Topics (384k) Articles (398k) Users (410) Discussions (242) Comments (383) Files (13k) New article
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 6 c Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
All moments below are under the proper posterior distribution of part (b). Write the known covariance matrix in block formThe score covariance is positive definite in this model. Indeed, a further integration by parts gives . If , then almost surely, because . Multiplying by would give , so .
For a fixed coefficient vector , the control variate summand has varianceThus the unique minimum is attained at , givingThe coefficient is fixed because the covariance is assumed known, and the posterior score control variate has mean zero, so this estimator remains unbiased.
We can identify exactly when the improvement is strict. Integration by parts with the bounded smooth function givesThe scalar expectation is strictly positive. Therefore whenever , and positive definiteness gives a strictly smaller variance in that case. If , and both estimators already have variance zero. The printed request for a smaller variance therefore needs this nondegeneracy qualification; a non-increasing variance always holds.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 6 b Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Put and . The probit regression likelihood is , so with the stated normal prior,Differentiating gives the probit posterior scoreEquivalently, the th likelihood term is . The prior's gradient is essential.
This posterior density is smooth, positive everywhere, and bounded above by a constant times the Gaussian prior, since the likelihood is at most one and has positive normalizing constant. Its derivatives are integrable: differentiating a likelihood factor gives a bounded normal density, and differentiating the prior gives a linear factor times a Gaussian. Integration by parts therefore yieldsThus each component has a known zero expected value and can be used as a posterior score control variate. This identity differentiates the posterior density in the random parameter; it is distinct from the usual mean-zero score identity for a likelihood under repeated sampling.
The requisite second moments are finite. For , is bounded. For ,Hence the score grows at most exponentially in , and the Gaussian upper bound on the posterior gives finite second moments. For a fixed vector , averaging is therefore an unbiased Monte Carlo estimator of the same mean; suitable coefficients can reduce its variance.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 6 a Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Let . To distinguish the number of posterior draws from the training-sample size, write it as ; in the question's sample notation . The Monte Carlo estimator of the requested posterior mean isFor independent exact posterior draws it is unbiased, andThe features are conditioned on as fixed data. The estimated probability concerns a future genuine banknote under the posterior predictive probability, not an indicator sampled from that predictive distribution.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 4 c Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Fix the observed , let , and let . The E-step formsFor finite expected log-likelihoods and , the Jensen inequality givesThe first inequality allows additional mass outside the old complete-data support; it is equality when the relevant supports coincide. Ratios are taken on the support of . If a candidate assigns zero density on positive -mass, its is minus infinity and it cannot be an improving finite M-step.
An M-step of the expectation-maximization algorithm chooses with , soThis EM likelihood monotonicity also holds for a generalized M-step that merely increases , rather than maximizing it exactly. It is a statement about the observed-data likelihood function.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 4 b Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
For one factor, write for the loading vector and . Let be its current value and . The E-step of the expectation-maximization algorithm uses the conditional multivariate normal distribution to obtainThus . The terms of the expected complete-data log-likelihood that depend on a candidate areThis is a strictly concave quadratic in , since . Its unique maximizer isFor an explicit expression involving only the current loading and data, put . Substitution gives the EM update for a single-factor Gaussian modelNo empirical centering is applied because the model fixes the mean to zero. If initialized at zero, this update remains at zero; monotonicity alone does not guarantee that such a stationary point is a global maximum. This part is present in the PDF but missing from the TeX transcription.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 4 a Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
In factor analysis, the independent multivariate normal distributions of and imply that their linear combination is also multivariate normal. Its expected value is zero and its covariance matrix isThereforeThe noise variance is assumed positive. The factor contribution is positive semidefinite and has rank at most , while the noise makes the marginal covariance nonsingular.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 3 a Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Write , , and . We keep the model's printed coefficient literally. Its latent log odds can be writtenThe independence of the is conditional on : they are not marginally independent when the random effects are correlated.
Holding the other predictors and the random-effect realization fixed, increasing predictor by one shifts the latent log odds by , so it multiplies the conditional odds of a late departure by . For a binary snow indicator, this compares the snow and no-snow days with other predictors held fixed. This is a conditional odds ratio, not generally the same as a marginal odds ratio after integrating the random effects.
Randomizing allows unmeasured daily conditions to change the success probability beyond the systematic linear predictor. In this logistic-normal regression with autoregressive random effects, let and . The law of total variance givesso the mixture allows overdispersion beyond a binomial distribution with fixed probability. The correlated part also allows related outcomes on nearby days.
The stationary autoregressive process of order one has and . HenceThe term is unstructured daily variability, whereas is the temporally correlated component under the printed parameterization. Larger gives greater persistence: the lag- correlation of the latent log odds is .
The PDF itself prints both the coefficient and the requested combination . They are inconsistent as a marginal variance. If the intended coefficient were , the variance and off-diagonal covariance would instead be and . The prior's argument list also repeats ; its displayed density indicates the intended second variance parameter is .
