arXiv subjects

Ankur Gupta

Publications and source records attributed to Ankur Gupta.

At least 19 recordsLinked to original sources

arXiv2026

Switchcraft: AI Model Router for Agentic Tool Calling

Agentic AI systems that invoke external tools are powerful but costly, leading developers to default to large models and overspend inference budgets. Model routing can mitigate this, but existing routers are designed for chat completion rather than tool use. We present Switchcraft, the first (to the best of our knowledge) model router optimized for agentic tool calling. Switchcraft operates inline, selecting the lowest-cost model subject to correctness. We construct an evaluation framework on five function-calling benchmarks and train a DistilBERT-based classifier, deployed under a latency budget. Switchcraft achieves 82.9% accuracy -- matching or exceeding the best individual model -- while reducing inference cost by 84%, saving over $3,600 per million queries. We find that larger models do not consistently outperform smaller ones on tool-use tasks, and that nominally cheaper models can incur higher total cost due to token-intensive reasoning. Our work enables cost-aware agentic AI deployment without sacrificing correctness.

cs.AI↗

arXiv2026

A class and home problem on electrolyte transport: constant electric field implies electroneutrality, but electroneutrality does not imply a constant electric field

We present a class and home problem in graduate transport phenomena and electrochemical engineering that clarifies a common misconception: electroneutrality implies the electric field is constant. Starting with one-dimensional Poisson--Nernst--Planck equations for a silver electroplating cell, students obtain concentration and potential profiles. A companion home problem with added background electrolyte introduces a new dimensionless ratio and admits a closed-form solution. Students conclude that electroneutrality is necessary but not sufficient for a constant electric field.

physics.chem-ph↗

arXiv2026

Shape-dependence of electrophoretic mobility: an AI-assisted derivation

The electrophoretic mobility of a spherical particle is well understood, yet how particle shape modifies this mobility at arbitrary Debye length remains an open question. Here, we compute the electrophoretic mobility of a nearly spherical particle whose surface is described by $r_s(\theta) = a[1 + \varepsilon f(\theta)]$, with $\varepsilon \ll 1$, at arbitrary ratio of particle size to Debye length $\kappa a$. Using a volume-integral formulation combined with domain perturbation techniques, we derive a universal shape correction coefficient $\sigma_2(\kappa a)$ such that the mobility takes the compact form $C_\parallel = f_H(\kappa a)\,[1 + \varepsilon\,c_2\,\sigma_2(\kappa a)]$, where $f_H$ is Henry's function. We show that $\sigma_2$ interpolates between $+1/5$ in the thick-double-layer (H\"{u}ckel) limit, governed solely by the Stokes drag correction, and zero in the thin-double-layer (Smoluchowski) limit, recovering the classical shape-independence theorem. The perturbation theory agrees quantitatively with exact spheroid solutions for both prolate and oblate orientations. A key finding is that only the $P_2$ (quadrupolar) component of the particle shape affects the mobility at leading order; higher harmonics are electrophoretically silent due to angular selection rules governing the coupling between the dipolar applied field and the shape perturbation. The results in this paper were generated using Claude Code (Anthropic, Opus 4.6 model) with supervision from the authors. Our thoughts on the usage of AI for theoretical research, along with representative prompts from the development process, are provided in the manuscript and Appendix.

physics.flu-dyn↗

arXiv2026

Rethinking Balance Sheets: A Poisson-Nernst-Planck Based Approach for Modeling Concentration and Flux Profiles Inside an Electrochemical Cell

Electrochemical cells serve as a building block for producing and storing electrical energy from chemical reactions. The analysis of ion transport in these systems forms the foundation for understanding more complex electrochemical systems that are becoming increasingly present in the broader societal energy infrastructure. From a pedagogical perspective, the ``balance sheets" introduced in Chapter 4 of Electrochemical Methods: Fundamentals and Applications by Alan J. Bard, Larry R. Faulkner and Henry S. White (hereafter referred to as BFW) provides a first-pass approach to analyze ion transport in electrochemical cells. However, the balance sheet approach lacks first-principles justifications from the underlying equations that describe the transport processes in electrochemical cells. In this work, we compare a first-principles approach via the Poisson-Nernst-Planck equations to describe ion transport in electrochemical cells to that of the balance sheet approach. By re-working the examples presented in BFW, we illustrate that the balance sheet approach is only valid in limited scenarios. Furthermore, we show that the PNP equations provide a more physical route to analyze ion transport in electrochemical systems. We hope the approach outlined here will be adopted by electrochemical engineering researchers and instructors.

