FUN WITH PASCAL'S TRIANGLE

Paul Jackson, 2009

Consider Pascal's Triangle...

1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
1 5 10 10 5 1
6 15 20 15 6 1
7 21 35 35 21 7 1

. . . . . . .

. .

Let the diagonal elements be called tn so the triangle numbers would be t3 for example. Of course
these each form infinite sequences. Then in general tn = 
n−1r
r 
. We shall show that

  n−1 
∞

∑ t1 =  n−2 , for n≥3 .

r=0 n

We also know that the sums of the reciprocals of t1 and t2 are both divergent. Using elementary
methods we find that for n = 3, 4 , 5 the sums are 2/1, 3/2, 4/3 respectively, so we seek to generalize
on these first results. To find these sums we invert the general formula for tn and after factoring out
the common numerator we get a general term that is the reciprocal of a product of factors in
arithmetic progression.

1
Special case

So as an example; for n = 5 we have a general term,

 1
t5
= 1/  4r
r =
4!
 r1  r2  r3  r 4 
, for r = 0 , 1 , 2 ...
Now just to make the structure a little clearer we put r +1 = r , for r = 1 , 2 , 3 ,... That is we have
1
redefined r to begin at r = 1 instead of zero. Then we need to express in
 r  r 1  r 2  r3 
partial fractions. So we have a general term:


1 1
−
3

3
−
3! r  r1   r2   r3 
1
 . Then when we sum to infinity we find all except the first
few terms disappear, in fact we get the following pattern for the terms inside the brackets:

1 3 3 1
−  −
1 2 3 4

1 3 3 1
+ −  −
2 3 4 5

1 3 3 1
+ −  −
3 4 5 6

1 3 3
+ −  - ..
4 5 6

1
+ . . . etc. We see that the terms diagonally starting from the top
5
right corner must equal zero, as they all have the same denominator and the numerators are just the
value of 1−x 3 when x = 1. So we see that we are just left with the first upper triangle of terms.
1
Now curiously when we sum these we find the value is just , hence we see that the sum we
3
4
require is just .
3
Now for the general case we find the same pattern occurring that is the first upper triangle of terms
1
has value
 n−2  , we shall next prove this is the case.

2
Toward the general case
1
In general using the above example we would need to express in
 r  r 1 r 2  .........  r k 
partial fractions, that is we need:

1 A0 A1 A2 Ak
=   + .........+ . [1]
 r  r1 r2  .........  rk  r  r1   r 2   r k 
We have k terms and k Ai coefficients to evaluate. Say we wished to evaluate an arbitrary
coefficient Ai , then we simply multiply [1] throughout by (r+ i) . In the left hand side of [1] this
term will cancel out, and on the right hand side each term except the one with coefficient Ai will be
multiplied by (r+ i). Now when we put r = -i into both sides of our expression, we are left with the
value of Ai alone, the other terms on the right hand side being each zero.
So for each i = 0, 1, 2 , 3, .......k we can evaluate [1] and nowhere on the left hand side will we get a
zero in the denominator.

Also in general the sign of Ai depends on the sign of the denominator

 r  r1  r2  .........  ri−1 ri1  ri2 ............  rk 

in the left hand side of [1].
We see that the product of terms after r + i- 1 is positive, and as there are i terms before and
including this term which will all be negative, when we substitute for r = i, we see the sign of Ai is
just (-1)i .

−1    i 1
 .
i i
−1 k
So we find Ai = = = −1 k ! i
 i  i−1  i−2  ......... 3 . 2 .1 .1 . 2 . 3........  k −i  i! k −i !

Hence if we let Bi = ki  , then A = i

Bi
−1i
k!
. So we see that in our general expression the
numerators in the array of fractions with denominators in arithmetic progression starting 1, 2 3... as
in the case for n = 5 above, will be binomial coefficients, of the appropriate row of Pascal's
Triangle, and hence each diagonal of n terms will sum to zero as in the case illustrated above, and
thus the sum of the series will depend on the sum of the upper left triangle of terms, which we need
1
to show is always equal to .
 n−2

