HIR3X

HIR3X - Helmholtz Joint Lab

HIR3X is the Helmholtz International Laboratory on Reliability, Repetition, Results at the most Advanced X-Ray Sources. For a personal overview of the project’s activities, hear from our two co-PIs on their experience:

“You optimize from head to toe” – HIR3X co-PI Professor Mei Bai explains the main areas that the project worked to improve

“Everything's really turned up to 11 in terms of trying to do these experiments” - HIR3X co-PI Professor Henry Chapman walks through the potential offered by FEL improvements

The revolutionary new capabilities of X-ray free-electron lasers (FEL) have launched a new field of ultrafast X-ray science. FELs can do this by generating X-ray pulses every femtosecond – that’s every one quadrillionth of a second. At their peak, these laser-like ‘flashes’ of X-ray radiation can also reach a brightness more than a billion times higher than any previous X-ray source.

These two characteristics allow X-ray pulses generated by FELs to measure samples at far higher and clearer resolutions – sometimes even smaller than a nanometre in length. This in turn has led to:

  • humanity’s first direct measurements of chemistry and catalysis at the atomic scale
  • creating ‘movies’ of magnetization dynamics at the nanoscale
  • observing the evolution of exotic quantum dynamics (such as squeezed phonons) in solids
  • generating and studying extreme states of matter as found in the cores of stars and planets, or atoms stripped of electrons from the inside giving new insights into atomic structure
  • better images of proteins free from the radiation damage that plagues normal methods of X-ray crystallography and cryo-electron microscopy

But all this scientific potential means that demand for the already limited time slots at FELs has risen. Meanwhile users must also find a way to handle the large volume of data generated by FEL scans of their test samples.

The HIR3X project was an initiative of three world leading FEL research centres; DESY and European XFLE in Germany and SLAC in the United States, coordinated as a Helmholtz International Laboratory.

The objective of this joint lab is to push the performance of advanced accelerator-based X-ray sources and their applications in specific research areas to their full capability and potential. By using new optimization and automation strategies, the project aimed to make these advanced experimental techniques faster, more efficient, and open to a larger number of non-expert users.

Each of the project’s four work packages proposed a novel and bold approach to improve reliability, for example:

  • applying machine learning to areas like operating the accelerator, generating the X-ray pulses, and detecting and analyzing the X-ray signals
  • deploying robotic control of the delivery of samples to avoid interruptions and downtime, and to address challenges in the transport of high-power X-ray beams to experiments

HIR3X researchers believe common solutions will enable standardization of experiments and protocols which will further foster collaboration in other areas and promote reliability and ease of use.

MeiBai_upscale.jpg
 

“You optimize from head to toe” – Professor Mei Bai

HIR3X project co-PI Prof. Mei Bai looks back at the factors leading up to the project

In the world of physics, something can be around for decades and still feel new.

The technology to create  X-ray free-electron lasers (XFELs) has been around for over two decades now, though it still feels new in the scientific community. It remains a very delicate tool; even as we welcome people from all scientific backgrounds to run experiments with our lasers, XFEL reliability remains a challenge to solve with broad diversity of user demands and requests.

At DESY and SLAC, the XFEL facilities try to accommodating as many users as possible on just one x-ray beam line, but each group has their own requirements. This means each time swapping out their apparatus and accordingly adjusting accelerators and x-ray optics for each experiment.

On paper anyone can do it instantly. But imagine that in reality every minor change, no matter how small, eats into the overhead time for the next user.

Now imagine this happening 1000 times a week; this eats up availability of precious beam time. We try to minimise this by grouping users according to their similar needs, though this can lead to too many compromises that rarely work out the best for anyone. This minimize overhead time. In addition, one needs to optimize everything for experiments that require extra precision and sensitivities.

Fortunately, we believe there is another way. You optimize from head to toe. Since 2020, both DESY in Hamburg and SLAC in California have worked to improve the reliability, repetition and results of these x-ray sources in the HIR3X project. These ‘three Rs’ aren’t just about the machine itself. It’s intertwined with creating high-quality data that can be reliably fed out to many end stations, as if they independently have their own x-ray source.

The project is also strengthening the many grass-roots links between SLAC and DESY researchers. Call it a friendship or collaboration; either way, this way of working can hopefully go on even longer in the future.

HIR3X thrived thanks in particular to the dynamic minds of the new generation of researchers. A highlight of this project was being able to support that potential for the future. That’s worth every single penny.

Our two institutions brought together their individual experience to drive the technology forward. For example, SLAC was one of the early adopters of applying machine learning technology to accelerators. In this project they worked with visiting DESY students and post-docs on their Xopt framework to bring machine learning-empowered beam optimisation.

For example one project group harnessed machine learning to optimise the performance the FEL light sources. In one result, they applied a quite advanced large language model to tune the accelerator online, and use feedback to shape the proceeding laser pulses. This process is called ‘reinforcement learning trained optimisation.’

Another aspect of the project focused on handling high-throughput data, which included developing two new methods for on-the-fly data selection (vetoing) performance. This can help in ‘bad image’ rejection in serial crystallography to give researchers a clearer view of their sample.

On the hardware side, we installed new hardware for fully automatic sample exchange for users, and developed prototype and motion controls for high-speed sample scanning. Speeding up and removing human error during this phase means experiments can handle more samples at a faster rate.

Finally, DESY and SLAC worked together to advance their respective high-performance X-ray devices and components. Open and constructive meetings provided the inspiration and advice needed to help projects such as the Optical DelayLine Project (XFEL) and the Pulse-length preserving double monochromator beamline (FLASH) realise their full potential.

As the HIR3X project wraps up, our researchers are already brainstorming new ideas for future collaboration in accelerator development, sample preparation & sample delivery, and big data handling. Even as the project ends, our momentum continues.

