Uniform Random Walks On The Plane: A Case Study in Experimental Mathematics

Intro Random walk Densities Expectations 3 and 4 steps
Uniform Random Walks on the Plane

A case study in experimental mathematics
James Wan
14 January, 2013
1 / 21
What is Experimental Mathematics?
The use of computers beyond routine simulations and calculations.
As dened by J. Borwein and D. Bailey:
1
Use graphics to suggest underlying principles; test conjectures;
conrm analytical results.
2
Gain intuition; discover patterns; suggest approaches for proof.
Computation as the third form of discovery, after theory and
experiment:
Algorithms of Celine, Gosper, Wilf-Zeilberger completely
automate binomial sums etc.
Find answer rst, then reverse engineer.
tool < computer collaborator.
A journal since 1992.
2 / 21
1
2
experiment:
2 / 21
1
2
experiment:
2 / 21
1
2
experiment:
2 / 21
1
2
experiment:
2 / 21
1
2
experiment:
2 / 21
1
2
experiment:
2 / 21
1
2
experiment:
2 / 21
Examples of Experimental Mathematics
Area of parabola (weighing, c. 250BC, Archimedes)
AGM and elliptic integrals (1799, Gauss)
Feigenbaum constant (on HP calculator, 1975)
Four colour theorem (Appel and Haken, 1976)
Kepler conjecture (LP, Hales, 1992-8)
Independent computation of digits of (BBP, 1995)
Solving checkers (Schaeer, 2007)
Solving sudoku (Douglas-Rachford, convex optimization, 2010)
3 / 21
3 / 21
3 / 21
3 / 21
3 / 21
3 / 21
3 / 21
3 / 21
My use of Experimental Mathematics (1)
Gaussian quadrature:
Traditionally used to approximate integrals by nite sums and
orthogonal polynomials.
Computational insight: use discrete version to approximate sums;
use orthogonal rational functions.
Surprisingly good for lattice sums, e.g. 1.4 digits per term for
Madelung constant
m,n,p
(1)
m+n+p
_
m
2
+ n
2
+ p
2
.
4 / 21
Madelung constant
m,n,p
(1)
m+n+p
_
m
2
+ n
2
+ p
2
.
4 / 21
Madelung constant
m,n,p
(1)
m+n+p
_
m
2
+ n
2
+ p
2
.
4 / 21
Elliptic integrals: K(x) =
_
/2
0
dt
1x
2
sin
2
t
_
1
0
K(x)
3
dx =
3(1/4)
8
1280
2
7.090227004846.
Reversed engineered using the Inverse Symbolic Calculator (PSLQ,
can certies no closed form below a certain size exists).
RHS is the evaluation of a lattice sum, so proof found by bridging
two sides via functions.
Galilean experiment: either gives us condence in the view we are
taking or rules out some possibilities.
5 / 21
_
/2
0
dt
1x
2
sin
2
t
_
1
0
K(x)
3
dx =
3(1/4)
8
1280
2
7.090227004846.
5 / 21
_
/2
0
dt
1x
2
sin
2
t
_
1
0
K(x)
3
dx =
3(1/4)
8
1280
2
7.090227004846.
5 / 21
_
/2
0
dt
1x
2
sin
2
t
_
1
0
K(x)
3
dx =
3(1/4)
8
1280
2
7.090227004846.
5 / 21
Special functions:
Computer assisted discovery and automatic proof of the g.f.
(1 cxy)
_

n=0
u
n
x
n
__

n=0
u
n
y
n
_
=
n=0
u
n
P
n
_
(x + y)(1 + cxy) 2axy
(y x)(1 cxy)
__
y x
1 cxy
_
n
,
where (n + 1)
2
u
n+1
= (an
2
+ an + b)u
n
cn
2
u
n1
.
Brings together special functions, Apery-like sequences, and
Ramanujan-type series for 1/.
n=0
_
n
k=0
k
j=0
_
n
k
__
1
8
_
k
_
k
j
_
3
_
nP
n
_
5
3
3
__
4
3
3
_
n
=
9
3
2
.
6 / 21
Special functions:
(1 cxy)
_

