1 Introduction
In his paper [Reference Laurent1] Michel Laurent has proved that for any
$(\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2})\in \mathbb{R}^{2}$
with
$1,\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2}$
linearly independent over
$\mathbb{Q}$
, the four classical approximation constants
$\unicode[STIX]{x1D714},\unicode[STIX]{x1D714}^{\ast },\hat{\unicode[STIX]{x1D714}},\hat{\unicode[STIX]{x1D714}}^{\ast }$
satisfy the relations
Conversely, he proved also that for each quadruple of numbers
$\unicode[STIX]{x1D714},\unicode[STIX]{x1D714}^{\ast },\hat{\unicode[STIX]{x1D714}},\hat{\unicode[STIX]{x1D714}}^{\ast }$
satisfying (1.1), (1.2) and (1.3), there exists some
$(\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2})\in \mathbb{R}^{2}$
with
$1,\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2}$
linearly independent over
$\mathbb{Q}$
whose approximation constants are those numbers.
In [Reference Schmidt and Summerer4] an alternative approach to this and several other problems in simultaneous approximation was introduced that is based on the study of the successive minima of a suitable family of convex bodies with respect to a lattice defined in terms of the real numbers to be approximated.
In the context of approximation to two real numbers
$\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2}$
, the authors considered the lattice
$\unicode[STIX]{x1D6EC}(\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2})$
of points
$(y,\unicode[STIX]{x1D709}_{1}y-x_{1},\unicode[STIX]{x1D709}_{2}y-x_{2})$
where
$(y,x_{1},x_{2})$
runs through
$\mathbb{Z}^{3}$
and the box
$K(q)$
of points
$(z_{0},z_{1},z_{2})$
with
If
$\unicode[STIX]{x1D706}_{1}(q),\unicode[STIX]{x1D706}_{2}(q),\unicode[STIX]{x1D706}_{3}(q)$
denote the successive minima of
$\unicode[STIX]{x1D6EC}$
with respect to
$K(q)$
, we set
$L_{i}(q):=\log \unicode[STIX]{x1D706}_{i}(q)$
for
$i=1,2,3$
. The functions
$L_{i}$
are continuous, piecewise linear functions with slopes
$1$
and
$-2$
and
by Minkowski’s second theorem on convex bodies. For
$i=1,2,3$
we set
$\unicode[STIX]{x1D711}_{i}(q)=L_{i}(q)/q$
and
For
$i=1,2,3$
, the
$\text{}\underline{\unicode[STIX]{x1D711}}_{i}$
and
$\overline{\unicode[STIX]{x1D711}}_{i}$
will be called geometric approximation constants.
In [Reference Schmidt and Summerer4] it was shown that there is a direct correspondence between
$(\unicode[STIX]{x1D714},\hat{\unicode[STIX]{x1D714}},\unicode[STIX]{x1D714}^{\ast },\hat{\unicode[STIX]{x1D714}}^{\ast })$
and
$(\text{}\underline{\unicode[STIX]{x1D711}}_{1},\overline{\unicode[STIX]{x1D711}}_{1},\text{}\underline{\unicode[STIX]{x1D711}}_{3},\overline{\unicode[STIX]{x1D711}}_{3})$
through the relations
and
As a consequence, (1.1), (1.2) and (1.3) may be reformulated as
The reader will notice that the above inequalities do not involve
$\text{}\underline{\unicode[STIX]{x1D711}}_{2}$
,
$\overline{\unicode[STIX]{x1D711}}_{2}$
and the aim of our paper will be to establish bounds for
$\text{}\underline{\unicode[STIX]{x1D711}}_{2}$
and
$\overline{\unicode[STIX]{x1D711}}_{2}$
. Concretely we will prove the following.
Theorem 1.1. For any
$(\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2})\in \mathbb{R}^{2}$
with
$1,\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2}$
linearly independent over
$\mathbb{Q}$
, the approximation constants
$\text{}\underline{\unicode[STIX]{x1D711}}_{1},\overline{\unicode[STIX]{x1D711}}_{1},\text{}\underline{\unicode[STIX]{x1D711}}_{2},\overline{\unicode[STIX]{x1D711}}_{2},\text{}\underline{\unicode[STIX]{x1D711}}_{3},\overline{\unicode[STIX]{x1D711}}_{3}$
satisfy the relations (1.5), (1.6), (1.7) and, if
$\text{}\underline{\unicode[STIX]{x1D711}}_{3}<1$
, also
as well as the dual inequality
Moreover, these bounds are best possible in the sense that for given numbers
$\text{}\underline{\unicode[STIX]{x1D711}}_{1},\overline{\unicode[STIX]{x1D711}}_{1},\text{}\underline{\unicode[STIX]{x1D711}}_{3},\overline{\unicode[STIX]{x1D711}}_{3}$
with
$0\leqslant \text{}\underline{\unicode[STIX]{x1D711}}_{3}<1$
and (1.6) and (1.7) as well as
there are
$(\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2})\in \mathbb{R}^{2}$
with
$1,\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2}$
linearly independent over
$\mathbb{Q}$
whose geometric approximation constants are the given numbers.
Remark.
We had already pointed out at the end of [Reference Schmidt and Summerer4] (in terms of
$\unicode[STIX]{x1D713}_{i}=\unicode[STIX]{x1D711}_{i}/2$
) that, when
$\text{}\underline{\unicode[STIX]{x1D711}}_{3}=1$
, then
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}=-2$
,
$\overline{\unicode[STIX]{x1D711}}_{1}=\text{}\underline{\unicode[STIX]{x1D711}}_{2}=-1/3$
and
$\overline{\unicode[STIX]{x1D711}}_{2}=\text{}\underline{\unicode[STIX]{x1D711}}_{3}=\overline{\unicode[STIX]{x1D711}}_{3}=1$
.
Our theorem, together with the remark, contains Laurent’s [Reference Laurent1] results on
$\text{}\underline{\unicode[STIX]{x1D711}}_{1},\overline{\unicode[STIX]{x1D711}}_{1},\text{}\underline{\unicode[STIX]{x1D711}}_{3},\overline{\unicode[STIX]{x1D711}}_{3}$
. However it does not give a full description of the spectrum of
$\text{}\underline{\unicode[STIX]{x1D711}}_{1},\overline{\unicode[STIX]{x1D711}}_{1},\text{}\underline{\unicode[STIX]{x1D711}}_{2},\overline{\unicode[STIX]{x1D711}}_{2},\text{}\underline{\unicode[STIX]{x1D711}}_{3},\overline{\unicode[STIX]{x1D711}}_{3}$
, i.e. the set of all such six-tuples arising from Diophantine approximation exponents. Recently Roy [Reference Roy3], using a different approach, could give a detailed description of this spectrum as a semi-algebraic set.
