Maximally localized modes of a multimode fiber

This paper introduces an optimization method to generate maximally localized optical modes in multimode fibers—analogue to Wannier functions—which spontaneously organize into concentric rings with complex, non-regular patterns for large mode counts, thereby enabling the design of physically grounded photonic lanterns for arbitrary bundle geometries.

The Gist

Imagine you have a giant, magical flashlight (a multimode fiber) that can carry many different "colors" of light at the same time. In physics, these different colors are actually different patterns of light traveling down the wire.

Usually, scientists describe these patterns using a standard mathematical alphabet (like the Laguerre-Gaussian basis). But here's the catch: just like you can rearrange the letters in the alphabet to spell different words, you can mix and match these light patterns in infinite ways. They all carry the same amount of information, but they look very different.

The question this paper asks is: "What is the most compact, 'tightest' way to arrange these light patterns?"

Think of it like packing a suitcase. You have a bunch of clothes (the light modes). You could stuff them in randomly, or you could fold them perfectly to take up the least amount of space. This paper invents a super-smart algorithm that acts like a master packer, finding the arrangement where every single "piece of light" is as small and concentrated as possible.

Here is the breakdown of their discovery using everyday analogies:

1. The "Self-Organizing" Party

The researchers didn't tell the light how to arrange itself. They didn't say, "Okay, form a circle!" or "Make a square!" They just said, "Get as small as possible."

The Result: The light patterns spontaneously organized themselves into concentric rings, like ripples in a pond or the rings of a tree.

• The Analogy: Imagine a group of people in a dark room who are told to stand as close to the center as possible without bumping into each other. Without being told to form circles, they would naturally cluster into rings. The light did the exact same thing.

2. The "Snowflake" vs. The "Cookie Cutter"

Previous methods of designing these fiber bundles were like using a cookie cutter. They assumed every spot of light was a perfect, identical circle (like a cookie), and they tried to pack as many circles as possible into a space.

The New Discovery: The light patterns are not identical cookies.

• The Analogy: Think of the light spots as snowflakes. The ones in the center are small and round. As you move to the outer rings, the "snowflakes" get bigger and stretch out into ovals (ellipses). They change shape depending on where they are in the ring. The old "cookie cutter" methods couldn't predict this because they assumed everything was the same shape. This new method sees the unique personality of every single light spot.

3. The "Perfect" vs. The "Good Enough"

For smaller numbers of light patterns, the rings were perfectly symmetrical. But when they crammed in a lot of patterns (55 modes), the perfect symmetry broke down.

• The Analogy: Imagine trying to arrange 55 people in a perfect circle. Eventually, it becomes impossible to keep everyone perfectly equidistant. Some people have to stand slightly closer or further apart, and the circle gets a little wobbly. The light did the same thing: it found a "wobbly" arrangement that was actually more efficient (smaller total size) than a perfect circle.

4. The "Remote Control" for Designers

The researchers also built a "constrained" version of their tool. This is like giving the master packer a remote control.

• How it works: If a designer really wants a perfect circle of fibers (maybe for a specific camera or telescope), they can use this tool to force the light into that shape.
• The Trade-off: The tool calculates exactly how much "wasted space" (spread) you are paying for that perfect shape. It tells you: "Hey, if you force this perfect circle, your light will be only 1.5% less efficient than the natural, messy arrangement." This helps engineers decide if the perfect shape is worth the tiny loss in performance.

Why Does This Matter? (The "Photonic Lantern")

This research is crucial for building Photonic Lanterns.

• What is a lantern? It's a device that takes a big, messy bundle of light from a telescope (like looking at a star) and splits it into many tiny, clean beams for a computer to analyze.
• The Problem: If the tiny beams aren't arranged just right, you lose information or the signal gets messy.
• The Solution: This paper gives engineers a blueprint. Instead of guessing where to put the fibers, they can now calculate the exact natural shape the light prefers. It's like finally having a map of the "natural habitat" of light, rather than forcing it to live in a house that doesn't fit.

In a Nutshell

This paper teaches us that light has a natural instinct to organize itself into rings, but those rings aren't perfect circles made of identical dots. They are dynamic, changing shapes that evolve as you add more light. By letting the light "choose" its own best shape, we can build better, more efficient tools for astronomy and high-speed internet.

Technical Summary

1. Problem Statement

Multimode fibers (MMFs) support an orthonormal basis of spatial modes, but this basis is not unique; any unitary transformation of a valid basis yields another valid basis. In applications like Photonic Lanterns (devices converting MMFs to bundles of single-mode fibers for Space-Division Multiplexing or astrophotonics), the spatial arrangement of the output cores is a critical design parameter.

Current design approaches rely on two main assumptions:

1. Modal Compatibility: Arranging cores based on radial and azimuthal mode groups (e.g., one ring per radial group).
2. Geometric Packing: Treating cores as identical disks and optimizing for maximum packing density (circle packing).

The Gap: Neither approach asks what geometry the fiber modes themselves intrinsically prefer. Existing methods impose geometric constraints on the modes rather than deriving the optimal geometry from the modal physics. This paper addresses the question: What is the most spatially concentrated basis for a given set of fiber modes, and what geometry does it spontaneously adopt?

2. Methodology

The author proposes an optimization framework to find Maximally Localized Fiber Modes (MLFM).