n=0
u
n
x
n
__

n=0
u
n
y
n
_
=
n=0
u
n
P
n
_
(x + y)(1 + cxy) 2axy
(y x)(1 cxy)
__
y x
1 cxy
_
n
,
where (n + 1)
2
u
n+1
= (an
2
+ an + b)u
n
cn
2
u
n1
.
n=0
_
n
k=0
k
j=0
_
n
k
__
1
8
_
k
_
k
j
_
3
_
nP
n
_
5
3
3
__
4
3
3
_
n
=
9
3
2
.
6 / 21
Special functions:
(1 cxy)
_

n=0
u
n
x
n
__

n=0
u
n
y
n
_
=
n=0
u
n
P
n
_
(x + y)(1 + cxy) 2axy
(y x)(1 cxy)
__
y x
1 cxy
_
n
,
where (n + 1)
2
u
n+1
= (an
2
+ an + b)u
n
cn
2
u
n1
.
n=0
_
n
k=0
k
j=0
_
n
k
__
1
8
_
k
_
k
j
_
3
_
nP
n
_
5
3
3
__
4
3
3
_
n
=
9
3
2
.
6 / 21
Random walks:
Very basic problem; sum of n random complex numbers.
Not much known computationally or analytically before 2009.
Application: Brownian motion, superposition of waves and
vibrations, quantum chemistry, migration, cryptography.
All our discoveries were experimental.
Hypergeometric series:
p
F
q
_
a
1
, . . . , a
p
b
1
, . . . , b
q
z
_
=
n=0
(a
1
)
n
(a
p
)
n
(b
1
)
n
(b
q
)
n
z
n
n!
.
7 / 21
Random walks:
p
F
q
_
a
1
, . . . , a
p
b
1
, . . . , b
q
z
_
=
n=0
(a
1
)
n
(a
p
)
n
(b
1
)
n
(b
q
)
n
z
n
n!
.
7 / 21
Random walks:
p
F
q
_
a
1
, . . . , a
p
b
1
, . . . , b
q
z
_
=
n=0
(a
1
)
n
(a
p
)
n
(b
1
)
n
(b
q
)
n
z
n
n!
.
7 / 21
Random walk integrals
Denition: For complex s,
W
n
(s) :=
_
[0,1]
n
k=1
e
2x
k
i
s
dx
W
n
(1) is the expectation.
Denition:
The density p
n
is the (unique) function that satises
W
n
(s) =
_
n
0
p
n
(x)x
s
dx.
Dimension reduction: let x
1
= 0.
8 / 21
W
n
(s) :=
_
[0,1]
n
k=1
e
2x
k
i
s
dx
W
n
Denition:
The density p
n
W
n
(s) =
_
n
0
p
n
(x)x
s
dx.
1
= 0.
8 / 21
W
n
(s) :=
_
[0,1]
n
k=1
e
2x
k
i
s
dx
W
n
Denition:
The density p
n
W
n
(s) =
_
n
0
p
n
(x)x
s
dx.
1
= 0.
8 / 21
Computational challenge
W
1
(s) = 1, p
1
(x) =
1
(x).
Maple 13 and Mathematica 7 think W
2
= 0.
p
2
(x) =
2
4x
2
, W
2
(s) =
_
s
s/2
_
, W
2
(1) =
4
.
Tanh-sinh quadrature gives 175 digits for W
3
(1), but
everything fails for W
4
(1). 256 cores at LBNL:
W
5
(1) 2.0081618.
9 / 21
W
1
(s) = 1, p
1
(x) =
1
(x).
2
= 0.
p
2
(x) =
2
4x
2
, W
2
(s) =
_
s
s/2
_
, W
2
(1) =
4
.
3
(1), but
4
W
5
(1) 2.0081618.
9 / 21
W
1
(s) = 1, p
1
(x) =
1
(x).
2
= 0.
p
2
(x) =
2
4x
2
, W
2
(s) =
_
s
s/2
_
, W
2
(1) =
4
.
3
(1), but
4
W
5
(1) 2.0081618.
9 / 21
W
1
(s) = 1, p
1
(x) =
1
(x).
2
= 0.
p
2
(x) =
2
4x
2
, W
2
(s) =
_
s
s/2
_
, W
2
(1) =
4
.
3
(1), but
4
W
5
(1) 2.0081618.
9 / 21
Jan Cornelius Kluyver & John William Strutt
p
n
(t) =
_

