a birth-death process
Imagine a population, or the number of people in a queue, or the count of busy phone lines, that goes up and down one at a time over continuous time. An arrival or a birth pushes the count up by one; a departure or a death pulls it down by one; nothing jumps by two at once. A birth-death process is the continuous-time model of exactly such a count: a number that lives on 0, 1, 2, ... and changes only by single steps up or down.
It works by rates, not fixed timetables. When the count is in state n, two clocks compete: 'next birth' arrives at rate lambda_n (taking n to n+1) and 'next death' arrives at rate mu_n (taking n to n-1), and these compete memorylessly like exponential alarms — whichever rings first wins. The rates are allowed to depend on the current state n, which is what makes the model flexible: births might slow as a population gets crowded, or deaths might speed up. A pure birth process with mu_n = 0 and constant lambda is just the Poisson process counting up; allowing deaths lets the count come back down. This makes a birth-death process a special, especially tractable continuous-time Markov chain — one whose only moves are to neighbors.
Birth-death processes are the backbone of elementary queueing and population modeling, and their great gift is a simple equilibrium condition: in steady state the flow up across any boundary balances the flow down (detailed balance), giving the long-run probability of each state as a tidy product of rate ratios lambda/mu. From that you read off practical numbers — the chance the queue is empty, the average number waiting, how likely the system is overloaded. The key modeling assumption to keep honest is the one-step, memoryless structure: real systems with batch arrivals (a tour bus dumping 40 customers at once) or with strong memory in service times do not fit the plain birth-death frame.
A small shop has at most a few customers. With state-dependent birth rate lambda_n (new customers arriving) and death rate mu_n (customers leaving served), detailed balance gives the steady-state probabilities: P(state 1)/P(state 0) = lambda_0/mu_1, P(state 2)/P(state 1) = lambda_1/mu_2, and so on. Multiplying these ratios and normalizing yields the long-run chance of finding exactly n customers present.
Count moves up by births (rate lambda_n), down by deaths (rate mu_n); detailed balance gives the steady state.
Births and deaths happen one at a time and memorylessly. Batch arrivals or non-exponential service times break the simple birth-death structure and need a more general model.