Making films of atomic-scale processes as they happen makes huge demands on any imaging system. One approach combines the advantages of pulsed laser harmonics and computerized image reconstruction.
To make a movie of molecular rearrangements, the criteria are strict. Enough of the illuminating beam should be scattered off the imaged object during each frame to form an image (at least one particle for each pixel), so the beam must be very intense and have a high scattering probability. This scattering should also be mostly elastic, and not transfer too much damaging energy to the sample. But most crucially, the imaging camera should have a frame speed of a few femtoseconds, coupled with atomic-scale spatial resolution.
These dual demands rule out most conventional imaging techniques. Optical lasers, for example, can offer the right sort of speed (a single period of laser light lasts about 2.5 fs), but they fall down on spatial resolution. (Spatial resolution is generally limited to about the wavelength of the probe, and the wavelength of optical light lies in the region of 400–700 nanometres.) Conversely, electrons have a small enough wavelength, but lack the requisite speed owing to the added complication of their charge interactions. Meanwhile, X-rays are limited by the aberrations and fabrication difficulties of the 'zone-plate' lenses that focus them.
Writing in Physical Review Letters, Sandberg et al.1 report taking a valuable step towards circumventing these problems. They combine recent breakthroughs in lensless imaging — which avoids the problems associated with lens aberrations by using a computer for image reconstruction — with advances in laser-driven X-ray generation to overcome the problem of spatial resolution, while preserving the laser's innate speed.
This advance is just the latest act in a fascinating story of the replacement of lenses by computers in imaging technology. The origins lie in the realms of signal processing, X-ray crystallography and electron microscopy2, and the breakthrough for X-rays came in 1999, with the first non-holographic reconstruction by numerical means of an image made by scattering X-rays from a non-periodic sample3. The current state-of-the-art4 fast, lensless imaging technique uses radiation produced by a free-electron laser at a synchrotron facility to make, in a single shot, images with a temporal resolution of 25 fs and a spatial resolution of 90 nm.
The secret behind all these techniques is an iterative phase-retrieval algorithm2. Iterative phase retrieval is one answer (various forms of holography and X-ray crystallography use other approaches) to the notorious 'phase problem' — that all detectors record only the intensity of the radiation that impinges on them, throwing away the phase information. Under suitable experimental conditions, however, this phase information is encoded in the intensity, and may be recovered if the intensity is sampled correctly. The algorithms iterate between the image and the scattering pattern (which are related by a mathematical operation known as a Fourier transform), while imposing known information, such as the approximate boundary of the object, on each. The great strength of such an algorithm is that it can be implemented for any type of imaging particle of any wavelength. Each particle interacts differently with a sample, and so can potentially provide new information about it. On the downside, such algorithms introduce constraints on the sample geometry, and coherence and aberrations in the illuminating wavefield become important.
Sandberg et al.1 generate very 'soft' X-rays (actually, extreme ultraviolet radiation) with a wavelength of about 30 nm by scattering intense pulses of infrared laser light of a much longer wavelength (800 nm) on gas atoms (Fig. 1). These X-rays scatter from the object, and are combined into an image with 214-nm resolution using a phase-retrieval algorithm. The imaging technique exploits high-frequency harmonics produced when laser light of energy hw (w is the laser frequency and h is Planck's constant) passes through a nonlinear medium. An atomic electron in the medium absorbs n laser photons before spitting out a single high-energy photon of n times the energy (and a similarly increased frequency nw), but the same properties of phase coherence and pulse duration as the driving laser.
Classically, we can think of the atomic electron being initially ejected by the laser pulse, before being returned to the atom during the second half of the laser cycle when the electric field reverses direction. The resulting acceleration produces radiation (bremsstrahlung) at the high-harmonic frequency. Importantly, this beam of radiation is directed forwards, and its phase coherence and conversion efficiency are greatly enhanced if generated inside a waveguide. High harmonics extending into soft X-ray frequencies were first observed5 in 1988, and used 5 years ago for holographic imaging with a resolution of about 10 µm (ref. 6).
Might a high-harmonic technique such as that of Sandberg's group one day provide competition for the large synchrotron particle accelerators currently used for molecular crystallography and the like? Synchrotrons provide tunable radiation with wavelengths from tens of nanometres to less than a tenth of a nanometre by collecting the bremsstrahlung from high-energy electrons accelerated over an optimized path. The wavelengths at which the high-harmonic technique is viable are continuing to fall (a collisional X-ray laser of wavelength 13 nm seeded by high-harmonic radiation is on the cards), pulsing rates are increasing and pulse duration is decreasing. A similar scheme using laser standing waves to undulate electron beams produces tunable, directed 35-kiloelectronvolt bremsstrahlung X-rays (which have wavelengths of a few hundredths of a nanometre). This would be useful for protein crystallography, but the apparatus occupies a room rather than a table. Then there is wakefield acceleration, in which laser pulses running through a plasma are used to accelerate electrons to gigaelectronvolt energies over a few centimetres.
This cornucopia of techniques is starting to produce viable competitors of synchrotrons in the effort to obtain higher-resolution, faster images. But the competition between the alternative techniques is intense. Sandberg and colleagues' method1, although promising, has some way to go. Its spatial resolution is still not sharp enough to see atoms, and the images required one and a half hours' exposure time with continuous 25-fs pulsing, owing to losses in the optics. One way around this problem for inorganic samples, in which repetitive processes such as electronic excitation and atomic motion can be triggered by another synchronized laser, is 'stroboscopic' imaging, which builds up pictures at different instants during the repeated cycle. In this case, fewer scattered particles are needed in each pulse, because many noisy images can be added together. A moving picture can thus be constructed by varying the delay between images. With biological samples, the limiting factor for this technique is the radiation damage caused by high exposure.
Despite these outstanding problems, lensless imaging in biology and materials science using electrons, neutrons and X-rays spans a wide and increasing array of techniques and capabilities. It is still early days — atoms were first seen7 in the field-ion microscope in 1951, and soon after with electron microscopy, whereas atomic-resolution lensless images of a single carbon nanotube were first reconstructed8 in 2003. One viable proposal for a non-damaging, sub-femtosecond, atomic-scale imaging technique based on self-diffraction of high-harmonic electrons from laser-aligned gas molecules already exists9. That is indeed a goal worth striving for.