There is a further substantive issue with that improper prior. After integrating the latent variables, the observed-data likelihood function is positive and continuous at for fixed finite , positive , and . On a compact positive-volume set of those parameters it therefore has a positive lower bound for small . The prior density is , so integrating over gives . Thus the stated joint posterior distribution is improper, an instance of improper posterior from a log-uniform random-effect scale prior. The requested fixed-parameter conditional laws below are nevertheless proper; their existence does not remedy the joint impropriety.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 2 d Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Let and let be the uniform distribution on the body. The hit-and-run kernel density lower bound impliesFor the last range, lies in a radius- disc about any of its points, so and . This global Doeblin condition proves irreducibility with respect to and aperiodicity. The entire state space is a small set; , , and satisfy the geometric drift condition.
In fact the conclusion is stronger. Write , where and is the residual Markov kernel. Since , also . Each transition has a probability of resetting to stationarity, and after such a reset every subsequent distribution remains stationary. Equivalently,Thus the algorithm has uniform geometric ergodicity, using the full-dimensional and boundary conventions stated in part (a).
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 2 c Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
The geometric drift condition requires a measurable , a small set , constants and , andSmallness means that for some , , and probability measure , for every . With irreducibility and aperiodicity, this drift-minorisation condition implies geometric ergodicity. The function controls the chain away from the small set; inside that set the additive constant allows unrestricted bounded drift.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 2 b Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Write and letThis is the accepted fraction of directions when zero-length chords are redrawn; it equals one in the interior. The Markov kernel integrates a test function asThe signed polar representation has Jacobian determinant . Each point is represented twice, once with positive and once with negative and the opposite direction. Since the chord is the same in either representation, the hit-and-run kernel density isfor almost every , and it is zero outside the body. Thus the transition is absolutely continuous with respect to planar Lebesgue measure. Redrawing removes the only potential zero-chord atoms.
Let . For distinct , the chord contains their segment, so it has positive length, andHence . Values on the diagonal, and on other null sets where a density is only specified almost everywhere, can be assigned this same lower bound. This gives a version satisfyingFor interior , and the chord length is unchanged on interchanging its endpoints, so . A convex body's boundary has area zero; for example contractions about an interior point fill the interior and their areas tend to the body's area. Thus this symmetry holds almost everywhere for two uniform points, proving detailed balance and stationarity of the uniform distribution. Without positive area or the boundary convention in part (a), the literal assertion need not hold.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 2 a Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
For the planar uniform distribution with a Lebesgue measure density, assume that the compact convex set has nonempty interior, equivalently positive area. This assumption is missing from the wording: a line segment or singleton would not have such a planar uniform density.
The hit-and-run sampler at a current point chooses with uniform on , finds the intervaland samples uniformly on that interval, setting . Convexity makes the intersection an interval and compactness makes it bounded.
If is in the interior, every direction has positive chord length, and the next point is in the interior almost surely. For a definition valid at every boundary point as well, redraw the direction whenever the chord has zero length. We use this convention in the following parts. Positive-length directions have positive angular probability at every , since an interior ball is visible from , so the redraw terminates almost surely.
The Markov kernel has the uniform distribution on as its invariant law, as follows from the density symmetry proved next. Repeating the step therefore produces approximate uniform samples after convergence. The boundary convention matters: if singleton chords were instead treated as holding moves, a square's corner would have a positive atom at its starting point, contradicting part (b)'s claim of an everywhere absolutely continuous kernel.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 1 c Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Let and . Since , . The Cauchy-Schwarz inequality applied to the centered sums givesDivide by . The first term is at most , while part (b) and give a limsup of at most for the second. Squaring provesThis argument also covers ; it does not require independence between the discarded prefix and the retained path. The same chain-dependent as in part (b) suffices.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 1 b Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Put , , and . We first obtain a uniform-in- stationary bound, then compare the burned-in trajectory with a stationary one.
For a reversible Markov chain, geometric ergodicity gives an absolute L2 spectral gap of a reversible Markov chain. Thus, for some , the Markov kernel viewed as an operator satisfiesOne can use the rate in the total-variation estimate. To see why, reversibility makes a self-adjoint operator. For bounded mean-zero supported on a set where is bounded,The spectral theorem for normal operators writes the left side as . Any positive spectral mass where would contradict the inequality as . These test functions are dense in : truncate a function and restrict it to , then correct its mean using a bounded function supported on a fixed positive-measure set where is bounded. Hence the spectral restriction extends to the entire mean-zero space.
Let be the stationary chain. For , the stationary covariance bound for reversible Markov chains givesConsequently
The initial law of the burned-in trajectory is . Its total variation distance from is at most . Attaching the same conditional future evolution cannot increase that distance, so the entire length- path laws differ by at most . For either path the sum has . The dual bound for total variation distance gives a difference of at most in the second moments, and at most in the squares of the means. This burn-in total variation comparison yieldsSince , . Dividing by givesThe constant depends on the chain, not on .
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 216 1 a Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Let be the Markov kernel and its invariant probability measure. Using the conventiongeometric ergodicity means that there are and a finite measurable state-dependent prefactor such thatHere total variation distance is used, and the rate does not depend on . The prefactor need not be uniformly bounded; that stronger property is uniform geometric ergodicity. For a deterministic starting state, this estimate must hold at that state. If the definition is made only -almost everywhere, the subsequent assertions concern those starting states where the prefactor is finite. The reversible Markov chain assumption means .