2 The setting
We will start by recalling some needed notions. In a more general situation such an exposition has already been given in [Reference Schmidt and Summerer5]. We will thus content ourselves to present it here for the two-dimensional approximation problem. Let
$\unicode[STIX]{x1D6E5}\subseteq \mathbb{R}^{3}$
consist of the points
$(\unicode[STIX]{x1D6FC}_{1},\unicode[STIX]{x1D6FC}_{2},\unicode[STIX]{x1D6FC}_{3})$
with
$\unicode[STIX]{x1D6FC}_{1}\leqslant \unicode[STIX]{x1D6FC}_{2}\leqslant \unicode[STIX]{x1D6FC}_{3}$
. Observe that
${\mathcal{L}}_{\unicode[STIX]{x1D709}}:=(L_{1},L_{2},L_{3})$
with
$L_{1},L_{2},L_{3}$
as in §1 is a map
$(0,\infty )\rightarrow \unicode[STIX]{x1D6E5}$
.
A 3-system will be a map
${\mathcal{P}}:(0,\infty )\rightarrow \unicode[STIX]{x1D6E5}$
, where
${\mathcal{P}}=(P_{1},P_{2},P_{3})$
with each
$P_{i}$
continuous and piecewise linear with slopes among
$1,-2$
with
$P_{1}(q)+P_{2}(q)+P_{3}(q)=0$
, and with a further condition formulated below.
The combined graph of
${\mathcal{P}}$
is the union of the graphs of
$P_{1},P_{2},P_{3}$
. When for some point
$(q,y)$
in the combined graph of
${\mathcal{P}}$
the set of slopes of the graph emanating to the left of
$(q,y)$
is not the same as the set of slopes to the right, then
$q$
is called a switch number. There will be some
$P_{u}$
which changes slope at
$q$
, and since
$P_{1}+P_{2}+P_{3}=0$
there will be some
$P_{v}$
,
$v\neq u$
, which changes slope the “other way”. Say
$P_{u}$
changes from rising to falling,
$P_{v}$
from falling to rising. For a
$3$
-system the extra condition alluded to above is in this case always
$u>v$
. For given
$u>v$
, the number
$q$
is named a switch number of type
$\binom{u}{v}$
. As a consequence, when
$P_{1}$
has a local maximum at
$q$
, then also the graph of
$P_{2}$
passes through
$(q,P_{1}(q))$
and
$P_{1}(q)=P_{2}(q)$
; when
$P_{3}$
has a local minimum at
$q$
, then
$P_{2}(q)=P_{3}(q)$
.
Call
${\mathcal{P}}$
proper if
$q-P_{3}(q)$
tends to infinity with
$q$
. Call
${\mathcal{P}}$
agreeable if there is no
$q$
with
$P_{1}(q)=P_{2}(q)=P_{3}(q)$
(hence
$=0$
), and there is no interval of positive length where
$P_{1}(q)=P_{2}(q)$
or
$P_{2}(q)=P_{3}(q)$
.
Theorem 2.1.
-
(a) For each
$\boldsymbol{\unicode[STIX]{x1D709}}=(\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2})\in \mathbb{R}^{2}$
with
$1,\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2}$
linearly independent over
$\mathbb{Q}$
, there is an agreeable proper
$3$
-system
${\mathcal{P}}:(0,\infty )\rightarrow \unicode[STIX]{x1D6E5}$
with
${\mathcal{P}}-{\mathcal{L}}_{\boldsymbol{\unicode[STIX]{x1D709}}}$
bounded on
$(0,\infty )$
. -
(b) Given a proper
$3$
-system
${\mathcal{P}}:(0,\infty )\rightarrow \unicode[STIX]{x1D6E5}$
, there exists some
$\boldsymbol{\unicode[STIX]{x1D709}}=(\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2})$
with
$1,\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2}$
linearly independent over
$\mathbb{Q}$
having
${\mathcal{P}}-{\mathcal{L}}_{\boldsymbol{\unicode[STIX]{x1D709}}}$
bounded on
$(0,\infty )$
.
Proof. As shown in Theorem 3.1 of our paper [Reference Schmidt and Summerer6], assertion (b) is an easy consequence of the fundamental work [Reference Roy2] of Roy.
As for (a), it was shown in [Reference Schmidt and Summerer6] that for each
$\boldsymbol{\unicode[STIX]{x1D709}}\in \mathbb{R}^{2}$
there is a 3-system
${\mathcal{P}}:(\unicode[STIX]{x1D702}_{0},\infty )\rightarrow \unicode[STIX]{x1D6E5}$
for some
$\unicode[STIX]{x1D702}_{0}\geqslant 0$
such that
${\mathcal{P}}-{\mathcal{L}}_{\boldsymbol{\unicode[STIX]{x1D709}}}$
is bounded on
$(\unicode[STIX]{x1D702}_{0},\infty )$
. But [Reference Roy2, Theorem 1.3] asserts the existence of a rigid system
${\mathcal{P}}$
of positive mesh with
${\mathcal{P}}-{\mathcal{L}}_{\boldsymbol{\unicode[STIX]{x1D709}}}$
bounded, and such a system is certainly agreeable. By the argument in [Reference Schmidt and Summerer6], such a system leads to an agreeable
$3$
-system
${\mathcal{P}}$
in our context with
${\mathcal{P}}-{\mathcal{L}}_{\boldsymbol{\unicode[STIX]{x1D709}}}$
bounded. Note that an agreeable
$3$
-system
${\mathcal{P}}:(\unicode[STIX]{x1D702}_{0},\infty )\rightarrow \unicode[STIX]{x1D6E5}$
, where
$\unicode[STIX]{x1D702}_{0}>0$
can be extended to an agreeable
$3$
-system
${\mathcal{P}}:(0,\infty )\rightarrow \unicode[STIX]{x1D6E5}$
, and then
${\mathcal{P}}-{\mathcal{L}}_{\boldsymbol{\unicode[STIX]{x1D709}}}$
will be bounded on
$(0,\unicode[STIX]{x1D702}_{0}]$
, hence on
$(0,\infty )$
.
