L'apprentissage Automatique Révèle Des Composants Cachés Des Impulsions De Rayons

Machine learning reveals hidden components of X-ray pulses

Measuring the properties of ultrafast pulses from X-ray lasers is a major challenge for physicists. Research published in the scientific journal Express Opticals Describes a method that can help you with this task.

These pulses reveal how the atoms move at femtosecond intervals – the equivalent of a quadrillion second. Although the determination of the maximum strength – or amplitude – of a pulse is accurate, it is often unclear when it reaches its maximum degree, or “phase”.

In the new study, scientists from the US Stanford Linear Acceleration Center (SLAC) and Germany’s Electronic Synchrotron (DESY) used machine learning and trained neural networks to analyze the pulse and reveal these hidden subcomponents, which are divided into “real” and “phantom.”

The X-ray pulse (white line) consists of “real” and “imaginary” components (red and blue traces, respectively), which determine the quantum effects. The neural network analyzes the low-resolution measurements (black shadow) to reveal the high-resolution pulse and its hidden components. Credit: Stanford Linear Acceleration Center

From the low-resolution measurements, neural networks reveal minute details of each of the impulses, which are analyzed millions of times faster than previous methods, with up to three times greater accuracy.

Knowing the components of each X-ray pulse leads to better and more accurate data. According to the authors, this will expand the understanding of fields using ultrafast X-ray lasers, including basic research in chemistry, physics, materials science and quantum computing.

“The neural network approach used here could have broad applications in X-ray accelerator science, including learning the shape of proteins or electron beam properties,” the researchers said.

Characterizations of system dynamics are important applications of X-ray free electron laser (XFEL). Diagnosing the characteristics of each individual XFEL pulse can enable a new class of simpler and more accurate dynamic experiments.

According to the site physicalNeural network architecture based on “physical studied” models can be trained directly on unlabeled experimental data and is fast enough for real-time analysis on the next generation Megahertz XFEL.

The method also recovers the phase, enabling consistent control experiments with XFELs and shaping the complex motion of electrons in molecules and condensed matter systems.

Have you seen our new videos on Youtube? Subscribe to our channel!

Leave a Reply

Your email address will not be published. Required fields are marked *