Universal Record Statistics of Random Walks GGI Workshop in Advances - - PowerPoint PPT Presentation
Universal Record Statistics of Random Walks GGI Workshop in Advances - - PowerPoint PPT Presentation
Universal Record Statistics of Random Walks GGI Workshop in Advances in Non-Equilibrium Statistical Mechanics Grgory Schehr, LPTMS (Orsay) Universal Record Statistics of Random Walks GGI Workshop in Advances in Non-Equilibrium Statistical
Universal Record Statistics of Random Walks
GGI Workshop in Advances in Non-Equilibrium Statistical Mechanics
Grégory Schehr, LPTMS (Orsay)
Collaborators:
- C. Godrèche (IPhT, Saclay)
- S. N. Majumdar (LPTMS, Orsay)
- G. Wergen (Uni. of Cologne)
Statement of the problem
random variables (e.g. time series)
n i xi
Statement of the problem
random variables (e.g. time series) is a record iff
n i xi
Statement of the problem
random variables (e.g. time series) is a record iff
n i xi
Questions: Statistics of the number of records ?
Statement of the problem
random variables (e.g. time series) is a record iff Questions: Statistics of the number of records ?
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
Statistics of the ages of records ?
Some recent applications of records in physics
Avalanches Le Doussal & Wiese ‘09 Financial data Wergen, Bogner & Krug ‘11, Wergen ‘13 Domain wall dynamics Alessandro et al. ‘90 Spin-glasses Sibani ‘07 Global warming Redner & Petersen ‘06, Wergen & Krug ‘10 Evolutionary biology Jain & Krug ‘05 Growing networks Godrèche & Luck ‘08 Random walks Majumdar & Ziff ‘08, Wergen, Majumdar, G. S. ‘12
Some recent applications of records in physics
Avalanches Le Doussal & Wiese ‘09 Financial data Wergen, Bogner & Krug ‘11, Wergen ‘13 Domain wall dynamics Alessandro et al. ‘90 Spin-glasses Sibani ‘07 Global warming Redner & Petersen ‘06, Wergen & Krug ‘10 Evolutionary biology Jain & Krug ‘05 Growing networks Godrèche & Luck ‘08 Random walks Majumdar & Ziff ‘08, Wergen, Majumdar, G. S. ‘12
Record statistics of i.i.d. random variables
i.i.d. random variables with PDF Nber of records
if is a record if is NOT a record
y i xi n k
Record statistics of i.i.d. random variables
i.i.d. random variables with PDF Nber of records
if is a record if is NOT a record is the proba. that a record is broken at step
y i xi n k
Record statistics of i.i.d. random variables
i.i.d. random variables with PDF Nber of records
if is a record if is NOT a record is the proba. that a record is broken at step
y i xi n k
Record statistics of i.i.d. random variables
i.i.d. random variables with PDF Nber of records
if is a record if is NOT a record is the proba. that a record is broken at step
y i xi n k
Record statistics of i.i.d. random variables
i.i.d. random variables with PDF Nber of records
if is a record if is NOT a record is the proba. that a record is broken at step
Universal !
y i xi n k
Record statistics of i.i.d. random variables
i.i.d. random variables with PDF
y i xi n k
Record statistics of i.i.d. random variables
i.i.d. random variables with PDF Average nber of records
y i xi n k
Record statistics of i.i.d. random variables
i.i.d. random variables with PDF Average nber of records Variance
y i xi n k
Record statistics of i.i.d. random variables
i.i.d. random variables with PDF Average nber of records Variance Universal probability distribution
y i xi n k
Record statistics of i.i.d. random variables
i.i.d. random variables with PDF Average nber of records Variance Universal probability distribution
Stirling numbers: number of permutations of elements with disjoint cycles
y i xi n k
Record statistics of i.i.d. random variables
i.i.d. random variables with PDF Average nber of records Variance Universal probability distribution
Stirling numbers: number of permutations of elements with disjoint cycles
Gaussian for large
y i xi n k
Record statistics of random walks
where the jumps s are i.i.d. with PDF continuous & symmetric
Record statistics of random walks
where the jumps s are i.i.d. with PDF
Including Ordinary random walks
continuous & symmetric
Lévy flights is the Lévy index
Record statistics of random walks
where the jumps s are i.i.d. with PDF
Including Ordinary random walks
continuous & symmetric
Lévy flights is the Lévy index
Q: Dependence of records on the jump distribution ?
Mean record number of random walks
is the proba. that a record is broken at step
y i xi n k
Mean record number of random walks
is the proba. that a record is broken at step
y i xi n k y xi k y xi k ⇐ ⇒
Mean record number of random walks
is the proba. that a record is broken at step
y xi k y xi k ⇐ ⇒
- Proba. that the walker stays negative up to step
starting from the origin
Mean record number of random walks
is the proba. that a record is broken at step
y xi k y xi k ⇐ ⇒
- Proba. that the walker stays negative up to step
starting from the origin is given by the Sparre Andersen Theorem
is given by the Sparre Andersen Theorem
Mean record number of random walks
is given by the Sparre Andersen Theorem For symmetric RW
Mean record number of random walks
is given by the Sparre Andersen Theorem For symmetric RW
Universal, i.e. independent
- f the jump distribution !
Mean record number of random walks
is given by the Sparre Andersen Theorem For symmetric RW
Universal, i.e. independent
- f the jump distribution !
Mean record number of random walks
Majumdar, Ziff `08
Record statistics of random walks with a drift
where the jumps s are i.i.d. with PDF continuous & symmetric
RW with a drift
Record statistics of random walks with a drift
where the jumps s are i.i.d. with PDF continuous & symmetric
RW with a drift Mean number of records of :
Record statistics of random walks with a drift
where the jumps s are i.i.d. with PDF continuous & symmetric
RW with a drift Mean number of records of : (Generalized) Sparre Andersen theorem
Record statistics of random walks with a drift
RW with a drift ,
c µ
1 2
I III IV V II
Majumdar, G. S., Wergen `12
Record statistics of random walks with a drift
RW with a drift ,
c µ
1 2
I III IV V II
Majumdar, G. S., Wergen `12
Record statistics of random walks with a drift
RW with a drift ,
c µ
1 2
I III IV V II
Majumdar, G. S., Wergen `12
Record statistics of random walks with a drift
RW with a drift ,
c µ
1 2
I III IV V II
Majumdar, G. S., Wergen `12
Record statistics of random walks with a drift
RW with a drift ,
c µ
1 2
I III IV V II
Majumdar, G. S., Wergen `12
Record statistics of random walks with a drift
RW with a drift ,
c µ
1 2
I III IV V II
Majumdar, G. S., Wergen `12
Record statistics of random walks with a drift
RW with a drift ,
c µ
1 2
I III IV V II
What about the full distribution of ?
Majumdar, G. S., Wergen `12
Renewal approach to records of RW
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
Joint distribution of ?
Renewal approach to records of RW
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
Joint distribution of ? RW is a Markov process independent except for the global constraint are
Renewal approach to records of RW
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
Joint distribution of ? RW is a Markov process independent except for the global constraint are are identical RW is translationally invariant s while has different statistics
Two main objects: Persistence (or survival) probability
- indep. of
Distribution of first-passage time (from below) Joint distribution of ?
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
Renewal approach to records of RW
- indep. of
Joint distribution of
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
Renewal approach to records of RW
survival proba. first passage proba.