It had been shown in [Reference Schmidt and Summerer4] that when
$1,\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2}$
are linearly independent over
$\mathbb{Q}$
, then there are arbitrarily large numbers
$r$
with
$L_{2}(r)=L_{3}(r)$
as well as arbitrarily large
$r^{\ast }$
with
$L_{1}(r^{\ast })=L_{2}(r^{\ast })$
. Since
${\mathcal{P}}-{\mathcal{L}}_{\boldsymbol{\unicode[STIX]{x1D709}}}$
is bounded, these relations give
$P_{2}(r)=P_{3}(r)+O(1)$
,
$P_{1}(r^{\ast })=P_{2}(r^{\ast })+O(1)$
with the implied constants independent of
$r,r^{\ast }$
. The second relation gives
$P_{2}(r^{\ast })\ll 1$
, hence
$P_{2}(q)<q-r^{\ast }+O(1)$
for
$q>r^{\ast }$
. When
$r>r^{\ast }$
, the first relation gives
$P_{3}(r)<P_{2}(r)+O(1)<r-r^{\ast }+O(1)$
, and when
$q>r$
,
$P_{3}(q)<q-r^{\ast }+O(1)$
. Since
$r^{\ast }$
is arbitrarily large,
$q-P_{3}(q)\rightarrow \infty$
.◻
The argument in the above proof shows that if
$P_{2}(r)=P_{3}(r)$
and
$P_{1}(r^{\ast })=P_{2}(r^{\ast })$
for certain arbitrarily large
$r,r^{\ast }$
, then a
$3$
-system
${\mathcal{P}}$
is proper. Conversely, when
${\mathcal{P}}$
is proper, then there are intervals
$[s,r]$
with large
$s$
where
$P_{3}$
declines, and has a local minimum at
$r$
, so that
$P_{2}(r)=P_{3}(r)$
. Now
$P_{1}(r)=P_{2}(r)=P_{3}(r)$
, or
$P_{2}$
declines to the right of
$r$
, hence
$P_{1}$
inclines, and reaches a local maximum at some
$r^{\ast }$
, so that
$P_{1}(r^{\ast })=P_{2}(r^{\ast })$
.
We set
$\unicode[STIX]{x1D711}_{i}(q):=P_{i}(q)/q$
for
$i=1,2,3$
and note that the quantities
$\text{}\underline{\unicode[STIX]{x1D711}}_{i}$
and
$\overline{\unicode[STIX]{x1D711}}_{i}$
defined by (1.4) in terms of
${\mathcal{P}}$
are the same as the quantities defined in terms of
${\mathcal{L}}_{\unicode[STIX]{x1D709}}$
when
${\mathcal{P}}-{\mathcal{L}}_{\boldsymbol{\unicode[STIX]{x1D709}}}$
is bounded. By Theorem 2.1, to prove Theorem 1.1 up to the best possible assertion, we may deal with an agreeable, proper
$3$
-system
${\mathcal{P}}$
. For such a system, the numbers
$r$
with
$P_{2}(r)=P_{3}(r)$
form a sequence tending to infinity, and so do the numbers
$r^{\ast }$
with
$P_{1}(r^{\ast })=P_{2}(r^{\ast })$
, and these sequences have no common element. In the mixed combined sequence of these
$r^{\ast }$
and
$r$
arranged in increasing order, blocks of
$r^{\ast }$
will alternate with blocks of
$r$
(possibly reduced to one single
$r^{\ast }$
respectively
$r$
). Starting with a block of
$r^{\ast }$
, in the
$j$
th block of
$r^{\ast }$
denote the first and last element by
$b_{j}^{\ast }$
and
$a_{j}^{\ast }$
. In the directly following block of
$r$
denote the first and last element by
$a_{j}$
and
$b_{j}$
. Note that if
$b_{j}^{\ast }=a_{j}^{\ast }$
or
$a_{j}=b_{j}$
, then the corresponding block consists of only one element. Set
${\mathcal{I}}_{j}:=[b_{j-1},b_{j+1}^{\ast }]$
, and note that
${\mathcal{I}}_{j},{\mathcal{I}}_{j+1}$
overlap on
$[b_{j},b_{j+1}^{\ast }]$
.
In
$(a_{j}^{\ast },b_{j+1}^{\ast })$
we have
$P_{1}(q)\neq P_{2}(q)$
, so that
$P_{1}$
has no local maximum.
$P_{1}$
decreases up to some
$p_{j}^{\ast }$
, then increases. Similarly,
$P_{3}$
increases up to some
$p_{j}$
, then decreases up to
$a_{j}$
. In the case when
$P_{2}$
increases throughout
$[a_{j}^{\ast },a_{j}]$
, then
$p_{j}=p_{j}^{\ast }$
and this number is a switch number of type
$\binom{3}{1}$
. Otherwise,
$P_{2}$
increases in
$[a_{j}^{\ast },p_{j}^{\ast }]$
, decreases in
$[p_{j}^{\ast },p_{j}]$
where
$p_{j}^{\ast }<p_{j}$
, and increases in
$[p_{j},a_{j}]$
. In this case
$p_{j}^{\ast },p_{j}$
are respectively switch numbers of type
$\binom{2}{1}$
and
$\binom{3}{2}$
.
When
$b_{j}^{\ast }<a_{j}^{\ast }$
, then
${\textstyle \frac{1}{2}}(P_{1}+P_{2})$
will decrease in
$(b_{j}^{\ast },a_{j}^{\ast })$
with slope
$-1/2$
, and
$P_{1},P_{2}$
will zigzag around the dotted line of Figure 1. Any local minimum of
$P_{1}$
in the interval will be assumed at a switch number of type
$\binom{2}{1}$
. Similarly, when
$a_{j}<b_{j}$
, then
$P_{2},P_{3}$
will zigzag around a line of slope
$-1/2$
and a local maximum of
$P_{3}$
in
$[a_{j},b_{j}]$
will be asumed at a switch number of type
$\binom{3}{2}$
.
For the convenience of the reader we illustrate the described situation within some interval
${\mathcal{I}}_{j}$
in Figure 1.
Figure 1

Now let
$c_{j}^{\ast }\in {\mathcal{I}}_{j}$
be a number where
$P_{1}$
assumes a local minimum. Then
$b_{j}^{\ast }<c_{j}^{\ast }\leqslant p_{j}^{\ast }$
. We will always have such a local minimum within
$[a_{j}^{\ast },a_{j}]$
, namely at
$c_{j}^{\ast }=p_{j}^{\ast }$
and when
$b_{j}^{\ast }<a_{j}^{\ast }$
there will be at least one more such
$c_{j}^{\ast }$
within
$(b_{j}^{\ast },a_{j}^{\ast })$
. For any
$c_{j}^{\ast }\in (b_{j}^{\ast },a_{j}^{\ast })$
let
$d_{j}^{\ast }<c_{j}^{\ast }$
be the largest number where
$P_{1}(d_{j}^{\ast })=P_{2}(d_{j}^{\ast })$
. For
$c_{j}^{\ast }=p_{j}^{\ast }$
let
$d_{j}^{\ast }=a_{j}^{\ast }$
.
