If the sum of Lebesgue outer measures on the right is infinite, the inequality is automatic. Otherwise, fix and cover each by a countable family of intervals whose total length is less than . These covers exist by the definition of infimum. The combined countable family covers and has total length at most . Take the infimum over covers of the union, then let tend to zero. No measurability assumption on the is needed.
Doob-Dynkin lemma Created 2026-10-25 Updated 2026-10-31
For real random variables , if is measurable with respect to the sigma-algebra generated by a random variable , there is a Borel measurable function with . To prove the bounded case, approximate pointwise by finite-valued simple functions . Each level set is for a Borel set , so each approximation factors as with a Borel measurable function. Apply scalar clipping to with a common bounded interval and use its limit superior to obtain . For unbounded finite , apply the bounded case to the one-to-one bounded continuous function , then invert on and assign arbitrary values elsewhere. Exact measurability gives exact factorization; assumptions after adjoining null sets give an almost surely equal version.
Define the head-count random variable by
The four elementary outcomes are inputs to this function; its three numerical values are outputs. Each fiber of a function collects all outcomes giving one value. Under the fair independent model, adding the masses in each fiber gives the probability mass function
Consequently the output values are not equally likely, although the inputs are. Using the reduced sample space with these masses gives the same head-count probability distribution; assigning to each value would describe a different model.
A question about a random variable's value becomes an event on the original sample space by taking an inverse image:
For a real-valued random variable, this is an event whenever is a Borel set, by measurability. In this finite example every subset of is allowed. In particular,
This is the same “at least one head” event counted above, expressed through an observable.[ref]
Figure 1.
Head count merges outcomes into fibers and adds their probabilities
. The random variable sends to zero, both and to one, and to two. The inverse image of is the highlighted event of at least one head. Two arrows meeting at one explain its probability . Details.
Every event determined by is a union of complete fibers. There are such events, whereas the full power set of has . In particular, the first-head event is not determined by : it contains but excludes , although both give . Choosing an observable makes calculations simpler precisely by forgetting distinctions; the useful choice is one that retains the distinctions the question needs.
On the full Cartesian product with its product sigma-algebra, take the coordinate random variables . A product-measurable set is a tail event if and only if membership is unchanged whenever two sequences differ in finitely many coordinates. The measurability condition is essential: invariance alone is not a license to assign a probability to an arbitrary subset.
For the forward implication, fix . An event in cannot distinguish points with the same remaining coordinates. Since a tail event belongs to this sigma-algebra for every , every finite change preserves membership.
For the converse, freeze the first coordinates to zero. The insertion map
is a measurable function for the product sigma-algebras. Thus is a measurable set. Invariance gives , where the coordinate projection discards the first entries. Hence for every . The same argument works on . On a general probability space, coordinates need not admit arbitrary replacements, so the defining intersection of sigma-algebras remains the reliable formulation.
Functional delta method Created 2026-10-07 Updated 2026-10-14
If converges in distribution to and is Hadamard differentiable at tangentially to a space containing the limit, the displayed transformation rule holds under the usual measurability and tight separable-support conditions. The variables must belong to the domain of .
On , the event of infinitely many heads is
It says that after every index there is another head. For each , the same intersection can start at , so ; therefore is a tail event. This verifies measurability directly using countable unions and intersections.
Figure 1.
A finite prefix admits both infinitely many heads and only finitely many heads
. The two illustrated extensions have the same initial observations. An alternating continuation belongs to the tail event of infinitely many heads; an all-tails continuation does not. Changing the highlighted finite prefix cannot change that infinite property, but it can change the exact first-head time. Details.
Assume now that the coordinates are fair independent Bernoulli trials. For fixed , let be the event of at least one head among . Its complement requires tails, so
The last equality follows from continuity of probability. Now , a countable intersection of almost sure events, and hence . This determines which of the two possibilities in Kolmogorov's zero–one law occurs. Equivalently, the Borel-Cantelli second lemma applies to the independent events .
Nevertheless, no finite prefix certifies membership in : every finite record can be extended by all tails or by alternating heads and tails. Under the fair independent model, conditioning on any exact finite prefix leaves , since is independent of the initial sigma-algebra. The excluded all-tails continuation has probability zero, but remains a possible member of the sample space. Logical determination by a finite record and a conclusion holding almost surely are different statements.
For a regular outer measure and measurable , the forward implication is countable additivity. For the reverse, choose a measurable hull of and intersect it with to obtain and . Measurability of splits into and , so
The assumed equality and finiteness force . Every outer-measure-null set is measurable by Caratheodory measurability criterion, so is measurable. Finiteness prevents invalid subtraction of infinite masses.
If , every countable interval cover of is also a cover of . Taking the infimum of the lengths over the larger family of covers can only decrease the result. Thus Lebesgue outer measure is monotone on all subsets, with no measurability assumption.
Use the characterization of measurability by preimages of open half-lines, allowing intermediate extended-real-valued functions. Put . For each real ,
Thus every is measurable. Countable suprema are measurable too, since and the upper half-lines also generate the Borel sigma-algebra. Therefore
is measurable. Its values are real by the stated hypothesis, so it is the required ordinary real-valued measurable function, rather than only an extended-real-valued one.
If , all indicator functions vanish and . If , the partition in part (i) puts it in exactly one , so exactly one term can be nonzero and the sum is a finite real number at that point. There is no ambiguous infinite sum of positive and negative terms.
For measurability, each symmetric partial sum
is Borel measurable by closure under finite linear combinations. For each , these partial sums eventually include its unique possible term and converge to . The measurable-limit result from part (a) therefore proves is Borel measurable.
measurability with respect to gives a Borel set with . Restrict it to the actual image:
This is Borel by the assumed Borel property of , and its preimage under is unchanged.
The intervals partition the real line, so their preimages under partition . If two of the restricted sets overlapped, a point in their intersection would equal for some , making lie in two disjoint intervals, a contradiction. Similarly any has a preimage , whose real value belongs to one of those intervals; that puts in the corresponding . Thus the restricted sets are disjoint and their union is exactly . This restriction is what removes arbitrary overlaps of the initially chosen Borel sets outside the image.
For two real-valued random variables and each real ,
If , density of the rationals supplies with ; conversely either pair of inequalities in the union implies . Every set in the countable union is measurable, so the half-line criterion proves measurability of . Its values are real, and induction gives every finite sum is a real-valued random variable.
Let . It is open: it is the union of all open intervals for which . This also proves measurability of the uncentered maximal function. Triangle inequality inside each average proves sublinearity and absolute homogeneity.
Take a compact subset , and choose finitely many such intervals covering it. The one-dimensional Wiener covering lemma selects pairwise disjoint intervals whose concentric triples cover . For a finite family this follows by choosing a longest remaining interval and discarding all intervals intersecting it: each discarded interval has no greater length and lies in the triple of the chosen one. Thus
Taking the supremum over compact subsets, by inner regularity of Lebesgue measure, gives
This establishes weak type (1,1). It is the one-dimensional uncentered maximal weak-type inequality for the finite measure ; the constant need not be optimal for this question.
There is a genuine qualification: the first printed inequality is valid as a useful general assertion for , but is false for arbitrary . Also the displayed likelihood ratio presupposes . We first prove the intended bound and then give a counterexample to the unrestricted quantifier.
For any estimator , set . By the triangle inequality, on , while on . Hence
Let with . On , . Therefore
The last step is Markov inequality. Taking the supremum over the model and then the infimum over estimators proves the intended metric two-point risk bound. At the right side is zero. For a negative right side is also harmless, but a positive right side need not be a lower bound.
