My team (Tasha Pais, Shivam Syal, Thomas Breydo) and I are developing a moonshot company leveraging quantum machine learning to find a 10x more efficient catalyst to use in direct air carbon capture, ultimately creating a more scalable and cost-effective process.
Introduction: Climate Change
Climate change is an unimaginably large global problem. 20 million people lose their homes each year due to climate change induced natural disasters. Half of all deaths in children under the age of five are because of climate related hunger. Increasing temperatures are melting ice sheets and rising sea levels. Further, the extra carbon dioxide in the air is reducing the pH of oceans, harming coral reefs and aquatic life. If this trend continues, up to 50% of species around the world are projected to lose their suitable climate conditions by 2100, leading to significant biodiversity losses.
Climate change is a natural cycle. However, excessive greenhouse gases released by humans are accelerating this process due to its heat trapping nature. A prevalent greenhouse gas is carbon dioxide, which is released through both natural processes (such as respiration) as well as human activities, such as deforestation and burning fossil fuels. The issue is that humans have created an imbalance in the atmospheric CO2 levels — its concentration has increased by 47% since the Industrial Revolution, making it a significant contributor to climate change.
Direct Air Capture and Carbon Conversion
To offset carbon emissions, it is vital that carbon dioxide from the air is removed. Direct air capture (DAC) is a promising method of pulling in air from the atmosphere, separating the carbon dioxide, and storing it underground or in a commercially used material.
Since CO2 is a stable molecule (making it unreactive), the chemical reaction in DAC needs a catalyst for it to occur. This catalyst serves to speed up the reaction by “setting it up”.
The reaction above shows carbon dioxide and hydrogen being converted to methanol with the help of a ruthenium-based catalyst.
The issue is, metal oxide catalysts such as the ruthenium-based one above are incredibly energy intensive (requiring temperatures of over 300° C to operate). Not only is this inefficient reaction costly, but it also creates heat and carbon monoxide as a byproduct.
Finding more efficient catalysts requires lengthy trial-and-error in a lab. Additionally, classical computers are unable to accurately simulate these reactions because they are incapable of precisely calculating the Hamiltonian (energies) of many electrons. These types of problems can get very complex very quickly as computers must simultaneously calculate the interactions between electrons in multiple atoms. In fact, simulating such a molecular reaction on a classical computer can take more time than the age of the universe!
Luckily, quantum computers exist.
Quantum computers run off of “qubits,” which can be represented as a linear combination of 0 and 1 (called “superposition”), rather than a rigid 0 or 1 value used in classical computers.
Because of this superposition property, quantum computers are excellent at performing many calculations simultaneously. This means it can easily solve the Schrödinger Equation, which is a physics equation whose solution gives the total energy of a system (the Hamiltonian).
A quantum machine learning algorithm called the Variational Quantum Eigensolver (VQE) can assist in finding the lowest energy configuration of an atom. Fundamentally, it can calculate the eigenvalues of a large Hamiltonian matrix. Here’s how it works:
- Information about a particle’s electron orbitals are encoded into qubits. From this, a circuit architecture (called the ansatz) prepares an initial “trial” wave function representing the particle. This will eventually serve as a guess for the ground state energy.
- Then, the ground state energy of this wavefunction is calculated and sent to a classical optimizer as an “expectation value”.
- The classical optimizer will adjust certain parameters in the quantum circuit to get a lower expectation value. Repeat!
Eventually, this process will iterate until the expectation value converges with the ground state energy, which is the smallest eigenvalue in the Hamiltonian.
So how does this relate to finding an ideal, efficient ruthenium catalyst?
In the diagram above, the steps on the right reflect the VQE. A classical computer prepares a Hamiltonian matrix in an active space, which is a certain section of the reaction that is the most relevant. This simplification is used to reduce computational speeds. Then, the quantum computer solves this Hamiltonian and passes on the expectation value to the classical computer. Now that the classical computer understands the energy levels of the catalyst, it will perform a kinetic analysis of the reaction and gauge the rate of the reaction. Next, it can modify the catalyst to make it more efficient, and repeat the process with the new catalyst structure. After a number of iterations, an “ideal” catalyst is found.
Today’s quantum computers are unable to simulate more than a few atoms because larger particles require more qubits. As more qubits are added to a circuit, error also increases because it is harder for qubits to maintain their entangled, or connected state. This loss of entanglement is called decoherence. Environmental disturbances, such as temperature fluctuations or radiation, can also lead to error in the delicate circuit. Currently, the computer with the most amount of functioning qubits is IBM’s 65-qubit computer. However, IBM is projected to create a functioning computer with upwards of 1000 qubits by 2023. Additionally, various experts in the field have validated this idea, and claim that quantum chemistry is the most promising application of quantum computing.