What is your subitizing range?
(and other numerical games)
40
34
(4,0)
1/16
(3,1)
1/4
(2,2)
3/8
(1,3)
1/4
(0,4)
1/16
State
Probability
Some simple mathematics
W (number of ways)
P(nR) =
N!
-----------nL! nR!
2 -N W
This is the binomial distribution, and it is not hard to show that:
<nR> = N/2
and
(nR) = N1/2 / 2
Observation count:
N=500 000
NR=0
NR=1
NR=2
NR=3
NR=4
26386
122844
198143
126360
26267
0.053
0.245
0.396
0.253
0.052
0.06
0.25
0.38
0.25
0.06
0.062
0.253
0.371
0.249
0.062
0.06
0.25
0.38
0.25
0.06
N=5 000 000
NR=0
NR=1
NR=2
NR=3
NR=4
314123
1267215
1857280
1246962
314420
Fifty fish in the tank:
State
(50,0)
(40,10)
(30,20)
(25,25)
W
1
1.027x1010
4.713x1013
1.264x1014
Probability
9x10 -16
9x10-6
0.0418
0.1123
Fifty fish in the tank:
State
(50,0)
(40,10)
(30,20)
(25,25)
W
1
1.027x1010
4.713x1013
1.264x1014
>40,<10
>45,<5
>30,<20
Between 20 and 30
Probability
9x10 -16
9x10-6
0.0418
0.1123
5.6x10 -6
4.5x10 -10
0.119
0.881
More trips to the pet store leads to problems with the factorial button
Stirlings Approximation:
N! NN e-N (2N)1/2 [1 + 1/12N + ] (asymptotic expansion)
N! NN e-N (2N)1/2
ln(N!)
(simplified form, SA1)
N ln (N) - N
N
N!
5
120
10 3628800
50 3.041x1014
(convenient form, SA2)
SA1
SA2
119.99
21.05
3598695 453999
3.039x1014 1.71x1013
Using SA1, it is fairly easy to show (but not now) that:
P(config) = exp{ N ln (N) - nR ln(nR) - nL ln(nL)
- 1/2 ln( 2 NR NL / N) - N ln (2) }
Fifty fish in the tank:
State
(50,0)
(40,10)
(30,20)
(25,25)
W
1
1.027x1010
4.713x1013
1.264x1014
>40,<10
>45,<5
>30,<20
Between 20 and 30
Probability
5x10 -14
3x10-16
9x10-6
9x10-6
0.0418
0.0420
0.1123
0.1128
5.6x10 -6 5.7x10-6
4.5x10 -10 4.5x10-10
0.119
0.119
0.881
0.881
Fifty fish in the tank:
State
(50,0)
(40,10)
(30,20)
(25,25)
W
1
1.027x1010
4.713x1013
1.264x1014
>40,<10
>45,<5
>30,<20
Between 20 and 30
Probability
5x10 -14
4x10-14
9x10-6
9x10-6
0.0418
0.0420
0.1123
0.1128
5.6x10 -6 5.7x10-6
4.5x10 -10 4.5x10-10
0.119
0.119
0.881
0.881
Good enough for 50, necessary
for a lot more
One thousand fish in the tank:
P(500,500)= 0.025
One thousand fish in the tank:
P(500,500)= 0.025
P(1000,0) = 0.0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000000
0000000000000000000000000000000000000000000000020
aka 2.024x10-298
A rather unlikely event, it would seem
Perhaps a better question is:
What is the chance that the numbers on each side of the tank differ by some
relative amount?
Ex (490,510) => A fluctuation of 2% with respect to the mean (1:50)
(480,520) => A fluctuation of 4% with respect to the mean (1:25)
Fluctuation
> 1:10
> 1:25
> 1:50
> 1:100
> 1:500
Probability
0.001
0.195
0.507
0.728
0.975
Examination
Which side of the tank has more fish????
Which side of the tank has more fish????
999
1001
Which side of the tank has more fish????
Which side of the tank has more fish????
1010
990
Which side of the tank has more fish????
Which side of the tank has more fish????
980
1020
Which side of the tank has more fish????
980
1020
Which side of the tank has more fish????
Which side of the tank has more fish????
950
1050
Which side of the tank has more fish????
950
1050
A 10% Difference!
10% Difference (1:10)
15% Difference (1:6.7)
20% Difference (1:5)
10% Difference
15% Difference
20% Difference
Bumping - 1000 steps - 0.1 per frame - reflections boundaries
<xR> = 498.46
R = 11.76
Binomial distribution
If the probability of a given result is p, and the event is repeated N
times, then:
<n()> = p N
n() = [p(1-p)N]1/2
In this case: p refers to the probability of the particle (fish) being on the
right side, and is 0.5
So
Avg => N/2 (obvious)
Std. Dev => N1/2/2
N=1000, p=0.5 gives:
<n(left)> = 500 v. 498.5 (close)
n(left) = 11.2 v. 11.8 (close)
Number of points outside 1 =
2 =
3 =
713 (35.65 %)
97 ( 4.85 %)
1 ( 0.05 %)
Number of points outside 1 =
2 =
3 =
713 (35.65 %)
97 ( 4.85 %)
1 ( 0.01 %)
(31.73%)
( 4.55%)
( 0.03%)
Suggests that distribution is Gaussian!
The system is not ergodic
Particles are not sampling the entire region available to them during the time
of observation!
How does behavior change as we increase the number of steps?
10000 steps
20000 steps
50000 steps
A glass of orange soda contains about 1018 dye molecules
Fluctuation
Probability
> 1:50
10 -86843247937480
A glass of orange soda contains about 1018 dye molecules
Fluctuation
Probability
> 1:50
10 -86843247937480
2.5 million years ago!
A glass of orange soda contains about 1018 dye molecules
Fluctuation
Probability
> 1:50
10 -86843247937480
> 1:2500
10-34737299180
2.5 million years ago!
Since 922 A.D.
A glass of orange soda contains about 1018 dye molecules
Fluctuation
Probability
> 1:50
10 -86843247937480
> 1:2500
10-34737299180
> 1:500000
10
-868435
2.5 million years ago!
Since 922 A.D.
10 days ago
A glass of orange soda contains about 1018 dye molecules
Fluctuation
Probability
> 1:50
10-86843247937480
> 1:2500
10 -34737299180
> 1:500000
10
> 1:500 000 000
2.5 million years ago!
Since 922 A.D.
-868435
5x10-89
Suppose you observed the glass once each femtosecond*
Time required to see > 1:500 000 000 fluctuation:
5 x 1071 years
Age of universe:
Mass of universe:
Hydrogen atoms in univ:
ca. 1.3 x 1010 years
ca. 1050 kg
ca. 1077
10 days ago
Summary
Since (N)/<N> = N-1/2, all states except the statistical one occur with
zero probability in the thermodynamic limit (N
).
Any isolated system prepared with W < W max will evolve such that W
increases. Movement towards smaller W becomes successively less
probable as the number of particles in the system increases.
As W increases, so does ln(W), and therefore kB ln(W).
Boltzmann:
S = kB ln (W)
[second law of thermodynamics]