# successive iteration

I hope you can help with a problem I am having. I want to solve this equation

$$\left( n^2 \Theta^2 + 2\alpha n\Theta -1\right)R_n = \frac{1}{R_0^2}\left( 2 \sum_{m=1}^\infty R_{n+m}R_m + \sum_{m=1}^n R_{n-m}R_m\right)$$,

for different value of $\Theta$ and $\alpha$, here 0.1 and 1. This is done by guessing a trial function $R_n \propto \exp(-n)$ and doing successive iteration. Thereby calculating the sums for a fixed value. Here I choose $m \leq 256$ divided by iteration number. So first one is 256 next is 128 etc. In the end I want to sum over all the number R_n to get a single value. My idea so far is to do the following code, by brute force

    f[n_] = Exp[-n]
E2 = (Sum[2*f[n + m]*f[m], {m, 256}] + Sum[f[n - m]*f[m], {m, 256}])/(n^2*0.1^2 + 2*n*0.1*1-1)*1/(f[0]^2)
D1[n_] = E2;


and then doing it all again

      E3 = (Sum[2*D1[n + m]*D1[m], {m, 256/2}] + Sum[D1[n - m]*D1[m], {m, 256/2}])/(n^2*0.1^2 + 2*n*0.1*1 - 1)*1/(D1[0]^2);
D2[n_] = E3;


and again

      E4 = (Sum[2*D2[n + m]*D2[m], {m, 256/4}] + Sum[D2[n - m]*D2[m], {m, 256/4}])/(n^2*0.1^2 + 2*n*0.1*1 - 1)*1/(D2[0]^2);
D3[n_] = E4;


The problem now is that the evaluation of E5 etc. take a long time. Am I doing it the right clever way? I want in the end a list of R_n values. Hope I have provided enough information.

I am trying to replicate the calculation in this paper https://journals.aps.org/prb/abstract/10.1103/PhysRevB.36.1931, from equation 7.1, where they for $\alpha=1$ and $\theta=0.1$ find that

$$s=2\pi/\Theta \left( 1+1/8 \sum_{n,m=-\infty}^{\infty} R_{n+m} R_n R_m \right)=22.2$$

• yes it should be and is in my code, will change in question – Morten Sode Apr 1 '18 at 21:23
• I am confused. Negative indices of R are allowed? – Henrik Schumacher Apr 1 '18 at 21:35
• in the article they state that $R_{n}=R_{-n}$, so that negative indices are just the positive ones – Morten Sode Apr 1 '18 at 21:37
• Ah. Okay. And we can expect that $R_n \to 0$ for $n \to \infty$? – Henrik Schumacher Apr 1 '18 at 21:38
• yes, exactly so we are only interested in a fixed value of $R_n$ – Morten Sode Apr 1 '18 at 21:40

I am not completely sure but I think, you are looking for the roots of the function F below. For given R, F[R] is the difference of both side of the equation

$$\left( n^2 \Theta^2 + 2\alpha n\Theta -1\right)R_n = \frac{1}{R_0^2}\left( 2 \sum_{m=1}^\infty R_{n+m}R_m + \sum_{m=1}^n R_{n-m}R_m\right)$$

Caution: R[[1]] corresponds to $R_0$, R[[2]] to $R_1$, R[[3]] to $R_2$ and so on.

nn = 100;
R0 = Exp[-Range[0., nn - 1]];
Θ = 0.1;
α = 1.;

F = R \[Function]
Join[{(-1) R[[1]] - 2. R[[1 ;;]].R[[1 ;; nn]] / R[[1]]^2},
Table[(n^2 Θ^2 + 2 α n Θ - 1) R[[n + 1]] - (2. R[[n + 1 ;;]].R[[1 ;; nn - n]] + R[[2 ;; n + 1]].R[[n ;; 1 ;; -1]])/R[[1]]^2, {n, 1, nn - 1}]
];


We can employ FindRoot to search for roots as follows

R = Array[r, nn];
FindRoot[Evaluate[F[R]], Evaluate@Transpose[{R, Evaluate@R0}],
MaxIterations -> 1000]


In a nutshell, FindRoot applies Newton's method

$$R^{(k+1)} = R^{(k)} - DF(R^{(k)})^{-1} \, F(R^{(k)})$$

Unfortunately, $R^{(k)}$ seems to converge towards the constant vector $0$ for $k \to \infty$.

Edit

If I got it correctly, the fixed point iteration they use in the article is generated by the following function:

G = R \[Function]
Join[{ 2. R[[1 ;;]].R[[1 ;; nn]]/(-1 R[[1]]^2)},
Table[(2. R[[n + 1 ;;]].R[[1 ;; nn - n]] +
R[[2 ;; n + 1]].R[[n ;; 1 ;; -1]])/((n^2 Θ^2 +
2 α n Θ - 1) R[[1]]^2), {n, 1,
nn - 1}]];


This is how you can optain the list of the first 1000 iterates:

sequence = NestList[G, R0, 1000];


And indeed, the iterates seem to converge to a fixed point (the differences converge to the zero vector):

ListPlot[Max /@ Abs[Differences[sequence[[1 ;; 100]]]], PlotRange -> All]


I am surprised.

• Would you be able to explain how and what the F function represents. If it converges towards zero, then it could be due to $\frac{1}{F_0}$. – Morten Sode Apr 2 '18 at 0:26
• I am not entirely sure, but the F function looks a bit weird in the table part. it seems like you also divide with $1/F_0^2$ on the bit to the left which you subtract. The bits with $((n^2 Θ^2 + 2 α n Θ - 1))$. – Morten Sode Apr 2 '18 at 9:13
• Ah, right. I corrected it. Still, FindRoot doesn't find a nontrivial solution with the given initial conditions... – Henrik Schumacher Apr 2 '18 at 10:41
• okay, I will have a go at it and accept your answer. I think the method is right, it is just confusing me because the article says it is a easy iteration . – Morten Sode Apr 2 '18 at 11:18