A stochastic process on a discrete state space evolving in continuous time with the Markov property: the future state probabilities depend only on the current state and time increments via transition rates given by an infinitesimal generator matrix (Q-matrix).

Principle

Memorylessness at the state level implies that holding times in each state are exponentially distributed with parameters equal to the exit rates, and transitions between states are governed by the generator's off-diagonal rates, producing a semigroup of transition probability matrices P(t)=exp(Qt).

Demonstration

Situation: A server can be in states {idle, busy, failed} with constant transition rates between them. Recognition: State changes occur stochastically in continuous time and satisfy the Markov property. Action: Specify Q with exit rates and off-diagonal transition rates, compute P(t)=exp(Qt) to find state distribution at future times or simulate sample paths using exponential holding times. Consequence: One can compute time-dependent probabilities, steady-state distributions (if ergodic), and expected sojourn times using Q.

Misapplication

Assuming that observed empirical waiting-time distributions that deviate from exponentials can still be modeled as CTMC by adjusting transition probabilities rather than rates. The semantic error is conflating non-exponential holding-time evidence with Markovian structure; non-exponential sojourns violate the CTMC assumption.

Consequence

When valid, CTMCs yield tractable analytic calculations for transient and stationary behavior and support efficient event-driven simulation. When applied to systems with non-Markovian memory or state-dependent duration distributions, results (e.g., predicted occupancy, passage times) will be biased and may misinform design or control decisions.

Reversal

If holding-time distributions are not exponential or if the state space is continuous, the CTMC framework does not apply; one must use semi-Markov processes, Markov jump processes with supplementary variables, or continuous-state Markov processes (diffusions) instead.

Boundary

Clearly within: A queue with Poisson arrivals and exponential service times where transitions between queue lengths follow CTMC rates. Boundary case: A system with phase-type approximable service times: can be embedded into a larger CTMC but requires increased state space. Clearly outside: A renewal process with general interarrival and service time distributions that are not phase-type without embedding.

Semantic Tension

Analytic tractability ↔ Realistic sojourn distributions — CTMCs are mathematically convenient but require exponential holding times, which may not match empirical duration distributions without state-space augmentation.

Synthesis

CTMCs formalize memoryless state transitions via generator rates and matrix exponentials; modeling judgment requires testing exponential sojourn assumptions and, when violated, choosing augmented or alternative stochastic frameworks that preserve necessary temporal structure.