dft_zmp/n2v.patched/inverter.py
2024-02-26 10:44:51 +01:00

542 lines
21 KiB
Python
Executable file
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

"""
Inverter.py
"""
from warnings import warn
from dataclasses import dataclass
import numpy as np
from opt_einsum import contract
from .methods.zmp import ZMP
from .methods.wuyang import WuYang
from .methods.pdeco import PDECO
from .methods.oucarter import OC
from .methods.mrks import MRKS
from .methods.direct import Direct
#Grider was imported by Ehsan
from .grid.grider import Grider
@dataclass
class V:
"""Stores Potentials on AO"""
T : np.ndarray
class E:
"""Stores Energies"""
# Grider was added by Ehsan
class Inverter(Direct, ZMP, WuYang, PDECO, OC, MRKS, Grider):
"""
Attributes:
----------
mol : Engine.molecule
Molecule class of engine used
basis : Engine.basis
Basis class of engine used
basis_str : str
Basis set
nbf : int
Number of basis functions for main calculation
nalpha : int
Number of alpha electrons
nbeta : int
Number of beta electrons
ref : {1,2}
Reference calculation
1 -> Restricted
2 -> Unrestricted
Dt : List
List of np.ndarray for target density matrices (on AO).
ct : List
List of np.ndarray for input occupied orbitals. This might not be correct for post-HartreeFock methods.
pbs_str: string
name of Potential basis set
pbs : Engine.basis
Basis class for Potential basis set of the engine used.
npbs : int
the length of pbs
v_pbs : np.ndarray shape (npbs, ) for ref==1 and (2*npbs, ) for ref==2.
potential vector on the Potential Baiss Set.
If the potential is not represented on the basis set, this should
remain 0. It will be initialized to a 0 array. One can set this
value for initial guesses before Wu-Yang method (WY) or PDE-Constrained
Optimization method (PDE-CO). For example, if PDE-CO is ran after
a WY calculation, the initial for PDE-CO will be the result of WY
if v_pbs is not zeroed.
S2 : np.ndarray
The ao overlap matrix (i.e. S matrix)
S3 : np.ndarray
The three ao overlap matrix (ao, ao, pbs)
S4 : np.ndarray
The four ao overlap matrix, the size should be (ao, ao, ao, ao)
jk : Engine.jk
Engine jk object.
T : np.ndarray
kinetic matrix on ao
V : np.ndarray
external potential matrix on ao
T_pbs: np.ndarray
kinetic matrix on pbs. Useful for regularization.
guide_potential_components: list of string
guide potential components name
va, vb: np.ndarray of shape (nbasis, nbasis)
guide potential Fock matrix.
"""
def __init__( self, engine='psi4' ):
engine = 'psi4'
self.eng_str = engine.lower()
if engine.lower() == 'psi4':
from .engines import Psi4Engine
self.eng = Psi4Engine()
elif engine.lower() == 'pyscf':
from .engines import PySCFEngine
self.eng = PySCFEngine()
else:
raise ValueError("Engine name is incorrect. The availiable engines are: {psi4, pyscf}")
def __repr__( self ):
return "n2v.Inverter"
def set_system( self, molecule, basis, ref=1, pbs='same' , **kwargs):
"""
Stores relevant information and intitializes Engine
Parameters
----------
molecule: Engine.molecule
Molecule object of selected engine
basis: str
Basis set of the main calculation
ref: int
reference for system. Restricted -> 1
Unrestricted -> 2
pbs: str, default='same'
Basis set for the potential
**kwargs:
Optional Parameters for different Engiens
Psi4 Engine:
wfn : psi4.core.{RHF, UHF, RKS, UKS, Wavefunction, CCWavefuncion...}
Psi4 wavefunction object
PySCF Engine:
None
"""
# Communicate TO engine
self.eng.set_system(molecule, basis, ref, pbs, **kwargs)
self.ref = ref
#added by Ehsan
self.basis = basis
#added by Ehsan
self.molecule = molecule
self.nalpha = self.eng.nalpha
self.nbeta = self.eng.nbeta
# Initialize ecompasses everything the engine builds with basis set
self.eng.initialize()
self.set_basis_matrices()
# Receive FROM engine
self.nbf = self.eng.nbf
self.npbs = self.eng.npbs
self.v_pbs = np.zeros( (self.npbs) ) if self.ref == 1 \
else np.zeros( 2 * self.npbs )
@classmethod
def from_wfn( self, wfn, pbs='same' ):
"""
Generates Inverter directly from wavefunction.
Parameters
----------
wfn: Psi4.Core.{RHF, RKS, ROHF, CCWavefunction, UHF, UKS, CUHF}
Wavefunction Object
Returns
-------
inv: n2v.Inverter
Inverter Object.
"""
from .engines import Psi4Engine
inv = self( engine='psi4' )
inv.eng = Psi4Engine()
ref = 1 if wfn.to_file()['boolean']['same_a_b_dens'] else 2
inv.set_system( wfn.molecule(), wfn.basisset().name(), pbs=pbs, ref=ref, wfn=wfn )
# done by Ehsan
#inv.Dt = [ np.array(wfn.Da()), np.array(wfn.Db()) ]
self.Dt = [ np.array(wfn.Da()), np.array(wfn.Db()) ]
# done by Ehsan
#inv.ct = [ np.array(wfn.Ca_subset("AO", "OCC")), np.array(wfn.Cb_subset("AO", "OCC")) ]
# ct contains matrices of occupied orbitals alpah and betta (n x m)
self.ct = [ np.array(wfn.Ca_subset("AO", "OCC")), np.array(wfn.Cb_subset("AO", "OCC")) ]
inv.et = [ np.array(wfn.epsilon_a_subset("AO", "OCC")), np.array(wfn.epsilon_b_subset("AO", "OCC")) ]
inv.eng_str = 'psi4'
inv.eng.wfn = wfn
return inv
def set_basis_matrices( self ):
"""
Generate basis dependant matrices
"""
self.T = self.eng.get_T()
self.V = self.eng.get_V()
self.A = self.eng.get_A()
self.S2 = self.eng.get_S()
self.S3 = self.eng.get_S3()
if self.eng.pbs_str != 'same':
self.T_pbs = self.eng.get_Tpbas()
self.S4 = None
def compute_hartree( self, Cocc_a, Cocc_b ):
"""
Computes Hartree Potential on AO basis set.
Parameters
----------
Cocc_a, Cocc_b: np.ndarray (nbf, nbf)
Occupied orbitals in ao basis
Returns
-------
J: List of np.ndarray
Hartree potential due to density from Cocc_a and Cocc_b
"""
return self.eng.compute_hartree(Cocc_a, Cocc_b )
def diagonalize( self, matrix, ndocc ):
"""
Diagonalizes Fock Matrix
Parameters
----------
marrix: np.ndarray
Matrix to be diagonalized
ndocc: int
Number of occupied orbitals
Returns
-------
C: np.ndarray
Orbital Matrix
Cocc: np.ndarray
Occupied Orbital Matrix
D: np.ndarray
Density Matrix
eigves: np.ndarray
Eigenvalues
"""
# np.linalg.eigh() gives eigenvalues and eigenvectors for a symmetric matrix of choice
Fp = self.A.dot(matrix).dot(self.A)
# eigvecs must be eigenvalues or energies here!
eigvecs, Cp = np.linalg.eigh(Fp)
C = self.A.dot(Cp)
Cocc = C[:, :ndocc]
# contract converts pi and qi to pq . here two matrices with n x m dimension
#are converted to one matrix with n x n shape,
#In fact it gives the product of Cocc matrix and its transpose matrix
D = contract('pi,qi->pq', Cocc, Cocc)
return C, Cocc, D, eigvecs
def diagonalize_with_potential_vFock(self, v=None):
"""
Diagonalize Fock matrix with additional external potential
Stores values in object.
Parameters
----------
v: np.ndarray
Additional external potential to be added to hamiltonian along with:
Kinetic_nm
External_nm
Guide_Potential_nm
"""
if v is None:
fock_a = self.V + self.T + self.va
else:
if self.ref == 1:
fock_a = self.V + self.T + self.va + v
else:
valpha, vbeta = v
fock_a = self.V + self.T + self.va + valpha
fock_b = self.V + self.T + self.vb + vbeta
self.Ca, self.Coca, self.Da, self.eigvecs_a = self.diagonalize( fock_a, self.nalpha )
if self.ref == 1:
self.Cb, self.Cocb, self.Db, self.eigvecs_b = self.Ca.copy(), self.Coca.copy(), self.Da.copy(), self.eigvecs_a.copy()
else:
self.Cb, self.Cocb, self.Db, self.eigvecs_b = self.diagonalize( fock_b, self.nbeta )
# Actual Methods
def generate_components(self, guide_components, **keywords):
"""
Generates exact potential components to be added to
the Hamiltonian to aide in the inversion procedure.
Parameters:
-----------
guide_potential_components: list
Components added as to guide inversion.
