A function that allocates a nonnegative weight to each frequency value characterizing how variance or power of a stationary signal or process is distributed across temporal frequencies, serving as the frequency-domain description of second-order structure.

Principle

The second-order statistical dependencies of a time series—its covariances at different lags—admit a representation in which contributions are apportioned by frequency; the spectral density summarizes these contributions so that filtering and resonance properties are expressed as frequency-selective weighting.

Demonstration

For an idealized uncorrelated white sequence, the spectral density is constant across all frequencies, indicating equal contribution to variance from every temporal frequency; for a narrowband oscillation, the density is concentrated near its carrier frequency.

Misapplication

Treating a single finite, nonstationary record's periodogram estimate as the true spectral density without accounting for bias, leakage, or nonstationarity and thereby inferring spurious persistent spectral peaks.

Consequence

A well-defined spectral density permits design and analysis of frequency-selective operations (e.g., amplifying or attenuating bands) and yields parsimonious characterizations of stochastic dynamics by locating dominant oscillatory components and bandwidths.

Reversal

A purely time-domain description such as the sequence of raw covariances without projecting onto frequency components; this emphasizes temporal-lag relations rather than frequency allocation of variance.

Boundary

Applies in the sense of second-order stationary processes or signals with finite variance whose covariance structure admits a frequency representation; excludes signals whose moments do not exist, strongly nonstationary records, or deterministic finite-length sequences without an appropriate limiting interpretation.

Semantic Tension

Often confused with short-time or finite-sample spectral estimates (periodograms, windowed spectra); the tension is between the idealized ensemble spectral density as a population object and concrete, noisy estimates derived from finite data.

Synthesis

The spectral density is the frequency-wise distribution of a stationary signal's second-order variability, encoding how different temporal frequencies contribute to overall variance and informing frequency-targeted analysis and design.