The posterior distribution has a long right tail. Its posterior mean is pulled upward by relatively rare large endpoints and is sensitive to the artificial upper cutoff; its untruncated posterior mean does not exist. The median is a central quantile that stays finite as the cutoff is removed and is much less sensitive to extreme draws. It is also the Bayes estimator under absolute-error loss, whereas a mean corresponds to squared-error loss when the relevant expectation exists.
The median better represents the typical posterior endpoint in this strongly right-skewed distribution. It need not be preferable for every decision problem, since the loss function matters. For comparison, exact summation of the truncated kernel at cutoff gives median and mean about ; the displayed simulation summaries are Monte Carlo estimates, not exact posterior calculations.
The WinBUGS model uses a finite truncation of the reciprocal prior. The numbered lines have the following roles:
Combining that categorical distribution observation likelihood with the truncated reciprocal prior produces
The Markov chain Monte Carlo output approximates this posterior. The actual PDF separates the observation line from the preceding comment; it must remain an active stochastic statement.
Multiplying the reciprocal improper prior by the likelihood function gives
Thus this discrete uniform endpoint posterior is proper. Its exact cumulative probability is . Integral approximation gives and , so
These approximations are useful when is moderately large and are not exact discrete identities. For example the exact untruncated median at is . Although its quantiles are finite, its posterior mean is infinite, since its expectation numerator is the divergent sum .
Keep fixed. Integral bounds for the decreasing function give , so . Consequently
The posterior mean thus grows like instead of approaching a data-determined limit. Even the posterior distribution mass below a fixed endpoint is . Hence increasing the supposedly innocuous prior cutoff moves the inference to larger and larger endpoints. In an integral approximation its median is of order , also depending strongly on the cutoff. This explains the sensitivity of the entire inference, not just of its posterior mean.
Write , with . The truncated uniform prior gives the posterior distribution
Its posterior mean is
Each term in the numerator of the expectation is , which explains the exact count .
A constant improper prior would give the formal posterior density kernel . However,
There is no normalized posterior distribution under this prior. The failure is not simply that the prior is improper: some improper priors give proper posteriors. Here posterior propriety itself fails after this single observation, so probabilities, quantiles and a posterior mean cannot be assigned to that kernel.
On its admissible support the likelihood function strictly decreases. Its maximum-likelihood estimate is consequently the smallest admissible endpoint:
Under the discrete uniform distribution, the observation has probability when its value is an available label, and zero otherwise. Therefore the likelihood function is
This is the discrete uniform endpoint posterior model's likelihood; all endpoints smaller than the observed label are excluded.
For the uniform prior, write for the pooled success count in trials. Conditional on , exchangeability makes all placements of the successes equally likely. The successes in the second batch therefore have a hypergeometric distribution, and the factor in question is
This is hypergeometric allocation of exchangeable Bernoulli counts: sample positions without replacement from a population of positions containing successes. The uniform-count predictive property gives and , so
As varies with fixed, also varies. Therefore alone is not a normalized hypergeometric distribution over ; its normalization factor is essential.
Before collecting any observations, a uniform success-chance prior distribution gives
Thus the uniform-count predictive property assigns equal probability to every possible number of successes in a fixed-size batch under the prior predictive distribution. It is uniform over counts, not over all ordered binary sequences. After observing successes in trials, its familiar one-step posterior predictive probability is ; the count-uniform property refers to prediction before those data are observed.
With a uniform beta distribution prior, the posterior predictive distribution for another batch of the same size is
But . Hence
The displayed expression is symmetric in . This is the uniform-count predictive property: the joint batch counts are exchangeable and both marginal counts are uniform. Conditional reversal is not a general property of arbitrary priors; it holds here because those marginal probabilities coincide.
Put , , and . Integrate the future binomial distribution against the current beta distribution posterior density:
This is the beta-binomial distribution. For positive integer , the gamma function values reduce to factorials, giving
Using and rewrites this as
Substitution of gives exactly the two required factors. The probability is zero outside .
For the independent Bernoulli trials, the log likelihood function, apart from its constant, is . Its negative expected second derivative, the Fisher information, is
For , the Jeffreys prior is proportional to its square root. Since , its normalized form is
The Jeffreys prior is invariant under smooth one-to-one reparameterization as a measure. If , then , so . Thus recomputing it in the new coordinate gives precisely the transformed prior, rather than a new prior. In several dimensions the corresponding density is , with the same Jacobian transformation property.
The binomial likelihood is . Multiplication by the beta distribution prior density and Bayes' theorem give the posterior density kernel . Normalizing by the beta function yields Beta-binomial conjugacy:
This is proper for and , including counts at either boundary.
Represent a feasible allocation by a label for each item saying which bidder receives it. Every such label vector gives disjoint bundles covering all items, so the following heuristic optimization methods can search without violating allocation feasibility.
In local search, begin with any feasible partition and repeatedly make an improving move, such as transferring one item between bidders or exchanging two items. For a transfer , evaluate the change
Stop when no allowed move improves the objective. Several starting partitions can avoid dependence on one local optimum, but no guarantee of a global optimum follows for general bids.
