1

2 Journal of Integer Sequences, Vol. 24 (2021),

3
47 6
Article 21.6.6
23 11

New Recursion Formulas for the Partition Function

Teerapat Srichan, Watcharapon Pimsert1 , and Vichian Laohakosol

Department of Mathematics
Faculty of Science
Kasetsart University
Bangkok 10900
Thailand
[email protected]
[email protected]
[email protected]

Abstract
We derive several new recursion formulas for the unrestricted partition function as
applications of a general recursion formula for a novel restricted partition function.

1 Introduction and results

A partition of a positive integer n, [1, Chapter 14], is a finite sequence of positive integers
λ1 , λ2 , . . . , λm such that
λ1 + λ2 + · · · + λm = n.
The λi ’s are called the parts of the partition. The number of parts is unrestricted, repetition
is allowed and the order of the parts is not taken into account. The corresponding partition
function p(n) is defined as the number of unordered partitions of n. Some well-known values
of p(n) are ([6, A000041])

p(−n) = 0 (n ∈ N), p(0) = 1 (by convention),

p(1) = 1, p(2) = 2, p(3) = 3, p(4) = 5, p(5) = 7, p(6) = 11,
p(7) = 15, p(8) = 22, p(9) = 30, p(10) = 42, p(11) = 56. (1)

Finding explicit formulas for p(n) is one of the most fundamental problems in additive
number theory, and there have appeared a good deal of formulas used to compute p(n).
1
Corresponding author.

1
More recent ones in the spirit of our work here are e.g., the two works in [4, 5]. Added
to this list, we present here a seemingly new recursion formula connecting the unrestricted
partition function p(n) with a restricted partition function P (n, k); see also [6, A008284].
Throughout, N denotes the set of all positive integers and N0 := N ∪ {0}.
Definition 1. Let n ∈ N, and k ∈ N0 . Define
(
the number of partitions (λ1 , λ2 , . . . ) of n with k < λi < n, if 0 < k < n;
P (n, k) =
0, otherwise.

From this definition, note that P (n, k) = 0, when k > ⌊n/2⌋ − 1. Moreover, for a fixed
n ∈ N, the restricted partition P (n, k) is a decreasing function in k. Our main result reads
as follows.
Theorem 2. Let p(0) = 1 and define p(n) to be 0 if n < 0. Then for n ≥ 4 and k ≥ 1 we
have ∞
X
ak,i p(n − i) = P (n, k) − P (n − 1, k), (2)
i=0

where the coefficients ak,i are obtained from the expansion

k ∞
n n−1 n−2

Y −i
X
ak,i xn−i ,


x − 2x +x 1−x = (3)
i=2 i=0

where empty product is taken to be 1.

Comparing with existing schemes, Theorem 2 can be employed competitively to deter-
mine the values of p(n) efficiently subject to the ease of computing P (n, k) − P (n − 1, k) and
ak,i . When k is large, the expression P (n, k)−P (n−1, k) is easy to find while the coefficients
ak,i are not (because the product expansion in (3) is lengthy). In the last section, several
applications of Theorem 2 for both large and small values of k are worked out. In addition,
optimal values of k, in terms of n, which render easy and effective applications of Theorem
2 are derived.

2 Lemmas
The following lemma, whose simple straightforward computation check is omitted, provides
information for some particular cases needed in the proof of Theorem 2.
Lemma 3. For k ∈ N, we have

P (2k + 1, k) = 0, P (2k + 2, k) = 1, P (2k + 3, k) = 1.

The main ingredient in our proof of Theorem 2 is:

2
Lemma 4. For given k, n ∈ N with n > k, we have

P (n, k) = P (n, k + 1) + P (n − k − 1, k) + 1.

