This action might not be possible to undo. Are you sure you want to continue?

NORMAL LAWS

WLODEK BRYC AND VIRGIL PIERCE

Abstract. This is a seminar talk of February 13, 2008.

1. Wick’s theorem for R-valued normal vectors

The following is known as Wick’s theorem.

Theorem A (Wick [9]). If (X

1

, . . . , X

2n

) is multivariate normal with mean zero,

then

E(X

1

X

2

. . . X

2n

) =

V

{j,k}∈V

E(X

j

X

k

),

where the sum is taken over all pair partitions V of ¦1, 2, . . . , 2n¦.

1

For example, E(X

1

X

2

X

3

X

4

) = E(X

1

X

2

)E(X

3

X

4

)+E(X

1

X

3

)E(X

2

X

4

)+E(X

1

X

4

)E(X

2

X

3

)

as ¦1, 2, 3, 4¦ has three pair partitions V

1

= ¦¦1, 2¦, ¦3, 4¦¦, V

1

= ¦¦1, 3¦, ¦2, 4¦¦,

V

1

= ¦¦1, 4¦, ¦3, 2¦¦. In particular, if X

1

= X

2

= X

3

= X

4

= X and E(X

2

) = 1,

then the formula gives E(X

4

) = 3.

Proof. (This is a consequence of moments-cumulants relation [4]; the connection

is best visible in the partition formulation of [7]. For another proof, see [2, page

12].) Suppose ﬁrst that X

1

. . . X

n

are of very special form. Namely, suppose they

are selected with repetition from a ﬁxed ﬁnite i.i.d. N(0, 1) family Z

1

, Z

2

, . . . .

Then E(X

1

. . . X

n

) is zero if some of the Z

**s enter the product an odd number of
**

times. If all Z’s appear an even number of times, then the answer is E(X

1

. . . X

n

) =

(2n

1

−1)!!(2n

2

−1)!! . . . (2n

k

−1), where say Z

1

is repeated 2n1 times, Z

2

is repeated

2n

2

times, etc.

This matches the answer from the pair partitions: the only contributing par-

titions are those that pair ¦1, . . . , 2n

1

¦ within itself, and ¦n

1

+ 1, . . . , n

1

+ n

2

¦

within itself, etc. The number of pairings of ¦1, . . . , 2n¦ is (2n − 1) matches

for 1 times the number of pairings of the remaining 2n − 2 elements. Thus it

is (2n −1) (2n −3) . . . 3 1.

Suppose now that we have general multivariate normal X

j

=

j

A

i,j

Z

j

. Then

C

a,b

= E(X

a

X

b

) =

A

a,j

A

b,j

.

Date: Created: November 16, 2007. Printed: February 12, 2008 File: quaaternion-wick-08.tex.

1

That is, partitions into two-element sets, so each V has the form

V = {{j

1

, k

1

}, {j

2

, k

2

}, . . . , {j

n

, k

n

}} .

1

2 WLODEK BRYC AND VIRGIL PIERCE

Now by the previous part,

E(X

1

, . . . , X

2n

) =

i

j

A

ij

Z

j

=

j1,...,j2n

A

1j1

. . . A

2n,j2n

E(Z

j1

. . . Z

j2n

)

=

j

1

,...,j

2n

A

1j

1

. . . A

2n,j

2n

V

(a,b)∈V

δ

j

a

=j

b

=

j

1

,...,j

2n

V

(a,b)∈V

δ

j

a

=j

b

A

a,j

a

A

b,j

b

=

V

(a,b)∈V

_

_

ja,j

b

δ

j

a

=j

b

A

a,j

a

A

b,j

b

_

_

=

V

(a,b)∈V

ja

A

a,j

a

A

b,j

a

=

V

(a,b)∈V

C

a,b

=

V

(a,b)∈V

E(X

a

, X

b

).

**Example 1.1. The above proof works also for X
**

1

, . . . , X

n

that are linear combina-

tions of i.i.d. N(0, 1) r.v. with complex coeﬃcients A

ij

. So with Z = X + iY ,

where X, Y are independent N(0, 1) random variables, we get E(Z

2n

) = 0 and

E([Z[

2n

) = n!.

2. Wick’s Theorem for quaternion normal variables

2.1. Quaternions. Recall that a quaternion q ∈ H can be represented as q =

x

0

+ix

1

+jx

2

+kx

3

with i

2

= j

2

= k

2

= ijk = −1 and real coeﬃcients x

0

, . . . , x

3

.

The conjugate quaternion is q = x

0

−ix

1

−jx

2

−kx

3

, so [q[

2

:= qq ≥ 0. Quaternions

with x

1

= x

2

= x

3

are usually identiﬁed with real numbers; the real part of a

quaternion is '(q) = (q + ¯ q)/2.

It is well known that quaternions can be identiﬁed with the set of certain 2 2

complex matrices:

(2.1) H ÷ x

0

+ix

1

+jx

2

+kx

3

∼

_

x

0

+ix

1

x

2

+ix

3

−x

2

+ix

3

x

0

−ix

1

_

∈ /

2×2

(C)

where on the right hand side i is the usual imaginary unit of C. Note that since

'(q) is twice the trace of the matrix representation in (2.1), this implies cyclic

property '(q

1

q

2

) = '(q

2

q

1

) which we will use freely.

2.2. Quaternion normal law. The (standard) quaternion normal random vari-

able is a H-valued random variable which can be represented as

(2.2) Z = ξ

0

+iξ

1

+jξ

2

+kξ

3

with independent real normal N(0, 1) random variables ξ

0

, ξ

1

, ξ

2

, ξ

3

. Due to sym-

metry of the centered normal laws on R, the law of (Z, Z) is the same as the law of

(Z, Z). A calculation shows that if Z is quaternion normal then for ﬁxed q

1

, q

2

∈ H,

E(Zq

1

Zq

2

) = E(Z

2

)¯ q

1

q

2

, E(Zq

1

Zq

2

) = E(Z

¯

Z)'(q

1

)q

2

.

For future reference, we insert explicitly the moments:

E(Zq

1

Zq

2

) = −2¯ q

1

q

2

, (2.3)

E(Zq

1

Zq

2

) = 2(q

1

+ ¯ q

1

)q

2

. (2.4)

SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 3

By linearity, these formulas imply

E('(Zq

1

)'(Zq

2

)) = '(q

1

q

2

), (2.5)

E('(Zq

1

)'(Zq

2

)) = '(¯ q

1

q

2

). (2.6)

2.3. Wick’s formula. In view of the Wick formula for real-valued jointly Gauss-

ian random variables, formulas (2.3) and (2.4) allow us to compute moments of

certain products of quaternion normal random variables. Suppose the n-tuple

(X

1

, X

2

, . . . , X

n

) consists of random variables taken, possibly with repetition, from

the set ¦Z

1

,

¯

Z

1

, Z

2

,

¯

Z

2

, . . . ¦ where Z

1

, Z

2

, . . . are independent quaternion normal.

Consider an auxiliary family of independent pairs ¦(Y

(r)

j

, Y

(r)

k

) : r = 1, 2, . . . ¦ which

have the same laws as (X

j

, X

k

), 1 ≤ j, k ≤ n and are independent for diﬀerent r.

Then the Wick formula for real-valued normal variables implies E(X

1

, X

2

, . . . , X

n

) =

0 for odd n, and

(2.7) E(X

1

X

2

. . . X

n

) =

f

E(Y

(f(1))

1

Y

(f(2))

2

. . . Y

(f(n))

n

),

where the sum is over all two-to-one valued functions f : ¦1, . . . , n¦ → ¦1, . . . , m¦

for n = 2m. Formulas (2.3) and (2.4) then show that the Wick reduction step takes

the following form.

(2.8) E(X

1

X

2

. . . X

n

) =

n

j=2

E(X

1

X

j

)E(U

j

X

j+1

. . . X

n

)

where

U

j

=

_

¸

_

¸

_

'(X

2

. . . X

j−1

) if X

j

=

¯

X

1

¯

X

j−1

. . .

¯

X

2

if X

j

= X

1

0 otherwise

For example, if Z is quaternion normal then applying (2.7) with the pairings

¦1, 2¦, ¦1, 3¦, ¦1, 4¦ we get

E(Z

4

) = E(X

2

)(E(Z

2

) +E(

¯

ZZ) +E(

¯

Z

2

)) = 0.

This suggests that one can do inductively the calculations, but it is not quite clear

what answers to expect. We begin with the one-variable warm-up.

Proposition 2.1. If Z is quaternion normal (2.2), then:

E([Z[

2k

) = 2

k

(k + 1)! (2.9)

E(Z

2

[Z[

2k−2

) = −2

k−1

(k + 1)! (2.10)

E(Z

2m

[Z[

2k

) = 0, m > 1. (2.11)

Proof of (2.9), (2.10), (2.11). Clearly, formula (2.9) holds true when k = 0, 1. For-

mula (2.10) holds for k = 0. Formula (2.11) holds for k = 0, m = 2 by inspection.

Suppose there is a K ≥ 0 such that the formulas already hold for ALL k, m such

that k +m ≤ K. Then the recurrence step is based on the expansions:

E(Z

2

[Z[

2K

) = E(ZZ

K+1

¯

Z

K

) (2.12)

E([Z[

2K+2

) = E(ZZ

K

¯

Z

K+1

) (2.13)

E(Z

2+2m

[Z[

2k

) = E(ZZ

k+2m+1

¯

Z

k

) (2.14)

(2.15)

4 WLODEK BRYC AND VIRGIL PIERCE

Each of the inductive steps applies (2.9), (2.10), (2.11) to the expressions with the

total exponent of at most K.