In simulated annealing, use the same feasible neighborhood but also accept a worsening move of loss with probability at temperature . Begin with a relatively high temperature, gradually cool, and retain the best allocation seen. Worsening moves can cross a barrier between local optima; a practical finite cooling schedule does not guarantee optimality.
In tabu search, retain a short memory of recent transfers or exchanges and temporarily prohibit their reversal. Select the best admissible move, possibly worsening the objective, to escape a local optimum. Permit an otherwise tabu move if it improves the best solution found, an aspiration rule. For the winner determination problem, bidder/item pairs or recent reverse transfers provide convenient tabu attributes. All three methods preserve the partition, but for arbitrary bids they are heuristics rather than algorithms with a proved constant approximation ratio.
The log-likelihood in the normal linear model, up to an additive constant, is
Full column rank makes positive-definite, so minimizing the residual sum of squares gives the normal equations . Maximizing over afterward gives the maximum-likelihood estimators
The residual sum of squares is positive almost surely because and , so this maximization has an interior variance solution almost surely. The first estimator is a linear Gaussian statistic with mean and covariance , giving
Let be the orthogonal projection onto the design column space, and let . This is a symmetric idempotent matrix of rank . Choose an orthonormal basis for its range and collect its columns in , so , and . Then
The definition of the chi-squared distribution therefore gives
The maximum-likelihood denominator is ; the unbiased residual-variance estimator uses .
Finally put . It has rank and , because . Part (a) implies that is independent of . Multiplication by a fixed constant preserves independence, so the normal linear model maximum-likelihood sampling distributions satisfy
For every nonzero , full column rank of gives , and hence
Thus the Gram matrix is a positive-definite matrix and is invertible. Using and ,
so
A linear image of a multivariate normal vector is again multivariate normal. Therefore the stacked vector has mean and covariance
In particular,
The two diagonal blocks are positive-definite by the rank assumptions. The two blocks are jointly Gaussian and have zero cross covariance. The fact that uncorrelated jointly Gaussian variables are independent makes and independent. Since
is a measurable function of the second block, independence is preserved under this transformation. This proves Gaussian independence of linear and quadratic statistics:
An open migration process has finitely many colonies and state . External arrivals to colony form independent Poisson processes with rates . When , individuals leave that colony at total rate , with . A departing individual moves to with probability or exits with probability . The routing choices and all arrival and service randomness are independent. The routing matrix is substochastic and transient: every individual eventually exits, equivalently its spectral radius is less than one. Self-routing is allowed, but creates no state change.
The traffic equations of an open migration process, using row vectors, are
The inverse matrix exists because converges; its entry counts expected visits to by an individual entering at . Thus includes both external arrivals and internal migrations. Restrict first to the accessible colonies with .
Set
If each , the candidate product-form stationary distribution of an open migration process is
Here is a global balance for a continuous-time Markov chain proof, which does not incorrectly assume that the routing is reversible. The weight ratio is . After dividing the incoming balance equation at by , external arrivals contribute , exits contribute , and migrations from to contribute . Combining the first and third contributions using the traffic equations of an open migration process yields . Summing those same traffic equations of an open migration process gives . The incoming rate is therefore exactly the outgoing rate
This proves stationarity. With finite arrival rates, finite service rates in each state and transient routing, the process is nonexplosive: only finitely many individuals enter in a finite interval, and each makes finitely many routing transitions before exit, almost surely. On the accessible irreducible Markov chain, a finite normalizer gives the unique stationary distribution and positive recurrence. Colonies with zero effective arrivals are empty in this equilibrium. For a nonempty accessible colony, a divergent prevents normalization of the equilibrium weights.
For the single-server specialization, for . Each colony has the stationary distribution of an M-M-1 queue:
requiring on every colony with positive traffic. The service rates do not change the traffic equations of an open migration process.
For positive , write and . A stable allocation requires and , with . The Cauchy-Schwarz inequality gives
Equality holds precisely when is proportional to . Hence the unique optimal service-capacity allocation in an open migration process and its mean population are
There is no stable allocation when and some traffic is positive. If some , those empty colonies contribute no population; allocate the spare capacity to the positive-traffic colonies by the same rule. Allowing zero capacity at an unused colony attains this optimum. If strictly positive are required even there, the displayed value is only an infimum, approached as their capacities tend to zero. If every , the equilibrium is empty and any admissible allocation minimizes its mean population at zero.
Put . In this heterogeneous two-server queue, the total population alone is not a Markov process: when only one customer is present, its departure rate depends on which server is occupied. Use states , where and distinguish the two single-customer possibilities, and records the total population. The memoryless property of the exponential distributions gives these nonzero transition rates:
and for , for . In particular, a departure from state leaves the other server occupied.
The detailed balance equations suggest
For example, both and hold; these are the two transitions that would be lost by using only population as the state. All other adjacent-state detailed balance equations hold as well. Since , the geometric series converges, and normalization gives
Thus the stationary distribution exists, and the requested probability is
With equal server rates , this reduces to , as for an M-M-s queue with two servers.
