A few weeks ago I flagged up a glitch in a non linear analysis where it had given me the wrong answer to the change in length of a bar whose CTE changed with temperature.
I flagged this up to the VAR who gave me a reply this morning.
When inputting a CTE curve for a material, don't put in the CTE curve. At least don't put in the instantaneous CTE curve. You need to put in the average CTE curve, a bit like a secant modulus curve I suppose. Of course this curve will change depending upon where the stress free temperature is and so you may well have to construct a new curve for each problem.
Explanation in the knowledge base here: s-019444