2D IR line-shape correlation describes how excitation- and detection-frequency coordinates vary together within a spectral feature in two-dimensional infrared spectroscopy. An elongated feature can retain a relation between the initially selected frequency and the frequency read out later. Its shape can change with 2D IR waiting time, but a fitted geometric parameter needs a response model before it is assigned a molecular dynamical meaning.
An original Gaussian-envelope example makes the geometry precise. Let centered frequency offsets be , with widths and correlation coefficient :
The associated covariance matrix is
At fixed , maximizing over gives the line
Thus a 2D IR center-line slope equals only for this orientation with equal widths in compatible frequency coordinates. For stipulated widths and in and , the slope is . A slope greater than one is not an impossible correlation coefficient; the two quantities have different width dependence.
For equal widths , rotated coordinates and have variances and . At , the diagonal-to-antidiagonal width ratio is . This geometric identity also helps explain why 2D IR diagonal-peak overlap depends on orientation as well as marginal width.
A 2007 primary analysis by Kyungwon Kwak, Sungnam Park, Ilya J. Finkelstein and M. D. Fayer extracted a waiting-time-dependent observable from maxima of spectral slices and tested it against calculated spectra with known frequency dynamics. Its center-line convention explicitly depends on the direction of the slices.[ref]
A vibrational frequency-frequency correlation function instead concerns frequency fluctuations at two times. Under an appropriate stationary model, its normalized form is with . Relating it to spectral geometry requires the pulse sequence, broadening model and slope convention. Multiple overlapping peaks, signed responses and finite acquisition windows must be addressed before interpreting a change of shape as a unique molecular time scale.
Acquisition-time sensitivity to tail-model parameters measures how an inferred recording extension changes when the assumed optical tail changes. In the positive-duration branch of acquisition duration from an exponential tail-energy tolerance, write
Here bounds the initial tail power, is its decay time and the tolerated missed energy. Their first-order relative effects are
These logarithmic condition numbers of a tail-acquisition requirement distinguish a fractional amplitude error from a fractional decay-time error. For a stringent tolerance with large , decay-time sensitivity stays near unity while amplitude sensitivity decreases.
For an original example, take , and . Then and . Let the independent standard uncertainties of and be and , with the tolerance fixed. The amplitude contribution to is ; the decay-time contribution is . Adding their variances gives , about .
Correlated fitted parameters require covariance terms in this propagation.[ref] A standard uncertainty is not a guaranteed recording margin: a worst-case requirement needs a justified allowed parameter set, as in the parent article. The zero-duration boundary is also outside this smooth positive-branch calculation.
Physically, these parameters must describe the unrecorded optical power, not merely an electrical detector tail. NIST separately measured the detector/cable/oscilloscope response before correcting pulse measurements.[ref] A precise fit to the wrong response component does not justify the optical envelope or its acquisition-time bound.
Adam 2026-11-05
Adam, named for adaptive moment estimation, is an adaptive gradient method that scales a running gradient estimate by a running squared-gradient estimate. Diederik P. Kingma and Jimmy Lei Ba presented it as a first-order method for stochastic optimization. Its state makes the next optimizer update depend on gradient history as well as the current parameters.[ref]
Starting from , the basic minimization rule is
Here is the supplied gradient, is the learning rate, , and . The Hadamard product, square root and division act coordinatewise. Adam bias correction supplies the two normalization factors. The accumulator concerns a raw second moment, rather than a centered variance.[ref]
For an original scalar calculation, choose , , and , and supply , . The first step has
The second has
Equal and opposite supplied gradients have not undone one another: the running histories give the second step a different magnitude. Plain gradient descent with the same constant learning rate would return to zero for this sequence.
The calculation specifies the recurrence, rather than a convergence guarantee for an unspecified objective function. An optimization checkpoint needs these histories and the step count to reproduce subsequent steps. AdamW adds decoupled weight decay; feeding a penalty gradient into Adam instead implements coupled L2 regularization and changes the histories.
AdamW 2026-11-05
AdamW combines the adaptive update of Adam with decoupled weight decay. It maintains exponentially weighted histories of the supplied data gradient and its componentwise square, while applying parameter shrinkage separately. Ilya Loshchilov and Frank Hutter presented the method in their work on decoupling decay from gradient-based regularization.[ref]
For minimization without optional algorithm variants, initialize . In the convention documented by PyTorch, step uses
The product is a Hadamard product; square roots and division are also coordinatewise. Assume , and the stated learning rate . The decay coefficient is . This explicitly fixes epsilon placement and the decay convention.[ref]
The denominators and perform Adam bias correction for zero-initialized histories. For a constant supplied scalar gradient , the corresponding geometric sums give and . Their corrected values are therefore and . The squared-gradient quantity concerns a raw second moment, rather than a centered variance: a constant gradient of 2 has square 4 even though its variance is zero.
