Portuguese 
Spanish 
Colossal simulation reached 55.5 trillion unknowns
Scientists led by the University of Texas at Austin have created an extreme-scale digital twin capable of predicting tsunami propagation in real-time, combining seabed sensors, advanced physical models, and high-performance supercomputing to anticipate waves generated by underwater earthquakes.
To tackle one of the biggest challenges of coastal alerts, the system transforms signals captured in the ocean into useful forecasts before the wave reaches land, drastically reducing the interval between the detection of an underwater event and the operational response.
According to the University of Texas at Austin, a task that could require 50 years of supercomputing time was reduced to a response in a fraction of a second, with a solution time of 0.2 seconds in one of the central steps of the process.
-
Collector adapts forklift electric motor in a Kombi Corujinha pickup, hides 16 batteries under the bed, and makes the Brazilian classic run silently with up to 90 km of range.
-
Goodbye traditional vacuum: Samsung surprises and launches 2-in-1 robot with steam, AI that recognizes floors and stains, self-cleaning station, hot air drying, automatic emptying, water refill, and app control.
-
Brazil is transforming acerola waste discarded by the pulp industry into an ambitious and surprising project that could forever change the way the country utilizes one of the most invisible by-products of the fruit sector.
-
Brazil anchored 42 floating concrete caissons in the Atlantic, used the world’s largest caisson dock, and erected more than 3.8 km of dikes to shield the Açu superport on the coast of Rio de Janeiro.
Digital twin uses seabed sensors to predict tsunamis
In practice, the technology functions as a virtual copy of a real physical phenomenon, capable of representing the ocean’s behavior after an underwater earthquake and estimating how the energy released at the seabed can transform into a tsunami.
Instead of relying solely on traditional seismic readings, the digital twin uses pressure data obtained from the seabed and three-dimensional physical equations to infer how the ocean floor moved after the rupture.
From this reconstruction, the platform simulates the formation and advance of the tsunami towards coastal areas, allowing the wave’s behavior to be analyzed before it reaches vulnerable points on the coast.
Recognized with the ACM Gordon Bell Prize, one of the main international awards in supercomputing, the research demonstrated significant advances in high-performance computing applied to science, engineering, and large-scale data analysis.
Leading the project is Omar Ghattas, professor of mechanical engineering and researcher at the Oden Institute for Computational Engineering and Sciences, University of Texas at Austin, who led the team responsible for the computational framework.
Colossal simulation reached 55.5 trillion unknowns
The size of the experiment helps to gauge the complexity of the work, even for those who do not follow the field of supercomputing, as the team executed a finite element simulation with 55.5 trillion unknowns.
According to the university, this calculation was run on 43,520 GPUs in the supercomputer El Capitan, at the Lawrence Livermore National Laboratory, with support from other high-performance systems, including Perlmutter, Alps, and Frontera.
As the main application, the researchers chose the Cascadia subduction zone, a seismic region that extends from northern California to British Columbia, Canada, monitored for its potential to generate large earthquakes and tsunamis.
In this Pacific Northwest strip, response time is a critical factor for coastal communities, as events near the coast can leave only a few minutes between the underwater tremor and the arrival of the first waves.
The University of Texas at Austin states that tsunamis of this type can reach the coast in up to 15 minutes after an earthquake, making any system that takes too long to transform raw data into operational guidance insufficient.
Supercomputer El Capitan accelerates coastal alert in real-time
The digital twin’s differential lies in the combination of anticipatory calculation and real-time response, an architecture that separates the heavier mathematical preparation from the emergency phase that needs to function while the threat is still in the ocean.
Before any emergency, the offline part prepares mathematical representations capable of accelerating the analysis, while the online phase uses data captured by sensors to produce probabilistic forecasts in less than a second when an earthquake occurs.
This division allows for the union of two difficult-to-reconcile requirements: high physical fidelity and speed, as detailed models usually require great computational power, while emergency alerts need to be issued almost immediately.
By bringing these two worlds closer, the project puts extreme supercomputing at the service of an operational decision that depends on seconds, especially in areas where evacuation needs to begin before the wave approaches the coast.
The structure also incorporated the so-called Bayesian inference, a statistical method used to update estimates as new information is received and to deal with natural uncertainties in complex events like tsunamis.
In the case of the tsunami, this method interprets pressure signals on the seabed to estimate the displacement of the ocean floor and calculate, with quantified uncertainties, the expected wave height at specific points on the coast.
Technology promises probabilistic predictions in less than a second
The University of Texas at Austin claims that the advancement represents a 10 billion times acceleration compared to methods considered state-of-the-art, a mark associated with the use of new algorithms designed for GPUs.
Among the technical results highlighted by the institution is the solution of a Bayesian inverse problem with 1 billion parameters in 0.2 seconds, a performance that allowed extremely complex scientific calculations to be approximated to a rapid alert application.
The project involved researchers from the University of Texas at Austin, the Scripps Institution of Oceanography at the University of California in San Diego, and the Lawrence Livermore National Laboratory, bringing together experts in oceanography, physical modeling, and supercomputing.
Among the names cited by the university are Stefan Henneking, Sreeram Venkat, Milinda Fernando, Alice-Agnes Gabriel, Veselin Dobrev, John Camier, Tzanio Kolev, and Omar Ghattas, who signed the award-winning research.
The proposal goes beyond reproducing water advancing over the ocean, as the system starts with the mechanism that gives rise to the tsunami, relating seismic rupture, seabed movement, and wave propagation.
This physical chain is important because small differences in initial displacement can alter the height, direction, and arrival time of the waves, directly affecting how coastal areas prepare for impact.
Coastal cities may gain more time before the arrival of waves
For emergency authorities, faster and more detailed predictions can help guide evacuations, position response teams, and reduce uncertainties in vulnerable regions, especially when the reaction margin is short.
The university claims that the framework was created to provide actionable alerts before the tsunami arrives, with wave height predictions and uncertainty margins at defined coastal locations.
The use of El Capitan shows how supercomputers, typically associated with physics research, national security, or climate modeling, can also be applied to problems directly related to protecting populations in coastal areas.
In this case, thousands of GPUs were used to solve an oceanic problem on an unprecedented scale, transforming nearly impracticable calculations into a rapid response tool for events that develop in a few minutes.
According to the University of Texas at Austin, extensions of the digital twin framework can serve predictive alert systems for other risks, including wildfires, severe weather, contaminant dispersion, and threat detection.
The common basis of these applications lies in the ability to cross-reference sensors, physics, and computing to represent an ongoing event before its effects spread, offering a faster reading of complex phenomena.
In the case of tsunamis, the potential impact is especially sensitive because the reaction window can be short and the threat crosses borders, affecting ports, coastal cities, industrial areas, and evacuation routes.
By combining ocean sensors, comprehensive physical models, and extreme processing, the digital twin developed by the team led by the University of Texas at Austin places tsunami forecasting on a new computational scale.
If a system can transform the invisible movement of the seabed into an almost instant alert, how much time will coastal cities still have to prepare before the next big wave?
