6

While solving this post, I came across the Somos-5 sequence again (since it appeared in another post). Given the Heronian triangle with half-sides $(a,b,c)$, area $A$, and medians $(m_1,m_2,m_3)$,

$$A=\sqrt{(a + b + c) (b + c - a) (a + c - b) (a + b - c)}$$ $$m_1=\sqrt{2 (b^2 + c^2) - a^2}$$ $$m_2=\sqrt{2 (c^2 + a^2) - b^2}$$ $$m_3=\sqrt{2 (a^2 + b^2) - c^2}$$

Let $(a,b,c)$ be integers. By solving an elliptic curve, it is possible to make all $m_k$ as integers. However, if $A$ is to be a non-zero integer as well, that is still an open problem.

If we start with the area $A$, Schubert conjectured in 1905 that solving the first three equations in non-zero integers was impossible, but was proven wrong by Buchholz's 1989 thesis.


I. Three quadratics

We do the cyclic substitution $(a,b,c) = (p+q,\,q-r,\,r-p)$ to simplify the equations,

\begin{align} |16\, p\, q\, r\, (-p - q + r)| &=\, A^2\\[5pt] (p - q)^2 - 4 (p + q) r + 4 r^2 &= m_1^2\\[5pt] (2 p + q)^2 - 2 (2 p - q) r + r^2 &= m_2^2\\[5pt] (p + 2 q)^2 + 2 (p - 2 q) r + r^2 &= m_3^2 \end{align}

So four quadratics in $r$ to be made squares. For this system, three at a time is infinitely solvable.


II. Somos-5 Sequences

Define the two Somos-5 sequences (starting with $n=0$),

$S_n = 1,\, 1,\, 1,\, 2,\, 3,\, 5,\, 11,\, 37,\, 83,\, 274,\, 1217,\; 6161,\; 22833,\dots$ (A006721)

$T_n = 0, 1, -1, 1, 1, -7, 8, -1, -57, 391, -455, -2729, 22352,\dots$ (A360381)

Then in page 6 of Andrew Hone's 2022 paper, we find the formulas (modified),

$$A = |4\alpha\beta|$$

$$\alpha = S_{n}\,S_{n+1}\,S_{n+2}^2\, S_{n+3}\, S_{n+4}$$ $$\beta = T_{n}\,T_{n+1}\,T_{n+2}^2\, T_{n+3}\, T_{n+4}$$

and the $(p,q,r)$ as,

$$p_n = S_{n+1}\,S_{n+2}^3\, S_{n+3}\, T_{n+2}$$

$$q_n = \;S_{n}^2\, S_{n+1}\, T_{n+3}\, T_{n+4}^2$$

$$r_n = T_{n+1}\, T_{n+2}^3 \,T_{n+3}\, S_{n+2}$$

$$s_n = -\,T_{n}^2\, T_{n+1}\, S_{n+3}\, S_{n+4}^2$$

with the formula for $s_n$ added by yours truly. So these $(A,p,q,r,s)$ are highly factorable, being products of 12 and 6 terms, respectively (counting multiplicity). Specifically,

$$p_n = -2,\, 24,\, 270, -28875, 1969880, -46246189,\dots$$

$$q_n\; =\; 1,\; 49,\; -896,\; 96,\; -146205, -2396409675,\dots$$

$$r_n\; = -1,\, -2,\; -21, -13720,\; 39424,\; 16872,\dots\;\quad$$

$$\quad s_n \,=\; 0,\; 75, -605, -15059, 1784251, -2442672736,\dots$$

Example:

$(p_1,q_1,r_1) = (24,\,49,-2)$ solves the first three quadratics above while $s_n \to r_n$ $(p_1,q_1,s_1) = (24,\,49,\,75)\,$ solves the first two and the fourth. And so on for all $n$.


III. Question

Q: What are the recurrence relations of the $(p,q,r,s)$? Are they also Somos-5 sequences but with different initial values?

P.S. Part 2 in in this post.

