Let be the absorber mass path along a ray. If the spectral line profile is independent of position, Beer-Lambert law gives . A finite-band mean and the line's equivalent width are respectivelyWhen the band contains all significant absorption, . An unqualified integral of transmission over the entire infinite line domain diverges because transmission tends to one far from line centre; is the finite and physically useful line deficit.
This is nonexponential band transmission: each monochromatic transmission is exponential, but their average generally is not. Differentiating with respect to mass path givesMore decisively, define the absorption coefficient averaged over the transmitted spectrum byThen . The core saturates first and the surviving radiation increasingly samples weakly absorbing wings. The effective attenuation per additional mass therefore decreases. Jensen's inequality also gives , with equality for a spectrally constant coefficient. Thus the band transmission decays more slowly than the exponential with its initial mean attenuation.
The corresponding curve of growth has for an optically weak line, and sublinear growth after saturation. For the Gaussian profile, with , absorption extends to , so . For a strongly saturated pure Lorentz profile the wings instead give . These asymptotes apply while the line remains isolated and its important wings remain within the range where the assumed profile and nearly constant Planck function are appropriate. The integrated absorption is not an exponential either: and .
Line saturation produces nonexponential band transmission
. An original fixed Gaussian line example shows its opaque core expanding with absorber path. The mean transmission across a fixed band remains above the single exponential determined by its initial mean attenuation. Details.For , the bound ensures the finite moment . By Jensen's inequality, . The process is an adapted, nonnegative integrable random variable at each time, and independence givesThus is a submartingale, and . The Doob maximal inequality for a nonnegative submartingale states that for . Apply it with . Since the power function is increasing on ,Combining this with the preceding bound givesThe result is a bound on the running maximum, which is why a maximal inequality for a submartingale, rather than an estimate of the single terminal random variable, is needed.[ref]
Probability: Theory and Examples (fifth edition) by Rick Durrett. §4.4, Theorem 4.4.2, printed p. 235: the submartingale maximal inequality
If future Bernoulli random variables are conditionally independent given a shared parameter , with success probability , their all-success conditional probability is . Taking a posterior expectation proves the identity. For , Jensen's inequality and give . For two distinct future trials the posterior covariance is . Thus shared parameter uncertainty can induce positive dependence even when trials are conditionally independent.
Pulse-contrast denominator acceptance rule 2026-11-05
A pulse-contrast denominator acceptance rule decides which shots may be normalized by their measured central signal. For background-corrected side signal and central signal , considerThe rule prevents normalization by an arbitrarily small accepted denominator. It also changes the estimand: averaging accepted ratios estimates , a conditional expectation. An unconditional expected value is a different target and need not exist. The pulse-contrast ratio denominator noise explains why even the unfiltered ratio mean can differ from the ratio of mean signals.
The rule provides a precise stability guarantee when its moment assumptions hold. Interpret the indicator-weighted integrand below as zero outside . If and , thenThis proves finite ratio moments under a positive denominator cutoff. The accepted ratio has finite variance and finite first absolute moment, the latter by the Cauchy-Schwarz inequality. No independence of and was used. The bound concerns mathematical stability, not whether accepted shots represent the full pulse population.
For an original brightness-selection example, suppose half the shots have and half have . The two ratios are and , so their expected value over all shots is . A cutoff retains only the bright group, giving accepted conditional expectation . Meanwhile the ratio of overall mean signals is . These three quantities answer different questions. Physical dependence between brightness and temporal shape can make the choice consequential even without readout noise.
Selection bias from a noisy denominator threshold arises even at fixed true brightness. Consider another original model with constant and measured , where . Write , with the standard normal density and the standard normal distribution function. The accepted central measurements have a truncated normal distribution, withTo derive the second identity, standardize and use , so . The density-to-tail quotient is an inverse Mills ratio.
Choose , , and in arbitrary signal units. The true-signal ratio is . Only of measured shots pass, and the conditional expectation of their measured denominator is . Since is a convex function for , Jensen's inequality and the cutoff implyThe accepted ratios are necessarily biased downward relative to this model's true-signal ratio. The cutoff selects upward central-readout errors; increasing the accepted-shot count does not remove that conditioning.