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 214 4 d Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
On a tree, every finite connected induced subgraph is itself a tree, and its only spanning tree contains all its edges. Thus the free uniform spanning forest is deterministic:Suppose for a contradiction that is a transient graph. By part (c), choose whose removal leaves two transient graph components . Let be their effective resistances to infinity measured from their respective endpoints.
In a wired finite exhaustion, the edge of resistance one is in parallel with the route from through to the wired boundary and back through to . The latter route has resistance , converging to . Consequently the limiting wired effective resistance isThe edge-inclusion formula for a uniform spanning tree, the one-edge case of the transfer-current theorem, states that an edge of conductance is present with probability . Passing to the wired limit and using givesBut under the free uniform spanning forest, is present with probability one. This contradicts . Therefore the tree is recurrent. Combined with part (b), this characterizes equality of the free and wired uniform spanning forests on locally finite unweighted trees.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 214 4 c Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Assume first that deleting an edge leaves two transient graph components. One component supports a finite-energy unit flow from its endpoint to infinity by the finite-energy flow criterion for transience. Extend that flow by zero across the deleted edge and throughout the other component. It remains a finite-energy unit flow on the whole tree, so the tree is a transient graph.
Conversely, root a transient tree at and take its finite-energy unit electrical flow to infinity. On a tree this flow can be chosen nonnegative in every direction away from the root: it is obtained as the limit of the currents to wired boundaries, each of which sends nonnegative current into descendant subtrees. At each nonroot vertex its incoming current equals the sum of its outgoing currents, by flow conservation.
If every vertex reached by positive current had exactly one positive-current child, the unit current would run along a single infinite ray. Every edge of that ray would carry current one, so its energy of a flow would be , a contradiction. Thus there is a vertex with two positive-current child edges, say and .
The restricted flow below , rescaled by its incoming current, is a finite-energy unit flow from to infinity; hence the component below is a transient graph. After removing , the other component contains , the edge , and the descendant subtree below . Send unit current along and then use the rescaled descendant flow below , with zero current elsewhere. Its energy of a flow is finite, so this component is a transient graph too. ThereforeUnit edge resistances are essential here: an infinite ray with sufficiently rapidly decreasing resistances can be transient without any such splitting edge.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 214 4 b Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Fix a root . On a recurrent graph, a simple random walk started at any vertex hits almost surely. Run Wilson's algorithm with root : successively attach the loop-erased random walk from each new starting vertex, stopped when it hits the existing tree. Every walk terminates almost surely, because that tree already contains .
To compare the two finite approximations, fix a finite set of edges and put all their endpoints first in the ordering of starting vertices. In the infinite graph, these finitely many Wilson's algorithm walks have finite lengths almost surely. Their visited vertices, together with their neighbors, are therefore contained in for all sufficiently large on each realization.
Couple the walks in , in , and in using the same neighbor choices while they are in the interior of . On the event just described, they encounter neither boundary, so the walks, their chronological loop erasures, and the resulting partial trees agree. Once every endpoint of has entered the partial tree, later walks cannot add any edge of : they stop on their first hit of the existing tree. Thus the indicators of membership for all edges of agree in the free and wired uniform spanning trees, with probability tending to one.
The two limiting measures agree on every finite-edge event, and thereforeThis argument also shows that the common uniform spanning forest is a single spanning tree on a recurrent graph. The equality refers to probability measures, not to independently sampled forests being identical.
Past exam of the mathematics course of the University of Cambridge 2018 iii Paper 214 4 a Solution by
Codex 0 Created 2026-10-03 Updated 2026-10-05
Use the usual implicit convention that the infinite connected graph is a locally finite graph. This is needed for the simple random walk and the finite wired constructions to be defined. Choose finite vertex sets exhausting , with each induced subgraph connected.
For the free uniform spanning forest, sample a uniform spanning tree of , using only the internal edges. Its limiting law on edge subsets of is the free uniform spanning forest measure:The limit is weak convergence of probability measures on , equivalently convergence of every finite-edge event.
For the wired uniform spanning forest, form by identifying every vertex outside to one boundary vertex . Retain all edges with at least one endpoint in , preserve parallel edges, and discard loops at . Sample a uniform spanning tree of this finite multigraph and restrict it to the original internal edges. The limiting law isFor each fixed finite set of edges the restriction is unambiguous once is large. The standard exhaustion theorem guarantees that both limits exist and do not depend on the chosen connected exhaustion. Both are random spanning forests; despite their names, they need not be connected.
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!
Intro to OurBigBook
. Source. We have two killer features:
- 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-calculusArticles 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/derivativeVideo 2. OurBigBook Web topics demo. Source. - 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.
- to OurBigBook.com to get awesome multi-user features like topics and likes
- as HTML files to a static website, which you can host yourself for free on many external providers like GitHub Pages, and remain in full control
Figure 2. You can publish local OurBigBook lightweight markup files to either OurBigBook.com or as a static website.Figure 3. Visual Studio Code extension installation.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. - Infinitely deep tables of contents:
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





