Accidental coincidence 2026-10-18
Suppose two independent background streams follow Poisson processes with rates and . Define a coincidence by : is the full accepted time-difference width. Around each first-counter signal, the expected number of second-counter signals is . The expected pair rate is therefore . A circuit producing at most one output per first signal instead gives
The distinction matters when multiple detections occur in one window. If a different convention accepts , its full width is and the dilute expression is .
For a chosen numerical model, take background rates , and . The dilute pair rate is . These are stipulated teaching inputs, not measurements of the historical chamber.
Making smaller also rejects some genuine events. Suppose true pairs arrive at rate before the timing cut, and their time difference has a centered normal distribution with standard deviation . Its accepted fraction, using the error function, is
For chosen and , the window retains about of true pairs. In the dilute background model the fraction of accepted events that are true is . A narrow window trades detection efficiency against background; it does not make either quantity equal to one. Delaying one counter's signal far beyond genuine event correlations gives a proposed background check under stationarity, without assuming that the zero-delay excess itself is accidental.
Amortization gap 2026-11-05
The amortization gap measures how much evidence lower bound is lost by using the distribution returned by amortized inference instead of the best distribution available for that observation in the same variational family. Hold the generative parameters fixed. Write for the evidence lower bound and suppose the optimum is attained. For the shared rule ,
The remaining variational approximation gap is , so the total loss is
Thus a restrictive recognition network or incomplete training can leave an amortization gap even when the variational family can represent the exact posterior distribution.[ref]
For an original example, stipulate for equally weighted observations and . The variational family represents both posterior distributions exactly when fitted separately, so . Restrict the shared rule to : it can adjust the intercept but its slope is fixed at one. The Kullback-Leibler divergence between these unit-variance normal distributions is . Averaging the amortization gap over the two observations gives
The best shared intercept still loses nat on average. A current rule with loses one nat: half a nat can be removed by training this rule better, while the remaining half requires changing its fixed-slope restriction. This separates the variational family's expressive power from the shared mapping's capacity and current training state.
In a numerical comparison, a locally improved per-observation distribution need not be the global optimum . Its improvement supplies a lower bound on the true amortization gap; failing to find an improvement does not establish a zero gap. Comparisons must keep both and fixed.
An anisotropic atomic displacement parameter describes direction-dependent positional spread about an atom's mean position in a crystal structure. In Cartesian coordinates a physical displacement model has a symmetric covariance matrix
The six independent entries have units of squared length. The crystallographic components stored in CIF use a cell-dependent convention, including reciprocal-cell lengths in the structure factor attenuation; they must be transformed before being interpreted as Cartesian covariance entries.[ref]
For an original Cartesian example, let
The principal root-mean-square displacements are0.1,0.2 and0.3 angstroms. An atomic displacement ellipsoid displays those preferred directions; its displayed boundary also depends on the chosen probability level. Rotating the coordinate axes generally introduces off-diagonal entries without changing the physical spread or eigenvalues.
Under a centered multivariate normal distribution for , its contribution to the scattering amplitude is multiplied by
where is the Cartesian reciprocal vector in cycles per unit length. Thus reflections probing different directions can distinguish directional spread. A physically valid covariance is a positive semidefinite matrix.
The equivalent isotropic atomic displacement parameter is in Cartesian coordinates. In this example it is ; that average does not make the original distribution isotropic. Positional spread can include unresolved static variation as well as motion, and differs from uncertainty in the refined mean coordinate.[ref]
Aperture photometry Created 2026-10-14 Updated 2026-10-20
Aperture photometry estimates brightness by summing calibrated detector pixels in an extraction region and subtracting the estimated background in that region. For equally weighted pixels, let be their total accumulated signal in electrons, the estimated background electrons per pixel over the same exposure, and the exposure duration. Then is a background-subtracted electron rate. Kepler applied aperture photometry to calibrated pixels with a fitted sky background. [ref].
This measurement trades source collection against background: enlarging the extraction region can retain more starlight but includes more sky brightness and neighbouring sources. Photon shot noise, read noise, and uncertainty in background subtraction contribute to its uncertainty. Dividing by a consistently measured reference rate gives a dimensionless light curve; it does not supply an absolute observed radiative flux. If contaminating light remains, the reference includes it and the inferred exoplanet transit depth suffers transit dilution.
In the first toy figure, the source's reference rate is electrons per second, the summed sky contribution is electrons per second, and a known multiplicative response is with in hours from mid-transit. The artificial exoplanet transit depth is , its total duration four hours, and its flat-bottom duration hours; the straight ingress and egress are a toy approximation, not an exact exoplanet transit photometry model. Each synthetic sample averages a -hour exposure and adds seeded normal distribution noise with relative standard deviation . Subtracting the known sky rate and dividing by the known response and reference rate recovers the dip. Real observations require estimating those corrections and carrying their uncertainty.
Consider a two-sided test rejecting when , with having a normal distribution of mean and variance one. Here is the cumulative distribution function of the standard normal distribution. Under a nonzero alternative , the type II error probability is the displayed function. It tends to the nonrejection probability under the null as , and to zero as . A particular observed has positive density under all these alternatives. Its null-calibrated p-value therefore does not select one value of . The power function describes repeated-sampling performance at specified alternatives.