A central-signal cutoff selects upward readout fluctuations and lowers the accepted normalized ratio
. Original Gaussian readout model with true central signal 5, standard deviation 1, cutoff 6 and constant side signal 0.05. Shading marks the accepted tail. The ratio panel shows pointwise accepted ratios and bounds on their conditional mean; it is not measured correlator data. Details.In the historical single-shot correlator, nonlinear mixing encoded delays spatially and an imaging lens recorded the third-harmonic signal on a filtered CCD detector.[ref]
A Single Shot 3rd Order Cross-Correlator for Pulse Contrast and Pulse Shape Measurements by J. Collier and others. Printed pp. 194–195, Description and Figure 2
Selecting a trace by its central pixels conditions subsequent normalization on the same recorded data. Preserve the threshold, background treatment, number attempted and number accepted alongside the shot-resolved pulse-contrast estimator. For independent shots sharing the same acceptance probability, the accepted count has a binomial distribution, carrying information that disappears if only normalized accepted traces are retained.Reciprocal-estimation bias in randomized atom inclusion quantifies the nonlinear plug-in loss beyond the sign supplied by Jensen's inequality. Keep the independent-calibration model of atom-probability error in a randomized endpoint rule: true atom probability , unbiased estimate , and included probability , with the upper endpoint calibrated to actual probability . Assume , , and .
The exact algebraic identitygivesThus the loss depends on squared errors weighted by the reciprocal estimated probability. The second-order delta method approximation is , but the exact remainder shows why small estimated probabilities deserve particular attention.
If almost surely for , a nonasymptotic bound isFor an original small-error example, let , let measurement calibration return or equally often, and take , . Here and the variance is . The exact included probability is about ; the second-order approximation is , while the bounded-error calculation puts the coverage loss between about and .
The bounded-relative-error assumption can fail for a finite binomial measurement calibration: an empirical atom fraction can be zero with positive probability. Such near-zero probability estimates in reciprocal plug-in rules require an explicitly defined guard or different measurement calibration model, not division by zero. The auxiliary endpoint decision must remain independent of the future duration and measurement calibration as stipulated. The documented uniform random-variates construction supplies that auxiliary randomness; it does not validate the atom estimate.[ref]
SciPy 1.16.2 documentation. scipy.stats.uniform, support and rvs
A dye-loss interaction is a departure from additivity on a specified response scale. Changing from fractional loss to survival, logarithmic survival or a fitted rate can change that departure. The contrast must therefore be defined before it is interpreted as evidence about coupled chemical mechanisms.
For a crossed initiation/illumination experiment, write fractional losses at a common observation time and defineAn original example takes losses . The contrast is . Yet the surviving fractions obey , with . They are exactly multiplicative on the survival scale.
For first-order survival , transform to . The rate-scale interaction isThe example gives . More generally, additive rates yieldwhich is negative for positive . The fractional-loss ceiling creates the contrast without an additional rate interaction.
The transformation also changes noise. First-order uncertainty propagation gives , so near-complete loss can strongly amplify uncertainty. In addition, the mean of a nonlinear transformation need not equal the transformation of the mean; Jensen's inequality makes that distinction explicit. Averaging transformed replicates and transforming an average need not produce the same result.
The STY-BODIPY readout is a coupled oxidation signal. Its primary method distinguishes radical-trapping kinetics from stoichiometry.[ref]
A Continuous Visible Light Spectrophotometric Approach To Accurately Determine the Reactivity of Radical-Trapping Antioxidants by Evan A. Haidasz and others. Abstract, co-autoxidation method and kinetic/stoichiometric interpretation
A rate transformation requires a justified survival model after product absorption and background have been addressed. Report both the response definition and the uncertainty model; a transformed zero is conditional evidence about that model, rather than a unique identification of the chemistry. Sharp two-term convex power bound 2026-10-14
Apply Jensen's inequality to the two equally weighted values and the convex function . Equality for proves that is the smallest possible constant. For , strict convexity makes equality equivalent to ; for , equality holds for every pair.
Stochastic thermodynamics Created 2026-09-28 Updated 2026-11-05
Stochastic thermodynamics assigns work, heat and entropy production to individual fluctuating sample trajectories. It is useful for a small system, such as a trapped colloidal particle, whose exchanged energy is comparable to the thermal scale . The surroundings remain a thermal reservoir at temperature , while the observed coordinates follow a stochastic process.[ref]
For a one-dimensional optical trap with potential energy , moving its centre does incremental work on the particle. The coordinate change contributes , where the circle denotes the Stratonovich integral convention. The ordinary chain rule then gives the trajectory first law of thermodynamics, . A particle ahead of the moving trap centre can give a negative instantaneous work increment, even during a protocol with positive mean dissipated work.
In the optical-trap experiment of G. M. Wang and collaborators, a recorded colloidal particle trajectory allowed short-time entropy-consuming realizations to be compared with entropy-producing ones. Repeating the same driving protocol measures a probability distribution rather than only a mean response.[ref]
Experimental Demonstration of Violations of the Second Law of Thermodynamics for Small Systems and Short Time Scales by G. M. Wang and others. Abstract: translated optical trap, recorded particle trajectories and finite-time fluctuation test
The integral fluctuation theorem recovers nonnegative mean entropy production through Jensen's inequality, while allowing individual negative values. Strict Jensen equality criterion 2026-10-25
Let be an integrable random variable taking values in an open interval . Suppose is a differentiable strictly convex function. Then its expected value has equality in Jensen's inequality exactly when is constant almost surely.
