A Simple Formula for the Hilbert Metric with Respect to a Sub-Gaussian Cone

Similar documents
A Simple Formula for the Hilbert Metric with Respect to a Sub-Gaussian Cone

Limiting value of higher Mahler measure

Spectral Synthesis in the Heisenberg Group

Construction of asymmetric orthogonal arrays of strength three via a replacement method

Basic Polyhedral theory

Equidistribution and Weyl s criterion

Research Article Norm and Essential Norm of an Integral-Type Operator from the Dirichlet Space to the Bloch-Type Space on the Unit Ball

The Matrix Exponential

Derangements and Applications

LINEAR DELAY DIFFERENTIAL EQUATION WITH A POSITIVE AND A NEGATIVE TERM

The Matrix Exponential

Cramér-Rao Inequality: Let f(x; θ) be a probability density function with continuous parameter

Recall that by Theorems 10.3 and 10.4 together provide us the estimate o(n2 ), S(q) q 9, q=1

ECE602 Exam 1 April 5, You must show ALL of your work for full credit.

2008 AP Calculus BC Multiple Choice Exam

Introduction to Arithmetic Geometry Fall 2013 Lecture #20 11/14/2013

10. The Discrete-Time Fourier Transform (DTFT)

Ewald s Method Revisited: Rapidly Convergent Series Representations of Certain Green s Functions. Vassilis G. Papanicolaou 1

Homotopy perturbation technique

Deift/Zhou Steepest descent, Part I

ON A SECOND ORDER RATIONAL DIFFERENCE EQUATION

Exercise 1. Sketch the graph of the following function. (x 2

NIL-BOHR SETS OF INTEGERS

A Propagating Wave Packet Group Velocity Dispersion

ABEL TYPE THEOREMS FOR THE WAVELET TRANSFORM THROUGH THE QUASIASYMPTOTIC BOUNDEDNESS

Solution: APPM 1360 Final (150 pts) Spring (60 pts total) The following parts are not related, justify your answers:

NEW APPLICATIONS OF THE ABEL-LIOUVILLE FORMULA

Supplementary Materials

cycle that does not cross any edges (including its own), then it has at least

Lie Groups HW7. Wang Shuai. November 2015

COMPUTER GENERATED HOLOGRAMS Optical Sciences 627 W.J. Dallas (Monday, April 04, 2005, 8:35 AM) PART I: CHAPTER TWO COMB MATH.

PROOF OF FIRST STANDARD FORM OF NONELEMENTARY FUNCTIONS

INCOMPLETE KLOOSTERMAN SUMS AND MULTIPLICATIVE INVERSES IN SHORT INTERVALS. xy 1 (mod p), (x, y) I (j)

VII. Quantum Entanglement

Inference Methods for Stochastic Volatility Models

Abstract Interpretation: concrete and abstract semantics

A Sub-Optimal Log-Domain Decoding Algorithm for Non-Binary LDPC Codes

CPSC 665 : An Algorithmist s Toolkit Lecture 4 : 21 Jan Linear Programming

Hardy-Littlewood Conjecture and Exceptional real Zero. JinHua Fei. ChangLing Company of Electronic Technology Baoji Shannxi P.R.

BINOMIAL COEFFICIENTS INVOLVING INFINITE POWERS OF PRIMES

EEO 401 Digital Signal Processing Prof. Mark Fowler

Einstein Equations for Tetrad Fields

EE140 Introduction to Communication Systems Lecture 2

On the irreducibility of some polynomials in two variables

REPRESENTATIONS OF LIE GROUPS AND LIE ALGEBRAS

Discrete Hilbert Transform. Numeric Algorithms

LINEAR SYSTEMS OF DIFFERENTIAL EQUATIONS DONE RIGHT KYLE ORMSBY

2 JOAQUIN BUSTO AND SERGEI K. SUSLOV l (.8) l cos nx l sin mx l dx m 6 n: In th prsnt papr w discuss a -vrsion of th Fourir sris (.) with th aid of ba

Finding low cost TSP and 2-matching solutions using certain half integer subtour vertices

On spanning trees and cycles of multicolored point sets with few intersections

DISTRIBUTION OF DIFFERENCE BETWEEN INVERSES OF CONSECUTIVE INTEGERS MODULO P

Fourier Transforms and the Wave Equation. Key Mathematics: More Fourier transform theory, especially as applied to solving the wave equation.