An atomic displacement ellipsoid visualizes the directional positional spread represented by an anisotropic atomic displacement parameter. For a centered three-dimensional normal distribution with positive-definite Cartesian covariance matrix , its probability surfaces have the form
The eigenvectors of give the principal directions; if its eigenvalues are , the semiaxis lengths are . The atomic displacement probability level specifies how much modeled probability is enclosed.[ref]
The squared Mahalanobis distance has a chi-squared distribution with3 degrees of freedom in this Gaussian model. A50% ellipsoid uses . Taking instead encloses only about19.9%, so a three-dimensional probability ellipsoid cannot be read as a one-dimensional one-standard-deviation interval.
For an original example, gives50% semiaxes approximately0.154,0.308 and0.461 angstroms. An isotropic matrix with the same equivalent isotropic atomic displacement parameter, , instead gives a sphere of radius about0.332 angstroms.
Figure 1.
Equal equivalent displacement does not imply equal ellipsoid shape
. Original Cartesian Gaussian models at50% enclosed probability. Both have the same equivalent isotropic atomic displacement parameter, but only the right-hand distribution is isotropic. Axes show displacements from the mean, not atomic radii. Details.
The ellipsoid represents modeled positional spread, including possible static disorder, rather than the atom's physical surface or a confidence region for its refined mean position. Its interpretation depends on the displacement model; the Gaussian probability construction requires a positive-definite covariance and does not turn a negative fitted eigenvalue into a physically meaningful radius.[ref]
An atomic displacement probability level is the fraction of modeled positional probability enclosed by an atomic displacement ellipsoid. It specifies the size of the drawn surface for a given anisotropic atomic displacement parameter; changing the level does not change the fitted displacement matrix.[ref]
For a centered three-dimensional normal distribution with positive-definite Cartesian covariance matrix , define
The variable has a chi-squared distribution with3 degrees of freedom. To enclose probability , choose satisfying and draw . If are the eigenvalues of , its semiaxis lengths are .
An explicit calculation uses the cumulative distribution function
where is the error function. Solving gives ; for , .
For an original comparison, keep . The50% semiaxes are approximately0.154,0.308 and0.461angstroms. The90% semiaxes are approximately0.250,0.500 and0.750angstroms. Every semiaxis grows by the same factor, about1.625, although the underlying directional spread is unchanged.
This also explains why a surface drawn at one root-mean-square displacement in each principal direction, , encloses only about19.9% in three dimensions. The probability inside a three-dimensional ellipsoid differs from the probability inside a one-dimensional standard-deviation interval. Comparisons of displayed atomic sizes therefore need the stated probability levels as well as the displacement parameters.
For Brownian motion, write the later maximum as , where has a nondegenerate normal distribution and . The independent increments make independent of the earlier history and of . Conditional on those two objects, equality of the two maxima prescribes one value of , an event of probability zero. This argument uses the positive gap and does not assume independence of the two absolute maxima.
Autocorrelation Created 2026-09-28 Updated 2026-10-14
The autocorrelation function of a real weakly stationary process with positive variance is its normalized autocovariance:
It is dimensionless, with , , and by the Cauchy-Schwarz inequality. It is unchanged by replacing with for fixed . This makes autocorrelation useful when comparing records measured on different scales. The ACF is a property of the model; the sample autocorrelation function estimates it and will generally not reproduce theoretical zeros exactly.
An AR(1) with coefficient has autocorrelations at lags one through three. With these become : the same diminishing magnitude, alternating in sign. An MA(1) with instead has lag-one autocorrelation and zero at every higher lag. Equal numerical coefficients in an autoregressive model and a moving-average model do not mean equal dependence. See Bartlett, lecture 3: theoretical ACFs and prediction.
Figure 1.
Theoretical autocorrelation: AR(1) with coefficients 0.7 and -0.7, compared with MA(1) with coefficient 0.7
. Details.
The bars show exact model autocorrelations, not estimates from the illustrated sample paths in time series. The AR(1) retains a diminishing trace of a disturbance at arbitrarily large lags; the MA(1) stops sharing driving noise after lag one. A zero autocorrelation means uncorrelated random variables, and does not generally imply independent random variables. For a Gaussian process it does imply independence of the two coordinates.
There is a direct prediction interpretation. Among affine predictions based on just , the best linear predictor of is , with mean squared prediction error . Thus negative autocorrelation is also useful predictive information. A prediction using the whole observed past can do better; best linear prediction from a finite past describes that problem. Interpreting the affine prediction as a conditional expectation requires additional assumptions, such as a joint normal distribution.
After integrating out other variables, let include the constrained entropy and the measure appropriate to the chosen coordinate . Its equilibrium probability density function is proportional to . Near a stable interior minimum, with . The resulting normal distribution has variance . This is the leading Gaussian fluctuation approximation; a coordinate-dependent measure contributes to the effective , and higher-order terms matter near a vanishing curvature.