Here is an explicit counterexample with positive probability density functions and . Let , , and take the model consisting of
Use the L2 norm metric, for which , and take for every . The majority-half decision selects if at least two observations lie in . Each error probability is , so its maximum normalized risk function is . This is also the likelihood-ratio test, giving . At the proposed right side is , contradicting the risk of this explicit estimator. Thus the intended range restriction cannot be omitted.
For the second request use the intended L2 norm metric . The PDF defines its square, not the square itself as a metric. Put
Both probability density functions belong to the model, and . The two-point choice may depend on , because the lower bound is applied separately at each sample size. Under , the product likelihood ratio satisfies
This is the product chi-squared divergence. By Cauchy-Schwarz inequality, . With , the proved metric two-point risk bound has normalized right side greater than . Consequently
The proof works for every despite the introductory . Measurability of the functions defining the probability density functions is understood. This establishes the requested lower bound; it does not assert a matching upper bound over this unrestricted class.
For fixed , define the left-limit sigma-algebra by . The section at of a generating rectangle of the predictable sigma-algebra is when , and empty otherwise. In the first case . Sections of the time-zero generators are empty for .
The collection of subsets of sample-time space whose section at belongs to is a sigma-algebra: sections commute with complements and countable unions. It therefore contains all of . Apply this to for every Borel set . Thus predictable sections at deterministic times satisfy
At the same argument gives -measurability, using the convention . The PDF asks for information strictly before ; the TeX aid drops the minus sign.
A discrete random variable on a probability space is a real-valued measurable function whose probability distribution is concentrated on some finite or countable set :
Measurability means for every real . Its probability mass function is , and . Requiring a countable image everywhere, rather than outside a null set, is an equivalent convention after harmless modification on a null set.
True in both cases. For every open , continuity of makes open. Then
is Lebesgue measurable by measurability of . The value has no bearing on this preimage argument.
Lebesgue measurability of means that is a Lebesgue measurable subset of for every open subset . Equivalently one can check all sets with real .
There are two conventions for a step function on the whole real line. Under the bounded-support integration convention it is a finite linear combination of indicators of bounded intervals; under the unrestricted piecewise-constant convention nonzero constant tails are allowed. The conclusions below state both conventions explicitly, since only the first assertion depends on this choice.
Solutions to every part of all three questions of the probability, measure and martingales course of the University of Oxford Part B examination, sat on 5 June 2023. Source: original released PDF, bundle pages 288–291; paper pages 1–4. The native PDF confirms the endpoint defect in Question 2: its slope formula evaluates outside when . We specify values on that null set below. Question 1(b)'s literal measurability assertion uses everywhere sign-valued random variables; for arbitrary versions with the stated probability distributions, it holds up to null sets, as explained in the solution.
For , fix signs . The constraints leave exactly two choices for the signs : choose freely, then set and for . Each input tuple has probability by independence. Consequently
Every finite subfamily is a marginal distribution of one such initial segment, so the entire sequence is independent, with fair sign probability distributions.
Using the usual everywhere sign-valued representatives, for every we have
The right side is measurable with respect to . Therefore is measurable with respect to their intersection .
Write . The Kolmogorov zero-one law applies to the independent random variables , so every event has probability zero or one. Also is independent of : this follows first for finite intersections of sign events from the given independence, then for their generated sigma-algebra by part (a).
The sets with and form a pi-system generating . If , the event has probability zero and is independent of every event in . If , it differs from by a null set, so the same independence follows from that of and . Another application of part (a) proves
In particular cannot contain the nontrivial event , whereas does. Thus adjoining a sigma-algebra need not commute with a decreasing intersection.
There is a small version qualification if the assumptions specify only the probability distributions, allowing exceptional values other than signs. Put when the denominator is nonzero, and otherwise. For , is measurable with respect to . The limit superior of , after scalar clipping to , is therefore -measurable and equals almost surely. Literal measurability of an arbitrary version of need not follow without choosing sign-valued representatives or adjoining the ambient null sets. All the independence conclusions are unaffected.