Hacker News Reader

Comment by smblau | Hacker News Reader

Wow - excited to see our work posted here! I’m the last author on this paper, happy to answer any questions :)

Interesting paper, thanks for sharing! Whenever I see physics-based inverse design using ML surrogates, I always ask, “why not optimize the problem directly” (and eg compute the gradient using an adjoint-variable method)? The paper implies that the forward simulation process isn’t differentiable, but is this true? Thanks!

Thanks for the question!

First off, the ML surrogate is many orders of magnitude faster than direct kMC simulation. Each global optimization involved up to 500 sequential predictions of emission intensity with different nanoparticle structures (and gradients - I'll come back to those). For the largest particles that we optimized, running kMC on only the final optimized structure (for validation) took three months! So running 500 sequential kMC simulations would take approximately 125 years - not ideal. In contrast, all 81 global optimizations took two days with our trained hetero-GNN.

Back to the gradients - running the kMC simulations involves generating four different randomly doped 3D nanoparticle structures according to the shell thicknesses and dopant concentrations for the structure of interest, and then running four kMC trajectories with different random seeds for each structure. The final predicted value for emission (based on the number of emitted photons within the energy range of interest) is averaged over all sixteen trajectories. kMC is fundamentally stochastic, and not differentiable. But even if you could make it so with the adjoint-variable method (which I am admittedly not super familiar with), or maybe by writing your kMC code in a language with foundational auto-diff capabilities, once you move from a description of layer thicknesses and dopant concentrations to an ensemble of 3D structures/trajectories, I still don't think that the gradient of emission with respect to layer thicknesses and dopant concentrations would be accessible.

Thoughts? Thanks!

Thanks for the reply!

> So running 500 sequential kMC simulations would take approximately 125 years - not ideal.

Ah I see. Ya it’s hard to get around this…

> is fundamentally stochastic, and not differentiable

This is actually a common “inverse design pattern” for a variety of applications, and luckily there are tricks to efficiently compute gradients here. In my domain (nanophotonics) we’re often simulating incoherent sources, which are similarly stochastic. But there are ways to reformulate the problem to drastically reduce the number of forward and adjoint solves you need during the design process [1].

That being said, given the cost of the forward problem (3 months per iteration!) this doesn’t help you much…

[1] https://doi.org/10.1007/s00158-022-03389-5

Thanks for the helpful citation! Looking forward to reading and learning more.

> UCNP heterostructures present a stringent test for any new DL model and representation.

Are UCNP heterostructures also the most promising area of application, or are you looking to apply the method in other areas now that you've validated it in the stringent case?

Very cool! Are you trying to synthesize the new nanoparticles that you found to see if they're actually as good as you predicted?

Yes! Emory Chan's group at the Molecular Foundry at LBL is trying to synthesize our champion particle. (Emory was the other co-corresponding author on the paper). As far as I understand, the small core size and sheer number of different layers has been challenging, but they're getting close! We're very excited to see if the simulated emission intensity validates in a real physical particle :)

Reply on news.ycombinator.com