In the paper BIP32 and deterministic key derivation are mentioned. But why given that two peers directly exchange messages (i.e., there is no need to obscure anything)? Alice always exchanges state with Bob and that commitment transactions usually don't land on the blockchain.
Punishments for transmitting an old state can be done by attaching to the previous "breach-remedy" transaction using a valid preimage (that was exchanged while getting the new "commitment"). This means the script has to check the signature and invoke the hash function. An optimization is to use elliptic curve trickery and thus construct something similar to pay2contract scheme. I believe this way you still have to store all old values or am I mistaken? I've heard about ELKREM and optimizations in that part. But my question here is more basic, why did Poon and Dryja even consider BIP32? I understand the need for different settlement keys in Eltoo or the additional privacy for on-chain transactions, but what about Lightning?