Henry_Chapman-3204.jpg

Q&A with Prof. Henry Chapman, HIR3X co-lead

Q: What challenges exist with the reliability of X-ray free electron lasers at the moment?

A: The accelerator that drives reactions and emits the X-rays are reliable, but the experiments can be challenging. It's been a big step from what people are used to with synchrotrons, where you have an X-ray beam that you can turn on and off to expose your sample over whatever time you may need.

With an FEL, these extremely powerful and short pulses usually destroys the sample during the experiment, essentially vaporising it. The idea is that researchers get structural information before the destruction has taken place.

We call this ‘diffraction before destruction.’ Although it can provide clear and precise measurements, each measurement is different because you need a constant supply of samples to replenish into the beam.

There's a lot of fluctuations that we have to worry about; the beam itself fluctuates because it's also born out of a random process, the samples can be different and be in different positions.

Each shot becomes a new experiment. There's all this randomness that we have to have to deal with. Then things have to go quickly especially at the European XFEL because these pulses come at megahertz rates.

Everything's really turned up to 11 in terms of trying to do these experiments. For example if we're doing crystallography on macromolecular crystals, we have to move them across at 100 meters per seconds. We do this in a micro jet of liquid with the sample in it. If that breaks down or gets clogged, then you're losing measurement time.

It's also making sure that FELs remain reliable so that users’ experiments don’t break down too often. We want them to have a good experience using this relatively new technology. Nobody wants to see their sample that they spent months preparing get wasted because the detection was not working, or was recorded in the wrong way.

What does that mean for researchers?

Ideally, these experiments should be pretty fast. We should be able to get all the information we want in just a few minutes. But all XFEL operators lose time trying to get everything in the experiment to align, so usually it takes days instead. So there's a lot of saving we can provide everyone through good engineering in the whole experimental setup and understanding the reasons why things might go wrong.

Part of the HIR3X project was bringing together expertise in those sorts of experiments, and thinking how to make the instruments that sit at the end of that beam line work much better.

For example, one of their main concerns has always been that they need to produce too many protein samples. We're trying to reduce the amount of sample that is required, and to make sure that they can go home with real structural information rather than a bunch of diffraction patterns that take a long time to process. I think this still a challenge, but our work in HIR3X means it’s less so than before.

For example, we’ve researched how to best put a sample on a substrate and then do rapid scans around it, so that you know that your sample is in the right position to be scanned. This means users in future are not going to waste any time before the scan; if there is a problem that causes the scan to stop, users can restart it without wasting any sample.

We’re trying to find combinations of these ideas that could work well. One idea is instead of moving these samples at 100 meters per second past the beam to match the wavelength of X-ray pulses, we could instead ‘sweep’ the beam across the samples instead, so that each pulse ‘sees’ the sample at a different angle. Alke Meents, who runs the work package on this idea is making a ‘beam sweeper’ - essentially a rotating mirror – which the X-rays reflect off.

Another example is tackling the wasted time tuning up the machine to the right pulse structure. A big work package in HIR3X was on machine learning to running the machines efficiently. We are taking advantage of these new digital capabilities to do many fancy things like changing wavelengths very quickly, so users can get chemical information as well as structural information out of their samples.

Finally, another big research area in the project is time-resolved crystallography, where users first excite their sample into some new state or reaction, and then hit it with the X-rays at a defined time delay. If you then repeat this at different time delays, you build up a ‘movie’ of the reaction or change of the protein. It adds more complexity to your experiment, but with a big payoff; understanding the exact process of your sample’s reaction.

The project relied on research happening at FELs continents apart. How do you feel the cooperation was like between DESY and SLAC?

We've known each other for a long time, since building our respective FELs has complemented our understanding of how they work. In California, SLAC developed the ‘hard’ x-ray Linac Coherent Light Source (LCLS), which demonstrated that the whole concept of free electron lasers could work. These ‘hard’ x-rays have a high energy and short wavelength, which makes them better to penetrate deeply into materials.

Here in Germany, DESY built the world’s first ‘soft’ x-ray FEL called FLASH. These x-rays have lower energy and longer wavelengths, which is better for probing the nano- and meso-scale structure of materials.

At  the time that FLASH came online, LCLS was being constructed. They were obviously very interested in FLASH to see what works and what doesn't. There was always that  connection.

We did experiments first at FLASH to understand what happens when the beam hits the sample, and how you're going to deal with all the information.

We started our imaging experiments first at the soft X-rays at FLASH. There was a German contingent who did experiments first at FLASH, and then at LCLS. So ideas have always been moving from one place to the other.

We all know each other very well, and it's a very close and cooperative community. Time is precious at all these facilities, and both sides have always found ourselves asking where it's best to do experiments. That means that teams always come together to find the nest solution. We've had a long experience of  working together and understanding what the issues are at different places. What I find really helps us both is that we don't hide the difficulties of our work from each other. We're happy to face them together.

What do you think will benefit most from HIR3X?

I think the non-expert FEL users who will benefit most will be the structural biologists. We’ve already made a lot of progress in developing better time-resolved scans of protein dynamics proteins.

How these scans are employed in screening experiments are pretty interesting, because you may have thousands of compounds in your experiment, but slightly different each time. These can all be tested nicely with automated measurement schemes that we have developed through HIR3X, for example in serial crystallography. Samples can even be mixed on the fly or automatically deposited onto substrate chips for testing. These chips can then be changed in the chamber with robots.

All this could really speed up screening experiments for drug discovery by orders of magnitude, and that’s not even mentioning the financial savings.

In actual fact, some of the techniques that we've developed are also being ported back to other synchrotrons to try and increase their throughput of measurements.  

Suggested research: