density functional theory

advantage.

So on a different note, actually, I mean, as we were discussing earlier about the AI, right? This artificial slash super intelligence that really comes back to the where these kind of sophisticated systems can help to discover the better materials, like be it quantum computing, be it DFT calculations, whatever so on.

So what's your view on that? How this super-intrusion is going to impact the best performing materials, particularly from the battery perspective, in your opinion?

Yeah, I think that's a great question.

You need good data, right? And where do you get the data from? So typically you get the data either from computational models like density functional theory.

But density functional theory is a very inefficient way to create data because it's very slow.

It takes a long time to run DFT calculations.

So DFT is not a smart way to create a lot of data.

machine learning abbreviated as ML interatomic potentials, the atomic cluster expansion abbreviated as ACE and its graph extension, GRACE the message passing network, MACE neural equivariant interatomic potentials abbreviated as NEQUIP deep potential molecular dynamics abbreviated as DEE PMD and others.

learn the energy landscape from density functional theory, abbreviated as DFT data, and reach near-DFT accuracy at the core.

Foundation models trained on the Materials Project and similar databases cover most of the periodic table out of the box, and a potential for silica, or a magnesium-aluminum-silicon-oxygen system, is now a published commodity.

For slags, this is a change of era.

In the bulk, screening makes this adequate, and they reproduce melt structure, diffusion, and elastic constants well.

At a crack or a polar surface, the unscreened field is missing.

Parker and colleagues showed that short-range ML potentials can generate configurations that look reasonable, but whose electrostatic potential varies so strongly that DFT would predict spurious metallization, a failure the potential cannot see.

Long-range terms are being added.

The math must predict the reality.

And to do that, they turn to a mathematical framework called time-dependent density functional theory, or TDDFT.

Yes.

Which sounds pretty intimidating.

Exactly.

That is where the time-dependent extension comes in.

TDDFT introduces the variable of time, allowing physicists to mathematically simulate how that 3D electron cloud sloshes, shifts, and responds to an external perturbation.

In this case, the tunneling electrons from the microscope tip?

I'm just gonna click through this because the shortcut is to compute it.

And, um, uh, density functional theory is a standard way to, uh, from first principles, not perfectly, but rather computationally efficience- efficiently, you can get the work function, you get formation energy.

And so with, uh, uh, her, um, uh, diving in with Jens Norskov in, uh, chemical engineering in a group at SLAC, uh, uh, was able to gear up on this and adapt some code and end up, uh, uh, using DFT and, and there's some, you know, feedback loop of optimizing the pseudopotentials [laughs] and all.

There's some front work that's needed.

This may be a good call back to our, actually our second episode ever, Heather Kulik had said, um, uh, she famously said, "There is no AlphaFold for materials," not just computationally, but the ground truth just doesn't exist.

Like, we don't really know what material crystal structures look like, and DFT is generally our best route towards it.

Yeah, I think so.

I mean, maybe two things I would add is DFT by definition doesn't have to be an approximation.

The Kohn-Hohenberg theorem shows that-

It doesn't

But I think as you're pointing out, the exchange correlation functional, we don't yet have access to.

And then for the, uh, charge density based models, we don't know what the kinetic energy functional is.