Some basic inequalities. Definition. Let V be a vector space over the complex numbers. An inner product is given by a function, V V C

Similar documents
APPENDIX A Some Linear Algebra

REAL ANALYSIS I HOMEWORK 1

More metrics on cartesian products

8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS

Math 217 Fall 2013 Homework 2 Solutions

Homework Notes Week 7

Section 8.3 Polar Form of Complex Numbers

Linear, affine, and convex sets and hulls In the sequel, unless otherwise specified, X will denote a real vector space.

Another converse of Jensen s inequality

Affine transformations and convexity

COMPUTING THE NORM OF A MATRIX

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017

Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011

n α j x j = 0 j=1 has a nontrivial solution. Here A is the n k matrix whose jth column is the vector for all t j=0

BOUNDEDNESS OF THE RIESZ TRANSFORM WITH MATRIX A 2 WEIGHTS

THE CHINESE REMAINDER THEOREM. We should thank the Chinese for their wonderful remainder theorem. Glenn Stevens

Exercise Solutions to Real Analysis

MAT 578 Functional Analysis

10-801: Advanced Optimization and Randomized Methods Lecture 2: Convex functions (Jan 15, 2014)

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix

Solutions HW #2. minimize. Ax = b. Give the dual problem, and make the implicit equality constraints explicit. Solution.

Appendix B. Criterion of Riemann-Stieltjes Integrability

1 Matrix representations of canonical matrices

Section 3.6 Complex Zeros

MULTIPLICATIVE FUNCTIONS: A REWRITE OF ANDREWS CHAPTER 6

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg

COMPLEX NUMBERS AND QUADRATIC EQUATIONS

Curvature and isoperimetric inequality

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X

Spectral Graph Theory and its Applications September 16, Lecture 5

Difference Equations

Determinants Containing Powers of Generalized Fibonacci Numbers

CSCE 790S Background Results

Research Article An Extension of Stolarsky Means to the Multivariable Case

Lecture 2: Gram-Schmidt Vectors and the LLL Algorithm

The Order Relation and Trace Inequalities for. Hermitian Operators

U.C. Berkeley CS294: Spectral Methods and Expanders Handout 8 Luca Trevisan February 17, 2016

2.3 Nilpotent endomorphisms

Math 261 Exercise sheet 2

COS 521: Advanced Algorithms Game Theory and Linear Programming

DIFFERENTIAL FORMS BRIAN OSSERMAN

P A = (P P + P )A = P (I P T (P P ))A = P (A P T (P P )A) Hence if we let E = P T (P P A), We have that

332600_08_1.qxp 4/17/08 11:29 AM Page 481

MA 323 Geometric Modelling Course Notes: Day 13 Bezier Curves & Bernstein Polynomials

Lecture 10 Support Vector Machines II

Lecture 20: Lift and Project, SDP Duality. Today we will study the Lift and Project method. Then we will prove the SDP duality theorem.

Lecture 12: Discrete Laplacian

Games of Threats. Elon Kohlberg Abraham Neyman. Working Paper

Linear Algebra and its Applications

Canonical transformations

Complex Numbers. x = B B 2 4AC 2A. or x = x = 2 ± 4 4 (1) (5) 2 (1)

Bezier curves. Michael S. Floater. August 25, These notes provide an introduction to Bezier curves. i=0

and problem sheet 2

Perron Vectors of an Irreducible Nonnegative Interval Matrix

Cauchy-Schwarz Inequalities Associated with Positive Semidenite. Matrices. Roger A. Horn and Roy Mathias. October 9, 2000

8.6 The Complex Number System

THE SUMMATION NOTATION Ʃ

FACTORIZATION IN KRULL MONOIDS WITH INFINITE CLASS GROUP

Linear Feature Engineering 11

Formulas for the Determinant

Math 426: Probability MWF 1pm, Gasson 310 Homework 4 Selected Solutions

Inner Product. Euclidean Space. Orthonormal Basis. Orthogonal

Lecture 5 Decoding Binary BCH Codes

A note on almost sure behavior of randomly weighted sums of φ-mixing random variables with φ-mixing weights

Bernoulli Numbers and Polynomials

Lecture 21: Numerical methods for pricing American type derivatives

Communication Complexity 16:198: February Lecture 4. x ij y ij

Randić Energy and Randić Estrada Index of a Graph

Vapnik-Chervonenkis theory

Zhi-Wei Sun (Nanjing)

Edge Isoperimetric Inequalities

Foundations of Arithmetic

Representation theory and quantum mechanics tutorial Representation theory and quantum conservation laws

Genericity of Critical Types

The internal structure of natural numbers and one method for the definition of large prime numbers

Quantum Mechanics I - Session 4

SELECTED SOLUTIONS, SECTION (Weak duality) Prove that the primal and dual values p and d defined by equations (4.3.2) and (4.3.3) satisfy p d.