Let
$c_{j}\in {\mathcal{I}}_{j}$
be a number where
$P_{3}$
assumes a local maximum. Then
$p_{j}\leqslant c_{j}<b_{j}$
. Thus
$c_{j}$
may be
$p_{j}$
, and when
$a_{j}<b_{j}$
there will be at least one such
$c_{j}$
within
$(a_{j},b_{j})$
. For any
$c_{j}\in (a_{j},b_{j})$
let
$d_{j}>c_{j}$
be the least number where
$P_{3}(d_{j})=P_{2}(d_{j})$
. For
$c_{j}=p_{j}$
let
$d_{j}=a_{j}$
.
Let
$\overline{\unicode[STIX]{x1D711}}_{ij}$
be the maximum of
$\unicode[STIX]{x1D711}_{i}$
in
${\mathcal{I}}_{j}$
respectively
$\text{}\underline{\unicode[STIX]{x1D711}}_{ij}$
the minimum. Note that when
$\overline{\unicode[STIX]{x1D719}}_{i}>\overline{\unicode[STIX]{x1D711}}_{i}$
,
$\text{}\underline{\unicode[STIX]{x1D719}}_{i}<\text{}\underline{\unicode[STIX]{x1D711}}_{i}$
, then
$\overline{\unicode[STIX]{x1D711}}_{ij}<\overline{\unicode[STIX]{x1D719}}_{i}$
,
$\text{}\underline{\unicode[STIX]{x1D711}}_{ij}>\text{}\underline{\unicode[STIX]{x1D719}}_{i}$
when
$j$
is large.
In what follows, we will assume
3 Basic inequalities
In this section we derive a few relations that result from the basic properties of proper 3-systems, in particular from
$P_{1}+P_{2}+P_{3}=0$
and the knowledge of the possible slopes for these functions.
For any
$c_{j}^{\ast }$
as defined in Section 2 we have
$$\begin{eqnarray}\displaystyle P_{3}(c_{j}^{\ast }) & = & \displaystyle P_{3}(b_{j}^{\ast })+c_{j}^{\ast }-b_{j}^{\ast },\nonumber\\ \displaystyle P_{1}(c_{j}^{\ast }) & = & \displaystyle P_{1}(b_{j}^{\ast })-{\textstyle \frac{1}{2}}(d_{j}^{\ast }-b_{j}^{\ast })-2(c_{j}^{\ast }-d_{j}^{\ast })\nonumber\\ \displaystyle & = & \displaystyle P_{1}(b_{j}^{\ast })-2(c_{j}^{\ast }-b_{j}^{\ast })+{\textstyle \frac{3}{2}}(d_{j}^{\ast }-b_{j}^{\ast }),\nonumber\end{eqnarray}$$
and therefore
hence upon division by
$c_{j}^{\ast }$
Also
and division by
$c_{j}^{\ast }$
yields
Combining (3.1) and (3.2) we obtain
$$\begin{eqnarray}\displaystyle 2\unicode[STIX]{x1D711}_{3}(c_{j}^{\ast })+\unicode[STIX]{x1D711}_{1}(c_{j}^{\ast }) & = & \displaystyle -\unicode[STIX]{x1D711}_{1}(b_{j}^{\ast })\biggl(\!3+\unicode[STIX]{x1D711}_{3}(c_{j}^{\ast })+2\unicode[STIX]{x1D711}_{1}(c_{j}^{\ast })-3\frac{d_{j}^{\ast }-b_{j}^{\ast }}{c_{j}^{\ast }}\biggr)\!+\frac{3}{2}\frac{d_{j}^{\ast }-b_{j}^{\ast }}{c_{j}^{\ast }}\nonumber\\ \displaystyle & = & \displaystyle -\unicode[STIX]{x1D711}_{1}(b_{j}^{\ast })(3+\unicode[STIX]{x1D711}_{3}(c_{j}^{\ast })+2\unicode[STIX]{x1D711}_{1}(c_{j}^{\ast }))\,+\,3\frac{d_{j}^{\ast }-b_{j}^{\ast }}{c_{j}^{\ast }}\biggl(\frac{1}{2}+\unicode[STIX]{x1D711}_{1}(b_{j}^{\ast })\!\biggr)\nonumber\\ \displaystyle & {\geqslant} & \displaystyle -\unicode[STIX]{x1D711}_{1}(b_{j}^{\ast })(3+\unicode[STIX]{x1D711}_{3}(c_{j}^{\ast })+2\unicode[STIX]{x1D711}_{1}(c_{j}^{\ast })),\nonumber\end{eqnarray}$$
since
$\unicode[STIX]{x1D711}_{1}(b_{j}^{\ast })\geqslant -1/2$
. Here
in view of
$\unicode[STIX]{x1D711}_{1}(c_{j}^{\ast })\geqslant -2$
and
$\unicode[STIX]{x1D711}_{2}(c_{j}^{\ast })\leqslant 1$
. Moreover
$\unicode[STIX]{x1D711}_{1}(b_{j}^{\ast })\leqslant \overline{\unicode[STIX]{x1D719}}_{1}$
, so that
which may also be written as
Thus
$(\unicode[STIX]{x1D711}_{1}(c_{j}^{\ast }),\unicode[STIX]{x1D711}_{3}(c_{j}^{\ast }))$
lies on or above the line
$L^{\ast }$
with equation
Dually to (3.4),
so that
$(\unicode[STIX]{x1D711}_{1}(c_{j}),\unicode[STIX]{x1D711}_{3}(c_{j}))$
lies on or below the line
$L$
with equation
Now choose
$c_{j}^{\ast }$
to be a number where
$\unicode[STIX]{x1D711}_{1}$
assumes its minimum in
${\mathcal{I}}_{j}$
. Then
$\unicode[STIX]{x1D711}_{1}(c_{j}^{\ast })=\text{}\underline{\unicode[STIX]{x1D711}}_{1j}$
and
$\unicode[STIX]{x1D711}_{3}(c_{j}^{\ast })<\overline{\unicode[STIX]{x1D719}}_{3}$
, so that (3.4) yields
Since
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}=\liminf _{j}\text{}\underline{\unicode[STIX]{x1D711}}_{1j}$
, we obtain
or, since
$\overline{\unicode[STIX]{x1D719}}_{i}$
may be chosen arbitrarily close to
$\overline{\unicode[STIX]{x1D711}}_{i}$
for
$i=1,3$
,
DuallyFootnote 1 to (3.6),
4 Bounds for
$\overline{\unicode[STIX]{x1D711}}_{2j}$
and
$\text{}\underline{\unicode[STIX]{x1D711}}_{2j}$
$\unicode[STIX]{x1D711}_{2}$
does have local maxima where
$P_{2}$
does. In
${\mathcal{I}}_{j}$
these local maxima are at
$a_{j}$
, at
$p_{j}^{\ast }$
if
$p_{j}^{\ast }<p_{j}$
and at numbers
$c_{j}^{\ast }$
in
$(b_{j}^{\ast },a_{j}^{\ast })$
.
