generalized ramanujan primes - williams college...introduction c-ramanujan primes results...

Introduction c-Ramanujan primes Results Distribution Conclusion

Generalized Ramanujan Primes

Nadine Amersi, Olivia Beckwith, Ryan Ronan

Advisor: Steven J. Miller

http://web.williams.edu/Mathematics/sjmiller/

Young Mathematicians ConferenceOhio State, August 2011

1


Prime Numbers

Any integer can be written as a unique product ofprime numbers (Fundamental Theorem of Arithmetic).

Exact spacing of primes unknown but roughly growinglogarithmically (Prime Number Theorem).

We expect linearly increasing intervals to contain anincreasing amount of primes.

2


Prime Numbers




3


Prime Numbers




4


Historical Introduction

Bertrand’s Postulate (1845)For all integers x ≥ 2, there exists at least one prime in(x/2, x ].

5


Ramanujan Primes

DefinitionThe n-th Ramanujan prime Rn: smallest integer such thatfor any x ≥ Rn, at least n primes in any (x/2, x ].

TheoremRamanujan: For each integer n, Rn exists.Sondow: Rn ∼ p2n.Sondow: As n→∞, 50% of primes are Ramanujan.

6


Ramanujan Primes


TheoremRamanujan: For each integer n, Rn exists.

Sondow: Rn ∼ p2n.Sondow: As n→∞, 50% of primes are Ramanujan.

7


Ramanujan Primes


TheoremRamanujan: For each integer n, Rn exists.Sondow: Rn ∼ p2n.

Sondow: As n→∞, 50% of primes are Ramanujan.

8


Ramanujan Primes


TheoremRamanujan: For each integer n, Rn exists.Sondow: Rn ∼ p2n.Sondow: As n→∞, 50% of primes are Ramanujan.

9


c-Ramanujan Primes

DefinitionThe n-th c-Ramanujan prime Rc,n: smallest integer suchthat for any x ≥ Rc,n have at least n primes in (cx , x ].

For each c and integer n, does Rc,n exist? Yes!

10


c-Ramanujan Primes


For each c and integer n, does Rc,n exist?

Yes!

11


c-Ramanujan Primes


For each c and integer n, does Rc,n exist? Yes!

12


c-Ramanujan Primes


Does Rc,n exhibit asymptotic behavior?

For c → 0, the interval is (0, x ] and Rc,n = pn.For c → 1, the interval would be (x , x ] so Rc,n doesnot exist.

Rc,n ∼ p n1−c

13


c-Ramanujan Primes



For c → 0, the interval is (0, x ] and Rc,n = pn.

For c → 1, the interval would be (x , x ] so Rc,n doesnot exist.

Rc,n ∼ p n1−c

14


c-Ramanujan Primes




Rc,n ∼ p n1−c

15


c-Ramanujan Primes




Rc,n ∼ p n1−c

16


c-Ramanujan Primes


What percent of primes are c-Ramanujan?

Given Rc,n ∼ p n1−c

, it follows Rc,N(1−c) ∼ pN .

1-c

17


c-Ramanujan Primes





1-c

18


c-Ramanujan Primes





1-c

19


Preliminaries

Let π(x) be the prime-counting function that gives thenumber of primes less than or equal to x .

Rosser’s Theorem states:

pm = m log m + O(m log log m).

The Prime Number Theorem states:

limx→∞

π(x)x/ log(x)

= 1.

20


Preliminaries





limx→∞

π(x)x/ log(x)

= 1.

21


Preliminaries





limx→∞

π(x)x/ log(x)

= 1.

22


The Prime Number Theorem

The logarithmic integral function Li(x) is defined by

Li(x) =∫ x

2

1log t

dt .

The Prime Number Theorem gives us

π(x) = Li(x) + O(

xlog2 x

),

i.e., there is a C > 0 such that for all x sufficiently large

−Cx

log x≤ π(x)− Li(x) ≤ C

xlog x

.

23


The Prime Number Theorem

The logarithmic integral function Li(x) is defined by

Li(x) =∫ x

2

1log t

dt .

