Error: cannot reshape array size ...

[steto123]

Dear devs.

I use a simple script for calculating nmr shieldings:

xyzin="propanol-g_opt1.xyz" # running

xyzin ="clprop1-g_opt1.xyz"
logfile="nmrtest2.log"

import pyscf

from pyscf import gto, dft

from gpu4pyscf.dft import rks
from pyscf.prop import nmr

mol = pyscf.M(atom=xyzin, basis='6-31g*', output=logfile)

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

nmr.RKS(mf).kernel()

Using the optimized geometry for 2-chloropropen fails in this calculation (with and without gpu support):

---

ValueError Traceback (most recent call last)
Cell In[9], line 24
21 mf.xc = 'b3lyp'
22 mf.run()
---> 24 nmr.RKS(mf).kernel()

File ~/.conda/envs/mycuda11/lib/python3.12/site-packages/pyscf/prop/nmr/rhf.py:287, in NMR.kernel(self, mo1)
286 def kernel(self, mo1=None):
--> 287 return self.shielding(mo1)

File ~/.conda/envs/mycuda11/lib/python3.12/site-packages/pyscf/prop/nmr/rhf.py:297, in NMR.shielding(self, mo1)
294 msc_dia = self.dia(self.gauge_orig)
296 if mo1 is None:
--> 297 self.mo10, self.mo_e10 = self.solve_mo1()
298 mo1 = self.mo10
299 msc_para, para_vir, para_occ = self.para(mo10=mo1)

File ~/.conda/envs/mycuda11/lib/python3.12/site-packages/pyscf/prop/nmr/rks.py:151, in solve_mo1(nmrobj, mo_energy, mo_coeff, mo_occ, h1, s1, with_cphf)
149 libxc = nmrobj._scf._numint.libxc
150 with_cphf = with_cphf and libxc.is_hybrid_xc(nmrobj._scf.xc)
--> 151 return rhf_nmr.solve_mo1(nmrobj, mo_energy, mo_coeff, mo_occ,
152 h1, s1, with_cphf)

File ~/.conda/envs/mycuda11/lib/python3.12/site-packages/pyscf/prop/nmr/rhf.py:209, in solve_mo1(nmrobj, mo_energy, mo_coeff, mo_occ, h1, s1, with_cphf)
207 else:
208 vind = gen_vind(nmrobj._scf, mo_coeff, mo_occ)
--> 209 mo10, mo_e10 = cphf.solve(vind, mo_energy, mo_occ, h1, s1,
210 nmrobj.max_cycle_cphf, nmrobj.conv_tol,
211 verbose=log)
212 else:
213 mo10, mo_e10 = _solve_mo1_uncoupled(mo_energy, mo_occ, h1, s1)

File ~/.conda/envs/mycuda11/lib/python3.12/site-packages/pyscf/scf/cphf.py:48, in solve(fvind, mo_energy, mo_occ, h1, s1, max_cycle, tol, hermi, verbose, level_shift)
45 return solve_nos1(fvind, mo_energy, mo_occ, h1,
46 max_cycle, tol, hermi, verbose, level_shift)
47 else:
---> 48 return solve_withs1(fvind, mo_energy, mo_occ, h1, s1,
49 max_cycle, tol, hermi, verbose, level_shift)

File ~/.conda/envs/mycuda11/lib/python3.12/site-packages/pyscf/scf/cphf.py:130, in solve_withs1(fvind, mo_energy, mo_occ, h1, s1, max_cycle, tol, hermi, verbose, level_shift)
128 v[:,occidx,:] = 0
129 return v.reshape(-1, nmonocc)
--> 130 mo1 = lib.krylov(vind_vo, mo1base.reshape(-1, nmonocc),
131 tol=tol, max_cycle=max_cycle, hermi=hermi, verbose=log)
132 mo1 = mo1.reshape(-1, nmo, nocc)
133 mo1[:,occidx] = mo1base[:,occidx]

File ~/.conda/envs/mycuda11/lib/python3.12/site-packages/pyscf/lib/linalg_helper.py:1307, in krylov(aop, b, x0, tol, max_cycle, dot, lindep, callback, hermi, max_memory, verbose)
1305 max_cycle = min(max_cycle, ndim)
1306 for cycle in range(max_cycle):
-> 1307 axt = aop(x1)
1308 if axt.ndim == 1:
1309 axt = axt.reshape(1,ndim)

File ~/.conda/envs/mycuda11/lib/python3.12/site-packages/pyscf/scf/cphf.py:124, in solve_withs1..vind_vo(mo1)
122 def vind_vo(mo1):
123 mo1 = mo1.reshape(-1, nmo, nocc)
--> 124 v = fvind(mo1).reshape(-1, nmo, nocc)
125 if level_shift != 0:
126 v -= mo1 * level_shift

File ~/.conda/envs/mycuda11/lib/python3.12/site-packages/pyscf/prop/nmr/rhf.py:241, in gen_vind..vind(mo1)
239 def vind(mo1):
240 dm1 = [reduce(numpy.dot, (mo_coeff, x*2, orbo.T.conj()))
--> 241 for x in mo1.reshape(3,nmo,nocc)]
242 dm1 = numpy.asarray([d1-d1.conj().T for d1 in dm1])
243 v1mo = lib.einsum('xpq,pi,qj->xij', vresp(dm1), mo_coeff.conj(), orbo)

ValueError: cannot reshape array of size 2800 into shape (3,70,20)

The same calculation for 1-propanol is finished without any errors.

Can anybody help?

[fishjojo]

It seems like you are using an earlier version of the code. This issue should be fixed in commit 4eee5a4.

[steto123]

Many thanks for your fast answer. I have installed pyscf and gpu4pyscf two days before in a fresh conda environment. This environment uses cuda11.
Using a other conda environment after (!) deinstalling gpu4pyscf and cutensor, the following code works:

`
xyzin ="clprop1-g_opt1.xyz"
logfile="clprop1_g.log"

import pyscf
from pyscf import gto, dft, prop
from pyscf.prop import nmr

mol = pyscf.M(atom=xyzin, basis='6-31g*', output=logfile)

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

nmr.RKS(mf).kernel()

`