Somos
  • 37,457
  • 3
  • 35
  • 85
  • As Part 2 asks a slightly more difficult question, I separated it in this post. – Tito Piezas III Feb 02 '25 at 06:44
  • 1
    The same bilinear recurrence that defines both sequences $S_n$ and $T_n$ has 3 terms. The same bilinear recurrence for both sequence $p_n$ and $r_n$ has 7 terms. Similarly for $q_n$ with 8 terms and integer coefficients over 20 digits. – Somos Feb 02 '25 at 21:47
  • @Somos No wonder my initial attempts to find them were not successful. Can you convert your comment into an answer so I can accept it? Also, can you include just the form of the recurrence (without the 20-digit coefficients). Thanks. – Tito Piezas III Feb 03 '25 at 03:13
  • @Somos One of the needed variables is actually $a=p+q$. Will it have a simpler recursion or make it more complicated? – Tito Piezas III Feb 03 '25 at 04:08
  • I am working on the square roots sign determination. I will submit my answer later. The $a,b,c$ sequences all have the same 5 term recurrence with 7 digits coefficient max. – Somos Feb 03 '25 at 22:41
  • @Somos Ah, the $(a,b,c)$ sequences have a simpler recurrence, but the $(p,q,r)$ have a simpler factorization. A trade-off, I guess. – Tito Piezas III Feb 04 '25 at 06:09
  • @Somos I edited the post to add a fourth formula $s_n$ so it is now $(p,q,r,s)$, but it yields the same $(a,b,c)$. – Tito Piezas III Feb 04 '25 at 10:13
  • 1
    Thanks for the formulas for $s$ and $\alpha,\beta$! I can now easily find 8 term recurrences for $s$ and the signed area. – Somos Feb 04 '25 at 12:09

1 Answers1

4

The context of this question is about generalized Somos sequences. They are usually defined by initial values and a single homogeneous bilinear recurrence somewhat similar to the definition of generalized Fibonacci sequences which also satisfy infinite families of linear recurrences. Similarly, generalized Somos sequences are best characterized by satisfying infinite families of homogeneous bilinear recurrences.

For example, the Somos-5 sequence $S_n$ and its companion sequence $T_n$ are defined by the same recurrence

$$ S_{n+3}S_{n-2} = S_{n+2}S_{n-1} + S_{n+1}S_n,\\ T_{n+3}T_{n-2} = T_{n+2}T_{n-1} + T_{n+1}T_n \tag1 $$

with initial values $\,S_0 = S_1 = S_2 = 1$ and $\,T_0 = 0,\, T_1 = T_3 = T_4 = 1,\, T_2 = -1.$ Note that $\,S_{-n} = S_n,\, T_{-n} = -T_n$ for all integer $n$.

These recurrences are special cases of the infinite family of recurrences

$$ S_{n+m+1}S_{n-m}=-T_{m+1}T_m S_{n+2}S_{n-1}+T_{m+2}T_{m-1}S_{n+1}S_n,\\ T_{n+m+1}T_{n-m}=-T_{m+1}T_mT_{n+2}T_{n-1}+T_{m+2}T_{m-1}T_{n+1}T_n\tag2$$

valid for all integers $n$ and $m$. Notice that the indices $\,(n+m+1,n-m)$ have opposite even/odd parity. Similarly, the Somos-4 sequence $U_n$ and its companion sequence $V_n$ are defined by the same recurrence

$$ U_{n+2}U_{n-2} = U_{n+1}U_{n-1} + U_{n}U_n,\\ V_{n+2}V_{n-2} = V_{n+1}V_{n-1} + V_{n}V_n \tag3 $$

with inital values $\,U_0 = U_1 = U_2 = U_3 = 1$ and $\,V_0 = 0,\,V_1 = V_2 = 1,\, V_3 = -1,\, V_4 = -5.$ Note that $\,U_{-n} = U_{n+3},\, V_{-n} = -V_n$ for all integer $n$.

These recurrences are special cases of the infinite family of recurrences

$$ U_{n+m}U_{n-m}=V_{m}V_m U_{n+1}U_{n-1}-V_{m+1}V_{m-1}U_{n}U_n,\\ V_{n+m}V_{n-m}=V_{m}V_m V_{n+1}V_{n-1}-V_{m+1}V_{m-1}V_{n}V_n\tag4$$

valid for all integers $n$ and $m$. Notice that the indices $\,(n+m,n-m)$ have the same even/odd parity. Notice also the two sequences satisfy the infinite family of recurrences

$$ U_{n+m+1}U_{n-m}=V_{m+1}V_m U_{n+2}U_{n-1}-V_{m+2}V_{m-1}U_{n+1}U_n,\\ V_{n+m+1}V_{n-m}=V_{m+1}V_m V_{n+2}V_{n-1}-V_{m+2}V_{m-1}V_{n+1}V_n\tag5$$

