X-ray lasers and computational nightmares
A duett by Janos Hajdu and Tomas Ekeberg
- Date and Time
Thursday, February 23rd, 2012 at 10:30
Polacksbacken, room 2245
X-ray free-electron lasers (FELs) are the most brilliant sources of X-rays to date, exceeding the peak brilliance of other X-ray sources by nearly 10 billion, and improving. In the duration of a single flash, the beam focused to a micron-sized spot has the same power density as all the sunlight hitting the Earth, focused to a millimetre square. Theory predicts that with an ultra-short and extremely bright coherent X-ray pulse, a single diffraction pattern may be recorded from a large macromolecule, a virus, or a cell before the sample explodes and turns into a plasma. The over-sampled diffraction pattern permits phase retrieval and hence structure determination. Today, 10 million exposures can be taken in 24 hours, corresponding to about 50 TB of data per day. This data rate is expected to increase thousand-fold by 2015. In the experiments, sample particles are intercepted randomly and in unknown orientations by the X-ray pulses. Many images are needed from new copies of structurally identical particles to derive a three-dimensional diffraction data set. This can be achieved if the orientation of the particle at the time of exposure can be determined for each of the patterns. This problem is however very hard both conceptually and computationally and can be complicated further if the patterns contain only a few photons per shot. We have demonstrated that it is possible to assemble a 3D data set and obtained a low resolution structure for a virus particle but as we try to reach higher resolution and include more data we quickly outgrow our current computational capabilities. The future of this field is dependent on an efficient implementation and the interplay between the hardware and software developments. It also depends on an understanding of the limits of these algorithms.