VSVansh Sharma

Vansh Sharma

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Vansh Sharma has a total of 18 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Kinetic Uncertainty in Hydrogen Jet Flames Using Lagrangian Particle Statistics

    by Shuzhi Zhang, Vansh Sharma, Venkat Raman

    Hydrogen-enriched fuel injection in staged gas-turbine combustors is commonly achieved through jet-in-crossflow (JICF) configurations, where flame stabilization is governed by a local balance between flow-induced strain/mixing and chemical reaction rates. This work investigates turbulent reacting JICF relevant to staged combustion conditions using high-fidelity simulations with adaptive mesh refinement (AMR) and differential-diffusion effects together with Lagrangian particle statistics. Chemistry model uncertainties are incorporated by using a projection method that maps uncertainty estimates from detailed mechanisms into the model used in this work. Results show that the macroscopic flame topology remains in a stable two-branch regime (lee-stabilized and lifted) and is primarily controlled by the jet momentum–flux ratio J. Visualization of the normalized scalar dissipation rate reveals that the flame front resides on the low-dissipation side of intense mixing layers, occupying an intermediate region between over-strained and under-mixed regions. While hydrogen content does not significantly change the global stabilization mode for the cases studied, uncertainty analysis reveals composition-dependent differences that are not apparent in the mean behavior alone. In particular, visualization in Eulerian (χ, T) state-space analysis and particle statistics conditioned on the stoichiometric surface demonstrate that higher-hydrogen cases observe a lower scalar dissipation rate and exhibit substantially reduced variability in OH production under kinetic-parameter perturbations, whereas lower-hydrogen blends experience higher dissipation and amplified chemical sensitivity. These findings highlight that, even in globally similar JICF regimes, the hydrogen content can modify the local response of the flame to kinetic-parameter uncertainty, motivating uncertainty-aware interpretation and design for hydrogen-fueled staging systems.

  • Reduced-Order Modeling of Turbulent Flows for High-Speed Aerothermoelastic Analysis

    by Elliot S. Kimmel, Daning Huang, Vansh Sharma, Jagmohan Singh, Venkat Raman, Peretz P. Friedmann

    This work presents a reduced-order model that efficiently generates unsteady aerothermal loads due to turbulent boundary layers for high-speed aerothermoelastic analysis. This prediction capability enables high-fidelity aerothermoelastic and structural fatigue analyses over long durations with tractable computational costs. Unsteady pressures over deformed structures are modeled by decomposing a turbulent boundary layer into temporal and spatial components. An unsteady pressure history over an undeformed structure is first reconstructed by superimposing spectral proper orthogonal decomposition modes and frequencies. Subsequently, the reconstructed pressures receive spatial corrections to model regions of flow compression and expansion. Heat fluxes over deformed structures are generated using a superposition/interpolation method that applies spatial corrections to the heat fluxes of an undeformed structure. An assumption required to construct the reduced-order model is verified: spatial pressure fluctuations and heat fluxes are a linear function of the modal coordinates of structural deformations in the linear regime. The reduced-order model produces accurate spectral contents of pressure fluctuations, which is critical for accurate structural excitation prediction; however, the model slightly overpredicts the fluctuation magnitudes. Likewise, the model produces accurate steady heat flux loads. The results enable high-fidelity unsteady aerothermoelastic simulations at a computational cost reduction of five orders of magnitude.

