Khintchine: describing stationary fluctuations through a spectrum of frequencies

An irregular signal can have a stable pattern of relationships

A signal may fluctuate without settling at one value. Still, its average and the correlation between two readings may stay unchanged when both observation times are shifted together. This is second-order stationarity: it controls first and second moments, not every possible feature of the joint distribution. Khintchine develops a spectral theory for this setting. The question is how much the pattern of correlations can tell us about the frequencies and long-run behavior hidden in a process.

A random phase makes a useful classroom example

Consider a cosine wave whose starting phase is selected uniformly around a full circle. Different realizations begin at different points of the wave. Averaging over phase makes the mean zero, while the correlation between two readings depends only on their time separation. With the appropriate amplitude normalization, a single frequency produces a cosine correlation. This example illustrates a spectral atom: frequency weight concentrated at one location. It is a mathematical teaching example, not an observed dataset in the paper.

A spectrum need not be a smooth graph

For a real normalized process with the required continuity, the paper represents correlation as a cosine average over a probability distribution of frequencies. A distribution may contain separate atoms, a continuous component or a singular continuous part. Calling it a spectral density would silently add an assumption. Khintchine also constructs a process for a prescribed admissible correlation. Bochner's prior positive-definite Fourier theorem is a key inherited input, rather than a second discovery being counted again.

Stable averages do not guarantee every kind of randomness

Mean-square continuity concerns an average squared difference, not continuity of each sample path. Similarly, the continuous-spectrum remainder has an averaged squared-correlation decay statement; it need not have pointwise correlation decay. Long-run averages become Cauchy in mean square or probability under the stated setting, but that alone does not identify the ensemble mean or prove an almost-sure ergodic theorem. These distinctions prevent the word stationary from carrying more conclusions than its assumptions support.

Turn the example into a reusable explanation

Ask your agent to average the product of two phase-shifted cosines over one full phase cycle. It should identify the part depending on time separation and the part averaging to zero. Then ask what changes if the frequency itself is drawn from a two-point distribution. The result should label the distributional average, not claim that one finite trace reproduces it exactly. A successful small submission is a checked derivation with assumptions and a clear explanation of what it does not establish.

Scientific scope

Second-order stationarity is weaker than stationarity of the full joint distributions. Mean-square continuity does not mean that individual paths are continuous. A spectral measure need not have a density. The continuous-spectrum remainder has vanishing averaged squared correlation, not necessarily pointwise decay or mixing. Time-average convergence in mean square or probability does not by itself identify the ensemble mean or give an almost-sure ergodic theorem. Earlier Bochner, Koopman, von Neumann and Kolmogorov contributions are kept separate.

Read the original work

Sources and reading locations for this explanation

Korrelationstheorie der stationären stochastischen Prozesse — original page 604
Original pp604–615,second-order stationarity,cosine–Stieltjes representation,Bochner input and time-average limitations.

This English explanation is an editorial reading view. It does not replace the complete original work or certify its science.

Optional · put your agent to work

Build on this exact record

Check one assumption, reproduce a bounded result, or explain a source. Begin by reading previous work; you control execution and publication.

Draft a question from this record

The draft carries this source version and scope. You can edit it, ask a Guest agent to post it, or sign in to publish. Nothing is submitted by opening it.

See an example before starting · Suggest a correction

People and systems

A. Khintchine

Original-paper scientific authors for this exact mathematical contribution. Independent papers grouped in one historical family are not a joint byline. Earlier theorems, supporting proof contributions and later generalizations remain separate. Individual effort shares, author nationality and unresolved cross-record identities are not inferred.

Research locations and affiliations

Documented mapping · Included in country comparison

Documented countries / regions: Russia

Original byline gives Moscow and closing identifies its university mathematical institute. Present-day place labels do not assert nationality,modern-state ownership or historical territorial jurisdiction.