Using (2.8), we expand the right hand side of (2.12):

(2.16) E(Z

2

[Z[

2K

) = E(ZZ

K+1

¯

Z

K

) =

K+1

j=1

E(Z

2

)E(

¯

Z

K+j−1

Z

K+1−j

)

+

1

2

K

r=1

E([Z[

2

)E(Z

K+1

¯

Z

K−1

) +

1

2

K

r=1

E([Z[

2

)E(

¯

Z

2K+1−r

Z

r−1

)

Substituting u = K + 2 −j in the ﬁrst sum, we get

(2.17)

E(Z

2

[Z[

2K

) = −2

K+1

u=1

E(

¯

Z

2u−2

[Z[

2K+2−2u

)+2KE(Z

2

[Z[

2K−2

)+2

K

r=1

E(

¯

Z

2K+2−2r

[Z[

2r−2

)

= −2E([Z[

2K

) −2E(

¯

Z

2

[Z[

2K−2

) −2K2

K−1

(K + 1)! + 2E(Z

2

[Z[

2K−2

)

= −2 2

K

(K + 1)! −2K2

K−1

(K + 1)! = −2

K

(K + 1)!(K + 2)

Note that the proof used (2.11) with pairs k, m such that m > 1 and k +m ≤ K.

Using (2.8), we expand (2.13). (This just requires us to shift the index K by 1

in the previous proof.)

E([Z[

2K+2

) = E(ZZ

K

¯

Z

K+1

) =

K

j=1

E(Z

2

)E(

¯

Z

K+j

Z

K−j

)

+

1

2

K+1

r=1

E([Z[

2

)E(Z

K

¯

Z

K

) +

1

2

K+1

r=1

E([Z[

2

)E(

¯

Z

2K+1−r

Z

r−1

)

= −2E(

¯

Z

K+1

Z

K−1

) + 2(K + 1)E(Z

K

¯

Z

K

) + 2E(

¯

Z

K

Z

K

) + 2E(

¯

Z

K+1

Z

K−1

)

= −2E(Z

2

[Z[

2K−2

) + 2(K + 1)E([Z[

2K

) + 2E([Z[

2K

) + 2E(Z

2

[Z[

2K−2

)

= 2(K + 2)E([Z[

2K

) = 2(K + 2)2

K

(K + 1)! = 2

K+1

(K + 2)!

Note that the proof used (2.11) for pairs k, m such that m > 1 and k +m ≤ K.

Suppose now m > 1 is such that k +m = K. Using (??), we expand (2.14):

(2.18)

E(Z

2+2m

[Z[

2k

) = E(ZZ

k+2m+1

¯

Z

k

) =

k+2m+1

j=1

E(Z

2

)E(

¯

Z

k+j−1

Z

k+2m+1−j

)

+

1

2

k

r=1

E([Z[

2

)E(Z

k+2m+1

¯

Z

k−1

) +

1

2

k

r=1

E([Z[

2

)E(

¯

Z

2k+2m+1−r

Z

r−1

)

= −2

k+2m+1

j=1

E(

¯

Z

2j−2

[Z[

2k+2m+2−2j

)+2kE(Z

2m+2

[Z[

2k−2

)+2

k

r=1

E(

¯

Z

2k+2m+2−2r

[Z[

2r−2

)

= −2E([Z[

2k+2m

) −2E(Z

2

[Z[

2k+2m−2

) + 0

= −2 2

k+m

(k +m+ 1)! + 4 2

k+m−1

(k +m+ 1)! = 0.

SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 5

Note that again the proof uses (2.11) only with k + m ≤ K. Similarly, (2.9) and

(2.10) are applied only to exponents ≤ K.

**Formula (2.8) implies that E(X
**

1

X

2

. . . X

n

) is real, so on the left hand side of

(2.8) we can write E('(X

1

X

2

. . . X

n

)); this form of the formula is associated with

one-vertex M¨obius graphs.

Furthermore, we have a Wick reduction in multiple vertex cases

(2.19)

E('(X

1

)'(X

2

X

3

. . . X

n

)) =

n

j=2

E('(X

1

)'(X

j

))E('(X

2

. . . X

j−1

X

j+1

. . . X

n

)),

(this is just the consequence of Wick formula for the R-valued case). Furthermore,

from (2.1) it is clear that we have cyclic symmetry,

(2.20) '(X

1

X

2

. . . X

n

) = '(X

2

X

3

. . . X

n

X

1

).

3. M¨ obius graphs and quaternion normal moments

3.1. M¨obius graphs. M¨obius graphs are graphs (with loops) that allow regular

as well as ”twisted” edges. For the formal deﬁnition, see [6, Section 6]. Here we will

draw the vertices as disks, the edges as ribbons which may preserve orientation, or

reverse it (twists).

3.1.1. Euler characteristics. Denote by v(Γ), e(Γ), and f(Γ) the number of vertices,

edges, and faces of Γ. The Euler characteristic is

χ(Γ) = v(Γ) −e(Γ) +f(Γ).

Note that the faces are the cycles of the graph. To illustrate the orientation-

changing edges, we draw ribbons for edges, and disks for the vertices. The conven-

tion is that the cycle is not completed until we return to the ”same side” of the

disk that represents a vertex. For example, in Fig. 1, the Euler characteristics are

χ

1

= 1 −1 + 2 = 2 and χ

2

= 1 −1 + 1 = 1.

If Γ decomposes into components Γ

1

, Γ

2

, then χ(Γ) = χ(Γ

1

) +χ(Γ

2

).

3.2. Quaternion version of Wick’s theorem. Suppose the n-tuple (X

1

, X

2

, . . . , X

2n

)

consists of random variables taken, possibly with repetition, from the set ¦Z

1

,

¯

Z

1

, Z

2

,

¯

Z

2

, . . . ¦

where Z

1

, Z

2

, . . . are independent quaternion normal. Fix a sequence j

1

, j

2

, . . . , j

m

of natural numbers such that j

1

+ +j

m

= 2n.

Consider the family M = M

j1,...,jm

(X

1

, X

2

, . . . , X

2n

), possibly empty, of M¨obius

graphs with mvertices of degrees j

1

, j

2

, . . . , j

m

which are labeled by X

1

, X

2

, . . . , X

2n

,

whose regular edges correspond to pairs X

i

=

¯

X

j

and ﬂipped edges correspond to

pairs X

i

= X

j

. No edges of Γ ∈ M can join random variables X

i

, X

j

that are

independent.

Theorem 3.1.

(3.1) E

_

'(X

1

X

2

. . . X

j

1

)'(X

j

1

+1

. . . X

j

1

+j

2

) . . .

'(X

j1+j2+jm−1+1

. . . X

2n

)

_

= 4

n−m

Γ∈M

(−2)

χ(Γ)

.

(The right hand side is interpreted as 0 when M = ∅.)

6 WLODEK BRYC AND VIRGIL PIERCE

Question 3.2. What would be the formula for E(q

1

X

1

q

2

X

2

. . . q

2n

X

2n

)? Or for

more general Q-Gaussian n-touples? (For more general deﬁnition, see [8])

Proof. In view of (2.7) and (2.8), it suﬃces to show that if X

1

, . . . , X

2n

consists of

n independent pairs, and each pair is either of the form (X, X) or (X,

¯

X), then

(3.2) E

_

'(X

1

X

2

. . . X

j

1

)'(X

j

1

+1

. . . X

j

1

+j

2

) . . .

'(X

j

1

+j

2

+j

m−1

+1

. . . X

2n

)

_

= 4

n−m

(−2)

χ(Γ)

,

where Γ is the M¨obius graph that describes all pairings of the sequence.

First we check the two M¨obius graphs for n = 1, m = 1:

E('(X

¯

X)) = (−2)

2

, and E('(XX)) = (−2)

1

.

One checks that these correspond to the M¨obius graphs in Figure 1, which gives a

sphere (χ = 2) and projective sphere (χ = 1) respectively.

Figure 1.

For n = 2, m = 1 we have the following twelve cases.

(1)

E('(XY

¯

Y

¯

X)) = E(X

¯

X)E(Y

¯

Y ) = (−2)

2

4

1

which matches the graph in Figure 2. This graph gives a sphere.

Figure 2. χ(Γ) = 2

(2)

E('(XY

¯

Y X)) = E(XX)E(Y

¯

Y ) = (−2)

1

4

1

which matches the graph in Figure 3. This graph gives a projective sphere.

SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 7

Figure 3. χ(Γ) = 1

(3)

E('(XY Y

¯

X)) = E(X

¯

X)E('(Y Y )) = (−2)

1

4

1

which matches the graph in Figure 4. This graph gives a projective sphere.

Figure 4. χ(Γ) = 1

(4)

E('(XY Y X)) = E(XX)E(

¯

Y

¯

Y ) = 4

1

which matches the graph in Figure 5. This graph gives a Klein bottle.

Figure 5. χ(Γ) = 0

(5)

E('(XY

¯

X

¯

Y )) = E(X

¯

X)E('(Y )

¯

Y ) = 4

1

which matches the graph in Figure 6. This graph gives a torus.

Figure 6. χ(Γ) = 0

8 WLODEK BRYC AND VIRGIL PIERCE

(6)

E('(XY X

¯

Y )) = E(XX)E(

¯

Y

¯

Y ) = 4

1

which matches the graph in Figure 7. This graph gives a Klein bottle.

Figure 7. χ(Γ) = 0

(7)

E('(XY

¯

XY )) = E(X

¯

X)E('Y Y ) = 4

1

which matches the graph in Figure 8. This graph gives a Klein bottle.

Figure 8. χ(Γ) = 0

(8)

E('(XY XY )) = E(XX)E(

¯

Y Y ) = (−2)

1

4

1

which matches the graph in Figure 9. This graph gives a projective sphere.

Figure 9. χ(Γ) = 1

(9)

E('(X

¯

XY

¯

Y )) = E(X

¯

X)E(Y

¯

Y ) = (−2)

2

4

1

which matches the graph in Figure 10. This graph gives a sphere.

(10)

E('(XXY

¯

Y )) = E(XX)E(Y

¯

Y ) = (−2)

1

4

1

which matches the graph in Figure 11. This graph gives a projective sphere.

SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 9

Figure 10. χ(Γ) = 2

Figure 11. χ(Γ) = 1

(11)

E('(X

¯

XY Y )) = E(X

¯

X)E(Y Y ) = (−2)

1

4

1

which matches the graph in Figure 12. This graph gives a projective sphere.

Figure 12. χ(Γ) = 1

(12)

E('(XXY Y )) = E(XX)E(Y Y ) = 4

1

,

which matches the graph in Figure 13. This graph gives a Klein bottle.

Figure 13. χ(Γ) = 0

We leave the cases of n = 2, m = 2, to the reader, they proceed as above using

only the relations (2.3-2.6).

(1) E('(XX)'(Y Y ) = 4, χ = 2.

10 WLODEK BRYC AND VIRGIL PIERCE

(2) E('(XX)'(Y

¯

Y ) = −8, χ = 3.

(3) E('(X

¯

X)'(Y

¯

Y ) = 16, χ = 4.

(4) E('(XY )'(X

¯

Y ) = −2, χ = 1.

(5) E('(XY )'(XY ) = 4, χ = 2.

(6) E('(X)'(

¯

XY

¯

Y ) = E('(X)'(XY

¯

Y ) = E('(X)'(Y

¯

X

¯

Y ) = E('(X)'(Y X

¯

Y ) =

4, χ = 2.

(7) E('(X)'(XY Y ) = E('(X)'(Y XY ) = E('(X)'(Y

¯

XY ) = −2, χ = 1

We now proceed with the induction step. One notes that by independence of

the pairs at diﬀerent edges, the left hand side of (3.2) factors into the product

corresponding to connected components of Γ. It is therefore enough to consider

connected Γ.

If Γ has two vertices that are joined by an edge, we can use cyclicity of ' to

move the variables that label the edge to the ﬁrst positions in their cycles, say X

1

and X

j1

and use (2.5) or (2.6) to eliminate this pair from the product. The use

of relation (2.5) is just that of gluing the two vertices together removing the edge

x which is labeled by the two appearances of Z. Relation (2.6) glues together the

two vertices, removing the edge x, and the reversal of orientation across the edge is

given by the conjugate (see Figure 14). These geometric operations reduce n and

m by one without changing the Euler characteristic: the number of edges and the

number of vertices are reduced by 1; the faces are preserved – in the case of edge

ﬂip in Fig. 14, the edges of the face from which we remove the edge, after reduction

follow the same order.

Therefore we will only need to prove the result for the single vertex case of the

induction step.

Figure 14.

We wish to show that

(3.3) E(X

1

X

2

X

3

X

4

. . . X

2n

) = (−2)

χ(Γ)

4

n−1

,

where Γ is a one vertex M¨obius graph with arrows (half edges) labeled by X

k

. The

trivial case is when X

1

=

¯

X

2

in which case we ﬁnd that (3.3) is

E(X

1

¯

X

1

)E(X

3

X

4

. . . X

2n

) = 4E(X

3

X

4

. . . X

2n

) (3.4)

= 4

_

(−2)

χ2

4

(n−1)−1

_

= (−2)

χ2

4

n−1

.

SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 11

Equation (3.4) corresponds to the reduction of the M¨obius graph given in Figure

15 and represents a net of 0 to the Euler characteristic, χ

2

= χ

1

+ 2, from which

we see that the result follows in this case.

Figure 15. χ

2

= v

2

−e

2

+f

2

= v

1

−(e

1

−1) + (f

1

−1) = χ

1

The second nearly trivial case is X

2

= X

1

in which case we ﬁnd that (3.3) is

E(X

1

X

1

)E(X

3

X

4

. . . X

2n

) = (−2) [E(X

3

X

4

. . . X

2n

)] (3.5)

= (−2)

_

(−2)

χ

2

4

(n−1)−1

_

(3.6)

= (−2)

χ2−1

4

n−1

. (3.7)

Equation (3.5) corresponds to the reduction of the M¨obius graph given in Figure

16 and represents a net change of 1 to the Euler characteristic, from which we see

that the result follows in this case.

In what follows we take j < k, and let q

1

= X

3

. . . X

j−1

, q

2

= X

j+1

. . . X

k−1

,

and q

3

= X

k+1

. . . X

2n

. Conjugation of q

i

in the following computations arises from

an orientation reversal. In the diagrams we have used an arrow to indicate an edge

which reverses the local orientation of the vertex, in other words a ribbon with

a ﬂip to it. Geometrically conjugation reverses the order of the edges as well as

switching local orientations. There are 8 non-trivial cases, corresponding precisely

to the n = 2, m = 1 cases 1-8:

12 WLODEK BRYC AND VIRGIL PIERCE

Figure 16.

(1) When X

1

=

¯

X

k

and X

2

=

¯

X

j

we ﬁnd that

E(X

1

X

2

q

1

¯

X

2

q

2

¯

X

1

q

3

)

= E(X

1

¯

X

1

)E('(X

2

q

1

¯

X

2

q

2

)'(q

3

))

= 4E(X

2

¯

X

2

)E('(q

1

)'(q

2

)'(q

3

))

= 4

2

E('(q

1

)'(q

2

)'(q

3

))

= 4

2

_

(−2)

χ

2

4

(n−2)−3

_

= (−2)

χ2−4

4

n−1

.

Figure 17.

SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 13

This corresponds to the reduction of the M¨obius graph given in Figure

17. We note that χ

2

= v

2

− e

2

+ f

2

= v

1

+ 2 − (e

1

− 2) + f

1

= χ

1

+ 4, as

the number of faces is not changed when q

1

, q

2

, q

3

, = 1.

(2) When X

1

= X

k

and X

2

=

¯

X

j

we ﬁnd that

'E(X

1

X

2

q

1

¯

X

2

q

2

X

1

q

3

)

= E(X

1

X

1

)'E(¯ q

2

X

2

¯ q

1

¯

X

2

q

3

)

= (−2)'E(X

2

¯ q

1

¯

X

2

q

3

¯ q

2

)

= (−2)E(X

2

¯

X

2

)E('(¯ q

1

)'(q

3

¯ q

2

))

= (−2)

3

E('(¯ q

1

)'(q

3

¯ q

2

))

= (−2)

3

_

(−2)

χ

2

4

(n−2)−2

_

= (−2)

χ2−3

4

n−1

.

Figure 18.

This corresponds to the reduction of the M¨obius graph given in Figure

18 and represents a net of 3 to the Euler characteristic.

(3) When X

1

=

¯

X

k

and X

2

= X

j

we ﬁnd that

E(X

1

X

2

q

1

X

2

q

2

¯

X

1

q

3

)

= E(X

1

¯

X

1

)E('(X

2

q

1

X

2

q

2

)'(q

3

))

= 4E(X

2

X

2

)E('(¯ q

1

q

2

)'(q

3

))

= (−2)

3

[E('(¯ q

1

q

2

)'(q

3

))]

= (−2)

3

_

(−2)

χ

2

4

(n−2)−2

_

= (−2)

χ2−3

4

n−1

.

This corresponds to the reduction for the M¨obius graph given in Figure

19, represents a net of 3 to the Euler characteristic, and the result follows.

14 WLODEK BRYC AND VIRGIL PIERCE

Figure 19.

(4) When X

1

= X

k

and X

2

= X

j

we ﬁnd that

E(X

1

X

2

q

1

X

2

q

2

X

1

q

3

)

= E(X

1

X

1

)E(¯ q

2

¯

X

2

¯ q

1

¯

X

2

q

3

)

= (−2)E(

¯

X

2

¯ q

1

¯

X

2

q

3

¯ q

2

)

= (−2)E(

¯

X

2

¯

X

2

)E(q

1

q

3

¯ q

2

)

= (−2)

2

E(q

1

q

3

¯ q

2

)

= (−2)

2

_

(−2)

χ

2

4

(n−2)−1

_

= (−2)

χ2−2

4

n−1

.

Figure 20.

SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 15

This corresponds to the reduction for the M¨obius graph given in Figure

20, represents a net of 2 to the Euler characteristic (χ

2

= χ

1

+ 2), and the

result follows.

(5) When X

1

=

¯

X

j

and X

2

=

¯

X

k

we ﬁnd that

'E(X

1

X

2

q

1

¯

X

1

q

2

¯

X

2

q

3

)

= E(X

1

¯

X

1

)E('(X

2

q

1

)'(q

2

¯

X

2

q

3

))

= 4E('(X

2

q

1

)'(

¯

X

2

q

3

q

2

))

= 4E('(q

1

q

3

q

2

))

= 4

_

(−2)

χ

2

4

(n−2)−1

_

= (−2)

χ2−2

4

n−1

.

Figure 21.

This corresponds to the reduction for the M¨obius graph given in Figure

21, represents a net of 2 to the Euler characteristic, and the result follows.

(6) When X

1

= X

j

and X

2

=

¯

X

k

we ﬁnd that

E(X

1

X

2

q

1

X

1

q

2

¯

X

2

q

3

)

= E(X

1

X

1

)E(¯ q

1

¯

X

2

q

2

¯

X

2

q

3

)

= (−2)E(

¯

X

2

q

2

¯

X

2

q

3

¯ q

1

)

= (−2)E(

¯

X

2

¯

X

2

)E(¯ q

2

q

3

¯ q

1

)

= (−2)

2

E(¯ q

2

q

3

¯ q

1

)

= (−2)

−2

_

(−2)

χ2

4

(n−2)−1

_

= (−2)

χ

2

−2

4

n−1

.

This corresponds to the reduction for the M¨obius graph given in Figure

22, represents a net of 2 to the Euler characteristic, and the result follows.

16 WLODEK BRYC AND VIRGIL PIERCE

Figure 22.

(7) When X

1

=

¯

X

j

and X

2

= X

k

we ﬁnd that

E(X

1

X

2

q

1

¯

X

1

q

2

X

2

q

3

)

= E(X

1

¯

X

1

)E('(X

2

q

1

)'(q

2

X

2

q

3

))

= 4E('(X

2

q

1

)'(X

2

q

3

q

2

))

= 4E(¯ q

1

q

3

q

2

)

= 4

_

(−2)

χ

2

4

(n−2)−1

_

= (−2)

χ

2

−2

4

n−1

.

