Let S be a set of three integers chosen from N = { 1 , 2 , … , 2 0 1 2 } . We say that S is 4-splittable if there is an n ∈ N ∖ S such that S ∪ { n } can be partitioned into 2 sets such that the sum of each set is the same. How many 3-element subsets of N are not 4-splittable?
Details and assumptions
You can refer to Set Notation for the definition of N ∖ S .
This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try
refreshing the page, (b) enabling javascript if it is disabled on your browser and,
finally, (c)
loading the
non-javascript version of this page
. We're sorry about the hassle.
Let S be a non-4-splittable set. Write S = { a , b , c } with a < b < c . Note that d = a + c − b satisfies a < d < c , and d would split S if it were in N ∖ S since a + c = b + d , so the only possibility is that d = b . So a + c = 2 b . Taking x = b − a , we can write S = { a , a + x , a + 2 x } .
Note that a − x would split S if it were in N ∖ S , since ( a − x ) + ( a + 2 x ) = a + ( a + x ) . Since a − x < a , the only possibility is that this integer is ≤ 0 , so x ≥ a .
But now note that x − a would split S if it were in N ∖ S , since ( x − a ) + a + ( a + x ) = a + 2 x . Since 0 ≤ x − a < a + x , there are only two possibilities left: x − a = 0 or x − a = a . This leads to S = { a , 2 a , 3 a } and S = { a , 3 a , 5 a } respectively.
It's easy to check in the first case that S is not splittable if and only if 3 a ≤ 2 0 1 2 (to make it a subset of N ) and 4 a ≥ 2 0 1 3 (so that it doesn't split S ), and in the second case S is not splittable if and only if 5 a ≤ 2 0 1 2 and 7 a ≥ 2 0 1 3 .
In the first case the admissible values of a are between 5 0 4 and 6 7 0 inclusive, and in the second case the admissible values are between 2 8 8 and 4 0 2 inclusive. This gives 1 6 7 + 1 1 5 = 2 8 2 sets in all.
Main idea : Brute force all numbers that a 4-splittable set candidate might "take" in order to form a set that can be split into equal sums. For example, with a set { a , b , c } and a < b < c , the number a + c − b might make the set to be a 4-splittable.
Complete proof :
Let's call a set of three integers from N be a 4-splittable set candidate, or candidate for short. For a candidate set S , let's call a number n (not necessarily in N \ S ) a witness if S ⊕ { n } is a 4-element multiset that can be partitioned into two sets with equal sums. A multiset is simply a set where the elements might repeat (like { 1 , 1 , 2 } ; as a set is it indistinguishable with { 1 , 2 } ). For a candidate set S and a witness n , call a partition of S ⊕ { n } into two subsets with equal sum as a split .
A set S is not 4-splittable if and only if its witnesses are not in N \ S . Call a witness that is in N \ S as a proof , and a fake otherwise.
Suppose that our candidate set is S = { a , b , c } where a < b < c ; we assume that it is not 4-splittable. Note that a + c − b is a witness as { a , c } , { b , a + c − b } is a split. Also, since a < b < c , we have a + c − b > a + b − b = a and a + c − b < b + c − b = c , so a < a + c − b < c . Every number between a and c must belong to N , and we have shown that a + c − b = a , c , so a + c − b is a proof unless b = a + c − b in which a + c − b ∈ S . Since we want to make all witnesses fakes, we need to have b = a + c − b . Then we can let b − a = k , so c − b = k and S = { b − k , b , b + k } .
Next, 2 k − b is a witness as { b + k } , { b , b − k , 2 k − b } is a split. Also, since b , b − k > 0 and b + k = b + ( b − k ) + ( 2 k − b ) , we must have 2 k − b < b + k . So this witness is a proof except in three cases:
Then b = k and S = { 0 , k , 2 k } , impossible.
Then b = 2 3 k . Letting k = 2 m , we have b = 3 m and so S = { m , 3 m , 5 m } .
Then b − 2 k ≥ 0 . Note that b − 2 k is also a witness as { b + k , b − 2 k } , { b , b − k } is a split, and since k > 0 , we have b − 2 k < b − k . Thus b − 2 k is a proof except when b − 2 k ≤ 0 . Combined with our assumption of this case, we need b − 2 k = 0 , so b = 2 k and so S = { k , 2 k , 3 k } .
Next, note that b + 2 k is a witness as { b + 2 k , b − k } , { b + k , b } is a split. This is equal to 7 m in case 2 or 4 k in case 3. Since b + 2 k > b + k , this is a proof unless b + 2 k > 2 0 1 2 . Thus we need 7 m > 2 0 1 2 ⟹ m ≥ 2 8 8 in case 2 and 4 k > 2 0 1 2 ⟹ k ≥ 5 0 4 in case 3.
Now, we also need b + k ≤ 2 0 1 2 . This means 5 m ≤ 2 0 1 2 ⟹ m ≤ 4 0 2 in case 2 and 3 k ≤ 2 0 1 2 ⟹ k ≤ 6 7 0 in case 3. So from case 2 we have 2 8 8 ≤ m ≤ 4 0 2 , for a total of 4 0 2 − 2 8 8 + 1 = 1 1 5 sets, and from case 3 we have 5 0 4 ≤ k ≤ 6 7 0 , for a total of 6 7 0 − 5 0 4 + 1 = 1 6 7 sets. It just remains to prove that they are indeed not 4-splittable.
Suppose we have a set S = { a , b , c } with 0 < a < b < c , a witness n , and a split into A , B where n ∈ A . Then there are only eight possible witnesses. Each of a , b , c goes into some of A , B , so each element has 2 choices; combined overall we have 2 3 = 8 possibilities. Also, after we have assigned a , b , c into the subsets, the value of n is unique: it's simply the sum of elements in B , minus the sum of elements in A not including n . Also, we can see that each term either contributes positively or negatively into the value of n , so the value of n can be expressed as ( − 1 ) p a + ( − 1 ) q b + ( − 1 ) r c where p , q , r are integers.
However, some witnesses are trivially excluded as potential proofs. The witnesses − a − b − c , a − b − c , − a + b − c are excluded as they must be less than zero. For example, for a − b − c , since a < c we have a − c < 0 , and adding with − b < 0 we have a − b − c < 0 . The other two cases are similar.
On the other end of the spectrum, if a + b + c is a proof then − a + b + c is also a proof. We prove this by contraposition: if − a + b + c is fake then a + b + c is fake. Since − a + b + c > − a + a + c = c , the only way − a + b + c can be a fake is if − a + b + c > 2 0 1 2 , but if so then a + b + c > − a + b + c > 2 0 1 2 is also a fake. So this makes our search to only four witnesses: a + b − c , − a − b + c , a − b + c , − a + b + c .
For case 2, namely S = { a , b , c } = { m , 3 m , 5 m } where 2 8 8 ≤ m ≤ 4 0 2 , the four witnesses we need to check are a + b − c = − m < 0 , − a − b + c = m ∈ S , a − b + c = 3 m ∈ S , or − a + b + c = 7 m ≥ 7 ⋅ 2 8 8 > 2 0 1 2 .
For case 3, namely S = { a , b , c } = { k , 2 k , 3 k } where 5 0 4 ≤ k ≤ 6 7 0 , the four witnesses we need to check are a + b − c = 0 , − a − b + c = 0 , a − b + c = 2 k ∈ S , or − a + b + c = 4 k ≥ 4 ⋅ 5 0 4 > 2 0 1 2 .
So we have shown that all these 1 1 5 + 1 6 7 sets are indeed not 4-splittable, and so the number of non-4-splittable sets is 1 1 5 + 1 6 7 = 2 8 2 .
Motivation : Well, this one is pretty straightforward, just bash all possible witnesses. To be honest, I didn't think of checking them when I did the problem; those last 5 paragraphs were written on the spot when I realized that there are only 8 possible witnesses for any candidate set.
Generalization : Of course, changing 2 0 1 2 to some other value is pretty trivial. It will simply change the bounds 2 8 8 , 4 0 2 , 5 0 4 , 6 7 0 . What's more interesting is changing 3 , as in the number of elements of S . However, for lower values, it becomes rather trivial (for 1 element all subsets are not "2-splittable" as the only two witnesses are the element itself or its negation which is obviously negative; for 2 elements the interesting witnesses are simply the sum and the difference, and a simple casework like above will easily lead to { k , 2 k } where 2 k ≤ 2 0 1 2 < 3 k ). And for higher values, it becomes a boring casework of checking the witnesses. An interesting result is that the sets { k , 2 k , … , n k } where ( n + 1 ) k > 2 0 1 2 and { m , 3 m , … , ( 2 n − 1 ) m } where ( 2 n + 1 ) m > 2 0 1 2 are always solutions, although there might be more; I'm already feeling lazy to check whether there are more. Or perhaps I miss something that can simplify it...
Clearly my solution is too rigorous compared to the (currently) two other solutions. Whee.
Suppose S={a,b,c} such that 1 ≤ a < b < c ≤ 2 0 1 2 . Therefore, S is not 4-splittable if all of the following conditions are satisfied:
a + b − c < 1 (1)
c + a − b = b (2)
c + b − a > 2 0 1 2 (3)
c − a − b = a or c − a − b = b or c − a − b < 1 (4)
From (2), we have c = 2 b − a , subtitute to (1),(4), we have
b − 2 a > − 1 (1')
b − 2 a = a or b − 2 a = b or b − 2 a < 1 (4')
b − 2 a = b leads to a = 0 , so we only have to consider 2 cases:
Case 1: b − 2 a = a , so b = 3 a and from (2), c = 5 a
Subtitute these equations to (3) and c ≤ 2 0 1 2 , we have 2 8 8 ≤ a ≤ 4 0 2
Therefore, there are 4 0 2 − 2 8 8 + 1 = 1 1 5 results in this case.
Case 2: b − 2 a < 1 . From (1'), we have b − 2 a = 0 , so b = 2 a and c = 3 a .
Subtitute these equations to (3) and c ≤ 2 0 1 2 , we have 5 0 4 ≤ a ≤ 6 7 0
Therefore, there are 6 7 0 − 5 0 4 + 1 = 1 6 7 results in this case.
The answer is 1 1 5 + 1 6 7 = 2 8 2 .
Problem Loading...
Note Loading...
Set Loading...
Let the desired 3 -element subset be S = a , b , c where a < b < c . Thus for any n ∈ N ∖ S , S ∪ { n } cannot be split into 2 sets such that the sum of each set is the same. First, S ∪ { n } cannot be split into 2 sets of 2 . Thus there does not exist n ∈ N ∖ S for any of the following to be true: a + b = c + n , b + c = a + n , c + a = b + n .
Thus each of a + b − c , b + c − a , c + a − b must be either out of the domain [ 1 , 2 0 1 2 ] or be equal to one of a , b , c . Since 1 ≤ a < c + a − b < c ≥ 2 0 1 2 , c + a − b must be equal to b . So c + a − b = b ⇒ c + a = 2 b . Now we let b = a + k , c = a + 2 k where k is a positive integer. Then since a + b − c < a and b + c − a > c must be out of the domain [ 1 , 2 0 1 2 ] , we must have a + b − c < 1 , b + c − a > 2 0 1 2 ⇒ a + a + k − ( a + 2 k ) < 1 , a + k + a + 2 k − a > 2 0 1 2 , which solves to 2 0 1 2 − 3 k < a ≤ k .
Now we consider the other possible way of partitioning the 4 -element set. If the 4 -element set is partitioned into 1 -element set and a 3 -element set such that the sum of each is equal, the element in the 1 -element set must be the biggest element among the 4 because they are all positive integers. Thus we cannot have any n ∈ N ∖ S such that any of the following is true: a + b + c = n , a + b + n = c .
So each of a + b + c , c − a − b must be either out of the range [ 1 , 2 0 1 2 ] or equal to one of a , b , c . Thus a + b + c > 2 0 1 2 ⇒ 3 a + 3 k > 2 0 1 2 ⇒ a ≥ 6 7 1 − k and either c − a − b = a ⇒ a + 2 k − a − ( a + k ) = a ⇒ a = 2 k or c − a − b = b ⇒ a + 2 k − a − ( a + k ) = a + k ⇒ a = 0 (reject) or c − a − b < 1 ⇒ a + 2 k − a − ( a + k ) ≤ 0 ⇒ a ≥ k ⇒ a = k because a ≤ k from earlier. As a result, a = k or a = 2 k , under the restriction 2 0 1 2 − 3 k < a , a ≥ 6 7 1 − k , and a + 2 k ≤ 2 0 1 2 .
When a = k , 2 0 1 2 − 3 k < a gives k ≥ 5 0 4 , a ≥ 6 7 1 − k gives k ≥ 3 3 6 , and a + 2 k ≤ 2 0 1 2 gives k ≤ 6 7 0 . So 5 0 4 ≤ k ≤ 6 7 0 , and for each k there is only one set { a , b , c } , thus there are 6 7 0 − 5 0 4 + 1 = 1 6 7 sets. [This gives us sets of the form { a , 2 a , 3 a } . - Calvin]
When a = 2 k , 2 0 1 2 − 3 k < a gives k ≥ 5 7 5 , a ≥ 6 7 1 − k gives k ≥ 4 4 8 , and a + 2 k ≤ 2 0 1 2 gives k ≤ 8 0 4 . So 5 7 5 ≤ k ≤ 8 0 4 . Since k must be even, there are 2 8 0 4 − 5 7 6 + 1 = 1 1 5 sets. [This gives us sets of the form { a , 3 a , 5 a } . - Calvin]
Thus there are 1 6 7 + 1 1 5 = 2 8 2 sets in total.