physics.chem-ph↗

arXiv2026

Effect of Nanopore Wall Geometry on Electrical Double-Layer Charging Dynamics

Confinement strongly influences electrochemical systems, where structural control has enabled advances in nanofluidics, sensing, and energy storage. In electric double-layer capacitors (EDLCs), or supercapacitors, energy density is governed by the accessible surface area of porous electrodes. Continuum models, built on first-principles transport equations, have provided critical insight into electrolyte dynamics under confinement but have largely focused on pores with straight walls. In such geometries, a fundamental trade-off emerges: wider pores charge faster but store less energy, while narrower pores store more charge but charge slowly. Here, we apply perturbation analysis to the Poisson-Nernst-Planck (PNP) equations for a single pore of gradually varying radius, focusing on the small potential and slender aspect ratio regime. Our analysis reveals that sloped pore walls induce an additional ionic flux, enabling simultaneous acceleration of charging and enhancement of charge storage. The theoretical predictions closely agree with direct numerical simulations while reducing computational cost by 5-6 orders of magnitude. We further propose a modified effective circuit representation that captures geometric variation along the pore and demonstrate how the framework can be integrated into pore-network models. This work establishes a scalable approach to link pore geometry with double-layer dynamics and offers new design principles for optimizing supercapacitor performance.

physics.chem-ph↗

arXiv2026

Imperfect Turing Patterns: Diffusiophoretic Assembly of Hard Spheres via Reaction-Diffusion Instabilities

Turing patterns are stationary, wave-like structures that emerge from the nonequilibrium assembly of reactive and diffusive components. While they are foundational in biophysics, their classical formulation relies on a single characteristic length scale that balances reaction and diffusion, making them overly simplistic for describing biological patterns, which often exhibit multi-scale structures, grain-like textures, and inherent imperfections. Here, we integrate diffusiophoretically-assisted assembly of finite-sized cells, driven by a background chemical gradient in a Turing pattern, while also incorporating intercellular interactions. This framework introduces key control parameters, such as the P\'{e}clet number, cell size distribution, and intercellular interactions, enabling us to reproduce strikingly similar structural features observed in natural patterns. We report imperfections, including spatial variations in pattern thickness, packing limits, and pattern breakups. Our model not only deepens our understanding but also opens a new line of inquiry into imperfect Turing patterns that deviate from the classical formulation in significant ways.

cond-mat.soft↗

arXiv2025

Continual Learning for Singing Voice Separation with Human in the Loop Adaptation

Deep learning-based works for singing voice separation have performed exceptionally well in the recent past. However, most of these works do not focus on allowing users to interact with the model to improve performance. This can be crucial when deploying the model in real-world scenarios where music tracks can vary from the original training data in both genre and instruments. In this paper, we present a deep learning-based interactive continual learning framework for singing voice separation that allows users to fine-tune the vocal separation model to conform it to new target songs. We use a U-Net-based base model architecture that produces a mask for separating vocals from the spectrogram, followed by a human-in-the-loop task where the user provides feedback by marking a few false positives, i.e., regions in the extracted vocals that should have been silence. We propose two continual learning algorithms. Experiments substantiate the improvement in singing voice separation performance by the proposed algorithms over the base model in intra-dataset and inter-dataset settings.

