Objective
Researchers at Argonne National Laboratory and collaborating institutions have developed RADAR, a federated, privacy-enhancing framework that enables observatories to coordinate gravitational-wave and radio follow-up without moving or exposing proprietary data. Running on NVIDIA-accelerated supercomputing systems, RADAR demonstrates how multi-messenger discovery can be accelerated through secure, cross-observatory workflows that preserve data ownership. The RADAR paper has been accepted for publication in The Astrophysical Journal Supplement Series, one of astronomy’s leading peer-reviewed journals.
Customer
Argonne National Laboratory
Partner
William H. Miller III Department of Physics and Astronomy at Johns Hopkins University
Department of Computer Science at The University of Chicago
Department of Physics and Department of Astronomy at the University of Illinois Urbana-Champaign
Use Case
Accelerated Computing Tools & Techniques
Products
Key Takeaways
5–10x Speedup
- Achieved monumental speedups using NVIDIA GPUs compared to CPU workloads.
Collaborative Sharing Enablement
- Established unified data sharing practices within the radio astronomy community.
Faster Processing of Raw Data
- Utilized NVIDIA GPUs on the Polaris, Delta, and DeltaAI supercomputers to process raw LIGO data at unprecedented speeds.
Too Much Data, Too Few Unified Systems
In multi-messenger astrophysics, rapid and locally executed data processing is essential for identifying gravitational-wave events. Detecting these signals promptly requires analyzing large, continuous data streams directly at the detector sites to minimize transfer delays and enable fast, reliable extraction of events.
Once a gravitational-wave trigger is identified, coordinated electromagnetic follow-up depends on fast, structured communication of event information. Radio afterglows evolve slowly and require long-term monitoring, which makes early contextualization of events crucial for guiding observing strategies and ensuring efficient use of limited telescope time.
The large localization areas associated with gravitational-wave detections, combined with the narrow fields of view of sensitive radio arrays, make timely and well-informed follow-up essential for maximizing scientific return.
This challenge is exacerbated by heterogeneous data-sharing policies across the radio astronomy community. Proprietary periods vary widely—from immediate public dissemination to embargoes lasting until peer-reviewed publication—rendering centralized data movement impractical and slowing the aggregation of observations needed to refine physical models in real time.
Without improved mechanisms for cross-observatory coordination, the community risks missing key multi-messenger opportunities.
Moreover, these constraints historically limited broad participation in radio follow-up campaigns, as institutions were understandably reluctant to transfer raw or unreleased data. The absence of a privacy-enhancing, federated approach meant that valuable measurements remained siloed, restricting the collective ability to model afterglows, update source parameters, and enable responsive multi-messenger science.
Argonne National Laboratory
Argonne National Laboratory
RADAR Paves the Way for Cohesive Sharing Policies
The RADAR framework is the product of a broad, multi-institutional collaboration and leverages computational resources from several U.S. national supercomputing facilities. These systems provide the distributed, high-performance infrastructure that enables RADAR’s federated, privacy-enhancing workflows, including:
- The Delta and DeltaAI supercomputers at the National Center for Supercomputing Applications
- The Polaris supercomputer at the Argonne Leadership Computing Facility
This framework is designed as an event-driven, federated system that operates within the multi-messenger astrophysics ecosystem to coordinate radio follow-up of gravitational-wave events while preserving data security and respecting diverse data-sharing policies.
Its lightweight, modular, and extensible design provides a generalizable foundation for distributed, collaborative science—one that can be readily adapted to additional messengers, electromagnetic bands, and evolving community norms for data exchange.
“RADAR gives us a way to plan and adapt follow-up strategies, even when the data itself can’t be shared directly,” said Alessandra Corsi, W. H. Miller professor of physics and astronomy at Johns Hopkins. “This capability will become increasingly critical as next-generation detectors transform today’s trickle of multi-messenger detections into a flood.”
A central design principle of RADAR is to enable community-driven sharing of high-level information needed to design and refine observing strategies, an essential requirement given limited radio observing resources and the anticipated rise in gravitational-wave detections.
RADAR is explicitly complementary to existing pipelines within the LIGO–Virgo–KAGRA collaboration, which comprises NASA’s Gravitational Wave Network, and other community initiatives—providing an integrative layer that enhances coordination without replacing established low-latency or data-analysis frameworks.
Integrating RADAR Into the System
The RADAR framework identifies gravitational-wave events using site-local AI inference and matches these detections with public LIGO–Virgo–KAGRA “superevent” alerts to determine whether coordinated follow-up is warranted. Rather than transferring raw data, RADAR adopts a privacy-enhancing approach in which all observatories retain ownership of their local datasets while exchanging only high-level parameters or model outputs needed for global inference and collaborative planning.
When a matching superevent is found, RADAR publishes a trigger to Octopus—a cloud-hosted, event-driven messaging fabric—often using ProxyStore to handle large intermediate results efficiently and securely. This trigger activates the radio and multi-messenger modules, which employ an AI-powered parser to convert unstructured General Coordinates Network (GCN) circulars into structured, machine-readable metadata. These public alerts are then combined with proprietary radio measurements aggregated from geographically distributed observatories participating in the federated workflow.
Subsequently, RADAR integrates information across messengers by running Dingo-BNS for gravitational-wave parameter estimation and afterglowpy for federated radio afterglow modeling. The resulting posteriors are compared and combined to refine key source parameters—demonstrating end-to-end multi-messenger inference while preserving data locality and respecting heterogeneous data-sharing policies.
“Advanced accelerated computing platforms have become essential for modern multi-messenger astrophysics.”
Eliu Huerta
Lead for Translational AI at Argonne National Laboratory
Accelerating AI-Driven Discovery
The researchers advanced their RADAR framework by using NVIDIA A100 Tensor Core GPUs, NVIDIA A40 GPUs, and NVIDIA GH200 GPUs. Upgrading to NVIDIA architectures delivered significantly higher performance that was essential for handling the escalating scale and complexity of data for modern, next-generation observatories.
Integrating NVIDIA GPUs on the Polaris, Delta, and DeltaAI supercomputers, the framework’s AI module for signal detection processed over an hour—4,096 seconds—of advanced LIGO data in under 4.5 minutes. This translates to a speedup of 5–10x compared to the initial timeframe when running on CPU workloads.
NVIDIA’s technology enabled collaborative, privacy-enhancing workflows across leading supercomputing centers, ensuring that sensitive data remains secure at local sites. Only selected results are shared within the broader radio astronomy community, promoting scientific cooperation without compromising data ownership.
Additionally, NVIDIA solutions provided the high-performance, on-site computational power necessary to process massive datasets in real time—allowing for immediate identification and deployment of resources during critical astrophysical events.
This combination of exceptional computational power, scalability, and privacy features set NVIDIA’s solution apart from competing alternatives, making it the clear choice for the RADAR project.
“Advanced accelerated computing platforms have become essential for modern multi-messenger astrophysics, enabling us to train and deploy AI models at the scale these challenges demand,” said Eliu Huerta, lead for translational AI at Argonne National Laboratory. “Our recent work demonstrates how high-performance AI frameworks can extract deeper insight from complex, distributed observatory data—helping the community move faster toward time-critical discovery.
Sharing Insights Without Spilling the Secrets
RADAR shows that meaningful collaboration is possible without compromising data rights. The framework enables model-informed, asynchronous planning across distributed facilities, allowing early-time radio results to inform later-time strategies and optimize telescope usage. By combining federated analysis with community-driven data sharing, RADAR boosts resource efficiency and system resilience, helping the community meet the growing demands of gravitational-wave monitoring.
Explore how accelerated computing can transform your data center.