Megahertz X-ray microscopy at X-ray Free-Electron Laser and Synchrotron sources

We demonstrate X-ray phase contrast microscopy performed at the European X-ray Free-Electron Laser sampled at 1.128 MHz rate. We have applied this method to image stochastic processes induced by an optical laser incident on water-filled capillaries with micrometer scale spatial resolution. The generated high speed water jet, cavitation formation and annihilation in water and glass, as well as glass explosions are observed. The comparison between XFEL and previous synchrotron MHz microscopy shows the superior contrast and spatial resolution at the XFEL over the synchrotron. This work opens up new possibilities for the characterization of dynamic stochastic systems on nanosecond to microsecond time scales at megahertz rate with object velocities up to few kilometers per second using X-ray Free-Electron Laser sources.

effective pixel size of the FT-CMOS camera Shimadzu HPV-X2. The X-ray pulse energy measured with a gas detection monitor at the end of the SASE 1 tunnel was about 750 µJ. The sample position in the interaction region of the X-ray and visible light lasers were locked optically using two high-resolution optical microscopes. The timing alignment was performed using temporal position of the scattered radiation of the X-ray and visible laser pulses from the scattering material placed at the sample position measured using a fast silicon diode connected to oscilloscope (LeCroy WaveRunner 8404M). The time alignment of the visible light pump laser with X-ray pump laser was done by shifting the synchronized 10Hz train signal, which indicates the start of each pulse train by a delay generator (Stanford Research DG645). As a sample we used a glass capillary with internal diameter of 300 µm and wall thickness of 25 µm. The capillary was filled with purified water and supplied via a remotely controlled syringe pump.
For the experiment at the ESRF synchrotron we used the 16bunch filling mode [6] providing a bunch separation of 176 ns and the MHz camera was recording every third frame with an inter-frame time of 530 ns. The harmonics of the undulator with central photon energy 32 keV were conditioned by a set of ten Beryllium compound refractive lenses with radius of curvature 0.5 mm and aperture of 1.3 mm focusing in vertical plane to enhance the flux density at the detector. The fast camera was coupled to the microscope with 4× magnification, 0.2 numerical aperture and 250 µm thick LYSO:Ce scintillator providing an effective pixel size of 8 µm and a FOV of 2.0 mm × 3.0 mm. The sample-to-detector distance was 1 m. As a sample we used a glass capillary with internal diameter of 500 µm and wall thickness of 270 µm. Purified water was mixed with Nile blue dye to enhance absorption of laser power in the water. A pulsed Nd:YAG laser (Minilite II, Continuum) was used to generate the water jet by power absorption in the water. The laser wavelength was 532 nm with a pulse duration 3-5 ns, and the attenuated laser beam energy was 2.03 mJ with a focal spot size of 300 µm. The 10 Hz laser flash lamp and Q-switch operation were synchronized with the RF system using the BCDU8 unit and a delay generator (DG 645, Stanford Instruments). Experimental arrangement for EuXFEL is shown in Fig. 2 and for ESRF setup in Fig. 3.

IMAGE FLOW ANALYSIS
In this section we describe the procedure for the optical flow analysis shown in the main paper. To remove normalization artifacts and high frequency noise for the EuXFEL data we performed an adaptive high pass filtering by subtracting a low-pass filtered (Gaussian convolution with the standard deviation σ = 5 pixels) image from its original version. This procedure significantly suppressed spatio-temporal image flickering. Using such image processing, we visualized the velocities of breaking glass reaching 35 m/s using flow analysis based on variational optical flow methods [7]. Such methods combine a data term that imposes constraints on image brightness and a smoothness term which regulates the properties of the flow field. A global model containing both constraints is then minimized to find a solution. Relying on spatial image gradients, the optical flow cannot be uniquely computed within homogeneous image regions. Thus the "edge enhancement" contrast which highlights sharp object boundaries is beneficial for the optical flow under the assumption of small displacements. The computed velocities provide quantitative information about complex kinematics of the burst process.

IMAGE QUALITY COMPARISON
To compare quantitatively the image quality between the two experiments, first the signal-to-noise ratio (SNR) was evaluated. Figures 4 and 5 show the results and frames used for the comparison. For each sequence frames with the sample were used and normalized by the frames without the sample. The SNR for each frame in the sequence was calculated by division of the mean values of normalized area containing the signal of the sample C c by the standard deviation of the normalized area B c without the sample. As can be clearly seen from the distribution of the mean values ( Fig.4d and Fig.5d) the pulse-to-pulse intensity fluctuations are strong in case of EuXFEL, which is due to the origin of the photons generation at XFELs using selfamplified spontaneous emission (SASE). Pointing fluctuations are pronounced at EuXFEL, as at all SASE XFELs. Modal composition of the spectrum and pulse-to-pulse energy fluctuations result in pulse-to-pulse wave front variations [8]. All such instabilities contribute to the measured image intensity distribution, which in turn varies from pulse to pulse.
This makes it difficult to perform a standard normalization of image frames by the background frame and it leads to the fluctuation of the standard deviation. Such fluctuations at the synchrotron are much smaller as it is clearly seen from the mean values in Fig. 5. Another source of standard deviation increase in case of the EuXFEL data is the strong high-frequency noise caused by the beam delivery KB optics manifested as vertical and horizontal stripes. This can be in future improved by better illumination, for example by using direct beam illumination. The longer-period oscillations in mean values in Fig. 5 are present due to the not exact synchronization of the camera to the master frequency of the X-ray pulses as the camera clock has a different frequency. Such oscillations are present as well in the EuXFEL data but with much smaller period resulting in only little or no illumination of certain frames in the sequence.
By comparing the SNR of the two data sets, we can conclude that the EuXFEL data have a clearly higher SNR by almost a factor of two. The results from the EuXFEL also indicate that despite using a much less efficient imaging system and a higher magnification, a stronger signal is detected. This performance can unlock fast dynamics at even higher spatial resolution. However, due to the very different experimental conditions (different photon energy, magnification, scintillator thickness, optics efficiency) such comparison is not entirely correct. A more objective comparison can be achieved by looking at the power spectrum. For this purpose we selected the frames with the highest SNR for both the ESRF and the EuXFEL data and calculated the power spectrum from normalized areas A c from both data sets. Using a Fourier transform, we calculated the power of the corresponding frames as      P(k x , k y ) = |FFT(A c (x, y))| 2 (1) and then integrated the spectrum in polar coordinates. The resulting graphs are shown in Fig. 6. The EuXFEL power spectrum is superior over the ESRF power spectrum as it is higher by one to almost two orders of magnitude over the entire range of spatial frequencies.