arXiv subjects

Saptarshi Saha

Publications and source records attributed to Saptarshi Saha.

At least 19 recordsLinked to original sources

arXiv2026

KisMATH: Do LLMs Have Knowledge of Implicit Structures in Mathematical Reasoning?

Chain-of-thought (CoT) traces have been shown to improve performance of large language models on a plethora of reasoning tasks, yet there is no consensus on the mechanism by which this boost is achieved. To shed more light on this, we introduce Causal CoT Graphs (CCGraphs), which are directed acyclic graphs automatically extracted from reasoning traces that model fine-grained causal dependencies in language-model outputs. A collection of 1671 mathematical reasoning problems from MATH500, GSM8K, and AIME, together with their associated CCGraphs, has been compiled into our dataset -- KisMATH. Our detailed empirical analysis with 15 open-weight LLMs shows that (i) reasoning nodes in the CCGraphs are causal contributors to the final answer, which we argue is constitutive of reasoning; and (ii) LLMs emphasize the reasoning paths captured by the CCGraphs, indicating that the models internally realize structures similar to our graphs. KisMATH enables controlled, graph-aligned interventions and opens avenues for further investigation into the role of CoT in LLM reasoning.

cs.CL↗

arXiv2026

Cyclic Counterfactuals under Shift-Scale Interventions

Most counterfactual inference frameworks traditionally assume acyclic structural causal models (SCMs), i.e. directed acyclic graphs (DAGs). However, many real-world systems (e.g. biological systems) contain feedback loops or cyclic dependencies that violate acyclicity. In this work, we study counterfactual inference in cyclic SCMs under shift-scale interventions, i.e., soft, policy-style changes that rescale and/or shift a variable's mechanism.

cs.AI↗

arXiv2026

A detailed study on various phases in dissipative anisotropic Dicke model

We present a comprehensive study of different phases in the Dicke model incorporating both anisotropy and dissipation. We begin with a concise review of the quantum phase transition in this setting, highlighting how these two parameters shift the critical point. We then perform a detailed investigation of the transition from ergodic to nonergodic phases by analyzing the eigenvalue and eigenvector properties of the Liouvillian with the aid of scaling of the Liouvillian gap and the average participation ratio. Our results show that the eigenvector properties of the Liouvillian are consistent with its spectral characteristics, leading to a phase diagram that has similarities with the closed counterpart. Furthermore, we demonstrate that the Liouvillian gap exhibits distinct scaling behaviors in these two phases. Finally, we extend our study to the driven case by applying a Thue-Morse quasiperiodic drive. In this case, we find that bosonic dissipation plays a crucial role in stabilizing the prethermal plateau, offering an effective mechanism to halt the heating effect arising from the quasi-periodic drive.

quant-ph↗

arXiv2025

Discrete Time Crystals in Noninteracting Dissipative Systems

Many-body quantum systems, under suitable conditions, exhibit time-translation symmetry breaking and settle in a discrete time crystalline (DTC) phase -- an out-of-equilibrium quantum phase of matter. The defining feature of DTC is a robust subharmonic response. However, the DTC phase is fragile in the presence of environmental dissipation. Here, we propose and exemplify a DTC phase in a noninteracting system that owes its stability to environmental dissipation. The lifetime of this DTC is independent of initial conditions and the size of the system, though it depends on the frequency of the external driver. We experimentally demonstrate this realization of DTC using Nuclear Magnetic Resonance spectroscopy.

quant-ph↗

arXiv2025

Emergence of Unruh prethermalization for uniformly accelerating many-atom system

A uniformly accelerated atom in an inertial vacuum generally thermalizes and reaches a Gibbs state. This phenomenon is commonly known as the Unruh effect. Here, we show that the situation is entirely different for the many-atoms problem. In the case of non-interacting accelerating atoms, we show that a regime exists where the entire system reaches a prethermal generalized Gibbs state before it thermalizes. The prethermal state is protected by emergent conserved quantities; hence, the system behaves like a nearly-integrable one, which shows a sharp distinction from the Unruh effect. We coin the term ``Unruh prethermalization" to characterize this phenomenon. The measure of entanglement is a good estimation of the lifetime of the prethermal state and is consistent with previous studies. Finally, we show that in such a regime, the dynamics show a Dicke superradiance-type radiation burst before reaching the prethermal state. In contrast, only a mono-exponential decay is observed for Unruh thermalization. In addition, to highlight the significance of our results, we compare them with existing experimental observations.