The Prime Number Theorem gives us

π(x) = Li(x) + O(

xlog2 x

),

i.e., there is a C > 0 such that for all x sufficiently large

−Cx

log x≤ π(x)− Li(x) ≤ C

xlog x

.

24


Existence of Rc,n

Theorem (Amersi, Beckwith, Ronan, 2011)For all n ∈ Z and all c ∈ (0,1), the n-th c-Ramanujanprime Rc,n exists.

Proof:The number of primes in (cx , x ] is π(x)− π(cx).Using the Prime Number Theorem and Mean ValueTheorem:

π(x)− π(cx) = Li(x)− Li(cx) + O(x log−2 x)= Li′(yc)(x − cx) + O(x log−2 x)

with yc(x) ∈ [cx , x ].

25


Existence of Rc,n


Proof:

The number of primes in (cx , x ] is π(x)− π(cx).Using the Prime Number Theorem and Mean ValueTheorem:



26


Existence of Rc,n


Proof:The number of primes in (cx , x ] is π(x)− π(cx).

Using the Prime Number Theorem and Mean ValueTheorem:



27


Existence of Rc,n



π(x)− π(cx) = Li(x)− Li(cx) + O(x log−2 x)

= Li′(yc)(x − cx) + O(x log−2 x)with yc(x) ∈ [cx , x ].

28


Existence of Rc,n





29


Existence of Rc,n

π(x)− π(cx) = (1−c)xlog yc

+ O(x log−2 x).

Since log yc = log x − bc, for bc ∈ [0,− log c],

π(x)− π(cx) =(1− c)x

log x − bc+ O

(x

log2 x

).

For sufficiently large x , π(x)− π(cx) is strictlyincreasing and π(x)− π(cx) ≥ n, for all integers n.

30


Existence of Rc,n


+ O(x log−2 x).


π(x)− π(cx) =(1− c)x

log x − bc+ O

(x

log2 x

).


31


Existence of Rc,n


+ O(x log−2 x).


π(x)− π(cx) =(1− c)x

log x − bc+ O

(x

log2 x

).


32


Bounds on Rc,n

Rc,n ≥ pn

≥ n log n.

Want Rc,n ≤ αcn log(αcn) for large n.

Conjecture αc = 21−c .

To prove, must showπ(x)− π(cx) ≥ n ∀x ≥ αcn log(αcn).

We then obtain bounds on log Rc,n :

(1− βc log log n

log n

)log n ≤ log Rc,n ≤

(1 +

βc log log nlog n

)log n.

33


Bounds on Rc,n

Rc,n ≥ pn ≥ n log n.





(1− βc log log n

log n


(1 +

βc log log nlog n

)log n.

34


Bounds on Rc,n






(1− βc log log n

log n


(1 +

βc log log nlog n

)log n.

35


Bounds on Rc,n






(1− βc log log n

log n


(1 +

βc log log nlog n

)log n.

36


Bounds on Rc,n






(1− βc log log n

log n


(1 +

βc log log nlog n

)log n.

37


Bounds on Rc,n






(1− βc log log n

log n


(1 +

βc log log nlog n

)log n.

38


Bounds on Rc,n






(1− βc log log n

log n


(1 +

βc log log nlog n

)log n.

39


Asymptotic Behavior

Theorem (Amersi, Beckwith, Ronan, 2011)For any fixed c ∈ (0,1), the nth c-Ramanujan prime isasymptotic to the n

