Hacker News new | past | comments | ask | show | jobs | submit login
Accelerated discovery of CO2 electrocatalysts using active machine learning (nature.com)
15 points by bookofjoe 18 days ago | hide | past | web | favorite | 3 comments



If an efficient way to use CO2 as an industrial feedstock can be found, then it opens the door for gigawatt scale gas turbine farms located at natural gas fields. (The advantage of location at the field is to minimize pipeline cost and methane leak detection cost.) Electricity generation with gas turbines is about as efficient as it gets https://en.wikipedia.org/wiki/Gas_turbine#Advances_in_techno... and natural gas is cheap. Since the byproducts are almost only water and CO2 (some amount, I know not how much, NO2) isolating the CO2 has to be about as cheap as it gets. Remaining economic problem, find a value add for pure CO2.


And not a word about machine learning. Sounds like density functional theory is the numerical model here. I don’t have access to the full article, but if you tell me it’s numerical optimization on a DFT objective, then that makes sense, but it’s not ML. If you tell me it’s machine learning, then I’ll be confused and ask you how you measured all those millions/billions of quantum states to get the training data. And, like, why are you using a ML model when the physical model (DFT) is known? Are you saying that you used ML to speed up simulation? Is this an ML paper or a materials science paper or maybe something about global warming? I’m confused.


Hard to imagine how any active system can ever be cheaper than an inactive system - ie. just leave the coal and oil in the ground!

Our current CO2 levels rely on events in Earth's history that cannot be recreated for millions of years. Its mindblowing that we still freely exhaust CO2 even for such mundane activities as private transport.

https://en.wikipedia.org/wiki/Azolla_event




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: