<<

Central Limit Theorem

Example: the distribution

Basic properties

nu nu-1 In[448]:= fgamma[x_, nu_, lambda_] := x Exp[-lambda x] Gamma[nu]

In[449]:= plotOrig = Plot[fgamma[x, 3, 1], {x, 0, 7}, Frame → True, AxesStyle → {Directive[Black, 15], Directive[Black, 15]}, LabelStyle → Directive[Black, 18], ImageSize → 800, FrameStyle → {Thick, Directive[Thick]}, PlotStyle → Black]

0.25

0.20

0.15

Out[449]=

0.10

0.05

0.00

0 1 2 3 4 5

Normalization : 2 L3m.nb

∞ In[450]:= FullSimplify fgamma[x, nu, lambda] ⅆ x, 0 Assumptions → {nu > 0, Re[lambda] > 0}

Out[450]= 1

Mean :

∞ In[451]:= = FullSimplify x fgamma[x, nu, lambda] ⅆ x, 0 Assumptions → {nu > 0, Re[lambda] > 0}

nu Out[451]= lambda Variance :

∞ 2 2 In[452]:= sigma2 = FullSimplify x fgamma[x, nu, lambda] ⅆ x - mu , 0 Assumptions → {nu > 0, Re[lambda] > 0}

nu Out[452]= lambda2 Skewness :

In[453]:= gamma1 = FullSimplify 1 ∞  (x - mu)3 fgamma[x, nu, lambda] ⅆ x, 3/2 (sigma2) 0 Assumptions → {nu > 0, Re[lambda] > 0}

2 Out[453]= nu

Kurtosis : L3m.nb 3

In[454]:= beta2 = FullSimplify 1 ∞  (x - mu)4 fgamma[x, nu, lambda] ⅆ x, 2 (sigma2) 0 Assumptions → {nu > 0, Re[lambda] > 0}

6 Out[454]= 3 + nu

In[455]:= gamma2 = beta2 - 3

6 Out[455]= nu

Central limit theorem As in lecture, let' s change variables to y = (x - mu) . Then characteristic function is : sigma2

In[456]:= FullSimplify ∞ (x - mu)  Expⅈ t  fgamma[x, nu, lambda] ⅆ x, 0 N sigma2 Assumptions → Re[nu] > 0 && ((Im[t] + Re[lambda] ⩵ 0 && Re[nu] < 1) || Im[t] + Re[lambda] > 0) && t Im  + Re[lambda] > 0 && lambda > 0 N nu lambda2

ⅈ -nu - nu t ⅈ t Out[456]= ⅇ N nu 1 - N nu 4 L3m.nb

Characteristic function defined :

-nu - ⅈ nu t ⅈ t In[457]:= [t_, nu_, N_] := ⅇ N nu 1 - N nu For our example let' s focus on nu = 1, the exponential distribution :

In[470]:= plotOrig2 = Plot[fgamma[x, 1, 1], {x, 0, 5}, Frame → True, AxesStyle → {Directive[Black, 15], Directive[Black, 15]}, LabelStyle → Directive[Black, 18], ImageSize → 800, FrameStyle → {Thick, Directive[Thick]}, PlotStyle → Black]

1.0

0.8

0.6

Out[470]=

0.4

0.2

0.0

0 1 2 3

Now let' s determine the distribution of N s that satisfy L3m.nb 5

distribution of N i.i.d' s that satisfy the underlying Gamma distribution : (this integrates analytially to the exponential integral but we' ll evaluate numerically) 1 In[458]:= fN[y_, N_] := NIntegrate 2 π Exp[-ⅈ t y] (phi[t, 1, N])N, {t, -20, 20} Let' s evaluate this for increasing N and compare to the normal distribution :

In[459]:= barfN1 = Table[{xx, Re[fN[xx - 1, 1]]}, {xx, -4, 4, .1}]; fN1plot = ListPlot[barfN1, PlotStyle → Red, Frame → True, Joined → True, PlotRange → {{-5, 5}, {-.1, 1.1}}]; barfN3 = Table[{xx, Re[fN[xx - 1, 3]]}, {xx, -4, 4, .1}]; fN3plot = ListPlot[barfN3, PlotStyle → Green, Frame → True, Joined → True]; barfN5 = Table[{xx, Re[fN[xx - 1, 10]]}, {xx, -4, 4, .1}]; fN5plot = ListPlot[barfN5, PlotStyle → Blue, Frame → True, Joined → True]; Here' s the normal distribution :

1 (xx - 1)2 In[465]:= fgauss[xx_] := Exp-  2 π 2

In[466]:= plotgauss = Plot[fgauss[xx], {xx, -5, 5}, PlotStyle → Black]; 6 L3m.nb

In[471]:= Show[plotOrig2, fN1plot, fN3plot, fN5plot, plotgauss, PlotRange → {{-2, 5}, {-.1, 1.1}},

FrameLabel → {"y", "fN(y)"}, LabelStyle → Directive[Black, 25]]

1.0

0.8

0.6 ) y ( N f Out[471]= 0.4

0.2

0.0

-2 -1 0 1 2 3 y

Example

Basic properties

3 1 In[472]:= ftest[x_] := 2 (1 + Abs[x])4 L3m.nb 7

In[473]:= plotOrig = LogPlot[ftest[x], {x, -5, 5}, Frame → True, AxesStyle → {Directive[Black, 15], Directive[Black, 15]}, LabelStyle → Directive[Black, 18], ImageSize → 800, FrameStyle → {Thick, Directive[Thick]}, PlotStyle → Black]

1

0.500

0.100

0.050 Out[473]=

0.010

0.005

0.001 -4 -2 0 2

Normalization :

∞ In[532]:=  ftest[x] ⅆ x -∞

Out[532]= 1

Mean : 8 L3m.nb

∞ In[475]:= mu =  x ftest[x] ⅆ x -∞

Out[475]= 0

Variance :

∞ 2 In[476]:= sigma2 =  x ftest[x] ⅆ x -∞

Out[476]= 1

Note that higher moments diverge!

∞ 3 In[477]:=  x ftest[x] ⅆ x -∞

3 x3 Integrate: Integral of does not converge on {-∞, ∞}. 2 (1 + Abs[x])4

∞ 3 x3 Out[477]=  ⅆx -∞ 2 (1 + Abs[x])4