cs.SD↗

arXiv2025

Characterizing Deep Research: A Benchmark and Formal Definition

Information tasks such as writing surveys or analytical reports require complex search and reasoning, and have recently been grouped under the umbrella of \textit{deep research} -- a term also adopted by recent models targeting these capabilities. Despite growing interest, the scope of the deep research task remains underdefined and its distinction from other reasoning-intensive problems is poorly understood. In this paper, we propose a formal characterization of the deep research (DR) task and introduce a benchmark to evaluate the performance of DR systems. We argue that the core defining feature of deep research is not the production of lengthy report-style outputs, but rather the high fan-out over concepts required during the search process, i.e., broad and reasoning-intensive exploration. To enable objective evaluation, we define DR using an intermediate output representation that encodes key claims uncovered during search-separating the reasoning challenge from surface-level report generation. Based on this formulation, we propose a diverse, challenging benchmark LiveDRBench with 100 challenging tasks over scientific topics (e.g., datasets, materials discovery, prior art search) and public interest events (e.g., flight incidents, movie awards). Across state-of-the-art DR systems, F1 score ranges between 0.02 and 0.72 for any sub-category. OpenAI's model performs the best with an overall F1 score of 0.55. Analysis of reasoning traces reveals the distribution over the number of referenced sources, branching, and backtracking events executed by current DR systems, motivating future directions for improving their search mechanisms and grounding capabilities. The benchmark is available at https://github.com/microsoft/LiveDRBench.

cs.CL↗

arXiv2025

pH-Dependent Zeta Potential Induces Diffusiophoretic Focusing in an Acid-Base Reaction

Diffusiophoresis of charged particles in the presence of electrolytes has been extensively studied in the literature. However, in these setups, particles typically move in a single direction, either up or down the electrolyte gradient. Here, we theoretically investigate the conditions under which a particle can reverse its diffusiophoretic direction within the same setup, leading to the formation of a focusing band under steady-state concentration gradients. Using multi-ion diffusiophoresis calculations, we simulate particle transport in an acid-based reaction system where salt is added alongside the acid. For a range of salt concentrations, particles focus within the channel. Our analysis reveals that a pH-dependent zeta potential is necessary for this focusing to occur, and determines where the particles focus, i.e., on or off the acid-base reaction front. We report qualitative agreement with prior experimental observations and derive analytical conditions governing particle focusing, highlighting the delicate balance between concentration gradients and zeta potential variations. The work elucidates the crucial physics of pH-dependent zeta potential and opens new avenues for exploring diffusiophoresis in acid-base systems, with implications for microfluidic design and biophysical transport processes.

cond-mat.soft↗

arXiv2025

Magnetically Coupled Circuits to Capture Dynamics of Ionic Transport in Nanopores

Ionic transport within charged nanopores is commonly represented by resistor-capacitor transmission line circuits, where charging electrical double layers are modeled as capacitors, and the resistance to ionic current is modeled as resistors. However, these circuits fail to account for oscillations observed in experimental Nyquist plots of impedance, which are attributed ad hoc to effects such as complex porous structures or chemical reactions. Here, we show that diffusivity asymmetry between ions in confinement - overlooked in previous studies - produces Nyquist plots with two turns. Additionally, we demonstrate that ionic transport is more accurately described by magnetically coupled inductor-resistor circuits than by a simple resistor-capacitor circuit. Our results show that an impedance response of ionic transport in nanopores for arbitrary Debye lengths is better captured by two Warburg elements in parallel than a single Warburg element.

physics.chem-ph↗

arXiv2025

Building English ASR model with regional language support

In this paper, we present a novel approach to developing an English Automatic Speech Recognition (ASR) system that can effectively handle Hindi queries, without compromising its performance on English. We propose a novel acoustic model (AM), referred to as SplitHead with Attention (SHA) model, features shared hidden layers across languages and language-specific projection layers combined via a self-attention mechanism. This mechanism estimates the weight for each language based on input data and weighs the corresponding language-specific projection layers accordingly. Additionally, we propose a language modeling approach that interpolates n-gram models from both English and transliterated Hindi text corpora. Our results demonstrate the effectiveness of our approach, with a 69.3% and 5.7% relative reduction in word error rate on Hindi and English test sets respectively when compared to a monolingual English model.

eess.AS↗

arXiv2024

CAVE-Net: Classifying Abnormalities in Video Capsule Endoscopy

Accurate classification of medical images is critical for detecting abnormalities in the gastrointestinal tract, a domain where misclassification can significantly impact patient outcomes. We propose an ensemble-based approach to improve diagnostic accuracy in analyzing complex image datasets. Using a Convolutional Block Attention Module along with a Deep Neural Network, we leverage the unique feature extraction capabilities of each model to enhance the overall accuracy. The classification models, such as Random Forest, XGBoost, Support Vector Machine and K-Nearest Neighbors are introduced to further diversify the predictive power of proposed ensemble. By using these methods, the proposed framework, CAVE-Net, provides robust feature discrimination and improved classification results. Experimental evaluations demonstrate that the CAVE-Net achieves high accuracy and robustness across challenging and imbalanced classes, showing significant promise for broader applications in computer vision tasks.

