arXiv subjects

Ashwin Ramachandran

Publications and source records attributed to Ashwin Ramachandran.

11 recordsLinked to original sources

arXiv2026

Transverse transport of microswimmers in oscillatory channel flows

Motile microorganisms inhabit oscillatory flows encountered in physiological and engineering systems, yet the influence of flow unsteadiness on their shear-induced preferential concentration remains less understood. We investigate transverse transport of elongated microswimmers in oscillatory pressure-driven channel flow using complementary Langevin simulations and two- and one-dimensional Fokker-Planck models. Agreement between particle-based and continuum descriptions across governing parameters establishes the fidelity of the Fokker-Planck formulation for swimmer transport in oscillatory shear flows. Our analysis demonstrates that oscillatory forcing fundamentally modifies classical steady-flow shear-trapping. We find that increasing Womersley number $Wo$ reduces centerline depletion by confining oscillatory shear to thinner near-wall regions. In contrast, increasing frequency ratio $β$, of swimmer rotational diffusion rate and flow oscillation frequency, promotes sustained orientational anisotropy, and leads to a saturating increase in centerline depletion. In the weak-swimming limit, we derive a hierarchy of coupled orientational moments, enabling analytical solutions at arbitrary temporal harmonic order. The asymptotic solutions establish a universal transfer law showing that, for a given swimmer shape, the normalized leading-order orientational response depends solely on $β$. Reconstruction of the orientational distribution yields closed-form expressions for the swimmer concentration profile and depletion index, $I_D$. Our asymptotic solution recovers the steady-flow weak-shear scaling of $I_D$ with flow Peclet number, $Pe_f$, of $I_D \propto Pe_f^2$, and reveals that oscillatory forcing attenuates this response by a factor $16β^2/(1+16β^2)$. These results provide a theoretical framework to predict transverse transport of microswimmers in oscillatory channel flows.

physics.flu-dyn↗

arXiv2026

HCCL: Collective Communication for Meta Training and Inference Accelerators

We present HCCL, a collective communication library co-designed with Meta's MTIA 300 accelerator, the first Meta chip to integrate backend networking directly on chip package. MTIA 300 includes dedicated message engines (MEs) with near-memory compute (NMC) that fully offload collective execution from the compute grid, enabling large overlap between computation and communication. HCCL uses a compiled communication model in which the host generates a complete description of each collective including dependencies. We describe the control and data path architecture, topology-aware algorithm selection across MTIA 300's asymmetric scale-up and scale-out network, and optimizations for both training and inference workloads. For training, HCCL achieves up to 940 GB/s on intra-rack collectives while introducing less than 0.5% degradation to concurrent compute throughput. For inference, we leverage one-sided communication primitives that bypass the scheduling path to minimize collective latency and describe collective designs that improve compute-communication pipelining for latency-sensitive workloads.

cs.NI↗

arXiv2026

MLCommons Chakra: Advancing Performance Benchmarking and Co-design using Standardized Execution Traces

The fast pace of artificial intelligence~(AI) innovation demands an agile methodology for observation, reproduction and optimization of distributed machine learning~(ML) workload behavior in production AI systems and enables efficient software-hardware~(SW-HW) co-design for future systems. We present Chakra, an open and portable ecosystem for performance benchmarking and co-design. The core component of Chakra is an open and interoperable graph-based representation of distributed AI/ML workloads, called Chakra execution trace~(ET). These ETs represent key operations, such as compute, memory, and communication, data and control dependencies, timing, and resource constraints. Additionally, Chakra includes a complementary set of tools and capabilities to enable the collection, analysis, generation, and adoption of Chakra ETs by a broad range of simulators, emulators, and replay tools. We present analysis of Chakra ETs collected on production AI clusters and demonstrate value via real-world case studies. Chakra has been adopted by MLCommons and has active contributions and engagement across the industry, including but not limited to NVIDIA, AMD, Meta, Keysight, HPE, and Scala, to name a few.

cs.DC↗

arXiv2025

Charting the Design Space of Neural Graph Representations for Subgraph Matching

Subgraph matching is vital in knowledge graph (KG) question answering, molecule design, scene graph, code and circuit search, etc. Neural methods have shown promising results for subgraph matching. Our study of recent systems suggests refactoring them into a unified design space for graph matching networks. Existing methods occupy only a few isolated patches in this space, which remains largely uncharted. We undertake the first comprehensive exploration of this space, featuring such axes as attention-based vs. soft permutation-based interaction between query and corpus graphs, aligning nodes vs. edges, and the form of the final scoring network that integrates neural representations of the graphs. Our extensive experiments reveal that judicious and hitherto-unexplored combinations of choices in this space lead to large performance benefits. Beyond better performance, our study uncovers valuable insights and establishes general design principles for neural graph representation and interaction, which may be of wider interest.

cs.LG↗

arXiv2025

Iteratively Refined Early Interaction Alignment for Subgraph Matching based Graph Retrieval

Graph retrieval based on subgraph isomorphism has several real-world applications such as scene graph retrieval, molecular fingerprint detection and circuit design. Roy et al. [35] proposed IsoNet, a late interaction model for subgraph matching, which first computes the node and edge embeddings of each graph independently of paired graph and then computes a trainable alignment map. Here, we present IsoNet++, an early interaction graph neural network (GNN), based on several technical innovations. First, we compute embeddings of all nodes by passing messages within and across the two input graphs, guided by an injective alignment between their nodes. Second, we update this alignment in a lazy fashion over multiple rounds. Within each round, we run a layerwise GNN from scratch, based on the current state of the alignment. After the completion of one round of GNN, we use the last-layer embeddings to update the alignments, and proceed to the next round. Third, IsoNet++ incorporates a novel notion of node-pair partner interaction. Traditional early interaction computes attention between a node and its potential partners in the other graph, the attention then controlling messages passed across graphs. In contrast, we consider node pairs (not single nodes) as potential partners. Existence of an edge between the nodes in one graph and non-existence in the other provide vital signals for refining the alignment. Our experiments on several datasets show that the alignments get progressively refined with successive rounds, resulting in significantly better retrieval performance than existing methods. We demonstrate that all three innovations contribute to the enhanced accuracy. Our code and datasets are publicly available at https://github.com/structlearning/isonetpp.

cs.LG↗

arXiv2025

ReFoRCE: A Text-to-SQL Agent with Self-Refinement, Consensus Enforcement, and Column Exploration

We present ReFoRCE, a Text-to-SQL agent that tops the Spider 2.0 leaderboard--a challenging benchmark reflecting complex, real-world Text-to-SQL scenarios. While Text-to-SQL systems enable natural language queries over structured databases, deploying them in enterprise environments remains difficult due to large, complex schemas (with over 1,000 columns), diverse SQL dialects (e.g., BigQuery, Snowflake), and sophisticated query requirements (e.g., transformations and analytics). ReFoRCE addresses these challenges through: (a) database information compression via pattern-based table grouping and LLM-guided schema linking to alleviate long-context issues; (b) self-refinement to iteratively correct syntax and semantic errors across dialects; (c) majority-vote consensus to select high-confidence candidates while deferring ambiguous cases arising from sophisticated queries; and (d) iterative column exploration guided by execution feedback to resolve those deferred cases. ReFoRCE achieves new state-of-the-art results, with scores of 35.83 on Spider 2.0-Snow and 36.56 on Spider 2.0-Lite.

cs.CL↗

arXiv2024

Text-to-SQL Calibration: No Need to Ask -- Just Rescale Model Probabilities

Calibration is crucial as large language models (LLMs) are increasingly deployed to convert natural language queries into SQL for commercial databases. In this work, we investigate calibration techniques for assigning confidence to generated SQL queries. We show that a straightforward baseline -- deriving confidence from the model's full-sequence probability -- outperforms recent methods that rely on follow-up prompts for self-checking and confidence verbalization. Our comprehensive evaluation, conducted across two widely-used Text-to-SQL benchmarks and multiple LLM architectures, provides valuable insights into the effectiveness of various calibration strategies.

cs.DB↗

arXiv2022

Microfluidic Isotachophoresis: Theory and Applications

Isotachophoresis (ITP) is a versatile electrophoretic technique which can be used for sample preconcentration, separation, purification, mixing, and control and acceleration of chemical reactions. Although the basic technique is nearly a century old and widely used, there has been a persistent need for an easily approachable, succinct, and rigorous review of ITP theory and analysis. This is important as interest and adoption of the technique has grown over the last two decades, especially because of its implementation into microfluidics and integration with on-chip chemical and biochemical assays. We here provide a review of ITP theory with a strong emphasis on steady and unsteady transport starting from physicochemical first principles including conservation of species, conservation of current, the approximation of charge neutrality, pH equilibrium of weak electrolytes, and so-called regulating functions governing transport dynamics. We combine these generally applicable (to all types of ITP) theoretical discussions with applications of ITP in the field of microfluidic systems, particularly on-chip biochemical analyses. Our discussion includes principles governing ITP focusing of weak and strong electrolytes, ITP dynamics in peak and plateau modes, review of simulation tools, experimental tools and detection methods, applications of ITP for on-chip separations and trace analyte manipulation, and design considerations and challenges for microfluidic ITP systems. We conclude with remarks on possible future research directions. The intent of this review is to help make ITP analysis and design principles more accessible to the scientific and engineering communities, and to provide a rigorous basis for increased adoption of ITP into microfluidics.

