Study the possibility of multi-tables indexes

Hello,

This is a very not finished idea.
On some queries, particularly when you model your data in a more academic way, minimizing denormalisation used for performance, I noticed that a multi-table index would greatly limit the volumetry after only a simple part of the filters in WHERE are applied to 2 or 3 columns that are in distinct tables.
The base volumetry of the joined tables (sometimes it is (not far from) a cartesian product) may be really huge.
It would be challenging to have such indexes for contention/locks problems,
but used with parcimony, it may still be a gain.
I have no clue if it has already been asked/studied before,
but it is typically a question that may have interested academic papers on databases,
contrary to my previous ticket on implict enums (but maybe I have a bias in my mind about what kind of database optimisation problem is more interesting and/or publishable in academic fields).
So maybe one day, this kind of thing will be possible.
I'm just thinking that maybe there are already indexes on materialized views in some DBMS.
But I don't know of materialized views that are "auto-refreshed".
Here the idea would be to have this multi-table index respect transaction levels semantic.

Best regards,
Laurent Lyaudet