Proof of Exponential Memorylessness

Why Memorylessness Matters

A waiting time is often observed only after it has already lasted for some period. For an exponential model, the probability of waiting at least a further $t$ time units is the same whether or not the process has already waited $s$ time units. This statement is called memorylessness. Its precise meaning comes from conditional probability, so the conditioning event and the time thresholds must be written carefully.

The property is special to the exponential distribution among continuous models on nonnegative waiting times. It is useful in probability calculations involving arrivals, service times, or lifetimes when an exponential model is appropriate. The focus here is the proof and correct use of the identity, not a general claim that all real waiting processes behave this way. This topic is aligned with the SOA VEE Mathematical Statistics syllabus.

Model: Let $X$ be an exponential random variable with rate $\lambda>0$, measured in inverse time units. Its support is $x\geq 0$, and its survival function is $$ S_X(x)=P(X>x)=e^{-\lambda x},\qquad x\geq 0. $$ For every finite $s\geq 0$, $P(X>s)=e^{-\lambda s}>0$, so conditioning on $X>s$ is valid.

The rate $\lambda$ must use the reciprocal of the time unit in which $X$, $s$, and $t$ are measured. For example, if $X$ is measured in days, then $\lambda$ is measured per day. The exponential distribution assigns positive probability to every positive-length interval contained in $[0,\infty)$; it assigns no probability to intervals lying wholly below zero.

The Memoryless Identity

Definition (memoryless property): A nonnegative random variable $X$ is memoryless if, for all $s,t\geq 0$, $$ P(X>s+t\mid X>s)=P(X>t). $$ For an exponential random variable with rate $\lambda>0$, this identity holds.

The threshold $s$ represents the time already elapsed in the event $X>s$, while $t$ is the further duration being considered. Thus the event in the numerator is $X>s+t$, not merely $X>t$. This distinction is essential whenever the conditioning threshold is positive.

Derivation from Conditional Probability

For events $A$ and $B$ with $P(B)>0$, conditional probability is defined by $P(A\mid B)=P(A\cap B)/P(B)$. Take $A=\{X>s+t\}$ and $B=\{X>s\}$. Because $t\geq0$, $X>s+t$ implies $X>s$; hence $A\cap B=A$. Therefore the desired conditional probability is a ratio of two survival probabilities.

Write the conditional probability.
Since $s,t\geq0$, the event $X>s+t$ is contained in $X>s$, and the conditioning event has positive probability.

Replace each probability by a survival probability.
The numerator is $P(X>s+t)=e^{-\lambda(s+t)}$, and the denominator is $P(X>s)=e^{-\lambda s}$.

Simplify the ratio.
Cancel the factor associated with the elapsed time $s$; what remains is exactly the probability of exceeding $t$ under the original exponential model.

\begin{aligned} P(X>s+t\mid X>s) &=\frac{P(X>s+t)}{P(X>s)}\\ &=\frac{e^{-\lambda(s+t)}}{e^{-\lambda s}}\\ &=\frac{e^{-\lambda s}e^{-\lambda t}}{e^{-\lambda s}}\\ &=e^{-\lambda t}\\ &=P(X>t). \end{aligned}

This proves the identity for every $s,t\geq0$. The cancellation explains why the elapsed time disappears: the conditional probability compares survival through both the elapsed and additional periods with survival through the elapsed period alone. Since the model's survival probabilities have the exponential form, their ratio depends only on the additional duration $t$.

The proof relies on the exponential survival formula and on $P(X>s)>0$. It does not say that the elapsed time is literally erased, nor that every waiting-time model has this property. In particular, substituting a different survival function into the ratio need not produce $P(X>t)$.

Worked Applications

Worked Example: Conditional Exceedance After an Elapsed Time

A hypothetical sensor reports a waiting time $X$, in minutes, until a signal arrives. Assume $X\sim\operatorname{Exponential}(\lambda=0.25\text{ per minute})$. Given that no signal has arrived by minute 4, find the probability that the waiting time exceeds minute 9. Then compare this with the probability that a new waiting time exceeds 5 minutes.

Method. The event $X>9$, under the condition $X>4$, means surviving an additional $9-4=5$ minutes. Thus take $s=4$ and $t=5$ in the memoryless identity.

\begin{aligned} P(X>9\mid X>4) &=P(X>4+5\mid X>4)\\ &=P(X>5)\\ &=e^{-0.25(5)}\\ &=e^{-1.25}\\ &\approx 0.2865. \end{aligned}

For a new waiting time, the probability of exceeding 5 minutes is also $P(X>5)=e^{-1.25}\approx0.2865$. The values agree, as the property predicts. The exponent is dimensionless: $0.25$ per minute multiplied by 5 minutes gives $1.25$.

Check. Directly, the conditional probability is $P(X>9)/P(X>4)=e^{-2.25}/e^{-1}=e^{-1.25}$, the same result. The common mistake to avoid is using 9 as the additional duration; 9 is the absolute threshold, while the additional duration after minute 4 is 5 minutes.