cs.CV↗

arXiv2024

Universal Translational and Rotational Mobility Expressions of Phoretic and Self-phoretic Particles with Arbitrary Interaction Potentials

The mobility of externally-driven phoretic propulsion of particles is evaluated by simultaneously solving the solute conservation equation, interaction potential equation, and the modified Stokes equation. While accurate, this approach is cumbersome, especially when the interaction potential decays slowly compared to the particle size. In contrast to external phoresis, the motion of self-phoretic particles is typically estimated by relating the translation and rotation velocities with the local slip velocity. While this approach is convenient and thus widely used, it is only valid when the interaction decay length is significantly smaller than the particle size. Here, by taking inspiration from Brady J. Fluid Mech. (2021), vol. 922, A10, which combines the benefits of two approaches, we reproduce their unified mobility expressions with arbitrary interaction potentials and show that these expressions can conveniently recover the well-known mobility relationships of external electrophoresis and diffusiophoresis for arbitrary double-layer thickness. Additionally, we show that for a spherical microswimmer, the derived expressions relax to the slip velocity calculations in the limit of the thin interaction lengthscales. We also employ the derived mobility expressions to calculate the velocities of an autophoretic Janus particle. We find that there is significant dampening in the translation velocity even when the interaction length is an order of magnitude larger than the particle size. Finally, we study the motion of a catalytically self-propelled particle, while it also propels due to external concentration gradients, and demonstrate how the two propulsion modes compete with each other.

physics.flu-dyn↗

arXiv2023

A universal approximation for conductance blockade in thin nanopore membranes

Nanopore-based sensing platforms have transformed single-molecule detection and analysis. The foundation of nanopore translocation experiments lies in conductance measurements, yet existing models, which are largely phenomenological, are inaccurate in critical experimental conditions such as thin and tightly fitting pores. Of the two components of the conductance blockade, channel and access resistance, the access resistance is poorly modeled. We present a comprehensive investigation into the access resistance and associated conductance blockade in thin nanopore membranes. By combining a first-principles approach, multi-scale modeling, and experimental validation, we propose a unified theoretical modeling framework. The analytical model derived as a result surpasses current approaches across a broad parameter range. Beyond advancing theoretical understanding, our framework's versatility enables analyte size inference and predictive insights into conductance blockade behavior. Our results will facilitate the design and optimization of nanopore devices for diverse applications, including nanopore base calling and data storage.

physics.bio-ph↗

arXiv2023

A Reaction-Diffusion-Chemotaxis Model for Human Population Dynamics over Fractal Terrains

Advection of entities induced by gradients in attractant concentration fields is observed via diffusiophoresis in colloids and via chemotaxis in microorganisms. Mathematically, both diffusiophoresis and chemotaxis follow similar mathematical descriptions and display a variety of interesting behaviors that are not observed through other transport mechanisms. However, the application of such a mathematical framework has largely been restricted to soft matter research. In this article, we argue that this framework is more general and can be expanded to study human population dynamics. We assert that human populations also migrate chemotactically, but by sensing concentrations gradients in attractants such as resource availability, social connections, and safety indices. Therefore, we extend the Fisher-KPP reaction-diffusion model, foundational to human population dynamics, to incorporate chemotactic advection. Furthermore, we introduce a fractal terrain to better mimic the human dispersal phenomena. Simulations demonstrate that, by including chemotaxis of a population toward attractants which are dispersed heterogenously over fractal terrains, population hotspots can appear from from initially uniformly dispersed states whereas Fisher-KPP without chemotaxis predicts a persistent tendency toward population uniformity. Varying the chemotactic migration yields fine control over inter- or intra-population segregation and thus the population growth rates may be substantially altered by considering the population-attractant coupling. This framework may be useful for characterizing historical population separations, and furthermore is particularly pertinent for predicting emergence of new population hotspots as climate change is expected to cause large-scale human displacement, which may be dictated by chemotactic movement of humans due to evolving concentration gradients in safety indices.

