Differentiable Quantum Chemistry (only Differentiable Density Functional Theory and Hartree Fock at the moment)

Last update: Dec 2, 2022

Related tags

Deep Learning automatic-differentiation density-functional-theory computational-chemistry computational-physics quantum-chemistry hartree-fock

Overview

DQC: Differentiable Quantum Chemistry

Differentiable quantum chemistry package. Currently only support differentiable density functional theory (DFT) and Hartree-Fock (HF) calculation.

Installation, tutorials, and documentations can be found at: https://dqc.readthedocs.io/

Applications

Here is a list of applications made easy by DQC. If you want your applications listed here, please contact us by opening an issue or make a pull request.

Applications	Repo	Paper
Learning xc functional from experimental data
Basis optimization
Alchemical perturbation

Citations

If you are using DQC for your publication, please kindly cite the following

@article{PhysRevLett.127.126403,
  title = {Learning the Exchange-Correlation Functional from Nature with Fully Differentiable Density Functional Theory},
  author = {Kasim, M. F. and Vinko, S. M.},
  journal = {Phys. Rev. Lett.},
  volume = {127},
  issue = {12},
  pages = {126403},
  numpages = {7},
  year = {2021},
  month = {Sep},
  publisher = {American Physical Society},
  doi = {10.1103/PhysRevLett.127.126403},
  url = {https://link.aps.org/doi/10.1103/PhysRevLett.127.126403}
}

If you want to read the paper in arxiv, you can find it here.

Comments

[BUG] different density matrix and Basis by using pscf and dqc

I discovered some bug or more or less something I don't understand. So first when I want to just read the Basis from Basis Set Exchange the Parameters are different. I discovered a normalization function in your code, but I don't know what it really does. Bur it changes the input coefficients of the Basis functions. Another Problem I got was by calculation the overlapmatix. In comparison to a similar calculation in pyscf the matrix values are sometime the same. Using another Basis, the overlapmatix differs from each others.

So here is a code to replicate the errors:


from pyscf import gto, scf
import dqc
import numpy as np
import xitorch as xt
from dqc.utils.datastruct import AtomCGTOBasis
import dqc.hamilton.intor as intor
from dqc.api.parser import parse_moldesc

#using pscf to calc:
#   the overlap Matrix
#   as well as print the coefficients of the basis sets

def overlapb_scf(basis):
    """
    just creates the overlap matrix for different input basis
    """
    mol = gto.Mole()
    mol.atom = [['H', [1.0, 0.0, 0.0]],
                ['H', [-1.0, 0.0, 0.0]]]
    mol.spin = 0
    mol.unit = 'Bohr' # in Angstrom
    mol.verbose = 6
    mol.output = 'scf.out'
    mol.symmetry = False
    mol.basis = basis
    mol.build()

    mf = scf.RHF(mol)
    mf.kernel()
    mf.mo_coeff
    dm = mf.make_rdm1()
    overlap = mf.get_ovlp()
    return overlap


########################################################################################################################

#using the dqc module to calc the basis and the overlap matrix

basis_dqc = [dqc.loadbasis("1:3-21G"), dqc.loadbasis("1:3-21G")]
basis2_dqc = [dqc.loadbasis("1:cc-pvdz"), dqc.loadbasis("1:cc-pvdz")]

def overlapb_dqc(basis):
    """
      just creates the overlap matrix for different input basis
    """
    bpacker = xt.Packer(basis)
    bparams = bpacker.get_param_tensor()

    atomstruc = "H 1 0 0; H -1 0 0"
    atomzs, atompos = parse_moldesc(atomstruc)
    atombases = [AtomCGTOBasis(atomz=atomzs[i], bases=basis[i], pos=atompos[i]) for i in range(len(basis))]

    wrap = dqc.hamilton.intor.LibcintWrapper(
            atombases)  # creates an wrapper object to pass informations on lower functions
    return intor.overlap(wrap)

print(f"the pyscf:\n3-21G: \n\t{gto.basis.load('3-21G',symb='H')}\n"
      f"coefficients of cc-pvdz: \n\t{gto.basis.load('cc-pvdz',symb='H')} ")
print(f"\nthe dqc:\n1:3-21G:\n\t {basis_dqc[0]},\n1:cc-pvdz\n\t{basis2_dqc[0]} ")

#checks that the overlap matrix is the same:
print(f"3-21G: {np.all(np.isclose(overlapb_scf('3-21G'),overlapb_dqc(basis_dqc)))}"
      f"\ncc-pvdz: {np.all(np.isclose(overlapb_scf('cc-pvdz'),overlapb_dqc(basis2_dqc)))}")

So the output is:


the pyscf:
3-21G: 
	[[0, [5.447178, 0.156285], [0.824547, 0.904691]], [0, [0.183192, 1.0]]]
coefficients of cc-pvdz: 
	[[0, [13.01, 0.019685], [1.962, 0.137977], [0.4446, 0.478148]], [0, [0.122, 1.0]], [1, [0.727, 1.0]]] 

the dqc:
1:3-21G:
	 [CGTOBasis(angmom=0, alphas=tensor([5.4472, 0.8245], dtype=torch.float64), coeffs=tensor([1.4079, 1.9778], dtype=torch.float64), normalized=True), CGTOBasis(angmom=0, alphas=tensor([0.1832], dtype=torch.float64), coeffs=tensor([0.7074], dtype=torch.float64), normalized=True)],
1:cc-pvdz
	[CGTOBasis(angmom=0, alphas=tensor([13.0100,  1.9620,  0.4446,  0.1220], dtype=torch.float64), coeffs=tensor([0.3407, 0.5779, 0.6577, 0.2614], dtype=torch.float64), normalized=True), CGTOBasis(angmom=0, alphas=tensor([0.1220], dtype=torch.float64), coeffs=tensor([0.5215], dtype=torch.float64), normalized=True), CGTOBasis(angmom=1, alphas=tensor([0.7270], dtype=torch.float64), coeffs=tensor([1.9584], dtype=torch.float64), normalized=True)] 
overwrite output file: scf.out
overwrite output file: scf.out
3-21G: True
cc-pvdz: False

So my question is when does the overlap matrix the same in pscf and dqc. As well as when will they differ from each other?