Proof. Let A be the set of unrestricted partitions of n having parts > k + 1 but < n, and B
the set of unrestricted partitions of n having parts > k with at least one part being k + 1.
By definition, we have

P (n, k) = |A| + |B|. (4)

By definition, we have
|A| = P (n, k + 1),
while |B| is equal to the number of unrestricted partitions of n − k − 1 having parts > k and
< n plus the partition n = (n − k − 1) + (k + 1). Putting these values into (4), Lemma 4
follows.
The next lemma relates the partition function p(n) with the initial value of the function
P (n, k).
Lemma 5. For n ≥ 2, we have

p(n) = P (n, 1) + p(n − 1) + 1. (5)

Proof. The case n = 2 follows immediately from the numerical values p(2) = 2, p(1) = 1
and P (2, 1) = 0.
For n ≥ 3, the proof follows from the observation that the set of all unrestricted parti-
tions of n comprises 3 disjoint classes. The first class consists of all partitions having parts
belonging to {2, 3, . . . , n − 1}, while the second class contains all partitions having at least
one part being 1, and the last class is simply the set {n}.
Our next lemma relates P (n, k) with the partition function ps (n) which counts the number
of partitions of n into s parts ([2, 3]).
Lemma 6. Let m, k ∈ N. We have
m−1
X
P (mk + r, k) = pi ((m − i)k + r), (6)
i=2

where 0 ≤ r < k.
Proof. For m ∈ {1, 2}, the relation (6) holds trivially because P (k, k) = P (2k, k) = P (2k +
1, k) = 0, and the right-hand sum is empty. Consider now m ≥ 3. By definition, we can
write
m−1
X
P (mk + r, k) = Pi (mk + r, k), (7)
i=2

3
where
(
the number of partitions (λ1 , λ2 , . . . , λi ) of n with k < λ• < n, if 0 < k < n;
Pi (n, k) =
0, otherwise.

For a fixed number of parts i ∈ {2, 3, . . . , m − 1}, to count Pi (mk + r, k), we first delete k
from each of the i parts. This gives rise to an equivalent counting of an unordered partition
of mk + r − ik = (m − i)k + r, yielding Pi (mk + r, k) = pi ((m − i)k + r).
The following lemma details precise values and certain relations of ps (n) for small s.
Trivial proofs of the first two assertions are omitted while the remaining two assertions are
quoted from [2].
Lemma 7. For n ≥ 0, we have
1) p2 (n) = ⌊n/2⌋.
2) p3 (n) = hn2 /12i, where hzi denotes the nearest integer of z ∈ R.
3) ([2, Proposition 5]) For n ≥ 3, we have
   
n+1 n+2
p3 (n) − p3 (n − 1) = − .
2 3

4) ([2, Proposition 9]) For n ≥ 5, we have

(
n
2
+ 1, if 2 | n;
p4 (n) − p4 (n − 1) = p3 (χ(n)), where χ(n) = n−1
2
, otherwise.

3 Proof of Theorem 2
From Lemma 4, we have
P (n, k + 1) = P (n, k) − P (n − k − 1, k) − 1, (8)
which shows that the second argument can be successively reduced to the case k = 1, i.e.,
each P (n, k) can be uniquely written as a Z-linear combination of finitely many elements
taken from P (n, 1), P (n − 1, 1), P (n − 2, 1), . . .. Applying the result of Lemma 5, the values
P (n, 1), P (n − 1, 1), P (n − 2, 1), . . . can be furthered uniquely reduced to the values of the
unrestricted partition function p(n), p(n − 1), p(n − 2), . . ., which implies that each P (n, k)
can be uniquely written as a Z-linear combination of finitely many elements taken from
p(n), p(n − 1), p(n − 2), . . .. To determine this explicitly, it is more convenient to write
∞
X
P (n, k) − P (n − 1, k) = ak,i p(n − i), (9)
i=0

4
where ak,i ∈ Z; this sum is finite because p(m) = 0 for integers m < 0. To compute the
coefficients ak,i , we resort to the use of generating series
∞
X
fkn (x) = ak,i xn−i . (10)
i=0

We claim that
k
Y
f1n (x) n
= x − 2x n−1
+x n−2
, fkn (x) n
= (x − 2x n−1
+x n−2
) (1 − x−i ) (2 ≤ k < n). (11)
i=2

For the case k = 1, Lemma 5 gives

P (n, 1) − P (n − 1, 1) = (p(n) − p(n − 1) − 1) − (p(n − 1) − p(n − 2) − 1)

= p(n) − 2p(n − 1) + p(n − 2),

yielding a1,0 = 1, a1,1 = −2, a1,2 = 1 and a1,n = 0 for all n ≥ 3, and so (11) holds when
k = 1. For 2 ≤ k < n, using Lemma 4, we get

P (n, k − 1) − P (n − 1, k − 1)
= (P (n, k) + P (n − k, k − 1) + 1) − (P (n − 1, k) + P (n − 1 − k, k − 1) + 1)
= (P (n, k) − P (n − 1, k)) + (P (n − k, k − 1) − P (n − k − 1, k − 1)) .

Using (9) to replace the differences of two restricted partitions, we get

∞
X ∞
X ∞
X
ak−1,i p(n − i) = ak,i p(n − i) + ak−1,i p(n − k − i). (12)
i=0 i=0 i=0

We claim now that

n
fk−1 n−k
(x) = fkn (x) + fk−1 (x). (13)
Using (10), the relation (13) holds if and only if the following system of equations holds

ak−1,i = ak,i (i = 0, 1, . . . , k − 1)
ak−1,i = ak,i + ak−1,i−k (i = k, k + 1, k + 2, . . .).

This system is identical with the one obtained from (12) by equating the coefficients of
p(n − i). As the coefficients aj,i are uniquely determined, the system of equation does indeed
hold which in turn affirms the relation (13). Since

n−k 1 n
fk−1 (x) = ak−1,0 xn−k + ak−1,1 xn−k−1 + · · · = f (x),
xk k−1

5
substituting this into (13), we get fkn (x) = 1 − x1k fk−1

n
(x). Hence,
 
1
f2 (x) = 1 − 2 f1n (x), . . . ,
n
x
     k
n 1 1 1 n n n−1 n−2

Y
1 − x−i ,


fk (x) = 1 − k 1 − k−1 · · · 1 − 2 f1 (x) = x − 2x +x
x x x i=2

which proves the second assertion in (11).

4 Applications
To apply Theorem 2 to compute values of p(n) subject to the knowledge of preceding values,
we subdivide our treatment into two separate subcases corresponding to large and small
values of k relative to n.

4.1 The case of large k

Here, we obtain several applications for k = ⌊n/3⌋ , ⌊n/4⌋ and ⌊n/5⌋. In these cases, explicit
formulas for the right-hand expression of (2) can be easily determined using Lemmas 6 and
7.

Theorem 8. For n ≥ 4, we have

∞
(
X 1, if n is even;
a⌊n/3⌋,i p(n − i) = (14)
i=0
0, if n is odd.

where the coefficients a⌊n/3⌋,i are obtained from the expansion

⌊n/3⌋ ∞
n n−1 n−2

Y −i
X
a⌊n/3⌋,i xn−i ,


x − 2x +x 1−x = (15)
i=2 i=0

with empty product being taken as 1.

Proof. Write n = 3k + r for some k ∈ N ∪ {0} and 0 ≤ r < 3. From Lemma 6 and Lemma
7 part 1), we get

P (n, k) − P (n − 1, k) = P (3k + r, k) − P (3k + r − 1, k)

    (
k+r k+r−1 1, if k + r is even;
= p2 (k + r) − p2 (k + r − 1) = − = (16)
2 2 0, if k + r is odd.

Applying Theorem 2 with k = ⌊n/3⌋ and using (16), the result follows.

