Aperture-scan separation of direct and diffuse nail transmission estimates the directly collected and diffusely transmitted fractions of light by changing detector acceptance while keeping the nail plate and illumination fixed. Separation depends on a stated angular model: a change in aperture is an experimental control, not a direct label attached to every photon.
In the original constant-diffuse-radiance model, let , where is the fully accepted direct fraction, the total diffuse fraction and the known diffuse acceptance. For a coaxial circular opening, the geometric approximation gives . Two distinct acceptances give
The distinction between these components is identifiable within this model only when the acceptances differ.
Suppose , , and . The fitted fractions are and . If independent signal errors each have standard uncertainty , with acceptance uncertainty temporarily neglected, linear propagation gives
and
The resulting uncertainties are about 0.00943 and 0.00359. Their negative covariance matters when adding the components: the total-transmission uncertainty is about 0.00629, rather than the value obtained by incorrectly treating the fitted components as independent.[ref]
Additional apertures give a straight-line fit of against , with intercept and slope . Weighted least squares can use independently characterized signal uncertainties. Systematic curvature would challenge the assumed angular law or fixed direct-beam acceptance. A plausible line alone does not prove that a finite source, changing geometry or residual fluorescence has been modeled correctly.
The primary nail-transmission study collected transmitted light using an integrating sphere, monochromator and photodiode.[ref] The aperture scan here is an original proposed comparison, not a procedure or angular scattering law reported in that experiment.
An integer pattern must survive the stated measurement uncertainty. For an original simulated example, choose a reference electric charge , define , and stipulate state means
Here each is an independent standard uncertainty of a state mean, before shared calibration effects. The generator constructs repeated noisy observations with these means; the numbers are deliberately chosen, not historical data. There is no conversion of to coulombs and no measured value of the elementary charge in this example.
For a candidate spacing in units of and proposed integers , use weighted least squares:
The last two expressions follow by differentiating the quadratic in and propagating the independent errors. They are conditional on fixed, correct integer assignments and the stated error model. They do not include uncertainty from choosing those assignments. For , they give , , and . Such a small residual is compatible with this chosen error scale; it is not evidence that the errors vanish. A chi-squared distribution interpretation additionally needs the appropriate normal distribution error model and allowance for the fitted parameter; searching many assignments changes that interpretation.
Figure 1.
Finite noisy electric charges fit a candidate unit and its subdivisions
. Original calculation for the simulated state means above. Horizontal bars show one standard uncertainty. Both spacings and have exactly the same fitted locations and residuals when the integer labels double. The lower curve minimizes over integers separately at every candidate spacing; a low residual alone cannot identify the elementary charge. Details.
The obstruction is identifiability: replacing by for any positive integer leaves every predicted reading unchanged. In particular, an -spaced pattern also fits an lattice using even indices. Even the coarsest well-fitting spacing need not be the physical unit: a dataset containing only even charge states can have twice that unit. Worse, if arbitrarily fine spacings and unrestricted integers are allowed, rounding each gives
Consequently a finite fit cannot prove a minimal transferable electric charge. Reproducible smallest observed jumps supply additional evidence; interpreting them as single-carrier transfers requires independent physical support. One visibly abrupt transition could contain several unresolved transfers. In the simulation the successive differences are , , and , each with standard uncertainty . These support changes near conditional on that unit; they equally support units of . Neighboring differences share a state reading and are therefore correlated, even though the state means were stipulated independent.
A shared scale error can preserve every apparent integer relation. To first order, an independent fractional scale standard uncertainty adds to the covariance matrix of the . For the simulation, stipulating changes the conditional uncertainty of to approximately , despite the same residual pattern. Repeating timings cannot average away this common scale contribution. In the uncorrected continuum model in elementary charge, holding the other measured inputs fixed gives , so a small fractional error in air dynamic viscosity causes times that fractional error in inferred electric charge. The gas slip correction to Stokes drag requires its own propagation. Pattern agreement, drop-stability controls and absolute measurement calibration together make the inference persuasive within resolution.
Effective kernel weight Created 2026-09-24 Updated 2026-11-05
The effective kernel weights express a local polynomial regression fit as a linear combination of the responses. At target , set , , and . If the local polynomial Gram matrix is invertible, then
where . This follows by solving the weighted least squares normal equations and taking the fitted intercept.
