This requires a great number of assumptions or an tremendous amount of work to prove them out and there are too many variables to make this viable in more than a few very specific edge cases.
Even if we somehow know that the account belongs to the person it appears to, that the account is active, that it doesn't just appear active due to automatic sign-ins, browser extensions, mobile apps, malware, that someone else doesn't have access to the account (any form of significant other, roommate, family member, etc), that the person actually ever reads their messages (I have an account just so nobody else can use the identity; I sign in maybe twice a year; I can count the number of times I have read my inbox there on one hand), that they read *that* message, etc.
If it were simply enough to say "we know this account really belongs to this person and that they actively login and use the account", then we wouldn't need certified mail or people to serve a summons in person. It would be enough to say "we have a mail address for this person and we'll just send a regular first class mail notification to them". But that *isn't* enough. And neither is saying "well, shit, we sent a facebook message".
And facebook can be more reliable than physical mail? We're going to bank all of this on the reliability of a single third party entity? Didn't we just go through like six months of "gosh, we lost every single email for all these important IRS people of the last five years, because email's supes unreliable, guys!"?