first-passage statistics of extreme values · nathan lemons (los alamos)! pearson miller (yale,...
TRANSCRIPT
![Page 1: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/1.jpg)
First-Passage Statistics of Extreme Values
Eli Ben-Naim Los Alamos National Laboratory
Talk, publications available @ http://cnls.lanl.gov/~ebn
Random Graph Processes, Austin TX, May 10, 2016
with: Paul Krapivsky (Boston University)
Nathan Lemons (Los Alamos) Pearson Miller (Yale, MIT)
![Page 2: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/2.jpg)
Plan
I. Motivation: records & their first-passage statistics as a data analysis tool
II. Incremental records: uncorrelated random variables
III.Ordered records: uncorrelated random variables
IV. Ordered records: correlated random variables
![Page 3: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/3.jpg)
I. Motivation: records & their first-passage
statistics as a data analysis tool
![Page 4: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/4.jpg)
Record & running record
• Record = largest variable in a series
!
• Running record = largest variable to date
!
• Independent and identically distributed variables
Statistics of extreme values
XN = max(x1, x2, . . . , xN )
Feller 68 Gumble 04
Ellis 05
Z 1
0dx �(x) = 1
X1 X2 · · · XN
![Page 5: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/5.jpg)
Average number of running records
• Probability that Nth variable sets a record
!
• Average number of records = harmonic number
!
• Grows logarithmically with number of variables
Behavior is independent of distribution function Number of records is quite small
MN = 1 +1
2+
1
3+ · · ·+ 1
N
MN ' lnN + � � = 0.577215
PN =1
N
![Page 6: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/6.jpg)
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010year
8.5
9
9.5
10M=8.5
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010year
88.59
9.510
M=8.0
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010year
7.58
8.59
9.510
M=7.5
Time series of massive earthquakes
Are massive earthquakes correlated?
Magnitude Annual #
9-9.9 1/20
8-8.9 1
7-7.9 15
6-6.9 134
5-5.9 1300
4-4.9 ~13,0003-3.9 ~130,000
2-2.9 ~1,300,000
1770 M>7 events 1900-2013
![Page 7: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/7.jpg)
100 101 102
N0123456789
MN
M>5 (37,000 events)M>7 (1,770 events)1+1/2+1/3+...+1/N
t1 t2 t3 t4
Records in inter-event time statistics
records indicate inter-event times uncorrelated massive earthquakes are random
Count number of running records in N consecutive events
attribute deviation to aftershocks
EB & Krapivsky PRE 13
![Page 8: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/8.jpg)
• Process by which a fluctuating quantity reaches a threshold for the first time
• First-passage probability: for the random variable to reach the threshold as a function of time.
• Total probability: that threshold is ever reached. May or may not equal 1
• First-passage time: the mean duration of the first-passage process. Can be finite or infinite
First-passage processes
S. Redner, A Guide to First-Passage Processes, 2001
![Page 9: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/9.jpg)
II. Incremental records: uncorrelated random variables
![Page 10: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/10.jpg)
Marathon world record
Incremental sequence of records
every record improves upon previous record by yet smaller amount
Are incremental sequences of records common?
Year Athlete Country Record Improvement
2002 Khalid Khannuchi USA 2:05:38
2003 Paul Tergat Kenya 2:04:55 0:43
2007 Haile Gebrsellasie Ethiopia 2:04:26 0:29
2008 Haile Gebrsellasie Ethiopia 2:03:59 0:27
2011 Patrick Mackau Kenya 2:03:38 0:21
2013 Wilson Kipsang Kenya 2:03:23 0:15
source: wikipedia
![Page 11: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/11.jpg)
Incremental records
Incremental sequence of records
every record improves upon previous record by yet smaller amount
y1
y2
y3y4
random variable = {0.4, 0.4, 0.6, 0.7, 0.5, 0.1}latest record = {0.4, 0.4, 0.6, 0.7, 0.7, 0.7} �
latest increment = {0.4, 0.4, 0.2, 0.1, 0.1, 0.1} ⇥
What is the probability all records are incremental?
![Page 12: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/12.jpg)
100 101 102 103 104 105 106 107 108N
10-3
10-2
10-1
100
S
simulationN-0.317621
Power law decay with nontrivial exponent Problem is parameter-free
SN ⇠ N�� � = 0.31762101
Miller & EB JSTAT 13
100 101 102 103
N10-1
100
S
N-0.3
Earthquake data
Probability all records are incremental
![Page 13: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/13.jpg)
Uniform distribution
• The variable x is randomly distributed in [0:1] !
• Probability record is smaller than x
• Average record
�(x) = 1 for 0 x 1
1/(N + 1) / 1/N
Distribution of records is purely exponential
AN =N
N + 1=) 1�AN ' N�1
RN (x) = x
N
![Page 14: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/14.jpg)
Distribution of records• Probability a sequence is incremental and record < x
!
• One variable !
• Two variables !
• In general, conditions are scale invariant
• Distribution of records for incremental sequences
!
• Distribution of records for all sequences equals
Statistics of records are standard
GN (x) =) SN = GN (1)
G1(x) = x =) S1 = 1
G2(x) =3
4x
2 =) S2 =3
4
GN (x) = SN x
N
x ! a x
x
N
x2 = x1
x2 = 2x1
x
x
x/2
Fisher-Tippett 28 Gumbel 35
x2 � x1 > x1
S1 = 1
S2 = 3/4
S3 = 47/72
![Page 15: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/15.jpg)
0 0.2 0.4 0.6 0.8 1x0
0.2
0.4
0.6
0.8
1
GN(x)/G
N(1)
N=1N=2N=3N=4xx2
x3
x4
Scaling behavior
• Distribution of records for incremental sequences
!
• Scaling variable
Exponential scaling function
GN (x)/SN = x
N = [1� (1� x)]N ! e
�s
s = (1� x)N
![Page 16: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/16.jpg)
Distribution of increment+records• Probability density SN(x,y)dxdy that:
1. Sequence is incremental 2. Current record is in range (x,x+dx) 3. Latest increment is in range (y,y+dy) with 0<y<x
• Gives the probability a sequence is incremental
!
• Recursion equation incorporates memory
!
!
• Evolution equation includes integral, has memoryold record holds a new record is set
S
N+1(x, y) = xS
N
(x, y) +
Zx�y
y
dy
0S
N
(x� y, y
0)
�SN
(x, y)
�N= �(1� x)S
N
(x, y) +
Zx�y
y
dy0 SN
(x� y, y0)
S
N
=
Z 1
0dx
Zx
0dy S
N
(x, y)
![Page 17: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/17.jpg)
Scaling transformation• Assume record and increment scale similarly
• Introduce a scaling variable for the increment
!
• Seek a scaling solution
!
• Eliminate time out of the master equation
s = (1� x)N and z = yN
y ⇠ 1� x ⇠ N
�1
SN (x, y) = N
2SN �(s, z)
✓2� � + s+ s
⇥
⇥s+ z
⇥
⇥z
◆�(s, z) =
Z 1
zdz0 �(s+ z, z0)
Reduce problem from three variables to two
![Page 18: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/18.jpg)
Factorizing solution• Assume record and increment decouple
• Substitute into equation for similarity solution
!
• First order integro-differential equation
!
• Cumulative distribution of scaled increment
• Convert into a second order differential equation
zf 0(z) + (2� �)f(z) = e�z
Z 1
zf(z0)dz0
✓2� � + s+ s
⇥
⇥s+ z
⇥
⇥z
◆�(s, z) =
Z 1
zdz0 �(s+ z, z0)
�(s, z) = e�s f(z)
g(0) = 1
g0(0) = �1/(2� �)zg00(z) + (2� �)g0(z) + e�zg(z) = 0
g(z) =
Z 1
zf(z0)dz0
Reduce problem from two variable to one
![Page 19: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/19.jpg)
Distribution of increment• Assume record and increment decouple
• Two independent solutions
!
• The exponent is determined by the tail behavior
!
• The distribution of increment has a broad tail
g(0) = 1
g0(0) = �1/(2� �)zg00(z) + (2� �)g0(z) + e�zg(z) = 0
g(z) = z⌫�1and g(z) = const. as z ! 1
PN (y) ⇠ N�1y��2
Increments can be relatively large problem reduced to second order ODE
� = 0.317621014462...
![Page 20: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/20.jpg)
0 2 4 6 8 10z0
0.2
0.4
0.6
0.8
1
g
theorysimulation
Numerical confirmation
Increment and record become uncorrelated
hszihsihzi ! 1
Monte Carlo simulation versus integration of ODE
g(0) = 1
g0(0) = �1/(2� �)
![Page 21: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/21.jpg)
Recap II
• Probability a sequence of records is incremental
• Linear evolution equations (but with memory)
• Dynamic formulation: treat sequence length as time
• Similarity solutions for distribution of records
• Probability a sequence of records is incremental decays as power-law with sequence length
• Power-law exponent is nontrivial, obtained analytically
• Distribution of record increments is broad
First-passage properties of extreme values are interesting
![Page 22: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/22.jpg)
III. Ordered records:
uncorrelated random variables
![Page 23: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/23.jpg)
xnyn
n
XnYn
Ordered records• Motivation: temperature records:
Record high increasing each year
• Two sequences of random variables
!
