Untangling complex syste.., p.101

Untangling Complex Systems, page 101

 

Untangling Complex Systems
Select Voice:
Brian (uk)
Emma (uk)  
Amy (uk)
Eric (us)
Ivy (us)
Joey (us)
Salli (us)  
Justin (us)
Jennifer (us)  
Kimberly (us)  
Kendra (us)
Russell (au)
Nicole (au)



Larger Font   Reset Font Size   Smaller Font  



  tion of the traveled distance and the time spent; if we want to know the density of a material we need

  to measure its mass and its volume.

  524

  Appendix D

  40

  3 )

  V = L 3

  V(cm 30

  ( Lbest + dL)3

  20

  ( Lbest)3

  ( Lbest − dL)3 10

  00.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4

  L (cm)

  Lbest − dL Lbest Lbest + dL

  FIGURE D.8 The propagation of the L uncertainty in the determination of the volume.

  best

  How do the errors made in the direct measurements propagate in the estimation of the

  derived variables? Imagine having measured the length L of a cube’s side directly and wanting

  to determine its volume. If Lbest = (2.6 ± .

  0 )

  1 cm, then the best estimate of the volume will be

  V

  3

  3

  best = ( Lbest ) = 17.576 cm . How much is the uncertainty in the volume? (Figure D.8) If the uncertainty in L is small, the values of V for L = Lbest + dL and L = Lbest − dL are pretty close to V . The function V( L), around V , can be approximated by a straight line. In other words, best

  best

  V( L) can be described by a truncated Taylor’s series expansion having V as its accumulation point: best

  dV 

  V( Lbest + dL) = V( Lbest) + 

  dL V

  

  

  = best + dV [D.6]

   dL  Lbest

  dV 

  V( Lbest − dL) = V( Lbest) − 

  dL V

  

  

  = best − dV [D.7]

   dL  Lbest

  It derives that the uncertainty in the volume is dV = ( dV dL)

  dL = (3 L 2 ) dL = 2 028 3

  .

  cm . The

  L

  best

  best

  final result is V

  3

  best = (17.6 ± .

  2 0) cm . This analysis of the error’s propagation can be generalized. For

  any function z of the variable x that is measured directly with an uncertainty dx, the error will be: dz 

  dz = 

  dx

  

  

  [D.8]

   dx  xbest

  Note that in equation [D.8], we have the absolute value of the first derivative because sometimes the

  first derivative may be negative, but the error never decreases when it propagates. If z is a function

  of two or more variables, x , , …., , the uncertainty will be:

  1 x 2

  xn

   dz 

   dz 

   dz 

  dz =

  dx 1

  dx

  

  

  + 

  

  2 +… + 

  

  dx [D.9]

   dx

  n

  1 

   dx 2 

   dx

  x

  

  1,

  x

  n

  best

  2, best

  xn, best

  Appendix D

  525

  The uncertainty in z determined through either the [D.8] or [D.9] formula is called the maximum

  a priori absolute error. In the absence of miscalibration errors, the z value will be surely within

  true

  the range  z

   best + dz, zbest − dz. When the uncertainties in the variables x

  that have been

  1 , x 2 , …, xn

  measured directly, are independent and random, we may use formula [D.10] for estimating the

  uncertainty in z.

  

  2

  2

  2

   dz 

  

    dz 

  

    dz 

  

  dz = 

  dx 

   +…+ 

  dx 

  1

  + 

  dx

  

  

  [D.10]

  

  

  

  

  

   dx

  2

  

  

   dx

  n

  n

  

  1 

  

   dx

  x

  2 

  

  

  1, best

  x 2, best

  

  

  

  

  

  xn, best

  

  Alternatively, we may calculate as many values of z as are those collected for the independent

  variables x

  . Let us assume to have repeated

  1 , x 2 , …, x

  m measurements for each variable. The best

  n

  estimation of z and its uncertainty is the average ( z

  m

  = (∑ z / and the standard deviation of the

  =1

  ) )

  j

  j

  m

  mean ( S

  m

  1

  2

  z =

  (

  ) /

  , respectively.

  1 ∑

  z − z

  m

  ( − =1

  ) )

  m

  j

  j

  d.6 sTudenT’s T-disTribuTion

  We have already learned that when we repeat the measurement of a value of the variable x, n times,

  with n → ∞, and the uncertainty is mainly random,3 the data are well-described by a Gaussian function. The Gaussian function will be centered on the mean value x , and its width will be the standard

  deviation of the mean, σ x. By exploiting the Gaussian distribution, we can calculate the probability

  that a further determination will be within the range x ± gσ x. Such probability is a function of the parameter g (see the values listed in Table D.1). For instance, the probability that a measurement is within the range  x +

  

  σ x, x −σ x , with g = 1, amounts to 68.27%.

  TABLE D.1

  Percent Probability (Prob) That a

  Measurement is within the Range

   x + gσ

  

  x, x − gσ x 

   as a Function of g and

  Calculated under the Hypothesis That the

  Measurements are Described by a

  Gaussian distribution

  g

  Prob (%)

  0.00

  0.00

  0.50

  38.29

  0.68

  50.35

  1.00

  68.27

  1.04

  70.17

  1.44

  85.01

  1.50

  86.64

  1.65

  90.11

  1.96

  95.00

  2.00

  95.45

  2.50

  98.76

  3.00

  99.73

  3 In other words, ε random >> ε sys.

  526

  Appendix D

  Most often, we cannot repeat the same measurement an infinite number of times (i.e., many

  times), but just a few. In such cases, our data are not properly described by a Gaussian distribution

  but a Student’s t- distribution. The Student’s t- distribution is defined for the continuous random variable t = ( x − xGauss) / Sx where xGauss is the expectation value of the corresponding Gaussian distribution and Sx is the experimental standard deviation of the mean x . The probability density function

  of the t variable is

  ν +1

  ν

  Γ

  

  +1

  1

  2

  

  2 − 2

  

  

  t

  f ( t,ν ) =

  1+

  [D.11]

  πν

  ν 

  ν 

  Γ  

  

   2 

  In equation [D.11], ν is the number of the degrees of freedom that is equal to m − 1, if m is the

  finite number of times a measurement has been repeated; Γ is the gamma function. The function

  f ( t,ν ) is symmetric and bell-shaped (like the Gaussian distribution), but it has heavier tails than the Gaussian distribution; its expectation value is zero, and its variance is ν / ν

  ( − 2) for ν > 2. If ν → ∞,

  the t-distribution approaches a Gaussian distribution centered at 0 and with a standard deviation

  σ = 1. The probability that the ( m + 1) - th measurement is within the interval  x + tS

  

  x , x − tSx 

   is a

  function of t and ν; some values are reported in Table D.2. The last row of Table D.2 (with ν = ∞) corresponds to the values defined for the Gaussian distribution.

  We may use the Student’s t- distribution for comparing the mean values of two independent sets

  of measurements, xA and xB, and ask us if they belong to the same distribution. This test is suitable to ascertain if the experimental conditions have not changed in the two series of measurements or if

  the measurements of the same physical quantity carried out in different laboratories can be merged

  to give a single best estimate. First, we calculate the t-value through the following formula

  xA x

  t

  B

  =

  −

  [D.12]

  S 2

  2

  A

  SB

  +

  nA nB

  TABLE D.2

  Values of the t Variable Expressed as a Function of the Percentage Probability (Prob) That

  a Further Measurement is within the Range  x + tS

  

  x, x − tSx 

  . The t-Value also Depends

  on the Degrees of Freedom ν = m − 1

  Prob (%)

  ν

  50

  60

  70

  80

  90

  95

  99

  1

  1.000

  1.376

  1.963

  3.078

  6.314

  12.706

  63.657

  2

  0.816

  1.061

  1.386

  1.886

  2.920

  4.303

  9.925

  3

  0.765

  0.978

  1.250

  1.638

  2.353

  3.182

  5.841

  4

  0.741

  0.941

  1.190

  1.533

  2.132

  2.776

  4.604

  5

  0.727

  0.920

  1.156

  1.476

  2.015

  2.571

  4.032

  6

  0.718

  0.906

  1.134

  1.440

  1.943

  2.447

  3.707

  7

  0.711

  0.896

  1.119

  1.415

  1.895

  2.365

  3.499

  8

  0.706

  0.889

  1.108

  1.397

  1.860

  2.306

  3.355

  9

  0.703

  0.883

  1.100

  1.383

  1.833

  2.262

  3.250

  10

  0.700

  0.879

  1.093

  1.372

  1.812

  2.228

  3.169

  20

  0.687

  0.860

  1.064

  1.325

  1.725

  2.086

  2.845

  30

  0.683

  0.854

  1.055

  1.310

  1.697

  2.042

  2.750

  ∞

  0.674

  0.842

  1.036

  1.282

  1.645

  1.960

  2.576

  Appendix D

  527

  In [D.12], S 2

  2

  A and SB are the variances (i.e., the standard deviations S and S , both to the power of A

  B

  two) of the two series of measurements, A and B, whose numbers of elements are n and n , respec-

  A

  B

  tively. If the estimated t-value is larger than the expected t- value for a confidence level of the 95%, we may assert that the experimental conditions have been changed. Otherwise, the two series may

  be considered as belonging to the same distribution and may be used to determine a better estimate

  through the calculation of the weighted average:

  xA xB

  +

  2

  2

  x

  SA SB

  best =

  [D.13]

  1 + 1

  S 2

  2

  A

  SB

  The formula [D.13] can be generalized to more than two separate sets of measurements that are

  consistent among them. If the number of independent sets of measurements is k, x 1, x 2,…, xk are the corresponding mean values, and S , , …, are their standard deviations, the best estimate is

  1 S 2

  Sk

  k wixi

  ∑

  x

  i 1

  best =

  =

  [D.14]

  k wi

  ∑ i=1

  The terms w

  2

  i = 1 / Si are called weights. The larger the standard deviation, the smaller the weight.

  The sets of measurements that are much less precise than the others contribute very much less to

  xbest. For example, if one set of measurements is three times less precise than the others, its weight is

  nine times smaller than the other weights. The uncertainty in the weighted average xbest is

  k

  

  −1/2

  S

  

  

  x =

  wi

  ∑ 

  [D.15]

   i=1

  

  d.7 leasT-sQuares fiTTing

  Sometimes the final goal of an experiment is the determination of the mathematical equation relat-

  ing two variables. For instance, a chemist may be interested in determining the order of a chemical

  reaction with respect to a reagent X. In other words, the chemist may be interested in determining

  the exponent that appears in the equation v k X nX

  = [ ] . The chemist measures the reaction rates ( v ,

  1

  v , …, ) at different concentrations of

  , [

  , …, [

  ) maintaining constant all the other

  2

  v

  X ([ X]

  X]

  X]

  n

  1

  2

  n

  experimental conditions. It is convenient to work with the logarithmic form of the kinetic equa-

  tion ( v = k[ X ] nX ), i.e., log( v) = log( k) + nX log[ X ]. In fact, by plotting log( v ), log( ), …, log( ) vs.

  1

  v 2

  vn

  log[ X] , log[

  , …, log[

  , the experimental data can be described by a straight line, y = A + Bx.

  1

  X]2

  X] n

  The slope is B = nX, and the intercept is A = log( k). The best estimates of the coefficients A and B

  are obtained by the linear regression, also called the least-squares method.

  More in general, imagine having collected pairs of data ( x

  ), (

  ), …, (

  ). The uncer-

  1 , y 1

  x 2 , y 2

  x , y

  n

  n

  tainties on the x values are negligible with respect to those on the y values, and the uncertainties on i

  i

  the y values are of the random type. Each y value belongs to a Gaussian distribution centered on its i

  i

  true value that is ( A + Bx ). Therefore, the probability of collecting y is:

  i

  i

  ( y

  2

  i − A− Bxi )

  −

  1

  2

  σ

  P y

  yi

  ( i) ∝

  e

  [D.16]

  σ yi

  528

  Appendix D

  The probability of obtaining the complete set of independent measurements, y , , …, , is the

  1 y 2

  yn

  product of the probabilities of obtaining every single value:

  n ( y

  2

  i − A− Bx )

  n

  i

  

  1  −∑

  2

  σ

  P( y , y , ,

  … y

  i 1

  yi

  ) ∝ 

  

  1

  2

  n

  e

  ∏

  =

  [D.17]

  σ 

   i

  yi

  1

  =

  

  According to the principle of maximum likelihood, the best estimates of the A and B parameters are

  those that maximize the total probability, P( y

  )

  1, y 2,

  ,

  … yn , or minimize the exponent4

  n

  2

  ( iy − A− Bxi)

  χ2 = ∑

  [D.18]

  σ 2

  i 1

  =

  yi

  For the determination of these values, we differentiate χ 2 with respect to A and B and set the derivatives equal to zero:

  ∂χ2

  n

  = −

  wi yi A Bx

  A

  ∑

  2

  ( − − i ) = 0

  ∂

  i=1

  [D.19]

  ∂χ2

  n

  = −

  wixi yi A Bx

 

Add Fast Bookmark
Load Fast Bookmark
Turn Navi On
Turn Navi On
Turn Navi On
Scroll Up
Turn Navi On
Scroll
Turn Navi On
183