quant-ph↗

arXiv2025

Determination of Fluctuation correlation time in solid-state Nuclear Magnetic Resonance

Solid-state NMR provides a wide variety of experimental techniques to detect and analyze a material's chemical and physical environment. Here, we offer a theoretical demonstration of a new approach that could be a promising candidate for characterizing the local environments and shifts. Prethermalization by applying a spin-locking pulse brings a new research paradigm in solid-state NMR and has not been hitherto explored for such purposes. We show that a prethermal state can also be effective in this case. A prethermal state is described using its lifetime and the value of transverse magnetization. Using these two variables, we successfully detect the changes in the environmental parameter and chemical and dynamical shifts (such as Lamb shifts). Our results exhibit that the lifetime increases with increasing environmental correlation time. On the other hand, the transverse magnetization decreases with the increase in the strength of the shift parameter. Based on these observations, we propose that the prethermalization dynamics can yield important information on local environment.

quant-ph↗

arXiv2025

Can a pure state remain pure in the Unruh effect?

A uniformly accelerated detector in an inertial vacuum undergoes an unavoidable dissipation, and the final steady-state becomes thermal. However, to attain such a mixed state, there is no bound for the acceleration of the single atomic detector. Here we show that the scenario is entirely different for two atoms with the same energy levels. There exists a critical limit of the acceleration for two atomic detectors, below which the purity of a particular initial state can be preserved. We observe that the generator of the dissipative dynamics (Lindbladian) is invariant under a weak symmetry transformation at this limit. Hence one of the eigenstates of the symmetry operator is unchanged during the evolution. This kind of state is called a quantum dark state, which is essentially a decoherence-free subspace. As a consequence, the system becomes localized, and it can skip the Unruh thermalization. Beyond the critical limit, the symmetry is explicitly broken. Therefore our results suggest that the system goes through a first-order dissipative phase transition from a localized to a thermal phase.

quant-ph↗

arXiv2025

On Measuring Intrinsic Causal Attributions in Deep Neural Networks

Quantifying the causal influence of input features within neural networks has become a topic of increasing interest. Existing approaches typically assess direct, indirect, and total causal effects. This work treats NNs as structural causal models (SCMs) and extends our focus to include intrinsic causal contributions (ICC). We propose an identifiable generative post-hoc framework for quantifying ICC. We also draw a relationship between ICC and Sobol' indices. Our experiments on synthetic and real-world datasets demonstrate that ICC generates more intuitive and reliable explanations compared to existing global explanation techniques.

stat.ML↗

arXiv2025

Crowdsource, Crawl, or Generate? Creating SEA-VL, a Multicultural Vision-Language Dataset for Southeast Asia

Southeast Asia (SEA) is a region of extraordinary linguistic and cultural diversity, yet it remains significantly underrepresented in vision-language (VL) research. This often results in artificial intelligence (AI) models that fail to capture SEA cultural nuances. To fill this gap, we present SEA-VL, an open-source initiative dedicated to developing high-quality, culturally relevant data for SEA languages. By involving contributors from SEA countries, SEA-VL aims to ensure better cultural relevance and diversity, fostering greater inclusivity of underrepresented languages in VL research. Beyond crowdsourcing, our initiative goes one step further in the exploration of the automatic collection of culturally relevant images through crawling and image generation. First, we find that image crawling achieves approximately ~85% cultural relevance while being more cost- and time-efficient than crowdsourcing. Second, despite the substantial progress in generative vision models, synthetic images remain unreliable in accurately reflecting SEA cultures. The generated images often fail to reflect the nuanced traditions and cultural contexts of the region. Collectively, we gather 1.28M SEA culturally-relevant images, more than 50 times larger than other existing datasets. Through SEA-VL, we aim to bridge the representation gap in SEA, fostering the development of more inclusive AI systems that authentically represent diverse cultures across SEA.