For an original first-step example, choose , , , , , and . Then
and
The constants are chosen for transparent arithmetic. Feeding into the same adaptive formulas instead, with no separate shrinkage, would give approximately . That coupled L2 regularization also changes the stored histories. Resuming AdamW therefore requires the relevant optimizer state and step count as well as the parameter values and settings.
Independent observations with known errors add a positive-semidefinite Fisher information matrix to the existing information . Writing shows its inverse covariance cannot increase in any parameter direction. Improvement can be zero along an uninformative direction. This statement does not assume that data-dependent residual variance estimates or model systematic errors must decrease.
Adjacent products of independent Gaussian variables Created 2026-10-25 Updated 2026-11-05
For independent standard normal random variables , define the displayed errors for . Each has mean zero and variance one. Nonadjacent errors depend on disjoint sets of independent variables. Adjacent covariances vanish because their product contains an independent centered factor. Thus their covariance matrix is .
They are nevertheless dependent: the Gaussian fourth moment gives , and hence . Constant variance and zero pairwise covariance can conceal dependence in magnitudes. This is useful when interpreting a residual plot in linear regression.
In a normal linear model with an intercept, observations and independent mean coefficients, adjusted is
It compares the residual variance estimate with the unmodelled sample variance, adjusting the raw coefficient of determination for fitted dimension. It can be negative and is not guaranteed to increase when another regressor is added. In penalized regression a related convention uses the fit's effective degrees of freedom instead of ; the precise smoothing-software variance convention should be specified.
Changing units or shifting the origin preserves kurtosis. For real constants and , the mean of is , its variance is , and its fourth central moment is . The factors cancel in the ratio, even for negative . Hence excess kurtosis is also unchanged. For example, and have the same kurtosis, although their means and variances differ. The case produces zero variance and is excluded.
Allocation of a deterministic bias budget among baseline constraints assigns how much residual signal-response and background-leakage error each constraint may contribute. Let the allowed total bias be , and let coefficient bounds multiply response errors . Choose nonnegative fractions with and set
If every computed response obeys , the triangle inequality gives . The response-normalized tolerance for baseline-feasibility residuals checks those component limits. Positive tolerances require positive allocated fractions; a zero allocation demands exact cancellation in the stated model.
For an original example, take and bounds of for signal amplitude, for constant background and for linear drift. Allocate fractions , and . The respective bias shares are , and , giving tolerances
Equal bias shares would instead allow each, relaxing the signal-response tolerance to while tightening the drift tolerance to . Both allocations guarantee the same total bound when their assumptions hold; their numerical and experimental demands differ.
An allocation should reflect attainable constraint accuracy and the cost of tightening each response. Cost-weighted allocation of baseline-response tolerances can formalize that tradeoff. Equal numerical tolerances across differently dimensioned conditions do not mean equal physical bias contributions. If a bound is uncertain, use a justified upper bound or include its uncertainty explicitly before calling the total a guarantee.
NIST's pulse-energy apparatus used before/after measurements for baseline and drift correction.[ref] The bounds above are illustrative design choices. This budget controls deterministic leakage within the assumed background family. It is separate from measurement-noise variance, whose covariance contributions must still be propagated.[ref]
Amplitude-squared squeezing Created 2026-10-20 Updated 2026-11-05
A field-squared quadrature variance below the equal-variance coherent-state reference in this normalization. Its conjugate variance must remain compatible with the state-dependent commutator bound. This is not the same observable as ordinary field-amplitude squeezing or a reduction of mean intensity.
Analytical sampling uncertainty is the contribution to measurement uncertainty arising when collected material is used to infer a property of a specified sampling target. The target might be a bulk lot or a defined location and time interval. Its composition need not equal that of one convenient portion. Sampling and physical preparation therefore belong in the measurement model when the measurand concerns the larger target.[ref]
A duplicated experiment can separate two random components. Within each target, collect two independent samples by the stated protocol, prepare them separately, and analyse each twice. For a simple model,
where labels collection and labels analysis; all random terms are centered and independent except for the shared within a sample. The within-sample difference has variance . The difference between the two sample means has variance . Repeating this design across targets supports estimation of both components; different target means cancel in each within-target difference.[ref]
For original stipulated difference standard deviations2and5milligrams per kilogram, respectively,
in squared concentration units. One analysis on one sample has combined standard deviation milligrams per kilogram. Averaging two analyses of that same sample gives ; averaging one analysis from each of two independent samples gives . More laboratory replication and more independent collection address different contributions.
