Publication reference for the ML-DFT Density Predictor

[Lim changmin]

Hello,

I am currently using the ML-DFT DensityPredictor introduced in QuantumATK Y-2026.03.

The release notes explain that it uses a graph neural network to predict DFT-quality electron densities and can provide an initial density for SCF calculations. However, I could not find a publication reference describing the underlying methodology.

Is the Density Predictor based on any previously published method or research paper? If so, could you please provide the recommended reference(s)?

I would also like to know what reference would be most appropriate when mentioning the use of the QuantumATK Density Predictor in a research paper.

Thank you.

[yuby]

There doesn’t appear to be a published paper specifically describing the QuantumATK Y-2026.03 DensityPredictor yet.

For now, in a paper, the safest approach is to:

Cite the standard QuantumATK software/reference paper.
Cite the Y-2026.03 DensityPredictor documentation.
State that you used the pretrained ML DensityPredictor to generate the initial electron density.

Don’t cite a generic GNN/ML-density paper as the methodology reference unless QuantumATK explicitly identifies it as the basis of their implementation.

[Troels-Markussen]

Hi,

We don't have a paper reference paper about the ML-DFT capabilities in QuantumATK. Hopefully we will make one soon.
We use a modified version of the DeepDFT code (https://doi.org/10.1038/s41524-022-00863-y) that you may cite in addition to the normal QuantumATK citations.

Best regards,
Troels