cs.CV↗

arXiv2025

Language Models are Crossword Solvers

Crosswords are a form of word puzzle that require a solver to demonstrate a high degree of proficiency in natural language understanding, wordplay, reasoning, and world knowledge, along with adherence to character and length constraints. In this paper we tackle the challenge of solving crosswords with large language models (LLMs). We demonstrate that the current generation of language models shows significant competence at deciphering cryptic crossword clues and outperforms previously reported state-of-the-art (SoTA) results by a factor of 2-3 in relevant benchmarks. We also develop a search algorithm that builds off this performance to tackle the problem of solving full crossword grids with out-of-the-box LLMs for the very first time, achieving an accuracy of 93% on New York Times crossword puzzles. Additionally, we demonstrate that LLMs generalize well and are capable of supporting answers with sound rationale.

cs.CL↗

arXiv2024

Region Mixup

This paper introduces a simple extension of mixup (Zhang et al., 2018) data augmentation to enhance generalization in visual recognition tasks. Unlike the vanilla mixup method, which blends entire images, our approach focuses on combining regions from multiple images.

cs.CV↗

arXiv2024

Emergence of superradiance in dissipative dipolar-coupled spin systems

In the superradiance phenomenon, a collection of non-interacting atoms exhibits collective dissipation due to interaction with a common radiation field, resulting in a non-monotonic decay profile. This work shows that dissipative dipolar-coupled systems exhibit an identical collective dissipation aided by the nonsecular part of the dipolar coupling. We consider a simplified dipolar network where the dipolar interaction between the spin-pairs is assumed to be identical. Hence the dynamics remain confined in the block diagonal Hilbert spaces. For a suitable choice of the initial condition, the resulting dynamics require dealing with a smaller subspace which helps extend the analysis to a larger spin network. To include the nonsecular dipolar relaxation, we use a fluctuation-regulated quantum master equation. We note that a successful observation of superradiance in this system requires a weak system-bath coupling. Moreover, we find that for an ensemble of N spins, the maximum intensity of the radiation exhibits a nearly quadratic scaling (N^2), and the dipolar relaxation time follows an inverse square proportionality (1/N^2); these two observations help characterize the emergence of superradiance. Our results agree well with the standard results of pure spin superradiance observed experimentally in various systems.

quant-ph↗

arXiv2024

VALUED -- Vision and Logical Understanding Evaluation Dataset

Starting with early successes in computer vision tasks, deep learning based techniques have since overtaken state of the art approaches in a multitude of domains. However, it has been demonstrated time and again that these techniques fail to capture semantic context and logical constraints, instead often relying on spurious correlations to arrive at the answer. Since application of deep learning techniques to critical scenarios are dependent on adherence to domain specific constraints, several attempts have been made to address this issue. One limitation holding back a thorough exploration of this area, is a lack of suitable datasets which feature a rich set of rules. In order to address this, we present the VALUE (Vision And Logical Understanding Evaluation) Dataset, consisting of 200,000$+$ annotated images and an associated rule set, based on the popular board game - chess. The curated rule set considerably constrains the set of allowable predictions, and are designed to probe key semantic abilities like localization and enumeration. Alongside standard metrics, additional metrics to measure performance with regards to logical consistency is presented. We analyze several popular and state of the art vision models on this task, and show that, although their performance on standard metrics are laudable, they produce a plethora of incoherent results, indicating that this dataset presents a significant challenge for future works.

cs.CV↗

arXiv2024

Prethermalization in an open quantum system coupled to a spatially correlated Bosonic bath

