If supercomputers had a motto, it would be “live fast and die young” as rapid advances mean that supercomputing technology becomes dated within just a few years.
KAUST’s supercomputer, Shaheen II, has powered diverse research from climate forecasting to molecular modeling, but is now approaching the end of its cycle. Shaheen means falcon in Arabic. Shaheen II is due for replacement in 2023. Meanwhile, the machine continues to punch above its weight in terms of the world-class research it powers across different disciplines and diverse projects.
Like the very first supercomputers, Shaheen II is used to simulate weather. Researchers led by mathematician Ibrahim Hoteit use the capacity of Shaheen II to simulate, understand and predict atmospheric and oceanic circulations over and around the Arabian Peninsula.
Engineers planning KAUST’s neighboring King Abdullah Economic City’s coastal development originally foresaw potential future storm surges 4 meters high. Exploiting Shaheen’s capacity, Hoteit’s models, however, showed that a nearby coral reef would dissipate waves from the Red Sea to reduce the storm-surge risk. This meant the elevation of the project’s foundations could be reduced by 2 meters, saving around US$600 million.
Risk levels were also clarified for the coastal megacity NEOM, being constructed on the northeastern shores of the Red Sea. Hoteit’s simulations predict freak wind episodes, now factored into the new plans.
Hoteit claims his research group’s biggest achievement has been the discovery and modeling of the Red Sea’s natural circulatory flow. This helped direct plans for oil spill cleanup and explain patterns of seasonal circulation cycles driven by the Indian monsoon and their impact on the biological productivity of the Sea’s deep-water basin. This natural circulation occurs when surface waters are blown northwards during winter, where it becomes saltier and cooler and sinks at the northern reaches of the Red Sea, before returning southwards as a deep current. This cycle reverses during the summer.
Sporadic events were also examined. Faraway volcanic eruptions in 1982 and 1991 produced unusually cold winters over the Sinai Peninsula. During these events, the water in the northern Red Sea sank deeper, with knock-on benefits across the ecosystem. “The colder water went deep down, taking oxygen with it. It sped up the cycle, mixed the water column and ventilated the basin sea,” says Hoteit.
In 2014, NASA called for a revolution in computational fluid dynamics predicting that by 2030 there would be new algorithms running on supercomputers to simulate the flow of air around spacecraft and high-performance aircraft more comprehensively and accurately.
However, KAUST researchers ran ahead of NASA’s prediction. Led by aerospace engineer and computational scientist Matteo Parsani developed the first prototype of next-generation fluid dynamic solver and tested it on Shaheen II early in 2021. Parsani’s team is now collaborating with scientists from NASA, and other aerospace heavyweights Boeing and Airbus.
“We developed the world’s first fully discrete entropy-stable adaptive solver for complex geometry,” says Rasha Al Jahdali, applied mathematician on this forward-looking project. “The aerospace industry knows that we have this unique capability. They want us to run simulations to see how it performs.”
Developing a new aircraft model is expensive. “Engineers select one option from several designs and then go through the costly process of building a prototype to test in a wind tunnel,” says Al Jahdali. Parsani’s system enables this to be done virtually, simulating many designs and tweaking them during the process. It allows developers to look at airflow at every point of their design, and under conditions impossible to replicate in a wind tunnel.
“The aerospace industry knows that we have this unique capability. They want us to run simulations to see how it performs.”
The science can also help build faster cars; Formula 1 developer McLaren already licenses Parsani’s software. KAUST’s research puts Saudi Arabia in the vanguard of fundamental upstream contributions to the software infrastructure of computational fluid dynamics. As part of the Kingdom’s plans to diversify its economy, this know-how will support indigenously developed aircraft by the mid-2030s. In line with Vision 2030, Saudi Arabia plans to localize aerospace industries. KAUST will be able to support these translation opportunities.
Computational scientist Peter Richtarik and his team develop algorithms that train machine learning applications. The team, including several graduate students, has also demonstrated their own successful learning: in the past three years, they have contributed five papers to the prestigious International Conference on Machine Learning.
The team’s machine learning applications perform tasks such as recognizing faces or translating text into another language. Currently, algorithms learn by looking and learning from millions of examples. However, this process is inefficient. Richtarik has developed “arbitrary sampling,” which uses more efficient methods so less examples are needed to train machine learning applications.
Richtarik’s original papers on the machine learning subfield have been followed and cited by thousands. Now, the biggest tech companies, including Google, Apple, Facebook, Samsung, Huawei and Tencent, have established dedicated federated learning research teams.