6
Example 9. As an illustration, we use Theorem 8 to compute p(9). Here, n = 9, k =
⌊n/3⌋ = 3. The left-hand expression of (15) is
3
Y
9 8 7
(x − 2x + x ) (1 − x−i ) = x9 − 2x8 + x6 + x5 − 2x3 + x2 ,
i=2

so that

a3,0 = 1, a3,1 = −2, a3,2 = 0, a3,3 = 1, a3,4 = 1, a3,5 = 0, a3,6 = −2, a3,7 = 1.

From (14), we have

0 = p(9) − 2p(8) + p(6) + p(5) − 2p(3) + p(2),

and using known values from (1), we get

p(9) = 2p(8) − p(6) − p(5) + 2p(3) − p(2) = 2(22) − 11 − 7 + 2(3) − 2 = 30.

Similarly, to find p(10) and p(11), we take n = 10 and n = 11 yielding in both cases
k = ⌊n/3⌋ = 3. Using (14), the left-hand expressions of (15) are, respectively,

x10 − 2x9 + x7 + x6 − 2x4 + x3 and x11 − 2x10 + x8 + x7 − 2x5 + x4 .

In view of (14) and using known preceding values from (1), we get

p(10) = 2p(9) − p(7) − p(6) + 2p(4) − p(3) + 1 = 2(30) − 15 − 11 + 2(5) − 3 + 1 = 42

p(11) = 2p(10) − p(8) − p(7) + 2p(5) − p(4) = 2(42) − 22 − 15 + 2(7) − 5 = 56.

For the remaining two cases, it is more convenient to work with k and r instead of n.

Theorem 10. For k ≥ 1 and 0 ≤ r < 4, we have

∞  
jrk 
  
X r−1 k+r+1 k+r+2
ak,i p(4k + r − i) = − + − (17)
i=0
2 2 2 3

where the coefficients ak,i are obtained from the expansion

k ∞
4k+r 4k+r−1 4k+r−2

Y −i
X
ak,i x4k+r−i ,


x − 2x +x 1−x = (18)
i=2 i=0

with empty product being taken as 1.

7
Proof. Writing n = 4k + r, 0 ≤ r < 4, using Lemma 6 and Lemma 7 parts 1) and 3), we
have

P (4k + r, k) − P (4k + r − 1, k) = p2 (2k + r) − p2 (2k + r − 1) + p3 (k + r) − p3 (k + r − 1)

       
2k + r 2k + r − 1 k+r+1 k+r+2
= − + −
2 2 2 3
jr k r − 1 k + r + 1 k + r + 2
= − + − .
2 2 2 3

Example 11. We illustrate the use of Theorem 10 to compute p(8). Here n = 8, k = 2 and
r = 0. The left-hand expression of (18) is
2
Y
8 7 6
(x − 2x + x ) (1 − x−i ) = x8 − 2x7 + 2x5 − x4 ,
i=2

so that

a2,0 = 1, a2,1 = −2, a2,2 = 0, a2,3 = 2, a2,4 = −1.

In view of (17), we have

       
0 0−1 2+0+1 2+0+2
1= − + − = p(8) − 2p(7) + 2p(5) − p(4),
2 2 2 3

and using known values from (1), we get

p(8) = 2p(7) − 2p(5) + p(4) + 1 = 2(15) − 2(7) + 5 + 1 = 22.

Similarly, to find p(9), p(10) and p(11), we take n = 9, 10 and n = 11. In the three cases we
have r = 1, 2, 3, and the left-hand expressions of (18) are, respectively,

x9 − 2x8 + 2x6 − x5 , x10 − 2x9 + 2x7 − x6 and x11 − 2x10 + 2x8 − x7 .

In view of (17), we have

       
1 1−1 2+1+1 2+1+2
1= − + − = p(9) − 2p(8) + 2p(6) − p(5),
2 2 2 3
       
2 2−1 2+2+1 2+2+2
1= − + − = p(10) − 2p(9) + 2p(7) − p(6),
2 2 2 3
       
3 3−1 2+3+1 2+3+2
1= − + − = p(11) − 2p(10) + 2p(8) − p(7),
2 2 2 3

8
and using known values from (1), we get

p(9) = 2p(8) − 2p(6) + p(5) + 1 = 2(22) − 2(11) + 7 + 1 = 30

p(10) = 2p(9) − 2p(7) + p(6) + 1 = 2(30) − 2(15) + 11 + 1 = 42
p(11) = 2p(10) − 2p(8) + p(7) + 1 = 2(42) − 2(22) + 15 + 1 = 56.

