Expected Maximum of Several Dice

Problem. Roll $n$ fair six-sided dice. What's the expected value of the largest number showing? For $n=1$ it's obviously $3.5$ (the standard expected value of one die), but the answer for $n=2$, $n=3$, and beyond isn't obvious by symmetry the way single-die expectation is — and trying to sum "probability of each value times that value" directly gets messy fast. There's a cleaner route.

The complement trick: $E[X] = \sum_k P(X \geq k)$

For a random variable $X$ taking positive integer values from $1$ to $6$, there's a useful identity:

$$E[X] = \sum_{k=1}^{6} P(X \geq k) .$$

In plain English: instead of weighting each outcome by its value and summing, you can equivalently sum, for each threshold $k$ from 1 to 6, the probability that $X$ reaches at least that threshold. This is easier here because "the max of $n$ dice is at least $k$" has an easy complement: "the max is at least $k$" fails only if every die shows less than $k$, i.e., every die is in $\{1, \dots, k-1\}$, which happens with probability $\left(\frac{k-1}{6}\right)^n$. So

$$P(\max \geq k) = 1 - \left(\frac{k-1}{6}\right)^n .$$

Worked example 1: two dice

For $n=2$: $P(\max \geq 1) = 1$ (trivially, since max is always $\geq 1$), $P(\max \geq 2) = 1 - (1/6)^2 = 35/36$, $P(\max \geq 3) = 1 - (2/6)^2 = 32/36$, $P(\max \geq 4) = 1-(3/6)^2 = 27/36$, $P(\max \geq 5) = 1-(4/6)^2 = 20/36$, $P(\max \geq 6) = 1-(5/6)^2 = 11/36$. Summing:

$$E[\max] = \frac{36+35+32+27+20+11}{36} = \frac{161}{36} \approx 4.472 .$$

Compare to the single-die expectation of $3.5$: taking the max of two dice pulls the expectation up noticeably, which matches intuition — the larger of two rolls tends to land above the midpoint.

Worked example 2: how it scales with more dice

The same formula extended to $n=3$ dice gives $E[\max] \approx 4.958$, and for $n=4$, $E[\max] \approx 5.27$. As $n$ grows, the expected maximum climbs toward 6 but never reaches it, since there's always some (shrinking) chance every die falls short. This "expected max grows and saturates near the top of the range" pattern is the same shape you'd expect from order statistics in general — the maximum of more samples from a bounded distribution creeps toward the upper bound but the increments shrink (going from $n=1$ to $n=2$ adds about $0.97$; from $n=3$ to $n=4$ adds only about $0.31$).

Use the explorer above to build intuition for how a bounded outcome's distribution compresses toward one tail as you effectively "select the best of several trials" — the same compression that drives $E[\max]$ toward 6 as $n$ grows, even though no single die's distribution changes.