0
xtJ
0
(xt)J
n
0
(x) dx.
Probability of returning to the unit disk:
_
1
0
p
n
(t)dt =
_

0
J
1
(x)J
n
0
(x) dx =
_
J
0
(x)
n+1
n + 1
_
0
=
1
n + 1
.
Rayleigh (multivariate CLT): p
n
(x)
2x
n
e
x
2
/n
.
10 / 21
p
n
(t) =
_

0
xtJ
0
(xt)J
n
0
(x) dx.
_
1
0
p
n
(t)dt =
_

0
J
1
(x)J
n
0
(x) dx =
_
J
0
(x)
n+1
n + 1
_
0
=
1
n + 1
.
n
(x)
2x
n
e
x
2
/n
.
10 / 21
p
n
(t) =
_

0
xtJ
0
(xt)J
n
0
(x) dx.
_
1
0
p
n
(t)dt =
_

0
J
1
(x)J
n
0
(x) dx =
_
J
0
(x)
n+1
n + 1
_
0
=
1
n + 1
.
n
(x)
2x
n
e
x
2
/n
.
10 / 21
p
n
with approximations superimposed.
11 / 21
Probability
We condition the distance z of an (n + m)-step walk on x (n
steps), followed by y (m steps).
Cosine rule, z
2
= x
2
+ y
2
+ 2xy cos().
\
x
z
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
y

So W
n+m
(s) =
1
_
n
0
_
m
0
_
_

0
z
s
d
_
p
n
(x)p
m
(y) dxdy.
Change of variable:
W
n+m
(s) =
_
n+m
0
z
s
__
n
0
_

0
z
y
p
n
(x)p
m
(y) ddx
_
. .
p
n+m
(z)
dz.
12 / 21
Probability
Cosine rule, z
2
= x
2
+ y
2
+ 2xy cos().
\
x
z
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
y

So W
n+m
(s) =
1
_
n
0
_
m
0
_
_

0
z
s
d
_
p
n
(x)p
m
(y) dxdy.
Change of variable:
W
n+m
(s) =
_
n+m
0
z
s
__
n
0
_

0
z
y
p
n
(x)p
m
(y) ddx
_
. .
p
n+m
(z)
dz.
12 / 21
Probability
Cosine rule, z
2
= x
2
+ y
2
+ 2xy cos().
\
x
z
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
y

So W
n+m
(s) =
1
_
n
0
_
m
0
_
_

0
z
s
d
_
p
n
(x)p
m
(y) dxdy.
Change of variable:
W
n+m
(s) =
_
n+m
0
z
s
__
n
0
_

0
z
y
p
n
(x)p
m
(y) ddx
_
. .
p
n+m
(z)
dz.
12 / 21
Probability
Cosine rule, z
2
= x
2
+ y
2
+ 2xy cos().
\
x
z
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
y

So W
n+m
(s) =
1
_
n
0
_
m
0
_
_

0
z
s
d
_
p
n
(x)p
m
(y) dxdy.
Change of variable:
W
n+m
(s) =
_
n+m
0
z
s
__
n
0
_

0
z
y
p
n
(x)p
m
(y) ddx
_
. .
p
n+m
(z)
dz.
12 / 21
Recursion for p
n
p
n
is a single integral over p
n1
. So
p
3
(x) =
2
3x
(3 + x
2
)
2
F
1
_
1
3
,
2
3
1
x
2
(9 x
2
)
2
(3 + x
2
)
3
_
.
(Found experimentally, proof by DE.)
Pearson posed the problem (1905), thought p
5
had a straight
line. Disproved in 1963.
p
4
hard to compute; we resort to moments and analytic
continuation.
0.5 1.0 1.5 2.0 2.5 3.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1 2 3 4
0.1
0.2
0.3
0.4
0.5
13 / 21
Recursion for p
n
p
n
n1
. So
p
3
(x) =
2
3x
(3 + x
2
)
2
F
1
_
1
3
,
2
3
1
x
2
(9 x
2
)
2
(3 + x
2
)
3
_
.
5
had a straight
p
4
continuation.
0.5 1.0 1.5 2.0 2.5 3.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1 2 3 4
0.1
0.2
0.3
0.4
0.5
13 / 21
Recursion for p
n
p
n
n1
. So
p
3
(x) =
2
3x
(3 + x
2
)
2
F
1
_
1
3
,
2
3
1
x
2
(9 x
2
)
2
(3 + x
2
)
3
_
.
5
had a straight
p
4
continuation.
0.5 1.0 1.5 2.0 2.5 3.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1 2 3 4
0.1
0.2
0.3
0.4
0.5
13 / 21
Combinatorics and analysis
Binomial expansion:
W
n
(s) = n
s
m0
(1)
m
n
2m
_
s
2
m
_
I
n,m
.
I
3,m
found experimentally on the OEIS, generalized to I
n,m
guessing, then proven combinatorially.
It follows that
W
n
(2k) =
a
1
+...+a
n
=k
_
k
a
1
, ..., a
n
_
2
.
Has a recursion lifts to a functional equation W
n
(s) has
analytical continuation to C with poles at negative integers.
14 / 21
Binomial expansion:
W
n
(s) = n
s
m0
(1)
m
n
2m
_
s
2
m
_
I
n,m
.
I
3,m
n,m
It follows that
W
n
(2k) =
a
1
+...+a
n
=k
_
k
a
1
, ..., a
n
_
2
.
n
(s) has
14 / 21
Binomial expansion:
W
n
(s) = n
s
m0
(1)
m
n
2m
_
s
2
m
_
I
n,m
.
I
3,m
n,m
It follows that
W
n
(2k) =
a
1
+...+a
n
=k
_
k
a
1
, ..., a
n
_
2
.
n
(s) has
14 / 21
Binomial expansion:
W
n
(s) = n
s
m0
(1)
m
n
2m
_
s
2
m
_
I
n,m
.
I
3,m
n,m
It follows that
W
n
(2k) =
a
1
+...+a
n
=k
_
k
a
1
, ..., a
n
_
2
.
n
(s) has
14 / 21
Three steps
Recursion + subtle analysis convolution formula for W
4
(s)
in terms of W
3
.
By playing around,
W
3
(k) = Re
3
F
2
_
1/2, k/2, k/2
1, 1
4
_
.
Theorem (1), Borwein, Nuyens, Straub, W. (2009)
W
3
(1) =
16
3
4
2
(
1
3
)
6
+
3(
1
3
)
6
8
3
4
4
1.57459723755.
Proven using elementary manipulation of integrand and the
transform Re K(1/x) = xK(x).
15 / 21
Three steps
4
(s)
in terms of W
3
.
By playing around,
W
3
(k) = Re
3
F
2
_
1/2, k/2, k/2
1, 1
4
_
.
W
3
(1) =
16
3
4
2
(
1
3
)
6
+
3(
1
3
)
6
8
3
4
4
1.57459723755.
15 / 21
Three steps
4
(s)
in terms of W
3
.
By playing around,
W
3
(k) = Re
3
F
2
_
1/2, k/2, k/2
1, 1
4
_
.
W
3
(1) =
16
3
4
2
(
1
3
)
6
+
3(
1
3
)
6
8
3
4
4
1.57459723755.
15 / 21
Three steps
4
(s)
in terms of W
3
.
By playing around,
W
3
(k) = Re
3
F
2
_
1/2, k/2, k/2
1, 1
4
_
.
W
3
(1) =
16
3
4
2
(
1
3
)
6
+
3(
1
3
)
6
8
3
4
4
1.57459723755.
15 / 21
Four steps
Theorem (2), Borwein, Straub, W., Zudilin (2010)
W
4
(1) 1.79909248 is given by
3
4
7
F
6
_
7
4
,
3
2
,
3
2
,
3
2
,
1
2
,
1
2
,
1
2
3
4
, 2, 2, 2, 1, 1
1
_
3
8
7
F
6