physics.chem-ph↗

arXiv2018

Performance Metrics for the Objective Assessment of Capacitive Deionization Systems

In the growing field of capacitive deionization (CDI), a number of performance metrics have emerged to describe the desalination process. Unfortunately, the separation conditions under which these metrics are measured are often not specified, resulting in optimal performance at minimal removal. Here we outline a system of performance metrics and reporting conditions that resolves this issue. Our proposed system is based on volumetric energy consumption (Wh/m$^3$) and throughput productivity (L/h/m$^2$) reported for a specific average concentration reduction, water recovery, and feed salinity. To facilitate and rationalize comparisons between devices, materials, and operation modes, we propose a nominal standard testing condition of removing 5 mM from a 20 mM NaCl feed solution at 50% water recovery for CDI research. Using this separation, we compare the desalination performance of a flow-through electrode (fte-CDI) cell and a flow between membrane (fb-MCDI) device, showing how significantly different systems can be compared in terms of generally desirable desalination characteristics. In general, we find that performance analysis must be considered carefully so to not allow for ambiguous separation conditions or the maximization of one metric at the expense of another. Additionally, for context we discuss a number of important underlying performance indicators and cell characteristics that are not performance measures in and of themselves but can be examined to better understand differences in performance.

physics.chem-ph↗

arXiv2018

Frequency analysis and resonant operation for efficient capacitive deionization

Capacitive deionization (CDI) performance metrics can vary widely with operating methods. We here develop a theory around and experimentally demonstrate a new operation for CDI that uses sinusoidal forcing voltage (or sinusoidal current). We use a dynamic system modeling approach, and quantify the frequency response (amplitude and phase) of CDI effluent concentration, and demonstrate that CDI can be modeled as a linear time invariant system. We validate this model with experiments, and show that a sinusoid voltage operation can simultaneously achieve high salt removal and strong energy performance, thus very likely making it superior to other conventional operating methods. Based on the underlying coupled phenomena of electrical charge (and ionic) transfer with bulk advection in CDI, we derive and validate experimentally the concept of using sinusoidal voltage forcing functions to achieve resonance-type operation for CDI. Despite the complexities of the system, we find a simple relation for the resonant time scale: the resonant time period is proportional to the geometric mean of the flow residence time and the electrical (RC) charging time. Operation at resonance implies the optimal balance between absolute amount of salt removed (in moles) and dilution (depending on the feed volume processed), thus resulting in the maximum average concentration reduction for the desalinated water. Using a generalized tool based on Fourier analysis, we show that most of the salt removal (~95%) for square and triangular voltage forcing waveforms is achieved by the fundamental Fourier (sinusoidal) mode. The frequency of higher Fourier modes precludes high flow efficiency for these modes, so these modes consume additional energy for minimal additional salt removed. This deficiency of higher frequency modes further highlights the advantage of DC-offset sinusoidal forcing for CDI operation.

physics.chem-ph↗

arXiv2018

Thermodynamics of Ion Separation by Electrosorption

We present a simple, top-down approach for the calculation of minimum energy consumption of electrosorptive ion separation using variational form of the (Gibbs) free energy. We focus and expand on the case of electrostatic capacitive deionization (CDI), and the theoretical framework is independent of details of the double-layer charge distribution and is applicable to any thermodynamically consistent model, such as the Gouy-Chapman-Stern (GCS) and modified Donnan (mD) models. We demonstrate that, under certain assumptions, the minimum required electric work energy is indeed equivalent to the free energy of separation. Using the theory, we define the thermodynamic efficiency of CDI. We explore the thermodynamic efficiency of current experimental CDI systems and show that these are currently very low, less than 1% for most existing systems. We applied this knowledge and constructed and operated a CDI cell to show that judicious selection of the materials, geometry, and process parameters can be used to achieve a 9% thermodynamic efficiency (4.6 kT energy per removed ion). This relatively high value is, to our knowledge, by far the highest thermodynamic efficiency ever demonstrated for CDI. We hypothesize that efficiency can be further improved by further reduction of CDI cell series resistances and optimization of operational parameters.

physics.chem-ph↗