1−c th prime as n→∞

By the triangle inequality∣∣∣Rc,n − p n1−c

∣∣∣ ≤∣∣∣∣Rc,n −

n1− c

log Rc,n

∣∣∣∣+ ∣∣∣∣ n1− c

log Rc,n −n

1− clog n

∣∣∣∣+

∣∣∣∣ n1− c

log n − n1− c

logn

1− c

∣∣∣∣+

∣∣∣∣ n1− c

log n − p n1−c

∣∣∣∣

≤ γcn log log n

Since n log log npn

→ 0 as n→∞⇒ Rc,n ∼ p n1−c

40


Asymptotic Behavior




∣∣∣ ≤∣∣∣∣Rc,n −

n1− c

log Rc,n

∣∣∣∣+ ∣∣∣∣ n1− c

log Rc,n −n

1− clog n

∣∣∣∣+

∣∣∣∣ n1− c

log n − n1− c

logn

1− c

∣∣∣∣+

∣∣∣∣ n1− c

log n − p n1−c

∣∣∣∣≤ γcn log log n

Since n log log npn

→ 0 as n→∞⇒ Rc,n ∼ p n1−c

41


Asymptotic Behavior




∣∣∣ ≤∣∣∣∣Rc,n −

n1− c

log Rc,n

∣∣∣∣+ ∣∣∣∣ n1− c

log Rc,n −n

1− clog n

∣∣∣∣+

∣∣∣∣ n1− c

log n − n1− c

logn

1− c

∣∣∣∣+

∣∣∣∣ n1− c

log n − p n1−c


Since n log log npn

→ 0 as n→∞

⇒ Rc,n ∼ p n1−c

42


Asymptotic Behavior




∣∣∣ ≤∣∣∣∣Rc,n −

n1− c

log Rc,n

∣∣∣∣+ ∣∣∣∣ n1− c

log Rc,n −n

1− clog n

∣∣∣∣+

∣∣∣∣ n1− c

log n − n1− c

logn

1− c

∣∣∣∣+

∣∣∣∣ n1− c

log n − p n1−c


Since n log log npn

→ 0 as n→∞⇒ Rc,n ∼ p n1−c

43


Frequency of c-Ramanujan Primes

Theorem (Amersi, Beckwith, Ronan, 2011)In the limit, the probability of a generic prime being ac-Ramanujan prime is 1− c.

Define N = b n1−c c.

|aN

|pN

|

bN

Worst cases:Rc,n = aN and every prime in (aN ,pN ] is c-Ramanujan,Rc,n = bN and every prime in [pN ,bN) is c-Ramanujan.

Goal: π(bN)−π(aN)π(pN)

→ 0 as N →∞.

44




Define N = b n1−c c.|

aN|

pN|

bN



→ 0 as N →∞.

45





aN|

pN|

bN

Worst cases:

Rc,n = aN and every prime in (aN ,pN ] is c-Ramanujan,Rc,n = bN and every prime in [pN ,bN) is c-Ramanujan.


→ 0 as N →∞.

46





aN|

pN|

bN

Worst cases:Rc,n = aN and every prime in (aN ,pN ] is c-Ramanujan,

Rc,n = bN and every prime in [pN ,bN) is c-Ramanujan.


→ 0 as N →∞.

47





aN|

pN|

bN



→ 0 as N →∞.

48





aN|

pN|

bN



→ 0 as N →∞.

49



1 Rosser’s Theorem:

pm = m log m + O(m log log m) x

2 Riemann Hypothesis:

π(x) = Li(x) + O(√

x log x) x

3 Prime Number Theorem X

⇒ π(bN)− π(aN)

π(pN)≤ ξ

log log Nlog N

→ 0 as N →∞

50




pm = m log m + O(m log log m)

x


π(x) = Li(x) + O(√

x log x) x



π(pN)≤ ξ

log log Nlog N

→ 0 as N →∞

51






π(x) = Li(x) + O(√

x log x) x



π(pN)≤ ξ

log log Nlog N

→ 0 as N →∞

52






π(x) = Li(x) + O(√

x log x) x



π(pN)≤ ξ

log log Nlog N

→ 0 as N →∞

53






π(x) = Li(x) + O(√

x log x)

x



π(pN)≤ ξ

log log Nlog N

→ 0 as N →∞

54






π(x) = Li(x) + O(√

x log x) x



π(pN)≤ ξ

log log Nlog N

→ 0 as N →∞

55






π(x) = Li(x) + O(√

x log x) x

3 Prime Number Theorem

X


π(pN)≤ ξ

log log Nlog N

→ 0 as N →∞

56






π(x) = Li(x) + O(√

x log x) x



π(pN)≤ ξ

log log Nlog N

→ 0 as N →∞

57






π(x) = Li(x) + O(√

x log x) x



π(pN)≤ ξ

log log Nlog N

→ 0 as N →∞

58


Prime Numbers

2 3 5 7 11 13 1719 23 29 31 37 41 4347 53 59 61 67 71 7379 83 89 97 101 103 107109 113 127 131 137 139 149151 157 163 167 173 179 181191 193 197 199 211 223 227

