In a SQL Server 2016 environment, I have a `SELECT` query which queries one table, which joins to itself on 2 indexed columns (both non-clustered):

    SELECT
        A.REFERENCE,
        A.MEMBERID
    FROM 
    	TRANS A INNER JOIN 
    	TRANS B ON A.REFERENCE = B.REFERENCE AND A.MEMBERID = B.MEMBERID

This one above returns in about 2 seconds.

But, when using the same query, but looking for matching references across different memberids e.g. changing `=` to `<>`, it takes about 20 seconds. 

    SELECT
        A.REFERENCE,
        A.MEMBERID,
        B.MEMBERID AS MEMBERID_B
    FROM 
    	TRANS A INNER JOIN 
    	TRANS B ON A.REFERENCE = B.REFERENCE AND A.MEMBERID <> B.MEMBERID

I realize that this 2nd query uses a negative comparison and thus, must fully review table B vs. the 1st query which can just find the single match and move on (I think). In the query plan, it uses the same index and both are seeks, but the `=` estimates a single row involved, and the `<>` estimates 20 million rows (entire table).

**My question is**: without being able to alter any indexes or table structure, etc. how might I go about optimizing the query with the negative comparison? (or re-writing it to achieve the same results?) I've searched google, and found lots of info saying not to use negative comparisons where possible, but not much on optimizing when you must.