_
7
4
,
3
2
,
3
2
,
1
2
,
1
2
,
1
2
,
1
2
3
4
, 2, 2, 2, 2, 1
1
_
.
Guessed using PSLQ (based on W
4
(1)).
Meijer G-function: dened as a contour integral of ratios of
s. The mother of all special functions.
Important in CAS: many denite integrations are Meijer G
transformations.
W
4
(s) =
2
s
3
(1 + s/2)
(s/2)
G
2,4
4,4
_
1, (1 s)/2, 1, 1
1/2, s/2, s/2, s/2
1
_
.
16 / 21
Four steps
W
4
(1) 1.79909248 is given by
3
4
7
F
6
_
7
4
,
3
2
,
3
2
,
3
2
,
1
2
,
1
2
,
1
2
3
4
, 2, 2, 2, 1, 1
1
_
3
8
7
F
6
_
7
4
,
3
2
,
3
2
,
1
2
,
1
2
,
1
2
,
1
2
3
4
, 2, 2, 2, 2, 1
1
_
.
4
(1)).
transformations.
W
4
(s) =
2
s
3
(1 + s/2)
(s/2)
G
2,4
4,4
_
1, (1 s)/2, 1, 1
1/2, s/2, s/2, s/2
1
_
.
16 / 21
Four steps
W
4
(1) 1.79909248 is given by
3
4
7
F
6
_
7
4
,
3
2
,
3
2
,
3
2
,
1
2
,
1
2
,
1
2
3
4
, 2, 2, 2, 1, 1
1
_
3
8
7
F
6
_
7
4
,
3
2
,
3
2
,
1
2
,
1
2
,
1
2
,
1
2
3
4
, 2, 2, 2, 2, 1
1
_
.
4
(1)).
transformations.
W
4
(s) =
2
s
3
(1 + s/2)
(s/2)
G
2,4
4,4
_
1, (1 s)/2, 1, 1
1/2, s/2, s/2, s/2
1
_
.
16 / 21
Four steps
W
4
(1) 1.79909248 is given by
3
4
7
F
6
_
7
4
,
3
2
,
3
2
,
3
2
,
1
2
,
1
2
,
1
2
3
4
, 2, 2, 2, 1, 1
1
_
3
8
7
F
6
_
7
4
,
3
2
,
3
2
,
1
2
,
1
2
,
1
2
,
1
2
3
4
, 2, 2, 2, 2, 1
1
_
.
4
(1)).
transformations.
W
4
(s) =
2
s
3
(1 + s/2)
(s/2)
G
2,4
4,4
_
1, (1 s)/2, 1, 1
1/2, s/2, s/2, s/2
1
_
.
16 / 21
Experimental proofs
Transform to G
2,2
4,4
.
Nesterenkos theorem: nice G
2,2
4,4
= triple integral.
a := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
= 2W
4
(1) not nice.
c := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
is nice. Experimentally a = 4c.
Once found, easy to prove. Introduce parameter z as
argument in a dierentiation.
Split triple integral in 2, Zudilins theorem: =
7
F
6
.
17 / 21
Experimental proofs
Transform to G
2,2
4,4
.
2,2
4,4
= triple integral.
a := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
= 2W
4
(1) not nice.
c := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
7
F
6
.
17 / 21
Experimental proofs
Transform to G
2,2
4,4
.
2,2
4,4
= triple integral.
a := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
= 2W
4
(1) not nice.
c := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
7
F
6
.
17 / 21
Experimental proofs
Transform to G
2,2
4,4
.
2,2
4,4
= triple integral.
a := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
= 2W
4
(1) not nice.
c := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
7
F
6
.
17 / 21
Experimental proofs
Transform to G
2,2
4,4
.
2,2
4,4
= triple integral.
a := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
= 2W
4
(1) not nice.
c := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
7
F
6
.
17 / 21
Experimental proofs
Transform to G
2,2
4,4
.
2,2
4,4
= triple integral.
a := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
= 2W
4
(1) not nice.
c := G
2,2
4,4
_
0,1,1,1
1
2
,
1
2
,
1
2
,
1
2
1
_
7
F
6
.
17 / 21
Closed form for p
3
p
3
(x) =
2x
k=0
W
3
(2k)
_
x
3
_
2k
.
With care, for small > 0,
_