The effective kernel weights are not the raw weights . They depend on the whole local design and satisfy the polynomial reproduction identities and for . Even when , individual can be negative. For a local linear regression at through observations at , the weights are , giving intercept . The negative weight is what permits exact extrapolation of a straight line; forcing all weights positive would lose that property at this target.
G. Fermi 2026-11-05
G. Fermi is the first author of the 1984 study refining human deoxyhemoglobin at resolution, and is named with Max Perutz as a depositor of its primary structure record, 4HHB.[ref] The published initial-only identity is retained.
The model allowed the investigators to examine local geometry around iron in the oxygen-binding groups of hemoglobin. Their analysis also tested whether conclusions about that geometry persisted when selected model restraints were removed.[ref] Such a comparison asks how strongly the observations constrain a feature independently of assumptions imposed during fitting.
An original one-parameter calculation illustrates this issue without recreating the historical refinement. Suppose a scalar coordinate is constrained by an observed value with uncertainty , and by a preferred value with strength . Minimize
For , , and in compatible units, . Removing the restraint gives . A fitted value can therefore depend on both observations and a modeling choice. This toy weighted least squares problem is not the algorithm used for 4HHB.
Agreement with data should consequently be examined together with restraint sensitivity and independent physical evidence. A precise-looking coordinate is not established merely by printing many decimal places. Fermi's documented work exemplifies using X-ray crystallography to make a molecular arrangement quantitatively testable.
Inverse-variance weighting combines estimates of the same quantity by giving a more precise estimate a larger weight. Suppose are independent, unbiased estimators of , with known variances . For normalized weights,
The inverse is taken after squaring the standard uncertainty: twice the standard uncertainty gives one quarter as much weight. The corresponding weighted least squares criterion is ; differentiating it with respect to gives the same weighted arithmetic mean.[ref]
The optimality statement is precise. Any linear unbiased estimator requires . By the Cauchy-Schwarz inequality,
Equality holds for the displayed weights. Thus they minimize variance among linear unbiased combinations under these assumptions; a normal distribution is not needed for that proof.
For an original example, combine observed values 10 and 14 with standard uncertainties 1 and 2 in the same units. The normalized weights are and , giving estimate , variance and standard uncertainty approximately . Equal weights give 12 and standard uncertainty .
Correlated errors change the problem. A contribution common to every reading survives every weighting whose coefficients sum to one; precision limits of weighting correlated repeats derives the resulting floor. More generally, use the full covariance matrix and generalized least squares rather than treating correlated readings as independent.[ref] Uncertainty in estimated inverse-variance weights is another limitation: variances estimated from the same data are not fixed known quantities, and an accidentally tiny estimated variance does not justify arbitrary confidence.
Kernel for nonparametric regression Created 2026-10-05 Updated 2026-11-05
A nonnegative regression kernel supplies weights in a local weighted least squares fit. Unlike a probability density function, its integral need not equal one, because rescaling every weight does not change the fitted coefficients.
Local linear regression Created 2026-10-06 Updated 2026-11-05
local linear regression fits to responses near each target by weighted least squares, then returns the fitted intercept. With weights , put and . If , then . Its polynomial reproduction property of local polynomial regression reproduces affine functions, reducing boundary bias relative to a local constant fit.
Local polynomial regression Created 2026-09-24 Updated 2026-11-05
Nuclear level reconstruction from energy differences Created 2026-10-20 Updated 2026-11-05
To leading order neglecting nuclear recoil, a measured gamma ray has energy equal to the difference between the upper and lower nuclear energy levels. For a freely recoiling atomic nucleus initially at rest, four-momentum conservation instead gives
where and are the rest masses of the upper and lower nuclear states. Some transition energy becomes daughter kinetic energy, so high-precision level reconstruction must account for nuclear recoil.
Reconstructing a nuclear gamma ray radiation spectrum amounts to assigning measured photon energies to differences between ordered nuclear energy levels with such corrections where needed. Cascade energy sums constrain the graph, while population-flow balances constrain its orientation and feeding. Experimental central energies need not close exactly; a weighted least squares energy-network fit uses their uncertainties consistently. Multiple graphs can satisfy the same radiation spectrum when coincidence or lifetime information is absent.
Choose the diagonal matrix . Multiplying a vector by scales its th component by . Therefore
The Euclidean norm objective for ordinary linear least squares on is exactly the weighted least squares objective, so the sets of minimizing vectors coincide. Zero weights simply discard the corresponding residuals.
Using the paper's named weights , take
The least-squares normal equations for weighted least squares become
Multiplying by four gives and . Their determinant is , giving
Thus, to two significant digits, , with the slope measured in expression units per hour.
There is a wording convention to make explicit. Because the stated objective squares , the choice above gives the final two squared residuals one-quarter of the influence of the others. If “half as much” instead means half the coefficient of each squared residual, use . Then the least-squares normal equations are
giving and . Stating which quantity is halved resolves the ambiguity.
The missing output is needed to identify the actual regression formula, response column, transformations and coefficient estimates. A useful possible analysis regresses the maintained-school percentage among acceptances on the corresponding percentage among applications, allowing for year; it must be described as a model of college-level composition, not a direct estimate of individual acceptance probabilities.
For instance, if and are these two percentages, a normal linear model might specify , with a slope-by-year interaction term if needed. The intercept is the expected response at zero predictor, often an extrapolation; the slope is a percentage-point change in acceptance composition per percentage-point change in application composition. A slope of one is not by itself evidence of equal acceptance chances: equal chances within every college would give the entire relation , including zero intercept and no year shift. A fit with positive slope mainly confirms that applicant composition helps predict acceptance composition.
Regression diagnostics should include residuals versus fitted values and predictors, a residual normal Q-Q plot, checks of heteroscedasticity, regression leverage, studentized residuals and Cook's distance. The aggregate mature-college category merits an influence check and a sensitivity analysis: combining several colleges can conceal different within-college patterns. Repeated observations from the same college across years may have correlated errors, so treating them as independent needs justification. Check nonlinearity, the year interaction and the effects of excluding no observation except with a substantive reason.
Percentages are bounded, and their precision depends on their denominators. A simple weighted least squares or binomial analysis requires the underlying numbers accepted, not merely percentages. For a proportion based on people, the working variance is under independent sampling, rather than constant across colleges. If the predictor percentage is itself estimated, an ordinary regression also ignores its uncertainty. The supplied summaries cannot recover all four cells of each college contingency table or the actual fitted numerical output.
The published Cambridge percentage table nevertheless allows a descriptive comparison of acceptance rates. Write for the maintained-school proportion among applicants and for its proportion among acceptances. The relative risk from group compositions gives
The overall acceptance probability cancels by Bayes' theorem. This risk ratio compares acceptance probabilities; it is not the acceptance odds ratio, which also needs the overall acceptance probability.
For Christ's, the two published application/acceptance pairs give risk ratios and . If the displayed percentages were rounded to the nearest whole percentage point, the relative risk bounds from rounded group compositions give ranges approximately and . Thus both comparisons remain below one under that rounding assumption. These are descriptive comparisons of observed groups, without adjustment for applicant qualifications or a causal effect interpretation. The published compositions determine approximate relative acceptance rates, but not their standard errors or the original regression results.
The response errors have standard deviations proportional to . Thus, when fitted means approximate the true means, increasing absolute fitted values tends to accompany increasing residual spread. A residual-versus-fitted plot shows a fan for positive means, or spread increasing away from zero for means of either sign. Standardization by one common residual standard deviation, even with the usual regression leverage correction, does not remove this systematic heteroscedasticity.
Under independent errors the exact covariance matrix after fitting is
where is the common proportionality constant. Hence the simple fan description is a diagnostic tendency, not an assertion that each fitted residual has exactly variance .
For positive responses, the intended variance-stabilizing transformation is the natural logarithm. The delta method gives
Use for positive data with modest relative error, then refit and check the transformed mean and residual structure. In particular, if the original mean was , the transformed mean is approximately , not automatically a linear function of the original covariates.
There is a domain qualification in the literal printed model: a nondegenerate normal distribution has negative support. For positive , the probability of is , so a real is not defined throughout the stated sampling model. When is small this is a negligible-tail approximation, not an exact transformation to normal errors. If signed values must be retained and all , write with standard normal . Then
The second term has a common distribution with finite variance, so log absolute value stabilization of a normal scale family is exact, but loses the sign of the mean and does not produce normal errors. Zeros and a zero mean are excluded from this transform. Alternatively retain the response scale and fit the stated variance model with weighted least squares weights proportional to where the means stay away from zero. No arbitrary addition of a constant before taking logarithms exactly solves the given variance relation.
At a target , local polynomial regression fits a polynomial of degree at most by weighted least squares:
Define the response vector , the design matrix by for , and the diagonal matrix by . The normal equations for linear least squares give
This formula assumes the local polynomial Gram matrix is invertible. It is invertible whenever at least distinct sites have positive weights: a nonzero degree- polynomial cannot vanish at all of them. With fixed , a nonzero continuous regression kernel, and , this condition holds eventually at the target points considered here. Using weights instead gives exactly the same fit.
Use the usual unit-integral normalization . A regression kernel need not intrinsically have unit integral, but multiplying all weights by a positive constant leaves local polynomial regression unchanged. With unnormalized raw moments the interior coefficient below is ; the version written using alone requires this normalization. A zero regression kernel cannot define an estimator.
Write the local moment sums as
The local constant estimator is . Since the errors have zero mean, its bias of an estimator is the corresponding weighted average of . By the second-order Taylor expansion, uniformly for contributing sites with ,
where . This uniform remainder follows from uniform continuity of on . Nonnegativity of the regression kernel therefore gives
whenever tends to a positive limit and .
For fixed interior , the moment approximations give , , and . The hypothesis makes , so the odd-moment contribution is negligible at second order. Thus
The symmetry of the regression kernel is what removes the first-order interior bias of an estimator.
At , the rescaled observation window is truncated to . Put and . Symmetry, nonnegativity, and unit mass give . The supplied moment approximations already imply the leading boundary bias of local constant regression:
To retain the second-order term rigorously, we need slightly more precise control of the first moment. The rescaled grid has spacing . Each truncated function has bounded variation, because is continuously differentiable on its support. The Riemann sum error is bounded by this variation times the grid spacing: summing the oscillation bounds on the grid cells proves
Endpoint cells, including the omitted site at zero, contribute the same error order. Consequently
Both discretization errors are under . The preceding Taylor expansion therefore proves the second-order boundary expansion for local constant regression:
Equivalently, expanding the derivatives about zero gives
In general the truncated first moment is nonzero, so the boundary bias of an estimator has order , although it can be smaller when or the omitted part of the regression kernel carries no mass.
For local linear regression, the polynomial reproduction property of local polynomial regression removes the constant and linear terms exactly. Let . Its effective kernel weights are
so , , and
At either an interior or the specified boundary target, is asymptotic to a positive constant times . At the boundary that constant is , the mass squared times the variance of the continuous normalized truncated regression kernel. Moreover , as follows by bounding in the window. Thus the Taylor remainder still contributes , even though some effective kernel weights may be negative. In particular,
At an interior point this coefficient reduces to by symmetry and normalization. Hence local linear regression has bias at both interior and boundary points; its linear reproduction removes the generic order- boundary term.
For ordinary homoscedastic fixed-design nonparametric regression, take observations at known distinct sites, with independent mean-zero errors of common variance. A degree- local polynomial estimator at a target solves the finite-dimensional weighted least squares problem
Here is a nonnegative regression kernel, is the smoothing bandwidth, and sufficient local design rank ensures uniqueness. The estimate of the regression function is . Each target has its own locally fitted polynomial; this is not one global polynomial fit. The bandwidth controls the localization and the bias-variance tradeoff.
A natural cubic smoothing spline instead solves the global penalized problem
The Sobolev space here consists of functions with square-integrable weak derivatives through order two. The minimizer is a natural cubic spline with knots at the observation sites and linear tails outside the extreme sites. Equivalently one may perform the minimization on that finite-dimensional spline space: the minimum roughness property of the natural cubic spline interpolant shows that replacing a candidate by the natural spline with the same site values preserves the data loss and decreases the second-derivative penalty. The smoothing parameter controls curvature; it differs from the local bandwidth .
For the interval-average observation problem, the printed upper limit in the loss is an indexing error: there are only responses and intervals, so a term with would involve the undefined and . The coherent loss uses the upper limit . Put and define the linear observation functionals
The intended criterion on is therefore
Unlike the ordinary cubic smoothing spline, this uses interval averages as data and a first-derivative penalty; the resulting spline degree will be two.
Fix any candidate . By the supplied interpolation property, there is a unique quadratic spline with knots , constant on both exterior intervals, such that for every . Put . Then
On each interior knot interval is a constant , and on the two exterior intervals . Integrate by parts separately on the intervals. At every knot the boundary contributions cancel because both and are continuous; the contributions at and vanish because is zero there. Thus
This orthogonality is the key to the quadratic smoothing spline for interval averages. Expanding the roughness penalty gives
The data-fit terms are equal because the interval averages are equal. Consequently,
Equality can occur only if everywhere: the derivative is continuous and its squared integral is zero. Then is constant, and any of its zero interval averages forces that constant to be zero. Therefore every minimizer must equal its matching quadratic spline .
