在Help里，这样说明smearing

In Dmol3, when checked indicates that thermal smearing will be applied to the orbital occupation to speed up convergence. Typically, We can encounter problems with SCF convergence for high symmetry, open-shell, or metallic systems

Smear
The Smear parameter allows electrons to be smeared out among all orbitals within a given range of the Fermi level. The range is defined in hartrees and typically is from 0.001 to 0.1 au. A linear function is used to smear the Fermi edge linearly in the interval specified. This procedure improves convergence of the SCF procedure by allowing orbitals to relax more rapidly. However, the energy derivatives with respect to the atomic coordinates are only approximate. Therefore, if smearing of electrons is needed, the Thermal smearing is recommended.

Thermal smearing

The Thermal smearing uses a finite-temperature Fermi function to compute fractional occupation numbers. This is the default smearing used by DMol3 in MS Modeling. When using thermal smearing the energy A is calculated as:

A = Usmear - TS

with Usmear being the electronic energy, T the electronic temperature, and S the entropy. In the case of thermal smearing, A is a free energy. Due to the entropic contribution, A is lower than or equal to Uno smear, the total electronic energy without smearing. However, Usmear is higher than Uno smear.

Contrary to the Smear smearing, the Thermal smearing yields the exact energy derivatives for A. Therefore, if smearing of electrons is needed, the Thermal smearing is recommended.

smearing参数是虚拟的"电子温度", 是人为引入的一种算法,使得费米能级之上的电子也可以被占据，在费米能级附近的电子态会出现分数占据.这样做的原因是加速收敛,尤其是费米能级电子结构比较复杂(如金属体系)的情况;但若准确DFT计算的话应该选择smearing=0,而如果直接这样选择对于一些体系会不收敛， 因此有文献提到几何优化时应该先选择大的smearing参数，优化之后再使用小的参数继续优化，直到逐渐降低为0

个人感觉最好不要使用smearing
----------------引自zzgyb，

谢谢！请问“先选择大的smearing参数，优化之后再使用小的参数继续优化”这样逐步减小smearing参数的方法，能保证收敛吗？ 如果可以的话，每次将smearing参数减少多少，一个数量级可以吗？

应该能，只要大的smearing参数收敛了，就继续减小，如果上到某个值不收敛了，就取前一个。减少多少，你可以结合你的体系试着计算了。