Euro-Q-Exa in the Experimental Phase

The research team led by atmospheric physicist Dr Mierk Schwabe from the German Aerospace Centre (DLR) is working to harness the potential of quantum computers for Earth systems and climate models. One aim of the DLR’s quantum computing initiative (DLR QCI) – and specifically the Klim-QML project – is to use quantum machine learning (QML) to improve and expand existing models, thereby enabling more accurate forecasts and projections, as well as allowing for a more precise analysis and assessment of the impacts of weather and other natural phenomena on the aerospace, transport and energy sectors. “These developments for climate protection are still in their infancy,” says Schwabe. “Climate models are often too large even for supercomputers; their resolution is therefore often still very coarse, and processes such as clouds, their formation or turbulence cannot be resolved at all.” However, because they are important for weather conditions and climate, they are parameterised – that is, simplified in separate models and simulated using smaller computational scales: “These ‘sub-models’ are not perfect; they lead to errors, which we traditionally compensate for through machine learning,” explains Schwabe. Because these processes are nevertheless important for weather patterns and climate, they are parameterized—that is, simplified in separate models: “These ‘sub-models’ are not perfect and lead to errors that we classically compensate for using machine learning,” explains Schwabe. Her team is now transferring this process to quantum computing, also because the computing power of new technologies raises hopes for more detailed and comprehensive climate and Earth system models while enabling new computational methods for simulations. “We were very happy to get the opportunity to test these methods on Euro-Q-Exa, a very performant superconducting quantum computer that offers us new possibilities besides the access to the DLR QCI quantum computers”, says Schwabe.

Dr. Hedwig Keller, a mathematician in Schwabe’s group who develops methods to couple QML-models to the climate model, therefore conducted experiments on Euro-Q-Exa with two smaller models. “The transition from simulated to real quantum computing was much easier than expected,” she says. “We ran two quantum models, a simple, two dimensional wave function and a larger model for cloud cover. We had previously trained both in a quantum computing simulation.” In March, Keller ran these programs on Euro-Q-Exa on several days, working with 2 or 6 qubits, testing computational units of varying quality, and attempting to parallelise jobs—that is, to run multiple models with different values simultaneously. “Although the Euro-Q-Exa system was recalibrated every day, the results were already quite similar,” the mathematician observed. “Quantum hardware is still noisy, which affects models and results, but this influence seems consistent to a certain degree.” Consequently, it can probably be minimised through changes in programming or through a hardware-specific optimisation.

Euro-Q-Exa provides 54 qubits; computers of this size are still considered “noisy” (Noisy Intermediate-Scale Quantum, NISQ) and error-prone. Despite regular calibration, the quality of the qubits fluctuates. Therefore, during the test phase of Euro-Q-Exa, several groups also developed strategies for error mitigation and correction — forming the basis for developing their own programs and more reliable hardware. Euro-Q-Exa is based on superconducting circuits. Its processing units, the qubits, are cooled and stabilised by a cryostat and activated by microwave pulses to assume states between 0 and 1, or superposition, and to become entangled with one another via quantum gates. Quantum circuits then arrange these gates in the correct temporal sequence for algorithms and functions.

Through entanglement and superposition, quantum computers gain computing power and speed: they can process calculations with multiple inputs simultaneously, operate exponentially faster, and each additional qubit multiplies performance. With 54 qubits, Euro-Q-Exa is already approaching the limits of the size of a quantum state which, classically, the Random Acess Memory (RAM) of a high-performance computer (HPC) like the LRZ flagship system SuperMUC-NG would not be able to handle.

For the development of quantum software, Euro-Q-Exa offers adapters to the most used software tools like Qiskit and PennyLane as part of the Munich Quantum Software Stack (MQSS), which adds further tools and interfaces for developing applications, including for hybrid high-performance computing that combines or accelerates supercomputing with quantum computing.