- Proba. distribution of the number of records
with
- Proba. distribution of the number of records
with Generating function w.r.t. the number of steps
- Proba. distribution of the number of records
with Generating function w.r.t. the number of steps
(for symmetric jumps)
- Proba. distribution of the number of records
By ``inverting’’ the GF (for symmetric jumps):
Majumdar, Ziff `08
- Proba. distribution of the number of records
By ``inverting’’ the GF (for symmetric jumps): For :
Majumdar, Ziff `08
- Proba. distribution of the number of records
By ``inverting’’ the GF (for symmetric jumps): For : RW with a drift
c µ
1 2
I III IV V II
Majumdar, Ziff `08 Majumdar, G. S., Wergen `12
- Proba. distribution of the number of records
By ``inverting’’ the GF (for symmetric jumps): For : RW with a drift
c µ
1 2
I III IV V II
Majumdar, Ziff `08 Majumdar, G. S., Wergen `12
- Proba. distribution of the number of records
By ``inverting’’ the GF (for symmetric jumps): For : RW with a drift
c µ
1 2
I III IV V II
e.g. for
Majumdar, Ziff `08 Majumdar, G. S., Wergen `12
Statistics of the ages of records
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
- sym. RW
Statistics of the ages of records
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
Typical age of a record:
- sym. RW
Statistics of the ages of records
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
Typical age of a record:
- sym. RW
What about the longest or shortest age of a record ?
Statistics of the ages of records
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
Typical age of a record:
- sym. RW
What about the longest or shortest age of a record ? What is the proba. that the current record is the oldest one ?
Godrèche, Majumdar, G. S., `14
Statistics of the ages of records
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
Typical age of a record:
- sym. RW
What about the longest or shortest age of a record ? What is the proba. that the current record is the oldest one ?
Godrèche, Majumdar, G. S., `14
Statistics of the ages of records
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
- sym. RW
= ?
Statistics of the ages of records
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
- sym. RW
= ?
Statistics of the ages of records
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
- sym. RW
= ?
Statistics of the ages of records
Statistics of the ages of records
Statistics of the ages of records
Statistics of the ages of records
with
Statistics of the ages of records
with
Generating function
Statistics of the ages of records
with
Generating function
Statistics of the ages of records
Statistics of the ages of records For symmetric RW
Statistics of the ages of records For symmetric RW One finds
Statistics of the ages of records
0.625 0.626 0.627 0.628 10 20 30 40 50 60 70 80 90 100
Q (n) n 0.626508...
Godrèche, Majumdar, G. S.,`14
Statistics of the ages of records
0.625 0.626 0.627 0.628 10 20 30 40 50 60 70 80 90 100
Q (n) n 0.626508...
Godrèche, Majumdar, G. S.,`14
Statistics of the ages of records
0.625 0.626 0.627 0.628 10 20 30 40 50 60 70 80 90 100
Q (n) n 0.626508...
Godrèche, Majumdar, G. S.,`14 Pitman, Yor, `97 Godrèche, Majumdar, G. S., `09
New observable...new universal constant
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
New observable...new universal constant
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
New observable...new universal constant
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
New observable...new universal constant
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6
n
Godrèche, Majumdar, G. S.,`14
Conclusions
Exact results for records of strongly correlated time series Universal records statistics for (symmetric) RWs Extension to multiparticle systems
Wergen, Majumdar, G. S. `12
Extension to Continuous Time Random Walks (CTRWs)
- S. Sabhapandit `12
see arXiv:1305.0639 for a short review
Conclusions
Exact results for records of strongly correlated time series Universal records statistics for (symmetric) RWs Extension to multiparticle systems
Wergen, Majumdar, G. S. `12
Extension to Continuous Time Random Walks (CTRWs)
- S. Sabhapandit `12
High sensitivity to the definition of the age of the last record
see arXiv:1305.0639 for a short review
Sensitivity to the definition of the age of the last record
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6 τ6 = 7
n
Godrèche, Majumdar, G. S., `14
Sensitivity to the definition of the age of the last record
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6 τ6 = 7
n
Godrèche, Majumdar, G. S., `14
Sensitivity to the definition of the age of the last record
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6 τ6 = 7
n
Godrèche, Majumdar, G. S., `14
Sensitivity to the definition of the age of the last record
An = 3
i xi
τ1 = 4 τ2 = 3 τ3 = 2 τ4 = 5 τ5 = 6 τ6 = 7