Volume 95 Issue 43 | pp. 27-31
Issue Date: October 30, 2017

Cover Story

Chemistry is quantum computing’s killer app

Quantum computers could help chemists better understand and develop catalysts, photovoltaics, and more
Department: Science & Technology
Keywords: Computational chemistry, quantum computers, quantum chemistry, nitrogenase, catalysts, photovoltaics, superconductors
IBM scientists Hanhee Paik and Sarah Sheldon examine the hardware inside an open refrigeration tank used to cool quantum computing hardware at the IBM Q Lab in Yorktown Heights, N.Y.
Credit: Connie Zhou/IBM
Scientists examine hardware for IBM's quantum computer.
IBM scientists Hanhee Paik and Sarah Sheldon examine the hardware inside an open refrigeration tank used to cool quantum computing hardware at the IBM Q Lab in Yorktown Heights, N.Y.
Credit: Connie Zhou/IBM
In brief

A flurry of developments in quantum computing hardware and software has occurred this year, suggesting useful quantum computers are within reach. The first killer app of quantum computers could involve big chemistry problems. In fact, researchers have dreamed about using quantum computers to simulate chemical systems since the 1980s. Quantum chemists think these machines could help researchers develop new catalysts, high-temperature superconductors, and more.

In an IBM lab in Yorktown Heights, N.Y., a circuit board sits at the bottom of a large tub cooled to temperatures colder than those of outer space. Shiny, solderlike zigzags of niobium cover the board, connecting 16 unusual computing elements. These elements, called quantum bits, or qubits, may form the basis of future computers that will solve problems by taking advantage of quantum mechanics.

So far, quantum computers like IBM’s 16-qubit system can solve only simple problems, all of them easy enough to be handled by traditional computers. This spring, IBM set a quantum-computing record by modeling beryllium hydride—which contains a mere three atoms (Nature 2017, DOI: 10.1038/nature23879). Earlier prototypes modeled hydrogen and factored the number 15.

As quantum computers become more sophisticated, they will allow chemists to solve problems that are intractable on today’s desktops and even supercomputers. In fact, physicists expect that chemistry simulations will be quantum computing’s killer app.

For example, quantum computers could aid development of catalysts for clean energy and renewable chemical manufacturing, enable deeper understanding of the enzymes that underlie photosynthesis and the nitrogen cycle, power the discovery of high-temperature superconductors and new materials for solar cells, and much more.

These computers could provide new insights into such chemical systems by doing something today’s computers cannot: Perform exact simulations of molecules with complex electron behavior. Today’s simulation methods, such as density functional theory, work well for many problems, especially in organic chemistry. But they fall short when it comes to inorganic systems, blurring important details about their electronic structure through approximations needed to simplify calculations so that traditional processors can handle them.

The simulations chemists currently rely on are like when you don’t have your glasses on, says Alán Aspuru-Guzik, a quantum chemist at Harvard University. “You’re touching the walls and slowly advancing” by feeling your way along. Doing chemical simulations on a quantum computer, he says, would be like putting on a new pair of glasses and seeing the world of chemistry anew, in sharp focus.

The question is not whether quantum computers will be useful, but when, say experts in the field. Although plenty of difficult problems wait to be solved, these researchers have been emboldened by recent successes.


To theorists, “Chemistry is just quantum mechanics and electrons moving around,” says Ryan Babbush, who is developing chemistry software for quantum computers at Google. Using the laws of quantum mechanics, researchers should be able to solve every chemistry problem. Once you know all the possible transitional structures a molecule might assume during each phase of a chemical reaction and the energies of all those structures, you can predict exactly what will happen.

That’s how pioneering theoretical physicist Paul Dirac saw it. The underlying physical laws necessary for a theoretical understanding of “the whole of chemistry” are known, he wrote in 1929. Unfortunately, he lamented, “the difficulty is only that the exact application of these laws leads to equations much too complicated to be soluble.”

In 2017, even with powerful supercomputers and simulation methods, this is still the case. Today’s quantum chemistry modeling algorithms can provide usually good enough, albeit inexact, predictions.

“The more electrons there are in a system, the harder it is to describe on a classical computer,” says Markus Reiher, a quantum chemist at the Swiss Federal Institute of Technology (ETH), Zurich. “You hit a wall that is not surmountable.”

To make these programs run on today’s computers, quantum chemists leave out some details about the behavior of some electrons—especially when they have strong interactions with each other. For many applications, good enough is fine. When simulating organic molecules, for example, the nitty-gritty quantum details are not of great practical importance. But for other systems, the details of the quantum dance of electrons are the very essence of the chemistry. The primary example of this is metals. “The thing about metals is you have a lot of electrons in a small space,” Babbush says. “Some crazy chemistry happens with these things. It’s hard to simulate and hard to wrap your head around.”