Desktop (please complete the following information):

DQC version: [e.g. 0.1.0]
Python version: [e.g. 3.8]
PyTorch version: [e.g. 1.9]
OS: [Ubuntu 18.04]

opened by Jaikinator 10

[ENH] Geometry optimization
Is your feature request related to a problem? Please describe. I would like to optimize the geometry of a molecule from an initial configuration. This would then allow me to compute vibrational frequencies and IR spectra at a local minimum of the potential energy surface.

Describe the solution you'd like

In PySCF I can use their geometry_optimizer to optimize the geometry of a molecule, for example something like

from pyscf import gto from pyscf.geomopt.geometric_solver import optimize as geometric_opt mol = gto.Mole(atom=atom, basis = 'sto-3g') # atom is some initial starting description of the molecule rhf = scf.RHF(mol) mol_opt = geometric_opt(rhf)

I would like to do something similar in dqc, for example something like

import dqc mol = dqc.Mol(moldesc=moldesc, basis="sto-3g") # moldesc is some initial starting description of the molecule mol.atompos.requires_grad_() hf = dqc.HF(mol).run() mol_opt = dqc.optimize_geometry(hf) # new proposed function

Describe alternatives you've considered Alternatively this would have to be implemented outside of dqc.

Additional context I know there are many algorithms possible for geometry optimization and I am very open to whichever is recommended.
opened by sofroniewn 8
[BUG] calculate density matrix
Normally, for example in pyscf the density Matrix P is defined by the expansion coefficient Matrix.
So the trace of P*S (where S is the overlap Matrix) should be equal to the number of Electrons in the system. When I try your density Matrix given by the code below, the trace of the density Matrix itself result in the number of electrons. So, can you explain to me why you use it like this?

The Program basis = [dqc.loadbasis("1:3-21G"), dqc.loadbasis("1:3-21G")] atomstruc = "H 1 0 0; H -1 0 0" m = dqc.Mol(atomstruc, basis = basis) qc = dqc.HF(m).run() dm = qc.aodm() print(torch.trace(dm)) #gives 2 in this example

Desktop (please complete the following information):

DQC version: 0.1.0

Python version: 3.8.1

PyTorch version: 1.9

OS: Ubuntu
opened by Jaikinator 5
Add optimal_geometry method

This PR closes #10 by adding an optimal_geometry method to the properties API. It also adds a make_copy method to the Mol and Sol systems. It adds basic tests to cover the make_copy and optimal_geometry.

Overall I tried to match the code, naming, and commenting style of the repo as best I could - feel free to change or request changes on anything you don't like.

One thing I am wondering is if you still like the optimal_geometry name and function descriptions. Alternatives could emphasize that it is a position that is a local minimum of the energy. I don't really mind.

opened by sofroniewn 4

[BUG] Big difference in PySCF vs dqc vibrational frequencies

Describe the bug When I compute vibrational frequencies using PySCF and dqc using the same level of theory I see very different results, and I'm not sure how to interpret the dqc frequencies.

To Reproduce To compute the dqc frequencies I do

import dqc

moldesc = 'C          -1.28652        0.25524        0.08013;C           0.07762        0.16535       -0.63495;O           0.22083        0.47033       -1.80613;C           1.24790       -0.33747        0.23567;H          -2.04623        0.61618       -0.60539;H          -1.22080        0.93293        0.92681;H          -1.57890       -0.72292        0.45183;H           1.03268       -1.33368        0.61210;H           2.16265       -0.36812       -0.34714;H           1.39077        0.32217        1.08708'

mol = dqc.Mol(moldesc=moldesc, basis="sto-3g")
mol.atompos.requires_grad_()
hf = dqc.HF(mol).run()
vib = dqc.vibration(hf, freq_unit='cm^-1') # calculate vibrational frequencies
freq_dqc = vib[0].detach().numpy()
print(freq_dqc)

# Which gives
array([ 1.70294616e+04,  1.69718735e+04,  1.68860090e+04,  1.68255299e+04,
        1.68225380e+04,  1.67753119e+04,  1.44644965e+04,  9.52891028e+03,
        9.07347876e+03,  1.88365866e-04,  1.02303786e-04, -6.48131738e-05,
       -5.43764168e+03, -6.09925459e+03, -6.22499857e+03, -6.99690325e+03,
       -9.01539897e+03, -1.08179418e+04, -1.08671386e+04, -1.08868149e+04,
       -1.11379683e+04, -1.12761626e+04, -1.14500718e+04, -1.19746118e+04,
       -1.25462017e+04, -1.26196271e+04, -1.27329229e+04, -1.27626848e+04,
       -1.29649564e+04, -1.29682415e+04])

To compute the PySCF frequencies I do

from pyscf import gto
from pyscf.hessian import thermo


moldesc = 'C          -1.28652        0.25524        0.08013;C           0.07762        0.16535       -0.63495;O           0.22083        0.47033       -1.80613;C           1.24790       -0.33747        0.23567;H          -2.04623        0.61618       -0.60539;H          -1.22080        0.93293        0.92681;H          -1.57890       -0.72292        0.45183;H           1.03268       -1.33368        0.61210;H           2.16265       -0.36812       -0.34714;H           1.39077        0.32217        1.08708'

mol = gto.Mole(atom=moldesc, basis='sto-3g')
mol.build()
mf = mol.RHF().run()
hessian = mf.Hessian().kernel()
freq_pyscf = thermo.harmonic_analysis(mf.mol, hessian)['freq_wavenumber']
print(freq_pyscf)

# Which gives
array([ 117.92712198,  163.62556757,  394.58567756,  531.09155165,
        557.37323521,  905.59174999, 1100.86423523, 1103.22104045,
       1263.96843255, 1299.84707843, 1432.34457908, 1705.964129  ,
       1706.20007991, 1806.21035873, 1810.62541007, 1810.74231001,
       1814.3241799 , 2129.55729236, 3564.58835085, 3565.97676807,
       3740.70588451, 3742.69670576, 3761.55937183, 3763.21894889])

The molecule I'm using is acetone which I have done geometry optimization on in PySCF, also using HF 'sto-3g'. The values in wavenumber are roughly in line for what I'd expect for acetone based on the literature. The dqc numbers are over a wide range, both positive and negative, despite using the same level of theory.

Expected behavior I expected the dqc numbers to be more inline with the PySCF numbers

Desktop (please complete the following information):

DQC version: 0.1.0
Python version: 3.9
PyTorch version: 1.10.2
OS: macOS

Additional context I am pretty new to PySCF, dqc and computational chemistry overall so I apologies if my problem here is a real beginners mistake and I am not understanding things correctly. I hope you also don't mind all the issues I am creating - I'm really impressed with dqc and am hoping to use it in one of my projects!

opened by sofroniewn 4

[BUG] Differentiable tensor atom position error if molecule initialized from string

Describe the bug I am unable to perform a vibration calculation when creating a molecule from a string description

To Reproduce Steps to reproduce the behavior:

import torch
import dqc
moldesc = "H 1 0 0; H -1 0 0"
mol = dqc.Mol(moldesc=moldesc, basis="3-21G")
qc = dqc.HF(mol).run()
ene = qc.energy()  # calculate the energy
print(ene)
vib = dqc.vibration(qc) # calculate vibrational frequencies
print(vib)

Gives

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
/Users/nsofroniew/Documents/code/chem/molecular_vibrations.ipynb Cell 50' in <module>
      [6](vscode-notebook-cell:/Users/nsofroniew/Documents/code/chem/molecular_vibrations.ipynb#ch0000060?line=5)[ ene = qc.energy()  # calculate the energy
      ]()[7](vscode-notebook-cell:/Users/nsofroniew/Documents/code/chem/molecular_vibrations.ipynb#ch0000060?line=6)[ print(ene)
----> ]()[8](vscode-notebook-cell:/Users/nsofroniew/Documents/code/chem/molecular_vibrations.ipynb#ch0000060?line=7)[ vib = dqc.vibration(qc) # calculate vibrational frequency
      ]()[9](vscode-notebook-cell:/Users/nsofroniew/Documents/code/chem/molecular_vibrations.ipynb#ch0000060?line=8)[ print(vib)

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py:66, in vibration(qc, freq_unit, length_unit)
     ]()[41](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=40)[ def vibration(qc: BaseQCCalc, freq_unit: Optional[str] = "cm^-1",
     ]()[42](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=41)[               length_unit: Optional[str] = None) -> Tuple[torch.Tensor, torch.Tensor]:
     ]()[43](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=42)[     """
     ]()[44](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=43)[     Calculate the vibration mode of the system based on the Hessian of energy
     ]()[45](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=44)[     with respect to atomic position.
   (...)
     ]()[64](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=63)[         to each axis sorted from the largest frequency to smallest frequency.
     ]()[65](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=64)[     """
---> ]()[66](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=65)[     freq, mode = _vibration(qc)
     ]()[67](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=66)[     freq = freq_to(freq, freq_unit)
     ]()[68](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=67)[     mode = length_to(mode, length_unit)

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py:32, in memoize_method.<locals>.new_fcn(self)
     ]()[30](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=29)[     return self.__dict__[cachename]
     ]()[31](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=30)[ else:
---> ]()[32](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=31)[     res = fcn(self)
     ]()[33](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=32)[     self.__dict__[cachename] = res
     ]()[34](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=33)[     return res

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py:269, in _vibration(qc)
    ]()[264](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=263)[ @memoize_method
    ]()[265](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=264)[ def _vibration(qc: BaseQCCalc) -> Tuple[torch.Tensor, torch.Tensor]:
    ]()[266](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=265)[     # calculate the vibrational mode and returns the frequencies and normal modes
    ]()[267](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=266)[     # freqs are sorted from largest to smallest
--> ]()[269](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=268)[     hess = hessian_pos(qc)  # (natoms * ndim, natoms * ndim)
    ]()[270](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=269)[     mass = qc.get_system().atommasses  # (natoms)
    ]()[271](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=270)[     ndim = hess.shape[0] // mass.shape[0]

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py:37, in hessian_pos(qc, unit)
     ]()[19](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=18)[ def hessian_pos(qc: BaseQCCalc, unit: Optional[str] = None) -> torch.Tensor:
     ]()[20](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=19)[     """
     ]()[21](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=20)[     Returns the Hessian of energy with respect to atomic positions.
     ]()[22](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=21)[ 
   (...)
     ]()[35](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=34)[         of the energy with respect to the atomic position
     ]()[36](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=35)[     """
---> ]()[37](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=36)[     hess = _hessian_pos(qc)
     ]()[38](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=37)[     hess = length_to(hess, unit)
     ]()[39](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=38)[     return hess