To prove the output claim, put the continuous-time Markov chain in its stationary distribution. The preceding detailed balance equations make it a reversible Markov chain: time reversal of a continuous-time Markov chain has the same transition rates. A forward departure becomes a reversed arrival. In every reversed state, the total rate of such arrivals is : from the rates add, from or the upward rate is , and from every it is also . The reversed process can therefore be constructed with an independent rate- Poisson process of arrival epochs, choosing the idle server by a fair coin when appropriate. Consequently reversed arrival counts on disjoint time intervals are independent Poisson random variables with the corresponding means. Reversing those intervals proves that the stationary departure stream is a rate- Poisson process, rather than merely a stream with mean rate .
Stationarity matters here. If the queue starts empty, a departure by time requires both an arrival and a completed service time, so its probability is ; a rate- Poisson process would have probability . Thus the output assertion is the equilibrium assertion in the context of the preceding stationary calculation.
Fix and . For the construction take , the range relevant to localization at zero, and define
This drift is bounded by and globally Lipschitz with constant at most . Together with the constant diffusion coefficient , the global existence theorem for stochastic differential equations with Lipschitz coefficients gives a unique strong stochastic solution of
for the prescribed driving Brownian motion. Let and choose as the stopped continuation. Before this time the drift is exactly . Uniqueness here means uniqueness up to the stopping time; the equation imposes no restriction on arbitrary continuations after it.
For , the two equations have the same coefficients while both solutions exceed . By local pathwise uniqueness they coincide until that level is reached. Thus the first hit of by is , and its later first hit of satisfies
Patch along a deterministic sequence : for , choose with and set . Compatibility makes the definition independent of . It gives a continuous adapted positive process satisfying the original equation on every compact time interval before . The same compatibility holds for other threshold choices, so the lifetime is the limit as , not a property of the chosen sequence.
To prove , put and . Its generator vanishes. On each finite horizon, Itô's formula makes a true martingale, because its stochastic integrand is bounded by . Positivity therefore implies
Since is contained in each , letting gives . Taking a countable union over integer proves almost surely. This is the truncation construction of a positive Bessel strong solution.
For two positive solutions driven by the same Brownian motion, localize at their first hits of . The cutoff equations and pathwise uniqueness make them equal up to those times. On every finite interval both positive continuous paths have positive minima, so decreasing removes the localization and proves indistinguishability. Thus
The resulting global solution also supplies the stopped solution for any threshold , completing the literal all-threshold existence request. The monotonicity used to construct the zero-boundary lifetime concerns the thresholds tending to zero; upper-threshold hitting times have the opposite ordering. Strong adaptation is preserved because every cutoff solution is strong and the patch uses their stopping times in the given filtration.

Pinned article: Introduction to the OurBigBook Project

Welcome to the OurBigBook Project! Our goal is to create the perfect publishing platform for STEM subjects, and get university-level students to write the best free STEM tutorials ever.
Everyone is welcome to create an account and play with the site: ourbigbook.com/go/register. We belive that students themselves can write amazing tutorials, but teachers are welcome too. You can write about anything you want, it doesn't have to be STEM or even educational. Silly test content is very welcome and you won't be penalized in any way. Just keep it legal!
We have two killer features:
  1. topics: topics group articles by different users with the same title, e.g. here is the topic for the "Fundamental Theorem of Calculus" ourbigbook.com/go/topic/fundamental-theorem-of-calculus
    Articles of different users are sorted by upvote within each article page. This feature is a bit like:
    • a Wikipedia where each user can have their own version of each article
    • a Q&A website like Stack Overflow, where multiple people can give their views on a given topic, and the best ones are sorted by upvote. Except you don't need to wait for someone to ask first, and any topic goes, no matter how narrow or broad
    This feature makes it possible for readers to find better explanations of any topic created by other writers. And it allows writers to create an explanation in a place that readers might actually find it.
    Figure 1.
    Screenshot of the "Derivative" topic page
    . View it live at: ourbigbook.com/go/topic/derivative
  2. local editing: you can store all your personal knowledge base content locally in a plaintext markup format that can be edited locally and published either:
    This way you can be sure that even if OurBigBook.com were to go down one day (which we have no plans to do as it is quite cheap to host!), your content will still be perfectly readable as a static site.
    Figure 5. . You can also edit articles on the Web editor without installing anything locally.
    Video 3.
    Edit locally and publish demo
    . Source. This shows editing OurBigBook Markup and publishing it using the Visual Studio Code extension.
  3. https://raw.githubusercontent.com/ourbigbook/ourbigbook-media/master/feature/x/hilbert-space-arrow.png
  4. Infinitely deep tables of contents:
    Figure 6.
    Dynamic article tree with infinitely deep table of contents
    .
    Descendant pages can also show up as toplevel e.g.: ourbigbook.com/cirosantilli/chordate-subclade
All our software is open source and hosted at: github.com/ourbigbook/ourbigbook
Further documentation can be found at: docs.ourbigbook.com
Feel free to reach our to us for any help or suggestions: docs.ourbigbook.com/#contact