Should We Relax Our Server Contempt If Used to Understand Batteries and Solar Cells?

[NNadir]

The paper to which I'll refer in this post is this one: Mailoa, J.P., Li, X. & Zhang, S. 3T-VASP: fast ab-initio electrochemical reactor via multi-scale gradient energy minimization. Nat Commun 15, 10140 (2024).

The paper is open for public reading for free.

It's become fashionable to bash computer servers since they use energy, and people complain about servers on blogs relying on, um, servers. It is also fashionable to say that solar cells and batteries are "green."

None of these claims actually appeal to me. I don't think that solar cells and batteries have proved to be sustainable, are proving to be sustainable, and I don't expect they will ever be sustainable. I don't consider either to be "green."

I also believe that computational power is overall a good thing, which may be a function of the famous Upton Sinclair remark, “It is difficult to get a man to understand something, when his salary depends on his not understanding it.” The scientific work in which I'm employed is very dependent on computational power, and instruments that generate the data that computers crunch. I fully understand that like any human activity, computational power can be and is abused to talk about celebrity marriages, rehabs and recidivism, blah, blah, watching movies and texting one another in lieu of talking or meeting, but no, I do not have a knee jerk belief that my use of a computer is the only use acceptable.

Other folks may find other things more important, like, um, blogging for instance.

Anyway.

VASP is a commercial software package that is computationally intensive, and is used to perform DFT, Density Functional Theory Calculations.

So how about raising the point, by arguing that computational power is great for understanding all those magical objects that are widely believed to have something to do with environmental protection in the common imagination, even though they have actually done nothing to protect the environment at all? Would that make everyone happy?

Maybe this paper will help:

Ab-initio methods such as density functional theory (DFT) are instrumental in the study of physical sciences for many material systems across different scientific fields1. Unfortunately, the cost of these ab-initio methods scales quickly with the system size; for example, DFT generally scales as N³ where N is a measure of the system size (such as the number of electrons or basis functions). While the computation cost of individual ab-initio steps is generally still accessible using hardware available today, the large number of DFT calls necessary to observe the desired event in a material simulation may limit its scalability. For example, accelerated high-temperature DFT-based ab-initio molecular dynamics (AIMD) is often used to simulate chemical reactions within Li-ion battery electrolytes and may require 10 ps simulation (10 or 20k DFT steps depending on the hydrogen isotope mass used) to observe 0–3 electrochemical reaction or redox events2,3. This may take several weeks per AIMD trajectory with the best hardware available today. Significant efforts have been made to alleviate this problem. Some of the most promising efforts to date involve the usage of reactive classical force field for enabling pre-determined chemical reactions4,5,6, neural network force field (NNFF) to replace the computation cost of each DFT step with a significantly lower-cost neural network potential energy equivalent7,8,9,10,11,12,13, and DFT-based metadynamics to reduce the number of DFT steps (typically into just several ps despite lower temperatures) by bypassing energy barriers and enabling atoms to transition between states more quickly than it would otherwise have under accelerated high-temperature AIMD14,15,16. These state-of-the-art methods are crucial for studying chemical reactions involving temperature dependencies. Other promising reaction exploration approaches are temperature-independent, and they work by modifying the steady-state structure optimization method to escape the local energy minimum. For example, the artificial force-induced reaction (AFIR) approach works by adding multi-parameter empirical artificial forces between all-atom pairs17. The approach is further extended in the implementation of Chemoton 2.0 software, where the combination of brute force conformer sampling and empirical artificial force added on chosen reactive sites between candidate reactant molecule pairs are used to accelerate a reaction18...

...In this work, we propose that in some limited circumstances such as the exploration of elementary electrochemical reaction products in complex systems without considering temperature dependencies, it is sufficient to simply perform multi-scale ab-initio energy minimization. We can enable the system to quickly reach lower-energy structure configurations typically inaccessible without tens of thousands of AIMD steps, mostly within just ~100 DFT steps (several hours). We utilize a previously developed tiered tensor transform (3T) method to transform material structure geometry in a multi-scale manner, enabling the system to explore some lower-energy structure configurations quickly28...

