SF-NOCI #126
SF-NOCI #126
Conversation
pyscf/sfnoci/direct_sfnoci.py
Outdated
| @@ -0,0 +1,702 @@ | |||
| #!/usr/bin/env python | |||
| # | |||
| # Copyright 2024 The PySCF Developers. All Rights Reserved. | |||
There was a problem hiding this comment.
Change copyright to 2025.
pyscf/sfnoci/sfnoci.py
Outdated
| @@ -0,0 +1,1270 @@ | |||
| #!/usr/bin/env python | |||
| # | |||
| # Copyright 2024 The PySCF Developers. All Rights Reserved. | |||
There was a problem hiding this comment.
Change copyright to 2025.
|
The failed CI unit tests are not your fault. Sorry, I keep overestimating the numerical precision of MC-PDFT on these runners. But please do fix the issues raised by the Lint. Re unit tests and examples, I notice you already have an example calculation in a |
|
|
||
| def tearDownModule(): | ||
| global mol, mo0, setocc, ro_occ | ||
| mol.stdout.close() |
There was a problem hiding this comment.
Don't close mol.stdout here. This object points to the sys.stdout
| @@ -138,3 +138,12 @@ set_target_properties (clib_dsrg PROPERTIES | |||
| CLEAN_DIRECT_OUTPUT 1 | |||
| LIBRARY_OUTPUT_DIRECTORY ${PROJECT_SOURCE_DIR} | |||
| OUTPUT_NAME "dsrg") | |||
|
|
|||
There was a problem hiding this comment.
Move these settings in a separated "CMakeLists.txt" file under the sfnoci directory. Then include this directory in this file
| + aa * ncas * ncas * ncas | ||
| + ia * ncas *ncas | ||
| + ab * ncas + ib] | ||
| * t1a[aa * ncas * na * na + ia * na * na + str1a * na + str0a] |
There was a problem hiding this comment.
aa*ncas*na*na and ia*na*na may cause integer overflow. Make a conversion such as (size_t)na for na
| + ia * ncas *ncas | ||
| + ab * ncas + ib] | ||
| * t1a[aa * ncas * na * na + ia * na * na + str1a * na + str0a] | ||
| * t1b[ab * ncas * nb * nb + ib * nb * nb + str1b * nb + str0b] |
| + i1 * ncas * ncas | ||
| + a2 * ncas | ||
| + i2] | ||
| * t2aa[a1 * ncas * ncas * ncas * na * na + i1* ncas * ncas * na * na + a2 * ncas* na * na + i2* na * na + str1a * na + str0a] |
| + i1 * ncas * ncas | ||
| + a2 * ncas | ||
| + i2] | ||
| * t2bb[a1 * ncas * ncas * ncas * nb * nb + i1* ncas * ncas * nb * nb + a2 * ncas* nb * nb + i2* nb * nb + str1b * nb + str0b] |
| def python_list_to_c_array(python_list): | ||
| if python_list is None: return ctypes.c_void_p(None), ctypes.c_void_p(None), 0 | ||
| else: | ||
| num_groups = len(python_list) |
There was a problem hiding this comment.
I'd recommend to change this to np.ndarray. E.g.
python_list = [np.ndarray(x) for x in python_list]
flat_list = np.asarray(np.hstack(pythn_list), dtype=np.int32)
group_sizes = np.asarray([x.size for x in python_list], dtype=np.int32)
Then passing the .ctypes attributes of flat_list and group_sizes to the C functions
| arrays.append(numpy.array(combination)) | ||
| return arrays | ||
| result_arrays=find_arrays(n_as,n_ase) | ||
| array_list=[] |
There was a problem hiding this comment.
why not directly calling
concatenated_array = np.array(result_arrays, dtype=np.int32) ?
| return concantenated_array | ||
|
|
||
| def fill_array_with_sum_N(length, N): | ||
| result = [0] * length |
There was a problem hiding this comment.
result = np.zeros(length, dtype=int)
...
result[:N] = 1
else:
result[:num_twos] = 2
result[num_twos:] = 1
This PR adds SF-NOCI, which was originally introduced by Mayhall and co-workers:
And this PR is particularly based on a recent extension:
for an efficient grouped bath ansatz.
We plan to add the following improvements based on this PR.
add examples and more unit tests (@jiseong0626)
add analytic energy gradient and X-ray spectrum calculation (@jiseong0626)
add DFT correction based on MC-PDFT (@BKim0828)