Maths Olympiad Prep

Library / /23 of 55

, 2019

Algebra Difficulty 8.7 Shortlist Prove it IMO

Let n3n \geqslant 3 be a positive integer and let (a1,a2,,an)(a_{1}, a_{2}, \ldots, a_{n}) be a strictly increasing sequence of nn positive real numbers with sum equal to 22. Let XX be a subset of {1,2,,n}\{1,2, \ldots, n\} such that the value of
1iXai \left|1-\sum_{i \in X} a_{i}\right|
is minimised. Prove that there exists a strictly increasing sequence of nn positive real numbers (b1,b2,,bn)(b_{1}, b_{2}, \ldots, b_{n}) with sum equal to 22 such that
iXbi=1 \sum_{i \in X} b_{i}=1

Solutions — 4

Solution 1

Solution 1. Without loss of generality, assume iXai1\sum_{i \in X} a_{i} \leqslant 1, and we may assume strict inequality as otherwise bi=aib_{i}=a_{i} works. Also, XX clearly cannot be empty.
If nXn \in X, add Δ\Delta to ana_{n}, producing a sequence of cic_{i} with iXci=iXcci\sum_{i \in X} c_{i}=\sum_{i \in X^{c}} c_{i}, and then scale as described above to make the sum equal to 22. Otherwise, there is some kk with kXk \in X and k+1Xck+1 \in X^{c}. Let δ=ak+1ak\delta=a_{k+1}-a_{k}.
- If δ>Δ\delta>\Delta, add Δ\Delta to aka_{k} and then scale.
- If δ<Δ\delta<\Delta, then considering X{k+1}\{k}X \cup\{k+1\} \backslash\{k\} contradicts XX being (ai)(a_{i})-minimising.
- If δ=Δ\delta=\Delta, choose any jk,k+1j \neq k, k+1 (possible since n3n \geqslant 3), and any ϵ\epsilon less than the least of a1a_{1} and all the differences ai+1aia_{i+1}-a_{i}. If jXj \in X then add Δϵ\Delta-\epsilon to aka_{k} and ϵ\epsilon to aja_{j}, then scale; otherwise, add Δ\Delta to aka_{k} and ϵ/2\epsilon / 2 to ak+1a_{k+1}, and subtract ϵ/2\epsilon / 2 from aja_{j}, then scale.

Solution 2

Solution 2. This is similar to Solution 1, but without scaling. As in that solution, without loss of generality, assume iXai<1\sum_{i \in X} a_{i}<1.
Suppose there exists 1jn11 \leqslant j \leqslant n-1 such that jXj \in X but j+1Xcj+1 \in X^{c}. Then aj+1ajΔa_{j+1}-a_{j} \geqslant \Delta, because otherwise considering X{j+1}\{j}X \cup\{j+1\} \backslash\{j\} contradicts XX being (ai)(a_{i})-minimising.
If aj+1aj>Δa_{j+1}-a_{j}>\Delta, put
bi={aj+Δ/2, if i=jaj+1Δ/2, if i=j+1ai, otherwise  b_{i}= \begin{cases}a_{j}+\Delta / 2, & \text{ if } i=j \\ a_{j+1}-\Delta / 2, & \text{ if } i=j+1 \\ a_{i}, & \text{ otherwise }\end{cases}
If aj+1aj=Δa_{j+1}-a_{j}=\Delta, choose any ϵ\epsilon less than the least of Δ/2,a1\Delta / 2, a_{1} and all the differences ai+1aia_{i+1}-a_{i}. If X2|X| \geqslant 2, choose kXk \in X with kjk \neq j, and put
bi={aj+Δ/2ϵ, if i=jaj+1Δ/2, if i=j+1ak+ϵ, if i=kai, otherwise  b_{i}= \begin{cases}a_{j}+\Delta / 2-\epsilon, & \text{ if } i=j \\ a_{j+1}-\Delta / 2, & \text{ if } i=j+1 \\ a_{k}+\epsilon, & \text{ if } i=k \\ a_{i}, & \text{ otherwise }\end{cases}
Otherwise, Xc2|X^{c}| \geqslant 2, so choose kXck \in X^{c} with kj+1k \neq j+1, and put
bi={aj+Δ/2, if i=j;aj+1Δ/2+ϵ, if i=j+1;akϵ, if i=k;ai, otherwise.  b_{i}= \begin{cases}a_{j}+\Delta / 2, & \text{ if } i=j ; \\ a_{j+1}-\Delta / 2+\epsilon, & \text{ if } i=j+1 ; \\ a_{k}-\epsilon, & \text{ if } i=k ; \\ a_{i}, & \text{ otherwise. }\end{cases}
If there is no 1jn1 \leqslant j \leqslant n such that jXj \in X but j+1Xcj+1 \in X^{c}, there must be some 1<kn1<k \leqslant n such that X=[k,n]X=[k, n] (certainly XX cannot be empty). We must have a1>Δa_{1}>\Delta, as otherwise considering X{1}X \cup\{1\} contradicts XX being (ai)(a_{i})-minimising. Now put
bi={a1Δ/2, if i=1an+Δ/2, if i=nai, otherwise  b_{i}= \begin{cases}a_{1}-\Delta / 2, & \text{ if } i=1 \\ a_{n}+\Delta / 2, & \text{ if } i=n \\ a_{i}, & \text{ otherwise }\end{cases}