Crazy behavior that defies current modeling methods also occurs in other systems. Quantum chemistry models hit a wall when it comes to simulating high-temperature superconductors; the active sites of enzymes containing metal complexes; and radicals, with their unpaired electrons. Chemists are also limited in their ability to delve into the precise behavior of photovoltaic materials and to predict how an inorganic catalyst will behave.

To properly model these chemical systems, a computer must calculate the positions and energy levels of electrons. These properties are described by mathematical functions called orbitals. And that’s why quantum mechanics is key.

“If you have 125 orbitals and you want to store all possible configurations, then you need more memory in your classical computer than there are atoms in the universe,” says Matthias Troyer, who develops algorithms and applications for quantum computers at Microsoft Research in Zurich. But a quantum computer could model such a system with just 250 qubits.

The difference comes down to how each type of computer represents the states of electrons. Conventional computers use ones and zeros to encode an abstract description of all the possible states of an electron. Meanwhile, the qubits in quantum computers can represent these states as they exist in the molecular systems—as superpositions.

What is a superposition? Before researchers observe or measure a molecule, its electrons are not in any particular state. They are in a collection, or superposition, of all possible places and energies available to them. It’s only when you observe the molecule that they are in a particular location. Quantum computers can represent this phenomenon naturally. “The superposition states in the molecule are represented as actual superpositions in the quantum computer,” says Peter Love, a physicist who works on quantum computation at Tufts University.

Love recommends not thinking too hard about the nature of a superposition unless you are fluent in the math of theoretical physics, which is the only language in which to understand quantum mechanics. “There is no English sentence you can write that will explain superposition,” Love says. “It’s just outside our experience of the world.”

To perform calculations with these superpositions, quantum computers require different hardware from what is used in conventional computers. The digital ones and zeros in a conventional computer usually get encoded in transistors and memory cells.

In a quantum computer, qubits can take many forms. “Anything that can be put in a superposition can be a qubit,” says Danna Freedman, an inorganic chemist at Northwestern University. Some computers use arrays of individual photons or ions, for example. IBM, Google, and others use superconducting sandwich structures called Josephson junctions. To read data from and write data to these qubits, current quantum computers rely on microwaves, magnetic pulses, or laser pulses, depending on the type of qubit.

Pedal to the metal

Physicists have been talking about quantum computing’s potential knack for chemistry for at least 35 years. In a 1982 paper, California Institute of Technology physicist and Nobel laureate Richard Feynman proposed doing quantum simulations on a quantum computer.

About three years ago, researchers at Microsoft and ETH Zurich decided the field had progressed far enough to put these ideas to the test. They embarked on a very meta quest: a simulation of a simulation. They wanted to find out what it would take for a quantum computer to solve a chemistry problem that is intractable with a traditional computer. “There were just too many claims about what quantum computers can do,” Reiher says. He wanted to know how many qubits the computer would need and whether it would truly solve a real, important problem in a reasonable amount of time.

Top targets

Quantum computers could model chemical systems that are unsolvable for conventional computers. Here are some of the systems that quantum chemists would love to get quan- tum computers working on.

The group chose the problem of nitrogenase. Bacteria use this enzyme to make ammonia from atmospheric nitrogen in ambient conditions. The mechanism by which the enzyme achieves this feat is unknown. At nitrogenase’s heart is an iron-molybdenum cofactor called FeMoco that traditional computers can’t model. “We picked nitrogen fixation because it’s a hard problem where the classical approaches reach their limits,” Reiher says.

It’s also an important problem. Humans use metal catalysts and the Haber-Bosch process to produce ammonia fertilizer from atmospheric nitrogen at high temperatures and pressures. The process therefore uses a lot of energy and produces an abundance of greenhouse gases. Because of this, the carbon footprint of a loaf of bread—from growing and harvesting the wheat to baking the dough—is about 590 g, 40% of which is carbon dioxide emitted during nitrogen fertilizer production. If chemists knew the details of the nitrogenase mechanism—and more about how transition metals behave in general—they could make better, more efficient catalysts for producing fertilizer, as well as renewable fuels and many other industrially important chemicals.

The ETH Zurich team published its results this summer (Proc. Natl. Acad. Sci. USA 2017, DOI: 10.1073/pnas.1619152114). The researchers predicted that quantum computers with 100 qubits working perfectly could solve the FeMoco mechanism within days or months. (The model assumed multiple computers were working in parallel to crunch the numbers several times to ensure accuracy.) “This for us is the first commercially relevant demonstration that quantum computing can have a big impact,” Troyer says.