Legendre Wavelets for Systems of Fredholm Integral Equations of the Second Kind

The Equitable Dominating Graph

2.3 Matrix Formulation

Bifurcation Theory. , a stationary point, depends on the value of α. At certain values

Symmetric Interior Penalty Galerkin Method for Elliptic Problems

Another view for a posteriori error estimates for variational inequalities of the second kind

Mutually Independent Hamiltonian Cycles of Pancake Networks

Section 11.6: Directional Derivatives and the Gradient Vector

ON THE DISTRIBUTION OF THE ELLIPTIC SUBSET SUM GENERATOR OF PSEUDORANDOM NUMBERS

COUNTING TAMELY RAMIFIED EXTENSIONS OF LOCAL FIELDS UP TO ISOMORPHISM

The van der Waals interaction 1 D. E. Soper 2 University of Oregon 20 April 2012

Journal of Mathematical Analysis and Applications

Port Hamiltonian Formulation of Infinite Dimensional Systems I. Modeling

State-space behaviours 2 using eigenvalues

(Upside-Down o Direct Rotation) β - Numbers

Application of Vague Soft Sets in students evaluation

International Journal of Scientific & Engineering Research, Volume 6, Issue 7, July ISSN

Self-Adjointness and Its Relationship to Quantum Mechanics. Ronald I. Frank 2016

EXISTENCE OF POSITIVE ENTIRE RADIAL SOLUTIONS TO A (k 1, k 2 )-HESSIAN SYSTEMS WITH CONVECTION TERMS

Random Process Part 1

Computing and Communications -- Network Coding

Propositional Logic. Combinatorial Problem Solving (CPS) Albert Oliveras Enric Rodríguez-Carbonell. May 17, 2018

Two Products Manufacturer s Production Decisions with Carbon Constraint

1 Minimum Cut Problem

Separating principles below Ramsey s Theorem for Pairs

A LOCALLY CONSERVATIVE LDG METHOD FOR THE INCOMPRESSIBLE NAVIER-STOKES EQUATIONS

ON RIGHT(LEFT) DUO PO-SEMIGROUPS. S. K. Lee and K. Y. Park

BINOMIAL COEFFICIENTS INVOLVING INFINITE POWERS OF PRIMES. 1. Statement of results

Particular solutions of a certain class of associated Cauchy-Euler fractional partial differential equations via fractional calculus

Chapter 10. The singular integral Introducing S(n) and J(n)

Stochastic Submodular Maximization

Content Based Status Updates

BSc Engineering Sciences A. Y. 2017/18 Written exam of the course Mathematical Analysis 2 August 30, x n, ) n 2

Math 102. Rumbos Spring Solutions to Assignment #8. Solution: The matrix, A, corresponding to the system in (1) is

Chapter 6. The Discrete Fourier Transform and The Fast Fourier Transform

Content Skills Assessments Lessons. Identify, classify, and apply properties of negative and positive angles.

sets and continuity in fuzzy topological spaces

Mathematics. Complex Number rectangular form. Quadratic equation. Quadratic equation. Complex number Functions: sinusoids. Differentiation Integration

SECTION where P (cos θ, sin θ) and Q(cos θ, sin θ) are polynomials in cos θ and sin θ, provided Q is never equal to zero.

