NVIDIA Demonstrates How AI Coding Agents and ALCHEMI Toolkit Accelerate GPU-Based Materials Simulations

NVIDIA Demonstrates How AI Coding Agents and ALCHEMI Toolkit Accelerate GPU-Based Materials Simulations NVIDIA has detailed how AI coding agents combined with the NVIDIA ALCHEMI Toolkit can overcome interface barriers in Machine Learning Interatomic Potentials (MLIP) simulations. By equipping agents like Claude Code with specialized skills and runnable environments, researchers can reliably generate and run GPU-accelerated simulation pipelines from natural-language descriptions. ALCHEMI Toolkit lowers MLIP simulation barriers using PyTorch-native, GPU-accelerated blocks with in-flight batching, paired with agent skills to guide AI coding tools. A systematic benchmark evaluated 45 simulation pipelines across three tasks (silicon equation of state, oxygen adsorption on Cu(111), and liquid lithium self-diffusion) and five prompt levels. Letting the agent run generated scripts in an active environment prevented broken imports, though GPU-specific paths still accounted for seven screening failures. Prompt specificity improved code structure, token cost, and interface reusability, but did not alter the core underlying physics outcomes. Agents do not assess physical validity on their own; researcher oversight, explicit protocol constraints (e.g., specifying NVE over Langevin dynamics), and independent reference comparisons remain critical. Bridging the Interface Barrier in MLIP Simulations Atomistic simulations using MLIP have traditionally lacked the accessible interfaces common in classical force fields. While ALCHEMI Toolkit provides GPU-accelerated, batch-enabled building blocks, general-purpose AI coding agents often hallucinate or misuse its APIs. Providing agents with dedicated ALCHEMI Toolkit agent skills and reference files addresses this issue, allowing researchers to specify scientific requirements rather than implementation syntax. Environment Setup and Prompting Practices The recommended setup requires Python 3.11 to 3.13, PyTorch 2.8 or higher, and CUDA 12 or 13 on compatible NVIDIA GPUs. Installing ALCHEMI Toolkit (version 0.2.0) with uv and pairing it with an agent capable of executing code—such as Claude Code running Claude Opus 4.8—eliminated broken imports across the benchmark. Key prompt strategies include naming the material, phase, scale, and reference conventions explicitly, describing constraints rather than internal API classes, and directly requesting validation checks. Three Benchmark Workflows on NVIDIA H200 GPUs The benchmark tested 45 pipelines across three problems using the MACE-MPA-0 checkpoint. For bulk silicon equation of state, all production representatives yielded identical lattice constants (a0 = 5.4661 Å) and bulk moduli (B0 = 88.15 GPa). For oxygen adsorption on Cu(111), scripts consistently identified the fcc hollow site as most stable (-4.799 ± 0.004 eV). For liquid lithium self-diffusion, prompt instructions proved crucial: unconstrained prompts defaulted to Langevin dynamics that suppressed diffusion by 3-5× due to thermostat friction, whereas explicit NVE requests matched experimental expectations. Benchmark Insights and Scientific Limitations Increased prompt specificity improved API-pattern coverage and code reusability—rising to 1.00 at the contract prompt level—at the expense of higher token counts and script lengths. However, agents exhibited biases, such as defaulting to the FIRE algorithm over FIRE2, and never questioned whether requested simulations were physically meaningful. Furthermore, MLIP models like MACE-MPA-0 carry inherent accuracy bounds, underscoring that coding agents cannot replace independent validation against DFT or experimental standards. Ecosystem Adoption Industry partners are incorporating these workflows into production platforms. In May 2026, Matlantis introduced a public Skills library on GitHub with Claude Code integration for its universal simulator. Additionally, Dassault Systèmes' BIOVIA is integrating NVIDIA ALCHEMI NIM into its virtual companion, MARIE, to expand molecular simulation and Virtual Twin capabilities.