Now when we form upper left triangle of terms as we did above, for the next case, n = 6, that is
k =4, and when we sum the diagonals of non disappearing terms we find for we have;
1 3 3 1
−  − , that is the first three non disappearing terms for the triangle for k = 3, plus the
1 2 3 4
extra term. The pattern is the same for the next few values of k. It seems possible that we can
generate the next sum of the upper right triangle from the previous one. This should not be too
surprising as we are dealing with binomial coefficients, in the form of Pascal's triangle, that seems
to be mapped in slightly strange ways. It would appear that as we manipulate this structure the same
coefficients keep popping up.
3
Now consider the diagonal elements in the triangle for k =4.
The first diagonal sum d1 = 1,
1 4 1−4 −3
second diagonal sum d2 = − = = ,
2 2 2 2
the third d3 = 1−46 = −36 = 3 , and lastly
3 3 3
1−46−4 −1
the fourth d4 = = .
4 4

But of course 4 = 1+3, so we are undoing the formation of the next line of coefficients in Pascal's
triangle. In fact we can reverse the procedure we have just accomplished to generate the upper right
triangle of terms for k = 4 from k = 3.

1 3 3
So we have; −  and we expand these to give:
1 2 3
1
, the next diagonal has two terms, so from Pascal's triangle 1+3 = 4, so – 3 = 1 – 4 , so this is
1
the next coefficient to be expanded

−3 1−4 1 4
= = − , the next term we need is the sum of 3 + 3 = 6, so 3 = -3 + 6, using the
2 2 2 2
last starting number.

3 −36 1−46
= = ,as -3 = 1- 4 from above, we then plug 3 into the next diagonal
3 3 3
expansion as -1 = 3 – 4 .

For the last diagonal we know it has k terms, and we know that the sum of the terms 1 – 4 + 6 – 4
+1 is zero, so the sum of 1 – 4 + 6 – 4 = -1, therefore the sum of the upper left diagonal for k = 4 is
1 3 3 −1 1
just the sum of −  and which is as required.
1 2 3 4 4

So in this particular case we will be able to find the the value of the upper triangle for k = 4 if we
knew the value of the first row of the upper triangle for k = 3. This is because we can always easily
deduce the value of the last diagonal for the upper triangle for k = 4, using binomial theorem in the
expansion of (1- x )4 when x = 1.

To generalize this we need some notation.

Let the the first row of the upper triangle for k be denoted by Rk , and the upper triangle, and last
diagonal for k + 1 be denoted by Uk+1 and dk+1 respectively. Then the value of dk+1 is just the value
∓1
of (1 – x )k+1 when x = 1, (which is zero of course), without the last term this being .
k 1

4
Hence Uk+1 = Rk - dk+1 . So we would solve this problem if we knew in general the value of Rk .
This is easy to find for small values of k . Which we tabulate below.

k 1 2 3 4 5 6 7 8
Rk 1 0 1/2 0 1/3 0 1/4 0

This again seems to imply interesting symmetry properties of the expressions for Rk and the
binomial coefficients, but we leave that aside to solve the current problem.
Now recall the example for k = 3 at the start of the discussion above we had the expression:

1
 r  r1  r2  r3 
=
1 1
−
3


3
−
3! r  r1   r 2   r 3 
1

expressed on the right hand side
in partial fractions, now when we let r = 1, and rearrange we get
3!
4!
=
 1 3 3 1
−  −
1 2 3 4 
but this is just R3 with the extra term! So we see that R3 =
1 −1
4
−
4

1
= as we expected. It is not hard to extend this example to the general case, from which we find
2
2
that Rk has value zero if k is even, and value if k is odd, remembering from above that
k 1
B
Ai = −1i i so the Bi are the binomial coefficients, or numerators in Rk .
k!

1 1
Therefore we see that the result Uk+1 = = follows, and so does what we set out to
k n−2
initially show, that is;

  n−1 
∞
1
∑
tn
=
 n−2 
, for n3 .
r=0

Finally we note in passing the similarity of this expression to

 
∞ ∞
1
∑ −1 t n x = ∑
r r

r=0
−1r n−1r x r =
r  1x 
n , which if valid in the context of infinite
r=0
series, would give, when we put x = 1, the value of the reciprocal of the nth row in Pascal's triangle.
It is almost as if we have a mapping of the diagonals of Pascal's Triangle to the rows.
 n−1  1
We seem to have a function f :  , for n3 n∈ℕ .
 n−2  2n
The above expression is a generating function, so it would seem we can put :

     
∞ n ∞ n
1
∑ −1 x r r
∑ −1r n−1r
=
r
xr =
1 x
as the last expression is just the nth
r=0 r=0

power of the standard Geometric Series.

We also note that there other manipulations of Pascal's Triangle that lead to the Fibonacci sequence
and Lucas sequences in general, and thence to Pell's equation and second order recurrence relations
and beyond.