Unofficial archive — problems, solutions & results © IMC, reproduced with permission.

IMC / 2012 / Problems / Day 2, P10

IMC 2012 · Day 2 · P10

killer

Let c1c \ge 1 be a real number. Let GG be an abelian group and let AGA \subset G be a finite set satisfying A+AcA|A + A| \le c |A|, where X+Y:={x+yxX, yY}X + Y := \{ x + y \mid x \in X,\ y \in Y \} and Z|Z| denotes the cardinality of ZZ. Prove that A+A++Ak timesckA|\underbrace{A + A + \dots + A}_{k\ \text{times}}| \le c^k |A| for every positive integer kk. (Plünnecke's inequality)

(Proposed by Przemyslaw Mazur, Jagiellonian University)

Solution (official)

Let BB be a nonempty subset of AA for which the value of the expression c1=A+BBc_1 = \frac{|A + B|}{|B|} is the least possible. Note that c1cc_1 \le c since AA is one of the possible choices of BB.

Lemma 1. For any finite set DGD \subset G we have A+B+Dc1B+D|A + B + D| \le c_1 |B + D|.

Proof. Apply induction on the cardinality of DD. For D=1|D| = 1 the Lemma is true by the definition of c1c_1. Suppose it is true for some DD and let xDx \notin D. Let B1={yBx+yB+D}B_1 = \{ y \in B \mid x + y \in B + D \}. Then B+(D{x})B + (D \cup \{x\}) decomposes into the union of two disjoint sets: B+(D{x})=(B+D)((BB1)+{x})B + (D \cup \{x\}) = (B + D) \cup \bigl( (B \setminus B_1) + \{x\} \bigr) and therefore B+(D{x})=B+D+BB1|B + (D \cup \{x\})| = |B + D| + |B| - |B_1|. Now we need to deal with the cardinality of the set A+B+(D{x})A + B + (D \cup \{x\}). Writing A+B+(D{x})=(A+B+D)(A+B1+{x})A + B + (D \cup \{x\}) = (A + B + D) \cup (A + B_1 + \{x\})

we count some of the elements twice: for example if yB1y \in B_1, then A+{y}+{x}(A+B+D)(A+B+{x})A + \{y\} + \{x\} \subset (A + B + D) \cap (A + B + \{x\}). Therefore all the elements of the set A+B1+{x}A + B_1 + \{x\} are counted twice and thus A+B+(D{x})A+B+D+A+B+{x}A+B1+{x}==A+B+D+A+BA+B1c1(B+D+BB1)=c1B+(D{x}),\begin{align*} |A + B + (D \cup \{x\})| &\le |A + B + D| + |A + B + \{x\}| - |A + B_1 + \{x\}| = \\ &= |A + B + D| + |A + B| - |A + B_1| \le c_1 (|B + D| + |B| - |B_1|) = c_1 |B + (D \cup \{x\})|, \end{align*} where the last inequality follows from the inductive hypothesis and the observation that A+BB=c1A+B1B1\frac{|A + B|}{|B|} = c_1 \le \frac{|A + B_1|}{|B_1|} (or B1B_1 is the empty set).

Lemma 2. For every k1k \ge 1 we have A++Ak times+Bc1kB|\underbrace{A + \dots + A}_{k\ \text{times}} + B| \le c_1^k |B|.

Proof. Induction on kk. For k=1k = 1 the statement is true by definition of c1c_1. For greater kk set D=A++Ak1 timesD = \underbrace{A + \dots + A}_{k-1\ \text{times}} in the previous lemma: A++Ak times+Bc1A++Ak1 times+Bc1kB|\underbrace{A + \dots + A}_{k\ \text{times}} + B| \le c_1 |\underbrace{A + \dots + A}_{k-1\ \text{times}} + B| \le c_1^k |B|.

Now notice that A++Ak timesA++Ak times+Bc1kBckA.|\underbrace{A + \dots + A}_{k\ \text{times}}| \le |\underbrace{A + \dots + A}_{k\ \text{times}} + B| \le c_1^k |B| \le c^k |A|. Remark. The proof above due to Giorgios Petridis and can be found at http://gowers.wordpress.com/2011/02/10/a-new-way-of-proving-sumset-estimates/

How the field did

contestants scored
313
average (of 10)
0.03
solved (≥ 80%)
0.0%
near-0 (≤ 10%)
99.7%
discrimination
0.16

Score distribution (field cohort)

Computed on contestants with a meaningful total (field cohort); discrimination is the corrected item–total correlation.

Similar problems

IMC 2007 · Day 1 · P4hardavg 2.0/10 · solved 16% · near-0 69% · disc 0.61
IMC 1999 · Day 2 · P11killeravg 0.3/10 · solved 0% · near-0 90% · disc 0.41
IMC 2005 · Day 1 · P6killeravg 0.2/10 · solved 0% · near-0 94% · disc 0.26