This page contains explanations pertaining to the "'Obvious Remarks'" section of Roth's paper.
Alternative definition of A-set
$A(x)$ is equal to the greatest number of integers that can be selected from $x$ consecutive terms of an arithmetic progression to form an $\mathcal{A}$-set
This is easy to see by taking $x$ consecutive terms in arithmetic progression:
(1)and corresponding them to $x$ consecutive integers
(2)Then as $(a+bu_k) - (a+bu_l) = b(u_k-u_l)$ we see that a three-term sequence in $\mathcal{A}_1$ is arithmetic if and only if the corresponding three term sequence in $\mathcal{A}_2$ is arithmetic.
A(x+y) is less than A(x)+A(y)
For any two positive integers $x$ and $y$ we have that $A(x+y) \leq A(x) + A(y)$.
First we recall the definition of $A(x+y)$:
(3)and similarly
(4)We now apply the alternative definition of $A(y)$ to write
(5)Now let $S_1$ be a subset of $\{1, 2, \ldots, x+y\}$ that is a maximal $\mathcal{A}$-set, so that $|S_1| = A(x+y)$. Now a subset of an $\mathcal{A}$-set is definitely still an $\mathcal{A}$-set (if a set contains no three-term arithmetic progression, a subset of it cannot contain any either). We therefore have $S_1 \cap \{1, 2, \ldots, x\}$ being an $\mathcal{A}$-set contained in $\{1, 2, \ldots, x\}$. Therefore, from the definition of $A(x)$ we have
(6)Similarly, we have $S_1 \cap \{x+1, x+2, \dots, x+y\}$ being an $\mathcal{A}$-set contained in $\{x+1, x+2, \ldots, x+y\}$. From the alternative definition of $A(y)$ given in 5 we obtain
(7)Finally we can combine 6 and 7 to get
(8)as required.
A(x*y) is less than x*A(y)
For any two positive integers $x$ and $y$ we have $A(x \cdot y) \leq x \cdot A(y)$.
From the above lemma we know that
(9)and repeating $x$ times we see that
(10)Some upper bounds on A(x)
For any two positive integers $x$ and $y$ we have $A(x) \leq A\left( \left( \left[ \frac{x}{y} \right] \right) + 1 \right) \leq \frac{x+y}{y} \cdot A(y)$.
The first inequality follows as for positive $y$ we have
(11)and the function $A$ is non-decreasing.
The second inequality is a direct consequence from the above corollary upon noting that
(12)a(x*y) is less than a(y)
For any two positive integers $x$ and $y$ we have $a(x \cdot y) \leq a(y)$.
We now utilize the definition of the function $a(x)$ and the above corollary:
(13)An upper bound on a(x)
For any two positive integers $x$ and $y$ we have $a(x) \leq \left( 1 + \frac{y}{x} \right) \cdot a(y)$.
We now apply the upper bound on $A(x)$ to get
(14)A trivial bound on a(x)
For any positive integer $x$ we have $\frac{1}{x} \leq a(x) \leq 1$.
From the definition of $A(x)$ we see that $A(x)$ is the size of a non-empty subset of $\{1, 2, \ldots, x\}$ and so
(15)But we have $A(x) = x \cdot a(x)$ and so
(16)which, on division by the positive $x$, gives the desired result:
(17)