Let be the event that wins, and the event that his first toss is a tail. For a nontrivial starting state , a first tail has probability and changes his holding to . By the chain rule for probabilities,
Since , the definition of conditional probability gives
In particular, when , losing the first coin would end the game in defeat, so the conditional answer is zero. At conditioning on victory is undefined, while at the game has already ended and no first toss occurs.
The events and are disjoint and their union is . Additivity of probability gives
Both conditioning events have positive probability because . Apply the definition of conditional probability to each term to obtain the law of total probability:
Let mean the added ball was red, and mean the first draw was red. Their relevant probabilities are , , and . The law of total probability gives .
Given the observed red draw, the remaining ball is red precisely when occurred. Bayes' theorem therefore gives
The observation increases the weight of the two-red-ball box, because that box is more likely to yield a red first draw.
For the discrete uniform distribution, each value has probability . Using the given finite sums,
Subtracting the square of the expected value gives the variance:
Hence
This numbered item specifies the model used below. A convenient sample space is the collection of ordered pairs , each with probability . The corresponding random variables are determined by the sum and difference. Inverting the two equations gives
Thus a candidate pair is possible exactly when and are both integers between and . In particular, and must have the same parity. This inverse description will determine the joint probability mass function.
Count the ordered roll pairs producing each value. The probability mass functions are
and are zero outside these ranges. Both counts have a unique peak:
However, forces equal rolls, whose sum is even. Hence , whereas the product of the two marginal probabilities is . This disproves independence of random variables:
Figure 1.
The 36 possible sum and difference pairs for two independent fair dice; the two marginal modes cannot occur together
. Details.
First restrict to integers in the marginal supports: and . Their marginal probabilities are positive. By part (i), a feasible joint pair identifies a unique ordered pair of rolls, so its joint probability is ; an infeasible pair has joint probability zero.
Put and . If the two events were independent, a feasible pair would require
Since , equality forces , hence . But that pair is infeasible. An infeasible pair cannot give independence either, since its joint probability is zero while the product of its positive marginal probabilities is positive. Thus
If all real values of are allowed literally, an impossible event is independent of every event. The complete answer without a support restriction is therefore
The two definitions of conditional probability give
Substitution proves Bayes' theorem:
The positive probabilities assumed in the question make both conditional expressions defined.
Let be a finite or countable partition of a sample space into events of positive probability. The events are disjoint and their union is . Countable additivity and the definition of conditional probability therefore give the law of total probability
Cells of zero probability contribute zero through and can be omitted rather than assigning an undefined conditional probability.
Given the coin, the next toss is independent of its earlier tosses and has that coin's head probability. Averaging using the posterior distribution gives the posterior predictive probability
Observing two heads raises the probability of having chosen a high-bias coin.
Let identify the selected coin and let be the event of two heads. The prior coin probabilities are equal, and the conditional probabilities of are . By Bayes' theorem,
The full posterior distribution over the three biases is .
If the selected coin has bias , the other two biases sum to . A uniform choice between those two coins therefore has conditional head probability . Average over the posterior distribution, using the previous posterior predictive probability:
The observation makes a high-bias coin more likely to have been removed from the remaining pair.
The law of total expectation gives . The law of total variance separates the two sources of randomness:
Substitution gives
The binomial thinning of a geometric random variable identity supplies a further check: itself has a zero-based geometric distribution with success probability .
Given , the score has a binomial distribution with success probability . Its conditional expected value and variance are and . Thus
This is binomial thinning of the first-round count; the second formula uses the identity .
A partition of a sample space is a finite or countable family of nonempty events such that
Thus exactly one event in the family occurs in each outcome. To use ordinary event-conditioned conditional expectations, one usually assumes ; any zero-probability blocks can instead be omitted from the expectation formula.
Let be the absorption time. Its expectation is finite: from every interior state, a string of at most successive left moves reaches , and the probability of such a string is at least . Hence the Markov property gives , a geometric tail bound, and .
For , one step is always spent before absorption or the remaining walk. Apply first-step analysis and the law of total expectation to the three possible first moves. The conditional expectations of the remaining absorption time are , respectively; independence of future moves gives the same law as a fresh random walk from the new position. Thus
which rearranges to
The term also appears when the walk stays at , because staying still consumes a step.
Each has the same probability distribution as . By independence, the entire vector and its negative have the same joint law, so has a symmetric distribution about zero. The two disjoint tail events consequently have equal probabilities:
Apply Chebyshev's inequality with the mean and variance from (i):
Using only containment in the two-sided event would give ; the symmetry supplies the necessary extra factor .
Since the form a partition of a set of the sample space, the events are pairwise disjoint and their set union is . By countable additivity of probability, followed by the definition of conditional probability,
Every ratio is legitimate because each . This proves the countable law of total probability.