And on what is this "faster" DFT tested out?

Cool stuff that is supposed to save the world although there is no evidence, none, that this cool and much worshipped stuff is saving the world. On the contrary, the world is collapsing in flames, but, well it's the thought that counts.

Cool stuff from the paper:

...We first show a simpler example using a single complex floppy organic cation binding on FAPbI₃ vacancy surface defect to highlight the way multi-scale structure energy minimization is performed with the aid of PyTorch autograd functionality. We then show more complex examples such as large-scale lithium-ion battery liquid electrolyte electrochemical reduction and oxidation reaction trajectory generation, reproducing many experimentally known lithium-ion battery byproducts in just ~100 DFT steps as opposed to tens of thousands of steps commonly utilized in literature...

It behooves me to point out what "FAPbI₃" is. FAPbI₃ is formic acid lead triiodide, which falls into a class of compounds known as perovskites, which are the latest fad in making super duper highly efficient solar cells, if, and only if, some problems with the lack of stability of these cells can be overcome. Personally, having thought about the problem of distributing lead in the environment, I think this is a dubious quest, but no environmental drawback of any type, the need for vast mining, the toxicology of trichlorohydrosilane, cadmium or indium, vast copper mines and the requirement of backing up solar with dangerous fossil fuels can dissuade people from claiming solar energy is "green;" that's certainly been my experience anyway.

As for batteries, well, there's the problem of cobalt (an ethical problem as much as an environmental problem) fluorophosphates, nickel mining etc, and well degradation of organic compounds in the system.

A figure from the text:




The caption:

a Structure update procedure difference between typical structure energy relaxation, molecular dynamics-based structure exploration, and multi-scale 3T structure energy relaxation. b 3T multi-scale structure transformation projects the energy minimization problem into higher-dimension 3T parameter hyperspace, where it is possible to explore farther local energy minima landscape difficult to access by standard relaxation method or longer to reach using molecular dynamics. c 3T multi-scale structure transformation mode sequence used in this work. d Some of the 3T reduction and oxidation electrochemical reaction byproducts generated from the bulk electrolyte liquid mixture in this work.

Other figures are available at the full document.

It's all about solar cells and batteries, so in this case, the server on which this work was done is "green."

Right?

Well I guess it's more important (slightly) than talking about Kylie Jenner, whoever she is.

In any case, I personally believe that it is possible to produce reliable continuous electricity cleanly and without fossil fuels. The realization of this possibility involves to my mind, the fundamental equation of neutronics, the Bateman equation(s), a coupled series of differential equations generally solved by (gasp) computer power :



Have a nice day tomorrow.

[quaint]

1. No is my uneducated answer to your thread question.

I try to educate myself before condemning technological advances, but I only understand one or two percent.
However, my common sense still says, no.

[NNadir]

2. I would caution anyone from opposing a technology based on a lack of understanding it.

Common sense is, in my view, vastly overrated. The word "common" means ordinary. To evaluate extraordinary technology requires extraordinary insight, work, and education, whether formal or autodidactically obtained.

RFK is a commoner. Nobel Laureate Katalin Kariko, whose work ultimately made RNA vaccines available is uncommon.

I oppose wind and solar energy because I do understand them, not because I don't understand them.

That is the difference between me and the orange pedophile in the White House. His position comes out of ignorance; mine out of detailed investigation.

He, similarly, supports fossil fuels out of ignorance; I oppose them because I have deep understanding of the consequences of their use.

Similarly, as a onetime antinuke I support nuclear energy because I looked into the details on a very deep level that ultimately bordered on, I confess, obsessive.

I have yet to meet an antinuke here or nearly anywhere who knows anything at all about nuclear energy other than they hate it.

DFT described in the OP is a very powerful tool. It can be used either in a positive or negative way, for good or for bad. How it is used is an ethical question, not a technical question. It represents a subset of the use of computational power, the use of which also is defined of ethics.

That is what is missing in our times, senses of ethics.