In the paper Provable Data Possession at Untrusted Stores, there is a mechanism that allows a client to efficiently check if the server still has the client's data. They define two constraints for their system: server-side data access and communication complexity.
Also, they state that a scheme based on just Merkle tree cannot be secure (so they don't talk about the complexity here). They have cited a paper but cannot find any explanation for that.
Question: Why cannot we design a PDP using just a Merkle tree?