0
p
3
(x)x
s
dx =
2
s+2
3(s + 2)
+
2
s+4
3
3(s + 4)
+
Residues of W
3
(s) come from series coecients.
Also explains the shape of p
5
.
If p
4
admits a Taylor series around 0, this argument would
give simple poles for W
4
(s), but it has double poles. !?
18 / 21
Closed form for p
3
p
3
(x) =
2x
k=0
W
3
(2k)
_
x
3
_
2k
.
_

0
p
3
(x)x
s
dx =
2
s+2
3(s + 2)
+
2
s+4
3
3(s + 4)
+
Residues of W
3
5
.
If p
4
4
18 / 21
Closed form for p
3
p
3
(x) =
2x
k=0
W
3
(2k)
_
x
3
_
2k
.
_

0
p
3
(x)x
s
dx =
2
s+2
3(s + 2)
+
2
s+4
3
3(s + 4)
+
Residues of W
3
5
.
If p
4
4
18 / 21
Closed form for p
3
p
3
(x) =
2x
k=0
W
3
(2k)
_
x
3
_
2k
.
_

0
p
3
(x)x
s
dx =
2
s+2
3(s + 2)
+
2
s+4
3
3(s + 4)
+
Residues of W
3
5
.
If p
4
4
18 / 21
Closed form for p
3
p
3
(x) =
2x
k=0
W
3
(2k)
_
x
3
_
2k
.
_

