted $k(n)$ or $\mu(n)$, required for such deformations to exist, noting that the answer depends critically on which topology is considered and whether the normal frame is fixed or allowed to rotate.

Methodology and Key Contributions

1. Resolution of the Literal $C^n$ Curve Problem
The author first solves the problem in the literal $C^n$ topology, where the curve $\gamma$ must be close to the plane circle $c_m(t) = (\sin mt, \cos mt, 0, \dots, 0)$.

• Result: The paper establishes that for $n \ge 4$, a single turn ($m=1$) is sufficient to admit arbitrarily small nondegenerate perturbations.
• Construction: The proof utilizes an inductive construction based on Lemma 2.1. By constructing a smooth periodic curve $A$ in $\mathbb{R}^{n-2}$ with non-vanishing Wronskian and zero first Fourier harmonic, the author defines a perturbed curve $\gamma_\Sigma(t) = (\sin t, \cos t, \Sigma Y(t))$. This construction ensures the curve remains nondegenerate (the determinant of the first $n$ derivatives is non-zero) while converging to the plane circle in the $C^n$ norm as parameters $\Sigma \to 0$.
• Contrast: This result ($k(n)=1$ for $n \ge 4$) contrasts with the classical Fenchel–Milnor phenomenon in $\mathbb{R}^3$ where $k(3)=2$. The author emphasizes that this $C^n$ solution does not solve Agrachev's original control problem, as the Frenet controls of such perturbations do not necessarily converge to the degenerate controls.

2. The Obstruction in Fixed-Normal-Frame Control
The paper analyzes the "naive" interpretation of Agrachev's problem, where one attempts to approximate a multiply covered plane circle with a fixed normal frame using positive Frenet controls.

• Obstruction: Using a spherical Fenchel obstruction (Lemma 3.2), the author proves that for $n \ge 4$, no fixed normally framed multiply covered plane curve can be approached by positive Frenet controls in any norm topology that forces the last two controls ($u_{n-2}, u_{n-1}$) to vanish in $L^1$.
• Implication: The integral of the spherical length of the last frame vector must be at least $2\pi$. Consequently, in dimensions $n \ge 4$, the "fixed-normal-frame" interpretation yields no finite number of turns; the problem is ill-posed under these rigid constraints.

3. Decorated Turn Vectors and Accessible Data
To salvage a nontrivial turn-counting problem, the author introduces decorated turn data. Instead of fixing the normal frame, the boundary datum includes the rotation numbers of the normal planes.

• Dimension 4: The boundary datum is a pair $(p, q)$, where $p$ is the tangent turn and $q$ is the normal plane turn. The degenerate control is $(p, 0, q)$.
  • Theorem 1.4 / 4.2: The paper proves that every nonresonant pair $(p, q)$ with $p, q > 0$ and $p \neq q$ is strongly accessible. This means there exist positive constant controls $(a_\epsilon, \epsilon, c_\epsilon)$ that generate a closed curve with a closed Frenet frame, converging to the degenerate datum as $\epsilon \to 0$.
  • Resonance: The case $p=q$ (resonant) is shown to be inaccessible by constant controls because a positive middle control splits the double eigenvalue, preventing the frame from closing with the same frequency.
• Even Dimensions ($n=2r$): The concept generalizes to a vector $\mathbf{p} = (p_1, \dots, p_r)$.
  • Theorem 5.3: If the entries of $\mathbf{p}$ are pairwise distinct, the datum is strongly accessible via constant controls. The proof relies on an inverse-spectral argument using the implicit function theorem on the eigenvalues of the skew-symmetric Frenet matrix.
  • Model: The vector $(1, 2, \dots, r)$ is identified as the smallest nonresonant model with one tangent turn.
• Odd Dimensions ($n=2r+1$):
  • Obstruction: Proposition 5.6 demonstrates that no positive constant-control Frenet trajectory in odd dimensions can have both a closed frame and a closed base curve. This is due to the unavoidable drift in the zero-frequency direction (the kernel of the Frenet matrix).
  • Requirement: A solution in odd dimensions requires time-dependent controls to cancel this translational drift.

4. Frame Length Estimates
The paper provides bounds on the length of the Frenet frame $L_F(\gamma) = \int \sqrt{\sum u_i^2} dt$.

• For the decorated class $C_{(1,2)}$ in $\mathbb{R}^4$, the infimum of the frame length is bounded above by $2\pi\sqrt{5}$.
• Generally, for the model vector $(1, 2, \dots, r)$ in even dimensions, the upper bound is $2\pi\sqrt{\sum_{k=1}^r k^2}$.

Significance and Claims
The paper claims to clarify that Agrachev's original question hides three distinct problems:

1. The $C^n$ curve problem, which has a complete and somewhat surprising answer ($k(n)=1$ for $n \ge 4$).
2. The fixed-normal-frame control problem, which is impossible in dimensions $n \ge 4$ due to a spherical Fenchel obstruction.
3. The decorated turn problem, which the author proposes as the correct, nontrivial formulation.

The significance lies in shifting the boundary condition from a rigid fixed frame to a "decorated" frame that records the rotation of normal planes. This modification:

• Retains the geometric essence of counting turns.
• Removes the artificial obstruction found in the fixed-frame interpretation.
• Introduces a rich structure involving resonance (where constant controls fail) and dimension-dependent behavior (constant controls suffice for nonresonant even dimensions, while odd dimensions require time-dependent controls).

The author concludes that the decorated version serves as a natural finite-dimensional replacement for the original problem, offering a framework with genuine obstructions, constructive proofs for nonresonant cases, and open questions regarding resonant and odd-dimensional cases.