Our protocols for t fault tolerance in a system provide us with a guarantee that our system will not fail if no more than t replicas fail. With this guarantee, we must ensure that the number of faulty nodes in an ensemble of replicas does not exceed t. We can do this by replacing faulty replicas with non-faulty replicas. Let's formally discuss this.
Modeling replica replacement
We define P(τ) as the total number of nodes running state machine replicas in an ensemble of replicas and F(τ) as the number of faulty nodes in that ensemble at time tau. P(τ)−F(τ) must be greater than a certain number to guarantee that our system will produce the correct output. Here is how we can formally define this combining condition: