Hi, greetings:<br /><br />I hope to use CP2K to optimize the spatial structure of the high spin state of the spin crossover periodic crystal [Fe(pyrazine)][Ni(CN)4] (a MOF type material, the crystal structure is shown in <a href="https://imgur.com/a/2xhQjKi">Figure 1</a>).<br /><br />This material is in a low spin state at normal pressure and low temperature, which is a diamagnetic state. The d-layer electronic configuration of Fe(ii) is t2g6, spin is 0, and Ni spin is 0. It will automatically transform into a high spin state at normal pressure and 300K, which is a paramagnetic state, the d-layer electronic configuration of Fe(ii) becomes t2g4eg2, spin is 2, and Ni spin is 0; the difference between the high and low spin states in the spatial structure is mainly reflected in the different bond lengths of the Fe-N octahedron (high---2.1+A, low---1.9+A), and the bond lengths of other chemical bonds remain basically unchanged. The unit cell of this material has 20 atoms.<br /><br />There are several studies in the field of spin crossover research (such as a recent one DOI: 10.1063/5.0157187) showing that tpssh, b3lyp and m06l are three functionals that are fairly good at describing the properties of spin crossover materials, so I would like to use these three functionals to optimize the spatial structure of the high and low spin states of [Fe(pyrazine)][Ni(CN)4] and compare the simulation results of different functionals.<br /><br />Recently, a study (10.1021/acs.jpcc.2c01030) used quantum espresso to simulate the PBE+D2 calculation of a similar material [Fe(pyrazine)][Pt(CN)4]. Therefore, I used the method of this study to optimize the spatial structure of the high and low spin states of [Fe(pyrazine)][Ni(CN)4] using QE's PBE+D2. After setting different spin polarization parameters for the input files corresponding to the high and low spin state according to the experimental data (see attachment 1 for the high spin input file and attachment 2 for the low spin input file), the vc-relax calculations converged successfully, and the results also reflected the spatial structure and magnetic properties of the high and low spin states. The calculated bond length data was in line with expectations (see <a href="https://imgur.com/9hMUtYb">Figure 2</a> for the optimized unit cell structure for high spin state and <a href="https://imgur.com/feFHAYl">Figure 3</a> for low spin state).<br /><br />Then I tried to use tpssh, b3lyp and m06l functionals in QE to optimize these spatial structures. However, for tpssh and m06l functionals, I did not find a pseudopotential group that could make the optimization run normally without errors (for example, the most common error is "Error in routine cdiaghg (161): S matrix not positive definite"). On the other hand, although the SCF converged, the b3lyp had the problem that the calculation of "Using ACE for calculation of exact exchange" would keep looping for a long time on a hpc cluster(see attachment 3). Therefore, I decided to try to use CP2K to simulate the three functionals of tpssh, b3lyp and m06l.<br /><br />I used Multiwfn to generate input files for high and low spin state spatial structure optimization. The low spin state was successfully simulated using three functionals: tpssh (expanding the cell to 2*2*2 and only counting the Γ point, attachment 4), b3lyp (expanding the cell to 2*2*2 and only counting the Γ point), and m06l (using the primitive cell, with the K point of 6*6*6, attachment 5). The optimization results of the two hybrid functionals t and b were in line with expectations (see Figures <a href="https://imgur.com/H0NeEv1">4</a> and <a href="https://imgur.com/oYAENpH">5</a>, respectively). Although the Fe-N bond length calculated by m was in the low spin state range, the bond lengths that should have been equal were different (see Figure <a href="https://imgur.com/Jo2x2a9">6</a>).<br /><br />In the high spin state simulation, although b3lyp (attachment 6) was able to converge and the program successfully exited, the optimized structure obtained was not ideal. Some of the 8 primitive cells in the supercell were in the high spin state (Fe-N bond length was greater than 2), and some were in the low spin state (Fe-N bond length was less than 2), and the Fe-N bond lengths in the same primitive cell were not all equal (see Figure <a href="https://imgur.com/xm2jKsp">7</a>). When I used tpssh to optimize, the settings I used were almost the same as those of b3lyp. However, scf did not converge (it oscillated around 10E-5 in the output file), so I increased EPS_SCF from 10E-6 to 10E-5 (Appendix 7) and tried the calculation again. This time, SCF converged, but the forces in the geometry optimization (Maximum gradient data in the output file) could not converge to the default 4.5E-4 (the closest it could get was 8E-4) after running for more than 12 hours. After checking the pdb file of the output geometry structure change data, I found that the overall structure only had obvious changes in the initial optimization stage, and no longer changed in the later stage. In addition, the later structure had similar problems as the
b3lyp
optimization
result (see Figure <a href="https://imgur.com/rci6EyL">8</a>). When optimizing with m06l (input attachment 8), the SCF generally oscillates between 10-2 and 10-3 (output attachment 9). I tried some combinations of basis vectors and pseudopotentials (https://cp2k-basis.pierrebeaujean.net/), such as the basis vector and pseudopotentials for PBE and MGGA, but the SCF convergence problem was not solved.<br /><br />I also tried to use CP2K's PBE+D3 method (using primitive cells and 6*6*6 K points) to optimize the high-spin state (attachment 10). After running for more than 10 hours, the situation similar to tpssh high-spin state simulation in which the force cannot reach the convergence limit appeared. The output geometry structure change trajectory from pdb file is almost stable in the later stage and the bond length and other data are consistent with the high-spin state (see Figure <a href="https://imgur.com/n99RAWL">9</a>). In addition, the high-spin state optimization results of PBE with 2*2*2 expanded cells as the initial guess of tpssh optimization also produced the similar result that the force cannot reach the convergence limit in the later stage and the optimized structure in the later stage has similar problems as described above (see Figure <a href="https://imgur.com/Y6pIhtn">10</a>).<br /><br />Then I tried to use CDFT to fix the magnetic moment of Fe, but the two functionals pbe and tpssh (see attachment 11) I tried faced a problem that the "Target value of constraint" and "Current value of constraint" polarized after the program ran for a certain period of time (see attachment 12).<br /><br /><div>Dear CP2K users, are there any problems in my various input file parameter settings shown above? Is there any way to make the conventional DFT or CDFT structure optimization results of tpssh, m06l or b3lyp successfully converge to a satisfactory high spin state?</div><div><br /></div><div>Thank you very much, your help are much appreciated.</div><div><br /></div><div> Daniel Lee</div>
<p></p>
-- <br />
You received this message because you are subscribed to the Google Groups "cp2k" group.<br />
To unsubscribe from this group and stop receiving emails from it, send an email to <a href="mailto:cp2k+unsubscribe@googlegroups.com">cp2k+unsubscribe@googlegroups.com</a>.<br />
To view this discussion visit <a href="https://groups.google.com/d/msgid/cp2k/70fc5e61-3c7f-4ec2-a231-57ba20d21b22n%40googlegroups.com?utm_medium=email&utm_source=footer">https://groups.google.com/d/msgid/cp2k/70fc5e61-3c7f-4ec2-a231-57ba20d21b22n%40googlegroups.com</a>.<br />