Consider the infinite matrix starting:
$$\displaystyle T = -\left( \begin{array}{ccccccc} +1&+1&+1&+1&+1&+1&+1&\cdots \\ +1&-1&+1&-1&+1&-1&+1 \\ +1&+1&-2&+1&+1&-2&+1 \\ +1&-1&+1&-1&+1&-1&+1 \\ +1&+1&+1&+1&-4&+1&+1 \\ +1&-1&-2&-1&+1&+2&+1 \\ +1&+1&+1&+1&+1&+1&-6 \\ \vdots&&&&&&&\ddots \end{array} \right)$$
defined by either the recurrence:
$$\displaystyle T(n,1)=-1, T(1,k)=-1, n>=k:T(n,k) = -\sum\limits_{i=1}^{k-1} T(n-i,k), n<k:T(n,k) = -\sum\limits_{i=1}^{n-1} T(k-i,n)$$
or:
A191898=A051731*transpose(A143256)
or:
$$\displaystyle T(n,k) = -a(GCD(n,k))$$
where $a$ is the Dirichlet inverse of the Euler totient function.
Does the largest eigenvalue of the matrix T(n,k) approximate the previous prime number sequence?
The signs of the eigenvalues appear to agree with the Möbius function.
As a Mathematica program this is:
Clear[b, t, n, k, i, j]
t[n_, 1] = -1;
t[1, k_] = -1;
t[n_, k_] :=
t[n, k] =
If[n >= k, -Sum[t[n - i, k], {i, 1, k - 1}], -Sum[
t[k - i, n], {i, 1, n - 1}]];
nn = 42;
b = Range[1, nn]*0;
Do[m = Table[Table[t[n, k], {k, 1, j}], {n, 1, j}];
b[[j]] = RankedMax[Eigenvalues[m], 1], {j, 1, nn}]
Round[b]
Table[NextPrime[i, -1], {i, 2, 43}]
with the largest eigenvalues, rounded:
{-1, 1, 3, 3, 5, 5, 7, 7, 7, 8, 11, 11, 13, 13, 13, 13, 17, 17, 19, 19, 19, 19, 23, 23, 23, 23, 23, 23, 29, 29, 31, 31, 31, 31, 31, 31, 37, 37, 37, 37, 41, 41}
and the previous prime sequence:
{-2, 2, 3, 3, 5, 5, 7, 7, 7, 7, 11, 11, 13, 13, 13, 13, 17, 17, 19, 19, 19, 19, 23, 23, 23, 23, 23, 23, 29, 29, 31, 31, 31, 31, 31, 31, 37, 37, 37, 37, 41, 41}
as output.
Also, do the largest eigenvalues of this matrix approximate infinitely long sequences of consecutive prime numbers as the size of the matrix goes to infinity?
Again as a Mathematica program for a $200$ times $200$ matrix this is:
Clear[b, t, n, k, i, j]
t[n_, 1] = -1;
t[1, k_] = -1;
t[n_, k_] :=
t[n, k] =
If[n >= k, -Sum[t[n - i, k], {i, 1, k - 1}], -Sum[
t[k - i, n], {i, 1, n - 1}]];
nn = 200;
m = Table[Table[t[n, k], {k, 1, nn}], {n, 1, nn}];
N[Sort[Eigenvalues[m], Less]]
which outputs a long list of eigenvalues of which the last few are:
{..., 157.4, 163.302, 167.281, 173.217, 179.157, 181.163, 191.074, 193.065, 197.038, 199.026}
Rounding these gives us:
{..., 157, 163, 167, 173, 179, 181, 191, 193, 197, 199}
which are equal to the 37:th to 46:th primes.
