⚛️ quantum physics

Coherent advantage in the computational expressivity of excitonic networks

This paper demonstrates that coherence acts as a critical resource for enhancing the computational expressivity of driven-dissipative excitonic networks, enabling performance to scale with network size in a manner analogous to artificial neural networks, whereas strong dephasing suppresses both this scaling and overall expressivity.

Original authors: Matthew Du, Carlos Floyd, Dipti Jasrasaria, Suriyanarayanan Vaikuntanathan

Published 2026-09-10

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Matthew Du, Carlos Floyd, Dipti Jasrasaria, Suriyanarayanan Vaikuntanathan

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 race to build smarter machines, artificial intelligence has become a voracious consumer of electricity. As digital models grow larger to perform better, the energy required to run them has become a significant burden. This reality has sparked a search for new ways to compute, looking beyond the silicon chips that power our current computers. Scientists are now exploring physical systems that can perform calculations using the natural laws of physics rather than just software instructions. One promising avenue involves using networks of molecules that can absorb and pass along energy. These systems, known as excitonic networks, rely on tiny particles of light energy moving between molecules. In the quantum world, this movement can behave like a wave, spreading out and interfering with itself, or it can behave like a simple particle hopping from one spot to another. The question researchers are asking is whether the wave-like behavior offers a special advantage for computation that the particle-like behavior cannot match.

A team of researchers at the University of Chicago has investigated this question by treating these molecular networks as a type of computer. In their setup, the input to the system is not a stream of data but the physical connections between the molecules. By adjusting how strongly the molecules are linked to one another, they can change how energy flows through the network. The output is the final state of the system, specifically how likely it is to find energy at a particular molecule after the system has settled down. The researchers found that when the molecules interact in a wave-like, coherent manner, the system becomes remarkably powerful. As they added more molecules to the network, the complexity of the calculations the system could perform grew steadily. It was as if adding more pieces to the machine automatically made it capable of solving harder problems, much like how adding more layers to a digital neural network improves its ability to recognize patterns.

However, this power is fragile and depends entirely on the environment. The researchers discovered that if the system is disturbed by too much noise or heat, the wave-like behavior collapses. The molecules stop acting as a unified wave and begin acting like independent particles hopping randomly. In this noisy, incoherent state, the system loses its ability to scale. No matter how many molecules are added, the complexity of the calculation it can perform remains stuck at a low level. The study shows that the wave-like nature of the energy is not just a side effect but a crucial resource. It allows the system to use the connections between all the molecules to create complex responses, whereas the noisy version can only use a few direct connections.

To test these ideas, the team used computer simulations to model networks of molecules with varying sizes and levels of noise. They observed that in the quiet, coherent state, the relationship between the input connections and the output energy distribution could be described by a mathematical curve that becomes increasingly intricate as the network grows. This curve can twist and turn many times, allowing it to represent very complex functions. In contrast, when they introduced strong noise to simulate a warmer, messier environment, the curve flattened out. It became a simple shape that could not change its complexity, regardless of how large the network became. This confirms that the ability to perform complex computations is tied directly to the preservation of quantum coherence.

The implications of this work extend to how we might build future computers. The researchers suggest that these molecular networks could be trained to perform specific tasks, similar to how artificial intelligence models are trained today. By carefully arranging molecules on a scaffold, such as a strand of DNA, scientists could tune the connections between them to create a physical device that solves problems efficiently. The study highlights that while these systems are currently theoretical and simulated, the principles are grounded in real physical chemistry. The key finding is that coherence acts as a fuel for computational power. Without it, the system reverts to the limitations of classical physics, where adding more parts does not necessarily make the machine smarter. With it, the system gains a unique capacity to process information that scales with its size, offering a potential path toward energy-efficient computing that leverages the strange and powerful rules of the quantum world.