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py:32, in memoize_method.<locals>.new_fcn(self)
     ]()[30](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=29)[     return self.__dict__[cachename]
     ]()[31](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=30)[ else:
---> ]()[32](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=31)[     res = fcn(self)
     ]()[33](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=32)[     self.__dict__[cachename] = res
     ]()[34](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=33)[     return res

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py:257, in _hessian_pos(qc)
    ]()[254](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=253)[ atompos = system.atompos
    ]()[256](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=255)[ # check if the atompos requires grad
--> ]()[257](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=256)[ _check_differentiability(atompos, "atom positions", "hessian")
    ]()[259](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=258)[ # calculate the jacobian
    ]()[260](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=259)[ jac_e_pos = _jac(ene, atompos, create_graph=True)  # (natoms * ndim)

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py:418, in _check_differentiability(a, aname, propname)
    ]()[416](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=415)[ if not (isinstance(a, torch.Tensor) and a.requires_grad):
    ]()[417](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=416)[     msg = "Differentiable tensor %s is required to calculate the %s" % (aname, propname)
--> ]()[418](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=417)[     raise RuntimeError(msg)

RuntimeError: Differentiable tensor atom positions is required to calculate the hessian]()

Expected behavior I expect to be able to run the vibration calculation. If I use the following code from the examples

import torch
import dqc
atomzs = torch.tensor([1, 1])
atomposs = torch.tensor([[1, 0, 0], [-1, 0, 0]], dtype=torch.double).requires_grad_()
mol = dqc.Mol(moldesc=(atomzs, atomposs), basis="3-21G")
qc = dqc.HF(mol).run()
ene = qc.energy()  # calculate the energy
print(ene)
vib = dqc.vibration(qc) # calculate vibrational frequencies
print(vib)

Then it works. I will note that the requires_grad_() call is necessary, otherwise I get the same error as above. From an API standpoint I think it would be much cleaner if atomposs did not need the requires_grad_() call, but if that was something you did called during the init.

Desktop (please complete the following information):

DQC version: 0.1.0
Python version: 3.9
PyTorch version: 1.10.2
OS: macOS

Additional context Overall I'm very excited about dqc, you've done a really nice job from what I can tell so far!!

opened by sofroniewn 4

[BUG] Unable to compute ir_spectrum

Describe the bug I am unable to compute the ir_spectrum of a molecule

To Reproduce Either

moldesc = 'C          -1.28652        0.25524        0.08013;C           0.07762        0.16535       -0.63495;O           0.22083        0.47033       -1.80613;C           1.24790       -0.33747        0.23567;H          -2.04623        0.61618       -0.60539;H          -1.22080        0.93293        0.92681;H          -1.57890       -0.72292        0.45183;H           1.03268       -1.33368        0.61210;H           2.16265       -0.36812       -0.34714;H           1.39077        0.32217        1.08708'
efield = torch.tensor([0, 0, 0], dtype=torch.double)
mol = dqc.Mol(moldesc=moldesc, basis="sto-3g", efield=efield)
print(len(mol.atomzs))
hf = dqc.HF(mol).run()
mol.atompos.requires_grad_()
ir = dqc.ir_spectrum(hf, freq_unit='cm^-1') # calculate ir spectrum

which gives

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
/Users/nsofroniew/Documents/code/chem/molecular_vibrations.ipynb Cell [5](vscode-notebook-cell:/Users/nsofroniew/Documents/code/chem/molecular_vibrations.ipynb#ch0000059?line=4)0' in <module>
      5[ hf = dqc.HF(mol).run()
      ]()[6](vscode-notebook-cell:/Users/nsofroniew/Documents/code/chem/molecular_vibrations.ipynb#ch0000059?line=5)[ mol.atompos.requires_grad_()
----> ]()[7](vscode-notebook-cell:/Users/nsofroniew/Documents/code/chem/molecular_vibrations.ipynb#ch0000059?line=6)[ ir = dqc.ir_spectrum(qc, freq_unit='cm^-1')

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py:95, in ir_spectrum(qc, freq_unit, ints_unit)
     ]()[71](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=70)[ def ir_spectrum(qc: BaseQCCalc, freq_unit: Optional[str] = "cm^-1",
     ]()[72](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=71)[                 ints_unit: Optional[str] = "(debye/angst)^2/amu") \
     ]()[73](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=72)[         -> Tuple[torch.Tensor, torch.Tensor]:
     ]()[74](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=73)[     """
     ]()[75](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=74)[     Calculate the frequency and intensity of the IR vibrational spectra.
     ]()[76](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=75)[     Unlike ``vibration``, this method only returns parts where the frequency is
   (...)
     ]()[93](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=92)[         tensor is the IR intensity with the same order as the frequency.
     ]()[94](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=93)[     """
---> ]()[95](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=94)[     freq, ir_ints = _ir_spectrum(qc)
     ]()[96](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=95)[     freq = freq_to(freq, freq_unit)
     ]()[97](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=96)[     ir_ints = ir_ints_to(ir_ints, ints_unit)

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py:32, in memoize_method.<locals>.new_fcn(self)
     ]()[30](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=29)[     return self.__dict__[cachename]
     ]()[31](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=30)[ else:
---> ]()[32](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=31)[     res = fcn(self)
     ]()[33](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=32)[     self.__dict__[cachename] = res
     ]()[34](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=33)[     return res

