<div dir="ltr">The cause are the ludicrously diffuse functions on Na. If I remove the outermost S and P functions then I don't get the cholesky error. Whether there is any way of filtering the matrices to not do this manually I don't know.<div><br></div><div>You could also try the POB basis sets or others tweaked for solid state.<br><div><br></div><div><div>NA 6-311Gxx 6-311G**</div><div> 9</div><div> 1 0 0 8 2</div><div> 36166.40000000 0.00103200 0.00000000</div><div> 5372.58000000 0.00807100 0.00000000</div><div> 1213.21000000 0.04212900 0.00000000</div><div> 339.62300000 0.16978900 0.00000000</div><div> 109.55300000 0.51462100 0.00000000</div><div> 38.77730000 0.37981700 0.37476200</div><div> 14.57590000 0.00000000 0.57576900</div><div> 5.26993000 0.00000000 0.11293300</div><div> 1 0 0 1 1</div><div> 1.82777000 1.00000000</div><div> 1 0 0 1 1</div><div> 0.61994800 1.00000000</div><div> 1 0 0 1 1</div><div> 0.05724000 1.00000000</div><div> 1 1 1 4 1</div><div> 144.64500000 0.01148500</div><div> 33.90740000 0.08238300</div><div> 10.62850000 0.31965800</div><div> 3.82389000 0.70129500</div><div> 1 1 1 2 1</div><div> 1.44429000 0.63850600</div><div> 0.55262100 0.42536500</div><div> 1 1 1 1 1</div><div> 0.18872000 1.00000000</div><div> 1 1 1 1 1</div><div> 0.04650100 1.00000000</div><div> 1 2 2 1 1</div><div> 0.17500000 1.00000000</div><div><br></div><div>Matt</div><div><br></div>On Sunday, August 18, 2019 at 9:32:21 AM UTC+1, Daniele Passerone wrote:<blockquote class="gmail_quote" style="margin: 0;margin-left: 0.8ex;border-left: 1px #ccc solid;padding-left: 1ex;"><div dir="ltr">
<p><font face="arial, sans-serif"><span>Dear community, </span></font></p><p><font face="arial, sans-serif"><span><br></span></font></p><p><font face="arial, sans-serif"><span>I am running a simulation of a sodium aluminate (NaAlO2) crystal, orthorombic, and since I will need to obtain core-level spectroscopy results I am running this system with GAPW and all-electron scheme. </span></font></p><p><font face="arial, sans-serif"><span><br></span></font></p><p><font face="arial, sans-serif"><span>Unfortunately, I keep getting the following error at the very beginning of the SCF, using OT or diagonalization, with any preconditioning scheme:</span></font></p><p><br></p><p><font face="arial, sans-serif"><b><span>Cholesky decompose failed: the matrix is not positive definite or </span><span>ill-conditioned.</span><span> </span></b></font></p><p><font face="arial, sans-serif"><b><span><br></span></b></font></p><p><font face="arial, sans-serif"><span>I have tried the following basis sets:</span></font></p><p><font face="arial, sans-serif"><span><br></span></font></p><p><span>OK: 3-21Gx</span></p><p><span>6-31Gx</span></p><p><span>6-31Gxx</span></p><p><span>6-31++Gxx</span></p><p><span>6-311Gxx</span></p><p><span>6-311++G2d2p <span> </span></span></p><p><span>6-311++G3df3pd<span> </span></span></p><p><span>OK: Ahlrichs-VDZ</span></p><p><span>OK: Ahlrichs-VTZ</span></p><p><span>OK: Ahlrichs-pVDZ</span></p><p><span>Ahlrichs-TZV</span></p><p><span>Ahlrichs-pTZV</span></p><p><span>Roos-ADZ-ANO</span></p><p>
</p><p><span>Roos-ATZ-ANO</span></p><p><span><br></span></p><p><span>and I get the error for ALL of them excepted the ones preceded by "OK:" in the list. </span></p><p><span>The error is persisting both slightly randomizing the coordinates or with a 2x2x2 supercell.</span></p><p><span><br></span></p><p><span>Attached input file and coordinates. Without preconditioning it works, but with extremely difficult convergence. A wavefunction from the non-preconditioned run can be used as a GUESS, but when preconditioning is switched on, the error appears again.</span></p><p><span><br></span></p><p><span>Any help with this is appreciated. Thank you.</span></p><p><span>Best regards, </span></p><p><span><br></span></p><p><span>Daniele Passerone</span></p><p><span><br></span></p><p><span><br></span></p></div></blockquote></div></div></div>