Technical Summary: Coherent Advantage in the Computational Expressivity of Excitonic Networks

Problem Statement
The escalating energy consumption of artificial intelligence (AI) has spurred interest in physical substrates for trainable computation, specifically within the field of neuromorphic computing. While physical neural networks have been realized in electronic, optical, and chemical systems, a critical challenge remains: understanding how the intrinsic dynamics of a physical system control its "computational expressivity"—the complexity of functions it can represent. Recent experimental advances in arranging molecular chromophores on DNA scaffolds allow for precise control over inter-site couplings, creating driven-dissipative excitonic networks. However, it is unclear whether the quantum coherence inherent in these systems offers a computational advantage over classical, incoherent rate models, particularly regarding how expressivity scales with system size.

Methodology
The authors model a network of $N$ molecular chromophores under continuous excitation. The system is governed by a Hamiltonian describing coherent hopping between sites and a Lindblad master equation (later generalized to a Redfield description) accounting for pumping, decay, and dephasing.

Input/Output Definition: The computational input is defined as the tunable coupling strength ( $V_{12}$ ) between a pumped site and an unpumped site. The output is the steady-state excitation probability ( $p_i$ ) at a specific site.
Analytical Approach: The authors derive the steady-state density matrix to establish the mathematical form of the input-output relation. They analyze the degree of the resulting rational functions and count the number of "turning points" (local extrema) as a metric for complexity.
Regimes: The study compares two dynamical regimes:
1. Coherent Regime: Low dephasing, where wave-like transport dominates.
2. Incoherent Regime: High dephasing (or high temperature), where dynamics reduce to a classical rate model.
Numerical Verification: Random excitonic networks are simulated with site energies and couplings sampled from uniform distributions to verify analytical predictions across varying system sizes ( $N$ ) and dephasing rates ( $k_\phi$ ).

Key Contributions and Results

Rational Function Scaling: The authors prove that in the coherent regime, the input-output relation $p_i(V_{12})$ is a rational function where the degrees of the numerator and denominator polynomials scale linearly with the system size $N$ (specifically $\sim 4N$ ). This contrasts with incoherent systems or previous chemical reaction network models where the polynomial degree is independent of $N$ .
Complexity Growth: Consequently, the number of turning points in the input-output relation increases with system size in the coherent regime. This mirrors the scaling behavior of artificial neural networks, where expressivity grows with the number of parameters.
Suppression by Dephasing: The study demonstrates that increasing the dephasing rate suppresses this scaling. As dephasing increases, the system transitions to an incoherent regime where the input-output relation becomes a low-degree rational function (degree 0 or 2) with a single turning point, rendering the expressivity independent of system size.
Temperature Dependence: Using a Redfield master equation, the authors show that increasing temperature (which enhances dephasing) similarly suppresses the scaling of expressivity, confirming that the advantage is tied to the preservation of coherence.
Input Multiplicity: The framework naturally generates "input multiplicity," where a single physical parameter (e.g., the position of one chromophore) modifies multiple couplings simultaneously, further enhancing expressivity without requiring explicit input replication.

Significance and Claims
The paper establishes coherence as a physical resource for expressive computation in nonequilibrium quantum systems. The primary claim is that coherence enables computational expressivity to scale with network size, a property absent in classical, incoherent rate models under similar input-output encodings.

The authors position their work as identifying a concrete mechanism by which the dynamical regime of a physical system controls its computational capacity. They propose a hybrid quantum-classical architecture where the driven-dissipative excitonic network serves as an expressive forward map (the quantum substrate), while a classical optimizer updates the physical parameters (couplings) to approximate target functions.

The paper modestly notes that while coherence is necessary for this scaling, it is not sufficient on its own (as coherent wave scattering systems also require specific conditions to achieve scaling). Furthermore, the authors suggest that extending this framework to the multi-excitation manifold could yield polynomial rather than linear scaling of expressivity, potentially offering a path toward quantum advantage, though this remains a direction for future work. The study does not claim to have built a physical device but rather provides the theoretical foundation and scaling laws for such a platform, motivated by existing experimental capabilities in DNA-scaffolded chromophore arrays.

Technical Summary: Coherent Advantage in the Computational Expressivity of Excitonic Networks

More like this