Combinatorial Networks Week 1, March 11-12

ON A CONJECTURE OF RYSElt AND MINC

y = 2xe x + x 2 e x at (0, 3). solution: Since y is implicitly related to x we have to use implicit differentiation: 3 6y = 0 y = 1 2 x ln(b) ln(b)

Complex Powers and Logs (5A) Young Won Lim 10/17/13

A Prey-Predator Model with an Alternative Food for the Predator, Harvesting of Both the Species and with A Gestation Period for Interaction

STABILITY ANALYSIS OF FUZZY CONTROLLERS USING THE MODIFIED POPOV CRITERION

Problem Set 6 Solutions

COHORT MBA. Exponential function. MATH review (part2) by Lucian Mitroiu. The LOG and EXP functions. Properties: e e. lim.

1 N N(θ;d 1...d l ;N) 1 q l = o(1)

Transcription:

mathmatics Articl A Simpl Formula for th Hilbrt Mtric with Rspct to a Sub-Gaussian Con Stéphan Chrétin 1, * and Juan-Pablo Ortga 2 1 National Physical Laboratory, Hampton Road, Tddinton TW11 0LW, UK 2 Faculty of Mathmatics and Statistics, Univrsity of St. Galln, CH-9000 St. Galln, Switzrland; juan-pablo.ortga@unisg.ch * Corrspondnc: stphan.chrtin@npl.co.uk Rcivd: 11 January 2018; Accptd: 19 Fbruary 2018; Publishd: 2 March 2018 Abstract: Th Hilbrt mtric is a widly usd tool for analysing th convrgnc of Markov procsss and th rgodic proprtis of dtrministic dynamical systms. A usful rprsntation formula for th Hilbrt mtric was givn by Livrani. Th goal of th prsnt papr is to xtnd this formula to th non-compact and multidimnsional stting with a diffrnt con, taylord for sub-gaussian tails. Kywords: dynamical systms; Hilbrt mtric; Livrani s formula 1. Introduction Lt V b a topological vctor spac and C a closd convx con insid V njoying th proprty that C C =. Thn, C dfins a partial ordring by stting Using this ordring, th Hilbrt smi-mtric Θ on C is givn by f g g f C 0}. 1) Θ f, g) = log [ ] β f,g), 2) α f,g) whr and α f, g) = supλ R + λ f g}, 3) β f, g) = µ R + g µ f } 4) and α = 0 and β = + if th corrsponding sts ar mpty. In th squl, w will focus on th cas whr V is th spac of continuous intgrabl functions on R n with Euclidan norm dnotd by. Th Hilbrt mtric, originally introducd in [1], has bn vry usful in rgodic thory for dtrministic dynamical systms [2 4], for th analysis and application of Markov Chains to control systms statistics and ormation thory [5 7]. Robust routing problms hav also bn studid using th Hilbrt mtric viw point [8]. Th first application of th Hilbrt mtric to th fild of dynamical systms is Birkhoff s approach to th Prron-Frobnius thorm [9]; s also [10]. Th ground-braking rsult of Birkhoff is th following. Thorm 1. Lt C b a con in a vctor spac V and K b a con in a vctor spac W. If L : V W is a linar mapping with LC) K and projctiv diamtr L) = sup x, y C, Lx) K Ly) dlx, Ly). Thn ) 1 dlx, Ly)/dx, y) tanh x C y C 4 L) 5) Mathmatics 2018, 6, 35; doi:10.3390/math6030035 www.mdpi.com/journal/mathmatics