This corresponds to the reduction for the M¨obius graph given in Figure

23, represents a net of 2 to the Euler characteristic, and the result follows.

(8) When X

1

= X

j

and X

2

= X

k

we ﬁnd that

E(X

1

X

2

q

1

X

1

q

2

X

2

q

3

)

= E(X

1

X

1

)E(¯ q

1

¯

X

2

q

2

X

2

q

3

)

= (−2)E(X

2

q

3

¯ q

1

¯

X

2

q

2

)

= (−2)E(X

2

¯

X

2

)E('(q

3

¯ q

1

)'(q

2

))

= (−2)

3

E('(q

3

¯ q

1

)'(q

2

))

= (−2)

3

_

(−2)

χ2

4

(n−2)−2

_

= (−2)

χ2−3

4

n−1

.

This corresponds to the reduction for the M¨obius graph given in Figure

24, represents a net of 3 to the Euler characteristic, and the result follows.

SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 17

Figure 23.

Figure 24.

Note that taking away an oriented ribbon creates a new vertex. The remaining

graph might be still connected, or it may split into two components. If taking

away a loop makes the graph disconnected, then the counts of changes to edges and

vertices are still the same. But the faces need to be counted as follows: The inner

face of the removed edge becomes the outside face of one component, and the outer

face at the removed edge becomes the outer face of the other component. Thus the

counting of faces is not aﬀected by whether the graph is connected – no change.

With each of these cases checked, by the induction hypothesis, the proof is

completed.

**18 WLODEK BRYC AND VIRGIL PIERCE
**

4. Duality between GOE and GSE ensembles

The duality between β = 1 and β = 4 asserts that the formulas for moments

of traces of polynomials in H-valued Gaussian ensembles can be obtained from the

corresponding formulas for the R-valued Gaussian ensembles by replacing the di-

mension parameter N by (−2N). The duality for one-matrix Wishart ensembles

was discovered in [1, Corollary 4.2]. The duality for one-matrix GOE/GSE ensem-

bles appears in [5]. A derivation based on recurrence formulas for moments is in

[3].

4.1. Quaternion matrices. By /

M×N

(H

β

) we denote the set of all M N

matrices with entries from H

β

. For A ∈ /

M×N

(H

β

), the conjugate matrix is

A

∗

i,j

:= A

j,i

. The trace is tr(A) =

N

j=1

A

jj

.

When β = 4, traces tr(A) may fail to commute, so in the formulas we will use

'(tr(A)). In fact, the natural deﬁnition of trace would be tr

β

() := β'(tr()). Such

a nonstandard deﬁnition is consistent with the treatment of A ∈ /

N×N

(H

β

) as

an element of /

βN×βN

(R) via transformation (??). Non-standard deﬁnition of

quaternionic trace is used in [1]

4.1.1. GSE Ensembles. We will be interested GUE

β

ensembles. These are square

self-adjoint matrices

(4.1) Z = [Z

i,j

]

where ¦Z

i,j

: i < j¦ is a family of independent H

β

-normal random variables, ¦Z

i,i

¦

is a family of independent real normal of variance 2, and Z

i,j

=

¯

Z

j,i

. The law of

such a matrix has a density C exp(−

1

2

tr(x

2

)), which is supported on the self-adjoint

subset of /

N×N

(H

β

). (C = C(N, β) is a normalizing constant.)

4.2. A multi-matrix version of Mulase and Waldron. In this section we will

demonstrate a multi-matrix version of the theorem of Mulase and Waldron [5]

from a heuristic argument using our Wick formula and quadratic relations. This

represents an improvement to the argument of Mulase and Waldron as we will not

need to rely on labelings of the vertices by the quaternions 1, i, j, and k. This part

of their argument is encoded in relations (2.5) and (2.6).

We will compute here the expected values of the real traces of powers of a

quaternionic self-dual matrix in the Gaussian symplectic ensemble (4.1) where the

oﬀ-diagonal matrix entries are (2.2).

The basic theorem we will prove is

Theorem 4.1.

1

(4N)

n−m

E('(tr(Z

j

1

))'(tr(Z

j

2

)) . . . '(tr(Z

j

m

))) =

Γ

(−2N)

χ(Γ)

,

where the sum is over labeled M¨obius graphs Γ with m vertices of degree j

1

, j

2

, . . . ,

j

m

, χ(Γ) is the Euler characteristic and j

1

+j

2

+ +j

m

= 2n.

More generally, suppose Z

1

, . . . , Z

s

are independent N N GSE ensembles and

t : ¦1, . . . , 2n¦ → ¦1, . . . , s¦ is ﬁxed. Let α

0

= 1, α

k

= j

1

+ j

2

+ + j

k−1

+ 1,

SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 19

β

k

= j

1

+j

2

+ +j

k

denote the ranges under traces. Then

(4.2)

1

(4N)

n−m

E

_

'(tr(Z

t(1)

. . . Z

t(β

1

)

))'(tr(Z

t(α

2

)

. . . Z

t(β

2

)

)) . . . '(tr(Z

t(α

m

)

. . . Z

t(β

m

)

))

_

=

Γ

(−2N)

χ(Γ)

,

where the sum is over labeled color-preserving M¨obius graphs Γ with vertices of

degree j

1

, j

2

, . . . , j

m

that are colored with s colors by the mapping t. As previously,

χ(Γ) is the Euler characteristic and j

1

+j

2

+ +j

m

= 2n. (If there are no Γ that

are consistent with the coloring t, we interpret the sum as 0.)

Perhaps instead of

the example, we

should clarify how

come the Theorem

IS the ”duality”?

Example 4.2. To illustrate the multi-matrix aspect of the theorem, suppose Z

1

and

Z

2

are independent N GSE and X

1

, X

2

be independent N GOE ensembles.

Fix r, m, k and let f(N) := E(tr(X

2m

1

X

2k

2

)

r

). Since f(N) = N

(m+k−1)r

Γ

N

χ(Γ)

,

[reference?] therefore Theorem 4.1 implies that the moments for the independent

GSE ensembles are determined from the corresponding moments of independent

GOE ensembles by the “dual formula”

E(tr(Z

2m

1

Z

2k

2

)

r

) = (4)

(m+k−1)r

f(−2N).

Proof. We begin by expanding out the traces in terms of the matrix entries

N

m

4

n−m

E('(tr(Z

j1

))'(tr(Z

j2

)) . . . '(tr(Z

jm

))) =

(4.3)

=

1 ≤ a

1

, a

2

, . . . , a

j

1

≤ N

1 ≤ b

1

, b

2

, . . . , b

j2

≤ N

.

.

.

1 ≤ c

1

, c

2

, . . . , c

j

m

≤ N

1

(4n)

n−m

E('(Z

a

1

,a

2

Z

a

2

,a

3

. . . Z

a

j

1

,a

1

)'(Z

b

1

,b

2

. . . Z

b

j

2

,b

1

) . . .

. . . '(Z

c1,c2

. . . Z

cj

m

,c1

)) .

Note that essentially the same expansion applies to (4.2), except that the consecu-

tive entries Z

(t)

i,j

must now be labeled also by the ”color” t.

Colors make for cumbersome notation! It is then better to index the products by

the cycles of a permutation, so that the formula is

a:[1...2n]→[1...N]

c∈σ

E('(

j∈c

Z

t(j)

a(j),a(σ(j))

)).

We then need a special convention to follow the right order for products over the

cycles. Luckily, ' of quaternions has cyclic property so such products are well

defnied.

From (2.8) it follows that the right hand side of (4.3) can be expanded as sum

over all pairings, and we can assume that the pairs are independent.

Of course, pairings that match two independent random variables do not con-

tribute to the sum. The pairings that contribute to the sum are of three diﬀerent

types: pairs that match Z with another Z at a diﬀerent position in the product,

pairs that match Z with

¯

Z, and pairings that match the diagonal (real) entries.

20 WLODEK BRYC AND VIRGIL PIERCE

We ﬁrst dispense with the pairings that match a diagonal entry Z

i,i

= ξ with

another diagonal entry Z

j,j

= ξ. Since real numbers commute with quaternions,

'(E(q

0

ξq

1

ξq

2

)) = E(ξ

2

)'(q

0

q

1

q

2

) = 2'(q

0

q

1

q

2

).

On the other hand, using the cyclic property of '() and adding formulas (2.3) and

(2.4), we see that the same answer arises from

E'(q

0

Zq

1

Zq

2

) +E'(q

0

Zq

1

¯

Zq

2

) = E'(Zq

1

Zq

2

q

0

) +E'(Zq

1

¯

Zq

2

q

0

) = 2'(q

1

q

2

q

0

).

So the contribution of each such diagonal pairing is the same as that of two matches

of quaternionic entries (Z, Z) and (Z,

¯

Z).

Thus, once we replace all the real entries that came from the diagonal entries

by the corresponding quaternion-normal pairs, we get the sum over all possible

pairs of matches of the ﬁrst two types only. We label all such pairings by M¨obius

graphs, with the interpretation that pairings or random variables Z, Z correspond

to twisted ribbons, while pairings Z,

¯

Z correspond to ribbons without a twist. The

pairs of variables at diﬀerent ribbons can now be assumed independent. In the

multi-matrix case, the edges of the graph at each vertex are colored according to

function t, which restricts the number of available pairings.

