If you don't want to use rules, you could do something like this:
- Pipe a Data Management extract job into an Import workflow, to copy and aggregate data from A1 to A2 into the newly created A3
- Verify all the data is there (all values in A should be doubled)
- Delete data pointing to A1 and A2. Now all values in A should be back to the correct state
- Orphan A1 and A2.
Remember that you cannot orphan members that contain data. Just switch the "In Use" property to False if you cannot remove data from them, without orphaning them.