q-bio.PE↗

arXiv2023

Modified Kirchhoff's Laws for Electric-Double-Layer Charging in Arbitrary Porous Networks

Understanding the dynamics of electric-double-layer (EDL) charging in porous media is essential for advancements in next-generation energy storage devices. Due to the high computational demands of direct numerical simulations and a lack of interfacial boundary conditions for reduced-order models, the current understanding of EDL charging is limited to simple geometries. Here, we present a theoretical framework to predict EDL charging in arbitrary networks of long pores in the Debye-Hückel limit without restrictions on EDL thickness and pore radii. We demonstrate that electrolyte transport is described by Kirchhoff's laws in terms of the electrochemical potential of charge (the valence-weighted average of the ion electrochemical potentials) instead of the electric potential. By employing this equivalent circuit representation with modified Kirchhoff's laws, our methodology accurately captures the spatial and temporal dependencies of charge density and electric potential, matching results obtained from computationally intensive direct numerical simulations. Our framework provides results up to five orders of magnitude faster, enabling the efficient simulation of thousands of pores within a day. We employ the framework to study the impact of pore connectivity and polydispersity on electrode charging dynamics for pore networks and discuss how these factors affect the timescale, energy density, and power density of the capacitive charging. The scalability and versatility of our methodology make it a rational tool for designing 3D-printed electrodes and for interpreting geometric effects on electrode impedance spectroscopy measurements.

physics.chem-ph↗

arXiv2023

Asymmetric rectified electric and concentration fields in multicomponent electrolytes with surface reactions

Recent studies have utilized AC fields and electrochemical reactions in multicomponent electrolyte solutions to control colloidal assembly. However, theoretical investigations have thus far been limited to binary electrolytes and have overlooked the impact of electrochemical reactions. In this study, we address these limitations by analyzing a system with multicomponent electrolytes, while also relaxing the assumption of ideally blocking electrodes to capture the effect of surface electrochemical reactions. Through a regular perturbation analysis in the low-applied-potential regime, we solve the Poisson-Nernst-Planck equations and obtain effective equations for electrical potential and ion concentrations. By employing a combination of numerical and analytical calculations, our analysis reveals a significant finding: electrochemical reactions alone can generate asymmetric rectified electric fields (AREFs), i.e., time-averaged, long-range electric fields, even when the diffusivities of the ionic species are equal. This finding expands our understanding beyond the conventional notion that AREFs arise solely from diffusivity contrast. Furthermore, we demonstrate that AREFs induced by electrochemical reactions can be stronger than those resulting from asymmetric diffusivities. Additionally, we report the emergence of asymmetric rectified concentration fields (ARCFs), i.e., time-averaged long-range concentration fields, which supports the electrodiffusiophoresis mechanism of colloidal assembly observed in experiments. We also derive analytical expressions for AREFs and ARCFs, emphasizing the role of imbalances in ionic strength and charge densities, respectively, as the driving forces behind their formation. The results presented in this article advance the field of colloidal assembly and also have implications for improved understanding of electrolyte transport in electrochemical devices.

cond-mat.soft↗

arXiv2023

Diffusiophoresis-Enhanced Turing Patterns

Turing patterns are fundamental in biophysics, emerging from short-range activation and long-range inhibition processes. However, their paradigm is based on diffusive transport processes, which yields Turing patters that are less sharp than the ones observed in nature. A complete physical description of why the Turing patterns observed in nature are significantly sharper than state-of-the-art models remains unknown. Here, we propose a novel solution to this phenomenon by investigating the role of diffusiophoresis in Turing patterns. The inclusion of diffusiophoresis enables one to generate patterns of colloidal particles with significantly finer length scales than the accompanying chemical patterns. Further, diffusiophoresis enables a robust degree of control that closely mimics natural patterns observed in species like the Ornate Boxfish and the Jewel Moray Eel. We present a scaling analysis indicating that chromatophores, ubiquitous in biological pattern formation, are likely diffusiophoretic, and that colloidal Péclet number controls the pattern enhancement. This discovery suggests important features of biological pattern formation can be explained with a universal mechanism that is quantified straightforwardly from the fundamental physics of colloids and inspires future exploration of adaptive materials, lab-on-a-chip devices, and tumorigenesis.

cond-mat.soft↗