结合 NVIDIA ALCHEMI Toolkit 与 AI 编程智能体加速材料模拟工作流 NVIDIA 探讨了通过 AI 编程智能体（如 Claude Code）配合 ALCHEMI Toolkit 构建 GPU 加速的机器学习原子间势（MLIP）模拟工作流。基准测试分析了 45 个端到端流水线，涵盖提示词工程、GPU 批处理性能及生成代码的可靠性。 ALCHEMI Toolkit 提供 PyTorch 原生的批处理模拟组件，结合 Agent Skills 可解决通用智能体不熟悉专用 API 的问题。 系统配置要求包括 Python 3.11–3.13、PyTorch ≥2.8、CUDA 12/13 及算力 ≥7.0 的 NVIDIA GPU，提供可执行环境能显著减少代码语法错误。 测试涵盖硅状态方程、铜表面氧吸附及液态锂自扩散三个任务，全部 45 个流水线均成功实现了 GPU 批处理执行。 提示词特异性主要提升代码结构与接口复用性，但不改变物理结果，且智能体缺乏自主物理判别能力，仍依赖人类科学判断。 环境配置与 Agent Skills 集成 原子模拟需克服科学认知、高效计算与易用接口三大障碍。针对 MLIP 软件栈接口门槛，开发者可通过 uv 安装 nvalchemi-toolkit 并同步下载 release 标签对应的 Agent Skills。基准测试使用 Claude Code，并在本地创建可运行的 Python 环境。实践表明，允许智能体直接执行代码能消除导入错误和不存在 API 的引用，比无运行环境的静态代码生成显著更可靠。 五级提示词实践与基准结果 测试建立了从草图（Sketch）到完整契约（Contract）的五级提示词梯队。在钻石立方硅状态方程、Cu(111) 氧吸附和液态锂自扩散三个工作流中，45 个脚本在 NVIDIA H200 上测试，其中 38 个通过筛选，15 个生产代表脚本全部完成运行。硅的晶格常数（5.4661 Å）与体弹性模量（88.15 GPa）在各提示词级别下完全一致，证明提示词特异性改变的是代码规模与架构，而非物理计算核心。 生成代码的算法偏好与硬件依赖 提示词详细度提升能使 API 模式覆盖率从低级别的 0.52 升至 0.96，但完全的命令行接口复用性会导致生成代码量翻倍且消耗约 4 倍的 token。同时，智能体倾向于使用预训练中熟悉的算法（如 38 个脚本选用 FIRE 算法而忽略文档提及的 FIRE2）。此外，虽然 CPU 沙盒自测排除了大部分机械错误，但仍有 7 个脚本在 GPU 专属路径上失败，凸显了在目标硬件上实际运行校验的必要性。 核心局限与生态落地 智能体目前缺乏物理合理性审查能力，无法质疑不符合物理现实的任务设定（如在短模拟时间内测量晶体 LiF 的锂扩散），若提示未指明系综，智能体默认使用的 Langevin 热浴会使扩散系数因人为阻尼偏低 3 至 5 倍。底层 MLIP 基础模型（如 MACE-MPA-0）亦存在分布外精度局限，因此数值模拟仍需独立基准比对。当前，Matlantis 与达索系统 BIOVIA MARIE 等平台已开始整合相关方案以推进材料虚拟孪生构建。

Atomistic simulation requires three things: knowledge of the science, compute-efficient implementation of simulations, and accessible interfaces to the simulation stack. The first remains the researcher’s domain, as no tool substitutes for knowing what to simulate or recognizing a physically meaningful result. NVIDIA ALCHEMI Toolkit, introduced earlier this year, has dramatically reduced the second barrier for Machine Learning Interatomic Potentials (MLIP) with composable, PyTorch-native building blocks for constructing GPU-accelerated simulation workflows with in-flight batching enabled. The third barrier has persisted. Unlike classical force fields, the MLIP ecosystem is still nascent, and the accessible interfaces that exist for classical simulations are very limited. They run on a different software stack than the tools many computational chemists are accustomed to with new data structures, composition patterns, and dependencies. AI coding agents offer a way through: They generate and execute code from natural-language descriptions written in the terms a researcher would use in daily technical discussions. But a general-purpose agent may not know the ALCHEMI Toolkit API, and can produce plausible-looking code that only appears to use it correctly. ALCHEMI Toolkit agent skills and reference files provide the missing API patterns on demand, leaving the prompt to your science: the material, the conditions, and the constraints on the simulation protocol. Building simulation workflows with a coding agent This post follows an end-to-end ALCHEMI Toolkit workflow: what the researcher starts with, how they prompt the agent, what code and simulation pipeline it produces, and how the results are validated on NVIDIA H200 GPUs. It also distills lessons from 45 generated pipelines into practical guidance for building trustworthy GPU-accelerated simulation workflows with coding agents. How to get started System and package requirements Python ≥3.11, <3.14 PyTorch ≥2.8 CUDA 12 or CUDA 13, with a compatible NVIDIA driver (570+ recommended) Operating System: Linux (primary), macOS NVIDIA GPU (RTX 20xx or newer), CUDA Compute Capability ≥ 7.0 Minimum 4 GB RAM (16GB recommended for large systems) Installation How you set up the agent’s environment meaningfully affects the reliability of generated code. We recommend installing the Toolkit in a runnable Python environment and letting the agent execute the scripts it generates. In the final 45-pipeline campaign, this setup produced no broken imports or references to nonexistent APIs. Step 1: Create a Python environment and install ALCHEMI Toolkit with uv package manager : # Installation through uv in a local folder # Create local environment at .venv uv venv --seed --python 3.12 # Install ALCHEMI Toolkit into .venv uv pip install "nvalchemi-toolkit[mace,ase]==0.2.0" To