We present an efficient computational protocol for robust transition state localization that can be routinely applied to complex (organometallic) reactions. The capabilities of the combination of extended tight-binding semiempirical methods (GFNn-xTB) with a state-of-the-art transition state localization algorithm (mGSM) is demonstrated on a modified version of the MOBH35 benchmark set, consisting of 29 organometallic reactions and transition states. Furthermore, for three examples we demonstrate how error-prone the conventional (manual) approach based on chemical intuition can be and how errors are avoided by a semiautomated generation of reaction profiles. The performance of the GFNn-xTB methods is carefully assessed and compared with that of the widely used PM6-D3H4 and PM7 semiempirical methods. The GFNn-xTB methods show much higher success rates of 89.7% (GFN1-xTB) and 86.2% (GFN2-xTB) compared with 72.4% for PM6-D3H4 and 69.0% for PM7. The barrier heights and reaction energies are computed with much better accuracy at reduced computational cost for the GFNn-xTB methods compared with the PMx methods, allowing a semiquantitative assessment of possible reaction pathways already at a semiempirical level. The mean error of GFN2-xTB for the barrier heights (8.2 kcal mol-1) is close to what low-cost density functional approximations provide and substantially smaller than the corresponding error of the competitor methods.

中文翻译：

---

使用半经验量子化学方法对有机金属配合物进行半自动过渡态本地化。

我们为鲁棒的过渡态本地化提出了一种有效的计算协议，可以常规地将其应用于复杂的（有机金属）反应。在由29种有机金属反应组成的MOBH35基准集的修改版上，证明了扩展的紧密结合半经验方法（GFNn-xTB）与最新的过渡状态定位算法（mGSM）结合的功能和过渡状态。此外，对于三个示例，我们演示了基于化学直觉的常规（手动）方法如何易于出错，以及如何通过半自动生成反应曲线来避免错误。仔细评估了GFNn-xTB方法的性能，并将其与广泛使用的PM6-D3H4和PM7半经验方法的性能进行了比较。GFNn-xTB方法的成功率分别为89.7％（GFN1-xTB）和86.2％（GFN2-xTB），而PM6-D3H4为72.4％，PM7为69.0％。与PMx方法相比，GFNn-xTB方法的势垒高度和反应能的计算精度更高，而且降低了计算成本，从而可以在半经验水平上对可能的反应路径进行半定量评估。GFN2-xTB的势垒高度（8.2 kcal mol-1）的平均误差接近低成本密度函数近似值所提供的误差，并且比竞争对手方法的相应误差小得多。与PMx方法相比，GFNn-xTB方法的势垒高度和反应能的计算精度更高，而且降低了计算成本，从而可以在半经验水平上对可能的反应路径进行半定量评估。GFN2-xTB的势垒高度（8.2 kcal mol-1）的平均误差接近低成本密度函数近似值所提供的误差，并且比竞争对手方法的相应误差小得多。与PMx方法相比，GFNn-xTB方法的势垒高度和反应能的计算精度更高，而且降低了计算成本，从而可以在半经验水平上对可能的反应路径进行半定量评估。GFN2-xTB的势垒高度（8.2 kcal mol-1）的平均误差接近低成本密度函数近似值所提供的误差，并且比竞争对手方法的相应误差小得多。