A nearly-integrable isolated quantum many-body system reaches a quasi-stationary prethermal state before a late thermalization. Here, we revisit a particular example in the settings of an open quantum system. We consider a collection of non-interacting atoms coupled to a spatially correlated bosonic bath characterized by a bath correlation length. Our result implies that the integrability of the system depends on such a correlation length. If this length is much larger than the distance between the atoms, such a system behaves as a nearly integrable open quantum system. We study the properties of the emerging prethermal state for this case, i.e., the state's lifetime, the extensive numbers of existing quasi-conserved quantities, the emergence of the generalized Gibbs state, and the scaling of von Neumann entropy, etc. We find that for the prethermal state, the maximum growth of entropy is logarithmic with the number of atoms, whereas such growth is linear for the final steady state, which is the Gibbs state in this case. Finally, we discuss how such prethermal states can have significant applications in quantum entanglement storage devices.

quant-ph↗

arXiv2023

Emergence and stability of discrete time-crystalline phases in open quantum systems

Here we provide a theoretical framework to analyze discrete time-crystalline phases (DTC) in open quantum many-body systems. As a particular realization, we choose a quantum many-body system that exhibits cascaded prethermalization . The analysis uses a fluctuation-regulated quantum master equation. The master equation captures the dissipative effects of the drive and dipolar coupling on the dynamics regularized by the thermal fluctuations. We find that the dissipators from the drive and the dipolar interactions lend stability to the dynamics and are directly responsible for the robustness. Specifically, we find that longer fluctuation correlation time enhances the stability of DTC. Our results are in good agreement with the experiments. Finally, we show and quantify how the DTC performance degrades with temperature.

quant-ph↗

arXiv2022

Cascaded dynamics of a periodically driven dissipative dipolar system

Recent experiments show that periodic drives on dipolar systems lead to long-lived prethermal states. These systems are weakly coupled to the environment and reach prethermal states in a timescale much shorter than the timescale for thermalization. Such nearly-closed systems have previously been analyzed using Floquet formalism, which shows the emergence of a prethermal plateau. We use a fluctuation-regulated quantum master equation (FRQME) to describe these systems. In addition to the system-environment coupling, FRQME successfully captures the dissipative effect from the various local interactions in the system. Our investigation reveals a cascaded journey of the system to a final steady state. The cascade involves a set of prethermal or arrested states characterized by a set of quasi-conserved quantities. We show that these prethermal states emerge in a timescale much shorter than the relaxation timescale. We also find and report the existence of a critical limit beyond which the prethermal plateau ceases to exist.

quant-ph↗

arXiv2022

Effects of dipolar coupling on an entanglement storage device

Quantum computation requires efficient long-term storage devices to preserve quantum states. An attractive candidate for such storage devices is qubits connected to a common dissipative environment. The common environment gives rise to persistent entanglements in these qubit systems. Hence these systems act efficiently as a storage device of entanglement. However, the existence of a common environment often requires the physical proximity of the qubits and hence results in direct dipolar coupling between the qubits. In this work, we investigate the total effect of the dipolar coupling on the environment-induced entanglement using a recently-proposed fluctuation-regulated quantum master equation [A. Chakrabarti and R. Bhattacharyya, Phys. Rev. A 97, 063837 (2018)]. We show that nonsecular part of the dipolar coupling results in reduced entanglement and hence less efficiency of the storage devices. We also discuss the properties of efficient storage that mitigates the detrimental effects of the dipolar coupling on the stored entanglement.

quant-ph↗

arXiv2021

Dissipative phase transition in a spatially-correlated bosonic bath

The presence of symmetries in a closed many-body quantum system results in integrability. For such integrable systems, complete thermalization does not occur. As a result, the system remains non-ergodic. On the other hand, a set of non-interacting atoms connected to a regular bosonic bath thermalizes. Here, we show that such atoms in a spatially-correlated thermal bath can show both the behavior depending on the temperature. At zero temperature, the bath has a large correlation length, and hence it acts as a common environment. In this condition, a set of weak symmetries exist, which prevent thermalization. The system undergoes a symmetry-broken dissipative phase transition of the first order as the temperature rises above zero.

quant-ph↗