1.

可以去一部分水再试。

另外整体结构可以先在UFF下优化，然后再尝试改用ONIOM；或者先尝试不用电子嵌入，优化成功了再开启

(来自计算化学公社)

2.

Actually I have a solution for this problem. Strangely, it seems there is no keyword to change the maximum number

of microiteration cycles. But there is a way to get around it. Use the keywords: " opt=quadmacro iop(1/98=66,1/19=7)"

Useful options might be iop(1/8=3,1/13=6). The deal is the following: "quadmacro" stands for quadratic macrostep.

This is an awesome feature in g09, I heavily encourage everybody to use it in all QM/MM calculations, especially for

TS searches. Iop(1/98=66) makes a twist in the optimization workflow, it disables microiterations, only second order steps

are made everytime in both the QM and MM regions. Keep in mind that the MM region needs much more steps to

converge than the QM region because of its sheer size. If you run an optimization like this, you will notice that it will take

steps faster than the standard protocol, and eventually the QM region will reach convergence, but the MM region will

most probably be far from convergence. So after like a few tens of steps (for example 50-100) stop, get the latest

structure, and start an optimization with microiterations. Mine looks like:

"opt=(quadmacro,readfc) iop(1/8=3,1/128=1,1/110=2,1/19=7,1/13=6)". 1/128 and 1/110 are microiteration realted options

I think they are useful too in a lot of cases. 1/19=7 is a must (at least for my calculations) with quadmacro because of a

bug. If you get still the same error, than get another round of pire quadmacro optimization. And try again microiterations

afterwards. You can reach full convergence without microiterations, I suppose, but it would be quite inefficient, I think.  

（from Researchgate Gergely Bánóczi）