Mathematica has been closing the kernel mid-way through evaluating a notebook, exactly as if I had done Evaluate > Quit Kernel. This happens seemingly at random every 3 hours or so and if I re-evaluate the notebook it usually runs fine. It has been happening ever since version 11.2 and now I am trying 12.1, but this still has the same problem. It has affected installations on several different Windows PCs and on many different notebooks, but never happened in 11.1 and prior. It has been the subject of numerous technical support inquiries, but its random nature means it is not repeatable. Has anyone else had this problem and found a solution?

Yes, it is possible. Indeed there has to be something very unusual about what I am doing. The only software I can think of common to all 4 PCs were MS Office, Visual Studio and Eclipse, Chrome and Acrobat Reader all of which I would expect many Mathematica users to also use. Also Norton Internet Security would be on all of them. One of the 4 PCs is only accessible through a remote desktop and has no graphics card, so I can cross off a wide range of video and graphics-intensive programmes or games that would never have been installed there. In the past, I thought it could be a MariaDB JDBC driver as the vast majority of my notebooks read from a centralised database and this driver would be on all PCs, but later I have had examples fail that didn't use the database at all and the failures were certainly not commonly happening during database functions or commands in Mathematica. I also wondered about network devices, especially printers and have tried removing these from the PC I am currently using, but haven't found a culprit yet. However, I think your direction is a good one to reflect on.

Is there any chance you can pare down to a smallish example? It is likely to be something along the lines suggested by Szabolcs Horvat. So the smaller the MCE ("minimal crashing example") the better, in terms of likelihood we might track it down.

Since people start to comment on this, it is good to point out that Roger may be using an early prerelease of 12.1, see my comment on https://community.wolfram.com/groups/-/m/t/1934409

Mathematica closing kernels