• Independent and identically distributed variables
• Two corresponding sequences of records
!
• Probability SN records maintain perfect order
{X1, X2, . . . , XN} and {Y1, Y2, . . . , YN}
xn = max{X1, X2, . . . , Xn} and yn = max{Y1, Y2, . . . , Yn}
x1 > y1 and x2 > y2 · · · and xN > yN
![Page 24: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/24.jpg)
Two sequences
• Survival probability obeys closed recursion equation
!
• Solution is immediate
!
• Large-N: Power-law decay with rational exponent
SN = SN�1
✓1� 1
2N
◆
SN =
✓2N
N
◆2�2N
SN ' ⇡�1/2 N�1/2
Universal behavior: independent of parent distribution!
identical to Sparre Andersen 53!
![Page 25: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/25.jpg)
Ordered random variables• Probability PN variables are always ordered
!
• Exponential decay
!
• Ordered records far more likely than ordered variables!
• Variables are uncorrelated
• Records are strongly correlated: each record “remembers” entire preceding sequence
X1 > Y1 and X2 > Y2 · · · and XN > YN
PN = 2�N
Ordered records better suited for data analysis
![Page 26: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/26.jpg)
xnynzn
n
XnYnZn
• Third sequence of random variables
!
• Probability SN records maintain perfect order
• Power-law decay with nontrivial exponent?
100 101 102 103 104 105 106 107 108N
10-1110-1010-910-810-710-610-510-410-310-210-1100
SN
simulationtheory
101 102 103 104 105
N1.26
1.28
1.30
1.32
σ
Three sequences
SN ⇠ N�� with � = 1.3029
xn > yn > zn n = 1, 2, . . . , N
![Page 27: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/27.jpg)
Rank of median record
• Closed equations for survival probability not feasible
• Focus on rank of the median record
• Rank of the trailing record irrelevant
• Joint probability PN,j that (i) records are ordered and (ii) rank of the median record equals j
• Joint probability gives the survival probability
leader median laggard | {z }
j
SN =NX
j=1
PN,j
![Page 28: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/28.jpg)
Closed recursion equations
•Closed recursion equations for joint probability feasible
!
!
!
•The survival probability for small N
| {z }j
PN+1,j =3N + 2� j
3N + 3
3N + 1� j
3N + 2
3N � j
3N + 1PN,j
+3N + 2� j
3N + 3
3N + 1� j
3N + 2
j
3N + 1PN,j�1
+3N + 2� j
(3N + 3)(3N + 2)(3N + 1)
N+1X
k=j
(3N � k)PN,k
+3N + 2� j
(3N + 3)(3N + 2)(3N + 1)
N+1X
k=j
k PN,k�1
N SN (3N)!SN
1 16 1
2 29360 58
3 4 59790 720 18 388
4 5 393149 688 17 257 600
5 179 828 1836 538 371 840 35 965 636 600
no new records
two new records
new leading records
new median records
O(N0)
O(N�1)
O(N�1)
O(N�2)
![Page 29: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/29.jpg)
0 1 2 3 4 5 6 7 8 9 10j
0
0.1
0.2
0.3
0.4
0.5
N=102
N=103
N=104
Key observation
Rank of median record j and N become uncorrelated!
Pj,N
SN
![Page 30: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/30.jpg)
Scaling exponent is an eigenvalue
• Rank of median record j and N become uncorrelated!
!
• Assume power law decay for the survival probability
!
• The asymptotic rank distribution is normalized
!
• Rank distribution obeys a much simpler recursion
Asymptotic analysis
� pj = (j + 1) pj �j
3pj�1 �
1
3
1X
k=j
pk
1X
j=1
pj = 1
�
SN ⇠ N��
PN,j ' SN pj as N ! 1
![Page 31: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/31.jpg)
• First-order differential equation for generating function
!
• Solution
!
• Behavior near z=3 gives tail of the distribution
!
• Behavior near z=1 gives the scaling exponent
The rank distribution
(3� z)dP (z)
dz+ P (z)
✓1
1� z� 3�
z
◆=
z
1� z
P (z) =
r1� z
3� z
✓z
3� z
◆� Z z
0
du
(1� u)3/2(3� u)��1/2
u��1
P (z) =X
j�1
pj zj+1
pj ⇠ j��1/23�j
2F1
�� 1
2 ,12 � �; 3
2 � �;� 12
�= 0 =) � = 1.302931 . . .