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py:303, in _ir_spectrum(qc)
    ]()[301](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=300)[ # get the derivative of dipole moment w.r.t. positions
    ]()[302](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=301)[ with torch.enable_grad():
--> ]()[303](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=302)[     mu = _edipole(qc)  # (ndim)
    ]()[304](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=303)[ dmu_dr = _jac(mu, atompos)  # (ndim, natoms * ndim)
    ]()[305](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=304)[ dmu_dq = torch.matmul(dmu_dr, normal_modes)  # (ndim, nfreqs)

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py:32, in memoize_method.<locals>.new_fcn(self)
     ]()[30](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=29)[     return self.__dict__[cachename]
     ]()[31](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=30)[ else:
---> ]()[32](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=31)[     res = fcn(self)
     ]()[33](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=32)[     self.__dict__[cachename] = res
     ]()[34](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/utils/misc.py?line=33)[     return res

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py:351, in _edipole(qc)
    ]()[349](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=348)[ system = qc.get_system()
    ]()[350](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=349)[ efield = system.efield
--> ]()[351](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=350)[ assert isinstance(efield, tuple)
    ]()[352](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=351)[ assert len(efield) > 0, "To calculate dipole, the constant electric field must be provided"
    ]()[354](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/dqc/api/properties.py?line=353) # check if the electric field requires grad

AssertionError:

This fails as efield is not a tuple

or if I leave efield unspecified

moldesc = 'C          -1.28652        0.25524        0.08013;C           0.07762        0.16535       -0.63495;O           0.22083        0.47033       -1.80613;C           1.24790       -0.33747        0.23567;H          -2.04623        0.61618       -0.60539;H          -1.22080        0.93293        0.92681;H          -1.57890       -0.72292        0.45183;H           1.03268       -1.33368        0.61210;H           2.16265       -0.36812       -0.34714;H           1.39077        0.32217        1.08708'
mol = dqc.Mol(moldesc=moldesc, basis="sto-3g")
print(len(mol.atomzs))
hf = dqc.HF(mol).run()
mol.atompos.requires_grad_()
ir = dqc.ir_spectrum(hf, freq_unit='cm^-1') # calculate ir spectrum

I get the same error message.

Expected behavior I should be able to pass an efield and then compute an ir spectrum (or ideally if I didn't pass one you could take a reasonable default value).

Desktop (please complete the following information):

DQC version: 0.1.0
Python version: 3.9
PyTorch version: 1.10.2
OS: macOS

opened by sofroniewn 3

[BUG] Customized meta-GGA functional
I tried to run the customized meta-GGA functional by setting dqc.xc.CustomXC.family =4, but it did not work. This is the code:

import torch import dqc import torch import dqc.xc import dqc.utils class MyXC(dqc.xc.CustomXC): def __init__(self): super().__init__() @property def family(self): # 1 for LDA, 2 for GGA, 4 for MGGA return 4 def get_edensityxc(self, densinfo): scan = dqc.get_xc("MGGA_X_SCAN + MGGA_C_SCAN") return scan.get_edensityxc(densinfo) myxc = MyXC() atomzs, atomposs = dqc.parse_moldesc("H -1 0 0; H 1 0 0") atomposs = atomposs.requires_grad_() # mark atomposs as differentiable mol = dqc.Mol((atomzs, atomposs), basis="6-311++G(3df,3pd)") qc = dqc.KS(mol, xc=myxc, variational=False).run()

Then I got the following error message:

Traceback (most recent call last): File "test_customxc.py", line 29, in qc = dqc.KS(mol, xc=myxc, variational=False).run() File "/usr/local/lib/python3.8/dist-packages/dqc/qccalc/scf_qccalc.py", line 90, in run scp0 = self._engine.dm2scp(dm) File "/usr/local/lib/python3.8/dist-packages/dqc/qccalc/ks.py", line 116, in dm2scp return self.__dm2fock(dm).fullmatrix() File "/usr/local/lib/python3.8/dist-packages/dqc/qccalc/ks.py", line 177, in __dm2fock vxc = self.hamilton.get_vxc(dm) # spin param or tensor (..., nao, nao) File "/usr/local/lib/python3.8/dist-packages/dqc/hamilton/hcgto.py", line 236, in get_vxc potinfo = self.xc.get_vxc(densinfo) # value: (*BD, nr) File "/usr/local/lib/python3.8/dist-packages/dqc/xc/base_xc.py", line 96, in get_vxc raise NotImplementedError("Default vxc for family %d is not implemented" % self.family) NotImplementedError: Default vxc for family 4 is not implemented

DQC version: 0.1.0

Python version: 3.3

PyTorch version: 1.10.1

OS: Ubuntu 20.04 LTS
opened by ml-electron-project 2

[BUG] Wrong number of vibrational frequencies

Describe the bug The number of vibrational frequencies do not correspond to natoms * ndim.

To Reproduce Steps to reproduce the behavior:

moldesc = 'C          -1.28652        0.25524        0.08013;C           0.07762        0.16535       -0.63495;O           0.22083        0.47033       -1.80613;C           1.24790       -0.33747        0.23567;H          -2.04623        0.61618       -0.60539;H          -1.22080        0.93293        0.92681;H          -1.57890       -0.72292        0.45183;H           1.03268       -1.33368        0.61210;H           2.16265       -0.36812       -0.34714;H           1.39077        0.32217        1.08708'
mol = dqc.Mol(moldesc=moldesc, basis="sto-3g")
print(len(mol.atomzs))
hf = dqc.HF(mol).run()
mol.atompos.requires_grad_()
vib = dqc.vibration(qc, freq_unit='cm^-1') # calculate vibrational frequencies
print(vib[0].shape, vib[1].shape)

Here I am trying to calculate the vibrational spectrum of acetone, which as 10 atoms in it (I'm using positions I obtained from a previous geometry optimization in PySCF). The output of the above script is

10
torch.Size([6]) torch.Size([6, 6])

So I have 10 atoms as expected, but I only have six vibrational frequencies.