Bartlett correction for a normal variance with unknown mean Created 2026-10-14 Updated 2026-11-05
For a sample of size from a normal distribution with unknown mean, the null variance likelihood-ratio test statistic is , where . With , its expected Taylor terms start with . Here , , and . Higher fixed-order terms contribute only under valid termwise asymptotic integration. Thus and .
Basic inequality for the Lasso Created 2026-09-24 Updated 2026-11-05
Suppose and minimizes the Lasso objective . With , comparing its value with the value at gives
Indeed, expand and cancel . This deterministic inequality needs neither a normal distribution nor a full-rank design matrix.[ref]
The noise term is at most by Holder inequality. Controlling that term probabilistically leads to a Lasso cone condition; a Compatibility condition for the Lasso then converts coefficient error into prediction error. The stronger prediction inequality from Lasso stationarity uses subgradient optimality and has a different leading constant.
For any real separable Hilbert space , any orthonormal basis and independent standard normal random variables , the displayed series converges almost surely for each fixed and in L2 space. Indeed, the Parseval identity for a Hilbertian basis gives , so the partial sums form an L2 space-bounded martingale. The martingale convergence theorem supplies both limits. Every finite linear combination has the normal distribution with variance . Thus its finite-dimensional characteristic function is , independent of the chosen orthonormal basis. This proves equality of the finite-dimensional distributions of the resulting isonormal Gaussian processes; pathwise equality needs a coupling.
A Bayesian hierarchical model is a statistical model used in Bayesian statistics, in which parameters at one level have a prior distribution controlled by parameters at another level. Those shared hyperparameters can themselves receive a hyperprior. For grouped observations , group parameters and shared parameters , one common conditional-independence structure is
The hierarchy specifies how information can be shared; the data are not assumed to become independent merely because they are listed in different groups.[ref]
A normal distribution example makes partial pooling explicit. Let and . Conditional on known , completing the square gives the posterior mean
For , , , and , this is4, halfway between the group mean and the shared mean. More precise group data increase . In a full hierarchy with uncertain hyperparameters, their posterior distribution must also be integrated over; simply fixing fitted values omits that uncertainty.
A measurement application appears in Blaza Toman's2017 calibration study. Shared uncertain calibration inputs induce correlations among results; the Bayesian hierarchical model represents this structure within the model rather than treating all replicates as independent.[ref]
Bayesian normal model Created 2026-09-24 Updated 2026-11-05
A Bayesian normal model combines a normal distribution for observations with a prior distribution for its unknown parameters. For independent with known and prior , normal-normal conjugacy gives posterior variance . The posterior mean is the corresponding precision-weighted average of and the sample mean. Unknown observation variance requires an additional prior; treating it as known yields different uncertainty. The model makes prior information and measurement variation explicit.
When a person's latent slope has normal distribution , its th quantile is . Multiply by an elapsed time to obtain the corresponding latent change quantile. This describes population variation between people; it is distinct from a confidence interval for the mean slope and from a prediction interval that includes measurement error.
For with an independent equiprobable drift , the likelihood ratio between the two drifts is . Multiplying the normal distribution densities of successive increments proves this ratio for every observation partition; continuity then gives the full observed-path posterior. The Bayes' theorem gives . Thus the observed drift is . The process is an observed-filtration martingale with quadratic variation , so the Lévy characterization of Brownian motion makes it a Brownian motion. The observed state is a scaled diffusion with hyperbolic tangent drift.
Box-Muller transform 2026-10-06
Independent uniform distribution draws on produce independent standard normal distribution draws and . The proof transforms an independent uniform angle and radial density through the polar-coordinate Jacobian.
Bradley Efron Created 2026-10-18 Updated 2026-11-05
Bradley Efron's work connects statistical theory, computation and applications. His bootstrap estimates a statistic's sampling uncertainty by repeatedly sampling from the empirical distribution and recalculating that statistic. For example, resampling the observed data and recomputing a median produces a simulated sampling distribution without requiring a normal population model. Validity still depends on the statistic and sampling setting; the method is not automatically correct for every extreme-value or dependent-data problem.
His bootstrap work has distinct records as a July 1977 Stanford technical report and a January 1979 journal article.[ref][ref] His Stanford profile describes research spanning theoretical statistics, biostatistics and astrophysical applications.[ref]
Figure 1.
Portrait of Bradley Efron at the 2007 National Medal of Science ceremony
. Ryan K. Morris, National Science and Technology Medals Foundation; cropped by Wikimedia user Dudek1337. Public domain, as recorded on Wikimedia Commons. Source usage terms. Source.
At integer times, has the same standard normal distribution for every . The event that its limit superior exceeds a fixed finite level is a tail event of the independent unit-time Brownian increments, since changing any finite initial segment contributes only a term tending to zero. The Kolmogorov zero-one law applies. Its probability is positive because each fixed-time exceedance has the same positive probability, giving a positive probability of infinitely many exceedances by the decreasing union-of-tail-events argument. Thus its probability is one. Intersecting over integer levels proves the stated divergence, without the lower iterated-logarithm bound.