1 / 41

Moment Problem and Density Questions Akio Arimoto

Moment Problem and Density Questions Akio Arimoto. Mini-Workshop on Applied Analysis and Applied Probability March 24-25,2010 at National Taiwan University. March 24-25,2010 at N T U. Stationary Stochastic Process PredictionTheory Truncated Moment Problem Infinite Moment Problem

tiger
Télécharger la présentation

Moment Problem and Density Questions Akio Arimoto

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Moment Problem and Density QuestionsAkio Arimoto Mini-Workshop on Applied Analysis and Applied Probability March24-25,2010 at National Taiwan University March24-25,2010 at N T U

  2. Stationary Stochastic Process PredictionTheory Truncated Moment Problem Infinite Moment Problem Polynomial Dense N-extreme Measure Conclusion Topics ,Key words

  3. Let Stationary Stochastic Sequences Discrete Time Case(Time Series) Probability space Random variables with time variable n weakly stationary Spectral representation Positive Borel Measure March24-25,2010 at N T U

  4. Stationary stochastic process Continuous Time Case Spectral representation (Bochner’s theorem) March24-25,2010 at N T U

  5. Conditions of deterministic is deterministic is deterministic Conformal mapping from the unit circle to upper half plane March24-25,2010 at N T U

  6. Transform the probability space into the function space Discrete time case Space of random variables with finite variance Space of square summable functions March24-25,2010 at N T U

  7. isometry Discrete time case Statistical Estimation error = Approximation error March24-25,2010 at N T U

  8. Kolmogorov-Szego’s Theoremof Prediction Discrete time Szegö’s Theorem:(Kolmogorov refound) Kolmogorov’s Theorem March24-25,2010 at N T U

  9. Prediction Error indeterministic deterministic March24-25,2010 at N T U

  10. History A.N.Kolmogorov , Interpolation and Extrapolation of Stationary Sequences, Izvestiya AN SSSR (seriya matematicheskaya),5 (1941), 3-14 (Wiener also had obtained the same results independently during the World War II and published later the following ) N. Wiener, Extrapolation, Interpolation, and Smoothing of Statioanry Time Series, MIT Technology Press (1950) Kolmogorov Hilbert Space (astract Math.) Wiener Fourier Analysis (Engineering sense) March24-25,2010 at N T U

  11. Szegö’sAlternative Continuous time • Either indeterministic and where Absolute continuous part of March24-25,2010 at N T U

  12. or else Deterministic case Continuous time then We can have an exact prediction from the past March24-25,2010 at N T U

  13. This book deals with the relation between the past and future of stationary gaussian process, Kolmogorov and Wiener showed ・・・The more difficult problem, when only a finite segment of past known, was solved by Krein....spectral theory of weighted string by Krein and Hilbert space of entire function by L. de Branges… Academic Press,1976 Dover edition,2008 March24-25,2010 at N T U

  14. Problem of Krein Finite Prediction From finite segment of past Predict the future value Compute the projection of on Krein’s idea=Analyze String and spectral function Moment Problem Technique ( see Dym- Mckean book in detail) March24-25,2010 at N T U

  15. Moment Problem uniquely determined indeterminated March24-25,2010 at N T U

  16. Representing measure is called the representing measure of if a set of representation measures( convex set) We particularly have an interest to find the extreme points of March24-25,2010 at N T U

  17. Truncated Moment Problem Positive definite such taht for any Find representing measures of which moments are And characterize the totality of representation measures March24-25,2010 at N T U

  18. Properties of Extreme Points is an extreme point of conves set Polynomial dense in is the representing measure for a singular extension of March24-25,2010 at N T U

  19. Singularlypositive definitesequence Trucated Moment Problem • Arimoto,Akio; Ito, Takashi, Singularly Positive Definite Sequences andParametrization of Extreme Points. Linear Algebra Appl. 239, 127-149(1996). March24-25,2010 at N T U

  20. Singular positive definite sequence Is singular positive definite is positive definite is nonegative definite but positive definite March24-25,2010 at N T U

  21. Theorem: extreme measures is an extreme point of is singularextenstion of March24-25,2010 at N T U

  22. Extreme points of representing measures • Let Orthonormal polynomials Singularly Positive Sequence determines uniquely measure as where are zeros of a polynomial simple roots on the unit circle . March24-25,2010 at N T U

  23. Hamburger Moment Problem Infinite Moment Problem where has infinite support Find satisfying (*) is a moment sequence of March24-25,2010 at N T U

  24. Achiezer : Classical Moment Problem March24-25,2010 at N T U

  25. Riesz’s criterion (1) For some (1’) For any March24-25,2010 at N T U

  26. The Logarithmic Integral • (2) This is a common formula which appears in the moment problem and the prediction theory. March24-25,2010 at N T U

  27. (3) Is determinate (4)       is dense in (5) is densein March24-25,2010 at N T U

  28. (1) (2) (3) (4) (5) are equivalent Equivalence has been proved by Riesz, Pollard and Achiezer March24-25,2010 at N T U

  29. Important Inequality by Professor Takashi Ito polynomials March24-25,2010 at N T U

  30. Key Inequality • If we take in the above inequality we have We can easily prove the above results when we use this inequality March24-25,2010 at N T U

  31. Theorem • Let We can apply this theorem to characterize N-extreme measures. March24-25,2010 at N T U

  32. Proof of Theorem • trivial Proof of We shall prove which implies March24-25,2010 at N T U

  33. Proof of Theorem By Minkowskii’s inequality March24-25,2010 at N T U

  34. closed linear hull of Proof of Theorem In order to prove that we can only notice Hahn-Banach theorem that imply In fact, for any complex March24-25,2010 at N T U

  35. N-extremal measure • Achiezerdefined N-extreme measure Is one point set determinate contains more than two points indeterminate • Indeterminate • Polynomial dense in is N-extremal March24-25,2010 at N T U

  36. Characterization by Geometry Meaning Is N-extremal if and only if Is co-dimension one in March24-25,2010 at N T U

  37. Characterization of N-extremal measure • N-extremeness implies the measure is atomic ( due to L.de Brange ) the set of zeros of the entire function i.e. discrete or isolated point set March24-25,2010 at N T U

  38. Theorem . (Borichev,Sodin) A positive measure is N-extremal if and only if for some B(z) and its zero set , we have (1) (2) ( ) (3) ( ) Entire Function March24-25,2010 at N T U

  39. we can find an entire function of exponential type 0 such that A.Borichev, M.Sodin, The Hamburger Moment Problem and Weighted Polynomial Approximation on the Discrete Subsets of the RealLine,J.Anal.Math.76(1998),219-264 March24-25,2010 at N T U

  40. Conclusion We saw a connection between moment problem theory and prediction theory. Much remains to be done to clarify the statistical content of the whole subject. March24-25,2010 at N T U

  41. Thank you March24-25,2010 at N T U

More Related