Hacker News Reader

Comment by carabiner | Hacker News Reader

More DFT? Haven't we concluded that has little predictive power?

More people saying it sounds feasibly like it could be a superconductor is great, but agreed - https://nitter.net/MichaelSFuhrer/status/1686267690770739200 and similar threads by actual experts in the field have outlined that DFT studies ignore a lot of variables that need to line up for superconductivity, in harsh contrast to techbros on Twitter claiming they're proof of the next thousand year golden era.

Yeah the techbros have been the worst. Coffee guy went from saying the first DFT paper should deserve a Nobel to, "it's not a strong signal of anything" in 12 hours. WTF. All the actual scientists have been rock steady at, "it is probably not going to replicate," especially CMTC.

Techbro speculation on physics should be considered harmful.

CMTC comes of as a whiny little POS tbh. They have the exact opposite of the attitude I would have expected from a scientist. In their first thread regarding this topic they were almost flat out belittling the og authors. WTF.

Because CMTC has high theory standards and the authors, to put it charitably, didn’t

This is the one from the all infamous University of Maryland correct? lol.

They may have been throwing a bit of shade around lately but they are by and large reputable scientists barring whatever SDS’s past MA announcement involvement was. It’s more than just him there.

I do not (almost everyone else too) associate "reputable scientist" with University of Maryland. So they should show a bit of humility.

Show me where on the doll they touched you.

Hey no need to get sensitive about it. I am seriously saying University of Maryland (chuckle) is just not enough for bragging rights. Do not worry may he when you grow up you will end up in an ivy.

I don’t have anything to do with them and I went to Caltech.

*MS not MA

>All the actual scientists have been rock steady at, "it is probably not going to replicate," especially CMTC.

Sinéad Griffin isn't an actual scientist?

https://en.wikipedia.org/wiki/Sin%C3%A9ad_Griffin

Plenty of real scientists are excited by the DFT results while cautioning they are merely supporting evidence and not conclusive evidence.

If DFT agrees with reality, it's fine

If it produces unphysical nonsense, you can always say "You should have considered xyz!"

Computational chemistry inherently cannot consider everything we understand about chemistry or physics or consider beyond certain scales, so its accuracy and precision are limited. You have to constrain the space properly, which makes people a little dismissive.

> tech bro speculation on physics should be considered harmful

why?

Unfounded hype is bad and they're usually low-key trying to gain attention for their failing startups.

Some of the tech bros are actually into technology though.

Thanks for linking someone who knows what he's talking about, very painful to see these DFT papers overhyped as some kind of proof... It proves almost nothing

What does DFT prove beyond the pedantic "it proves what it proves"? Or rather, does it not heavily lean towards some reality?

I’d be interested in seeing a source on that. My understanding was that it was a powerful and popular computational tool. The theory is approximate but approximate theories can still be quite useful if the approximation is a good one.

It's basically an open secret, no one really believes DFT has much predictive power in strongly correlated systems (where relevant electrons cannot be treated as independent). It's not useless, but DFT calculations require many arbitrary choices and are hardly "first principles" in a meaningful way.

The LBNL DFT paper isn't claiming to have modeled the strong correlations. Rather, they observe a specific feature of the band structure (flat bands crossing the Fermi level formed from d-orbitals) that have also appeared in DFT calculations of other high-temp superconductors and are considered a signature of those materials. In the other high-Tc materials, the calculations are consistent with experiment, and so the unexpected appearance of these bands in the reported crystal structure is seen as a signal of "hey, if this turns out to be real, we can explain it with current physics." That's something that previous fraudulent claims of room temperature ambient pressure superconductors (University of Rochester group) couldn't do.

The other interesting conclusion of the LBNL paper is that the low-energy physics of the electronic structure can be effectively modeled by a two-level effective Hamiltonian. This is a common pathway to bridging the gap between DFT calculations and theoretical understanding of strong correlations.

None of the serious scientists are saying "hey, look, DFT predicted it so it's true." Rather, it doesn't rule it out, and it suggests that it could be understood with our current physics. That's significant in itself.

From my understanding of DFT that makes sense but it seems like in that case you are using the wrong tool. Approximate theories are good when the approximations are good. If your phenomena is strongly correlated, throwing out those correlations is bad.

I have no idea if the use of DFT here is appropriate or not.

DFT is generally good for well characterized phenomenon. Superconductivity is a bit of an edge case, really dependent on quantum effects that you need ab-initio quantum mechanics simulations to predict.

Reply on news.ycombinator.com