🔬 mesoscale physics

Near-field maps of topological edge states of molecular aggregates

This paper demonstrates that spatially resolved near-field spectroscopy, combined with a convolutional neural network, can robustly reconstruct the complex spatial structure of topological edge states in molecular aggregates, overcoming the limitations of traditional far-field optical probes.

Original authors: Aritra Mishra, Sidhartha Nayak, Alexander Eisfeld

Published 2026-10-26

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Aritra Mishra, Sidhartha Nayak, Alexander Eisfeld

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

In the microscopic world of materials science, researchers often look for ways to control how energy moves through a structure. Imagine a crowd of people passing a secret message; in some arrangements, the message gets stuck in the middle, while in others, it flows effortlessly along the edges. This behavior is known as topology, a concept from mathematics that describes how shapes and connections hold together even when they are twisted or stretched. In the realm of molecular physics, scientists have discovered that certain arrangements of molecules can create "edge states," where energy becomes trapped along the boundary of a material, protected from the chaos and disorder that might disrupt it in the center. These states are promising for future technologies because they could allow for highly efficient and robust energy transport, but they are notoriously difficult to see. Traditional methods of looking at materials from a distance often fail to reveal these edge states because the signals are too faint, and the distance blurs the specific location of the energy.

A team of researchers at the Max Planck Institute for the Physics of Complex Systems and the Technical University of Dresden has found a way to peer directly at these hidden states using a theoretical model of near-field spectroscopy. Instead of simulating light shining from far away, they model a tiny, needle-like probe hovering just nanometers above a flat sheet of molecules. This probe acts like a sensitive ear, listening to the energy states of the molecules as it scans across the surface. The researchers focused on a specific type of molecular sheet made of metalloporphyrins or metallophthalocyanines, which are ring-shaped molecules capable of holding energy. By applying a strong magnetic field in their simulations, they forced these molecules to behave in a very specific way, creating a complex interaction that naturally leads to the formation of these protected edge states.

When the researchers simulated scanning their probe across the molecular sheet, they expected to see a simple map showing where the energy was most likely to be found. However, the results were far more complicated. The maps they produced did not look like a straightforward picture of the energy's location. Instead, the signals showed strange, interference-like patterns, with bright spots and dark patches that seemed to contradict the actual distribution of the energy. In some cases, the signal was strong on one side of the edge but weak on another, even though the energy was actually spread out along the entire boundary. This happened because the energy in these molecules is described by complex numbers that include both a size and a phase, much like a wave that can be high or low. When the probe interacts with these waves, the phases mix together, creating a pattern that hides the true location of the energy.

To solve this puzzle, the researchers turned to a powerful tool from the field of artificial intelligence known as a convolutional neural network. This is a type of computer program designed to recognize patterns in images, similar to how a human brain learns to identify faces. The team trained this computer program using thousands of simulated maps, teaching it to look past the confusing interference patterns and deduce the underlying structure of the energy states. They tested the program by feeding it maps that included random noise, mimicking the imperfections and errors found in real-world experiments. Remarkably, the computer was able to reconstruct the full, complex description of the energy states with high accuracy, even when the input data was messy and incomplete. It successfully identified the specific coefficients that define how the energy is distributed across the molecules, effectively reversing the confusing process that the probe had created.

The study demonstrates that while the raw data from near-field spectroscopy can be misleading, it contains all the necessary information to understand these exotic states if analyzed correctly. The researchers showed that their method works even when the signal is weak or when the data is corrupted by noise, suggesting that this approach could be used in real laboratories to study topological materials. They also noted that while their current work relied on simulations and a specific setup involving a strong magnetic field, the principles could be extended to other types of molecular arrangements and even to different ways of exciting the molecules, such as using an electrical current instead of light. By combining precise physical measurements with machine learning, the team has opened a new window into the hidden world of topological edge states, turning a confusing blur of signals into a clear picture of how energy behaves at the edge of matter.