• Objective Function: The method minimizes the total spatial spread ($\Omega$) of the modes. For a basis $\{\psi_i\}$, the spread is defined as the sum of the variances of the intensity-weighted spatial coordinates:
  $$\Omega(U) = \sum_{i=1}^N \sigma_i^2, \quad \text{where } \sigma_i^2 = 2(\langle x^2 \rangle_i - \langle x \rangle_i^2 + \langle y^2 \rangle_i - \langle y \rangle_i^2)$$
  This is the optical analogue of the Marzari–Vanderbilt functional used to find maximally localized Wannier functions in solid-state physics.

• Optimization Space: The search is conducted over the unitary group $U(N)$. To avoid manifold constraints, the unitary matrix $U$ is parameterized as $U = \exp((A - A^\dagger)/2)$, where $A$ is an unconstrained complex matrix.

• Implementation:

  • Framework: Implemented using FluxOptics.jl, a differentiable wave optics framework in Julia.
  • Solver: Parameters are updated using Nesterov-accelerated gradient descent.
  • Basis: The study applies the method to the Laguerre-Gaussian (LG) basis of a graded-index fiber, covering mode counts $N$ from 6 to 55 (complete mode groups).

• Constrained Variant: A penalty term is added to the functional to force the mode centroids toward a target geometry $(x_0, y_0)$:
  $$\Omega_\beta(U) = \Omega(U) + \beta \sum_{i=1}^N [(\langle x \rangle_i - x_{0,i})^2 + (\langle y \rangle_i - y_{0,i})^2]$$
  This allows for "inverse design," quantifying the localization cost of imposing a specific bundle geometry.

3. Key Contributions

1. Definition of MLFM: Introduction of a rigorous method to derive the most spatially concentrated basis for MMFs without geometric priors, analogous to Wannier functions.
2. Discovery of Spontaneous Self-Organization: Demonstration that minimizing spatial spread causes modes to spontaneously organize into concentric rings, a structure emerging purely from modal physics rather than geometric constraints.
3. Characterization of Heterogeneity: Identification that MLFM spots are heterogeneous (unlike the identical disks assumed in circle packing). Outer spots are larger and more elliptical than inner ones, with azimuthal variance exceeding radial variance ($\sigma_\theta^2 > \sigma_r^2$).
4. Breakdown of Regularity: Discovery that for large mode counts ($N \geq 36$), the regular recurrence relation for ring populations ($n_k = n_{k-1} + 4$) breaks down. The optimizer finds asymmetric configurations where spots within a ring are not identical, and multiple topologically distinct solutions exist as stable local minima.
5. Quantitative Design Tool: The constrained functional $\Omega_\beta$ provides a metric to evaluate how "expensive" (in terms of spatial spread) a specific photonic lantern geometry is, bridging modal physics and device design.

4. Results

• Spontaneous Ring Formation: For all tested $N$ (6 to 55), the unconstrained optimizer converged to concentric ring structures.
  • For $N \leq 28$, the ring populations followed two regular families: starting with 1 spot ($n_1=1$) or 3 spots ($n_1=3$), with subsequent rings increasing by 4 spots ($n_k = n_{k-1} + 4$).
  • For $N \geq 36$, the regular progression failed. The optimizer found asymmetric solutions (e.g., for $N=36$, the distribution was $[3, 7, 13, 13]$ rather than the symmetric $[3, 7, 13, 13]$ or similar regular patterns).
• Spot Geometry Evolution: The spots are not circular. They become larger and more elliptical as the ring index increases. The elongation is azimuthal, reflecting the curvature of the ring.
• Constrained Optimization Performance:
  • When forcing the optimizer to target regular symmetric geometries (using $\beta=1$), the algorithm successfully recovered near-optimal symmetric solutions.
  • Cost: The total spread of these constrained symmetric solutions was only 0.8% to 1.7% higher than the unconstrained global minimum.
  • Quasi-Symmetry: The constrained solutions are "quasi-symmetric." While intensity profiles look identical, the phase structures (nodal lines) differ between spots in the same ring to maintain orthogonality. This reveals that the unconstrained problem does not have a true continuous rotational symmetry minimum; the symmetric solutions are local minima only when constrained.
• Robustness: The method is robust to imprecise target geometry specifications; it preserves angular symmetry while freely adjusting ring radii to minimize spread.

5. Significance

• Fundamental Physics: The work reveals that the intrinsic spatial structure of graded-index fiber modes is a set of concentric, heterogeneous rings. This challenges the assumption that fiber modes can be perfectly mapped to identical, geometrically packed cores.
• Photonic Lantern Design: The method offers a new paradigm for designing photonic lanterns. Instead of assuming a geometry and checking compatibility, designers can:
  1. Use the unconstrained method to find the "natural" geometry of the fiber.
  2. Use the constrained method to quantify the penalty of deviating from this natural geometry to meet manufacturing or system constraints.
• Beyond Circle Packing: The results demonstrate that geometric circle packing (treating cores as identical disks) is insufficient for high-mode-count fibers. The optimal packing is dictated by the modal content, leading to non-uniform spot sizes and shapes that geometric approaches cannot predict.
• Generalizability: While demonstrated on LG modes, the framework is basis-independent and applicable to any modal family or fiber geometry, providing a universal tool for mode analysis and device optimization.