Page 51 -
P. 51

Substituting (4) in (3) gives

                                                    The Basic Approximation Lemma      45
                                            m          m
                          q m (ω)   2α f         − C P     +(f (n−m)−C P (n−m)). (5)
                                            α          α
                           Now we find it useful to replace the function fàx) − C P x by its
                        “monotonemajorant”Fàx)   max t≤x (f (t)−C P t)andnotethatthis
                        Fàx) is nondecreasing and satisfies Fàx)   o(x) since fàx) − C P x
                        satisfies the same. So (5) can be replaced by

                                            m                           n
                           q m (ω)   2αF         + Fàn − m) ≤ 2αF           + F(n)    (6)
                                            α                           α
                        (a bound independent of m).
                           So choose n 0 so that x ≥ n 0 implies Fàx) ≤ 1x, and then choose
                                                           1
                        n 1 so that x ≥ n 1 implies Fàx) ≤   x. From now on we will pick
                                                           n 0
                                                       n
                        n ≥ n 1 and also will fix N   [   ].
                                                       n 0
                                              1
                           Dirichlet’s theorem on approximation by rationals now tells us
                        that the totality of arcs surrounding these ω with length 2  2π
                                                                                   αàN +1)
                        covers the whole circle. Thus using (2) for q(z), ζ   ω and 7
                        2   2π   ≤ 2  2πè 0  gives
                          αàN +1)     nα

                                                   n                     n 0
                                   q(z)   [2αF         + F(n)] 1 + 2π         .       à 7)
                                                   α                      α
                           We separate two cases:
                                                                                    n
                        Case I: α ≤ n 0 . Here we use Fà  n  ) ≤ F(n) and obtain [2αF( ) +
                                                        α                           α
                                   2πè 0                2πè 0               2πè 0
                        F(n)](1+       ) ≤ (2α+1)(1+        )F(n) ≤ 3αà 1+      )F(n)
                                    α                    α                    α
                                                                             1
                        (3α + 6πè 0 )F(n) ≤ (6π + 3)n 0 F(n) ≤ (6π + 3)n 0     n ≤ 221n.
                                                                             n 0
                                                     n               2πè 0          n
                        Case II: α> n 0 . Here [2αF( ) + F(n)](1 +       ) ≤ [2αF( ) +
                                                     α                α             α
                                                      n      n             n       n
                        F(n)](1 + 2πð . But still α ≤   ,or    ≥ n 0 .So Fà  ) ≤ 1 , and
                                                      n 0    α             α       α
                        the above is ≤ (21n + 1n)(1 + 2πð   (3 + 6πð1n < 221n.
                           In either case Dirichlet’s theorem yields our lemma.
                           So let P be any affine property, and denote by A   A(n;Pð the
                        number of arithmetic progressions from S(n;Pð (where order counts
                                                                                    N
                        1 Dirichlet’s theorem can be proved by considering the powers 1, z, z , ···, z for z
                                                                              2
                        any point on the unit circle. Since these are N + 1 points on the circle, two of them
                            j
                         i
                        z , j must be within arc length  2π  of one another. This means | arg z i−j  |≤  2π
                                                 N+1                                 N+1
                        and calling |i − j|  α gives the result.
   46   47   48   49   50   51   52   53   54   55   56