Blockchain consensus: proof of work versus proof of stake

The problem consensus solves

In a distributed ledger thousands of nodes may receive transactions in different orders, and some nodes may be faulty or dishonest. Without a referee, how do they agree on one history? This is a version of the Byzantine Generals Problem: reaching agreement when some participants may lie. A consensus mechanism is the set of rules that solves it well enough for the system to be trusted. In open, public blockchains anyone can join, so the rules must also stop one person from creating thousands of fake identities (a Sybil attack). Both proof of work and proof of stake do this by tying influence to a scarce resource.

Proof of work

In proof of work (PoW), nodes called miners compete to find a value (the nonce) that makes the block header's hash fall below a target. The only way is trial and error, so it takes a lot of computation. The first miner to succeed broadcasts the block, others check it instantly, and the winner earns a reward. Bitcoin uses PoW with SHA-256 and adjusts the difficulty so blocks arrive at a steady rate.

When two valid blocks appear at once, nodes follow the chain with the most accumulated work. To rewrite history, an attacker would need more computing power than the honest miners combined, which is the basis of the 51% attack discussed in blockchain security. The big drawback is energy consumption and the need for specialised hardware.

Proof of stake

In proof of stake (PoS), validators lock up (stake) some of the chain's native currency. The protocol chooses validators to propose blocks, and others vote (attest) that the blocks are valid. If a validator breaks the rules, for example by signing two conflicting blocks, part of its stake can be destroyed. This penalty is called slashing. Ethereum switched from proof of work to proof of stake in 2022, an upgrade known as "the Merge", which cut its energy use dramatically.

Voting-based consensus and finality

Permissioned blockchains know who their members are, so they do not need an expensive Sybil defence. They often use Byzantine Fault Tolerant (BFT) algorithms such as Practical BFT (PBFT), where nodes exchange votes and a block is final once enough agree. Classic BFT tolerates fewer than one third of nodes being faulty. Some systems use crash-fault-tolerant ordering such as Raft, which assumes nodes may fail but not lie. Proof of authority lets a small set of approved signers create blocks. These are fast, but trust sits with a known group.

Finality means a transaction cannot be reversed. With PoW it is probabilistic: each extra block on top makes a reversal less likely, which is why services wait for several confirmations. Many PoS and BFT systems give clearer, faster finality.

Tip: in exams, link each method to its Sybil defence: PoW uses computing cost, PoS uses financial stake, permissioned chains use known identities.

The same ideas appear outside blockchains. Distributed databases use Raft and Paxos to keep replicas in agreement; see consensus in distributed systems.