14992701623

Committed 13 May 2025 09:07AM UTC coverage: 83.844%. First build

Build # 14992701623

Build Type

Pull #187

github

Committed by

web-flow

Commit Message

Merge e73393a0b into 20fe7ebf9

Pull Request Pull Request #187: Clean up Main

Run Details

955 of 1334 branches covered (71.59%)

Branch coverage included in aggregate %.

293 of 372 new or added lines in 47 files covered. (78.76%)

4515 of 5190 relevant lines covered (86.99%)

0.87 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

82.35

/qmctorch/wavefunction/orbitals/backflow/kernels/backflow_kernel_base.py

import torch
from torch import nn
from .....scf import Molecule
from .....utils import gradients, hessian

class BackFlowKernelBase(nn.Module):
    def __init__(self, mol: Molecule, cuda: bool):
        """Compute the back flow kernel, i.e. the function
        f(rij) where rij is the distance between electron i and j
        This kernel is used in the backflow transformation
        .. math:
            q_i = r_i + \\sum_{j\\neq i} f(r_{ij}) (r_i-r_j)
        """
        super().__init__()
        self.nelec = mol.nelec
        self.cuda = cuda
        self.device = torch.device("cpu")
        if self.cuda:
            self.device = torch.device("cuda")

    def forward(self, ree: torch.Tensor, derivative: int = 0) -> torch.Tensor:
        """Computes the desired values of the kernel
         Args:
            ree (torch.tensor): e-e distance Nbatch x Nelec x Nelec
            derivative (int): derivative requried 0, 1, 2

        Returns:
            torch.tensor : f(r) Nbatch x Nelec x Nelec
        """

        if derivative == 0:
            return self._backflow_kernel(ree)

        elif derivative == 1:
            return self._backflow_kernel_derivative(ree)

        elif derivative == 2:
            return self._backflow_kernel_second_derivative(ree)

        else:
            raise ValueError("derivative of the kernel must be 0, 1 or 2")

    def _backflow_kernel(self, ree: torch.Tensor) -> torch.Tensor:
        """Computes the kernel via autodiff

        Args:
            ree ([type]): [description]

        Returns:
            [type]: [description]
        """
        raise NotImplementedError("Please implement the backflow kernel")

    def _backflow_kernel_derivative(self, ree: torch.Tensor) -> torch.Tensor:
        """Computes the first derivative of the kernel via autodiff

        Args:
            ree ([type]): [description]

        Returns:
            [type]: [description]
        """
        if ree.requires_grad == False:
            ree.requires_grad = True

        with torch.enable_grad():
            kernel_val = self._backflow_kernel(ree)

        return gradients(kernel_val, ree)

    def _backflow_kernel_second_derivative(self, ree: torch.Tensor) -> torch.Tensor:
        """Computes the second derivative of the kernel via autodiff

        Args:
            ree ([type]): [description]

        Returns:
            [type]: [description]
        """
        if ree.requires_grad == False:
            ree.requires_grad = True

        with torch.enable_grad():
            kernel_val = self._backflow_kernel(ree)
            hess_val, _ = hessian(kernel_val, ree)

        # if the kernel is linear, hval is None
        if hess_val is None:
            hess_val = torch.zeros_like(ree)

        return hess_val

1	import torch	1✔
2	from torch import nn	1✔
3	from .....scf import Molecule	1✔
4	from .....utils import gradients, hessian	1✔
5
6	class BackFlowKernelBase(nn.Module):	1✔
7	def __init__(self, mol: Molecule, cuda: bool):	1✔
8	"""Compute the back flow kernel, i.e. the function
9	f(rij) where rij is the distance between electron i and j
10	This kernel is used in the backflow transformation
11	.. math:
12	q_i = r_i + \\sum_{j\\neq i} f(r_{ij}) (r_i-r_j)
13	"""
14	super().__init__()	1✔
15	self.nelec = mol.nelec	1✔
16	self.cuda = cuda	1✔
17	self.device = torch.device("cpu")	1✔
18	if self.cuda:	1!
19	self.device = torch.device("cuda")	×
20
21	def forward(self, ree: torch.Tensor, derivative: int = 0) -> torch.Tensor:	1✔
22	"""Computes the desired values of the kernel
23	Args:
24	ree (torch.tensor): e-e distance Nbatch x Nelec x Nelec
25	derivative (int): derivative requried 0, 1, 2
26
27	Returns:
28	torch.tensor : f(r) Nbatch x Nelec x Nelec
29	"""
30
31	if derivative == 0:	1✔
32	return self._backflow_kernel(ree)	1✔
33
34	elif derivative == 1:	1✔
35	return self._backflow_kernel_derivative(ree)	1✔
36
37	elif derivative == 2:	1!
38	return self._backflow_kernel_second_derivative(ree)	1✔
39
40	else:
41	raise ValueError("derivative of the kernel must be 0, 1 or 2")	×
42
43	def _backflow_kernel(self, ree: torch.Tensor) -> torch.Tensor:	1✔
44	"""Computes the kernel via autodiff
45
46	Args:
47	ree ([type]): [description]
48
49	Returns:
50	[type]: [description]
51	"""
52	raise NotImplementedError("Please implement the backflow kernel")	×
53
54	def _backflow_kernel_derivative(self, ree: torch.Tensor) -> torch.Tensor:	1✔
55	"""Computes the first derivative of the kernel via autodiff
56
57	Args:
58	ree ([type]): [description]
59
60	Returns:
61	[type]: [description]
62	"""
63	if ree.requires_grad == False:	1✔
64	ree.requires_grad = True	1✔
65
66	with torch.enable_grad():	1✔
67	kernel_val = self._backflow_kernel(ree)	1✔
68
69	return gradients(kernel_val, ree)	1✔
70
71	def _backflow_kernel_second_derivative(self, ree: torch.Tensor) -> torch.Tensor:	1✔
72	"""Computes the second derivative of the kernel via autodiff
73
74	Args:
75	ree ([type]): [description]
76
77	Returns:
78	[type]: [description]
79	"""
80	if ree.requires_grad == False:	1!
81	ree.requires_grad = True	×
82
83	with torch.enable_grad():	1✔
84	kernel_val = self._backflow_kernel(ree)	1✔
85	hess_val, _ = hessian(kernel_val, ree)	1✔
86
87	# if the kernel is linear, hval is None
88	if hess_val is None:	1!
NEW 89	hess_val = torch.zeros_like(ree)	×
90
91	return hess_val	1✔

NLESC-JCER / QMCTorch / 14992701623

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous