Scribd Logo
Upload

Welcome to Scribd!

  • 14 views

    1B40 Practical Skills: Properties of The Gaussian Distribution

    Uploaded by

    Roy Vesey
    The document discusses key properties of the Gaussian or normal distribution: 1) It is symmetric about the mean and its shape is determined by the standard deviation. 2) The fraction of values within ±z standard deviations of the mean can be calculated using the error function erf(z). Approximately two-thirds of values lie within ±1 standard deviation. 3) Outliers more than 3 standard deviations from the mean are quite unlikely based on the normal distribution, occurring in only about 1 in 400 cases.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    1B40 Practical Skills: Properties of The Gaussian Distribution

    Uploaded by

    Roy Vesey
    14 views3 pages
    The document discusses key properties of the Gaussian or normal distribution: 1) It is symmetric about the mean and its shape is determined by the standard deviation. 2) The fraction of values within ±z standard deviations of the mean can be calculated using the error function erf(z). Approximately two-thirds of values lie within ±1 standard deviation. 3) Outliers more than 3 standard deviations from the mean are quite unlikely based on the normal distribution, occurring in only about 1 in 400 cases.

    Original Description:

    1B40 Physics Practical Skills - Data Analysis Lecture 5 (UCL)

    Original Title

    1B40 DA Lecture 5v2

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    The document discusses key properties of the Gaussian or normal distribution: 1) It is symmetric about the mean and its shape is determined by the standard deviation. 2) The fraction of values within ±z standard deviations of the mean can be calculated using the error function erf(z). Approximately two-thirds of values lie within ±1 standard deviation. 3) Outliers more than 3 standard deviations from the mean are quite unlikely based on the normal distribution, occurring in only about 1 in 400 cases.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    14 views3 pages

    1B40 Practical Skills: Properties of The Gaussian Distribution

    Uploaded by

    Roy Vesey
    The document discusses key properties of the Gaussian or normal distribution: 1) It is symmetric about the mean and its shape is determined by the standard deviation. 2) The fraction of values within ±z standard deviations of the mean can be calculated using the error function erf(z). Approximately two-thirds of values lie within ±1 standard deviation. 3) Outliers more than 3 standard deviations from the mean are quite unlikely based on the normal distribution, occurring in only about 1 in 400 cases.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 3

    1B40 Practical Skills

    Properties of the Gaussian distribution


    The (Gaussian) normal distribution has the functional form
    1  − ( x − µ )2 
    p ( x) = exp  
    σ 2π  2σ
    2
    
    For simplicity we will set µ = 0.

    0.6

    X= 0, σ =5

    0.4
    f(x)

    0.2

    0
    -20 -15 -10 -5 0 5 10 15 20

    What fraction of the results lie between -x to +x?


    The probability density function f(x) is the fraction of results that lie between x and x + dx.
    Define φ(x) such that:
    x

    φ ( x) = ∫ p ( x′) dx′,
    −x

    1  1  x′  2 
    +x

    φ ( x) = ∫ exp  − 2  σ   dx′.
    σ 2π − x  
    Let t = x'/σ and z = x/σ then φ(x) becomes
    +z
    1  t2 
    φ (z) = ∫ exp − 2  σ dt.
    σ 2π −z
    This is symmetric about t = 0 and can be written as
    2 +z  t2 
    φ ( z) =
    2π ∫0
    exp −  .
     2
    The function erf(z) is referred to as the Error Function and is defined by
    2 z
    erf ( z ) = ∫ exp  −t 2  dt
    π 0

    and is tabulated in mathematical tables. Thus we have

    1
     z 
    φ ( z ) = erf  .
     2
    It represents the fraction of p(x) that lies within ± z standard deviations of the mean of the
    distribution.

    z = x/σ φ(z) Approximate fraction of readings


    outside ±z
    0 0 1
    1 0.683 1/3
    2 0.9540 1/20
    3 0.9973 1/400
    4 0.9994 1/16000

    The erf function is plotted below.


    1.0

    0.8

    0.6
    erf(z)

    0.4

    0.2

    0
    0 0 .5 1.0 1.5 2 .0

    z = x /σ

    Important points
    • Two out of three observations lie within ± σ (one in three outside).
    • About one in twenty observations lie outside ± 2σ
    • About one in 400 observations lie outside ± 3σ
    • About one in 16,000 observations lie outside ± 4σ, i.e. there is one chance in 16,000 that
    the true value lies outside this range.

    You can use these results to check that σ has been estimated correctly. Roughly two thirds of
    the readings should lie between x ± σ . When you quote a result as x ± σ m you imply the
    probability that the true value lies in the quoted range is roughly two thirds.

    2
    Treatment of suspect results
    The discussion above on the properties of the normal distribution can help guide us when we
    have “suspect” results. Consider the following set of experimental data:

    Residual ( t i − t ) (10-4 s2)


    2
    Time t (s)
    5.38 25
    5.42 1
    5.48 25
    5.30 169
    5.34 81
    5.29 196
    5.97 2916
    5.32 121
    5.40 9
    mean = 5.43 s sum=39367
    σ=(39367/9)1/2 =2 ×10-1 s
    σm=σ/3 =7×10-2 s

    The result is T = (5.43 ± 0.07) s. The results are well clustered about the mean value except for
    the 5.97 value. If you recalculated the mean omitting this result you get T = (5.36 ± 0.02) s.

    Are you justified in neglecting this one reading? In general you never ignore the result of an
    experiment in this way. However you can calculate the probability that the suspect result is
    valid. The standard deviation of the distribution is ±0.20 seconds. (Note in the final result the
    error quoted is standard error on the mean, not the standard deviation of the sample). The
    suspect reading is about 3 standard deviations away from the mean; the probability of its being a
    valid part of the distribution is about one part in 400 - i.e. not impossible, but unlikely!

    You may be justified in ignoring a particularly unusual result BUT ONLY IF YOU CAN
    SPOT THE SYSTEMATIC ERROR WHICH HAS LED TO THE ODD RESULT.

    You can usually do this as you are doing the experiment. If for example you are determining
    the value of some well known constant you may sometimes see (with hindsight) that a single
    unusual measurement has led you to an inaccurate value for the result. Your alternatives are:
    1. Repeat the entire experiment taking care to avoid what happened before (Best).
    2. Comment on what has happened, noting the likely source of error in your final result
    (Next best).
    3. Remove the single odd measurement and recompute the answer (OK sometimes but
    least preferable – could be dishonest!).

    You might also like