In passing, observe that to compute p(11), using Theorem 8 we need 5 preceding partitions
terms, while using Theorem 10 we need only 3 terms, which indicates that Theorem 10 seems
more effective than Theorem 8.

Theorem 12. For k ≥ 1 and 0 ≤ r < 5, we have

∞
(χ(k + r))2 
     
X  k+r r+1
ak,i p(5k + r − i) = k + + +
i=0
2 2 12
  k + r − 1   2k + r + 2  
− + (19)
2 3

where the coefficients ak,i are obtained from the expansion

k ∞
5k+r 5k+r−1 5k+r−2

Y −i
X
ak,i x5k+r−i


x − 2x +x 1−x = (20)
i=2 i=0

with empty product being taken as 1; the symbols hzi and χ(n) are defined as in Lemma 7
parts 2) and 4).

Proof. Writing n = 5k + r, 0 ≤ r < 5, using Lemma 6 and Lemma 7 parts 1), 2), 3) and 4),
we have

P (5k + r, k) − P (5k + r − 1, k)
= p2 (3k + r) − p2 (3k + r − 1) + p3 (2k + r) − p3 (2k + r − 1) + p4 (k + r) − p4 (k + r − 1)
       
3k + r 3k + r − 1 2k + r + 1 2k + r + 2
= − + − + p3 (χ(k + r))
2 2 2 3
(χ(k + r))2
         
k+r k+r−1 r+1 2k + r + 2
= − +k+ − + .
2 2 2 3 12

Example 13. We illustrate the use of Theorem 12 to compute p(10), p(11), p(12), p(13) and
p(14) with the following respective values:

n = 10, k = 2, r = 0; n = 11, k = 2, r = 1; n = 12, k = 2, r = 2;

n = 13, k = 2, r = 3; n = 14, k = 2, r = 4.

9
The left-hand expression of (20) is
2
Y
n n−1 n−2
(x − 2x +x ) (1 − x−i ) = xn − 2xn−1 + 2xn−3 − xn−4 ,
i=2

so that a2,0 = 1, a2,1 = −2, a2,2 = 0, a2,3 = 2, a2,4 = −1. From (19), we have
∞
(χ(2 + r))2 
   
 
X 2+r  r+1
a2,i p(10 + r − i) = 2 + + +
i=0
2 2 12

 2+r−1   
4+r+2 
− + =: Sr .
2 3

By simple computation, we get

S0 = 1, S1 = 1, S2 = 3, S3 = 1, S4 = 3.

Using known values from (1), we get

p(10) = 2p(9) − 2p(7) + p(6) + 1 = 2(30) − 2(15) + 11 + 1 = 42

p(11) = 2p(10) − 2p(8) + p(7) + 1 = 2(42) − 2(22) + 15 + 1 = 56
p(12) = 2p(11) − 2p(9) + p(8) + 3 = 2(56) − 2(30) + 22 + 3 = 77
p(13) = 2p(12) − 2p(10) + p(9) + 1 = 2(77) − 2(42) + 30 + 1 = 101
p(14) = 2p(13) − 2p(11) + p(10) + 3 = 2(101) − 2(56) + 42 + 3 = 135.

4.2 The case of small k

In this case, using a trick, the right-hand side of (2) can be slightly improved.

Theorem 14. For n ≥ 4 and k ≥ 1, we have

∞ ⌊n/(k+1)⌋
X X
ak,i p(n − i) = (pi (n − ki) − pi (n − ki − 1)) , (21)
i=0 i=2

where the coefficients ak,i are obtained from the expansions (3).