He cautions, though, that today’s quantum computers are far from being sophisticated enough to simulate FeMoco. Although it would be a snap to build 100 physical qubits like the ones in the IBM system, getting them to work together would be a challenge.

Quantum computers are divas. To get good performance out of a quantum computer, physicists need to make sure the machines’ qubits can maintain a certain quantum state for long enough to complete a computational task—in the parlance of the field, the qubits need a long coherence time. If qubits don’t have long coherence times and are unstable, they produce errors during calculations.

Qubit stability depends on the quality of the materials and methods used to make the qubits. And coherence is extremely sensitive to outside disturbances such as stray electromagnetic pulses. At the first hint of noise, the qubits “decohere,” or fall out of step, making computation impossible. The more qubits that engineers add to a computer, the harder it is to keep them all coherent for long periods.

The ETH Zurich simulation assumes all the qubits stayed calm, collected, and coherent, or logical. But no one expects to build 100 perfect qubits anytime soon. Instead, quantum physicists plan to use groups of physical qubits that act collectively like a single logical qubit. Special software will enable this. As qubit stability and software quality increase, fewer physical qubits will be needed to make one logical qubit.

Given the state of today’s technology, Troyer and Reiher believe hundreds of thousands to a million physical qubits working together would be required to make 100 logical qubits and solve nitrogenase.

What to do about qubits

Companies working on quantum computing hardware and software are split on how to deal with the divalike behavior of qubits.

Some are taking the long view. Microsoft is sponsoring basic research into a new kind of high-quality physical qubit. It cannot be made yet because its performance hinges on the existence of a fundamental particle, the Majorana, which has not been experimentally validated. These so-called topological qubits will stay calm and collected, remaining coherent in the face of disturbances.

Because of quantum effects mediated by this particle, which Ettore Majorana predicted in 1937, topological qubits will be “braided” together. This braiding involves a phenomenon called entanglement, in which the individual quantum states of each qubit will greatly influence those of other qubits.

This hardware helps cool Google’s quantum computer to temperatures colder than that of outer space.
Credit: Google, Erik Lucero
A photograph showing hardware from Google’s quantum computer.
This hardware helps cool Google’s quantum computer to temperatures colder than that of outer space.
Credit: Google, Erik Lucero

Researchers at the University of California, Santa Barbara, and Eindhoven and Delft Universities of Technology—some of them sponsored by Microsoft—are hot on the Majorana’s trail. This summer, the team, led by Erik Bakkers, reported the fabrication of a nanostructure that will allow them to test for braiding (Nature 2017, DOI: 10.1038/nature23468). The researchers built indium phosphide nanowires in a hashtag shape and expect to find braiding at the crossings. If they do, the nanohashtags could be used as topological qubits in the future.

“Topological qubits will blow superconductors out of the water, but first [researchers] need to show they have a single one,” Harvard’s Aspuru-Guzik says. For now, consider the topological qubit to be in a superposition of “very useful” and “completely impractical.”

Other researchers are trying to work with what they have—taking what computer scientists call a heuristic approach. IBM and Google continue to work with superconducting qubits because they are proven to work and are easy to build using existing fabrication equipment. Both companies use qubits based on Josephson junctions. At IBM, researchers stack a film of aluminum oxide between two layers of aluminum, which is a superconductor at extremely low temperatures. The qubits are cooled to 15 millikelvin and are controlled with microwave pulses.

IBM’s record-breaking simulation of beryllium hydride was done on a 7-qubit computer. The company has now built a 16-qubit system. Jerry Chow, head of IBM’s experimental quantum computing research, says the company’s current goal is not to keep pushing down the rows of the periodic table to calculate more complex molecules. Instead, it wants to learn how to make these systems work better.

“We’re dealing with the problem of noise,” Chow says. “A lot of our research comes down to understanding all the loss mechanisms and improving the materials so they are less sensitive.”

At Intel, engineers are taking a couple of approaches to building quantum computers. This month, they announced they had built a 17-qubit superconducting system and had sent it to the Netherlands for testing by collaborators. They’re also working on silicon-based qubits controlled by magnetic fields rather than microwave pulses. “We’re hedging our bets at this early stage,” says James Clarke, the company’s director of quantum hardware. He hopes the company’s expertise in fabricating high-end silicon transistors will translate to making quality qubits.

And although developing better qubits has been the province mainly of physicists, some chemists are delving into the field. Northwestern’s Freedman, for example, is using coordination chemistry to build qubits—a method she hopes will prove to be commercially viable while enabling greater stability and longer coherence times.

No matter what qubit type prevails, quantum computers will need good software. The IBM team was able to move into the second row of the periodic table with their noisy, small quantum computer because they developed clever algorithms to work with its limitations. “Right now the algorithms are improving faster than the hardware,” Tufts’s Love says.