Solution 3

Solution 3. Without loss of generality, assume iXai1\sum_{i \in X} a_{i} \leqslant 1, so Δ0\Delta \geqslant 0. If Δ=0\Delta=0 we can take bi=aib_{i}=a_{i}, so now assume that Δ>0\Delta>0.
Suppose that there is some knk \leqslant n such that X[k,n]>Xc[k,n]|X \cap[k, n]|>|X^{c} \cap[k, n]|. If we choose the largest such kk then X[k,n]Xc[k,n]=1|X \cap[k, n]|-|X^{c} \cap[k, n]|=1. We can now find the required sequence (bi)(b_{i}) by starting with ci=aic_{i}=a_{i} for i<ki<k and ci=ai+Δc_{i}=a_{i}+\Delta for iki \geqslant k, and then scaling as described above.
If no such kk exists, we will derive a contradiction. For each iXi \in X we can choose i<jini<j_{i} \leqslant n in such a way that jiXcj_{i} \in X^{c} and all the jij_{i} are different. (For instance, note that necessarily nXcn \in X^{c} and now just work downwards; each time an iXi \in X is considered, let jij_{i} be the least element of XcX^{c} greater than ii and not yet used.) Let YY be the (possibly empty) subset of [1,n][1, n] consisting of those elements in XcX^{c} that are also not one of the jij_{i}. In any case
Δ=iX(ajiai)+jYaj \Delta=\sum_{i \in X}\left(a_{j_{i}}-a_{i}\right)+\sum_{j \in Y} a_{j}
where each term in the sums is positive. Since n3n \geqslant 3 the total number of terms above is at least two. Take a least such term and its corresponding index ii and consider the set ZZ which we form from XX by removing ii and adding jij_{i} (if it is a term of the first type) or just by adding jj if it is a term of the second type. The corresponding expression of Δ\Delta for ZZ has the sign of its least term changed, meaning that the sum is still nonnegative but strictly less than Δ\Delta, which contradicts XX being (ai)(a_{i})-minimising.