  • Open Access

    Chain of Unit-Physics: A Primitive-Centric Approach to Scientific Code Synthesis

    by Vansh Sharma, Venkat Raman

    Agentic large language models are proposed as autonomous code generators for scientific computing, yet their reliability in high-stakes problems remains unclear. Developing computational scientific software from natural-language queries remains challenging broadly due to (a) sparse representation of domain codes during training and (b) the limited feasibility of RLHF with a small expert community. To address these limitations, this work conceptualizes an inverse approach to code design, embodied in the Chain of Unit-Physics framework: a first-principles (or primitives)-centric, multi-agent system in which human expert knowledge is encoded as unit-physics tests that explicitly constrain code generation. The framework is evaluated on a nontrivial combustion task (12 degrees-of-freedom), used here as a representative benchmark for scientific problem with realistic physical constraints. Closed-weight systems and code-focused agentic variants fail to produce correct end-to-end solvers, despite tool and web access, exhibiting four recurrent error classes: interface (syntax/API) hallucinations, overconfident assumptions, numerical/physical incoherence, and configuration fragility. Open-weight models with chain-of-thought (CoT) decoding reduce interface errors but still yield incorrect solutions. On the benchmark task, the proposed framework converges within 5–6 iterations, matches the human-expert implementation (mean error of 3.1 e-3%), with a 33.4% faster runtime and a 30% efficient memory usage at a cost comparable to mid-sized commercial APIs, yielding a practical template for physics-grounded scientific code generation. As datasets and models evolve, zero-shot code accuracy will improve; however, the Chain of Unit-Physics framework goes further by embedding first-principles analysis that is foundational to scientific codes, thereby guiding more reliable and interpretable human–AI collaboration.

  • High-Fidelity Numerical Analyses for Hydrogen Jet-Flames in Crossflow

    by Shuzhi Zhang, Vansh Sharma, Venkat Raman

    This study compares two advanced numerical approaches in the context of investigating flame stabilization in hydrogen jets in crossflow (JICF). The first approach utilizes a highly resolved large-eddy simulation (LES) coupled with a strain-based flamelet manifold combustion model. The second approach involves a high-fidelity direct numerical simulation (DNS) with detailed chemical kinetics for the same JICF case. To capture the complex flame dynamics, the LES solver uses a static, highly resolved mesh, whereas the DNS solver employs adaptive mesh refinement (AMR) based on flame thickness. Both simulations are validated against experimental data, with a specific focus on flow regimes characterized by the Reynolds number, particularly mixing effects on flame stabilization. The flame dynamics are further analyzed through hydroxyl radical mass fraction contours and spatiotemporal scatter plots of key quantities. The study reveals that turbulent mixing regions are captured well by the LES model. However, on the lee side of the flame, differential diffusion significantly affects flame stabilization, which occurs further downstream in the DNS simulations. This comparison presents a holistic perspective on the capabilities of these methods in terms of spatial and chemical basis, highlighting their potential to study the dynamics of practical combustion systems under varying conditions.

  • Three-dimensional analysis of hydrogen fuel effects in multi-tube combustor

    by Shuzhi Zhang, Vansh Sharma, Venkat Raman, Tristan T. Shahin, Alexander J. Hodge, Rohan M. Gejji, Robert P. Lucht, Carson D. Slabaugh


  • Open Access

    Steering Conceptual Bias via Transformer Latent-Subspace Activation

    by Vansh Sharma, Venkat Raman

    This work examines whether activating latent subspaces in language models (LLMs) can steer scientific code generation toward a specific programming language. Five causal LLMs were first evaluated on scientific coding prompts to quantify their baseline bias among four programming languages. A static neuron-attribution method, perturbing the highest activated MLP weight for a C++ or CPP token, proved brittle and exhibited limited generalization across prompt styles and model scales. To address these limitations, a gradient-refined adaptive activation steering framework (G-ACT) was developed: per-prompt activation differences are clustered into a small set of steering directions, and lightweight per-layer probes are trained and refined online to select the appropriate steering vector. In LLaMA-3.2 3B, this approach reliably biases generation towards the CPP language by increasing the average probe classification accuracy by 15% and the early layers (0-6) improving the probe classification accuracy by 61.5% compared to the standard ACT framework. For LLaMA-3.3 70B, where attention-head signals become more diffuse, targeted injections at key layers still improve language selection. Although per-layer probing introduces a modest inference overhead, it remains practical by steering only a subset of layers and enables reproducible model behavior. These results demonstrate a scalable, interpretable and efficient mechanism for concept-level control for practical agentic systems.

  • Top Cited Papers

    Structure of three-dimensional conical oblique detonation waves

    by Sebastian Abisleiman, Vansh Sharma, Ral Bielawski, Venkat Raman