Three sequences: scaling exponent is nontrivial
EB & Krapivsky PRE 15
![Page 32: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/32.jpg)
1 2 3 4 5 6 7 8m0
1
2
3
4
5
6
σ
• Probability SN that m records maintain perfect order
• Expect power-law decay with m-dependent exponent
!
• Lower and upper bounds
!
• Exponent grows linearly with number of sequences (up to a possible logarithmic correction)
Multiple sequences
In general, scaling exponent is nontrivial
SN ⇠ N��m
0 m� �m 1 +1
2+
1
3+ · · ·+ 1
m
�m ' m
![Page 33: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/33.jpg)
Family of ordering exponents
0 1 2 3 4 5 6m0
1
2
3
4
5 δ γ β α σ
•One sequence always in the lead: 1>rest !
•Two sequences always in the lead: 1>2>rest !
•Three sequences always in the lead: 1>2>3>rest
AN ⇠ N�↵m ↵m = 1� 1
m
BN ⇠ N��m2F1
✓� 1
m� 1,m� 2
m� 1� �;
2m� 3
m� 1� �;� 1
m� 1
◆= 0
CN ⇠ N��m
m ↵ � � �1 02 1/2 1/23 2/3 1.302931 1.3029314 3/4 1.56479 2.255 2.2555 4/5 1.69144 2.547 3.246 5/6 1.76164 2.680 3.53
![Page 34: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/34.jpg)
Recap III
• Probability multiple sequences of records are ordered
• Uncorrelated random variables
• Survival probability independent of parent distribution
• Power-law decay with nontrivial exponent
• Exact solution for three sequences
• Scaling exponent grows linearly with number of sequences
• Key to solution: statistics of median record becomes independent of the sequence length (large N limit)
• Scaling methods allow us to tackle combinatorics
![Page 35: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/35.jpg)
IV. Ordered records: correlated random variables
![Page 36: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/36.jpg)
x1x2
t
x1 x2 m1m2
t
Brownian recordsBrownian positions
![Page 37: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/37.jpg)
x1x2
t
• Universal probability
!
• Asymptotic behavior
First-passage kinetics: brownian positions
Universal first-passage exponent
S ⇠ t�1/2
St =
✓2t
t
◆2�t
Sparre Andersen 53
Behavior holds for Levy flights, different mobilities, etc
Probability two Brownian particle do not meet
S. Redner, A guide to First-Passage Processes 2001
Feller 68
![Page 38: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/38.jpg)
x1 x2 m1m2
t
First-passage kinetics: brownian records
Is ¼ exact? Is exponent universal? S ⇠ t�� � = 0.2503± 0.0005
100 101 102 103 104 105 106 107 108t
10-2
10-1
100
P
simulationt-1/4
Probability running records remain ordered
![Page 39: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/39.jpg)
x1 x2 m1m2
t
m1 > m2 if and only if m1 > x2
![Page 40: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/40.jpg)
From four variables to three
• Four variables: two positions, two records
!
• The two records must always be ordered
!
• Key observation: trailing record is irrelevant!
!
• Three variables: two positions, one record
m1 > x1 and m2 > x2
m1 > m2
m1 > m2 if and only if m1 > x2
m1 > x1 and m1 > x2
![Page 41: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/41.jpg)
From three variables to two• Introduce two distances from the record
!
• Both distances undergo Brownian motion
!
• Boundary conditions: (i) absorption (ii) advection
!
• Probability records remain ordered
u = m1 � x1 and v = m1 � x2
P (t) =
Z 1
0
Z 1
0du dv �(u, v, t)
⇥�(u, v, t)
⇥t= Dr2�(u, v, t)
���v=0
= 0 and
✓⇥�
⇥u�⇥�
⇥v
◆ ���u=0
= 0
![Page 42: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/42.jpg)
u
v
Diffusion in corner geometry
![Page 43: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/43.jpg)
“Backward” evolution• Study evolution as function of initial conditions
!
• Obeys diffusion equation
!
• Boundary conditions: (i) absorption (ii) advection
!
• Advection boundary condition is conjugate!
P ⌘ P (u0, v0, t)
�P (u0, v0, t)
�t= Dr2P (u0, v0, t)
P��v0=0
= 0 and
✓�P
�u0+�P
�v0
◆ ���u0=0
= 0
![Page 44: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/44.jpg)
Solution
• Use polar coordinates
!
• Laplace operator
!
• Boundary conditions: (i) absorption (ii) advection
!