The point
$(a_{j},P_{2}(a_{j}))$
is the intersection of the line of slope
$1$
through
$(p_{j},P_{2}(p_{j}))$
and the line of slope
$-2$
through
$(p_{j},P_{3}(p_{j}))$
. Therefore
and also
Twice the first of these equations added to the second gives
Moreover
by Figure 1, so that
so that in combination with (4.1)
with
Since
$f(x,y)$
is decreasing in
$x$
as well as in
$y$
when
$3+x+2y>0$
, and since
$\unicode[STIX]{x1D711}_{1}(p_{j})\geqslant \unicode[STIX]{x1D711}_{1}(p_{j}^{\ast })$
,
$\unicode[STIX]{x1D711}_{3}(p_{j})\geqslant \unicode[STIX]{x1D711}_{3}(p_{j}^{\ast })$
, we obtain
Given
$c_{j}^{\ast }\in (b_{j}^{\ast },a_{j}^{\ast })$
, let
$(\tilde{a}_{j},\tilde{P}_{j})$
be the point of intersection of the line of slope
$1$
through
$(c_{j}^{\ast },P_{2}(c_{j}^{\ast }))$
and the line of slope
$-2$
through
$(c_{j}^{\ast },P_{3}(c_{j}^{\ast }))$
. By the argument above, with
$c_{j}^{\ast }$
in place of
$p_{j}$
we obtain, in analogy to (4.3)
where
$\tilde{\unicode[STIX]{x1D711}}_{2}(\tilde{a}_{j})=\tilde{P}_{j}/\tilde{a}_{j}$
. Since clearly
$\unicode[STIX]{x1D711}_{2}(c_{j}^{\ast })\leqslant \tilde{\unicode[STIX]{x1D711}}_{2}(\tilde{a}_{j})$
, we may conclude that
$\unicode[STIX]{x1D711}_{2}(c_{j}^{\ast })\leqslant f(\unicode[STIX]{x1D711}_{1}(c_{j}^{\ast }),\unicode[STIX]{x1D711}_{3}(c_{j}^{\ast }))$
, hence (as
$p_{j}^{\ast }$
is also some
$c_{j}^{\ast }$
)
and therefore
with the maximum ranging over points
$(x,y)$
on or above
$L^{\ast }$
and with
$x\geqslant \text{}\underline{\unicode[STIX]{x1D719}}_{1}$
.
with
$f^{\ast }(x,y)=f(y,x)=-(2y+x)/(3+y+2x)$
, and
with the minimum ranging over points on or below
$L$
, and with
$y\leqslant \overline{\unicode[STIX]{x1D719}}_{3}$
.
5 Proof of (1.8) and (1.9)
Since
$f(x,y)$
increases as
$y$
decreases,
$f(x,y)$
with
$(x,y)$
as in (4.6) is
${\leqslant}f(x,y(x))$
with
$x\geqslant \text{}\underline{\unicode[STIX]{x1D719}}_{1}$
, where
$L^{\ast }$
is given by
Hence we obtain
$$\begin{eqnarray}\displaystyle f(x,y(x)) & = & \displaystyle -\frac{2x+y(x)}{3+x+2y(x)}\nonumber\\ \displaystyle & = & \displaystyle -\frac{2x-((1+2\overline{\unicode[STIX]{x1D719}}_{1})x+3\overline{\unicode[STIX]{x1D719}}_{1})/(2+\overline{\unicode[STIX]{x1D719}}_{1})}{3+x-2((1+2\overline{\unicode[STIX]{x1D719}}_{1})x+3\overline{\unicode[STIX]{x1D719}}_{1})/(2+\overline{\unicode[STIX]{x1D719}}_{1})}\nonumber\\ \displaystyle & = & \displaystyle -\frac{2x(2+\overline{\unicode[STIX]{x1D719}}_{1})-x(1+2\overline{\unicode[STIX]{x1D719}}_{1})-3\overline{\unicode[STIX]{x1D719}}_{1}}{6+3\overline{\unicode[STIX]{x1D719}}_{1}+x(2+\overline{\unicode[STIX]{x1D719}}_{1})-2x(1+2\overline{\unicode[STIX]{x1D719}}_{1})-6\overline{\unicode[STIX]{x1D719}}_{1}}\nonumber\end{eqnarray}$$
and simplification leads to
Since
$x\geqslant \text{}\underline{\unicode[STIX]{x1D719}}_{1}$
, this is at most
$(-\text{}\underline{\unicode[STIX]{x1D719}}_{1}+\overline{\unicode[STIX]{x1D719}}_{1})/(2-\overline{\unicode[STIX]{x1D719}}_{1}-\overline{\unicode[STIX]{x1D719}}_{1}\text{}\underline{\unicode[STIX]{x1D719}}_{1})$
, which is therefore an upper bound for
$\unicode[STIX]{x1D711}_{2j}$
. This is also an upper bound for
$\text{}\underline{\unicode[STIX]{x1D711}}_{2}$
, and since
$\overline{\unicode[STIX]{x1D719}}_{1},\text{}\underline{\unicode[STIX]{x1D719}}_{1}$
can be chosen arbitrarily close to
$\overline{\unicode[STIX]{x1D711}}_{1},\text{}\underline{\unicode[STIX]{x1D711}}_{1}$
, inequality (1.8) follows.
Here
$f^{\ast }(x,y)$
decreases as
$x$
increases, so that
$f^{\ast }(x,y)$
with
$(x,y)$
as in (4.6*) is
${\geqslant}f^{\ast }(x(y),y)$
where
$L$
consists of points
$(x(y),y)$
. Dually to (5.1)
Taking into account
$y\leqslant \overline{\unicode[STIX]{x1D719}}_{3}$
, inequality (1.9) easily follows.
6 Standard pieces
Before turning to the “best possible” assertion of Theorem 1.1 we need to define standard functions on some intervals.
Suppose
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}\overline{\unicode[STIX]{x1D711}}_{1},\text{}\underline{\unicode[STIX]{x1D711}}_{2},\overline{\unicode[STIX]{x1D711}}_{2},\text{}\underline{\unicode[STIX]{x1D711}}_{3},\overline{\unicode[STIX]{x1D711}}_{3}$
are numbers with (1.5),
$\ldots$
, (1.9) and with
$\text{}\underline{\unicode[STIX]{x1D711}}_{3}<1$
(hence
$\overline{\unicode[STIX]{x1D711}}_{1}>-1/2$
) and
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}<0<\overline{\unicode[STIX]{x1D711}}_{3}$
. Further let
$\text{}\underline{t},\overline{t}$
be numbers, not both zero, with
$\overline{t}<1$
,
so that (1.7) holds with
$\text{}\underline{t},\overline{t}$
in place of
$\text{}\underline{\unicode[STIX]{x1D711}}_{1},\overline{\unicode[STIX]{x1D711}}_{3}$
. Thus
$(\text{}\underline{t},\overline{t})$
is on or below the line
$L$
given by
$(2+\text{}\underline{\unicode[STIX]{x1D711}}_{3})x+(1+2\text{}\underline{\unicode[STIX]{x1D711}}_{3})y=-3\text{}\underline{\unicode[STIX]{x1D711}}_{3}$
, and on or above the line
$L^{\ast }$
with equation
$(1+2\overline{\unicode[STIX]{x1D711}}_{1})x+(2+\overline{\unicode[STIX]{x1D711}}_{1})y=-3\overline{\unicode[STIX]{x1D711}}_{1}$
. (Note that
$L,L^{\ast }$
are analogues to the lines of §3.) So
$(\text{}\underline{t},\overline{t})$
is also on or below the line
$y=1$
, hence is in the triangle
$T$
bounded by these three lines. A short computation shows that the vertices of
$T$
are
$(-2,1)$
,
$(\unicode[STIX]{x1D6FC},1)$
with
$\unicode[STIX]{x1D6FC}:=-(1+5\text{}\underline{\unicode[STIX]{x1D711}}_{3})/(2+\text{}\underline{\unicode[STIX]{x1D711}}_{3})>-2$
and
$(\text{}\underline{\unicode[STIX]{x1D6FD}},\bar{\unicode[STIX]{x1D6FD}})$
with
$\text{}\underline{\unicode[STIX]{x1D6FD}}:=-3\text{}\underline{\unicode[STIX]{x1D711}}_{3}/(1-\overline{\unicode[STIX]{x1D711}}_{1}\text{}\underline{\unicode[STIX]{x1D711}}_{3})$
,
$\bar{\unicode[STIX]{x1D6FD}}:=-3\overline{\unicode[STIX]{x1D711}}_{1}/(1-\overline{\unicode[STIX]{x1D711}}_{1}\text{}\underline{\unicode[STIX]{x1D711}}_{3})$
, so that
$\text{}\underline{\unicode[STIX]{x1D6FD}}\leqslant 0,\bar{\unicode[STIX]{x1D6FD}}\geqslant 0$
and
By (6.1),
hence
$2\overline{t}+\text{}\underline{t}\geqslant 0$
. Dually,
$2\text{}\underline{t}+\overline{t}\leqslant 0$
, so that
Not both relations can hold with equality, for then
$\text{}\underline{t}=\overline{t}=0$
, which was ruled out. We may infer that
$\text{}\underline{t}<0<\overline{t}$
. The case where both relations in (6.3) are strict inequalities will be called the generic case, the other case is the special case. This case can only occur when
$\overline{\unicode[STIX]{x1D711}}_{1}=0=\text{}\underline{\unicode[STIX]{x1D711}}_{3}$
.
After this tedious beginning we now come to the core of this section. For suitable numbers
with
$c^{\ast }<c$
(which have no relation with the numbers
$c_{j}^{\ast },c_{j}$
of §3) we will construct functions
$P_{1},P_{2},P_{3}$
on
${\mathcal{I}}(\text{}\underline{t},\overline{t}):=[c^{\ast },c]$
with
$P_{1}+P_{2}+P_{3}=0$
, where each
$P_{i}$
is continuous and piecewise linear with slopes among
$1,-2$
. We will have
$P_{1}$
will rise (with slope
$1$
) in
$[1,c]$
.
$P_{2}$
will rise,
$P_{3}$
will fall (with slope
$-2$
) to the right of
$q=1$
, and they will meet at
$q=a$
, so that
$P_{2}(a)=P_{3}(a)$
. Since
and also
we obtain
in fact
${>}1$
when
$2\overline{t}+\text{}\underline{t}>0$
. It follows that
Moreover
$P_{2}(b)=P_{3}(b)=\text{}\underline{\unicode[STIX]{x1D711}}_{3}b$
, and when
$a<b$
, then
$P_{2},P_{3}$
will zigzag with alternating slopes
$1,-2$
around a line of slope
$-1/2$
(indicated by a dotted line segment in Figure 2), so that
hence
so that
Finally
$P_{1}(c)=P_{2}(c)$
, and when
$b<c$
, then
$P_{1},P_{3}$
will rise,
$P_{2}$
will fall in
$[b,c]$
. Since
$P_{1}(b)=-2P_{2}(b)=-2\text{}\underline{\unicode[STIX]{x1D711}}_{3}b$
, we have
$3(c-b)=P_{3}(b)-P_{1}(b)=-3\text{}\underline{\unicode[STIX]{x1D711}}_{3}b$
, hence
$c=(1+\text{}\underline{\unicode[STIX]{x1D711}}_{3})b$
,
$c>b$
if
$\text{}\underline{\unicode[STIX]{x1D711}}_{3}>0$
. Moreover by (6.8), (1.6)
Also
$P_{2}(c)=P_{3}(b)-2(c-b)=\text{}\underline{\unicode[STIX]{x1D711}}_{3}b-2\text{}\underline{\unicode[STIX]{x1D711}}_{3}b=-\text{}\underline{\unicode[STIX]{x1D711}}_{3}b$
. So
But
$P_{1}(c)=\text{}\underline{t}+c-1$
since
$P_{1}$
rises in
$[1,c]$
, and we obtain again (6.9). By (6.7), (6.8), (6.1),
Figure 2

We give a dual construction on the left of
$q=1$
. In our duality, we have
$c^{\ast },b^{\ast },a^{\ast }$
in place of
$c,b,a$
, and
$\overline{\unicode[STIX]{x1D711}}_{1},\text{}\underline{\unicode[STIX]{x1D711}}_{3},\overline{t},\text{}\underline{t},P_{1},P_{2},P_{3}$
are replaced by
$\text{}\underline{\unicode[STIX]{x1D711}}_{3},\overline{\unicode[STIX]{x1D711}}_{1},\text{}\underline{t},\overline{t},P_{3},P_{2},P_{1}$
. Now
$P_{3}$
rises in
$[c^{\ast },1]$
,
$P_{1}(a^{\ast })=P_{2}(a^{\ast })$
and
$P_{2}$
rises,
$P_{1}$
falls in
$[a^{\ast },1]$
. Hence
Also
and
By (6.6), (6.8), (6.9), (6.6*), (6.8*), (6.9*) (and using (6.3)) we have in fact (6.4). Not both relations
$a=1$
and
$a^{\ast }=1$
can hold, so that
$a^{\ast }<a$
, hence
$c^{\ast }<c$
.
The triple of functions on the interval
${\mathcal{I}}(\text{}\underline{t},\overline{t})$
will be called a standard piece and denoted
${\mathcal{P}}(\text{}\underline{t},\overline{t})$
. When
$\overline{\unicode[STIX]{x1D711}}_{1}=0=\text{}\underline{\unicode[STIX]{x1D711}}_{3}$
, then
$c^{\ast }=b^{\ast }$
,
$b=c$
, and
$P_{1}(q)=P_{2}(q)=P_{3}(q)=0$
for
$q=c^{\ast }$
and
$q=c$
. Moreover there will be an interval
$[c^{\ast },g^{\ast }]$
with
$c^{\ast }<g^{\ast }$
where
$P_{1}(q)=P_{2}(q)$
, and an interval
$[g,c]$
with
$g<c$
where
$P_{2}(q)=P_{3}(q)$
. A degenerate version of Figure 2 will ensue.
Define
$\unicode[STIX]{x1D711}_{i}(q)=P_{i}(q)/q$
for
$i=1,2,3$
on
${\mathcal{I}}(\text{}\underline{t},\overline{t})$
, and let
$\overline{\unicode[STIX]{x1D711}}_{i}({\mathcal{P}}(\text{}\underline{t},\overline{t}))$
(respectively
$\text{}\underline{\unicode[STIX]{x1D711}}_{i}({\mathcal{P}}(\text{}\underline{t},\overline{t}))$
) be the maximum (respectively minimum) of
$\unicode[STIX]{x1D711}_{i}(q)$
in
${\mathcal{I}}(\text{}\underline{t},\overline{t})$
. Clearly
$\text{}\underline{\unicode[STIX]{x1D711}}_{3}({\mathcal{P}}(\text{}\underline{t},\overline{t}))=\text{}\underline{\unicode[STIX]{x1D711}}_{3}$
and
$\overline{\unicode[STIX]{x1D711}}_{1}({\mathcal{P}}(\text{}\underline{t},\overline{t}))=\overline{\unicode[STIX]{x1D711}}_{1}$
. But caution:
${\mathcal{P}}(\text{}\underline{t},\overline{t})$
is not completely determined if
$b^{\ast }<a^{\ast }$
or
$a<b$
. In the generic case,
$a^{\ast }<1<a$
,
$P_{1}(a^{\ast })>\text{}\underline{t}$
,
$P_{3}(a)<\overline{t}$
, and we can pick alternating functions in
$[b^{\ast },a^{\ast }]$
and
$[a,b]$
such that
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}({\mathcal{P}}(\text{}\underline{t},\overline{t}))=\text{}\underline{t}$
,
$\overline{\unicode[STIX]{x1D711}}_{3}({\mathcal{P}}(\text{}\underline{t},\overline{t}))=\overline{t}$
. In the special case we can for each
$\unicode[STIX]{x1D716}>0$
pick a standard piece
${\mathcal{P}}_{\unicode[STIX]{x1D716}}(\text{}\underline{t},\overline{t})\!)$
with
We will now suppose
$\overline{\unicode[STIX]{x1D711}}_{3}<1$
, hence
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}<0<\overline{\unicode[STIX]{x1D711}}_{3}<1$
. Three choices for
$(\text{}\underline{t},\overline{t})$
will be needed, denoted by
$(\text{}\underline{s}_{j},\overline{s}_{j})$
for
$j=0,1,2$
. First
which is allowed by (1.7). Here
in the generic case,
in the special case.
Next,
where
$y=y(x)$
is the equation of the line
$L^{\ast }$
. Thus
$(\text{}\underline{s}_{1},\overline{s}_{1})$
is the downward projection of
$(\text{}\underline{\unicode[STIX]{x1D711}}_{1},\overline{\unicode[STIX]{x1D711}}_{3})$
on
$L^{\ast }$
, hence is in
$T$
, hence satisfies the condition on pairs
$(\text{}\underline{t},\overline{t})$
. Moreover
$y(\text{}\underline{\unicode[STIX]{x1D711}}_{1})\leqslant \overline{\unicode[STIX]{x1D711}}_{3}$
, so that
respectively
in the generic respectively in the special case for sufficiently small
$\unicode[STIX]{x1D716}$
.
Moreover by (4.3) with
$p_{j}=1$
, and by (6.5),
hence by (5.1)
with
$\bar{\unicode[STIX]{x1D6FA}}$
as in (1.8).
Finally,
where
$x=x(y)$
is the equation of
$L$
. By an argument dual to the one for
$(\text{}\underline{s}_{1},\overline{s}_{1})$
,
in the general and the special case. By (4.3*) with
$p_{j}^{\ast }=1$
,
so that by (5.1*)
${\mathcal{P}}(\text{}\underline{s}_{0},\overline{s}_{0})$
is agreeable in an obvious sense unless
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}+2\overline{\unicode[STIX]{x1D711}}_{3}=0$
or
$2\text{}\underline{\unicode[STIX]{x1D711}}_{1}+\overline{\unicode[STIX]{x1D711}}_{3}=0$
, and
${\mathcal{P}}(\text{}\underline{s}_{1},\overline{s}_{1})$
,
${\mathcal{P}}(\text{}\underline{s}_{2},\overline{s}_{2})$
are agreeable when
$\overline{\unicode[STIX]{x1D711}}_{1}<0<\text{}\underline{\unicode[STIX]{x1D711}}_{3}$
but not otherwise, since
$\overline{\unicode[STIX]{x1D711}}_{1}=0=\text{}\underline{\unicode[STIX]{x1D711}}_{3}$
implies
$y(\overline{\unicode[STIX]{x1D711}}_{1})=-{\textstyle \frac{1}{2}}\overline{\unicode[STIX]{x1D711}}_{1}$
,
$x(\overline{\unicode[STIX]{x1D711}}_{3})=-12\overline{\unicode[STIX]{x1D711}}_{3}$
.
7 Best possible
Given a
$3$
-system
${\mathcal{P}}$
we had in §2 defined quantities
$\text{}\underline{\unicode[STIX]{x1D711}}_{i},\overline{\unicode[STIX]{x1D711}}_{i}$
,
$i=1,2,3$
, which will now be denoted
$\text{}\underline{\unicode[STIX]{x1D711}}_{i}({\mathcal{P}}),\overline{\unicode[STIX]{x1D711}}_{i}({\mathcal{P}})$
. To prove that the conditions (1.5),
$\ldots$
, (1.9) are best possible, we will for given numbers
$\text{}\underline{\unicode[STIX]{x1D711}}_{i},\overline{\unicode[STIX]{x1D711}}_{i}$
satisfying these five conditions, as well as
$\text{}\underline{\unicode[STIX]{x1D711}}_{3}<1$
, construct a proper
$3$
-system
${\mathcal{P}}$
with
and
with
$\text{}\underline{\unicode[STIX]{x1D6FA}},\overline{\unicode[STIX]{x1D6FA}}$
as in (1.8), (1.9). In the generic case, when (6.3) holds with strict inequalities,
${\mathcal{P}}$
will be self-similar, i.e. will be invariant under dilation by a factor
$\unicode[STIX]{x1D70C}>1$
.