Expected behavior I expect to get 10 * 3 or 30 frequencies (of which 24 should correspond to vibrations).

Desktop (please complete the following information):

DQC version: 0.1.0
Python version: 3.9
PyTorch version: 1.10.2
OS: macOS

opened by sofroniewn 1

[ENH] GPU support
I tried to run the calculation with the device of Mol set to be cuda.

atomzs, atomposs = dqc.parse_moldesc(atom_positions) mol = dqc.Mol((atomzs, atomposs), basis="6-311++G(3df,3pd)", device= torch.device("cuda")) qc=dqc.HF(mol, variational=False) qc.run()

However, it occurred errors because some internal torch.tensor such as "dm" are not copied to the cuda device. It worked by just modifying them to refer "Mol.device". I found it sped up HF, KS, and especially force calculation.

I suggest that it would be useful if this modification is officially implemented.
opened by ml-electron-project 1

[BUG] Bug in `optimal_geometry` for KS case

Describe the bug There is a bug in the optimal_geometry function add in #11 for DFT case

To Reproduce Steps to reproduce the behavior:

import dqc

moldesc = "O 0.0000 0.0000 0.2217; H 0.0000 1.4309 -0.8867; H 0.0000 -1.4309 -0.8867" 
m = dqc.Mol(moldesc, basis="cc-pvdz", grid=4, efield=efield)
m.atompos.requires_grad_()
# sys = dqc.HF(m).run() # WORKS
sys = dqc.KS(m, xc="lda_x+lda_c_pw").run() # DOESN'T WORK
dqc.optimal_geometry(sys)

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
/Users/nsofroniew/Documents/code/chem/dqc_0.ipynb Cell 24' in <module>
      [6](vscode-notebook-cell:/Users/nsofroniew/Documents/code/chem/dqc_0.ipynb#ch0000030?line=5)[ # sys = dqc.HF(m).run() # WORKS
      ]()[7](vscode-notebook-cell:/Users/nsofroniew/Documents/code/chem/dqc_0.ipynb#ch0000030?line=6)[ sys = dqc.KS(m, xc="lda_x+lda_c_pw").run() # DOESN'T WORK
----> ]()[8](vscode-notebook-cell:/Users/nsofroniew/Documents/code/chem/dqc_0.ipynb#ch0000030?line=7)[ dqc.optimal_geometry(sys)

File ~/GitHub/dqc/dqc/api/properties.py:339, in optimal_geometry(qc, length_unit)
    ]()[321](file:///~/GitHub/dqc/dqc/api/properties.py?line=320)[ def optimal_geometry(qc: BaseQCCalc, length_unit: Optional[str] = None) -> torch.Tensor:
    ]()[322](file:///~/GitHub/dqc/dqc/api/properties.py?line=321)[     """
    ]()[323](file:///~/GitHub/dqc/dqc/api/properties.py?line=322)[     Compute the optimal atomic positions of the system.
    ]()[324](file:///~/GitHub/dqc/dqc/api/properties.py?line=323)[ 
   (...)
    ]()[337](file:///~/GitHub/dqc/dqc/api/properties.py?line=336)[         of atoms at the optimal geometry.
    ]()[338](file:///~/GitHub/dqc/dqc/api/properties.py?line=337)[     """
--> ]()[339](file:///~/GitHub/dqc/dqc/api/properties.py?line=338)[     atompos = _optimal_geometry(qc)
    ]()[340](file:///~/GitHub/dqc/dqc/api/properties.py?line=339)[     atompos = convert_length(atompos, to_unit=length_unit)
    ]()[341](file:///~/GitHub/dqc/dqc/api/properties.py?line=340)[     return atompos

File ~/GitHub/dqc/dqc/utils/misc.py:32, in memoize_method.<locals>.new_fcn(self)
     ]()[30](file:///~/GitHub/dqc/dqc/utils/misc.py?line=29)[     return self.__dict__[cachename]
     ]()[31](file:///~/GitHub/dqc/dqc/utils/misc.py?line=30)[ else:
---> ]()[32](file:///~/GitHub/dqc/dqc/utils/misc.py?line=31)[     res = fcn(self)
     ]()[33](file:///~/GitHub/dqc/dqc/utils/misc.py?line=32)[     self.__dict__[cachename] = res
     ]()[34](file:///~/GitHub/dqc/dqc/utils/misc.py?line=33)[     return res

File ~/GitHub/dqc/dqc/api/properties.py:503, in _optimal_geometry(qc)
    ]()[500](file:///~/GitHub/dqc/dqc/api/properties.py?line=499)[     return ene
    ]()[502](file:///~/GitHub/dqc/dqc/api/properties.py?line=501)[ # get the minimal enery position
--> ]()[503](file:///~/GitHub/dqc/dqc/api/properties.py?line=502)[ minpos = xitorch.optimize.minimize(_get_energy, atompos, method="gd", maxiter=200, 
    ]()[504](file:///~/GitHub/dqc/dqc/api/properties.py?line=503)[                                    step=1e-2)
    ]()[506](file:///~/GitHub/dqc/dqc/api/properties.py?line=505)[ return minpos

