Ångstrom and attosecond are the fundamental spatiotemporal scales for electron dynamics in various materials. Although attosecond pulses with wavelengths comparable to the atomic scales are expected to be a key tool in advancing attosecond science, producing high-power hard X-ray attosecond pulses at ångstrom wavelengths remains a formidable challenge. Here, we report the generation of terawatt-scale attosecond hard X-ray pulses using a free-electron laser in a special operation mode. We achieved 9 keV single-spike X-ray pulses with a mean pulse energy of around 180 μJ, exceeding previous reports by more than an order of magnitude, and an estimated average pulse duration of 200 as at full-width at half-maximum. Exploiting the unique capability of the European XFEL, which can deliver ten pulse trains per second with each containing hundreds of pulses at megahertz repetition rates, this study demonstrates the generation of attosecond X-ray pulses at a 2.25 MHz repetition rate. These intense high-repetition-rate attosecond X-ray pulses present transformative prospects for structural and electronic damage-free X-ray measurements and attosecond time-resolved X-ray methodologies, heralding a new era in ultrafast X-ray science.

Particle-beam-driven plasma wakefield acceleration (PWFA) enables various novel high-gradient techniques for powering future compact light-source and high-energy physics applications. Here, a driving particle bunch excites a wakefield response in a plasma medium, which may rapidly accelerate a trailing witness beam. In this Letter, we present the measurement of ratios of acceleration of the witness bunch to deceleration of the driver bunch, the so-called transformer ratio, significantly exceeding the fundamental theoretical and thus far experimental limit of 2 in a PWFA. An electron bunch with ramped current profile was utilized to accelerate a witness bunch with a transformer ratio of $4.6_{-0.7}^{+2.2}$ in a plasma with length $\sim 10\mathrm{cm}$, also demonstrating stable transport of driver bunches with lengths on the order of the plasma wavelength.

High-resolution X-ray spectroscopy in the sub-nanosecond to femtosecond time range requires ultrashort X-ray pulses and a spectral X-ray flux considerably larger than that presently available. X-ray free-electron laser (XFEL) radiation from hard X-ray self-seeding (HXRSS) setups has been demonstrated in the past and offers the necessary peak flux properties. So far, these systems could not provide high repetition rates enabling a high average flux. We report the results for a cascaded HXRSS system installed at the European XFEL, currently the only operating high-repetition-rate hard X-ray XFEL facility worldwide. A high repetition rate, combined with HXRSS, allows the generation of millijoule-level pulses in the photon energy range of 6–14 keV with a bandwidth of around 1 eV (corresponding to about 1 mJ eV^–1 peak spectral density) at the rate of ten trains per second, each train including hundreds of pulses arriving at a megahertz repetition rate. At 2.25 MHz repetition rate and photon energies in the 6–7 keV range, we observed and characterized the heat-load effects on the HXRSS crystals, substantially altering the spectra of subsequent X-ray pulses. We demonstrated that our cascaded self-seeding scheme reduces this detrimental effect to below the detection level. This opens up exciting new possibilities in a wide range of scientific fields employing ultrafast X-ray spectroscopy, scattering and imaging techniques.

Temporally modulated electron beams have a wide array of applications ranging from the generation of coherently enhanced electromagnetic radiation to the resonant excitation of electromagnetic wakefields in advanced-accelerator concepts. Likewise producing low-energy ultrashort microbunches could be useful for ultrafast electron diffraction and new accelerator-based light-source concepts. In this Letter we propose and experimentally demonstrate a passive microbunching technique capable of forming a picosecond bunch train at $\sim 6\mathrm{MeV}$. The method relies on the excitation of electromagnetic wakefields as the beam propagates through a dielectric-lined waveguide. Owing to the nonultrarelativistic nature of the beam, the induced energy modulation eventually converts into a density modulation as the beam travels in a following free-space drift. The modulated beam is further accelerated to $\sim 20\mathrm{MeV}$ while preserving the imparted density modulation.

Modeling of large-scale research facilities is extremely challenging due to complex physical processes and engineering problems. Here, we adopt a data-driven approach to model the longitudinal phase-space-diagnostic beamline at the photoinector of the European XFEL with an encoder-decoder neural-network model. A deep convolutional neural network (decoder) is used to build images, measured on the screen, from a small feature map generated by another neural network (encoder). We demonstrate that the model, trained only with experimental data, can make high-fidelity predictions of megapixel images for the longitudinal phase-space measurement without any prior knowledge of photoinjectors or electron beams. The prediction significantly outperforms existing methods. We also show the scalability and interpretability of the model by sharing the same decoder with more than one encoder, used for different setups of the photoinjector, and propose a pragmatic way to model a facility with various diagnostics and working points. This opens the door to a way of accurately modeling a photoinjector using neural networks and experimental data. The approach can possibly be extended to the whole accelerator and even other types of scientific facility.