We now relabel the Z

i,j

by Z

a

k

,a

k+1

= X

k

, Z

b

k

,b

k+1

= X

j

1

+k

, . . . , M

c

k

,c

k+1

=

X

j

1

+j

2

+...j

m−1

+k

. Our claim is then that given a M¨obius graph Γ with vertices of

degree j

1

, j

2

, . . . , and j

m

with edges labeled by X

k

, satisﬁes

(4.4)

1

(4N)

n−m

E('(X

1

. . . X

j1

)'(X

j1+1

. . . X

j1+j2

) . . . '(X

j1+···+jm−1+1

. . . X

2n

))

= (−2N)

χ(Γ)

N

−f(Γ)

,

where f(Γ) is the number of faces of Γ. The N

−f(Γ)

terms are removed by the

summations in (4.3), the relations given by the edges of Γ reduce the number of

summations leaving us with f(Γ) sums from 1 to N. Collecting the powers of N

we ﬁnd that this is the same as (3.2).

References

[1] Hanlon, P. J., Stanley, R. P., and Stembridge, J. R. Some combinatorial aspects of the

spectra of normally distributed random matrices. In Hypergeometric functions on domains of

positivity, Jack polynomials, and applications (Tampa, FL, 1991), vol. 138 of Contemp. Math.

Amer. Math. Soc., Providence, RI, 1992, pp. 151–174.

[2] Janson, S. Gaussian Hilbert spaces, vol. 129 of Cambridge Tracts in Mathematics. Cambridge

University Press, Cambridge, 1997.

[3] Ledoux, M. A recursion formula for the moments of the gaussian orthogonal ensemble.

preprint: http://www.math.univ-toulouse.fr/ ledoux/GOE.pdf, 2007.

[4] Leonov, V. P., and Shiryaev, A. N. On a method of semi-invariants. Theor. Probability

Appl. 4 (1959), 319–329.

[5] Mulase, Motohico; Waldron, A. Duality of orthogonal and symplectic matrix inte-

grals and quaternionic Feynman graphs. Commun.Math.Phys. () 240 (2003), 553–586.

http://arxiv.org/abs/math-ph/0206011.

[6] Pierce, V. U. An algorithm for map enumeration, 2006.

[7] Speed, T. P. Cumulants and partition lattices. Austral. J. Statist. 25, 2 (1983), 378–388.

[8] Vakhania, N. N. Random vectors with values in quaternion hilbert space. Theory of Proba-

bility and its Applications 43 (1999), 99–115.

[9] Wick, G. C. The evaluation of the collision matrix. Phys. Rev. 80, 2 (Oct 1950), 268–272.

2

WLODEK BRYC AND VIRGIL PIERCE

**Now by the previous part, E(X1 , . . . , X2n ) =
**

i j

Aij Zj =

j1 ,...,j2n

**A1j1 . . . A2n,j2n E(Zj1 . . . Zj2n ) δja =jb = δja =jb Aa,ja Ab,jb
**

j1 ,...,j2n V (a,b)∈V

=

j1 ,...,j2n

**A1j1 . . . A2n,j2n
**

V (a,b)∈V ja ,jb V (a,b)∈V

=

δja =jb Aa,ja Ab,jb =

V (a,b)∈V ja

Aa,ja Ab,ja Ca,b = E(Xa , Xb ).

V (a,b)∈V

=

V (a,b)∈V

Example 1.1. The above proof works also for X1 , . . . , Xn that are linear combinations of i.i.d. N (0, 1) r.v. with complex coeﬃcients Aij . So with Z = X + iY , where X, Y are independent N (0, 1) random variables, we get E(Z 2n ) = 0 and E(|Z|2n ) = n!. 2. Wick’s Theorem for quaternion normal variables 2.1. Quaternions. Recall that a quaternion q ∈ H can be represented as q = x0 + ix1 + jx2 + kx3 with i2 = j 2 = k 2 = ijk = −1 and real coeﬃcients x0 , . . . , x3 . The conjugate quaternion is q = x0 −ix1 −jx2 −kx3 , so |q|2 := qq ≥ 0. Quaternions with x1 = x2 = x3 are usually identiﬁed with real numbers; the real part of a quaternion is (q) = (q + q )/2. ¯ It is well known that quaternions can be identiﬁed with the set of certain 2 × 2 complex matrices: (2.1) H x0 + ix1 + jx2 + kx3 ∼ x0 + ix1 −x2 + ix3 x2 + ix3 ∈ M2×2 (C) x0 − ix1

where on the right hand side i is the usual imaginary unit of C. Note that since (q) is twice the trace of the matrix representation in (2.1), this implies cyclic property (q1 q2 ) = (q2 q1 ) which we will use freely. 2.2. Quaternion normal law. The (standard) quaternion normal random variable is a H-valued random variable which can be represented as (2.2) Z = ξ0 + iξ1 + jξ2 + kξ3

with independent real normal N (0, 1) random variables ξ0 , ξ1 , ξ2 , ξ3 . Due to symmetry of the centered normal laws on R, the law of (Z, Z) is the same as the law of (Z, Z). A calculation shows that if Z is quaternion normal then for ﬁxed q1 , q2 ∈ H, ¯ E(Zq1 Zq2 ) = E(Z 2 )¯1 q2 , E(Zq1 Zq2 ) = E(Z Z) (q1 )q2 . q For future reference, we insert explicitly the moments: (2.3) (2.4) E(Zq1 Zq2 ) E(Zq1 Zq2 ) = −2¯1 q2 , q = 2(q1 + q1 )q2 . ¯

.14) (2. . This suggests that one can do inductively the calculations.11) holds for k = 0.8) where E(X1 X2 . .13) (2. Z2 . (2. X2 0 ¯ if Xj = X1 if Xj = X1 otherwise For example. . {1. . 4} we get ¯ ¯ E(Z 4 ) = E(X 2 )(E(Z 2 ) + E(ZZ) + E(Z 2 )) = 0. . . Suppose the n-tuple (X1 . 2}. In view of the Wick formula for real-valued jointly Gaussian random variables. .3) and (2. m = 2 by inspection.9) (2. Xn ) consists of random variables taken. 3}. where the sum is over all two-to-one valued functions f : {1. . Wick’s formula.15) E(|Z|2K+2 ) = E(Z 2+2m |Z|2k ) = ¯ E(ZZ K Z K+1 ) k+2m+1 ¯ k E(ZZ Z ) . Xn ) = j=2 E(X1 Xj )E(Uj Xj+1 . X2 . . possibly with repetition. . . Proposition 2. .12) E(Z 2 |Z|2K ) = E(ZZ K+1 Z K ) (2. . . n} → {1. .4) then show that the Wick reduction step takes the following form. Formulas (2. We begin with the one-variable warm-up. Then the Wick formula for real-valued normal variables implies E(X1 . Formula (2. . . . {1. . .1. } which have the same laws as (Xj . q 2. If Z is quaternion normal (2. Xj−1 ) ¯ ¯ Uj = Xj−1 . k ≤ n and are independent for diﬀerent r.3) and (2. Clearly. 2. Xn ) = 0 for odd n. . . from ¯ ¯ the set {Z1 . . are independent quaternion normal. Proof of (2. (r) (r) Consider an auxiliary family of independent pairs {(Yj . these formulas imply (2. then: (2. . but it is not quite clear what answers to expect. Suppose there is a K ≥ 0 such that the formulas already hold for ALL k.7) with the pairings {1. (¯1 q2 ). 1.9). . (2. Yn (n)) ). 1 ≤ j. . if Z is quaternion normal then applying (2. m such that k + m ≤ K. .10) (2.11) E(|Z|2k ) E(Z 2 |Z|2k−2 ) E(Z 2m |Z|2k ) = 2k (k + 1)! = −2k−1 (k + 1)! = 0. .5) (2. and (2.11). Formula (2. . Xn ) (X2 . X2 .9) holds true when k = 0. Yk ) : r = 1. . Z2 . formulas (2. . Xk ). n (2. . .10) holds for k = 0. formula (2. m > 1. . m} for n = 2m. Z2 . Z1 .10). Xn ) = f E(Y1 (f (1)) Y2 (f (2)) (f . } where Z1 .6) E( (Zq1 ) (Zq2 )) = E( (Zq1 ) (Zq2 )) = (q1 q2 ). .4) allow us to compute moments of certain products of quaternion normal random variables.7) E(X1 X2 .3. Then the recurrence step is based on the expansions: ¯ (2.2).SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 3 By linearity. .

m such that m > 1 and k + m ≤ K. Suppose now m > 1 is such that k + m = K. Using (2.8).8). we expand the right hand side of (2. (This just requires us to shift the index K by 1 in the previous proof. (2.13).10). Using (??).12): K+1 (2. m such that m > 1 and k + m ≤ K.11) to the expressions with the total exponent of at most K. we get (2. we expand (2. we expand (2.) K ¯ E(|Z|2K+2 ) = E(ZZ K Z K+1 ) = j=1 ¯ E(Z 2 )E(Z K+j Z K−j ) K+1 + 1 2 K+1 r=1 1 ¯ E(|Z|2 )E(Z K Z K ) + 2 ¯ E(|Z|2 )E(Z 2K+1−r Z r−1 ) r=1 ¯ ¯ ¯ ¯ = −2E(Z K+1 Z K−1 ) + 2(K + 1)E(Z K Z K ) + 2E(Z K Z K ) + 2E(Z K+1 Z K−1 ) = −2E(Z 2 |Z|2K−2 ) + 2(K + 1)E(|Z|2K ) + 2E(|Z|2K ) + 2E(Z 2 |Z|2K−2 ) = 2(K + 2)E(|Z|2K ) = 2(K + 2)2K (K + 1)! = 2K+1 (K + 2)! Note that the proof used (2.9).14): (2. (2.11) for pairs k.11) with pairs k. Using (2.18) k+2m+1 ¯ E(Z 2+2m |Z|2k ) = E(ZZ k+2m+1 Z k ) = j=1 ¯ E(Z 2 )E(Z k+j−1 Z k+2m+1−j ) 1 2 k + = −2 1 2 k ¯ E(|Z|2 )E(Z k+2m+1 Z k−1 ) + r=1 ¯ E(|Z|2 )E(Z 2k+2m+1−r Z r−1 ) r=1 k k+2m+1 ¯ E(Z 2j−2 |Z|2k+2m+2−2j )+2kE(Z 2m+2 |Z|2k−2 )+2 j=1 r=1 ¯ E(Z 2k+2m+2−2r |Z|2r−2 ) = −2E(|Z| 2k+2m k+m ) − 2E(Z |Z| 2 2k+2m−2 )+0 (k + m + 1)! = 0.17) K+1 K E(Z 2 |Z|2K ) = −2 u=1 ¯ E(Z 2u−2 |Z|2K+2−2u )+2KE(Z 2 |Z|2K−2 )+2 r=1 ¯ E(Z 2K+2−2r |Z|2r−2 ) ¯ = −2E(|Z|2K ) − 2E(Z 2 |Z|2K−2 ) − 2K2K−1 (K + 1)! + 2E(Z 2 |Z|2K−2 ) = −2 × 2K (K + 1)! − 2K2K−1 (K + 1)! = −2K (K + 1)!(K + 2) Note that the proof used (2.16) E(Z |Z| 2 2K ) = E(ZZ K K+1 ¯ ZK ) = j=1 ¯ E(Z 2 )E(Z K+j−1 Z K+1−j ) 1 2 K + 1 2 ¯ E(|Z|2 )E(Z K+1 Z K−1 ) + r=1 ¯ E(|Z|2 )E(Z 2K+1−r Z r−1 ) r=1 Substituting u = K + 2 − j in the ﬁrst sum.4 WLODEK BRYC AND VIRGIL PIERCE Each of the inductive steps applies (2. = −2 × 2 (k + m + 1)! + 4 × 2 k+m−1 .