run on an NVIDIA GPU, include the CUDA extra that matches your CUDA environment. For example, for CUDA 13, install nvalchemi-toolkit[mace,ase,cu13]==0.2.0 . Step 2: Download agent skills from the same release tag so they match the installed API: # Download the nvalchemi-toolkit skills npx degit NVIDIA/nvalchemi-toolkit/.claude/skills#v0.2.0 .claude/skills Step 3: Install a coding agent. This benchmark used Claude Code: # Install Claude Code through npm npm install -g @anthropic-ai/claude-code # Or natively through curl curl -fsSL https://claude.ai/install.sh | bash # Start a claude session in the local folder claude Open the agent in your project directory, allow it to execute code, and start describing simulations; the agent loads the relevant skills on demand. Letting the agent run what it writes removes almost all mechanical errors before you see the script. From there, the quality is decided mostly by your prompt. Without a runnable shell, a source git-checkout is a fallback: In earlier tests, reading the source eliminated broken imports across 617 import statements. A pip install with neither shell nor source was the weakest configuration. Any agent supporting the open Agent Skills standard works, including Cursor and OpenCode (see the Agent Skills user guide for configuration). Best practices for writing a good prompt The following principles are synthesized from measurable differences in the benchmark’s output quality, not from general coding-agent advice. The five prompt levels, from least to most specified, are shown in Figure 1, below. Figure 1. The five-level prompt ladder, illustrated on the liquid-lithium self-diffusion workflow. Each level strictly adds to the one before; Goal is a good starting point, Recipe is recommended when the result depends on simulation protocol Name the system, method, and scale always; add a CLI contract only for unattended operation . Prompts naming the material, method, and scale scored best. Full CLI contracts achieved complete reusability, but cost about 4× more tokens and produced 2.3× more code than Sketch prompts. Spec prompts were most fragile, accounting for three of seven screening failures. Specify the science and deliverables; let the skills supply API patterns. Name the material, phase, and reference convention explicitly . Under-specifying the system caused the clearest physics failures in earlier tests: “a transport property of a Li material” produced an argon demo, and two Cu scripts used different adsorption references, materially affecting the outcomes. The final prompts eliminate those failures by naming the material, phase, and reference convention. Protocol matters too: Scripts without a thermostat instruction used Langevin production dynamics that damped diffusion by 3-5×; requesting NVE changed every script to the appropriate measurement ensemble. Specify the constraint, not the implementation. Describe what the script must accomplish rather than internal API classes. In a controlled comparison, naming a pipeline construct changed none of 12 implementations. The API pattern comes from the examples and skills, not the prompt. Naming internals without a complete interface contract is where fragility concentrates. Ask explicitly for self-assessment and premise-checking. Agents did not question whether any requested property was physically well-posed. Ask explicitly for premise checks, validation, and uncertainty estimates, and require recovery of an independent known result; the agent won’t add these on its own. Three workflows, from prompt to GPU execution The following workflows illustrate what ALCHEMI Toolkit and a coding agent can handle end-to-end from a natural-language prompt. To ground these recommendations in measured results, we ran a systematic benchmark of 45 simulation pipelines across three workflows, five prompt levels, and three samples per level. Silicon equation of state (EOS) Oxygen adsorption on Cu(111) Li self-diffusion molecular dynamics (MD) Scripts were assessed two ways. Firstly, deterministic code-feature checks: property coverage (right quantity, right formula), API-pattern coverage (use of the toolkit’s batched API surface), and reusability (a parameterized, re-runnable interface). Secondly, execution on identical NVIDIA H200 GPUs as ground truth. All 45 scripts used batched GPU execution. L2-L5 prompts required batching; L1 adopted it from the Toolkit skills and examples. We screened all scripts at demonstration scale and ran one representative per workflow and level at production settings. Figure 2. H200 execution outcomes for 45 benchmark pipelines. Thirty-eight completed screening; all 15 production representatives completed Bulk silicon equation of state The task: The equation of state of diamond-cubic silicon The energy-volume curve whose minimum gives the lattice constant a0 and whose curvature gives the bulk modulus B0. At every prompt level the agent built the same pipeline: 50-60 strained volumes relaxed concurrently as one GPU batch, then a Birch-Murnaghan fit. All five production representatives agree to the last digit, a0 = 5.4661 Å and B0 = 88.15 GPa, inside the established all-electron PBE reference window . The lattice constant differs from the NIST experimental value , while the bulk-modulus comparison comes from McSkimin’s experimental measurements . These offsets are consistent with reported PBE behavior .  The takeaway mirrors the benchmark-wide pattern: prompt specificity changed the code’s structure and cost, never the physics. Figure 3. Diamond-cubic silicon lattice constant and bulk modulus across prompt levels Oxygen adsorption on Cu(111) The task: Rank the adsorption sites for atomic oxygen on Cu(111) The pipelines freeze the slab’s bottom layers, relax 24+ candidates across the four high-symmetry sites as one GPU batch, and compute E ads = E slab+ads − E clean slab − E O . Every production representative finds the fcc hollow most stable, E ads(fcc) = −4.799 ± 0.004 eV across the five levels, with the ordering fcc ≥ bridge > hcp >> top. Reported reference values are −4.31 eV at 0.25 monolayer. Copper microcalorimetry studies by Giamello et al. and Naumann d’Alnoncourt et al. reported values from −4.46 to −4.60 eV. Coverage differences and model error may contribute to the stronger binding. Millielectronvolt agreement across independent scripts is what a stated reference convention buys. Figure 4. Most-stable O/Cu(111) adsorption energy across prompt levels, compared with DFT-PW91 and microcalorimetry at their reported coverages Lithium self-diffusion via molecular dynamics The task: Estimate the self-diffusion coefficient D of liquid lithium. The pipelines build a bcc lithium supercell, melt and equilibrate it above lithium’s 454 K melting point, propagate three temperatures times three seeds (nine replicas) as a single batched system on the GPU, and extract D from the mean-squared displacement, MSD — the average squared distance that atoms travel — whose slope over time gives D through the Einstein relation. Prompt level Production ensemble D at 600 K (cm 2 /s) D at 800 K (cm 2 /s) D at 1000 K (cm 2 /s) L1 (Sketch) Langevin – 1.21 x 10 -4 – L2 (Goal) Langevin 0.80 x 10 -4 1.03 x 10 -4 1.55 x 10 -4 L3 (Recipe) NVE 2.91 x 10 -4 5.38 x 10 -4 7.22 x 10 -4 L5 (Contract) NVE 3.62 x 10 -4 4.32 x 10 -4 6.49 x 10 -4 Table 1. Liquid-lithium self-diffusion by prompt level and temperature. L4 is excluded because its generated script did not conserve energy Figure 5. Liquid-lithium self-diffusion versus temperature. NVE runs are within about 2× of the experimental extrapolation; Langevin runs are 3 to 5× lower than NVE because thermostat friction suppresses transport. L4 is excluded because its generated script did not conserve energy Technical specifications The following configuration was used to produce the workflow examples in this post: Coding agent: Claude (claude-opus-4-8) Effort level: High MLIP checkpoint: MACE-MPA-0 (medium-mpa-0) Toolkit version pin (0.2.0) Prompt level Total tokens processed (incl. cache) Generated tokens Iterations Script length (LOC) Sketch ~2.4M ~32k 44 498 Goal ~3.3M ~39k 48 451 Recipe ~4.4M ~43k 60 574 Spec ~8.9M ~81k 88 788 Contract ~10.0M ~107k 84 1,168 Table 2. Per-cell generation cost, iterations, and code length by prompt level. Prompt-cache reads dominate total token counts Figure 