Can be chosen from ["hartree", "fermi_amandi", "svwn"]
"""
self.guide_components = guide_components
self.va = np.zeros( (self.nbf, self.nbf) )
self.vb = np.zeros( (self.nbf, self.nbf) )
self.J0 = self.compute_hartree(self.ct[0], self.ct[1])
N = self.nalpha + self.nbeta
if self.eng_str == 'psi4':
J0_NO = self.eng.hartree_NO(self.Dt[0])
self.J0 = J0_NO if J0_NO is not None else self.J0
if guide_components == 'none':
warn("No guide potential was provided. Convergence may not be achieved")
elif guide_components == 'hartree':
self.va += self.J0[0] + self.J0[1]
self.vb += self.J0[0] + self.J0[1]
elif guide_components == 'fermi_amaldi':
v_fa = (1-1/N) * (self.J0[0] + self.J0[1])
self.va += v_fa
self.vb += v_fa
else:
raise ValueError("Guide component not recognized")
def invert(self, method,
guide_components = 'hartree',
opt_max_iter = 50,
**keywords):
"""
Handler to all available inversion methods
Parameters
----------
method: str
Method used to invert density.
Can be chosen from {wuyang, zmp, mrks, oc}.
See documentation below for each method.
guide_components: list, opt
Components added as to guide inversion.
Can be chosen from {"fermi_amandi", "svwn"}
Default: ["fermi_amaldi"]
opt_max_iter: int, opt
Maximum number of iterations inside the chosen inversion.
Default: 50
direct
------
Direct inversion of a set of Kohn-Sham equations.
$$v_{xc}(r) = \frac{1}{n(r)} \sum_i^N [\phi_i^{*} (r) \nabla^2 \phi_i(r) + \varepsilon_i | \phi_i(r)|^2] $$
Parameters:
-----------
grid: np.ndarray, opt
Grid where result will be expressed in.
If not provided, dft grid will be used instead.
wuyang
------
the Wu-Yang method:
The Journal of chemical physics 118.6 (2003): 2498-2509.
Parameters:
----------
opt_max_iter: int
maximum iteration
opt_method: string, opt
Method for scipy optimizer
Currently only used by wuyang and pdeco method.
Defaul: 'trust-krylov'
https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.minimize.html
reg : float, opt
Regularization constant for Wuyant Inversion.
Default: None -> No regularization is added.
Becomes attribute of inverter -> inverter.lambda_reg
tol: float
tol for scipy.optimize.minimize
gtol: float
gtol for scipy.optimize.minimize: the gradient norm for
convergence
opt: dict
options for scipy.optimize.minimize
Notice that opt has lower priorities than opt_max_iter and gtol.
return:
the result are stored in self.v_pbs
zmp
---
The Zhao-Morrison-Parr Method:
Phys. Rev. A 50, 2138
Parameters:
----------
lambda_list: list
List of Lamda parameters used as a coefficient for Hartree
difference in SCF cycle.
zmp_mixing: float, optional
mixing \in [0,1]. How much of the new potential is added in.
For example, zmp_mixing = 0 means the traditional ZMP, i.e. all the potentials from previous
smaller lambda are ignored.
Zmp_mixing = 1 means that all the potentials of previous lambdas are accumulated, the larger lambda
potential are meant to fix the wrong/inaccurate region of the potential of the sum of the previous
potentials instead of providing an entire new potentials.
default: 1
opt_max_iter: float
Maximum number of iterations for scf cycle
opt_tol: float
Convergence criteria set for Density Difference and DIIS error.
return:
The result will be stored in self.proto_density_a and self.proto_density_b
For zmp_mixing==1, restricted (ref==1):
self.proto_density_a = \sum_i lambda_i * (Da_i - Dt[0]) - 1/N * (Dt[0])
self.proto_density_b = \sum_i lambda_i * (Db_i - Dt[1]) - 1/N * (Dt[1]);
unrestricted (ref==1):
self.proto_density_a = \sum_i lambda_i * (Da_i - Dt[0]) - 1/N * (Dt[0] + Dt[1])
self.proto_density_b = \sum_i lambda_i * (Db_i - Dt[1]) - 1/N * (Dt[0] + Dt[1]);
For restricted (ref==1):
vxc = \int dr' \frac{self.proto_density_a + self.proto_density_b}{|r-r'|}
= 2 * \int dr' \frac{self.proto_density_a}{|r-r'|};
for unrestricted (ref==2):
vxc_up = \int dr' \frac{self.proto_density_a}{|r-r'|}
vxc_down = \int dr' \frac{self.proto_density_b}{|r-r'|}.