Table[Prime[i], {i, 37, 46}]
{157, 163, 167, 173, 179, 181, 191, 193, 197, 199}
Edit 1.8.2013: A more precise approximation of the primes by eigenvalues is given by this Mathematica program:
Clear[nn, n, k, d, kolumn];
a[n_] := If[n < 1, 0, Sum[d MoebiusMu@d, {d, Divisors[n]}]]
Do[nn = j;
A3 = Range[nn]*0;
Do[
kolumn = i;
A1 = Table[Table[a[GCD[n, k]], {k, 1, nn}], {n, 1, nn}];
MatrixForm[A1];
A1[[All, kolumn]];
MatrixForm[
Table[Table[
If[Mod[n, k] == 0, MoebiusMu[n/k]*A1[[All, kolumn]][[k]],
0], {k, 1, nn}], {n, 1, nn}]];
a1 = Table[
Total[Table[
If[Mod[n, k] == 0, MoebiusMu[n/k]*A1[[All, kolumn]][[k]],
0], {k, 1, nn}]], {n, 1, nn}];
a2 = Sign[a1]*Exp[Abs[a1]];
A2 = Table[
Table[If[Mod[n, k] == 0, a2[[n/k]], 0], {k, 1, nn}], {n, 1, nn}];
MatrixForm[A2];
a3 = Table[
Total[Table[If[Mod[n, k] == 0, a2[[n/k]], 0], {k, 1, nn}]], {n,
1, nn}];
A3[[i]] = a3;
, {i, 1, nn}]
MatrixForm[A3];
Print[N[Log[-Min[Eigenvalues[A3]]], 12]]
(*Print[N[Sign[Eigenvalues[A3]]Log[Abs[Eigenvalues[A3]]],12]]*)
, {j, 1, 24}]
which outputs:
{1.00000000000+3.14159265359 I, 1.71614308398, 2.89852243027, 2.94293185770, 4.98292271242, 4.98305676486, 6.99755465240, 6.99756090303, 6.99756737456, 6.99756777041, 10.9999546191, 10.9999546211, 12.9999938562, 12.9999938562, 12.9999938562, 12.9999938562, 16.9999998874, 16.9999998874, 18.9999999847, 18.9999999847, 18.9999999847, 18.9999999847, 22.9999999997, 22.9999999997}
Rounding these values we get:
{1, 2, 3, 3, 5, 5, 7, 7, 7, 7, 11, 11, 13, 13, 13, 13, 17, 17, 19, 19, 19, 19, 23, 23}
The program above basically uses the Riemann zeta function product that converges to the Dirichlet inverse of the Euler totient function with exponentiated divisors, analogously to the Riemann zeta function product that converges to the von Mangoldt function. I will explain more later, bedtime here now.
This latter matrix starts:
$$A_3 = \left( \begin{array}{cccccc} e & e & e & e & e & e \\ e & e-e^2 & e & e-e^2 & e & e-e^2 \\ e & e & e-e^3 & e & e & e-e^3 \\ e & e-e^2 & e & e-e^2 & e & e-e^2 \\ e & e & e & e & e-e^5 & e \\ e & e-e^2 & e-e^3 & e-e^2 & e & e-e^2-e^3+e^6 \end{array} \right)$$
and is defined by taking the Möbius transform of each column of the first matrix $T$, exponentiating the divisors, multiplying with the Möbius function and then taking the inverse Möbius transform.
$$\displaystyle T = \begin{bmatrix} +1&+1&+1&+1&+1&+1&+1 \\ +1&-1&+1&-1&+1&-1&+1 \\ +1&+1&-2&+1&+1&-2&+1 \\ +1&-1&+1&-1&+1&-1&+1 \\ +1&+1&+1&+1&-4&+1&+1 \\ +1&-1&-2&-1&+1&+2&+1 \\ +1&+1&+1&+1&+1&+1&-6 \end{bmatrix}$$
Example: For the 6-th column with the entries: {1,-1,-2,-1,1,2,...} by Möbius inversion we get: {1,-2,-3,0,0,6} which is equal to the divisors of $6$ times the elementwise multiplication of the Möbius function of the divisors of $6$: {1,-2,-3,0,0,6}={1,2,3,0,0,6} times {1,-1,-1,0,-1,1}. By then exponentiating the divisors we have: {e^1,e^2,e^3,0,0,e^6} times {1,-1,-1,0,-1,1}. The Inverse Möbius transform (=summing over the divisors) then gives: {e-e^2, e-e^3, e-e^2, e, e-e^2-e^3+e^6}
The claim is that the most negative eigenvalue of $A_3$ approximates the previous prime number sequence.
Of course this could be explained more clearly by starting with the exponentiated divisors, its matrix inverse and the row sums thereof, I believe.