6. Generation cost and script size versus prompt level Benchmark insights The benchmark produced three practical findings by systematically varying prompt level, tool access, and execution capability. Prompt specificity buys code structure. Property coverage was 1.00 at every level, so the science is right from the first prompt. What specificity brought was structure: API-pattern coverage doubles at L4-Spec (0.52 to 0.96) and full interface reusability arrives only with the L5-Contract (0.67 to 1.00) (Figure 7, below). Figure 7. Line chart of three coverage metrics versus five prompt levels Agents defaulted to familiar algorithms: 38 of 45 scripts used FIRE and none used FIRE2, despite FIRE2 being documented beside it as an improved variant. What the examples and skills showcase, the agent uses; what they do not, it fills from pretraining. Execution exposed failures that CPU self-tests could not. The runnable shell eliminated broken imports, but seven scripts still failed on GPU-specific paths. More elaborate prompts exercised more of these paths and failed more often. Self-testing improves code generation, but does not replace validation on target hardware. A generation sandbox with GPU access may close this gap, but our methodology accounts for what a typical user would do in practice: Use “Claude autorun” in the Claude app on their laptop and then submit to a GPU cluster after. Limitations Scientific judgment remains essential. Across every tested condition, zero configurations pushed back on a physically ill-posed task. In an earlier probe, every agent complied with a request for Li-ion diffusion in pristine crystalline LiF, a property not measurable at the simulated timescale. However, it is possible a web-equipped agent would run this check; the generation sandbox had no web access, so the agent could not have looked up whether the property was plausible. Agents also chose plausible timesteps and thermostats without assessing their suitability for the material or timescale. Agents accelerate translation from scientific intent to code, but cannot determine whether the intended simulation is physically meaningful. MLIP foundation models are not universally accurate. The benchmark used MACE-MPA-0 (medium-mpa-0) , whose accuracy varies outside its training distribution. Our silicon, copper, and liquid-lithium results are reasonable against their references, but every new chemistry should be validated against DFT or experiment. This is a model and sometimes reference DFT limitation, not an ALCHEMI Toolkit or agent limitation. Always compare numerical results with an independent reference. The Langevin-production lithium scripts reported diffusion coefficients that were statistically impeccable yet damped 3-5× by the thermostat; only comparison against the temperature-matched experimental anchor exposes it. The check must also be independent: one script validated its unit conversion by generating synthetic data with the same wrong constant, so its self-test passed. Statistically reliable and physically meaningful can be orthogonal. Ecosystem adoption Several companies have been exploring the approaches we discuss in this blog. In May 2026, Matlantis, an AI-powered simulation platform used by over 150 companies and an ALCHEMI ecosystem partner, released a public Skills library on GitHub and announced a Claude Code integration for its universal simulator. BIOVIA is also leveraging NVIDIA ALCHEMI NIM in Dassault Systèmes’s AI-powered virtual companion, MARIE, to democratize access to advanced simulation capabilities and enable convenient execution of molecular simulations, empowering more scientists to develop and validate Virtual Twins of materials. Get started building simulation workflows with ALCHEMI Toolkit and coding agents Visit the NVIDIA/nvalchemi-toolkit GitHub repo and NVIDIA ALCHEMI Toolkit documentation . The repository includes agent skills and more than 30 workflows across basic, intermediate, advanced, and distributed use cases. Acknowledgments We would like to thank Nikita Fedik, Susumu Ohno from Matlantis Corporation, and James Wescott from Dassault Systèmes – BIOVIA for their contributions to this post.