Mathmatics 2018, 6, 35 2 of 5 whr tanh ) = 1. Birkhoff s thorm provids an lgant way to prov that crtain maps btwn cons ar contracts and thrfor obtain xistnc and uniqunss for crtain problms such as in Prron-Frobnius thory for positiv oprators. Th goal of this short not is to xtnd to th noncompact and multidimnsional stting a usful formula for th Hilbrt smi-mtric which was prviously givn by Livrani [2] in th cas in which V is th spac L 1 [0, 1]) of intgrabl functions of th intrval [0, 1]. Th con C in that rsult, which will b dnotd in th squl by C a, a > 0, is givn by C a [0, 1]) = g C 0 [0, 1]) L 1 [0, 1]) x, y [0, 1] g > 0 and gx) gy) a x y }. 6) In that stup, th Hilbrt smi-mtric is givn by th following xprssion: 2. Main Rsult [ Θ f, g) = ln sup x,y,u,v [0,1] a x y 2 gy) gx) ) a u v 2 f v) f u) ) a x y 2 f y) f x) ) a u v 2 gv) gu) ) ]. 7) Th con w us in th squl is diffrnt from th con chosn by Livrani in [1]. C a R n ) = g C 0 R n ) L 1 R n ) x, y R n g > 0 and gx) gy) a x y 2} 8) This con is chosn to contain dnsitis with subgaussian tails on R n. Our main rsult is th following thorm. Thorm 2. Lt Θ b th Hilbrt smi-mtric associatd to C a, a > 0; whn f, g C a, [ Θ f, g) = ln sup x,y,u,v E a x y 2 gy) gx) ) a u v 2 f v) f u) ) a x y 2 f y) f x) ) a u v 2 gv) gu) ) ]. 9) Proof. W hav to comput α f, g) and β f, g). Just as in th proof p. 247 of Lmma 2.2 of [2], w obtain gx) α f, g) = min x R n f x) ; a x y 2 gy) gx) x, y R n a x y 2 f y) f x) }. 10) W now prov that gx) x R n f x) a x y 2 gy) gx) x, y R n a x y 2 f y) f x) 11) Lt x n ) n N R n b a minimizing squnc for th lft hand sid of Equation 11). W hav to split th analysis into two cass. First cas. Assum that x n ) n N has a boundd subsqunc, dnotd by x σn) ) n N. Morovr, by th Bolzano-Wirstrass Thorm w may assum without loss of gnrality that this subsqunc convrgs to som limit point x. Fix ɛ > 0 and lt N N b such that gx σn) ) f x σn) ) x R n gx) f x) + ɛ 12)

Mathmatics 2018, 6, 35 3 of 5 for all n N. Now tak x such that Thn, w gt sup a x x σn) 2 f x σn) ) f x) < +. 13) n N a x x σn) 2 gx σn) ) gx) a x x σn) 2 f x σn) ) f x) Lt n tnd towards +, and obtain = a x x σn) 2 gx σn) ) f x σn) ) f x σn)) gx) f x) f x) a x x σn) 2 f x σn) ) f x) a x x σn) 2 gx) f x) f x σn)) + ɛ ) gx) f x) f x) a x x σn) 2 f x σn) ) f x) = gx) f x) + ɛ a x xσn) 2 f xσn) ) a x x σn) 2 f x σn) ) f x). 14) a x x 2 gx ) gx) a x x 2 f x ) f x) gx) f x) + ɛ a x x 2 f x ) a x x 2 f x ) f x) 15) and th rsult follows by taking ɛ 0 du to th assumption of Equation 13) on x. Th rsult is thn asily sn to hold for all x by continuity. Scond cas. In this cas, x n x +. Considr th unit vctor d n givn by d n = x n x x n x 16) and xtract a convrgnt subsqunc d σn) ) n N dnoting its limit by d. Now tak z δ = x δd. Lt us prov that a z δ x σn) 2 gx σn) ) gz δ ) lim n + a z δ x σn) 2 f x σn) ) f z δ ) gz δ) f z δ ) Ltting δ tnd towards zro will prov th dsird rsult using continuity. As in th first cas, on asily finds that for ach ɛ > 0 thr xists N N such that for all n N, 17) a z δ x σn) 2 gx σn) ) gz δ ) a z δ x σn) 2 f x σn) ) f z δ ) gz δ) f z δ ) + ɛ 1 1 f z δ ) a z δ x σn) 2 f x σn) ) = gz δ) f z δ ) + ɛ 1 f x) 1 f z δ ) a x x σn) 2 a x x σn) 2 f x σn) ) a z δ x σn) 2 f x σn) ). 18) Looking at this xprssion, w obviously wondr about th asymptotic bhavior of c n = f x) f z δ ) a x x σn) 2 a x x σn) 2 f x σn) ) a z δ x σn) 2 f x σn) ). 19) For this purpos, first notic that sinc f blongs to C a, w hav f x) a x x 1. 20) σn) 2 f x σn) )