Solution 4

Solution 4. This uses some similar ideas to Solution 3, but describes properties of the index sets XX that are sufficient to describe a corresponding sequence (bi)(b_{i}) that is not derived from (ai)(a_{i}).
Note that, for two subsets X,YX, Y of [1,n][1, n], the following are equivalent:
- X[i,n]Y[i,n]|X \cap[i, n]| \leqslant|Y \cap[i, n]| for all 1in1 \leqslant i \leqslant n;
- YY is at least as large as XX, and for all 1jY1 \leqslant j \leqslant|Y|, the jthj^{\text{th}} largest element of YY is at least as big as the jthj^{\text{th}} largest element of XX;
- there is an injective function f:XYf: X \rightarrow Y such that f(i)if(i) \geqslant i for all iXi \in X.
If these equivalent conditions are satisfied, we write XYX \leq Y. We write X<YX<Y if XYX \leq Y and XYX \neq Y.
Note that if XYX \prec Y, then iXai<iYai\sum_{i \in X} a_{i}<\sum_{i \in Y} a_{i} (the second description above makes this clear).
We claim first that, if n3n \geqslant 3 and XXcX \prec X^{c}, then there exists YY with XYXcX \prec Y \prec X^{c}. Indeed, as XXc|X| \leqslant|X^{c}|, we have Xc2|X^{c}| \geqslant 2. Define YY to consist of the largest element of XcX^{c}, together with all but the largest element of XX; it is clear both that YY is distinct from XX and XcX^{c}, and that XYXcX \leq Y \leq X^{c}, which is what we need.
But, in this situation, we have
iXai<iYai<iXcai and 1iXai=(1iXcai), \sum_{i \in X} a_{i}<\sum_{i \in Y} a_{i}<\sum_{i \in X^{c}} a_{i} \quad \text{ and } \quad 1-\sum_{i \in X} a_{i}=-\left(1-\sum_{i \in X^{c}} a_{i}\right),
so 1iYai<1iXai\left|1-\sum_{i \in Y} a_{i}\right|<\left|1-\sum_{i \in X} a_{i}\right|.
Hence if XX is (ai)(a_{i})-minimising, we do not have X<XcX<X^{c}, and similarly we do not have Xc<XX^{c}<X.
Considering the first description above, this immediately implies the following Claim.
Claim. There exist 1k,n1 \leqslant k, \ell \leqslant n such that X[k,n]>nk+12|X \cap[k, n]|>\frac{n-k+1}{2} and X[,n]<n+12|X \cap[\ell, n]|<\frac{n-\ell+1}{2}.
We now construct our sequence (bi)(b_{i}) using this claim. Let kk and \ell be the greatest values satisfying the claim, and without loss of generality suppose k=nk=n and <n\ell<n (otherwise replace XX by its complement). As \ell is maximal, nn-\ell is even and X[,n]=n2|X \cap[\ell, n]|=\frac{n-\ell}{2}. For sufficiently small positive ϵ\epsilon, we take
bi=iϵ+{0, if i<δ, if in1γ, if i=n b_{i}=i \epsilon+ \begin{cases}0, & \text{ if } i<\ell \\ \delta, & \text{ if } \ell \leqslant i \leqslant n-1 \\ \gamma, & \text{ if } i=n\end{cases}
Let M=iXiM=\sum_{i \in X} i. So we require
Mϵ+(n21)δ+γ=1 M \epsilon+\left(\frac{n-\ell}{2}-1\right) \delta+\gamma=1
and
n(n+1)2ϵ+(n)δ+γ=2 \frac{n(n+1)}{2} \epsilon+(n-\ell) \delta+\gamma=2
These give
γ=2δ+(n(n+1)22M)ϵ \gamma=2 \delta+\left(\frac{n(n+1)}{2}-2 M\right) \epsilon
and for sufficiently small positive ϵ\epsilon, solving for γ\gamma and δ\delta gives 0<δ<γ0<\delta<\gamma (since ϵ=0\epsilon=0 gives δ=1/(n2+1)\delta=1 /\left(\frac{n-\ell}{2}+1\right) and γ=2δ\gamma=2 \delta), so the sequence is strictly increasing and has positive values.

Want a route through all this instead of an archive? The track puts 2,000 problems in a working order, from AMC 10 level to the IMO shortlist.

Source: MathNet, licensed CC-BY-4.0. Statement and solution reproduced as published; topic and difficulty added by this site.