Hi
I added into your file here "approach 3" which contains my answer. The modified file is attached.
Schematically is as follows
1. I use MetricCompute like your "approach 2", but I make the result for Riemann into kompRiemann which is inputted into another tensor called compRiemann. In this stage, I don't use your reduce function which simplify the expression a lot.
2. Then, I calculate the 1 covariant derivative of Riemann then put it into kompDRiemann and input it into compDRiemann. To check it, I just called kompDRiemann and use your reduce function.
3. Repeate the steps again to into kompDDRiemann and input it into compDDRiemann, then into kompDDDRiemann and input it into compDDDRiemann
I find this to be faster and it produce the missing term a2^2 v22 that you mention in the file. Hope it helps.
Best,
Ilham