First lies in : at each finite endpoint, taking the expected value of the strictly positive distance to that endpoint gives a strictly positive distance for . A tangent affine function supports at , soStrict convexity forces for . Indeed, equality at another point would force the graph to coincide with its supporting line on the connecting line segment, contradicting strict convexity. Thereforeand the right side is zero exactly when almost surely.
The tangent affine function of is an integrable random variable, so has an integrable negative part. Thus is well defined, possibly , even if absolute integrability was not assumed. Equality to automatically excludes that infinite case.
Uncertain variance in heterodyne lognormal bias correction matters when a measured overlap is corrected using an estimated, rather than known, logarithmic variance. The known-variance correction is a nonlinear function of that estimate, so replacing the variance by an unbiased estimate does not automatically preserve unbiasedness of the corrected overlap.
Suppose , with , and use . If the variance estimate is independent of the overlap readout,Even if , Jensen's inequality gives . Consequently the mean corrected estimate is at least under these independence and unbiased-variance assumptions. Equality requires a nonrandom variance estimate in the ordinary finite-variance case.
An original exact example uses a variance estimate from an independent normal calibration sample, with following a chi-squared distribution with degrees of freedom. Its Laplace transform givesThe logarithm of this factor is , which is positive for . With stipulated and , the factor is : about a 9.91% remaining mean shift. This intentionally broad statistical example is not a typical receiver specification.
For fixed and large , the logarithm is . More independent calibration information reduces this effect. Merely taking more overlap readings while retaining the same uncertain variance estimate does not remove its influence.
Dependent variance correction in heterodyne normalization requires the joint quantity instead of the product of expectations above. A variance inferred from the same readout cannot simply be declared independent. Propagate the joint calibration and readout model, with a stated target summary. JCGM 101 treats input distributions from prior uncertainty calculations and numerical propagation explicitly.[ref]
The optical gain-matching condition for heterodyne normalization still concerns physical calibration paths. This calculation addresses the statistical correction of residual log-errors; it does not imply that an imperfect RF or DC calibration can be repaired solely by subtracting a fitted variance.
Variance tensorization Created 2026-10-03 Updated 2026-10-18
Write for all inputs except . Under the Efron–Stein inequality hypotheses, variance tensorization is the equivalent conditional variance formConditioning freezes the other inputs. Then are independent with the same conditional probability distribution, givingTaking expected values explains the factor .[ref]
Lecture 7a: Efron Stein inequality by Purnamrita Sarkar. Lecture 7a: Efron Stein inequality, slides 2 and 7, PDF pp. 3 and 8: conditional variances and the independent-copy identity
For Bernoulli input with parameter , let be the value of with coordinate set to one minus its value with that coordinate set to zero, keeping fixed. The two-point conditional variance isFor the maximum example, and . The two averaged conditional variances are and , precisely half the respective resampling terms. In the dependent example , each other coordinate already determines , so both conditional variances vanish; variance tensorization also fails without independence.
For a short induction proof, separate the last input using the law of total variance. Apply the smaller-dimensional statement to the conditional expectation; Jensen's inequality for the square bounds its coordinate contributions by the original ones. One input gives equality, starting the induction.[ref]
Lecture 7a: Efron Stein inequality by Purnamrita Sarkar. Lecture 7a: Efron Stein inequality, slides 3–6, PDF pp. 4–7: variance decomposition and conditional Jensen step; the induction outline here reorganizes those ingredients
For nonnegative real numbers and positive weights with , the weighted geometric mean does not exceed the weighted arithmetic mean:For positive , this follows from Jensen's inequality and the concavity of the natural logarithm; a limit extends it to zero inputs. For two terms it gives when .
Weighted geometric mean Created 2026-10-14 Updated 2026-11-05
For positive real numbers and nonnegative weights with , the weighted geometric mean isIt takes a weighted mean of logarithms. A zero weight excludes that input; equal weights give the ordinary geometric mean. If occurs times in a list of total length , its correct weight is . For the list ,and replacing all four entries by preserves the product .
There is also an optimization interpretation. A logarithmic transformation turns multiplicative discrepancies into additive ones. For a positive representative , the weighted mean minimizes squared deviations identity givesThus uniquely minimizes weighted quadratic loss in logarithmic coordinates. The weighted arithmetic mean minimizes squared discrepancies in the original coordinates. These two objectives explain why different choices of what constitutes an error select different summaries.
The strictly concave function and the reversed Jensen's inequality give , where is the weighted arithmetic mean. Apply the same reasoning to the multiplicative inverses to get for the weighted harmonic mean. Thus ; equality holds exactly when all positive-weight inputs agree. For this list, , , and . The weights specify which inputs are represented and must stay the same across the comparison.