valid for all integers $n$ and $m$. Notice the similarity of equation $(4)$ to equations $(2)$ except for sign changes. Is there an variation of equation $(4)$ for the Somos-5 sequences? Yes, it is

$$ S_{n+m}S_{n-m}=e_{n,m}T_{m}T_m S_{n+1}S_{n-1}- f_{n,m}T_{m+1}T_{m-1}S_{n}S_n,\\ T_{n+m}T_{n-m}=g_{n,m}T_{m}T_m T_{n+1}T_{n-1}- h_{n,m}T_{m+1}T_{m-1}T_{n}T_n\tag6$$

where $\,e,f,g,h$ depend only on the parity of both $n$ and $m$. There are other recurrences whose coefficients do not depend on the parity of $n$ and $m$ but they have more terms. The question is

Q: What are the recurrence relations of the $(p,q,r,s)$? Are they also Somos-5 sequences but with different initial values?

The recurrence relations for the $p$ and $r$ sequences are the same and similarly for the pair of sequences $q$ and $s$. For example

$$ 133p_{n+7}p_{n-6} = \sum_{k=1}^6 d_k p_{n+k}p_{n-k+1} \tag7 $$

where $\,d_6 = -2131349,\,\dots,\,d_1 = 605797403016587130$ and

$$ q_{n+8}q_{n-7} = \sum_{k=1}^7 e_k q_{n+k}q_{n-k+1} \tag8 $$

where $\,e_7 = -175111,\,\dots,\,e_1 = 1620919496354218862230$.

The recurrence relations for the $\,a,b,c$ sequences are the same. For example:

$$ a_{n+5}a_{n-4} = \sum_{k=1}^4 f_k a_{n+k}a_{n-k+1} \tag9 $$

where $\,f_4 = -37,\, f_3 = 165585,\, f_2 = -175047,\, f_1 = 2567356.$

Initial values of the sequences (starting with $n=-2$),

$$ a_n = 1,\, 0, -1, 73, -626, -28779, 1823675, -2442655864,\dots \\ b_n = 1, -1, 2, 51, -875, 13816, -185629, -2396426547, \dots \\ c_n = 0, -1, 1, -26, -291, 15155, -1930456, 46263061, \dots $$

where $\,a_{-2+n} = b_{-2-n}$ and $\,c_{-2+n} = -c_{-2-n}$ for all integer $n$.

Somos
  • 37,457
  • 3
  • 35
  • 85
  • 1
    Great! So the $(a,b,c)$ are generalized Somos-9 sequences! And I noticed three of the four $e_k$ can be divided by $S_7=37$ and another three by $S_9=83$. (In fact, most of the divisors of $c_k$ and $d_k$ are appropriate $S_m$ and $T_n$.) Can you kindly include the initial $a_n$? (I know they need to be signed which makes it a bit tricky.) – Tito Piezas III Feb 05 '25 at 03:18
  • 1
    Thanks for the edit! By the way, I read the Somos-n sequences with all coefficients $e_k=1$ yield integers only for $n\leq7$. But the one for $(a,b,c)$ is a generalized Somos-9. So I guess the rules for generalized Somos-n sequences are more relaxed? – Tito Piezas III Feb 05 '25 at 13:11
  • 1
    @TitoPiezasIII Yielding integers is a "red herring". Please read my first paragraph. – Somos Feb 05 '25 at 14:50
  • 1
    Well, you gotta change the Wikipedia entry then. It’s quite emphasized there. Anyway,, that is not my main concern. What I’d like to tackle next is Part 2. Randall Rathbun gave me some new data, but I need to go over it first. – Tito Piezas III Feb 05 '25 at 15:31
  • 1
    @TitoPiezasIII Thanks for that comment! I just made a minor edit in Wikipedia. I would be interested in the new data. – Somos Feb 05 '25 at 15:41
  • New data and some analysis is now in Part 2. No formula yet. Kindly check this long comment. – Tito Piezas III Feb 06 '25 at 07:39
  • Any success with a second family? I cannot believe the factorization of $(p,q,r)$ of Rathbun’s data is just coincidence. Too bad there’s only four solutions as six would have been better. – Tito Piezas III Feb 07 '25 at 09:57