Accelerating GW calculations through machine learned dielectric matrices

JSON Export

{
  "id": "1837", 
  "updated": "2023-07-27T13:27:06.120638+00:00", 
  "metadata": {
    "version": 2, 
    "contributors": [
      {
        "givennames": "Mario", 
        "affiliations": [
          "Department of Materials, Imperial College London, United Kingdom"
        ], 
        "email": "mario.zauchner15@imperial.ac.uk", 
        "familyname": "Zauchner"
      }, 
      {
        "givennames": "Johannes", 
        "affiliations": [
          "Department of Materials, Imperial College London, United Kingdom"
        ], 
        "familyname": "Lischner"
      }, 
      {
        "givennames": "Andrew", 
        "affiliations": [
          "Department of Materials, Imperial College London, United Kingdom"
        ], 
        "familyname": "Horsfield"
      }
    ], 
    "title": "Accelerating GW calculations through machine learned dielectric matrices", 
    "_oai": {
      "id": "oai:materialscloud.org:1837"
    }, 
    "keywords": [
      "machine learning", 
      "ab initio", 
      "DFT"
    ], 
    "publication_date": "Jul 27, 2023, 15:27:06", 
    "_files": [
      {
        "key": "pseudo.tar.gz", 
        "description": "pseudopotentials", 
        "checksum": "md5:e2668e3bf973bc8f5747c70702d93ab8", 
        "size": 73146
      }, 
      {
        "key": "1-CDDRFs_train.tar.gz", 
        "description": "1-C DDRFs used for training the model.", 
        "checksum": "md5:84c25f3b1f74bea525b88473c8e23a34", 
        "size": 1368851006
      }, 
      {
        "key": "descriptors_H_train.tar.gz", 
        "description": "Hydrogen descriptors", 
        "checksum": "md5:071b40c03f38df2078ca63bf728189ed", 
        "size": 94224413
      }, 
      {
        "key": "descriptors_Si_train.tar.gz", 
        "description": "Silicon descriptors", 
        "checksum": "md5:c5c67683d210e00703d5f6c358ab8172", 
        "size": 214608047
      }, 
      {
        "key": "inputsperturbed.tar.gz", 
        "description": "Input files for perturbed structures.", 
        "checksum": "md5:f071f9e0446443fd6962c99ad077a975", 
        "size": 1221737
      }, 
      {
        "key": "inputsunperturbed.tar.gz", 
        "description": "Input files for unperturbed structures.", 
        "checksum": "md5:dacc067d87c27c40e179cf3edfcd63cd", 
        "size": 217294
      }, 
      {
        "key": "structuresunperturbed.tar.gz", 
        "description": "XYZ files of unperturbed structures.", 
        "checksum": "md5:80aadb8fc32a1946117a024ec0add23e", 
        "size": 102878
      }, 
      {
        "key": "structuresperturbed.tar.gz", 
        "description": "XYZ files for unperturbed structures.", 
        "checksum": "md5:617298a8698aeb128f47dab3b9ba19f9", 
        "size": 510413
      }, 
      {
        "key": "QPDATA.tar.gz", 
        "description": "Quasiparticle data.", 
        "checksum": "md5:82e34967aa85d9328b6d9b27507d7ed1", 
        "size": 205009
      }, 
      {
        "key": "README.txt", 
        "description": "Data description", 
        "checksum": "md5:a4f06420544fad4cf767199192b491f2", 
        "size": 1558
      }
    ], 
    "references": [
      {
        "doi": "https://doi.org/10.48550/arXiv.2305.02990", 
        "citation": "Mario G. Zauchner, Andrew Horsfield, and Johannes Lischner. Accelerating GW calculations\nthrough machine learned dielectric matrices, 2023.", 
        "type": "Preprint"
      }
    ], 
    "description": "The GW approach produces highly accurate quasiparticle energies, but its application to large systems is computationally challenging, which can be largely attributed to the difficulty in computing the inverse dielectric matrix. To address this challenge, we develop a machine learning approach to efficiently predict density-density response functions (DDRF) in materials. For this, an atomic decomposition of the DDRF is introduced as well as the neighbourhood density-matrix descriptor both of which transform in the same way under rotations. The resulting DDRFs are then used to evaluate quasiparticle energies via the GW approach. This technique is called the ML-GW approach. To assess the accuracy of this method, we apply it to hydrogenated silicon clusters and find that it reliably reproduces HOMO-LUMO gaps and quasiparticle energy levels. The accuracy of the predictions deteriorates when the approach is applied to larger clusters than those included in the training set. These advances pave the way towards GW calculations of complex systems, such as disordered materials, liquids, interfaces and nanoparticles.", 
    "status": "published", 
    "license": "Creative Commons Attribution 4.0 International", 
    "conceptrecid": "1833", 
    "is_last": true, 
    "mcid": "2023.119", 
    "edited_by": 98, 
    "id": "1837", 
    "owner": 1096, 
    "license_addendum": "", 
    "doi": "10.24435/materialscloud:gx-m3"
  }, 
  "revision": 7, 
  "created": "2023-07-25T11:33:38.995648+00:00"
}