0
$\begingroup$

In short, I want to solve a recurrence equation in two variables with multiple boundary conditions.

In more detail:

For instance, suppose I want to compute

$a[0,j]= 1 \quad \forall j$

$a[i,0]= 1 \quad \forall i$

$a[i,j] = a[i-1,j]+a[i-1,j-1]+a[i,j-1]\quad \forall i,j \geq 1$

My approach:

Clear["Global`*"]
RSolve[{a[i, j] == a[i - 1, j] + a[i - 1, j - 1] + a[i, j - 1], a[0, j] == 1, a[i, 0] == 1}, a[i, j], {i, j}]

Just yields the same expression as output -- without any error message or result:

RSolve[{a[i, j] == a[-1 + i, -1 + j] + a[-1 + i, j] + a[i, -1 + j], a[0, j] == 1, a[i, 0] == 1}, a[i, j], {i, j}]

I presume the error is because I somehow need to encode the case distinction (for $i,j \geq 1$, apply the recursive equation, else the constants), but I could not find any example code in the documentation on how to do this. In general, in the documentation, there is never a case distinction, so mathematica seems to ''find out'' which equation to use.

My attempt to fix this was a piecewise function:

RSolve[a[i, j] == Piecewise[{{a[i - 1, j], i > 0}, {1, i == 0} } ], a[i, j], {i, j}]

But also did not work.

Debugging the code further, I have the weird behaviour that this very basic version of the recurrence equation works:

RSolve[{a[i, j] == a[i - 1, j], a[0, j] == 1}, a[i, j], {i, j}]

{{a[i, j] -> 1}}

but as soon as I add the second base case:

RSolve[{a[i, j] == a[i - 1, j], a[0, j] == 1, a[i, 0] == 1},  a[i, j], {i, j}]

I get again the strange output

RSolve[{a[i, j] == a[-1 + i, j], a[0, j] == 1, a[i, 0] == 1}, a[i, j], {i, j}]

without any error message.

My questions would be:

(1) Does anyone know how to solve this problem?

(2) Going further, if we have another case distinction, e.g. the recurrence would change for $i > 10 $ to $ a[i,j] = 100*a[i-1,j]+ a[i-1,j-1] + a[i,j-1]$, is there a way to encode that case distinction?

$\endgroup$
4
  • $\begingroup$ RSolve[a[i, j] == a[i - 1, j], a, {i, j}] results in {{a -> Function[{i, j}, C[1][j]]}}. One may put C[1][j]=1. $\endgroup$
    – user64494
    Apr 30, 2022 at 11:47
  • $\begingroup$ I tried that trick of removing the boundary conditions for the full recursive formula, i.e. if you include $a[i-1,j-1]$ and $a[i,j-1]$, and it didn't work. Does it for you? $\endgroup$
    – black
    Apr 30, 2022 at 12:28
  • $\begingroup$ The question was essentially changed after my comment. The edit is not described at all. $\endgroup$
    – user64494
    Apr 30, 2022 at 15:46
  • $\begingroup$ The question itself was not changed; the difference is that I added my attempt using piecewise, but this is no affect on your comment. $\endgroup$
    – black
    Apr 30, 2022 at 16:22

1 Answer 1

1
$\begingroup$

Not a solution with RSolve, but get recurrenceTable with Reduce and find " Delannoy numbers d(i,j) " as generating formula.

eqs = {a[i, j] == a[i - 1, j] + a[i - 1, j - 1] + a[i, j - 1], 
       a[0, j] == 1, a[i, 0] == 1}

vars = Flatten[mat = Table[a[i, j], {i, 0, 6}, {j, 0, 6}]];

tab = Flatten@Table[eqs, {i, 0, 6}, {j, 0, 6}] // Union

red = Reduce[tab, vars, Reals]

sol = Solve[red]

(mat1 = mat /. sol[[1]]) // MatrixForm

(*   MatrixForm[{{1, 1, 1, 1, 1, 1, 1}, 
                 {1, 3, 5, 7, 9, 11, 13}, 
                 {1, 5, 13, 25, 41, 61, 85}, 
                 {1, 7, 25, 63, 129, 231, 377}, 
                 {1, 9, 41, 129, 321, 681, 1289}, 
                 {1, 11, 61, 231, 681, 1683, 3653}, 
                 {1, 13, 85, 377, 1289, 3653, 8989}}]   *)

And @@ (tab /. sol[[1]])

(*   True   *)

With the help of https://oeis.org/search?q=1%2C13%2C61%2C129&sort=&language=english&go=Search you find, the results are the " Delannoy numbers d(i,j) ".

d[n_, k_] := Binomial[n + k, k]*Hypergeometric2F1[-k, -n, -n - k, -1];

Table[d[n, k], {n, 0, 6}, {k, 0, n}] // MatrixForm

Since solution matrix is symmetric, get it with

(mat2 = Table[a[i, j], {i, 0, 6}, {j, 0, 6}] /. 
    a[i_, j_] -> d[i, j]) // MatrixForm

mat1 == mat2
(*   True   *)
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.