Properties of Strongly Harmonizable Processes...

Strongly harmonizable processes are a class of stochastic processes characterized by their unique spectral representation and a set of distinctive properties. These processes extend many results from stationary processes to a broader class, making them valuable tools for analyzing non-stationary phenomena in various fields of mathematics and applied sciences.

Spectral Representation of Processes

The spectral representation of strongly harmonizable processes provides a powerful framework for analyzing their behavior in the frequency domain. While the basic spectral representation was mentioned in a previous section, it's important to delve deeper into its implications and extensions.

For a strongly harmonizable process X(t), the spectral representation can be expressed as:

$X(t)=\int_{-\infty}^{\infty}e^{itλ}dZ(λ)$

where Z(λ) is a complex-valued random measure with orthogonal increments 1. This representation allows for a decomposition of the process into its frequency components, providing insights into its oscillatory behavior.

The spectral measure F(λ) associated with this representation has several important properties:

1. Finite Total Variation: Unlike weakly harmonizable processes, the spectral measure of strongly harmonizable processes has finite total variation 1. This property ensures better convergence properties and allows for more robust statistical analysis.
2. Symmetry: For real-valued processes, the spectral measure exhibits conjugate symmetry, meaning F(-λ) = F*(λ), where F* denotes the complex conjugate 1.
3. Spectral Distribution Function: The spectral measure can be characterized by a spectral distribution function, which is a non-decreasing, right-continuous function 2.

The spectral representation also extends to the covariance function. For a strongly harmonizable process, the covariance function R(s,t) can be expressed as:

$R(s,t)=\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}e^{i(sλ_1-tλ_2)}dF(λ_1,λ_2)$

where F(λ_1, λ_2) is a complex-valued measure of bounded variation 1. This two-dimensional spectral representation provides a more comprehensive characterization of the process's second-order properties.

An important aspect of the spectral representation is its connection to the Fourier transform. For a strongly harmonizable process, the finite-dimensional distributions can be obtained through the inverse Fourier transform of the spectral measure 2. This relationship facilitates the analysis of the process in both time and frequency domains.

The spectral representation of strongly harmonizable processes also has practical implications in signal processing and time series analysis. It allows for the development of spectral estimation techniques that can handle non-stationary signals, extending the applicability of Fourier-based methods to a broader class of processes 1.

Understanding the spectral representation of strongly harmonizable processes is crucial for their analysis and application in various fields, including communications, econometrics, and geophysics, where non-stationary phenomena are common.

---

Sources:

• (1) 10_xxx_joyeux.dvi
• (2) 1

Key Mathematical Properties

Strongly harmonizable processes possess several key mathematical properties that distinguish them from other stochastic processes and make them particularly useful in various applications. These properties extend beyond the basic spectral representation and correlation characteristics discussed in previous sections.

One fundamental property of strongly harmonizable processes is their mean square continuity. This property ensures that the process varies smoothly in the mean square sense, which is crucial for many analytical techniques. Mathematically, for a strongly harmonizable process X(t), we have:

$\lim_{h\to 0}E[|X(t+h)-X(t)|^2]=0$

This continuity property is directly related to the behavior of the spectral measure F(λ) at infinity 1.

Another important characteristic is the Karhunen representation of strongly harmonizable processes. This representation allows for the decomposition of the process into a series of uncorrelated random variables:

$X(t)=\sum_{n=1}^{\infty}\sqrt{\lambda_n}\xi_n \phi_n(t)$

where λ_n are the eigenvalues of the covariance operator, φ_n(t) are the corresponding eigenfunctions, and ξ_n are uncorrelated random variables 1.

Strongly harmonizable processes also exhibit a unique property related to their Fourier transforms. The Fourier transform of a strongly harmonizable process exists in the mean square sense and is itself a strongly harmonizable process in the frequency domain 2. This duality between time and frequency domains provides powerful tools for analysis and synthesis of these processes.

An important mathematical property of strongly harmonizable processes is their behavior under linear transformations. If X(t) is a strongly harmonizable process and L is a bounded linear operator, then L[X(t)] is also a strongly harmonizable process 2. This property ensures that many operations preserve the strongly harmonizable nature of the process.

Furthermore, strongly harmonizable processes possess a unique property called the "harmonizable dilation." This means that every strongly harmonizable process can be embedded in a larger space where it becomes a part of a stationary process 2. Mathematically, for a strongly harmonizable process X(t), there exists a stationary process Y(t) in a larger Hilbert space such that:

$X(t)=P[Y(t)]$

where P is a projection operator. This property provides a bridge between the theory of stationary and non-stationary processes, allowing for the extension of many results from stationary processes to strongly harmonizable ones.

Lastly, strongly harmonizable processes have well-defined integration properties. The stochastic integral of a strongly harmonizable process with respect to another strongly harmonizable process is itself strongly harmonizable 1. This property is crucial for developing stochastic calculus for this class of processes and has important applications in fields such as financial mathematics and signal processing.

These mathematical properties collectively make strongly harmonizable processes a powerful and flexible tool for modeling and analyzing a wide range of non-stationary phenomena, providing a rich framework that extends many classical results from stationary processes to a broader class of stochastic processes.