To get potential on grid, one needs to do
vxc = self.on_grid_esp(Da=self.proto_density_a, Db=self.proto_density_b, grid=grid) for restricted;
vxc_up = self.on_grid_esp(Da=self.proto_density_a, Db=np.zeros_like(self.proto_density_a),
grid=grid) for unrestricted;
mRKS
----
the modified Ryabinkin-Kohut-Staroverov method:
Phys. Rev. Lett. 115, 083001
J. Chem. Phys. 146, 084103p
Parameters:
-----------
maxiter: int
same as opt_max_iter
vxc_grid: np.ndarray of shape (3, num_grid_points), opt
When this is given, the final result will be represented
v_tol: float, opt
convergence criteria for vxc Fock matrices.
default: 1e-4
D_tol: float, opt
convergence criteria for density matrices.
default: 1e-7
eig_tol: float, opt
convergence criteria for occupied eigenvalue spectrum.
default: 1e-4
frac_old: float, opt
Linear mixing parameter for current vxc and old vxc.
If 0, no old vxc is mixed in.
Should be in [0,1)
default: 0.5.
init: string or psi4.core.Wavefunction, opt
Initial guess method.
default: "SCAN"
1) If None, input wfn info will be used as initial guess.
2) If "continue" is given, then it will not initialize
but use the densities and orbitals stored. Meaningly,
one can run a quick WY calculation as the initial
guess. This can also be used to user speficified
initial guess by setting Da, Coca, eigvec_a.
3) If it's not continue, it would be expecting a
method name string that works for psi4. A separate psi4 calculation
would be performed.
sing: tuple of float of length 4, opt.
Singularity parameter for _vxc_hole_quadrature()
default: (1e-5, 1e-4, 1e-5, 1e-4)
[0]: atol, [1]: atol1 for dft_spherical grid calculation.
[2]: atol, [3]: atol1 for vxc_grid calculation.
return:
The result will be stored in self.grid.vxc
oc
--
Ou-Carter method
J. Chem. Theory Comput. 2018, 14, 56805689
Parameters:
-----------
maxiter: int
same as opt_max_iter
vxc_grid: np.ndarray of shape (3, num_grid_points)
The final result will be represented on this grid
default: 1e-4
D_tol: float, opt
convergence criteria for density matrices.
default: 1e-7
eig_tol: float, opt
convergence criteria for occupied eigenvalue spectrum.
default: 1e-4
frac_old: float, opt
Linear mixing parameter for current vxc and old vxc.
If 0, no old vxc is mixed in.
Should be in [0,1)
default: 0.5.
init: string, opt
Initial guess method.
default: "SCAN"
1) If None, input wfn info will be used as initial guess.
2) If "continue" is given, then it will not initialize
but use the densities and orbitals stored. Meaningly,
one can run a quick WY calculation as the initial
guess. This can also be used to user speficified
initial guess by setting Da, Coca, eigvec_a.
3) If it's not continue, it would be expecting a
method name string that works for psi4. A separate psi4 calculation
would be performed.
wuyang
pdeco
-----
the PDE-Constrained Optimization method:
Int J Quantum Chem. 2018;118:e25425;
Nat Commun 10, 4497 (2019).
Parameters:
----------
opt_max_iter: int
maximum iteration
opt_method: string, opt
Method for scipy optimizer
Currently only used by wuyang and pdeco method.
Defaul: 'L-BFGS-B'
Options: ['L-BFGS-B', 'BFGS']
https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.minimize.html
reg : float, opt
Regularization constant for Wuyant Inversion.
Default: None -> No regularization is added.
Becomes attribute of inverter -> inverter.lambda_reg
gtol: float
gtol for scipy.optimize.minimize: the gradient norm for
convergence
opt: dict
options for scipy.optimize.minimize
Notice that opt has lower priorities than opt_max_iter and gtol.
return:
the result are stored in self.v_pbs
"""
self.generate_components(guide_components)
if method.lower() == "direct":
return self.direct_inversion(**keywords)
elif method.lower() == "wuyang":
self.wuyang(opt_max_iter, **keywords)
elif method.lower() == "zmp":
self.zmp(opt_max_iter, **keywords)
elif method.lower() == "mrks":
if self.eng_str == 'pyscf':
raise ValueError("mRKS method not yet available with the PySCF engine. Try another method or another engine.")
return self.mRKS(opt_max_iter, **keywords)
elif method.lower() == 'oc':
if self.eng_str == 'pyscf':
raise ValueError("OuCarter method not yet available with the PySCF engine. Try another method or another engine.")
return self.oucarter(opt_max_iter, **keywords)
elif method.lower() == 'pdeco':
return self.pdeco(opt_max_iter, **keywords)
else:
raise ValueError(f"Inversion method not available. Methods available: {['wuyang', 'zmp', 'mrks', 'oc', 'pdeco']}")