REFINING CBS INEQUALITY FOR DIVISIONS OF MEASURABLE SPACE

C/CS/Phy191 Problem Set 3 Solutions Out: Oct 1, 2008., where ( 00. ), so the overall state of the system is ) ( ( ( ( 00 ± 11 ), Φ ± = 1

The Second Eigenvalue of Planar Graphs

18.781: Solution to Practice Questions for Final Exam

STEINHAUS PROPERTY IN BANACH LATTICES

Topic 5: Non-Linear Regression

princeton univ. F 13 cos 521: Advanced Algorithm Design Lecture 3: Large deviations bounds and applications Lecturer: Sanjeev Arora

arxiv: v1 [math.co] 12 Sep 2014

MTH 263 Practice Test #1 Spring 1999

Kernel Methods and SVMs Extension

Polynomials. 1 More properties of polynomials

Week 2. This week, we covered operations on sets and cardinality.

ALGEBRA HW 7 CLAY SHONKWILER

Dirichlet s Theorem In Arithmetic Progressions

Characterizing the properties of specific binomial coefficients in congruence relations

9 Characteristic classes

arxiv: v1 [quant-ph] 6 Sep 2007

Math1110 (Spring 2009) Prelim 3 - Solutions

Vector Norms. Chapter 7 Iterative Techniques in Matrix Algebra. Cauchy-Bunyakovsky-Schwarz Inequality for Sums. Distances. Convergence.

EEE 241: Linear Systems

Chapter Twelve. Integration. We now turn our attention to the idea of an integral in dimensions higher than one. Consider a real-valued function f : D

Online Appendix. t=1 (p t w)q t. Then the first order condition shows that

Transcription:

Some basc nequaltes Defnton. Let V be a vector space over the complex numbers. An nner product s gven by a functon, V V C (x, y) x, y satsfyng the followng propertes (for all x V, y V and c C) (1) x + x, y = x, y + x, y (2) cx, y = c x, y (3) y, x = x, y (4) x, x 0 and x, x = 0 f and only of x = 0. Note that f, s an nner product then for each y the functon x x, y s a lnear functon. Also we have x, cy = c x, y and x, y + ỹ = x, y + x, ỹ. Remark: We can also defne nner products for vector spaces over R, but then the thrd axom s changed to the symmetry axom y, x = x, y for all x, y V. Thus f V s a vector space over the real numbers then then for each y the functon x x, y s a lnear functon, and for each x the functon y x, y s a lnear functon. The latter statement for y x, y fals n vector spaces over C. Defnton. A sem-norm on a vector space over C (or over R) s a functon : V [0, ) satsfyng the followng propertes for all x, y V. (1) x 0 (2) For scalars c, cx = c x. (3) x + y x + y (the trangle nequalty). If n addton we also have the property that and x = 0 only f x = 0 then we call a norm. 1. The Cauchy-Schwarz nequalty Theorem. Let, be an nner product on V. Then for all x, y V x, y x, x y, y. Proof. The nequalty s mmedate f one of the two vectors s 0. We may thus assume that y 0 and therefore y, y > 0. We shall frst show the weaker nequalty (1.1) Re x, y x, x y, y Let t R. We shall use that x + ty, x + ty 0. 1

2 Then compute x + ty, x + ty = x, x + t x, y + t y, x + t 2 y, y = x, x + 2t Re x, y + t 2 y, y. Here we used that for the complex number z = x, y we have z+z = 2 Re (z). We have seen that for all t R x, x + 2t Re x, y + t 2 y, y 0. Re ( x,y ) We use ths nequalty for the specal choce t = y,y (whch happens to be the choce of t that mnmzes the quadratc polynomal). Pluggng n ths value of t yelds the nequalty whch gves x, x (Re x, y )2 y, y 0 (Re x, y ) 2 x, x y, y and (1.1) follows. Fnally let z := x, y. If z = 0 there s nothng to prove, so assume z 0. Then we can wrte z n polar form,.e. z = z (cos φ + sn φ) for some angle φ. Let c = cos φ sn φ. Then cz = z and cz s real and postve. 1 Also c = 1. Hence we get x, y = c x, y = cx, y = Re cx, y. Applyng the already proved nequalty (1.1) for the vectors cx and y we see that the last expresson s cx, cx y, y = cc x, x y, y = x, x y, y. Ths fnshes the proof. Exercse: Show that equalty n Cauchy-Schwarz, x, y = x, x y, y, only happens f x and y are lnearly dependent (.e. one of the two s a scalar multple of the other). Defnton. We set x = x, x. Theorem The map x x, x defnes a norm on V. Proof. Settng x := x, x we clearly have that x 0 and x = 0 f and only f x = 0, by property (4) for the nner product. Also cx, cx = cc x, x = c x, x. It remans to prove the trangle nequalty. 1 If you prefer not to use polar notaton, another equvalent way to defne c, gven z = a+b wth z 0 s to set c = a a b,.e. c = z/ z. Note that cz = zz/ z = 2 +b 2 z 2 / z = z.