We conclude that the minimizer is a continuously differentiable quadratic spline with knots at the specified sites and constant tails on both exterior intervals. Existence and uniqueness also follow from the supplied interpolation property. The interval-average map identifies the finite-dimensional space of these quadratic splines with , and the criterion becomes a strictly convex quadratic function of that average vector: its data-fit term is , while its penalty is nonnegative and quadratic. It has a unique minimizer, and the strict equality argument above proves uniqueness in the full function class.
In fixed-design nonparametric regression, the covariates are deterministic and
For the usual finite-variance model assume independent errors with . In random-design nonparametric regression, the pairs are independent observations and ; equivalently with conditional mean-zero errors. Conditional variances are generally assumed bounded for risk bounds.
For a nonnegative regression kernel and a smoothing bandwidth , the Nadaraya–Watson estimator is
when the denominator is positive; replace by for random-design nonparametric regression. The local polynomial regression estimator of degree minimizes the weighted least squares criterion
and reports . If the weighted Gram matrix is invertible, this minimizer is unique. The Nadaraya-Watson estimator is exactly local polynomial regression of degree zero. Higher degrees reproduce nonconstant polynomials and reduce boundary bias of an estimator.
The requested error bound needs and mean-zero independent errors of uniformly bounded variance, along with the matrix condition in the hint. The degree requirement is missing from the printed question. To see the problem, take , , , zero errors, , and . This regression function has two bounded derivatives, but the local constant estimate is the average of for :
The scalar matrix is bounded away from zero for every . Nevertheless , so the printed conclusion with fails. A bounded-variance error hypothesis is also essential; otherwise an expectation bound need not even be finite.
Under the qualified hypotheses, use and
This symmetric Gram matrix has inverse operator norm bounded by the reciprocal of its uniform positive eigenvalue lower bound. If is supported in , only contribute. The equally spaced design and imply that there are at most such points. The bounded kernel and bounded on this interval therefore yield
For the stochastic part, variance additivity for independent random variables and Cauchy-Schwarz inequality give
For the deterministic part, let be the Taylor polynomial of at of degree . The polynomial reproduction property of local polynomial regression gives . By Taylor's theorem, on the kernel window. Thus
Combining the two parts proves
The argument also applies at the endpoints when the assumed matrix bound holds. With arbitrary degree, the generally valid smoothness term is ; extra symmetry can improve some interior biases but does not fix the general boundary counterexample.
For fixed-design nonparametric regression, let and . The degree- local polynomial estimator takes the fitted intercept , where
Use a nonnegative regression kernel and an invertible local polynomial Gram matrix to obtain a unique fit. For , solving the single weighted least squares normal equation gives the local constant estimator:
At the interior point , use the standard unit-integral regression kernel convention . Otherwise the leading variance below is ; unlike a kernel density estimator, the local constant estimator itself is unchanged by rescaling .
The local constant estimator is a linear estimator in nonparametric regression with weights . The independent unit-variance errors imply
To control the bias of an estimator, differentiability at this one point is enough. Write
Symmetry makes , and the supplied moment approximation gives . Therefore the linear contribution to the bias of an estimator is . The remaining contribution has absolute value at most , because only receive positive weight. This is the interior first-order bias cancellation for local constant regression, and gives squared bias of an estimator . Combining with the variance proves
For fixed , take . It satisfies both asymptotic smoothing bandwidth conditions, and the displayed remainder, multiplied by , tends to zero for this fixed . Thus
Letting proves the optimized scaled pointwise error tends to zero. The order of these limits matters: no remainder uniform in has been assumed.
For the lower bound over the whole function class, the absence of a uniform bound on derivatives permits a particularly direct argument. It also avoids assuming a normal distribution for errors when the model only supplies their first two moments. Fix and . Choose a smooth bump function on the interval, with , whose support contains no design point except possibly . Such a bump exists because the design is finite; at an endpoint use the restriction of a smooth bump on the real line. Every , , belongs to the differentiability class.
If is not a design point, all observation means are zero for every . The Le Cam two-point lemma applied to has total variation distance zero and gives a lower bound tending to infinity with . In fact, the supremum of the mean squared error is infinite at every such .