• Dimensional analysis + power law + separable form
r =q
u20 + v20 and � = arctan
v0u0
P���=0
= 0 and
✓r⇥P
⇥r�⇥P
⇥�
◆ ����=⇥/2
= 0
P (r, �, t) ⇠✓
r2
Dt
◆�
f(�)
r2 =⇥2
⇥r2+
1
r
⇥
⇥r+
1
r2⇥2
⇥�2
![Page 45: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/45.jpg)
Selection of exponent
• Exponent related to eigenvalue of angular part of Laplacian
!
• Absorbing boundary condition selects solution
!
• Advection boundary condition selects exponent
!
• First-passage probability
f 00(⇥) + (2�)2f(⇥) = 0
f(⇥) = sin (2�⇥)
tan (�⇡) = 1
P ⇠ t�1/4
EB & Krapivsky PRL 14
![Page 46: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/46.jpg)
• Particles have diffusion constants D1 and D2
!
!
• Condition on records involves ratio of mobilities !
!
• Analysis straightforward to repeat
!
• First-passage exponent: nonuniversal, mobility-dependent
General diffusivities
(x1, x2) ! (bx1, bx2) with (bx1, bx2) =
✓x1pD1
,
x2pD2
◆
rD1
D2bm1 > bm2
rD1
D2tan (�⇥) = 1
� =1
⇥arctan
rD2
D1
ben Avraham Leyvraz 88
![Page 47: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/47.jpg)
0 0.5 1 1.5 2 2.5 3D1/D2
0
0.1
0.2
0.3
0.4
0.5
β
simulationtheory
Numerical verification
perfect agreement
![Page 48: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/48.jpg)
Properties• Depends on ratio of diffusion constants
!
• Bounds: involve one immobile particle
!
• Rational for special values of diffusion constants
!
• Duality: between “fast chasing slow” and “slow chasing fast”
�(0) =1
2�(1) = 0
�(D1, D2) ⌘ �
✓D1
D2
◆
�
✓D1
D2
◆+ �
✓D2
D1
◆=
1
2
�(1/3) = 1/3 �(1) = 1/4 �(3) = 1/6
Alternating kinetics: slow-fast-slow-fast
![Page 49: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/49.jpg)
!
• Probability n Brownian positions are perfectly ordered
!
• Records perfectly ordered
!
• In general, power-law decay
Multiple particles
Fisher & Huse 88
m1 > m2 > m3 > · · · > mn
Sn ⇠ t�⌫n
Pn ⇠ t�↵n ↵n =n(n� 1)
4
n ⌫n �n/22 1/4 1/43 0.653 0.6514654 1.13 1.1285 1.60 1.626 2.01 2.10
Uncorrelated variables provide an excellent approximation Suggests some record statistics can be robust
![Page 50: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/50.jpg)
Recap IV• First-passage kinetics of extremes in Brownian motion
• Problem reduces to diffusion in a two-dimensional corner with mixed boundary conditions
• First-passage exponent obtained analytically
• Exponent is continuously varying function of mobilities
• Relaxation is generally slower compared with positions
• Open questions: multiple particles, higher dimensions
• Why do uncorrelated variables represent an excellent approximation?
![Page 51: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/51.jpg)
First-passage statistics of extreme values
• Survival probabilities decay as power law
• First-passage exponents are nontrivial
• Theoretical approach: differs from question to question
• Concepts of nonequilibrium statistical mechanics are powerful: scaling, correlations, large system-size limit
• Many, many open questions
• Ordered records as a data analysis tool
![Page 52: First-Passage Statistics of Extreme Values · Nathan Lemons (Los Alamos)! Pearson Miller (Yale, MIT) Plan I. Motivation: records & their first-passage statistics as a ... 0.2 0.4](https://reader036.vdocuments.site/reader036/viewer/2022071216/60469bfbc67b613d6e4973df/html5/thumbnails/52.jpg)
Publications1. Scaling Exponents for Ordered Maxima,
E. Ben-Naim and P.L. Krapivsky, Phys. Rev. E 92, 062139 (2015)
2. Slow Kinetics of Brownian Maxima, E. Ben-Naim and P.L. Krapivsky, Phys. Rev. Lett. 113, 030604 (2014)
3. Persistence of Random Walk Records, E. Ben-Naim and P.L. Krapivsky, J. Phys. A 47, 255002 (2014)
4. Scaling Exponent for Incremental Records, P.W. Miller and E. Ben-Naim, J. Stat. Mech. P10025 (2013)
5. Statistics of Superior Records, E. Ben-Naim and P.L. Krapivsky, Phys. Rev. E 88, 022145 (2013)