> His goal is nothing less than a digital twin of the real worm, accurate down to the molecule.
I am nitpicking this: The state of computational chemistry is not at a level to support this. I'm optimistic we'll get there eventually; need to find novel approaches, and current ones are too imprecise, or too slow.
I think the connectome aspect is more interesting, as it may be feasible to get there 100% without a computational chem breakthrough.
What's the latest and the best so far? Are they using GPGPU? Is quantum computing there yet, or would it help? Heuristics and sampling?
GPGPU is definitely mainstream for large scale quantum and molecular simulation. Quantum computing might help speed up electronic structure calculations, but my impression is that it’s still in its infancy
To give a sense of the scale of this problem, the largest frontier simulations I’m aware of are around the trillion atom scale. (On tens of thousands of GPUs [0])
Based on a quick web search, a c elegans cell is between 3 microns and 30 microns in diameter, so if we assume we can count atoms using the density of water then an all-atom simulation of a single neuron would need between 5e11 to 5e14 atoms. c. Elegans has 302 neurons so simulating the full neural network will be 2-5 orders of magnitude larger than current frontier simulations. Honestly more doable than I thought it would be, though all-atom simulation of a full organism still seems quite out of reach
This is all with classical force fields. Doing this simulation at the electronic structure level is much much harder with our current modeling capabilities
0: https://www.mrs.org/meetings-events/annual-meetings/archive/...
There's also interesting custom-made machines for molecular simulations that don't rely on GPGPUs and are significantly faster, e.g.: - https://arxiv.org/abs/2405.07898 - https://www.psc.edu/resources/anton/
is there any reasoning that for besides highly reductionistic and repetitive systems like crystals, quantum computing can compute quantum properties of molecules?
it seems to me "the quantum computer you seek" is the molecule + the medium (especially the medium) itself
So first I think I might need to apologize for some jargon collision - my background is mostly material simulation, and when I say “quantum simulation” I mostly mean using classical algorithms to solve the quantum mechanical wave equation describing a material or molecule.
I don’t pretend to have any particularly deep insight into quantum algorithms for chemistry, but [0] is a really nice review. It seems like there are a lot of possibilities for simulating general molecular and materials systems on quantum computers. The holy grail would be solving the exact quantum mechanical wave equation in sub-exponential time and space complexity. I don’t know how feasible that is, but it seems like people are making progress using quantum algorithms to accelerate approximate quantum simulation [1].
Back to all-atom c. elegans: I think quantum computing is more about accurate and scalable electronic structure modeling, and simulating enormous systems like this will still require fitting classical (meaning electrons are implicit) force fields and running them at scale for the foreseeable future. A lot of this is space complexity - I’m not sure how a quantum computer could do atomic simulations with sublinear scaling of qubits in the number of atoms being simulated, and were in the very early days of scaling quantum computers up
thanks for the reference.
yes. you nailed my point-that you will have to fit classical or quasi-classical fields, which is liable to require scads of qbits just to get close. qbits are just not "designed" to do that sort of thing.
in any case we ~solved protein folding heuristically and not using fields so i shouldn't be too pessimistic that it's impossible that quantum compute will help eventually.
And I don't think we will get there with digital computers either.
So, with what?
If we get there, and it's not guaranteed that we do, it will be a different kind of machine.