Worked Example: Finding an Additional Duration from a Conditional Probability

For a hypothetical equipment-repair process, let $X$ be the time to completion in days, with $X\sim\operatorname{Exponential}(\lambda=0.10\text{ per day})$. The repair is still incomplete after 3 days. Find the additional duration $t$ for which the conditional probability of remaining incomplete for at least that further duration is 0.25. State the corresponding absolute time threshold.

Method. The requested conditional event is $X>3+t$ given $X>3$. By memorylessness, its probability is $P(X>t)$. Set that survival probability equal to 0.25 and solve for $t$.

\begin{aligned} P(X>3+t\mid X>3)&=P(X>t)=0.25,\\ e^{-0.10t}&=0.25,\\ -0.10t&=\ln(0.25),\\ t&=\frac{-\ln(0.25)}{0.10} =\frac{\ln(4)}{0.10}\\ &\approx 13.8629\text{ days}. \end{aligned}

The absolute time measured from the start is $3+t\approx3+13.8629=16.8629$ days. A check gives $e^{-0.10(13.8629)}\approx e^{-1.38629}=0.25$. The reported additional duration is rounded to four decimal places.

Common mistake avoided. Solving $P(X>t\mid X>3)=0.25$ would use $t$ as an absolute threshold, not as the additional duration stated in the question. Writing the event as $X>3+t$ makes the intended meaning explicit.

Worked Example: A Conditional Probability for an Interval

In a hypothetical inspection system, the time $X$ until an alert is generated follows an exponential distribution with rate $\lambda=0.20$ per day. Given that no alert has occurred during the first 3 days, find the probability that it occurs after day 3 but no later than day 8.

Method. Under the condition $X>3$, the event of interest is $3<X\leq8$. Its conditional probability is one minus the conditional probability of still having no alert after day 8. The latter is a memoryless probability with additional duration $8-3=5$ days.

\begin{aligned} P(3<X\leq8\mid X>3) &=1-P(X>8\mid X>3)\\ &=1-P(X>5)\\ &=1-e^{-0.20(5)}\\ &=1-e^{-1}\\ &\approx 1-0.3679\\ &\approx 0.6321. \end{aligned}

The answer is approximately 0.6321, or 63.21% as a percentage. For a direct check using conditional probability, the interval event has probability $P(3<X\leq8)=e^{-0.6}-e^{-1.6}$. Dividing by $P(X>3)=e^{-0.6}$ gives $1-e^{-1}$, as above. The interval endpoints cause no extra term because a continuous random variable has probability zero at any single point.

Common Errors and Scope of the Result

Conditional probability problems are especially vulnerable to confusion between an absolute threshold and the amount of additional time. If the condition is $X>s$, then a further $t$ time units corresponds to $X>s+t$. By contrast, $X>t\mid X>s$ asks about the absolute threshold $t$. When $t\geq s$, the exponential model gives

P(X>t\mid X>s)=\frac{e^{-\lambda t}}{e^{-\lambda s}} =e^{-\lambda(t-s)}.

The difference $t-s$, rather than $t$, is the remaining duration in that expression. If $t<s$, the event $X>t$ already contains the conditioning event $X>s$, so $P(X>t\mid X>s)=1$. Always read the event before choosing a duration.

A second error is forgetting the denominator in conditional probability. The numerator $P(X>s+t)$ alone is an unconditional survival probability. The conditional probability requires dividing by $P(X>s)$. A third error is a unit mismatch: a rate of 0.2 per day must be multiplied by a duration in days, not hours. If the duration is given in hours, convert it to days first or convert the rate to a per-hour rate.

Key takeaway: For $X\sim\operatorname{Exponential}(\lambda)$, with $\lambda>0$ and $s,t\geq0$, conditioning on $X>s$ gives $$ P(X>s+t\mid X>s)=P(X>t)=e^{-\lambda t}. $$ The numerator's threshold is the elapsed time plus the additional duration; the conditioning event must have positive probability, as it does for every finite $s$ in this model.

Check Your Understanding

For each problem, identify the additional duration before applying the identity. Rates are stated in the same time units as the waiting times.

A hypothetical dispatch time $X$ is exponential with rate 0.40 per hour. Find $P(X>5\mid X>2)$. (Answer: $e^{-1.2}\approx0.3010$)
A hypothetical inspection duration $X$ is exponential with rate 0.08 per day. Given $X>5$, calculate $P(X>15\mid X>5)$. (Answer: $e^{-0.8}\approx0.4493$)
A hypothetical service time $X$ is exponential with rate 0.20 per minute. Find the additional time $t$ such that $P(X>6+t\mid X>6)=0.30$. (Answer: $t=-\ln(0.30)/0.20\approx6.0199$ minutes)
For an exponential waiting time $X$ with unknown rate $\lambda$ per hour, it is known that $P(X>10\mid X>4)=0.20$. Find $\lambda$, then find $P(X>16\mid X>4)$. (Answer: $\lambda=-\ln(0.20)/6\approx0.2682$ per hour, and the second probability is $0.0400$)
A hypothetical alert time $X$ is exponential with rate 0.50 per hour. Given $X>2$, find the probability that $2<X\leq7$. (Answer: $1-e^{-2.5}\approx0.9179$)

What You'll Learn