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py:275, in minimize(fcn, y0, params, bck_options, method, **fwd_options)
    ]()[270](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=269)[ # if it is just using the rootfinder algorithm, then the forward function
    ]()[271](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=270)[ # is the one that returns only the grad
    ]()[272](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=271)[ else:
    ]()[273](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=272)[     _fwd_fcn = _rf_fcn
--> ]()[275](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=274)[ return _RootFinder.apply(_rf_fcn, y0, _fwd_fcn, opt_method, fwd_options, bck_options,
    ]()[276](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=275)[                          len(params), *params, *pfunc.objparams())

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py:302, in _RootFinder.forward(ctx, fcn, y0, fwd_fcn, is_opt_method, options, bck_options, nparams, *allparams)
    ]()[300](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=299)[     name = "rootfinder" if not is_opt_method else "minimizer"
    ]()[301](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=300)[     method_fcn = get_method(name, methods, method)
--> ]()[302](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=301)[     y = method_fcn(fwd_fcn, y0, params, **config)
    ]()[304](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=303)[ ctx.fcn = fcn
    ]()[305](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=304)[ ctx.is_opt_method = is_opt_method

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_impls/optimize/minimizer.py:50, in gd(fcn, x0, params, step, gamma, maxiter, f_tol, f_rtol, x_tol, x_rtol, verbose, **unused)
     ]()[48](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_impls/optimize/minimizer.py?line=47)[ v = torch.zeros_like(x)
     ]()[49](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_impls/optimize/minimizer.py?line=48)[ for i in range(maxiter):
---> ]()[50](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_impls/optimize/minimizer.py?line=49)[     f, dfdx = fcn(x, *params)
     ]()[52](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_impls/optimize/minimizer.py?line=51)[     # update the step
     ]()[53](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_impls/optimize/minimizer.py?line=52)[     v = (gamma * v - step * dfdx).detach()

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_core/pure_function.py:34, in PureFunction.__call__(self, *params)
     ]()[33](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_core/pure_function.py?line=32)[ def __call__(self, *params):
---> ]()[34](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_core/pure_function.py?line=33)[     return self._fcntocall(*params)

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py:256, in minimize.<locals>._min_fwd_fcn(y, *params)
    ]()[254](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=253)[ with torch.enable_grad():
    ]()[255](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=254)[     y1 = y.clone().requires_grad_()
--> ]()[256](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=255)[     z = pfunc(y1, *params)
    ]()[257](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=256)[ grady, = torch.autograd.grad(z, (y1,), retain_graph=True,
    ]()[258](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=257)[                              create_graph=torch.is_grad_enabled())
    ]()[259](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/optimize/rootfinder.py?line=258)[ return z, grady

File ~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_core/pure_function.py:34, in PureFunction.__call__(self, *params)
     ]()[33](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_core/pure_function.py?line=32)[ def __call__(self, *params):
---> ]()[34](file:///~/opt/anaconda3/envs/chem/lib/python3.9/site-packages/xitorch/_core/pure_function.py?line=33)[     return self._fcntocall(*params)

File ~/GitHub/dqc/dqc/api/properties.py:498, in _optimal_geometry.<locals>._get_energy(atompos)
    ]()[496](file:///~/GitHub/dqc/dqc/api/properties.py?line=495)[ def _get_energy(atompos: torch.Tensor) -> torch.Tensor:
    ]()[497](file:///~/GitHub/dqc/dqc/api/properties.py?line=496)[     new_system = system.make_copy(moldesc=(system.atomzs, atompos))
--> ]()[498](file:///~/GitHub/dqc/dqc/api/properties.py?line=497)[     new_qc = qc.__class__(new_system).run()
    ]()[499](file:///~/GitHub/dqc/dqc/api/properties.py?line=498)[     ene = new_qc.energy()  # calculate the energy
    ]()[500](file:///~/GitHub/dqc/dqc/api/properties.py?line=499)[     return ene

TypeError: __init__() missing 1 required positional argument: 'xc']()

Expected behavior The method should work

Desktop (please complete the following information):

DQC version: master
Python version: 3.9
PyTorch version: 1.10.2
OS: macOS

Additional context I will try and fix and add a test when I get the chance - maybe later tonight or tomorrow depending on timing. Hopefully it will be an easy fix, just looks like something missing from an __init__

opened by sofroniewn 1

Add initial asv benchmarks

This PR begins work towards #12 by adding an initial airspeed velocity compatible benchmark. airspeed velocity must be installed and configured to run them.

Right now it includes just one benchmark file that has benchmarks for both dqc and PySCF. These could be split into two files if preferred.

I also just include a couple simple energy calculations for two molecules, based this notebook.

In my mind this PR is really just illustrative to give a sense of what might be possible. We can either continue discussion on what we want to include in #12 or here.

Here are the results of benchmarks on my 2019 macbook pro

(chem) laptop:dqc nsofroniew$ asv run -E existing -b TimePySCF
· virtualenv package not installed
· Discovering benchmarks
· Running 2 total benchmarks (1 commits * 1 environments * 2 benchmarks)
[  0.00%] ·· Benchmarking existing-py_Users_nsofroniew_opt_anaconda3_envs_chem_bin_python
[ 25.00%] ··· Running (benchmarks.TimePySCF.time_energy_HF--)..
[ 75.00%] ··· benchmarks.TimePySCF.time_energy_HF                                                                                                                                                         ok
[ 75.00%] ··· ========== ============
               molecule
              ---------- ------------
                 H2O       61.2±2ms
                C4H5N     1.52±0.01s
              ========== ============

[100.00%] ··· benchmarks.TimePySCF.time_energy_LDA                                                                                                                                                        ok
[100.00%] ··· ========== ============
               molecule
              ---------- ------------
                 H2O       316±3ms
                C4H5N     4.78±0.02s
              ========== ============

(chem) laptop:dqc:dqc nsofroniew$ asv run -E existing -b TimeDQC
· virtualenv package not installed
· Discovering benchmarks
· Running 2 total benchmarks (1 commits * 1 environments * 2 benchmarks)
[  0.00%] ·· Benchmarking existing-py_Users_nsofroniew_opt_anaconda3_envs_chem_bin_python
[ 25.00%] ··· Running (benchmarks.TimeDQC.time_energy_HF--).
[ 50.00%] ··· Running (benchmarks.TimeDQC.time_energy_LDA--).
[ 75.00%] ··· benchmarks.TimeDQC.time_energy_HF                                                                                                                                                           ok
[ 75.00%] ··· ========== ===========
               molecule
              ---------- -----------
                 H2O       63.4±1ms
                C4H5N     16.0±0.3s
              ========== ===========

[100.00%] ··· benchmarks.TimeDQC.time_energy_LDA                                                                                                                                                          ok
[100.00%] ··· ========== ============
               molecule
              ---------- ------------
                 H2O       414±4ms
                C4H5N     19.9±0.08s
              ========== ============

opened by sofroniewn 5

[ENH] Performance/ Benchmarking/ Scaling

Is your feature request related to a problem? Please describe. This is maybe more of a question than a feature request - but it could turn into one.

I saw you've got one of two files in the repo around benchmarks - like time_forward.py - but I was wondering if you have done or are interested in doing more widespread and systematic benchmarking - particularly with respect to how performance scales with variables like number of atoms for different basis sets on cpu and gpu.

I'm wondering how you expect performance to compare to a library like PySCF just knowing the architecture of dqc? I'm also curious if performance / scale we're things you were interested in or even motivated by when developing dqc too?

I havn't tried any comparisons yet and don't really know enough about the actual implementations to have any expectations one way or another. There is a little bit of PySCF benchmark data that could be compared to, or we could compute our own.

Ideally I'd like to push to as large systems as possible, but I am so new to this space that I'm really not sure what is possible. If I could do things on the scale of amino acids (~20 atoms) that would be a nice start - getting to ~100 atoms would be even better, and so it continues!

Describe the solution you'd like I could imagine developing a benchmarking suite of both basic calculations and calculations of properties (like IR spectra) and a set of molecules of increasing size and then measuring calculation time for a variety of different basis sets and hardware configurations, cpu/ gpu. The goal would be to assess performance as molecule size increased.

Describe alternatives you've considered I could do ad-hoc testing using basic timing functionality to try and build up an intuitive feel for scaling performance

Additional context If this was something you were interested in I'd appreciate your help in designing the best approach to benchmarking as I am so new to this space.

If benchmarking is successful one can then start doing profiling to try and identify performance gaps and ultimately work on improving performance

opened by sofroniewn 4

Differentiable Quantum Chemistry (only Differentiable Density Functional Theory and Hartree Fock at the moment)

Related tags

Overview

DQC: Differentiable Quantum Chemistry

Applications

Citations

Comments

[BUG] different density matrix and Basis by using pscf and dqc

[ENH] Geometry optimization

[BUG] calculate density matrix

Add optimal_geometry method

[BUG] Big difference in PySCF vs dqc vibrational frequencies

[BUG] Differentiable tensor atom position error if molecule initialized from string

[BUG] Unable to compute ir_spectrum

[BUG] Customized meta-GGA functional

[BUG] Wrong number of vibrational frequencies

[ENH] GPU support

[BUG] Bug in `optimal_geometry` for KS case

Add initial asv benchmarks

[ENH] Performance/ Benchmarking/ Scaling

Owner

A library for using chemistry in your applications

UMT is a unified and flexible framework which can handle different input modality combinations, and output video moment retrieval and/or highlight detection results.

Moment-DETR code and QVHighlights dataset

CVPR '21: In the light of feature distributions: Moment matching for Neural Style Transfer

Automatic Differentiation Multipole Moment Molecular Forcefield

[2021 MultiMedia] CONQUER: Contextual Query-aware Ranking for Video Corpus Moment Retrieval

Differentiable Neural Computers, Sparse Access Memory and Sparse Differentiable Neural Computers, for Pytorch

This YoloV5 based model is fit to detect people and different types of land vehicles, and displaying their density on a fitted map, according to their coordinates and detected labels.

Numenta Platform for Intelligent Computing is an implementation of Hierarchical Temporal Memory (HTM), a theory of intelligence based strictly on the neuroscience of the neocortex.

Numenta Platform for Intelligent Computing is an implementation of Hierarchical Temporal Memory (HTM), a theory of intelligence based strictly on the neuroscience of the neocortex.

This repository builds a basic vision transformer from scratch so that one beginner can understand the theory of vision transformer.

Drone-based Joint Density Map Estimation, Localization and Tracking with Space-Time Multi-Scale Attention Network

Details about the wide minima density hypothesis and metrics to compute width of a minima

Estimation of human density in a closed space using deep learning.

SMD-Nets: Stereo Mixture Density Networks

PyTorch implementations of algorithms for density estimation

MADE (Masked Autoencoder Density Estimation) implementation in PyTorch

This program presents convolutional kernel density estimation, a method used to detect intercritical epilpetic spikes (IEDs)

Official code of the paper "Expanding Low-Density Latent Regions for Open-Set Object Detection" (CVPR 2022)