If is a design point, only its response depends on . The resulting location family is the single observation ; all remaining observations are independent of and give no extra information. The single-observation location minimax bound is
Here is a proof covering discrete as well as continuous errors. Give the prior distribution , independently of the noise. Put and . Consider an easier experiment in which an oracle also reveals whether , and reveals itself on the complementary event. On the event and , every possible truncated noise value places within . The flat prior distribution then makes the Bayesian posterior noise law exactly its truncated original law. The smallest conditional squared-error loss is its conditional variance , attained by the posterior mean.
The probability of that event is at least for , by restricting further to . Thus every rule's supremum risk function is at least the original Bayes risk, which is at least the easier experiment's Bayes risk, and therefore at least . Let and then . Finite second moment and zero mean give . The rule has constant mean squared error one, proving the equality.
Consequently, at every , the unrestricted differentiability class satisfies the stronger conclusion
Thus works for every . There is no conflict with the preceding fixed-function limit: the worst functions, including arbitrarily narrow bumps, can vary with . This is the pointwise versus uniform risk distinction. A smoothness ball with a common derivative bound would pose a different minimax risk problem.
Put and , where is a nonnegative integer and is the smoothing bandwidth. The local polynomial estimator is the intercept of the weighted least squares fit
When the local polynomial Gram matrix is invertible, the normal equations give
A singular local polynomial Gram matrix requires a selection convention and may leave the intercept unidentified; an arbitrary nonnegative regression kernel alone does not guarantee uniqueness. At , provided , the Nadaraya–Watson estimator is
Multiplying every weight by changes neither fit.
For the uniform kernel, let and . Then . For and , the interval is inside , and the regular design has
The same lower bound holds when the interval starts at zero, despite the absence of the design point zero. In particular the Nadaraya–Watson estimator is defined throughout the integration interval.
For every Lipschitz continuous mean function in , the bias of an estimator and variance satisfy
Here the variance uses the independence of the errors in the fixed-design nonparametric regression model. The bias-variance decomposition of mean squared error and Tonelli theorem give a slightly stronger integrated mean squared error bound than required:
Thus
The printed integration interval represents a nonnegative risk function for . For , the written integral has reversed limits and is nonpositive, making the displayed upper bound trivial; it should not be interpreted as an integrated mean squared error in that range. For , evaluation outside would additionally require a specified extension of the mean function. The subsequent optimization restricts to .
Minimizing the stated right side gives
For sufficiently large , and . For example take . Therefore the upper bound holds with
For the lower bound, the PDF has the mean squared error ; the TeX erroneously moves the square outside the expected value. The PDF's suggested smooth bump function is , with value zero at the endpoints, rather than the corrupted TeX formula. A triangular alternative suffices because only Lipschitz continuity is needed.
The Le Cam two-point lemma says that for two data laws with parameter separation , every estimator satisfies
Indeed, classify according to the nearer parameter value. Misclassification incurs squared-error loss at least , and the sum of the two error probabilities is at least one minus the total variation distance. Averaging the two risk functions proves the stated form.
Let , , and use the two-point lower bound with a triangular bump
Both functions are in , and . Assume first that and . At most design points meet the bump's support. The Kullback-Leibler divergence between normal distributions and chain rule for relative entropy give
The Pinsker's inequality therefore gives . The Le Cam two-point lemma yields the pointwise minimax rate for Lipschitz regression
This universal constant is valid for .
If the printed lower bound is read as applying to all positive integers , it also holds with a smaller constant depending on the fixed . For , use the constant mean functions and . They have Kullback-Leibler divergence , so Pinsker's inequality and the Le Cam two-point lemma give minimax risk at least . Combining the two ranges gives the explicit all- choice
The universal asymptotic constant and this parameter-dependent all- constant are distinct conventions; the latter supplies the assertion without inserting an unstated large- restriction.
The weighted least squares objective is
Its gradient is
Since is positive diagonal and has full column rank, is positive definite. The unique stationary point is therefore the unique minimizer:
At a current value , put
and define the working response
Then the displayed formulas in the question can be written compactly as
and
Indeed, the th component of the latter is
which is precisely the stated score.
Substitution into the Fisher-scoring update gives
By part (b), this is the weighted least squares fit of the working response on . Both and are recomputed from the current fitted means after every update. Fisher scoring for a generalized linear model is therefore the iteratively reweighted least squares algorithm.