Download Probability and combinatorics Michael Anshelevich May 1, 2012 Texas A&M University

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
Probability and combinatorics
Michael Anshelevich
Texas A&M University
May 1, 2012
Michael Anshelevich
Probability and combinatorics
Probability spaces.
(Λ, M, P ) = measure space.
Probability space: P a probability measure, P (Λ) = 1.
Michael Anshelevich
Probability and combinatorics
Probability spaces.
(Λ, M, P ) = measure space.
Probability space: P a probability measure, P (Λ) = 1.
Algebra A = L∞ (Λ, P ) of bounded random variables.
R
E[X] = X dP = expectation functional on A.
Michael Anshelevich
Probability and combinatorics
Probability spaces.
(Λ, M, P ) = measure space.
Probability space: P a probability measure, P (Λ) = 1.
Algebra A = L∞ (Λ, P ) of bounded random variables.
R
E[X] = X dP = expectation functional on A.
For each real-valued X, have µX = probability measure on R
defined by
Z
Z
f (x) dµX (x) =
f (X) dP = E[f (X)]
Λ
R
for f ∈ C0 (R).
µX = distribution of X.
Michael Anshelevich
Probability and combinatorics
Independence.
More generally, if X1 , X2 , . . . , Xn random variables,
µX1 ,X2 ,...,Xn = measure on Rn = joint distribution.
Definition. X, Y are independent if
µX,Y = µX ⊗ µY (product measure)
i.e.
E[f (X)g(Y )] = E[f (X)]E[g(Y )]
Michael Anshelevich
Probability and combinatorics
Independence.
More generally, if X1 , X2 , . . . , Xn random variables,
µX1 ,X2 ,...,Xn = measure on Rn = joint distribution.
Definition. X, Y are independent if
µX,Y = µX ⊗ µY (product measure)
i.e.
E[f (X)g(Y )] = E[f (X)]E[g(Y )]
Remark. If X, Y independent
Z
Z
f (t) dµX+Y (t) = f (x + y) dµX,Y (x, y)
Z
Z
= f (x + y) d(µX ⊗ µY ) = f (t) d(µX ∗ µY ).
So in this case, µX+Y = µX ∗ µY .
Michael Anshelevich
Probability and combinatorics
Fourier transform.
Definition. Fourier transform
Z
FX (θ) = eiθx dµX (x) = E[eiθX ]
Michael Anshelevich
Probability and combinatorics
Fourier transform.
Definition. Fourier transform
Z
FX (θ) = eiθx dµX (x) = E[eiθX ]
Lemma. If X, Y independent,
FX+Y (θ) = FX (θ)FY (θ).
Michael Anshelevich
Probability and combinatorics
Fourier transform.
Definition. Fourier transform
Z
FX (θ) = eiθx dµX (x) = E[eiθX ]
Lemma. If X, Y independent,
FX+Y (θ) = FX (θ)FY (θ).
Proof.
E[eiθ(X+Y ) ] = E[eiθX eiθY ] = E[eiθX ]E[eiθY ]
Michael Anshelevich
Probability and combinatorics
Combinatorics.
Z
FX (θ) =
eiθx dµX (x) =
∞
X
(iθ)n
n=0
Z
mn =
n!
mn (X),
xn dµX (x) = E[X n ].
{m0 , m1 , m2 , . . .} = moments of X.
Michael Anshelevich
Probability and combinatorics
Combinatorics.
Z
FX (θ) =
eiθx dµX (x) =
∞
X
(iθ)n
n=0
Z
mn =
n!
mn (X),
xn dµX (x) = E[X n ].
{m0 , m1 , m2 , . . .} = moments of X.
For X, Y independent, mn (X + Y ) complicated. But:
FX+Y (θ) = FX (θ)FY (θ),
log FX+Y (θ) = log FX (θ) + log FY (θ).
Denote `X (θ) = log FX+Y (θ).
Michael Anshelevich
Probability and combinatorics
Cumulants.
`X (θ) =
∞
X
(iθ)n
n=1
n!
cn ,
{c1 , c2 , c3 , . . .} = cumulants of X.
cn (X + Y ) = cn (X) + cn (Y ).
Michael Anshelevich
Probability and combinatorics
Cumulants.
`X (θ) =
∞
X
(iθ)n
n=1
n!
cn ,
{c1 , c2 , c3 , . . .} = cumulants of X.
cn (X + Y ) = cn (X) + cn (Y ).
Relation between {mn }, {cn }?
Michael Anshelevich
Probability and combinatorics
Cumulants.
`X (θ) =
∞
X
(iθ)n
n=1
n!
cn ,
{c1 , c2 , c3 , . . .} = cumulants of X.
cn (X + Y ) = cn (X) + cn (Y ).
Relation between {mn }, {cn }?
A set partition
{(1, 3, 4), (2, 7), (5), (6)} ∈ P(7).
Michael Anshelevich
Probability and combinatorics
Moment-cumulant formula.
Proposition.
mn =
X Y
c|B| .
π∈P(n) B∈π
Michael Anshelevich
Probability and combinatorics
Moment-cumulant formula.
Proposition.
mn =
X Y
c|B| .
π∈P(n) B∈π
mean
m1 = c1 ,
c1 = m1
m2 = c2 + c21 ,
c2 = m2 − m21
m3 = c3 + 3c2 c1 +
c31 ,
c3 = m3 −
Michael Anshelevich
3m2 m21
variance
+
2m31
Probability and combinatorics
Moment-cumulant formula.
Proposition.
mn =
X Y
c|B| .
π∈P(n) B∈π
mean
m1 = c1 ,
c1 = m1
m2 = c2 + c21 ,
c2 = m2 − m21
c31 ,
c3 = m3 −
` = log F,
F = e` ,
m3 = c3 + 3c2 c1 +
Michael Anshelevich
3m2 m21
variance
+
2m31
F 0 = `0 F.
Probability and combinatorics
Moment-cumulant formula.
Proposition.
mn =
X Y
c|B| .
π∈P(n) B∈π
mean
m1 = c1 ,
c1 = m1
m2 = c2 + c21 ,
c2 = m2 − m21
c31 ,
c3 = m3 −
` = log F,
F = e` ,
m3 = c3 + 3c2 c1 +
mn+1 =
n X
n
k=0
Michael Anshelevich
k
3m2 m21
variance
+
2m31
F 0 = `0 F.
ck+1 mn−k .
Probability and combinatorics
Central limit theorem.
Theorem. Let {Xn : n ∈ N} be independent, identically
distributed, mean 0, variance v.
E[Xn ] = 0,
Let
Sn =
E[Xn2 ] = v.
X1 + X2 + . . . + Xn
√
.
n
Then the moments of Sn converge to the moments of the
normal distribution N (0, v).
Michael Anshelevich
Probability and combinatorics
Central limit theorem.
Proof. For each k,
ck (αX) = αk ck (X).
Michael Anshelevich
Probability and combinatorics
Central limit theorem.
Proof. For each k,
ck (αX) = αk ck (X).
ck (Sn ) = ck
X1 + X2 + . . . + Xn
√
n
Michael Anshelevich
n
= √ k ck (X1 ).
( n)
Probability and combinatorics
Central limit theorem.
Proof. For each k,
ck (αX) = αk ck (X).
ck (Sn ) = ck
X1 + X2 + . . . + Xn
√
n
n
= √ k ck (X1 ).
( n)
(k = 1) c1 (X1 ) = 0, c1 (Sn ) = 0.
Michael Anshelevich
Probability and combinatorics
Central limit theorem.
Proof. For each k,
ck (αX) = αk ck (X).
ck (Sn ) = ck
X1 + X2 + . . . + Xn
√
n
n
= √ k ck (X1 ).
( n)
(k = 1) c1 (X1 ) = 0, c1 (Sn ) = 0.
(k = 2) c2 (X1 ) = v, c2 (Sn ) = v.
Michael Anshelevich
Probability and combinatorics
Central limit theorem.
Proof. For each k,
ck (αX) = αk ck (X).
ck (Sn ) = ck
X1 + X2 + . . . + Xn
√
n
n
= √ k ck (X1 ).
( n)
(k = 1) c1 (X1 ) = 0, c1 (Sn ) = 0.
(k = 2) c2 (X1 ) = v, c2 (Sn ) = v.
n
(k > 2) k/2 → 0, ck (Sn ) → 0.
n
Michael Anshelevich
Probability and combinatorics
Central limit theorem.
Proof. For each k,
ck (αX) = αk ck (X).
ck (Sn ) = ck
X1 + X2 + . . . + Xn
√
n
n
= √ k ck (X1 ).
( n)
(k = 1) c1 (X1 ) = 0, c1 (Sn ) = 0.
(k = 2) c2 (X1 ) = v, c2 (Sn ) = v.
n
(k > 2) k/2 → 0, ck (Sn ) → 0.
n
In the limit, get whichever distribution has
(
v, k = 2,
ck =
0, otherwise.
Check: normal distribution.
Michael Anshelevich
Probability and combinatorics
Central limit theorem.
Proof. For each k,
ck (αX) = αk ck (X).
ck (Sn ) = ck
X1 + X2 + . . . + Xn
√
n
n
= √ k ck (X1 ).
( n)
(k = 1) c1 (X1 ) = 0, c1 (Sn ) = 0.
(k = 2) c2 (X1 ) = v, c2 (Sn ) = v.
n
(k > 2) k/2 → 0, ck (Sn ) → 0.
n
In the limit, get whichever distribution has
(
v, k = 2,
ck =
0, otherwise.
P
Check: normal distribution. Note mn = π∈P2 (n) v n/2 .
Michael Anshelevich
Probability and combinatorics
Operators.
H = real Hilbert space, e.g. Rn .
HC = its complexification (Cn ).
Michael Anshelevich
Probability and combinatorics
Operators.
H = real Hilbert space, e.g. Rn .
HC = its complexification (Cn ).
HC⊗n = HC ⊗ HC ⊗ . . . ⊗ HC
= symmetric tensor product
= Span ({h1 ⊗ h2 ⊗ . . . ⊗ hn , order immaterial})
with the inner product
hh1 ⊗ . . . ⊗ hn , g1 ⊗ . . . ⊗ gn i =
X
h1 , gσ(1) . . . hn , gσ(n)
σ∈Sym(n)
(degenerate inner product).
Michael Anshelevich
Probability and combinatorics
Creation and annihilation operators.
Symmetric Fock space
F(HC ) =
∞
M
HC⊗n = CΩ ⊕ HC ⊕ HC⊗2 ⊕ HC⊗3 . . . ,
n=0
Ω = vacuum vector.
Michael Anshelevich
Probability and combinatorics
Creation and annihilation operators.
Symmetric Fock space
F(HC ) =
∞
M
HC⊗n = CΩ ⊕ HC ⊕ HC⊗2 ⊕ HC⊗3 . . . ,
n=0
Ω = vacuum vector.
−
For h ∈ H, define a+
h , ah on F(HC )
a+
h (f1 ⊗ . . . ⊗ fn ) = h ⊗ f1 ⊗ . . . ⊗ fn ,
n
X
a−
(f
⊗
.
.
.
⊗
f
)
=
hfi , hi f1 ⊗ . . . ⊗ fˆi ⊗ . . . ⊗ fn ,
n
h 1
i=1
a−
h (f )
= hf, hi Ω
creation and annihilation operators.
Michael Anshelevich
Probability and combinatorics
Operator algebra.
+ ∗
Check: a−
h = (ah ) adjoint.
−
So Xh = a+
h + ah = self-adjoint.
Michael Anshelevich
Probability and combinatorics
Operator algebra.
+ ∗
Check: a−
h = (ah ) adjoint.
−
So Xh = a+
h + ah = self-adjoint.
a+ , a− do not commute:
+
+ −
a−
h ag − ag ah = hg, hi .
Michael Anshelevich
Probability and combinatorics
Operator algebra.
+ ∗
Check: a−
h = (ah ) adjoint.
−
So Xh = a+
h + ah = self-adjoint.
a+ , a− do not commute:
+
+ −
a−
h ag − ag ah = hg, hi .
But Xh , Xg commute.
A = Alg {Xh : h ∈ H} = commutative algebra.
Define the expectation functional on it by
E[A] = hAΩ, Ωi .
(A, E) = probability space.
Michael Anshelevich
Probability and combinatorics
Wick formula.
D
E
−
+
−
+
−
E[Xh1 Xh2 . . . Xhn ] = (a+
+
a
)(a
+
a
)
.
.
.
(a
+
a
)Ω,
Ω
h1
h1
h2
h2
hn
hn
X
Y
=
hhi , hj i .
π∈P2 (n) (i,j)∈π
Michael Anshelevich
Probability and combinatorics
Wick formula.
D
E
−
+
−
+
−
E[Xh1 Xh2 . . . Xhn ] = (a+
+
a
)(a
+
a
)
.
.
.
(a
+
a
)Ω,
Ω
h1
h1
h2
h2
hn
hn
X
Y
=
hhi , hj i .
π∈P2 (n) (i,j)∈π
Therefore:
(
khk2 , k = 2,
ck (Xh ) =
0,
otherwise,
and so Xh ∼ N (0, khk2 ).
Michael Anshelevich
Probability and combinatorics
Wick formula.
D
E
−
+
−
+
−
E[Xh1 Xh2 . . . Xhn ] = (a+
+
a
)(a
+
a
)
.
.
.
(a
+
a
)Ω,
Ω
h1
h1
h2
h2
hn
hn
X
Y
=
hhi , hj i .
π∈P2 (n) (i,j)∈π
Therefore:
(
khk2 , k = 2,
ck (Xh ) =
0,
otherwise,
and so Xh ∼ N (0, khk2 ).
If h ⊥ g, the Xh , Xg are independent.
Michael Anshelevich
Probability and combinatorics
Related documents