Proof. Write n = mk + r, 0 ≤ r < k. From Lemma 6, we have

m−1 ⌊n/(k+1)⌋
X X
P (n, k) = P (mk + r, k) = pi ((m − i)k + r) = pi (n − ki); (22)
i=2 i=2

10
since ps (n) counts the number of partitions of n into s parts, we must have s ≤ n. This
enables us to replace the upper limit of summation m − 1 = ⌊n/k⌋ − 1 by ⌊n/(k + 1)⌋ which
is better when k is small compared to n. Using Theorem 2 and (22), we have

∞ ⌊n/(k+1)⌋ ⌊(n−1)/(k+1)⌋
X X X
ak,i p(n − i) = pi (n − ki) − pi (n − 1 − ki),
i=0 i=2 i=2

and the result follows from the observation that lengthening the last upper limit of summation
yielding only zero value.

Example 15. We illustrate the use of Theorem 14 to compute p(14). If we take k = 2, then
⌊n/(k + 1)⌋ = ⌊14/3⌋ = 4. Thus, (21) and Lemma 7 give
∞
X 4
X
a2,i p(14 − i) = (pi (14 − 2i) − pi (13 − 2i))
i=0 i=2
= p2 (10) − p2 (9) + p3 (8) − p3 (7) + p4 (6) − p4 (5)
= ⌊10/3⌋ − ⌊9/2⌋ + ⌊9/2⌋ − ⌊10/2⌋ + p3 (χ(6)) = 3.

In view of (3), we have

2 ∞
14 13 12

Y −i 14 13 11 10
X
a2,i x14−i .


x − 2x + x 1−x = x − 2x + 2x − x =
i=2 i=0

Theorem 14 yields

p(14) = 2p(13) − 2p(11) + p(10) + 3 = 2(101) − 2(56) + 42 + 3 = 135.

If we take k = 3, then ⌊14/(3 + 1)⌋ = 3. From (21), we have

∞
X 3
X
a3,i p(14 − i) = (pi (14 − 3i) − pi (13 − 3i)) = p2 (8) − p2 (7) + p3 (5) − p3 (4) = 2.
i=0 i=2

In view of (3), we have

3 ∞
14 13 12

Y −i 14 13 11 10 8 7
X
a3,i x14−i .


x − 2x + x 1−x = x − 2x + x + x − 2x + x =
i=2 i=0

Theorem 14 and preceding values give

p(14) = 2p(13) − p(11) − p(10) + 2p(8) − p(7) + 2

= 2(101) − (56) − 42 + 2(22) − 15 + 2 = 135.

11
5 Acknowledgments
The authors gratefully acknowledge a financial support under the Basic Research Fund
(BRF) provided by the Faculty of Science, Kasetsart University.

References
[1] T. M. Apostol, Introduction to Analytic Number Theory, Springer, 1976.

[2] M. Barnabei, F. Bonetti, and M. Silimbani, Bijections and recurrences for integer parti-
tions into a bounded number of parts. Appl. Math. Lett. 22 (2009), 297–303.

[3] J. L. De Carufel, A few identities involving partitions with a fixed number of parts, Ars
Comb. 68 (2003), 125–130.

[4] M. Merca, A generalization of Euler’s pentagonal number recurrence for the partition
function, Ramanujan J. 37 (2015), 589–595.

[5] M. Merca and M. D. Schmidt, The partition function p(n) in terms of the classical Möbius
function, Ramanujan J. 49 (2019), 87–96.

[6] N. J. A. Sloane et al., The On-line Encyclopedia of Integer Sequences, 2020. Available
at https://oeis.org/.

2020 Mathematics Subject Classification: Primary 05A17; Secondary 11P81.

Keywords: recursion formula, restricted partition function.

(Concerned with sequences A000041 and A008284.)

Received March 23 2021; revised version received May 20 2021. Published in Journal of
Integer Sequences, May 20 2021.

Return to Journal of Integer Sequences home page.

12