PoS is a deceptive marketing ploy that the Ethereum "Foundation" has been using since 2014[1] to dupe people into buying ethereum believing that PoS will replace Proof-of-Work.
[1] https://blog.ethereum.org/2014/01/15/slasher-a-punitive-proo...
How is it a ploy?
There are some coins that use PoS as a voting mechanism not a consensus mechanisms. If you so desired, you could do the same in any PoW chain.
>How is it a ploy?
Broadly small-game fallacy[1]. If PoS would work at scale ethereum developers would have already deployed it. It doesn't work. Why?
An physical analogy for why PoS can never achieve BFT is: Assume 1) A cryptographic hash function is not broken 2) you live in a cave and you cannot talk the outside world
You are given a letter with a series of nested hash pointers that could have only been produced if they expended resources to created. There is no way this could have been faked, even though you live in a cave, you can trust the message. PoW is a stable BFT solution.
Another version of you in PoS land is also in a cave. You receive a message. It has a list of Arcade tickets that are signed by other Arcade tickets. In order to trust the veracity of this message you have to hope that someone else outside of your cave also saw this message to you because if they haven't, you could be being eclipse attacked, consequently defrauded. You cannot trust that this message hasn't been forged without contacting other people or hoping that other people have see this message. It's not stable it's not a solution to BFT. It's why after 4 years and counting Ethereum still hasn't implemented it.
[1] https://unenumerated.blogspot.com/2015/05/small-game-fallaci...
A few days ago Vitalik went through the history of Ethereum's proof of stake research at the annual Devcon, if anyone's interested: https://www.youtube.com/watch?v=Km9BaxRm1wA
At this point they're pretty happy with the protocol, the spec is mostly written, and eight teams are working on clients. The new design includes blockchain sharding for about 1000X better scaling.
Regarding your analogy, regardless of protocol you have a certain amount of "weak subjectivity," in that you at least have to know what client software to run. Otherwise you might be accepting invalid blocks.
The attacks you mention can be handled by requiring stake to be locked up for several months, and slightly broadening the subjectivity you accept to include a periodic known blockhash. You have to use off-chain social means to find out the correct recent hash, but you have to do the same to find out the correct software to run. https://blog.ethereum.org/2014/11/25/proof-stake-learned-lov...
The beauty of PoW is that you don't. By examining a PoW blockchain one can determine the relative amount of work expended. That provides trust with little to no external information. (All you need to know is the relative difficulty of hash functions and what info is hashed to form a block, both independently verifiable).
With PoS there is no analog. You must get your 'trust' externally, through deciding which codebase/client you want to use. Without external information / sidechannels, you have no way of determining which client nor which history to trust.
Also, if you really assume that only the chain with the most hashpower is the valid one, then you have to ignore all other blockchains. And even then, you're assuming you know the correct hash function; given ASICs it's not practical to know what hash function requires the most economic effort just by looking at it.
Utxo commitment is definitely a shortcut that bitcoin will eventually implement, but it can only work reliably for long timeframes especially when you have “disturbances in the force” like what happened with BCH last year.
“Known” blockhash from several months back is something that can easily be faked unless you actually fetch all the history and verify.
Voting is sybill vulnerable.
The point is the resource has to be external (i.e. outside of the consensus system) in order to be invulnerable to sybil attacks. In PoW this external resource is computing power, while in PoS it is the resource of the system itself (which might be internal to the consensus system, but has to necessarily be gained externally).