A collaborative wheat-and-coffee trial varied the sampler as well as the collected material. Its contrasting reproducibility and repeatability findings show why the uncertainty depends on which collection conditions are allowed to vary.[ref]
Duplicate sampling with one sampler and protocol does not establish absence of a shared sampling bias. Reference comparisons, alternative protocols or inter-organisational trials can test additional effects. Common preparation losses or measurement calibration errors also require covariance terms beyond the independent model.
Anderson–Darling test Created 2026-10-14 Updated 2026-10-18
The test rejects for large values of this tail-weighted empirical discrepancy. For a specified continuous null, its standard limiting statistic is , using a Brownian bridge. The reciprocal variance weight gives tail deviations more influence than the Cramér–von Mises criterion. If null parameters are estimated, fitting must be included when computing or simulating its reference law.
For standard Cauchy distribution probability density function and , this estimates its upper-tail probability without sampling a heavy tail. The function decreases on , while increases, so their covariance is nonpositive. The paired estimator is unbiased and has no larger variance than an estimator using the same number of independent probability density function evaluations.
Antithetic variates 2026-10-14
Antithetic variates deliberately couple unbiased estimators of the same expectation with negative covariance. Their average can have smaller variance than an average using independent draws. For example, evaluating a monotone integrand at and , where has a uniform distribution, gives opposite fluctuations while preserving both marginal distributions.
For total source electrons , aperture pixels, sky-plus-dark rate , time , reads with rms , and an independent mean-background estimate of variance per pixel, the aperture variance is before additional calibration terms. The last factor is because one uncertain background mean is subtracted from all aperture pixels. Pixel correlations require covariance terms.
Approximate experimental design Created 2026-10-14 Updated 2026-11-05
An approximate design is a probability measure on the design region, with weight specifying a proportion of observations at . For independent observations and common error variance , a linear regression with vector has normalized information matrix of an experimental design ; its least-squares estimator has covariance when the allocations realize these proportions. Exact designs restrict to integers.
Specify whether two parameters mean endpoints or location and width. With endpoints , define for a standard arcsine distribution variable . Then
This is a location-scale family with location and width . Its mean is and its variance is . A convention using parameters for location and width instead has right endpoint , so it must not be read as having endpoints .[ref]
For example, a chosen interval gives mean and variance , while its density still concentrates toward both endpoints. It is a bounded probability distribution; its endpoint singularities do not make its support infinite.
Inverting gives the quantile function
Consequently, inverse transform sampling uses with uniform on . The quartiles are , and .[ref]
The first and third quartiles are about and , so the central half of the probability spans about of the unit interval. Under a uniform distribution, its width would be .
The symmetric probability distribution gives . The angle representation gives a short calculation of the second moment:
Use to evaluate the integral. A uniform distribution on has variance ; the larger variance quantifies the greater spread toward the ends.
More generally, for a nonnegative integer , the beta function yields
For this is one; for it reproduces the values above. The central binomial coefficient thus appears in an exact moment calculation for a continuous probability distribution.
Arithmetic stock model with constant volatility Created 2026-10-06 Updated 2026-11-05
The arithmetic stock model with constant volatility has additive price dynamics , with constant and . Its absolute volatility is constant, and the model can take negative stock values. This differs from a geometric Brownian motion, whose noise is proportional to its current price.
With a bank account satisfying , assume a finite-horizon equivalent risk-neutral measure exists. The market price of risk is , and the pricing dynamics become . For , the conditional Gaussian law has
The variance is when and when . The zero-rate case is obtained by the continuous limit, not by division by zero.
Asymptotic distribution of a sample median Created 2026-09-24 Updated 2026-11-05
Let be independent and identically distributed random variables with unique median , and let their probability distribution have continuous probability density function with . For odd , the sample median satisfies
The influence function of the sample median has mean zero and variance : each of its two nonzero values occurs with probability . Its asymptotic linear representation, followed by the central limit theorem, gives the displayed limit.
For a standard normal distribution, , so the asymptotic variance of the unscaled sample median is . The sample mean instead has variance in that model. This quantifies an efficiency cost at the reference model without contradicting the sample median's bounded response to an outlier.