arXiv subjects

Kyle J. Ray

Publications and source records attributed to Kyle J. Ray.

15 recordsLinked to original sources

arXiv2026

Metastable Dynamical Computing with Energy Landscapes: A Primer

Smartphones, laptops, and data centers are CMOS-based technologies that ushered our world into the information age of the 21st century. Despite their advantages for scalable computing, their implementations come with surprisingly large energetic costs. This challenge has revitalized scientific and engineering interest in energy-efficient information-processing designs. One current paradigm -- dynamical computing -- controls the location and shape of minima in potential energy landscapes that are connected to a thermal environment. The landscape supports distinguishable metastable energy minima that serve as a system's mesoscopic memory states. Information is represented by microstate distributions. Dynamically manipulating the memory states then corresponds to information processing. This framing provides a natural description of the associated thermodynamic transformations and required resources. Appealing to bifurcation theory, a computational protocol in the metastable regime can be analyzed by tracking the evolution of fixed points in the state space. We illustrate the paradigm's capabilities by performing 1-bit and 2-bit computations with double-well and quadruple-well potentials, respectively. These illustrate how dynamical computing can serve as a basis for designing universal logic gates and investigating their out-of-equilibrium thermodynamic performance.

cond-mat.stat-mech↗

arXiv2026

Momentum-Driven Reversible Logic Accelerates Efficient Irreversible Universal Computation

We present implementations of two physically-embedded computation-universal logical operations using a 2-bit logical unit composed of coupled quantum flux parametrons -- Josephson-junction superconducting circuits. To illustrate universality, we investigate NAND gates built from these two distinct elementary operations. On the one hand, Controlled Erasure (CE) is designed using fixed-point analysis and assumes that information must be stored in locally-metastable distributions. On the other, Erasure-Flip (EF) leverages momentum as a computational resource and significantly outperforms the metastable approach, simultaneously achieving higher fidelity and faster computational speed without incurring any additional energetic cost. Notably, the momentum degree of freedom allows the EF to achieve universality by using both nontrivial reversible and irreversible logic simultaneously in different logical subspaces. These results not only provide a practical, high-performance protocol ripe for experimental realization but also underscore the broader potential of momentum-based computing paradigms.

cond-mat.stat-mech↗

arXiv2026

Transformers learn factored representations

Transformers pretrained via next token prediction learn to factor their world into parts, representing these factors in orthogonal subspaces of the residual stream. We formalize two representational hypotheses: (1) a representation in the product space of all factors, whose dimension grows exponentially with the number of parts, or (2) a factored representation in orthogonal subspaces, whose dimension grows linearly. The factored representation is lossless when factors are conditionally independent, but sacrifices predictive fidelity otherwise, creating a tradeoff between dimensional efficiency and accuracy. We derive precise predictions about the geometric structure of activations for each, including the number of subspaces, their dimensionality, and the arrangement of context embeddings within them. We test between these hypotheses on transformers trained on synthetic processes with known latent structure. Models learn factored representations when factors are conditionally independent, and continue to favor them early in training even when noise or hidden dependencies undermine conditional independence, reflecting an inductive bias toward factoring at the cost of fidelity. This provides a principled explanation for why transformers decompose the world into parts, and suggests that interpretable low dimensional structure may persist even in models trained on complex data.

cs.LG↗

arXiv2025

Optimal Computation from Fluctuation Responses

The energy cost of computation has emerged as a central challenge at the intersection of physics and computer science. Recent advances in statistical physics -- particularly in stochastic thermodynamics -- enable precise characterizations of work, heat, and entropy production in information-processing systems driven far from equilibrium by time-dependent control protocols. A key open question is then how to design protocols that minimize thermodynamic cost while ensur- ing correct outcomes. To this end, we develop a unified framework to identify optimal protocols using fluctuation response relations (FRR) and machine learning. Unlike previous approaches that optimize either distributions or protocols separately, our method unifies both using FRR-derived gradients. Moreover, our method is based primarily on iteratively learning from sampled noisy trajectories, which is generally much easier than solving for the optimal protocol directly from a set of governing equations. We apply the framework to canonical examples -- bit erasure in a double-well potential and translating harmonic traps -- demonstrating how to construct loss functions that trade-off energy cost against task error. The framework extends trivially to underdamped systems, and we show this by optimizing a bit-flip in an underdamped system. In all computations we test, the framework achieves the theoretically optimal protocol or achieves work costs comparable to relevant finite time bounds. In short, the results provide principled strategies for designing thermodynamically efficient protocols in physical information-processing systems. Applications range from quantum gates robust under noise to energy-efficient control of chemical and synthetic biological networks.

cond-mat.stat-mech↗

arXiv2025

Large Interconnected Thermodynamic Systems Nearly Minimize Entropy Production

Many have speculated whether nonequilibrium systems obey principles of maximum or minimum entropy production. In this work, we use stochastic thermodynamics to derive the condition for the minimum entropy production state (MEPS) for continuous-time Markov chains (CTMCs), even far from equilibrium. We show that real nonequilibrium steady states (NESS) generally violate both the MINEP and MAXEP principles. However, through numerical sampling of large interconnected CTMCs, we find that as system size increases, the steady-state entropy production tends to converge toward the minimum. This suggests that large nonequilibrium systems may self-organize to make efficient use of thermodynamic resources, offering a nuanced perspective on the longstanding debate between MAXEP and MINEP.

cond-mat.stat-mech↗

arXiv2025

Learning Stochastic Thermodynamics Directly from Correlation and Trajectory-Fluctuation Currents

Markedly increased computational power and data acquisition have led to growing interest in data-driven inverse dynamics problems. These seek to answer a fundamental question: What can we learn from time series measurements of a complex dynamical system? For small systems interacting with external environments, the effective dynamics are inherently stochastic, making it crucial to properly manage noise in data. Here, we explore this for systems obeying Langevin dynamics and, using currents, we construct a learning framework for stochastic modeling. Currents have recently gained increased attention for their role in bounding entropy production (EP) from thermodynamic uncertainty relations (TURs). We introduce a fundamental relationship between the cumulant currents there and standard machine-learning loss functions. Using this, we derive loss functions for several key thermodynamic functions directly from the system dynamics without the (common) intermediate step of deriving a TUR. These loss functions reproduce results derived both from TURs and other methods. More significantly, they open a path to discover new loss functions for previously inaccessible quantities. Notably, this includes access to per-trajectory entropy production, even if the observed system is driven far from its steady-state. We also consider higher order estimation. Our method is straightforward and unifies dynamic inference with recent approaches to entropy production estimation. Taken altogether, this reveals a deep connection between diffusion models in machine learning and entropy production estimation in stochastic thermodynamics.

cond-mat.stat-mech↗

arXiv2024

Nonequilibrium Thermodynamics of a Superconducting Szilard Engine

We implement a Szilard engine using a 2-bit logical unit consisting of inductively coupled quantum flux parametrons (QFPs) -- Josephson-junction superconducting circuits with applications in both the classical and quantum information processing regimes. Detailed simulations show that it is highly thermodynamically efficient while functioning as a Maxwell demon -- converting heat to work. The physically-calibrated design is targeted to direct experimental exploration. However, variations in Josephson junction fabrication introduce asymmetries that result in energy inefficiency and low operational fidelity. We provide a design solution that mitigates these practical challenges. The resulting platform is ideally suited to probe the thermodynamic foundations of information processing devices far from equilibrium.

cond-mat.stat-mech↗

arXiv2024

Controlled Erasure as a Building Block for Universal Thermodynamically-Robust Superconducting Computing

Reducing the energy inefficiency of conventional CMOS-based computing devices -- which rely on logically irreversible gates to process information -- remains both a fundamental engineering challenge and a practical social challenge of increasing importance. We extend an alternative computing paradigm that manipulates microstate distributions to store information in the metastable minima determined by an effective potential energy landscape. These minima serve as mesoscopic memories that are manipulated by a dynamic landscape to perform information processing. Central to our results is the control erase (CE) protocol that controls the landscape's metastable minima to determine whether information is preserved or erased. Importantly, successive protocol executions can implement a NAND gate -- a logically-irreversible universal logic gate. We show how to practically implement this in a device created by two inductively-coupled superconducting quantum interference devices (SQUIDs). We identify circuit parameter ranges that give rise to effective CEs and establish the device's robustness against logical errors. These SQUID-based logical devices are capable of operating above GHz frequencies and at the $k_\text{B} T$ energy scale. Due to this, optimized devices and associated protocols provide a universal-computation substrate that is both computationally fast and energy efficient.