We compute x + y 2 = x + y, x + y = x, x + x, y + y, x + y, y = x 2 + 2 Re ( x, y ) + y 2 and by the Cauchy-Schwarz nequalty the last expresson s x 2 + 2 x y + y 2 = ( x + y ) 2. So we have shown x + y 2 ( x + y ) 2 and the trangle nequalty follows. 2. Generalzed arthmetc and geometrc means Gven two nonnegatve numbers a, b we call ab the geometrc mean of a and b. The geometrc sgnfcance s that the rectangle wth sdes of length a and b has the same area as the square wth sdelength ab. The arthmetc mean s a+b 2. The arthmetc mean exceeds the geometrc mean: a + b ab. 2 Ths follows mmedately from ( a b) 2 0,.e. a + b 2 a b 0 (for nonnegatve a, b). A useful generalzaton s Theorem. Let a, b be nonnegatve numbers and let 0 < ϑ < 1. Then (2.1) a 1 ϑ b ϑ (1 ϑ)a + ϑb. 3 Proof. If one of a, b s zero then the nequalty s mmedate. Let s assume that a 0. Then settng c = b/a the asserton s equvalent wth (2.2) c ϑ (1 ϑ) + ϑc, for c 0. To prove (2.2) we set f(c) := (1 ϑ) + ϑc c ϑ and observe that f (c) = ϑ(1 c ϑ 1 ). Snce by assumpton 0 < ϑ < 1 we see that f (c) 0 for 0 c 1 and f (c) 0 for c 1. Hence f must have a mnmum at c = 1. Clearly f(1) = 0 and therefore f(c) 0 for all c 0. Thus (2.2) holds. 3. The nequaltes by Hölder and Mnkowsk For vectors x = (x 1,..., x n ) n R n (or n C n ) we defne ( x p = x p) 1/p. It s our ntenton to show that x p defnes a norm ehen p > 1. We shall use the followng result (Hölder s nequalty) to prove ths.

4 For p > 1 we defne the conjugate number p by 1 p + 1 p = 1 p p 1..e. p = Theorem: (Hölder s nequalty): Let 1 < p <, 1/p + 1/p = 1. For x, y C n, ( x y x p) 1/p( y p ) 1/p or n the above notaton x y x p y p. Remark. When p = 2, then p = 2 and Hölder s nequalty becomes the Cauchy-Schwarz nequalty for the standard scalar product x, y = n x y on R n (or the standard scalar product x, y = n x y on C n ). Proof of Hölder s nequalty. If we replace x wth x/ x p and y wth y/ y p then we see that t s enough to show that (3.1) x y 1 provded that x p = 1 and y p = 1 Also t s clearly suffcent to do ths for vectors x and y wth nonnegatve entres (smply replace x wth x etc.) Thus for the rest of the proof we assume that x, y are vectors wth nonnegatve entres satsfyng x p = 1, y p = 1. Set a = x p, b = y p. And set ϑ = 1 1/p. Snce we assume p > 1 we see that 0 < ϑ < 1. By the nequalty for the generalzed arthmetc and geometrc means we have a 1 ϑ b ϑ (1 ϑ)a + ϑb.e. Thus x y = a 1/p b 1 1/p 1 p a + (1 1 p )b = 1 p xp + (1 1 p )yp x y 1 p x p + (1 1 p ) y p = 1 p x p p + (1 1 p ) y p p = 1 p + (1 1 p ) = 1; here we have used that x p = 1, y p = 1. Remark: Hölder s nequalty has extensons to ther settngs. One s n Problem 6 on the frst homework assgnment. Here note that Remann ntegrals can be approxmated by sums, and so the Hölder nequalty wth n summands may be useful for smlar versons for ntegrals as well.

The followng result called Mnkowsk s nequalty 2 establshes the trangle nequalty for p. Theorem: For x, y C n ( x + y p) 1/p ( x p) 1/p ( + y p) 1/p or shortly, x + y p x p + y p. Proof. If x + y = 0 the nequalty s trval, thus we assume that x + y 0 and hence x + y p > 0 Wrte x + y p p = x + y p = x + y p 1 x + y x + y p 1 ( x + y ) = x x + y p 1 + By Hölder s nequalty ( x x + y p 1 x p) 1/p( x + y (p 1)p ) 1/p = x p x + y p 1 p snce (p 1)p = p. The same calculaton yelds y x + y p 1 y p x + y p p 1. We add the two nequaltes and we get x + y p p x + y p 1 p ( x p + y p ). 5 y x + y p 1 Dvde by x + y p 1 p and the asserted nequalty follows. Corollary. Let 1 p <. The expresson x p = ( n x p ) 1/p defnes a norm on C n (or R n ). 2 Mnkowsk s pronounced Mnkoffsk