59


Ramanujan Primes

2 3 5 7 11 13 1719 23 29 31 37 41 4347 53 59 61 67 71 7379 83 89 97 101 103 107109 113 127 131 137 139 149151 157 163 167 173 179 181191 193 197 199 211 223 227

60


Coin Flipping Model (Variation on Cramer Model)

We defineγ = 0.5772 . . ., the Euler-Mascheroni constant,

P, the probability of Heads,N, the number of trials,LN , the longest run of Heads.

E[LN ] ≈ log Nlog(1/P)

−(

12− log(1− P) + γ

log(1/P)

)Var[LN ] ≈ π2

6 log2 (1/P)+

112

61



We defineγ = 0.5772 . . ., the Euler-Mascheroni constant,P, the probability of Heads,

N, the number of trials,LN , the longest run of Heads.


−(

12− log(1− P) + γ

log(1/P)

)Var[LN ] ≈ π2

6 log2 (1/P)+

112

62



We defineγ = 0.5772 . . ., the Euler-Mascheroni constant,P, the probability of Heads,N, the number of trials,

LN , the longest run of Heads.


−(

12− log(1− P) + γ

log(1/P)

)Var[LN ] ≈ π2

6 log2 (1/P)+

112

63



We defineγ = 0.5772 . . ., the Euler-Mascheroni constant,P, the probability of Heads,N, the number of trials,LN , the longest run of Heads.


−(

12− log(1− P) + γ

log(1/P)

)Var[LN ] ≈ π2

6 log2 (1/P)+

112

64





−(

12− log(1− P) + γ

log(1/P)

)

Var[LN ] ≈ π2

6 log2 (1/P)+

112

65





−(

12− log(1− P) + γ

log(1/P)

)Var[LN ] ≈ π2

6 log2 (1/P)+

112

66


What is P?

Define Pc as the frequency of c-Ramanujan primesamongst the primes,

As N →∞,Pc = 1− c,For finite intervals [a,b], Pc is a function of a and b,Choose a = 105,b = 106.

67


What is P?

Define Pc as the frequency of c-Ramanujan primesamongst the primes,As N →∞,Pc = 1− c,

For finite intervals [a,b], Pc is a function of a and b,Choose a = 105,b = 106.

68


What is P?

Define Pc as the frequency of c-Ramanujan primesamongst the primes,As N →∞,Pc = 1− c,For finite intervals [a,b], Pc is a function of a and b,

Choose a = 105,b = 106.

69


What is P?

Define Pc as the frequency of c-Ramanujan primesamongst the primes,As N →∞,Pc = 1− c,For finite intervals [a,b], Pc is a function of a and b,Choose a = 105,b = 106.

70


Distribution of generalized Ramanujan primes

Length of the longest run in [105,106] ofc-Ramanujan primes Non-c-Ramanujan primes

c Expected Actual Expected Actual

0.50 14 20 16 36

71


Distribution of generalized c-Ramanujan primes

Length of the longest run in [105,106] ofc-Ramanujan primes Non-c-Ramanujan primes

c Expected Actual Expected Actual0.10 70 58 5 30.20 38 36 7 70.30 25 25 10 120.40 18 21 13 160.50 14 20 16 360.60 11 17 22 420.70 9 14 30 780.80 7 9 46 1540.90 5 11 91 345

72


Open Problems

1 Laishram and Sondow: p2n < Rn < p3n for n > 1.Can we find good choices of ac and bc such thatpacn ≤ Rc,n ≤ pbcn for all n?

2 For a given prime p, for what values of c is p ac-Ramanujan prime?

3 Is there any explanation for the unexpecteddistribution of c-Ramanujan primes amongst theprimes?

73


Open Problems




74


Open Problems




75


Acknowledgements

This work was supported by the NSF, Williams College,University College London.

We would like to thank Jonathan Sondow and ourcolleagues from the 2011 REU at Williams College.

76

generalized ramanujan primes - williams college...introduction c-ramanujan primes results...

Documents