0
p
3
(x)x
s
dx =
2
s+2
3(s + 2)
+
2
s+4
3
3(s + 4)
+
Residues of W
3
5
.
If p
4
4
18 / 21
Series for p
4
Plot p
4
(x) for small x was best done from rst principles.
Instead of using
lim
h0
p
4
(x + h) p
4
(x)
h
,
I foolishly used
lim
h0
p
4
(x + h) p
4
(h)
x
.
Amazingly, they produced almost the same plot, except mine
was translated up by r 0.14.
This means p
4
almost satises the dierential equation
f
(x) + r =
f(x)
x
,
Solution: f(x) = (a r log x)x, a 0.33, explaining the
double pole!
19 / 21
Series for p
4
Plot p
4
Instead of using
lim
h0
p
4
(x + h) p
4
(x)
h
,
I foolishly used
lim
h0
p
4
(x + h) p
4
(h)
x
.
This means p
4
f
(x) + r =
f(x)
x
,
double pole!
19 / 21
Series for p
4
Plot p
4
Instead of using
lim
h0
p
4
(x + h) p
4
(x)
h
,
I foolishly used
lim
h0
p
4
(x + h) p
4
(h)
x
.
This means p
4
f
(x) + r =
f(x)
x
,
double pole!
19 / 21
Series for p
4
Plot p
4
Instead of using
lim
h0
p
4
(x + h) p
4
(x)
h
,
I foolishly used
lim
h0
p
4
(x + h) p
4
(h)
x
.
This means p
4
f
(x) + r =
f(x)
x
,
double pole!
19 / 21
Series for p
4
Plot p
4
Instead of using
lim
h0
p
4
(x + h) p
4
(x)
h
,
I foolishly used
lim
h0
p
4
(x + h) p
4
(h)
x
.
This means p
4
f
(x) + r =
f(x)
x
,
double pole!
19 / 21
Closed form for p
4
To be consistent, we must have:
p
4
(x) =
n=1
_
a
4
(n) r
4
(n) log x
_
x
2n1
,
a
4
(n): residues at 2n; r
4
(n): coecients of the double pole.
Guessed that p
4
satises a DE, shared by the g.f. for W
4
(2k)
(c.f. p
3
), and is a solution with a logarithmic singularity.
DE rigorously produced by Mellin transform, PDE regularity,
and a Gosper type algorithm.
More work on modular forms:
Theorem (3) Borwein, Straub, W., Zudilin (2010)
p
4
(x) =
2
16 x
2
2
x
Re
3
F
2
_
1
2
,
1
2
,
1
2
5
6
,
7
6
(16 x
2
)
3
108x
4
_
.
20 / 21
Closed form for p
4
p
4
(x) =
n=1
_
a
4
(n) r
4
(n) log x
_
x
2n1
,
a
4
4
Guessed that p
4
4
(2k)
(c.f. p
3
p
4
(x) =
2
16 x
2
2
x
Re
3
F
2
_
1
2
,
1
2
,
1
2
5
6
,
7
6
(16 x
2
)
3
108x
4
_
.
20 / 21
Closed form for p
4
p
4
(x) =
n=1
_
a
4
(n) r
4
(n) log x
_
x
2n1
,
a
4
4
Guessed that p
4
4
(2k)
(c.f. p
3
p
4
(x) =
2
16 x
2
2
x
Re
3
F
2
_
1
2
,
1
2
,
1
2
5
6
,
7
6
(16 x
2
)
3
108x
4
_
.
20 / 21
Closed form for p
4
p
4
(x) =
n=1
_
a
4
(n) r
4
(n) log x
_
x
2n1
,
a
4
4
Guessed that p
4
4
(2k)
(c.f. p
3
p
4
(x) =
2
16 x
2
2
x
Re
3
F
2
_
1
2
,
1
2
,
1
2
5
6
,
7
6
(16 x
2
)
3
108x
4
_
.
20 / 21
Science is what we understand well enough to explain to a
computer. Art is everything else we do. Donald Knuth
Mathematics is much less formally complete and precise than
computer programs. William Thurston
Thank you!
J. M. Borwein, D. Nuyens, A. Straub, J. Wan
Some arithmetic properties of short random walk integrals.
Ramanujan Journal, 26, (2011), 109132.
J. M. Borwein, A. Straub, J. Wan
Three-step and four-step random walk integrals. Experimental
Mathematics, 22, (2013), 114.
J. M. Borwein, A. Straub, J. Wan, W. Zudilin (& D. Zagier)
Densities of short uniform random walks. Canadian Journal of
Mathematics, 64, (2012), 961990.
21 / 21
Science is what we understand well enough to explain to a
computer. Art is everything else we do. Donald Knuth
Mathematics is much less formally complete and precise than
computer programs. William Thurston
Thank you!
J. M. Borwein, D. Nuyens, A. Straub, J. Wan
Some arithmetic properties of short random walk integrals.
Ramanujan Journal, 26, (2011), 109132.
J. M. Borwein, A. Straub, J. Wan
Three-step and four-step random walk integrals. Experimental
Mathematics, 22, (2013), 114.
J. M. Borwein, A. Straub, J. Wan, W. Zudilin (& D. Zagier)
Densities of short uniform random walks. Canadian Journal of
Mathematics, 64, (2012), 961990.
21 / 21

Uniform Random Walks On The Plane: A Case Study in Experimental Mathematics

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Uniform Random Walks On The Plane: A Case Study in Experimental Mathematics

Uploaded by

Copyright:

Available Formats

Intro Random walk Densities Expectations 3 and 4 steps

Uniform Random Walks on the Plane

You might also like