Some beginner problems

[steto123]

After succesfully installing cuda, gpu4pscf und cutensor in my jupyter environment on my university hpc cluster i have problems running examples:

/home/sthomas/.conda/envs/myjupyter/lib/python3.10/site-packages/cupy/_environment.py:370: UserWarning: CuPy failed to preload library (/home/sthomas/.conda/envs/myjupyter/lib/python3.10/site-packages/cutensor/lib/libcutensor.so.2): OSError (libcublasLt.so.12: cannot open shared object file: No such file or directory)
warnings.warn(msg)
/home/sthomas/.conda/envs/myjupyter/lib/python3.10/site-packages/gpu4pyscf/lib/cutensor.py:138: UserWarning: using cupy as the tensor contraction engine.
warnings.warn(f'using {contract_engine} as the tensor contraction engine.')

NVRTCError Traceback (most recent call last)
File ~/.conda/envs/myjupyter/lib/python3.10/site-packages/cupy/cuda/compiler.py:692, in _NVRTCProgram.compile(self, options, log_stream)
691 nvrtc.addNameExpression(self.ptr, ker)
--> 692 nvrtc.compileProgram(self.ptr, options)
693 mapping = None

File cupy_backends/cuda/libs/nvrtc.pyx:125, in cupy_backends.cuda.libs.nvrtc.compileProgram()

File cupy_backends/cuda/libs/nvrtc.pyx:138, in cupy_backends.cuda.libs.nvrtc.compileProgram()

File cupy_backends/cuda/libs/nvrtc.pyx:53, in cupy_backends.cuda.libs.nvrtc.check_status()

NVRTCError: NVRTC_ERROR_COMPILATION (6)

During handling of the above exception, another exception occurred:

CompileException Traceback (most recent call last)

... (the full error listing is attached)

pip show cutensor-cu12 shows only one version of cutensor installed

---

alternatively i have started compiling gpu4pyscf.

After solving some little problems compiling stops:

[ 63%] Completed 'libxc'
[ 66%] Built target libxc
gmake: *** [Makefile:91: all] Error 2

Does anyone have some good ideas for me?

Thank you

errors.txt

[wxj6000]

It seems that nvrtc is missing in your CUDA environment. You can try to add it when you install cuda-toolkit (it is a part of cuda-toolkit).

[steto123]

Many thanks for your help.

I was able to use the precompiled cuda11x versions in a clean environment on our HPC Cluster without greater problems.

Many thanks

And another question:

Is there an example for calculating shieldings? I found test_shielding.py but i do not have an ide how to integrate this in a optimisation run. (for pyscf i found a simple example and i will inspect this in the next days more intense)

[sunqm]

In current version, you can run

from gpu4pyscf.properties.shielding import eval_shielding
mf = mol.RKS().to_gpu().run()
eval_shielding(mf)

[steto123]

Many thanks for your time to answer me.
I have used this in the following example:

`
xyzin="propanol-g_opt1.xyz" # funktioniert
logfile="test_g.log"

import gpu4pyscf
from gpu4pyscf.properties.shielding import eval_shielding
mol = pyscf.M(atom=xyzin, basis='6-31g*', output=logfile)

#mf = dft.RKS(mol)
#mf.xc = 'b3lyp'

mf = mol.RKS().to_gpu().run()
eval_shielding(mf)

`

i have uncommented the two lines alone and both at the same time. In all cases the error was the following:

---

ValueError Traceback (most recent call last)
Cell In[8], line 21
17 mf.xc = 'b3lyp'
20 mf = mol.RKS().to_gpu().run()
---> 21 eval_shielding(mf)

File ~/miniconda3/lib/python3.12/site-packages/gpu4pyscf/properties/shielding.py:205, in eval_shielding(mf)
202 dm0 = mf.make_rdm1()
203 natom = mf.mol.natm
--> 205 h1ao = get_h1ao(mf)
206 tmp = contract('xuv,ua->xav', s1ao, mvir)
207 s1ai = contract('xav,vi->xai', tmp, mocc)

File ~/miniconda3/lib/python3.12/site-packages/gpu4pyscf/properties/shielding.py:180, in get_h1ao(mf)
178 h1ao += -mf.mol.intor('int1e_ignuc')
179 h1ao = cupy.array(h1ao)
--> 180 h1ao += get_vxc(mf, dm0)
182 return h1ao

File ~/miniconda3/lib/python3.12/site-packages/gpu4pyscf/properties/shielding.py:160, in get_vxc(mf, dm0)
159 def get_vxc(mf, dm0):
--> 160 vxc = nr_rks(mf._numint, mf.mol, mf.grids, mf.xc, mf.make_rdm1())
161 # ! imaginary part
162 vj, vk = get_jk(mf.mol, dm0)

File ~/miniconda3/lib/python3.12/site-packages/gpu4pyscf/properties/shielding.py:100, in nr_rks(ni, mol, grids, xc_code, dms)
98 giao_aux = giao[:,:,index]
99 for idirect in range(3):
--> 100 vtmp = contract('pu,p,vp->uv', giao_aux[idirect], wv, ao)
101 vtmp = cupy.ascontiguousarray(vtmp)
102 add_sparse(vmat[idirect], vtmp, index)

File ~/miniconda3/lib/python3.12/site-packages/gpu4pyscf/lib/cutensor.py:151, in contract(pattern, a, b, alpha, beta, out)
146 def contract(pattern, a, b, alpha=1.0, beta=0.0, out=None):
147 '''
148 a wrapper for general tensor contraction
149 pattern has to be a standard einsum notation
150 '''
--> 151 return contraction(pattern, a, b, alpha, beta, out=out)

File ~/miniconda3/lib/python3.12/site-packages/gpu4pyscf/lib/cutensor.py:73, in contraction(pattern, a, b, alpha, beta, out, op_a, op_b, op_c, algo, jit_mode, compute_desc, ws_pref)
60 def contraction(
61 pattern, a, b, alpha, beta,
62 out=None,
(...)
69 ws_pref=WORKSPACE_RECOMMENDED
70 ):
72 pattern = pattern.replace(" ", "")
---> 73 str_a, rest = pattern.split(',')
74 str_b, str_c = rest.split('->')
75 key = str_a + str_b

ValueError: too many values to unpack (expected 2)

This makes me unhappy. Can you help me again?

Many thanks for your work and help.

[wxj6000]

This is a bug when you use LDA functional for NMR shielding calculation. It has been fixed in #256

Just a reminder. These modules (Polarizability, IR, and NMR shielding) are still experimental. They probably are not as efficient as SCF module.

[steto123]

Many thanks for your help and this reminder. I have editited shielding.py in the same way and now it works.

Steffen