cond-mat.supr-con↗

arXiv2024

Entropy Production by Underdamped Langevin Dynamics

Entropy production (EP) is a central quantity in nonequilibrium physics as it monitors energy dissipation, irreversibility, and free energy differences during thermodynamic transformations. Estimating EP, however, is challenging both theoretically and experimentally due to limited access to the system dynamics. For overdamped Langevin dynamics and Markov jump processes it was recently proposed that, from thermodynamic uncertainty relations (TUR), short-time cumulant currents can be used to estimate EP without knowledge of the dynamics. Yet, estimation of EP in underdamped Langevin systems remains an active challenge. To address this, we derive a modified TUR that relates the statistics of two specific novel currents -- one cumulant current and one stochastic current -- to a system's EP. These two distinct but related currents are used to constrain EP in the modified TUR. One highlight is that there always exists a family of currents such that the uncertainty relations saturate, even for long-time averages and in nonsteady-state scenarios. Another is that our method only requires limited knowledge of the dynamics -- specifically, the damping-coefficient to mass ratio and the diffusion constant. This uncertainty relation allows estimating EP for both overdamped and underdamped Langevin dynamics. We validate the method numerically, through applications to several underdamped systems, to underscore the flexibility in obtaining EP in nonequilibrium Langevin systems.

cond-mat.stat-mech↗

arXiv2023

Extracting Equations of Motion from Superconducting Circuits

Alternative computing paradigms open the door to exploiting recent innovations in computational hardware to probe the fundamental thermodynamic limits of information processing. One such paradigm employs superconducting quantum interference devices (SQUIDs) to execute classical computations. This, though, requires constructing sufficiently complex superconducting circuits that support a suite of useful information processing tasks and storage operations, as well as understanding these circuits' energetics. First-principle circuit design, though, leads to prohibitive algebraic complications when deriving the effective equations of motion -- complications that to date have precluded achieving these goals, let alone doing so efficiently. We circumvent these complications by (i) specializing our class of circuits and physical operating regimes, (ii) synthesizing existing derivation techniques to suit these specializations, and (iii) implementing solution-finding optimizations which facilitate physically interpreting circuit degrees of freedom that respect physically-grounded constraints. This leads to efficient, practical circuit prototyping and access to scalable circuit architectures. The analytical efficiency is demonstrated by reproducing the potential energy landscape generated by the quantum flux parametron (QFP). We then show how inductively coupling two QFPs produces a device that is capable of executing 2-bit computations via its composite potential energy landscape. More generally, the synthesis methods detailed here provide a basis for constructing universal logic gates and investigating their thermodynamic performance.

cond-mat.stat-mech↗

arXiv2022

The Thermodynamic Uncertainty Theorem

Thermodynamic uncertainty relations (TURs) express a fundamental tradeoff between the precision (inverse scaled variance) of any thermodynamic current by functionals of the average entropy production. Relying on purely variational arguments, we significantly extend these inequalities by incorporating and analyzing the impact of higher statistical cumulants of entropy production within a general framework of time-symmetrically controlled computation. This allows us to derive an exact expression for the current that achieves the minimum scaled variance, for which the TUR bound tightens to an equality that we name Thermodynamic Uncertainty Theorem (TUT). Importantly, both the minimum scaled variance current and the TUT are functionals of the stochastic entropy production, thus retaining the impact of its higher moments. In particular, our results show that, beyond the average, the entropy production distribution's higher moments have a significant effect on any current's precision. This is made explicit via a thorough numerical analysis of swap and reset computations that quantitatively compares the TUT against previous generalized TURs. Our results demonstrate how to interpolate between previously-established bounds and how to identify the most relevant TUR bounds in different nonequilibrium regimes.

cond-mat.stat-mech↗

arXiv2022

Gigahertz Sub-Landauer Momentum Computing

We introduce a fast and highly-efficient physically-realizable bit swap. Employing readily available and scalable Josephson junction microtechnology, the design implements the recently introduced paradigm of momentum computing. Its nanosecond speeds and sub-Landauer thermodynamic efficiency arise from dynamically storing memory in momentum degrees of freedom. As such, during the swap, the microstate distribution is never near equilibrium and the memory-state dynamics fall far outside of stochastic thermodynamics that assumes detailed-balanced Markovian dynamics. The device implements a bit-swap operation -- a fundamental operation necessary to build reversible universal computing. Extensive, physically-calibrated simulations demonstrate that device performance is robust and that momentum computing can support thermodynamically-efficient, high-speed, large-scale general-purpose computing that circumvents Landauer's bound.

cond-mat.stat-mech↗

arXiv2020

Szilard Engines as Quantum Thermodynamical Systems

We analyze an engine whose working fluid consists of a single quantum particle, paralleling Szilard's construction of a classical single-particle engine. Following his resolution of Maxwell's Second Law paradox using the latter, which turned on physically instantiating the demon (control subsystem), the quantum engine's design mirrors the classically-chaotic Szilard Map that operates a thermodynamic cycle of measurement, thermal-energy extraction, and memory reset. Focusing on the thermodynamic costs to observe and control the particle and comparing these in the quantum and classical limits, we detail the thermodynamic tradeoffs behind Landauer's Principle for information-processing-induced thermodynamic dissipation in the quantum and classical regimes. In particular, and as found with the classical engine, we show that the sum of the thermodynamic costs over a cycle obeys a generalized Landauer Principle, exactly balancing energy extraction from the heat bath. Thus, the quantum engine obeys the Second Law. However, the quantum engine does so via substantially different mechanisms: classically measurement and erasure determine the thermodynamics, while in the quantum implementation the cost of partition insertion is key.

quant-ph↗

arXiv2020

Non-Markovian Momentum Computing: Universal and Efficient

All computation is physically embedded. Reflecting this, a growing body of results embraces rate equations as the underlying mechanics of thermodynamic computation and biological information processing. Strictly applying the implied continuous-time Markov chains, however, excludes a universe of natural computing. We show that expanding the toolset to continuous-time hidden Markov chains substantially removes the constraints. The general point is made concrete by our analyzing two eminently-useful computations that are impossible to describe with a set of rate equations over the memory states. We design and analyze a thermodynamically-costless bit flip, providing a first counterexample to rate-equation modeling. We generalize this to a costless Fredkin gate---a key operation in reversible computing that is computation universal. Going beyond rate-equation dynamics is not only possible, but necessary if stochastic thermodynamics is to become part of the paradigm for physical information processing.

cond-mat.stat-mech↗

arXiv2020

Variations on a Demonic Theme: Szilard's Other Engines

Szilard's now-famous single-molecule engine was only the first of three constructions he introduced in 1929 to resolve several paradoxes arising from Maxwell's demon. We analyze Szilard's remaining two demon models. We show that the second one, though a markedly different implementation employing a population of distinct molecular species and semi-permeable membranes, is informationally and thermodynamically equivalent to an ideal gas of the single-molecule engines. Since it is a gas of noninteracting particles one concludes, following Boyd and Crutchfield, that (i) it reduces to a chaotic dynamical system---called the Szilard Map, a composite of three piecewise linear maps that implement the thermodynamic transformations of measurement, control, and erasure; (ii) its transitory functioning as an engine that converts disorganized heat energy to work is governed by the Kolmogorov-Sinai entropy rate; (iii) the demon's minimum necessary "intelligence" for optimal functioning is given by the engine's statistical complexity, and (iv) its functioning saturates thermodynamic bounds and so it is a minimal, optimal implementation. We show that Szilard's third model is rather different and addresses the fundamental issue, raised by the first two, of measurement in and by thermodynamic systems and entropy generation. Taken together, Szilard's suite of constructions lays out a range of possible realizations of Maxwellian demons that anticipated by almost two decades Shannon's and Wiener's concept of information as surprise and cybernetics' notion of functional information. This, in turn, gives new insight into engineering implementations of novel nanoscale information engines that leverage microscopic fluctuations and into the diversity of thermodynamic mechanisms and intrinsic computation harnessed in physical, molecular, biochemical, and biological systems.

cond-mat.stat-mech↗