Xn X1 ). this form of the formula is associated with one-vertex M¨bius graphs. j2 . (2. . M¨bius graphs are graphs (with loops) that allow regular o as well as ”twisted” edges. . × (Xj1 +j2 +jm−1 +1 . Z2 . (3. Furthermore. Section 6]. Z1 . . . e(Γ).8) implies that E(X1 X2 . X2n ).) .9) and (2. Denote by v(Γ). M¨bius graphs.1) E (X1 X2 . . . jm of natural numbers such that j1 + · · · + jm = 2n.SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 5 Note that again the proof uses (2. . X2 . . X2n ) = 4n−m Γ∈M (−2)χ(Γ) . and f (Γ) the number of vertices. Xn )). . X2 . . . Formula (2. . . . . . Consider the family M = Mj1 . For example. Xn )) = j=2 E( (X1 ) (Xj ))E( (X2 . .20) (X1 X2 . Xn ) = (X2 X3 . . . . j2 . edges.11) only with k + m ≤ K. . . we draw ribbons for edges. For the formal deﬁnition. possibly with repetition. (2. and faces of Γ.8) we can write E( (X1 X2 .1) it is clear that we have cyclic symmetry. .. Z2 . Quaternion version of Wick’s theorem. possibly empty.. Xn )). If Γ decomposes into components Γ1 . so on the left hand side of (2.1.1. Xj that are independent. . . . Similarly. . .1. and disks for the vertices. . Here we will draw the vertices as disks. . from the set {Z1 . see [6. Euler characteristics. . . o Furthermore.1.2. X2 . Mobius graphs and quaternion normal moments o 3. or reverse it (twists). Note that the faces are the cycles of the graph. No edges of Γ ∈ M can join random variables Xi . . 3.10) are applied only to exponents ≤ K. (this is just the consequence of Wick formula for the R-valued case). } where Z1 .. . 3. . The Euler characteristic is χ(Γ) = v(Γ) − e(Γ) + f (Γ). Theorem 3. X2n ) ¯ ¯ consists of random variables taken. (The right hand side is interpreted as 0 when M = ∅. . Xn ) is real. Fix a sequence j1 . . Xj1 +j2 ) × . then χ(Γ) = χ(Γ1 ) + χ(Γ2 ). .jm (X1 . Xj−1 Xj+1 . To illustrate the orientationchanging edges. . . we have a Wick reduction in multiple vertex cases (2. are independent quaternion normal. . X2n . . Xj1 ) (Xj1 +1 . jm which are labeled by X1 . .19) n E( (X1 ) (X2 X3 . Suppose the n-tuple (X1 . . from (2. Z2 . the edges as ribbons which may preserve orientation. of M¨bius o graphs with m vertices of degrees j1 . the Euler characteristics are χ1 = 1 − 1 + 2 = 2 and χ2 = 1 − 1 + 1 = 1.. . ¨ 3. Γ2 . ¯ whose regular edges correspond to pairs Xi = Xj and ﬂipped edges correspond to pairs Xi = Xj . . . in Fig. 1. The convention is that the cycle is not completed until we return to the ”same side” of the disk that represents a vertex. .

. . In view of (2. o First we check the two M¨bius graphs for n = 1. What would be the formula for E(q1 X1 q2 X2 . . Figure 1. then (3.8).6 WLODEK BRYC AND VIRGIL PIERCE Question 3. This graph gives a projective sphere. it suﬃces to show that if X1 . X) or (X. see [8]) Proof. Xj1 +j2 ) × . . q2n X2n )? Or for more general Q-Gaussian n-touples? (For more general deﬁnition. . and each pair is either of the form (X. .2. × (Xj1 +j2 +jm−1 +1 . . . X2n ) = 4n−m (−2)χ(Γ) . Figure 2. χ(Γ) = 2 (2) ¯ ¯ E( (XY Y X)) = E(XX)E(Y Y ) = (−2)1 41 which matches the graph in Figure 3. . . . m = 1: o ¯ E( (X X)) = (−2)2 . Xj1 ) (Xj1 +1 . . X).7) and (2. which gives a o sphere (χ = 2) and projective sphere (χ = 1) respectively. (1) ¯ ¯ ¯ ¯ E( (XY Y X)) = E(X X)E(Y Y ) = (−2)2 41 which matches the graph in Figure 2. X2n consists of ¯ n independent pairs. For n = 2. . One checks that these correspond to the M¨bius graphs in Figure 1. m = 1 we have the following twelve cases. where Γ is the M¨bius graph that describes all pairings of the sequence.2) E (X1 X2 . and E( (XX)) = (−2)1 . . This graph gives a sphere. .

This graph gives a projective sphere. χ(Γ) = 0 . χ(Γ) = 1 (4) ¯¯ E( (XY Y X)) = E(XX)E(Y Y ) = 41 which matches the graph in Figure 5. Figure 5. χ(Γ) = 1 (3) ¯ ¯ E( (XY Y X)) = E(X X)E( (Y Y )) = (−2)1 41 which matches the graph in Figure 4. Figure 4. This graph gives a torus.SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 7 Figure 3. Figure 6. This graph gives a Klein bottle. χ(Γ) = 0 (5) ¯¯ ¯ ¯ E( (XY X Y )) = E(X X)E( (Y )Y ) = 41 which matches the graph in Figure 6.

¯ ¯ E( (XXY Y )) = E(XX)E(Y Y ) = (−2)1 41 which matches the graph in Figure 11. Figure 8. (10) . This graph gives a Klein bottle. χ(Γ) = 0 (8) ¯ E( (XY XY )) = E(XX)E(Y Y ) = (−2)1 41 which matches the graph in Figure 9. χ(Γ) = 0 (7) ¯ ¯ E( (XY XY )) = E(X X)E( Y Y ) = 41 which matches the graph in Figure 8. Figure 9. This graph gives a sphere. This graph gives a projective sphere. Figure 7. This graph gives a Klein bottle. This graph gives a projective sphere. χ(Γ) = 1 (9) ¯ ¯ ¯ ¯ E( (X XY Y )) = E(X X)E(Y Y ) = (−2)2 41 which matches the graph in Figure 10.8 WLODEK BRYC AND VIRGIL PIERCE (6) ¯ ¯¯ E( (XY X Y )) = E(XX)E(Y Y ) = 41 which matches the graph in Figure 7.

6). to the reader. Figure 12. which matches the graph in Figure 13.SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 9 Figure 10. χ(Γ) = 1 (11) ¯ ¯ E( (X XY Y )) = E(X X)E(Y Y ) = (−2)1 41 which matches the graph in Figure 12. (1) E( (XX) (Y Y ) = 4. they proceed as above using only the relations (2. . This graph gives a Klein bottle.3-2. χ(Γ) = 0 We leave the cases of n = 2. This graph gives a projective sphere. χ = 2. Figure 13. χ(Γ) = 1 (12) E( (XXY Y )) = E(XX)E(Y Y ) = 41 . m = 2. χ(Γ) = 2 Figure 11.

If Γ has two vertices that are joined by an edge. ¯ (7) E( (X) (XY Y ) = E( (X) (Y XY ) = E( (X) (Y XY ) = −2.6) glues together the two vertices. Therefore we will only need to prove the result for the single vertex case of the induction step. after reduction follow the same order. X2n ) = (−2)χ(Γ) 4n−1 . χ = 4. where Γ is a one vertex M¨bius graph with arrows (half edges) labeled by Xk . . The o ¯ trivial case is when X1 = X2 in which case we ﬁnd that (3. It is therefore enough to consider connected Γ. . say X1 and Xj1 and use (2.3) is ¯ (3. the left hand side of (3. ¯ (Y Y ) = 16. χ = 2. χ = 3. the faces are preserved – in the case of edge ﬂip in Fig. X2n ) = 4E(X3 X4 . We wish to show that (3. . 14. ¯ ¯ ¯ ¯¯ ¯ E( (X) (XY Y ) = E( (X) (XY Y ) = E( (X) (Y X Y ) = E( (X) (Y X Y ) = 4. . One notes that by independence of the pairs at diﬀerent edges. These geometric operations reduce n and m by one without changing the Euler characteristic: the number of edges and the number of vertices are reduced by 1.3) E(X1 X2 X3 X4 .10 WLODEK BRYC AND VIRGIL PIERCE ¯ E( (XX) (Y Y ) = −8. The use of relation (2. χ = 2.5) is just that of gluing the two vertices together removing the edge x which is labeled by the two appearances of Z. X2n ) = 4 (−2)χ2 4(n−1)−1 = (−2)χ2 4n−1 .6) to eliminate this pair from the product.5) or (2. (2) (3) (4) (5) (6) Figure 14. the edges of the face from which we remove the edge. E( (XY ) (XY ) = 4. χ = 1. χ = 1 We now proceed with the induction step. ¯ E( (X X) ¯ E( (XY ) (X Y ) = −2. . we can use cyclicity of to move the variables that label the edge to the ﬁrst positions in their cycles. Relation (2. . removing the edge x.2) factors into the product corresponding to connected components of Γ. and the reversal of orientation across the edge is given by the conjugate (see Figure 14). .4) E(X1 X1 )E(X3 X4 .

and q3 = Xk+1 . corresponding precisely to the n = 2. In what follows we take j < k. . X2n )] = (−2) (−2)χ2 4(n−1)−1 = (−2)χ2 −1 4n−1 . Xk−1 . in other words a ribbon with a ﬂip to it. . Geometrically conjugation reverses the order of the edges as well as switching local orientations. . . . Figure 15. q2 = Xj+1 . from which we see that the result follows in this case. from which we see that the result follows in this case. m = 1 cases 1-8: . Xj−1 . X2n ) = (−2) [E(X3 X4 . In the diagrams we have used an arrow to indicate an edge which reverses the local orientation of the vertex. χ2 = v2 − e2 + f2 = v1 − (e1 − 1) + (f1 − 1) = χ1 The second nearly trivial case is X2 = X1 in which case we ﬁnd that (3.3) is (3.7) E(X1 X1 )E(X3 X4 . X2n . . .5) corresponds to the reduction of the M¨bius graph given in Figure o 16 and represents a net change of 1 to the Euler characteristic. Conjugation of qi in the following computations arises from an orientation reversal. There are 8 non-trivial cases.SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 11 Equation (3. . χ2 = χ1 + 2.6) (3. .5) (3. Equation (3.4) corresponds to the reduction of the M¨bius graph given in Figure o 15 and represents a net of 0 to the Euler characteristic. . and let q1 = X3 .

12 WLODEK BRYC AND VIRGIL PIERCE Figure 16. . Figure 17. ¯ ¯ (1) When X1 = Xk and X2 = Xj we ﬁnd that ¯ ¯ E(X1 X2 q1 X2 q2 X1 q3 ) ¯ ¯ = E(X1 X1 )E( (X2 q1 X2 q2 ) (q3 )) ¯ = 4E(X2 X2 )E( (q1 ) (q2 ) (q3 )) = 42 E( (q1 ) (q2 ) (q3 )) = 42 (−2)χ2 4(n−2)−3 = (−2)χ2 −4 4n−1 .

represents a net of 3 to the Euler characteristic. This corresponds to the reduction of the M¨bius graph given in Figure o 18 and represents a net of 3 to the Euler characteristic. Figure 18. q3 = 1. as the number of faces is not changed when q1 . ¯ (2) When X1 = Xk and X2 = Xj we ﬁnd that ¯ E(X1 X2 q1 X2 q2 X1 q3 ) = E(X1 X1 ) E(¯2 X2 q1 X2 q3 ) q ¯ ¯ = (−2) E(X2 q1 X2 q3 q2 ) ¯ ¯ ¯ ¯ = (−2)E(X2 X2 )E( (¯1 ) (q3 q2 )) q ¯ = (−2)3 E( (¯1 ) (q3 q2 )) q ¯ = (−2)3 (−2)χ2 4(n−2)−2 = (−2)χ2 −3 4n−1 . and the result follows. ¯ (3) When X1 = Xk and X2 = Xj we ﬁnd that ¯ E(X1 X2 q1 X2 q2 X1 q3 ) ¯ = E(X1 X1 )E( (X2 q1 X2 q2 ) (q3 )) = 4E(X2 X2 )E( (¯1 q2 ) (q3 )) q = (−2)3 [E( (¯1 q2 ) (q3 ))] q = (−2)3 (−2)χ2 4(n−2)−2 = (−2)χ2 −3 4n−1 . We note that χ2 = v2 − e2 + f2 = v1 + 2 − (e1 − 2) + f1 = χ1 + 4. This corresponds to the reduction for the M¨bius graph given in Figure o 19. q2 .SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 13 This corresponds to the reduction of the M¨bius graph given in Figure o 17. .

(4) When X1 = Xk and X2 = Xj we ﬁnd that E(X1 X2 q1 X2 q2 X1 q3 ) = E(X1 X1 )E(¯2 X2 q1 X2 q3 ) q ¯ ¯ ¯ ¯ ¯ ¯ ¯ = (−2)E(X2 q1 X2 q3 q2 ) ¯ ¯ = (−2)E(X2 X2 )E(q1 q3 q2 ) ¯ = (−2)2 E(q1 q3 q2 ) ¯ = (−2)2 (−2)χ2 4(n−2)−1 = (−2)χ2 −2 4n−1 . .14 WLODEK BRYC AND VIRGIL PIERCE Figure 19. Figure 20.

and the result follows. and the result follows. Figure 21. . ¯ ¯ (5) When X1 = Xj and X2 = Xk we ﬁnd that ¯ ¯ E(X1 X2 q1 X1 q2 X2 q3 ) ¯ ¯ = E(X1 X1 )E( (X2 q1 ) (q2 X2 q3 )) ¯ = 4E( (X2 q1 ) (X2 q3 q2 )) = 4E( (q1 q3 q2 )) = 4 (−2)χ2 4(n−2)−1 = (−2)χ2 −2 4n−1 .SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 15 This corresponds to the reduction for the M¨bius graph given in Figure o 20. represents a net of 2 to the Euler characteristic (χ2 = χ1 + 2). represents a net of 2 to the Euler characteristic. represents a net of 2 to the Euler characteristic. This corresponds to the reduction for the M¨bius graph given in Figure o 21. ¯ (6) When X1 = Xj and X2 = Xk we ﬁnd that ¯ E(X1 X2 q1 X1 q2 X2 q3 ) ¯ = E(X1 X1 )E(¯1 X2 q2 X2 q3 ) q ¯ ¯ ¯ ¯ = (−2)E(X2 q2 X2 q3 q1 ) ¯ ¯ = (−2)E(X2 X2 )E(¯2 q3 q1 ) q ¯ = (−2)2 E(¯2 q3 q1 ) q ¯ = (−2)−2 (−2)χ2 4(n−2)−1 = (−2)χ2 −2 4n−1 . and the result follows. This corresponds to the reduction for the M¨bius graph given in Figure o 22.

represents a net of 3 to the Euler characteristic. represents a net of 2 to the Euler characteristic.16 WLODEK BRYC AND VIRGIL PIERCE Figure 22. and the result follows. ¯ (7) When X1 = Xj and X2 = Xk we ﬁnd that ¯ E(X1 X2 q1 X1 q2 X2 q3 ) ¯ = E(X1 X1 )E( (X2 q1 ) (q2 X2 q3 )) = 4E( (X2 q1 ) (X2 q3 q2 )) = 4E(¯1 q3 q2 ) q = 4 (−2)χ2 4(n−2)−1 = (−2)χ2 −2 4n−1 . . This corresponds to the reduction for the M¨bius graph given in Figure o 24. This corresponds to the reduction for the M¨bius graph given in Figure o 23. (8) When X1 = Xj and X2 = Xk we ﬁnd that E(X1 X2 q1 X1 q2 X2 q3 ) = E(X1 X1 )E(¯1 X2 q2 X2 q3 ) q ¯ = (−2)E(X2 q3 q1 X2 q2 ) ¯ ¯ ¯ = (−2)E(X2 X2 )E( (q3 q1 ) (q2 )) ¯ = (−2)3 E( (q3 q1 ) (q2 )) ¯ = (−2)3 (−2)χ2 4(n−2)−2 = (−2)χ2 −3 4n−1 . and the result follows.

But the faces need to be counted as follows: The inner face of the removed edge becomes the outside face of one component. by the induction hypothesis. Note that taking away an oriented ribbon creates a new vertex. . Figure 24. the proof is completed. With each of these cases checked. or it may split into two components.SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 17 Figure 23. If taking away a loop makes the graph disconnected. and the outer face at the removed edge becomes the outer face of the other component. then the counts of changes to edges and vertices are still the same. Thus the counting of faces is not aﬀected by whether the graph is connected – no change. The remaining graph might be still connected.

