In laser experiments, and especially in high harmonic generation, it is important to know the time variation of the electric field of the laser. For long laser pulses (consisting of several optical cycles), this can be approximately determined from the laser’s intensity profile, which is measurable. However, for lasers consisting of a few optical cycles, the same temporal intensity profile can correspond to different temporal electric fields, which significantly modifies the spectrum of the generated high harmonics. It is the CEP value that shows how the oscillation of an electric field relates to the intensity shape of the pulse that envelops it. In other words, how the maxima of the pulse and the electric field are shifted relative to each other. In our MIR laser, this shift is controlled by one device and measured by another. In most cases, these devices are expensive, i.e. finding a substituting technique can save us money.
“It has long been known that the spectra of the generated harmonics in solids and gases depend on the state of the CEP. At certain values of the CEP of single-cycle optical pulses, the harmonic spectrum resembles the teeth of a comb, but if we adjust the device, thus the laser CEP, we get a continuous spectrum,” said Zsolt Divéki, leader of the SYLOS GHHG Compact and GHHG Long Research Group at ELI ALPS.
The harmonic spectrum is sensitive to the CEP value, and the underlying mathematical relationship is rather complex. Therefore, our physicists decided to deploy artificial intelligence. If we know that a given spectrum has a given CEP value, based on the knowledge of many spectra and their CEP values, the machine learning algorithm can learn the relationship between them. If we feed a new spectrum to the algorithm, it will provide the corresponding CEP value.
The algorithm is cumbersome to train, and we need a tool to initially check the predicted CEP value. However, once the simulations provide realistic harmonic spectra, the CEP measurement tool is no longer needed. This is exactly the subject of the paper published in Optics Express. “This idea has been mooted by other researchers, but we were the first to prove experimentally that the idea does work. There are indications that we can use artificial intelligence to predict not only the CEP value, but also other parameters of the laser. Characteristics that affect the shape of the harmonic spectrum,” said Zsolt Divéki, indicating that the research will continue.
Harmonic spectra are easy to generate with the MIR laser: one has to focus the laser light into a crystal with a lens or mirror. However, care must be taken, because too strong focusing can destroy the crystal. In other words, subtle alignment is a must to preserve the setup components while making the experiment successful.
Zsolt Divéki has confirmed that thanks to their research, artificial intelligence can be applied in a new field. (Our colleagues have some more ideas up their sleeves where artificial intelligence can be deployed.) The publication of their result indicates that the topic raised in Szeged is of interest for other researchers too.
If adequately efficient simulations could realistically predict the harmonic spectrum, our physicists could hit the market with a palm-sized (even patent-protected) instrument. The device would contain a crystal, a mirror and a spectrometer, which could estimate other, important laser parameters besides CEP.