Neural networks can study and model complex conditions at the macro scale, such as weather patterns or the movement of heavenly bodies, but researchers at Princeton University have also been applying this tool to ever smaller objects, yielding potentially valuable contributions for chemistry, physics and quantum computing.
In their project, “Seeking to Greatly Accelerate the Achievement of Quantum Many-body Optimal Control Through the Use of Artificial Neural Networks,”
Herschel Rabitz, the Charles Phelps Smyth '16 *17 Professor of Chemistry, and a team of researchers have harnessed neural networks to design, model, understand, and control quantum dynamics phenomena between atoms within molecules.
The researchers are using advanced data science techniques to accelerate this process by stimulating and tailoring the activity of many more particles (i.e., the many-body aspect) than previously possible. Results from this research, for example, can help in the development and control of novel materials that have useful industrial attributes.
“The calculations are difficult to scale satisfactorily on a digital computer by standard means,” Rabitz said about modeling and controlling quantum dynamics. “Our number one goal is to break that barrier and show that we can get much more favorable scaling by using tools from data science.”
This project was one of the first nine funded last year by Princeton University’s Schmidt DataX Fund, which aims to spread and deepen the use of artificial intelligence and machine learning across campus with the aim of accelerating discovery. In February last year, the University announced the new fund, which was made possible through a major gift from Schmidt Futures.
“We are excited about these DataX projects because they use data science to tackle problems that were once difficult to study or model or they help scale up experiments that are arduous to replicate at mass quantities in a laboratory setting,” said Peter Ramadge, director of the Center for Statistics and Machine Learning, which oversees parts of the DataX Fund. “The projects are also interdisciplinary in approach and cover a wide breadth of fields, showing the broad applicability of data science and machine learning.”
For an initial foray into the project, Rabitz and his research team first undertook a broad study of utilizing machine learning in time-dependent many-body quantum optimally controlled dynamics.
“We wanted to address the question of how complex is the actual problem,” said Rabitz about modeling and controlling quantum dynamics phenomena.
It’s nominally considered as highly complex because atoms exist in several different quantum states and modeling this becomes exponentially complicated when researchers try to observe or control the many atoms in a molecule, Rabitz said. A molecule, for example, can have 100 atoms and a piece of material can be composed of many molecules – leading to “a mind-boggling number” of atoms that need to be studied.
“Modeling these dynamics in classical computation is unreachable because of the quantum character of each particle,” said Rabitz.
The researchers’ results have shown that their time dependent many-body artificial neural-network method can model quantum dynamics, which lays the groundwork for performing “computationally demanding calculations of complex atomic and polyatomic control problems,” Rabitz said.
“Our initial results show that machine learning can really crack the problem of modeling and controlling quantum dynamics, a nominally exponentially difficult problem by conventional computational means,” Rabitz said.
This neural network could also potentially have implications in the field of quantum computing and quantum information sciences. The researchers’ algorithm is able to mathematically compress large amounts of data for study. Rabitz said they talked about their research with Google, whose researchers have been delving into quantum computing. A few Google scientists have already expressed interest in their study because quantum computing will require compression, as demonstrated in Rabitz’s research.
“In the future, quantum computing will be married to a machine learning algorithm and we can come up with even better results,” he said.
Rabitz said he is excited about this project because it gives scientists a deep look into how and why atoms form together to make molecules and materials.
“We have a laser laboratory and we do control experiments on highly complex molecules,” Rabitz said. “We can control large molecules in the laboratory but have no understanding how it works. We don’t know what’s really taking place. For very fundamental reasons, this opportunity to bring in machine learning into many-body quantum dynamics is lifting up the edge of the tent to peek under and see why it is working in the laboratory, and I am confident that this understanding will lead to new technologies.”