1) Z = [Zi. In fact.18 WLODEK BRYC AND VIRGIL PIERCE 4. In this section we will demonstrate a multi-matrix version of the theorem of Mulase and Waldron [5] from a heuristic argument using our Wick formula and quadratic relations. The duality for one-matrix GOE/GSE ensembles appears in [5]. . χ(Γ) is the Euler characteristic and j1 + j2 + · · · + jm = 2n. .1. . 1 E( (tr(Zj1 )) (tr(Zj2 )) . . (tr(Zjm ))) = (4N )n−m (−2N )χ(Γ) . . . {Zi. 2n} → {1. . so in the formulas we will use (tr(A)). the conjugate matrix is N A∗ := Aj. We will compute here the expected values of the real traces of powers of a quaternionic self-dual matrix in the Gaussian symplectic ensemble (4. and Zi. j. αk = j1 + j2 + · · · + jk−1 + 1. The duality for one-matrix Wishart ensembles was discovered in [1. i.j ] where {Zi. By MM ×N (Hβ ) we denote the set of all M × N matrices with entries from Hβ . . Corollary 4. A multi-matrix version of Mulase and Waldron. Γ where the sum is over labeled M¨bius graphs Γ with m vertices of degree j1 . 4.1. We will be interested GUEβ ensembles. .i } ¯ is a family of independent real normal of variance 2. Such a nonstandard deﬁnition is consistent with the treatment of A ∈ MN ×N (Hβ ) as an element of MβN ×βN (R) via transformation (??).i . the natural deﬁnition of trace would be trβ (·) := β (tr(·)).1. More generally. This represents an improvement to the argument of Mulase and Waldron as we will not need to rely on labelings of the vertices by the quaternions 1.2]. j2 . . For A ∈ MM ×N (Hβ ). and k.i . These are square self-adjoint matrices (4. which is supported on the self-adjoint subset of MN ×N (Hβ ). GSE Ensembles. . . . . Zs are independent N × N GSE ensembles and t : {1.1. A derivation based on recurrence formulas for moments is in [3]. . . The law of 1 2 such a matrix has a density C exp(− 2 tr(x )).5) and (2.j : i < j} is a family of independent Hβ -normal random variables.2). (C = C(N. . Let α0 = 1.j When β = 4. traces tr(A) may fail to commute.2. This part of their argument is encoded in relations (2.1) where the oﬀ-diagonal matrix entries are (2. The basic theorem we will prove is Theorem 4.6). o jm . Duality between GOE and GSE ensembles The duality between β = 1 and β = 4 asserts that the formulas for moments of traces of polynomials in H-valued Gaussian ensembles can be obtained from the corresponding formulas for the R-valued Gaussian ensembles by replacing the dimension parameter N by (−2N ). Non-standard deﬁnition of quaternionic trace is used in [1] 4. β) is a normalizing constant.) 4. i. . . The trace is tr(A) = j=1 Ajj . suppose Z1 . Quaternion matrices.j = Zj. s} is ﬁxed.

a(σ(j)) )). 1 2 [reference?] therefore Theorem 4. . . . We begin by expanding out the traces in terms of the matrix entries (4.c2 .c1 )) . so that the formula is E( ( a:[1. X2 be independent N × GOE ensembles. c2 . . . m. . (tr(Zt(αm ) . Luckily. . and pairings that match the diagonal (real) entries.) Example 4.. .3) Nm E( (tr(Zj1 )) (tr(Zj2 )) . . Zcjm . Fix r. and we can assume that the pairs are independent.2n]→[1. pairings that match two independent random variables do not contribute to the sum. . χ(Γ) is the Euler characteristic and j1 + j2 + · · · + jm = 2n. we should clarify how come the Theorem IS the ”duality”? Za(j). .SEMINAR TALK: WICK FORMULA FOR QUATERNION NORMAL LAWS 19 βk = j1 + j2 + · · · + jk denote the ranges under traces. . . . = . b2 . suppose Z1 and Z2 are independent N × GSE and X1 . 1 ≤ a1 . (If there are no Γ that are consistent with the coloring t.8) it follows that the right hand side of (4. (tr(Zjm ))) = 4n−m 1 (4n)n−m E( (Za1 . To illustrate the multi-matrix aspect of the theorem. . . Zt(β2 ) )) . Colors make for cumbersome notation! It is then better to index the products by the cycles of a permutation.b1 ) . t(j) We then need a special convention to follow the right order for products over the cycles. Zt(β1 ) )) (tr(Zt(α2 ) .. k and let f (N ) := E(tr(X2m X2k )r ). where the sum is over labeled color-preserving M¨bius graphs Γ with vertices of o degree j1 . Zt(βm ) )) = Γ (−2N )χ(Γ) . jm that are colored with s colors by the mapping t. Zaj1 . . . .b2 .a1 ) (Zb1 . As previously. ¯ pairs that match Z with Z. . . Zbj2 . .2).j must now be labeled also by the ”color” t. . The pairings that contribute to the sum are of three diﬀerent types: pairs that match Z with another Z at a diﬀerent position in the product. except that the consecu(t) tive entries Zi.. .N ] c∈σ j∈c Perhaps instead of the example. . Of course.a2 Za2 . .2) 1 E (4N )n−m (tr(Zt(1) . we interpret the sum as 0. .2.1 implies that the moments for the independent GSE ensembles are determined from the corresponding moments of independent GOE ensembles by the “dual formula” E(tr(Z2m Z2k )r ) = (4)(m+k−1)r f (−2N ). Since f (N ) = N (m+k−1)r Γ N χ(Γ) . . .a3 . a2 . (Zc1 . j2 . cjm ≤ N Note that essentially the same expansion applies to (4. . . . From (2. Then (4. .3) can be expanded as sum over all pairings. . . . 1 2 Proof. . 1 ≤ c1 . bj2 ≤ N . . of quaternions has cyclic property so such products are well defnied. aj 1 ≤ N 1 ≤ b1 .. .

2006. the relations given by the edges of Γ reduce the number of summations leaving us with f (Γ) sums from 1 to N . once we replace all the real entries that came from the diagonal entries by the corresponding quaternion-normal pairs. where f (Γ) is the number of faces of Γ. 1997. .. we get the sum over all possible pairs of matches of the ﬁrst two types only.. P. Math. http://arxiv. Probability Appl. Mck . (Xj1 +···+jm−1 +1 . [4] Leonov. [9] Wick. T.i = ξ with another diagonal entry Zj. Cambridge University Press. V. 25. 129 of Cambridge Tracts in Mathematics.. The pairs of variables at diﬀerent ribbons can now be assumed independent. RI.3). 4 (1959). 1991). [8] Vakhania.pdf. N. Cambridge.. Duality of orthogonal and symplectic matrix integrals and quaternionic Feynman graphs. Phys. [2] Janson. In Hypergeometric functions on domains of positivity.Phys.. 2007. Cumulants and partition lattices. . [3] Ledoux. [5] Mulase. 2 (Oct 1950). Thus. with the interpretation that pairings or random variables Z. . . So the contribution of each such diagonal pairing is the same as that of two matches ¯ of quaternionic entries (Z. and Stembridge. Amer.fr/ ledoux/GOE. . 268–272. Commun. 138 of Contemp. we see that the same answer arises from ¯ ¯ E (q0 Zq1 Zq2 ) + E (q0 Zq1 Zq2 ) = E (Zq1 Zq2 q0 ) + E (Zq1 Zq2 q0 ) = 2 (q1 q2 q0 ).univ-toulouse. J. Math. vol. N. Theor. R. and jm with edges labeled by Xk . Since real numbers commute with quaternions. .jm−1 +k . R. . [7] Speed. Gaussian Hilbert spaces. Our claim is then that given a M¨bius graph Γ with vertices of o degree j1 . Waldron.20 WLODEK BRYC AND VIRGIL PIERCE We ﬁrst dispense with the pairings that match a diagonal entry Zi. Some combinatorial aspects of the spectra of normally distributed random matrices. () 240 (2003). P. Xj1 +j2 ) . S. 151–174. pp. Motohico. N. X2n )) (4N )n−m = (−2N )χ(Γ) N −f (Γ) . On a method of semi-invariants. References [1] Hanlon.4). Providence. M.ck+1 = Xj1 +j2 +. . . 99–115.org/abs/math-ph/0206011. vol. j2 . (E(q0 ξq1 ξq2 )) = E(ξ 2 ) (q0 q1 q2 ) = 2 (q0 q1 q2 ). An algorithm for map enumeration. and Shiryaev. satisﬁes (4. Theory of Probability and its Applications 43 (1999). A recursion formula for the moments of the gaussian orthogonal ensemble. .j by Zak . Jack polynomials. 378–388.4) 1 E( (X1 . Z) and (Z. 553–586. 1992. 319–329. . The evaluation of the collision matrix.ak+1 = Xk .bk+1 = Xj1 +k . while pairings Z. .math. Statist. J. [6] Pierce. J. . . Rev. Xj1 ) (Xj1 +1 . 2 (1983). which restricts the number of available pairings. the edges of the graph at each vertex are colored according to function t. A. FL. . P. and applications (Tampa. Z). V..Math. using the cyclic property of (·) and adding formulas (2. C. We label all such pairings by M¨bius o graphs. Zbk .3) and (2. preprint: http://www. Soc. Stanley. P. We now relabel the Zi. G. A. 80. In the multi-matrix case. The N −f (Γ) terms are removed by the summations in (4. U. Z correspond ¯ to twisted ribbons.j = ξ.2). Random vectors with values in quaternion hilbert space. Z correspond to ribbons without a twist. . Austral. On the other hand. . Collecting the powers of N we ﬁnd that this is the same as (3.

Sign up to vote on this title

UsefulNot useful- journal of computer science Constraint Minimum Vertex Cover in K-Partite Graph: Approximation Algorithm and Complexity Analysisby ijcsis
- Kruskal Algorithmby Siti Hajar Ismail
- 10.1.1.130.7655 Attitude Determination.by Alee K. Obeid
- An open graph visualization system and its applications to software engineeringby Jeff Pratt

- Exercises about Cognitive Computing
- RESOLVING OF VERTEX DISCLOSURE AND MULTI COMMUNITY DISCLOSURE IN SOCIAL NETWORKS
- plugin-art%253A10.1007%252Fs10479-007-0181-5
- Archived
- Pajek Manual
- 183
- 10.1.1.146.4390
- Defending the Roman Empire—A new strategy
- chapter7a
- Graph Cluster Analysis
- Breadth First Search
- Sol 4
- GN Lecture 5
- Comparison between Vertex Merge Algorithm and Dsatur Algorithm
- Depthmap Paper VGA
- 9781783987344_Mastering_Gephi_Network_Visualization_Sample_Chapter
- HPhys Unit 01 CVPM Packet 2013
- pkdd13
- Hill Cclimbing Algorithm
- Dban Smart Camera Ready
- journal of computer science Constraint Minimum Vertex Cover in K-Partite Graph
- Kruskal Algorithm
- 10.1.1.130.7655 Attitude Determination.
- An open graph visualization system and its applications to software engineering
- Assignment
- Abinitio - Practitioner_v 1.0_day1
- warshall's algorith
- gb-2006-7-1-r7
- A Workbench for Emergent Urbanism and Architectural Form
- ijsrp-p2219
- ion Wick 08

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.