---

Sources:

• (1) 10_xxx_joyeux.dvi
• (2) 1

Relationships to Other Processes

Strongly harmonizable processes occupy a unique position in the broader landscape of stochastic processes, exhibiting relationships and connections to various other classes of processes. These relationships provide valuable insights into the nature of strongly harmonizable processes and their applicability in different contexts.

One of the most significant relationships is with stationary processes. Every stationary process is strongly harmonizable, but the converse is not true 1. This inclusion relationship allows for the extension of many results from stationary processes to the broader class of strongly harmonizable processes. For instance, the spectral representation of stationary processes can be seen as a special case of the strongly harmonizable spectral representation, where the spectral measure is concentrated on the main diagonal of the frequency plane.

Periodic stationary sequences form another important subclass of strongly harmonizable processes 1. These processes exhibit periodic behavior in their correlation structure and have a spectral measure concentrated on a finite number of parallel lines in the frequency plane. This relationship is particularly useful in modeling phenomena with recurring patterns, such as seasonal economic data or cyclic natural processes.

Oscillatory sequences, which are characterized by their oscillating correlation functions, are also strongly harmonizable 1. These processes are valuable in modeling systems with inherent oscillatory behavior, such as electromagnetic waves or mechanical vibrations. The strongly harmonizable framework provides a rigorous mathematical foundation for analyzing these oscillatory phenomena.

Slowly changing processes, both in continuous and discrete time, form another significant subclass of strongly harmonizable processes 1. These processes are characterized by correlation functions that change slowly over time, making them suitable for modeling systems that exhibit gradual, non-stationary behavior. The strongly harmonizable framework allows for a more precise characterization of these processes compared to traditional non-stationary models.

It's worth noting that while all strongly harmonizable processes are weakly harmonizable, the converse is not true 2. The key distinction lies in the properties of their spectral measures. Strongly harmonizable processes have spectral measures with finite total variation, whereas weakly harmonizable processes only require finite variation along horizontal and vertical lines.

The relationship between strongly harmonizable processes and Gaussian processes is also noteworthy. While not all strongly harmonizable processes are Gaussian, all Gaussian processes that are strongly harmonizable possess additional properties, such as the ability to be represented as a stochastic integral with respect to a Gaussian measure 2.

Lastly, strongly harmonizable processes have connections to more general classes of non-stationary processes. They can be viewed as a special case of harmonizable processes, which in turn are a subclass of processes with orthogonal increments 2. This hierarchical relationship places strongly harmonizable processes in a broader context of non-stationary process theory, highlighting their importance as a bridge between stationary and more general non-stationary processes.

These relationships to other processes underscore the versatility and theoretical importance of strongly harmonizable processes in the study of stochastic phenomena. By understanding these connections, researchers and practitioners can leverage the properties of strongly harmonizable processes to analyze and model a wide range of complex, non-stationary systems across various fields of science and engineering.

---

Sources:

• (1) 10_xxx_joyeux.dvi
• (2) 1

Special Cases in Correlation

Strongly harmonizable processes exhibit unique correlation properties that distinguish them from other stochastic processes. These special cases in correlation provide valuable insights into the behavior of these processes and their applications in various fields.

One notable special case is the concept of linear correlation in strongly harmonizable processes. A process X(t) is said to be linearly correlated if its correlation function R(s,t) can be expressed as:

$R(s,t)=\sum_{k=1}^na_k(s)b_k(t)$

where a_k(s) and b_k(t) are continuous functions 1. This representation allows for a simplified analysis of the process's second-order properties and has important implications for spectral analysis.

Interestingly, for strongly harmonizable processes that are linearly correlated, the support of their spectral measure is concentrated on a finite number of straight lines parallel to the main diagonal in the frequency plane 1. This property provides a geometric interpretation of the process's spectral characteristics and can be useful in identifying underlying periodicities or trends.

Another special case is convex correlation. Every strongly harmonizable process that is linearly correlated is also convexly correlated 1. This means that the correlation function can be expressed as a convex combination of simpler correlation functions, providing a more flexible framework for modeling complex correlation structures.

The concept of slowly changing correlation is particularly relevant for strongly harmonizable processes. In this case, the correlation function R(s,t) can be approximated by:

$R(s,t)\approx R(\frac{s+t}{2},s-t)$

This approximation is valid when the process changes slowly compared to the observation interval 2. The spectral measure F(du,dv) for such processes is concentrated on a band along the bisector of the frequency plane, with the width of the band determined by the rate of change of the process 2.

For periodic stationary sequences, which form a subclass of strongly harmonizable processes, the correlation function exhibits a periodic structure. This leads to a spectral measure concentrated on a finite number of parallel lines in the frequency plane, with the spacing between lines related to the period of the process 2.

Oscillatory sequences, another subclass of strongly harmonizable processes, display correlation functions with oscillatory behavior. These processes are characterized by spectral measures concentrated on curved lines in the frequency plane, reflecting the varying frequency content of the process over time 2.

Understanding these special cases in correlation is crucial for effectively modeling and analyzing strongly harmonizable processes in various applications, from signal processing to financial time series analysis. The rich correlation structure of these processes allows for the representation of a wide range of non-stationary phenomena, making them a powerful tool in the study of complex systems.

---

Sources:

• (1) 1
• (2) 10_xxx_joyeux.dvi