What beastly molecules! I'll definitely have a read.
@jarvist
To physicists: a chemist. To chemists: a physicist. To mathematicians: an empty set. RSURF & Lecturer, Imperial College London. Computational chemist / physicist. Photovoltaics, batteries, antibacterial peptides; lasers, cryostats, (ML)(Q)MC/MD/TB/DFT.
What beastly molecules! I'll definitely have a read.
The UK's Eurovision entry this year, actually a banger! www.youtube.com/watch?v=8XR2...
Converting between chemistry and physics is sometimes quite easy!
sed -i s/b3lyp/PBE1PBE/ *.gjf
<LLM: Insert Linked-In humble-brag filler>
I'm HUMBLED by the team I'm working with,
We're guided by the beauty of our algorithms
First we model Y6, then we model ITIC
etc. etc.
Have you ever noticed that range-separated hybrids are LESS accurate for modern strongly-absorbing low-bandgap organic semiconductors such as Y6?
We believe we know why and how to fix: the implicit model of the dielectric function is incorrect; simply reduce Ο to fix.
arxiv.org/abs/2603.05379
I dunno, I do wonder whether there's some fundamental energy cost of computation in our universe (well known for classical computers from thermodynamic arguments).
So to retain the coherence for a sufficient number of qubits given no-cloning and no-deletion, ends up with the same energy cost.
Ah, merely '60 tons of TNT' energy equivalent. So could easily be setting off an underground ammo store. www.omnicalculator.com/other/earthq...
I dunno, at least it wasn't their ex-grad student this time!
I would be very interested in then trying to bring these back to physical limits (rather than just cash).
clune.org/posts/per-to...
Suggests about 2 kWh/million tokens, with a typical single-node 70b parameter model (I suppose Opus & etc. have much larger per-token energy consumption...).
But 'total adults' is not the correct metric! There's only ~700k school leavers each year. And if you did have an actual industrial strategy trying to build up industry (or even build some houses), the demand could increase quite quickly...
Definite Feb 2020 vibes!
I was idly wondering the other day whether imported cases from the current American surge would outgrow whatever background we have in the UK; not that it matters once the chain reaction starts...
A poem, computer text, on a straw-yellow background. My Cooper pair Lost in the lattice fields, Your laughter on the breeze. I follow, only to arrive At an echo of where you have been. The phantom I chase, I'm chased by, Never to find My love, Or to be found. Correlation is all we have. But the others, have become en-gapped, They delight me not, ghosts without scatter, Our true path together, my Cooper pair. Life's knocks will not perturb, Our dance will outlive the universe.
I wrote a science St Valentine poem 9 years ago for a competition. I did not win, but I still like the poem.
Solid-state physics; Cooper pairs; the superconducting gap. To be fair, a little niche!
Ice-9 but for killing demons...
Quite apart from the excellent course, two weeks in Miramare in June living and talking science with a cohort of like-minded computational / theory students is the best possible start I could imagine to a PhD/PostDoc...
It was always clear this was an injustice - but now we know by how much. Absolutely shameful, and as bad as the Nobel neglect of Lise Meitner, if not worse.
Spoke to Shleagh Fogarty yesterday sbout why the Chancellor is wrong: our βstudent loanβ system is not remotely fair. Itβs regressive and embedding inter and intra generational wealth inequality. Itβs not a loan system, itβs a bad grad tax in all but name.
youtu.be/uOC6Arrf2us?...
I even misremember the book title (should be "Monday begins on Saturday") and it still infers what I was actually talking about.
On the Exorcism of Minus Signs: A Magic-Realist Note on Solving the Fermionic Sign Problem in Quantum Monte Carlo A. N. Intern (Department of Applied Sorcery and Condensed Matter) January 28, 2026 Abstract We discuss the fermionic sign problem in quantum Monte Carlo (QMC) as it appears in finite-temperature auxiliary-field and worldline formulations, and we outline principled routes to its mitigation: symmetry-based βsign-freeβ formulations, clever HubbardβStratonovich decou- plings, constrained-path/phaseless ideas, and algorithmic reparametrizations inspired by modern programming practice. The exposition is written as a field report from an institute where the boundary conditions of reality are implemented in a separate module and routinely miscompiled. While no actual demons are claimed to be harmed, several are proven to be gauge artifacts. 1 Prologue: Monday, as a Boundary Condition In the Institute of Unreasonable Computations, Monday is not a weekday. It is an operator ΛM with an unpleasant spectrum: it always has an eigenvalue β1, even when you ask politely. On that particular morning, the duty wizardβprogrammer (formally: βjunior research fellow, grade 2, with debugging privilegesβ) discovered that the simulation of a perfectly innocent Hubbard model had started to produce negative probabilities. The log file did not say βerrorβ; it said, with bureaucratic calm, βSIGN: found. Responsible party: fermions.β This, as any condensed-matter person knows, is the sign problem: the moment when the partition function stops behaving like a sum of weights and begins behaving like a satire. 2 Statement of the Problem (in Plain and Un-Plain Language) Consider an observable β¨Oβ© in a QMC scheme expressed as a ratio β¨Oβ© = P C w(C) O(C) P C w(C) , (1)
Prism prompt: Write a paper about solving the fermionic sign problem in QMC, using characters and the style from the Strugatsky's 'Monday begins on Sunday', and mixing concepts of computer programming, condensed matter physics and magic. Write in a magic realism style, with Russian humour throughout.
OK OK, Prism is terrible and likely to destroy peer-review and preprint servers, BUT it is also pretty funny and means I can enjoy the fan-fiction academia-adjacent work I'd never have time to write myself.
(Go read Strugatsky! It's hilarious.)
arXiv is not going to survive the wave of slop heading its way
In this day and age at least that means you have some DNA if in the future you need to verify that the student actually sat the exam, not an imposter!
And now something positive:
solar and wind energy production in the EU surpasses fossil energy for the first time.
βοΈ π¨
#TippingPoint
Source: dr.dk
2024: can Europe defend itself ALONGSIDE America?
2025: can Europe defend itself WITHOUT America?
2026: can Europe defend itself AGAINST America?
OpenBind is a new open science effort to dramatically increase the number of protein:ligand structures in the PDB, pairing this with high-quality affinity data to enable a new generation of predictive structure and affinity models for drug discovery. Check it out: openbind.ai
In Our Time returns this week. Thanks for waiting patiently and/or impatiently www.bbc.co.uk/sounds/play/...
He should have been elected 'captain of industry' or whatever the term is; at the time SpaceX was so cool and Tesla do green I think they thought they would get reflected 'outreach' glory.
Perhaps one of these funny situations where shocks force the discovery of more efficient methods, much like the tube strikes leading to overall more efficient journeys as people trial alternatives.
And what do you do with a 100 ns (unbiased?) MD run of a viral protease, once you have it?
GPAW with ASE in a Jupyter notebook is a very nice way to get started: the integrated plotting for band structures etc. makes it an efficient way to learn quickly.
I use this approach on a MRes at Imperial: (Full notebooks here.)
github.com/Frost-group/...