Lightnews — Scholar-powered news

The Matter Lab @thematterlab.bsky.social · 5h

Even better, our Hessian readout head is simple and can be added to any of your favorite equivariant MLIPs ✅

The brilliant team:
@andreasburger.bsky.social, Luca Thiede, Nikolaj Rønne, @variniabernales.bsky.social, Nandita Vijaykumar, @tvegge.bsky.social, Arghya Bhowmik, @aspuru.bsky.social
[4/4]

The Matter Lab @thematterlab.bsky.social · 5h

Compared to MLIPs with autograd, we achieve:

➡️ 2x lower error
➡️ 70x faster inference, more efficient parallelism, and better scaling with system size ⚡
➡️ Consistently higher success rates on all downstream tasks
[3/4]

1 1

The Matter Lab @thematterlab.bsky.social · 5h

In our new preprint, "Shoot from the HIP: Hessian Interatomic potentials without derivatives", we show that we can directly predict symmetry-preserving Hessians using an equivariant Hessian redout head.

📃 arxiv.org/abs/2509.21624
🖥️ github.com/BurgerAndrea...
[2/4]

Shoot from the HIP: Hessian Interatomic Potentials without derivatives

Fundamental tasks in computational chemistry, from transition state search to vibrational analysis, rely on molecular Hessians, which are the second derivatives of the potential energy. Yet, Hessians ...

arxiv.org

1 1

The Matter Lab @thematterlab.bsky.social · 5h

What do transition state search, geometry relaxation, zero point energy corrections, and extrema classification have in common? They all require Hessians!

The problem is, accurate Hessians are really expensive, even with MLIPs. We say, just shoot 'em from the HIP! 🤠
[1/4]

1 5

The Matter Lab @thematterlab.bsky.social · 8d

Kudos to authors: Danial Motlagh, Robert A. Lang, Paarth Jain, @ja-camga.bsky.social, William Maxwell, Tao Zeng, @aspuru.bsky.social, and Juan Miguel Arrazola.
[8/8]

3

The Matter Lab @thematterlab.bsky.social · 8d

- Resource estimates for anthracene-based chromophores highlight the path toward quantum-enabled discovery of new singlet fission materials.
[7/8]

1 2

The Matter Lab @thematterlab.bsky.social · 8d

- Key observables such as electronic state populations can be extracted simply by measuring the electronic register, bypassing the input–output bottleneck common in many quantum algorithms.
[6/8]

1 2

The Matter Lab @thematterlab.bsky.social · 8d

🔬 Highlights:

- Proof-of-principle simulations show that realistic vibronic models of exciton transport and charge transfer can be tackled with feasible quantum resources.
[5/8]

1 2

The Matter Lab @thematterlab.bsky.social · 8d

- Materials discovery pipeline — demonstrated how our algorithm integrates into computational workflows for singlet fission solar cells, a next-generation photovoltaic technology that can double exciton yield and potentially surpass the Shockley–Queisser limit.
[4/8]

1 2

The Matter Lab @thematterlab.bsky.social · 8d

- Algorithmic optimizations — introduced caching and block-diagonalization techniques that dramatically reduce the cost of exponentiating vibronic operators, making simulations more efficient.
[3/8]

1 2

The Matter Lab @thematterlab.bsky.social · 8d

🚀What we achieved:

- General vibronic dynamics — developed the first trotterization scheme for vibronic Hamiltonians with more than two electronic states, enabling scalable simulation of ultrafast non-adiabatic processes.
[2/8]

1 2

The Matter Lab @thematterlab.bsky.social · 8d

Excited to share our new article: “Quantum algorithm for vibronic dynamics: case study on singlet fission solar cell design”, now published in Quantum Science and Technology!

📜 Paper: doi.org/10.1088/2058...
📝 Blog post overview: pennylane.ai/blog/2025/02...

[1/8]

1 3 12

The Matter Lab @thematterlab.bsky.social · 14d

A big thank you to the authors: Jonas Elsborg, Luca Thiede, @aspuru.bsky.social, @tvegge.bsky.social, and Arghya Bhowmik.
[6/6]

The Matter Lab @thematterlab.bsky.social · 14d

💡Technical tidbit: To make this work, we had to carefully break the symmetry of the equivariant network; otherwise, "local" reflection symmetries constrain the network’s expressivity too much.
[5/6]

1 1

The Matter Lab @thematterlab.bsky.social · 14d

When used to initialize plane-wave DFT, our densities reduce the number of SCF cycles by an average of 50.7% on the QM9 dataset compared to standard initialization.
[4/6]

1

The Matter Lab @thematterlab.bsky.social · 14d

- Compared to previous models that represent charge densities using atom-centered spherical harmonics-based orbitals, our floating Gaussians are much more efficient to evaluate, yielding a 10x speedup at similar or better accuracy compared to the previous state of the art.
[3/6]

1 1

The Matter Lab @thematterlab.bsky.social · 14d

Electra is a machine learning model for charge density prediction.

- Inspired by the incredible efficiency of Gaussian splatting, we represent the density as a mixture of "floating" Gaussians.
- Then, an equivariant graph network predicts the Gaussians' parameters.
[2/6]

1 1

The Matter Lab @thematterlab.bsky.social · 14d

We're extremely happy to share that our paper, ⭐ELECTRA: A Cartesian Network for 3D Charge Density Prediction with Floating Orbitals⭐, was accepted to NeurIPS 2025 as a spotlight paper.

Link to preprint: arxiv.org/abs/2503.08305
[1/6]

1 2 10

The Matter Lab @thematterlab.bsky.social · 19d

Collaboration with Raul Ortega Ochoa (DTU), @realmantilla.bsky.social (UofT), Juan Bernardo Pérez Sánchez (UofT), Mohsen Bagherimehrab (UofT), @an-aldossary.bsky.social (UofT), @tvegge.bsky.social (DTU), Tonio Buonassisi (MIT), and @aspuru.bsky.social (UofT)
[5/5]

1

The Matter Lab @thematterlab.bsky.social · 19d

➡️ Deep tomography — a hypothesis claiming that latent vectors should be informationally-equivalent to k-RDMs and should be limited by the N-representability problem
➡️ Outlook — future paths for better data and benchmarking for foundation models in chemistry.
[4/5]

1 1

The Matter Lab @thematterlab.bsky.social · 19d

They act as compressed surrogates of quantum states, enabling generalization across chemistry.

What’s inside?
➡️ Bridging ideas — connects concepts from quantum state tomography and the platonic hypothesis in representation learning.
[3/5]

1 1

The Matter Lab @thematterlab.bsky.social · 19d

We propose the Deep Tomography Hypothesis: Large ML models trained on informationally-complete molecular properties will learn latent representations that are informationally equivalent to reduced quantum density matrices.
[2/5]

1 1

The Matter Lab @thematterlab.bsky.social · 19d

🎉Excited to share our new perspective: “Connecting the concepts of quantum state tomography and molecular representations for machine learning” — now on ChemRxiv!

Link: doi.org/10.26434/che...

[1/5]

1 2 8

The Matter Lab @thematterlab.bsky.social · 21d

and many more who have contributed to our lab’s vibrant science and community.

¡Viva Latinoamérica!
[3/3]

1

The Matter Lab @thematterlab.bsky.social · 21d

We also wish a happy Hispanic Heritage Month to current and recent members with Latin American roots, including Luis Calderon, Irene Zuniga, Juan Perez, Jorge Arturo Campos Gonzalez Angulo, Maria Luiza Linazzi Peracchi, Jesus Valdes, Andres Guzman Cordero, Ignacio Gustin, Manuel Drehwald -
[2/3]

1 1