Technical Summary: Near-Field Mapping and Wavefunction Reconstruction of Topological Edge States in Molecular Aggregates

Problem Statement
Two-dimensional molecular aggregates can host topologically nontrivial edge states, which are promising for controlled excitation dynamics due to their spectral isolation and robustness against disorder. However, characterizing these states experimentally presents significant challenges. First, edge states often exhibit weak transition strengths in far-field spectroscopy, making them optically difficult to access. Second, far-field spectra do not provide spatial information, failing to reveal whether an excitation is localized at the edge. While near-field spectroscopy offers a pathway to probe spatial structures, the resulting maps for systems with complex transition dipoles do not directly mirror site populations ( $|c_m^{(\ell)}|^2$ ). Instead, they display strongly nonuniform, interference-like features arising from the interplay between complex expansion coefficients and complex transition dipoles. This nonlinearity makes the direct interpretation of near-field maps to reconstruct the underlying wavefunction coefficients highly nontrivial, particularly in the presence of noise.

Methodology
The authors investigate a model system consisting of a rectangular lattice of $N=200$ molecules (two sublattices, $N_X=N_Y=10$ ) with degenerate optical transitions. A magnetic field is applied to lift the degeneracy, creating an effective single excited state per molecule with a genuinely complex transition dipole. The system is described by a Frenkel-exciton Hamiltonian where intermolecular coupling is governed by transition-dipole-dipole interactions.

To simulate experimental conditions, the authors calculate near-field absorption maps by scanning a metallic tip (modeled as a spherical nanoparticle with an oscillating point dipole) over the aggregate at a fixed height ( $Z_{tip} = 2\Delta$ ). The absorption strength $A^{(\ell)}$ is computed as a function of the tip position, incorporating the complex coefficients $c_m^{(\ell)}$ of the eigenstates and the complex transition dipoles.

To address the inverse problem of reconstructing the complex coefficients from these noisy, non-intuitive maps, the authors employ a Convolutional Neural Network (CNN). The network is trained on a dataset of near-field maps (75 $\times$ 75 grid points) generated with random variations in molecular orientation angles ( $\theta_A, \phi_B$ ) within the topological regime. Random noise (strength $s=0.1$ ) is added to both training and testing data. The loss function is defined as the Frobenius norm of the difference between the true and predicted density matrices.

Key Results

Complex Near-Field Signatures: The study confirms that near-field maps of topological edge states do not simply reflect the probability distribution of the excitation. Instead, the maps exhibit strong spatial nonuniformity and interference patterns. For instance, some edge states show strong absorption only on specific edges or corners, while others show suppressed signals in regions where the site population is high.
CNN Reconstruction: Despite the nonlinearity and the presence of noise, the CNN successfully reconstructs the complex expansion coefficients $c_m^{(\ell)}$ (up to an overall phase) from the near-field maps.
Robustness to Noise: The reconstruction remains accurate even when the input maps contain substantial random noise. The loss distribution for the test dataset peaks around 0.02, with nearly all values falling below 0.1. For loss values below 0.1, both the magnitudes and phases of the coefficients are accurately reproduced.
Edge State Identification: The reconstructed wavefunctions clearly reveal the edge-localized character of the states, distinguishing them from bulk states, even when the raw near-field maps do not visually suggest a uniform distribution along the boundary.

Significance and Claims
The paper claims that spatially resolved near-field spectroscopy encodes the full spatial structure of topological edge states, even though this information is obscured by interference effects in the raw data. The primary contribution is the demonstration that a convolutional neural network can effectively invert this complex, nonlinear relationship to reconstruct the full eigenstate wavefunction from noisy near-field maps.

The authors emphasize that this approach allows for the identification of topological edge states in regimes where direct optical interpretation fails. They note that while the current implementation uses a specific Hamiltonian derived from a magnetic-field-induced Zeeman splitting, the methodology is adaptable. The paper modestly suggests that future work could explore different noise types, multiple polarization directions, and other lattice geometries, but does not claim the method is currently applicable to all topological systems without further validation. The work establishes that machine learning can bridge the gap between complex near-field observables and the underlying quantum mechanical wavefunctions in molecular nanostructures.

Technical Summary: Near-Field Mapping and Wavefunction Reconstruction of Topological Edge States in Molecular Aggregates

More like this