Using the same algorithm as above and exponentiating and correspondingly logarithmating the terms in the eigenvalue sequence arbitrary many times, it appear that they - the logarithms of the eigenvalues - converge to the Möbius function times the natural numbers.
Example, double exponentiating of divisors and double logarithms of eigenvalues of a 12 times 12 matrix: {-11.0000000000,10.0000000000,-7.00000000000,6.00171666577,-5.00000000000,-3.03392765715,-2.13750865340,1.00344457743,0,0,0,0}
Example, double exponentiating of divisors and double logarithms of eigenvalues of a 13 times 13 matrix: {-13.0000000000,-11.0000000000,10.0000000000,-7.00000000000,6.00171666577,-5.00000000000,-3.03392765715,-2.13750865340,1.00344457743,0,0,0,0}
which appears to converge to: {1, -2, -3, 0, -5, 6, -7, 0, 0, 10, -11, 0, -13, 14, 15, 0, -17, 0, -19, 0, 21, 22, -23, 0, 0, 26, 0, 0, -29, -30, -31, 0}
Or as a Mathematica line:
Range[32]*MoebiusMu[Range[32]]
Link to Pastebin with Mathematica program
Clear[n, k, a1, A1, a2, nn]
nn = 8;
b1 = Expand[
Table[Limit[
Zeta[s]*Total[
MoebiusMu[Divisors[n]]*Exp[Exp[Exp[Divisors[n]]]]^(s - 1)],
s -> 1], {n, 1, nn}]];
b1[[1]] = Exp[Exp[Exp[0]]];
A1 = Table[Table[b1[[GCD[n, k]]], {k, 1, nn}], {n, 1, nn}];
MatrixForm[A1]
a2 = Eigenvalues[A1];
N[Table[Sign[a2[[i]]] If[a2[[i]] == 0, 0, Log[Log[Abs[a2[[i]]]]]], {i,
1, nn}], nn]
Output: {-7.0000000, 6.0000000, -5.0000000, -3.0000000, -2.1375087, 1.0034446, 0, 0}
which are:
$$\text{sign}(Eigenvalues(A_1))\log(\log(abs(Eigenvalues(A_1))))$$ of: $$A_1(n,k) = b_1(GCD(n,k))$$ where: $$b_1(n) = \lim_{s\to 1} \, \zeta (s) \sum\limits_{d|n}\mu (d(n)) (\exp(\exp (\exp (d(n)))))^{s-1}$$
So in terms of this limit with the zeta function, there is one logarithm less than the number of exponentials.
As already mentioned above, the sequence that begins to emerge is found in the oeis:
{1, -2, -3, 0, -5, 6, -7, 0, 0, 10, -11, 0, -13, 14, 15, 0, -17, 0, -19, 0, 21, 22, -23, 0, 0, 26, 0, 0, -29, -30, -31, 0}
I believe this tendency becomes clearer with infinitely many exponentials of the divisors and equally many, minus one, logarithms of the eigenvalues. So something like:
$$\text{sign}(Eigenvalues(A_1))\log(\log(\log(\log(\log(\log(\log(abs(Eigenvalues(A_1)))))))))$$ of: $$A_1(n,k) = b_1(GCD(n,k))$$ where: $$b_1(n) = \lim_{s\to 1} \, \zeta (s) \sum\limits_{d|n}\mu (d(n)) (\exp(\exp(\exp(\exp(\exp(\exp(\exp (\exp (d(n))))))))))^{s-1}$$
Edit 22.8.2013: Simplifying further to a infinite matrix
$$A_1 = \left( \begin{array}{cccccccccccc} e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} \\ e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 \\ e^{e^{e^e}} & e^{e^{e^e}} & 0 & e^{e^{e^e}} & e^{e^{e^e}} & 0 & e^{e^{e^e}} & e^{e^{e^e}} & 0 & e^{e^{e^e}} & e^{e^{e^e}} & 0 \\ e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 \\ e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & 0 & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & 0 & e^{e^{e^e}} & e^{e^{e^e}} \\ e^{e^{e^e}} & 0 & 0 & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & 0 & 0 & e^{e^{e^e}} & 0 \\ e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & 0 & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} \\ e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 \\ e^{e^{e^e}} & e^{e^{e^e}} & 0 & e^{e^{e^e}} & e^{e^{e^e}} & 0 & e^{e^{e^e}} & e^{e^{e^e}} & 0 & e^{e^{e^e}} & e^{e^{e^e}} & 0 \\ e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & 0 & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 \\ e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & e^{e^{e^e}} & 0 & e^{e^{e^e}} \\ e^{e^{e^e}} & 0 & 0 & 0 & e^{e^{e^e}} & 0 & e^{e^{e^e}} & 0 & 0 & 0 & e^{e^{e^e}} & 0 \end{array} \right)$$
equal to $$(\exp(\exp(\exp(\exp(\exp(\exp(\exp(\exp(1))))))))))$$ if GCD(n,k)=1 and 0 otherwise, and taking equally many logarithms of the eigenvalues:
$$\text{sign}(Eigenvalues(A_1))\log(\log(\log(\log(\log(\log(\log(\log(abs(Eigenvalues(A_1))))))))))$$
we get the (rounded) eigenvalues:
{1.00000,-1.00000,-1.00000,-1.00000,-1.00000,1.00000,-1.00000,1.00000,-1.00000,1.00000,-1.00000,-1.00000,-1.00000,-1.00000,1.00000,1.00000,1.00000,-1.00000,1.00000,-1.00000,0,0,0,0,0,0,0,0,0,0,0,0}
which appears to be a rearrangement of the Möbius function:
{1, -1, -1, 0, -1, 1, -1, 0, 0, 1, -1, 0, -1, 1, 1, 0, -1, 0, -1, 0, 1, 1, -1, 0, 0, 1, 0, 0, -1, -1, -1, 0}
Mathematica:
Do[
nn = ii;
A1 = Table[
Table[If[GCD[n, k] == 1, Exp[Exp[Exp[Exp[1]]]], 0], {k, 1,
nn}], {n, 1, nn}];
a2 = Eigenvalues[A1];
Print[N[Table[
Sign[a2[[i]]] If[a2[[i]] == 0, 0,
Log[Log[Log[Log[Abs[a2[[i]]]]]]]], {i, 1, nn}], 6]], {ii, 1, 32}]
MatrixForm[A1]
Mussardos paper:
http://lanl.arxiv.org/pdf/cond-mat/9712010.pdf
http://people.sissa.it/~mussardo/Professional_web/Quantum_Mechanics_and_Number_Theory.html
The eigenvalues can be arbitrary. Here the zeta zeros as input, and as output the zeta zeros as eigenvalues:
(*Mathematica 8 program start*)(*The Mobius function times "n" \
approximately as the eigenvalues of a matrix*)Clear[nn, n, k, d, \
kolumn]
a[n_] := If[n < 1, 0, Sum[Im[ZetaZero[d]], {d, Divisors[n]}]]
Do[nn = j;
A3 = Range[nn]*0;
Do[kolumn = i;
A1 = Table[Table[a[GCD[n, k]], {k, 1, nn}], {n, 1, nn}];
MatrixForm[A1];
A1[[All, kolumn]];
MatrixForm[
Table[Table[
If[Mod[n, k] == 0, MoebiusMu[n/k]*A1[[All, kolumn]][[k]],
0], {k, 1, nn}], {n, 1, nn}]];
a1 = Table[
Total[Table[
If[Mod[n, k] == 0, MoebiusMu[n/k]*A1[[All, kolumn]][[k]],
0], {k, 1, nn}]], {n, 1, nn}];
a2 = Sign[a1]*Exp[Abs[a1]];
A2 = Table[
Table[If[Mod[n, k] == 0, a2[[n/k]], 0], {k, 1, nn}], {n, 1, nn}];
MatrixForm[A2];
a3 = Table[
Total[Table[If[Mod[n, k] == 0, a2[[n/k]], 0], {k, 1, nn}]], {n,
1, nn}];
A3[[i]] = a3;, {i, 1, nn}] MatrixForm[A3];
(*Print[N[Log[Log[-Min[Eigenvalues[A3]]]],12]]*)
Print[N[Sign[Eigenvalues[A3]] Log[Abs[Eigenvalues[A3]]], 12]], {j, 1,
10}]
(*program end*)