Into the unknown

At Google, researchers’ immediate goal is to build a quantum computer so large that it cannot be simulated on a classical computer, a milestone called quantum supremacy.

Physicists believe a 49-qubit computer should be able to achieve quantum supremacy. At that point, quantum information scientists will be stepping into uncharted territory. By definition, once researchers achieve quantum supremacy, the computers cannot be simulated, so it’s difficult to make promises about what they will do.

Guided by mathematics and the principles of quantum mechanics, researchers have already been developing algorithms that will help quantum computers show off. Some who are working on chemistry problems believe it will be possible to solve useful, currently intractable, chemistry problems with 50 to 100 qubits.

Ken Brown, a computational chemist at Georgia Institute of Technology, expects that early systems of this size could simulate metals. Babbush believes a 70- to 80-qubit computer will be able to more accurately determine the spectra of small single molecules.

But quantum computers will enable researchers to do more than simulate larger, more complex systems. Brown says they will help chemists approach problems in different ways. Today, theorists can use quantum models to calculate the properties of a particular catalyst. Quantum computers will provide more accurate predictions. Theorists, Brown says, will be able to tell their experimental colleagues, “Make this, I can guarantee it will be a good catalyst.”

At that point, the question will be whether chemists will trust the results. Brown says a good early trust exercise would be to test the quantum computer on a problem with a known answer—a molecule whose energy is known from experiments but miscalculated by existing methods.

Whenever quantum computers start performing useful chemical calculations, only a few large computers will be available, Love says. Users will access quantum computers at companies like IBM and Google over the cloud or perhaps through supercomputing centers like those run by the U.S. Department of Energy.

IBM has already made its 16-qubit system available on the cloud through a project called the Quantum Experience. Chow says over 56,000 users have performed 1.5 million experiments—most of them on an earlier 5-qubit system—and that the project has yielded 30 research papers. And earlier this month, Google released an open-source software package called OpenFermion to help scientists translate existing quantum chemistry software into algorithms compatible with quantum hardware.

So when will quantum computers be useful for chemists? Depends on who you ask.

“I think that before 2035 we will be beating classical computers at chemistry,” Aspuru-Guzik says. He is so confident in the field’s progress that he has founded a start-up company called Zapata Computing to develop software for chemistry applications on quantum computers.

Freedman is more tempered in her expectations. “It’s really tricky to find the right problem for 50 to 100 qubits,” she says. “I think this will take 30 years.”

Troyer points out that 10 years ago, people in the field would say, “If we’re lucky, our grandchildren will see a quantum computer in their lifetimes.” In those 10 years, there has been remarkable progress in the development of hardware and software for quantum computers, he says. “Things have changed, and we’ve become much more optimistic. I want to do these calculations in my lifetime.”

Katherine Bourzac is a freelance science writer based in San Francisco.

CORRECTION: This story was updated on Nov. 6, 2017, to correct the type of quantum computing system used by 56,000 users via the cloud to perform the lion’s share of 1.5 million experiments. It was a 5-qubit system, not a 7-qubit system.

Chemical & Engineering News
ISSN 0009-2347
Copyright © American Chemical Society
Paul C Li (October 31, 2017 8:06 AM)
Dear C&EN Editors in Chief: Also Ms Katherine Bourzac,
The chemical bondings in the figurative nitrogenase shown in your publication is impressive if it were in existence under a thunderstorm. My botanical professor deMovre who told us that the biggest nitrogen fixer is "thunderstorm" who wakes up all
and stimulates the needed catalysts for plants and animals nutrients with bacteria ready for next day's photosynthetic greens etc.This was a lecture back in late 1959 at Taiwan Provincial Taichung Agricultural College .
I hope quantum computing can do more than nature simulating the thunderstorm in the hardware for such a short period of time.
Right now I am using three mathematical equations to calculate the bond stretching mode and the ring breathing mode of certain simple molecules in the mid-IR region using only bond length and the reduced mass of two to three atoms only. The v(nu bar) in wave numbers obtained by this simple technique satisfy the observed frequencies so well which promotes me to do it with your nitrogenase if I am allowed to, however, I need the density and molecular weight of that nitrogenase when available.
Best regards, submitted with highest honest and sincerity in my heart from a life long spectroscopic endeavor in better chemistry for better life, for life itself is chemistry.
jonathan ward (October 31, 2017 1:07 PM)
Fantastic Article - the best I've seen on quantum computing applications so far.
» Reply
Carol Bessel (November 1, 2017 7:06 PM)
Agree - fantastic article! Very timely given NSF's 10 Big Ideas!!
Leave A Comment