Mathmatics 2018, 6, 35 4 of 5 and but f z δ ) a x x σn) 2 a z δ x σn) 2 f x σn) ) a x x σn) 2 a x z δ 2 a z 21) δ x σn) 2 a x x σn) 2 a z δ x σn) 2 +a x z δ 2 = 2a x x σn) x z δ, 22) and sinc x n x +, w obtain that c n 0 as n +, implying that th valu on is not an accumulation point. Using this, w may dduc from Equation 18) that Ltting ɛ tnd towards zro w obtain a z δ x σn) 2 gx σn) ) gz δ ) a z δ x σn) 2 f x σn) ) f z δ ) gz δ) + ɛ. 23) f z δ ) a z δ x σn) 2 gx σn) ) gz δ ) n N a z δ x σn) 2 f x σn) ) f z δ ) gz δ) f z δ ). 24) Ltting δ tnd towards zro and using continuity, w finaly conclud that a x x σn) 2 gx σn) ) gx) n N a x x σn) 2 f x σn) ) f x) gx) f x), 25) implying th dsird rsult. 3. Conclusions In this short not, w prsntd an xtnsion of a formula for th Hilbrt mtric obtaind in [2] to th multidimnsional and non-compact cas for a diffrnt rfrnc con. Using th prsntd formula and Thorm 1, applications to th study of Markov Chain Mont Carlo mthods and th analysis of th Sinkhorn algorithms for Optimal Transportation will b undrtakn in futur work. Author Contributions: Stéphan Chrétin and Juan-Pablo Ortga workd qually on th contnt and on th writing of th papr. Conflicts of Intrst: Th authors dclar no conflicts of intrst. Rfrncs 1. Hilbrt, D. Übr di grad Lini als kürzst Vrbindung zwir Punkt. Math. Ann. 1895, 46, 91 96. 2. Livrani, C. Dcay of corrlations. Ann. Math. 1995, 142, 239 301. 3. Baladi, V. Positiv Transfr Oprators and Dcay of Corrlations; Advancd Sris in Nonlinar Dynamics, 16; World Scintific Publishing Co., Inc.: Rivr Edg, NJ, USA, 2000. 4. Naud, F. Birkhoff cons, symbolic dynamics and spctrum of transfr oprators. Discrt. Contin. Dyn. Syst. 2004, 11, 581 598. 5. Bodnart, R.; Stttnr, L. Asymptotics of controlld finit mmory filtrs. Syst. Control Ltt. 2002, 47, 181 190. 6. Di Masi, G.B.; Stttnr, L. Ergodicity of hiddn Markov modls. Math. Control Signals Syst. 2005, 17, 269 296. 7. L Gland, F.; Oudjan, N. Stability and uniform approximation of nonlinar filtrs using th Hilbrt mtric and application to particl filtrs. Ann. Appl. Probab. 2004, 14, 144 187. 8. Chn, Y.; Gorgiou, T.; Pavon, M.; Tannnbaum, A. Rlaxd Schrodingr bridgs and robust ntwork routing. arxiv 2018, arxiv:1801.07852.

Mathmatics 2018, 6, 35 5 of 5 9. Birkhoff, G. Extnsions of Jntzsch s thorm. Trans. Am. Math. Soc. 1957, 85, 219 227. 10. Kohlbrg, E.; Pratt, J. Th contraction mapping approach to th Prron-Frobnius thory: Why Hilbrt s mtric? Math. Opr. Rs. 1982, 7, 198 210. c 2018 by th authors. Licns MDPI, Basl, Switzrland. This articl is an opn accss articl distributd undr th trms and conditions of th Crativ Commons Attribution CC BY) licns http://crativcommons.org/licnss/by/4.0/).