Fall 2012 Analysis of Experimental Measurements B. Eisenstein/rev. S. Errede. with respect to λ. 1. χ λ χ λ ( ) λ, and thus:

Similar documents
Least squares. Václav Hlaváč. Czech Technical University in Prague

CISE 301: Numerical Methods Lecture 5, Topic 4 Least Squares, Curve Fitting

DCDM BUSINESS SCHOOL NUMERICAL METHODS (COS 233-8) Solutions to Assignment 3. x f(x)

Quiz: Experimental Physics Lab-I

6 Roots of Equations: Open Methods

Review of linear algebra. Nuno Vasconcelos UCSD

Fall 2012 Analysis of Experimental Measurements B. Eisenstein/rev. S. Errede

Principle Component Analysis

4. Eccentric axial loading, cross-section core

CENTROID (AĞIRLIK MERKEZİ )

Chapter Newton-Raphson Method of Solving a Nonlinear Equation

Rank One Update And the Google Matrix by Al Bernstein Signal Science, LLC

Effects of polarization on the reflected wave

Definition of Tracking

1B40 Practical Skills

Department of Mechanical Engineering, University of Bath. Mathematics ME Problem sheet 11 Least Squares Fitting of data

INTRODUCTION TO COMPLEX NUMBERS

GAUSS ELIMINATION. Consider the following system of algebraic linear equations

UNIVERSITY OF IOANNINA DEPARTMENT OF ECONOMICS. M.Sc. in Economics MICROECONOMIC THEORY I. Problem Set II

Chapter Newton-Raphson Method of Solving a Nonlinear Equation

Introduction to Numerical Integration Part II

Course Review Introduction to Computer Methods

6. Chemical Potential and the Grand Partition Function

Abhilasha Classes Class- XII Date: SOLUTION (Chap - 9,10,12) MM 50 Mob no

Generalized Least-Squares Regressions I: Efcient Derivations

Electrochemical Thermodynamics. Interfaces and Energy Conversion

Demand. Demand and Comparative Statics. Graphically. Marshallian Demand. ECON 370: Microeconomic Theory Summer 2004 Rice University Stanley Gilbert

( dg. ) 2 dt. + dt. dt j + dh. + dt. r(t) dt. Comparing this equation with the one listed above for the length of see that

Polynomial Approximations for the Natural Logarithm and Arctangent Functions. Math 230

ESCI 342 Atmospheric Dynamics I Lesson 1 Vectors and Vector Calculus

Support vector machines for regression

Work and Energy (Work Done by a Varying Force)

Chapter 5 Supplemental Text Material R S T. ij i j ij ijk

Mathematics. Area under Curve.

Physics 121 Sample Common Exam 2 Rev2 NOTE: ANSWERS ARE ON PAGE 7. Instructions:

x yi In chapter 14, we want to perform inference (i.e. calculate confidence intervals and perform tests of significance) in this setting.

SVMs for regression Non-parametric/instance based classification method

INTERPOLATION(1) ELM1222 Numerical Analysis. ELM1222 Numerical Analysis Dr Muharrem Mercimek

Correlation and Regression. Correlation 9.1. Correlation. Chapter 9

Quadratic Forms. Quadratic Forms

Model Fitting and Robust Regression Methods

Linear Regression & Least Squares!

ES.182A Topic 32 Notes Jeremy Orloff

LINEAR ALGEBRA APPLIED

Remember: Project Proposals are due April 11.

Multiple view geometry

PROPERTIES OF AREAS In general, and for an irregular shape, the definition of the centroid at position ( x, y) is given by

International Journal of Pure and Applied Sciences and Technology

Chapter Runge-Kutta 2nd Order Method for Ordinary Differential Equations

Chemical Reaction Engineering

Applied Statistics Qualifier Examination

ARITHMETIC OPERATIONS. The real numbers have the following properties: a b c ab ac

Chapter 6 Continuous Random Variables and Distributions

5.7 Improper Integrals

M344 - ADVANCED ENGINEERING MATHEMATICS

Solubilities and Thermodynamic Properties of SO 2 in Ionic

CONIC SECTIONS. Chapter 11

Lecture 4: Piecewise Cubic Interpolation

Fundamental Theorem of Calculus

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 9

COMPLEX NUMBER & QUADRATIC EQUATION

Torsion, Thermal Effects and Indeterminacy

( ) ( )()4 x 10-6 C) ( ) = 3.6 N ( ) = "0.9 N. ( )ˆ i ' ( ) 2 ( ) 2. q 1 = 4 µc q 2 = -4 µc q 3 = 4 µc. q 1 q 2 q 3

Logarithms. Logarithm is another word for an index or power. POWER. 2 is the power to which the base 10 must be raised to give 100.

Partially Observable Systems. 1 Partially Observable Markov Decision Process (POMDP) Formalism

, MATHS H.O.D.: SUHAG R.KARIYA, BHOPAL, CONIC SECTION PART 8 OF

Statistics 423 Midterm Examination Winter 2009

7.2 Volume. A cross section is the shape we get when cutting straight through an object.

Lecture 21: Order statistics

p-adic Egyptian Fractions

Mathematics Number: Logarithms

along the vector 5 a) Find the plane s coordinate after 1 hour. b) Find the plane s coordinate after 2 hours. c) Find the plane s coordinate

Chapter 3 Single Random Variables and Probability Distributions (Part 2)

Goals: Determine how to calculate the area described by a function. Define the definite integral. Explore the relationship between the definite

Chemical Reaction Engineering

The Schur-Cohn Algorithm

VECTORS VECTORS VECTORS VECTORS. 2. Vector Representation. 1. Definition. 3. Types of Vectors. 5. Vector Operations I. 4. Equal and Opposite Vectors

10.2 The Ellipse and the Hyperbola

P 1 (x 1, y 1 ) is given by,.

Homework Assignment 6 Solution Set

Introduction to Algebra - Part 2

SVMs for regression Multilayer neural networks

Discrete Mathematics and Probability Theory Spring 2013 Anant Sahai Lecture 17

Chapter I Vector Analysis

Econ107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)

COMPLEX NUMBERS INDEX

An Ising model on 2-D image

Resistors. Consider a uniform cylinder of material with mediocre to poor to pathetic conductivity ( )

Calculus Module C21. Areas by Integration. Copyright This publication The Northern Alberta Institute of Technology All Rights Reserved.

Substitution Matrices and Alignment Statistics. Substitution Matrices

Online Appendix to. Mandating Behavioral Conformity in Social Groups with Conformist Members

Fall 2012 Analysis of Experimental Measurements B. Eisenstein/rev. S. Errede

Geometric Correction or Georeferencing

Section 14.3 Arc Length and Curvature

8.6 The Hyperbola. and F 2. is a constant. P F 2. P =k The two fixed points, F 1. , are called the foci of the hyperbola. The line segments F 1

PHYS 2421 Fields and Waves

Math 497C Sep 17, Curves and Surfaces Fall 2004, PSU

The practical version

Continuous Random Variables Class 5, Jeremy Orloff and Jonathan Bloom

Jens Siebel (University of Applied Sciences Kaiserslautern) An Interactive Introduction to Complex Numbers

Transcription:

More on χ nd errors : uppose tht we re fttng for sngle -prmeter, mnmzng: If we epnd The vlue χ ( ( ( ; ( wth respect to. χ n Tlor seres n the vcnt of ts mnmum vlue χ ( mn χ χ χ χ + + + mn mnmzes χ, nd thus: Then: χ ( s the mnmum vlue of χ 0. : χ, whch s eplctl desgnted s such, s: mn χ. Thus, ner the mnmum of χ,.e. n the vcnt of, then: χ χ χmn ( + χ χ χ mn + ( χ + ( + Note tht the ove epresson s nlogous to one we developed when we were dscussng the M.L.M. for severl prmeters (ee 598AEM Lect. Notes 3, p. 3-4: M M M ( + ( k k + ( k k ( r r + k k k r k r For onl sngle prmeter, ths epresson reduces to: When then: + + + ( 0, nd ner the mmum of (,.e. n the vcnt of ( ( ( + : 598AEM Lecture Notes 7

The nlog cn e mde even stronger recllng tht for ndependent Gussn/normll-,,, of the rndom vrle (, wth..f. dstruted mesurements ( then: f ( ( ; e π L ( ; ( ( ( ; ( ; ln ( χ π ( ln ln π Thus, we see tht: χ ( ; whch phscll mens tht the (negtve curvture of ( of the (postve curvture of χ ( ; t ts mnmum., t ts mmum s equl to Recll tht when we were llowed to truncte the Tlor seres epnson for ( t the qudrtc term, when the numer of mesurements ws ver lrge, we found tht: ( ˆ V, j ( E j ws the -j th element of the nverse of the covrnce mtr Vˆ of the ftted prmeters: ( V E ( ( j j ˆ [ ] For the stuton here, wth just sngle prmeter, we hve: mn, j χ ( ; E E + E + χ χ χ χ ( + ( mn + where we hve ssumed tht we cn replce ( : ( χ ( ; E E + E + 598AEM Lecture Notes 7

Ths epresson now descres how χ : χ vres n the neghorhood of mn ( mn + χ χ In prtculr, when the -prmeter vres n,,3,... stndrd devtons: ( ± n ( ± mn + mn + n χ χ χ or: χ ( χ ( χ ( Δ ± ± mn n Thus, we see tht chnge of χ ( χ ( χ mn ( mnmum vlue mn Δ ± ± from ts χ corresponds to stndrd devton chnge n the -prmeter. Lter on/shortl, we wll generlze ths to the cse of mn -prmeters Lest-qures Ft to trght Lne: Now we wll eplore n some detl the use of the LQ prncple n the cse where the ; depends lnerl on the -prmeters. theor We wll egn wth the χ ( ft of strght lne to mesured dt ponts (,,,. We mke ndependent mesurements ( ± t the correspondng ponts (n.. whch re ssumed to e known precsel,.e. 0 here. We mke the nstz,.e. hpothess tht: ( +, whch s the theor tht we use to predct the ( ; ( ;,,.e. ( ;, +. Best Ft Lne 3 3 3 n.. The est ft lne s sometmes clled the regresson lne non-phscs tpes... We defne the χ (, χ for the LQ ft to strght lne ( + s: ( ( ;, ( ( ( + ( ( 598AEM Lecture Notes 7 3

The est estmtes of the slope nd the ntercept nmel I. II. (, ( ( ( χ 0 (, ( ( ( χ 0 We rewrte these equtons s: nd re otned from: 0 0 I. II. + + It s conventonl to defne: nd: Then the ove equtons cn e wrtten compctl s: I. II. + + Two equtons & two unknowns solve these two equtons smultneousl nd: where: Net, we work on determnng the elements of the covrnce mtr of the ftted prmeters: Vˆ ( cov, cov (, We cn use generl methods to clculte the mtr elements, ut here we wll revert to more prmtve technque for purposes of llustrton: ropgton of Errors. We lred know tht the covrnce mtr of the mesurements, whch we re lws ssumng re ndependent, s of the dgonl form: 0 0 0 0 V ( 0 0 598AEM Lecture Notes 7 4

We lso hve the two functons: ( (, (,, ( nd: ( (, (,, ( Then, usng error propgton to determne e.g. the vrnce ssocted wth : + + + And: Where: But: And: Thus: ( Then: ( + + And now, summng over ll terms, we otn: + + + + + Thus:. Usng error propgton on., we otn: cov,, whch s < 0, re negtvel/nt-correlted wth ech other. Wth t more effort, we lso otn the covrnce: ( hence we see tht nd Thus, the covrnce mtr of ftted prmeters ssocted wth LQ-ft to strght lne + s: Vˆ cov, cov (, 598AEM Lecture Notes 7 5

In prctce, we must NEVER forget tht nd re (nt-correlted. Ther (nt- correlton s mportnt when, for emple, we tr to nterpolte or etrpolte the ftted lne. The result of the LQ ft s the est-ft lne : + Usng error propgton, the vrnce of ( + s: ( ( ( ( cov (, + + + + + + Wth lttle lger, nd rememerng tht we cn wrte ths s: cov (, ( + + If we hd gnored the fct tht nd re (nt- correlted, we would hve nsted otned: ( + + whch s qute dfferent (nd ncorrect. However, compre these two results when Let s work through numercl strght-lne ft emple whch ws generted to e n del cse: ˆ + 4+. We egn wth the theor : ˆ Net, we choose the 0,,,3, 4,5 ( 6 nd t ech -pont, we specf vlue for. Fnll, fnd 4 + nd then dd to t the product of nd rndom numer R, chosen from the Gussn/norml dstruton N ( 0, ths smultes Gussn error, or kck to ech ndvdul mesurement. Then cll the fnl numer ( ( 4 ( R Thus, the set of Monte Crlo mesurements + +. s gurnteed to e set of Gussn/ normll-dstruted rndom vrles wth gven/specfed nd followng known theor. 0 3 4 5 4 + 5 9 3 7 / 3/4 5/4 3/ R 0.58 0.848.04.67 0.6 0.60 R 0.08 0.85 0.78.46 0.6 0.90 0.9 4.5 9.78 4.46 7.6.90 598AEM Lecture Notes 7 6

Let us now ft these mesurements to + : 8.86.70 6.47 84.40 40.98 835.63 Gvng: 4.7 nd: 0.878 wth: Compre these results to ther true vlues: ˆ 4.0 nd ˆ.0. 0.793 0 5 The est ft lne s: 4.7 + 0.878 0 5 0 0 3 4 5 We lso determne the numercl vlues of the elements of the covrnce mtr of ftted prmeters: Vˆ cov, 0.0439 0.069 0.069 0.03 cov (, 598AEM Lecture Notes 7 7

Tkng the squre root of the dgonl elements of the covrnce mtr of ftted prmeters nd gnorng the (nt- correlton etween nd, we would s: Ftted slope: ± 4.3 ± 0. Ftted ntercept: ± 0.88 ± 0.45 whch re certnl consstent wth ˆ 4.0 nd ˆ.0. The correlton coeffcent s: ρ Is ths n greement wth our ntuton? As The mnmum vlue of χ occurs t: χ ( mn ( cov, 0.67 ncreses, does 6 ( χ,.078 decrese, nd vce vers? Yes! nce there re M 6 4 degrees of freedom, from the Vlue CL upper vs. χ grph on p. 4 of 598AEM Lect. Notes 6, we would epect to fnd χ >.078 to occur out ~ 70% of the tme, upon repetng ths eperment gzllon tmes. The χ per degree of freedom s: χ N of.078 4 0.5 ( <. The grphcl LQ ft result (see ove looks good, ee nd s good, quntttvel, from the χ LQ ft result. Now we nvestgte t more nltcll ths specfc cse tht of LQ ft wth two lner prmeters nd. We hve: χ (, or: ( + + + χ, + + + nce nd must e postve, nd χ, s 3-dmensonl surfce whose cross-sectons ( slces or contours of constnt χ n the horzontl - plne re ellpses, s shown n the fgure elow: s lso postve, we cn consder ( ( ( ( χ (, χ + + ρ χ + n mn mn ρ ( ellpse lng n, plne t heght n ove χmn 598AEM Lecture Notes 7 8

χ (, (, n.. the mjor/mnor es of the ellpse ren t prllel to the nd es unless 0, n whch cse: cov (, 0. For ths emple: χ, + + + 835.63 368.8+ 40.98 + 5.40.94+ 8.86 598AEM Lecture Notes 7 9

If we collpse the contours of constnt χ χ mn +, χ χ mn +, χ χ mn + m onto the - plne, we get feelng for how the mnmum looks, usng onl -dmensonl plot: 5.8 5.4 "est ft" (, (0.88, 4.3 5.0 4.6 contours of constnt χ + m mn 4. 3.8 3.4 "true" (, (.0,4.0 3.0 -.0-0.6-0. 0. 0.6.0.4.8. Below, we plot the sngle contour χ ( the error ellpse would not e tlted, nd, χ +. If nd were uncorrelted, nd would e the sem-mjor/mnor es. mn 598AEM Lecture Notes 7 0

For fed/constnt vlue of χ, the ellpses contours of constnt χ n the - plne re: ( ( ( ( + ρ n ρ The ellpse contour(s mke n ngle ϕ wth respect to the horzontl -s of: ρ ( cov, ϕ tn tn Alterntvel, we cn slce the 3- fgure through plne of constnt or χ, s functon of, or vce vers, s shown n the two fgures elow: nd plot Ths nterpretton s clerl WRONG f correltons est etween nd! 598AEM Lecture Notes 7

Etrpolton Fnd outsde the rnge of mesured : Wht s the vlue of ( t 0? From the theor we know tht Wht does the LQ ft predct? From the LQ ft result: ( + we get: And, from p. 6 of these Lecture Notes (ove: We eplctl see tht the 0 4 0 + 4. 0 4.7 0 + 0.878 43.5. + +. Thus, here: cov (, + + 0 ( 0.0439 ( 0.03 0 ( 0.069 4.39 + 0.03.58 3.335 term s the domnnt contrutor to for lrge, whch s ssocted wth the - uncertnt on the slope of the ftted lne: 0.0439 0.095 The - uncertnt on ( 0 Thus, we quote: s: 0 43.5 ±.83. ( ( 598AEM Lecture Notes 7 ( 3.335.83. 0 0 n.. Hd we not ncluded the (nt- correlton etween the LQ-ftted slope nd the ntercept, we would hve nsted otned: 4.39 + 0.03.4 ( >.83. 0 Interpolton Fnd nsde the rnge of mesured : Wht s the vlue of ( t 0.5? From the theor we know tht From the LQ ft results we get: ( 0.5 4.7 0.5 + 0.878.99 nd + + 0.5 ( 0.0439 ( 0.03 0.5 ( 0.069 0.00 + 0.03 0.069 0.5 Thus, we eplctl see tht for ner zero, the domnnt contruton s from the - uncertnt on the ntercept of the ftted lne: 0.03 0.45. The - uncertnt on ( 0.5 Thus, we quote: nd 0.5 4 0.5 + 3.0. s: 0.5 0.39. ( 0.5 ( 0.5 0.5.99 ± 0.39. Hd we gnored the (nt- correlton etween, we would hve nsted otned: 0.00 + 0.03 0.46 > 0.39 0.5 Comment on the correlton term: In ths nce emple, the omsson of the cov (, would not hve chnged the estmtes of ( term ver much However, n the rel world, the error mde gnorng the correlton term s sometmes ver lrge. Morl: never gnore t!