When
${\mathcal{P}}(\text{}\underline{t},\overline{t})$
is as in the preceding section, let
${\mathcal{P}}^{-}(\text{}\underline{t},\overline{t})$
(respectively
${\mathcal{P}}^{+}(\text{}\underline{t},\overline{t})$
) be its part with
$q\in [c^{\ast },b^{\ast }]$
(respectively
$q\in [b,c]$
). For any
$(\text{}\underline{t},\overline{t}),(\text{}\underline{t}^{\prime },\overline{t}^{\prime })$
, the parts
${\mathcal{P}}^{-}(\text{}\underline{t},\overline{t}),{\mathcal{P}}^{+}(\text{}\underline{t},\overline{t}),{\mathcal{P}}^{-}(\text{}\underline{t}^{\prime },\overline{t}^{\prime }),{\mathcal{P}}^{+}(\text{}\underline{t}^{\prime },\overline{t}^{\prime })$
will all be similar, i.e. obtained from each other by multiplication by factors
$\unicode[STIX]{x1D70C}>0$
. In particular,
${\mathcal{P}}^{+}(\text{}\underline{t},\overline{t})$
is obtained from
${\mathcal{P}}^{-}(\text{}\underline{t},\overline{t})$
by a factor
$\unicode[STIX]{x1D70C}(\text{}\underline{t},\overline{t})>1$
. In fact,
$\unicode[STIX]{x1D70C}(\text{}\underline{t},\overline{t})=b/c^{\ast }=c/b^{\ast }$
.
Let
$(\text{}\underline{t}_{j},\overline{t}_{j})$
,
$j\in \mathbb{Z}$
, again be pairs with
$\overline{t}_{j}\leqslant 1$
and with (6.1). Then
so that
defines functions on
${\mathcal{I}}(\text{}\underline{t}_{j},\overline{t}_{j})\cup \unicode[STIX]{x1D70C}(\text{}\underline{t}_{j},\overline{t}_{j}){\mathcal{I}}(\text{}\underline{t}_{j+1},\overline{t}_{j+1})$
: we could say that the standard pieces
${\mathcal{P}}(\text{}\underline{t}_{j},\overline{t}_{j})$
,
${\mathcal{P}}(\text{}\underline{t}_{j+1},\overline{t}_{j+1})$
are patched together. Define
$\unicode[STIX]{x1D714}_{j}$
(
$j\in \mathbb{Z}$
) by
$\unicode[STIX]{x1D714}_{0}=1$
and
$\unicode[STIX]{x1D714}_{j+1}/\unicode[STIX]{x1D714}_{j}=\unicode[STIX]{x1D70C}(\text{}\underline{t}_{j},\overline{t}_{j})$
. If
$\unicode[STIX]{x1D714}_{j}\rightarrow 0$
(respectively
$\unicode[STIX]{x1D714}_{j}\rightarrow \infty$
) as
$j\rightarrow -\infty$
(respectively
$j\rightarrow \infty$
), then
defines a proper 3-system on
$(0,\infty )$
.
Again suppose
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}<0<\overline{\unicode[STIX]{x1D711}}_{3}<1$
. We now set
$(\text{}\underline{t}_{j},\overline{t}_{j})=(\text{}\underline{s}_{i},\overline{s}_{i})$
for
$i=0,1,2$
when
$j\equiv i(\text{mod}\,3)$
. In the case when
$\overline{\unicode[STIX]{x1D711}}_{1}<0<\text{}\underline{\unicode[STIX]{x1D711}}_{3}$
, it follows from (6.12), (6.15), (6.17), (6.15*), (6.17*) and the bounds (1.8), (1.9) that (7.1), (7.2) hold. Our
${\mathcal{P}}$
is agreeable and self similar with factor
$\unicode[STIX]{x1D70C}:=\unicode[STIX]{x1D70C}(\text{}\underline{s}_{0},\overline{s}_{0})\unicode[STIX]{x1D70C}(\text{}\underline{s}_{1},\overline{s}_{1})\unicode[STIX]{x1D70C}(\text{}\underline{s}_{2},\overline{s}_{2})$
. In the case when
$\overline{\unicode[STIX]{x1D711}}_{1}=0=\text{}\underline{\unicode[STIX]{x1D711}}_{3}$
we pick
$\unicode[STIX]{x1D716}_{j}>0$
with
$\unicode[STIX]{x1D716}_{j}\rightarrow 0$
as
$j\rightarrow \infty$
, and in (7.3) replace
${\mathcal{P}}(\text{}\underline{t}_{j},\overline{t}_{j})$
by
${\mathcal{P}}_{\unicode[STIX]{x1D716}}(\text{}\underline{t}_{j},\overline{t}_{j})$
. Again (7.1), (7.2) will hold. Thus Theorem 1.1 is true when
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}<0<\overline{\unicode[STIX]{x1D711}}_{3}<1$
.
Next suppose
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}<0\leqslant \text{}\underline{\unicode[STIX]{x1D711}}_{3}<\overline{\unicode[STIX]{x1D711}}_{3}=1$
. Pick a sequence of points
$(\text{}\underline{s}_{0j},\overline{s}_{0j})\in T$
tending to
$(\text{}\underline{\unicode[STIX]{x1D711}}_{1},\overline{\unicode[STIX]{x1D711}}_{3})$
as
$j\rightarrow \infty$
, with
$\text{}\underline{s}_{0j}>\text{}\underline{\unicode[STIX]{x1D711}}_{1}$
,
${\textstyle \frac{1}{2}}(\text{}\underline{\unicode[STIX]{x1D711}}_{3}+\overline{\unicode[STIX]{x1D711}}_{3})<\overline{s}_{0j}<\overline{\unicode[STIX]{x1D711}}_{3}=1$
. Also set
so that for
$i=0,1,2$
,
where
$(\text{}\underline{s}_{i},\overline{s}_{i})$
is given by (6.11), (6.14), (6.18). Set
$(\text{}\underline{t}_{j},\overline{t}_{j})=(\text{}\underline{s}_{ij},\overline{s}_{ij})$
for
$i=0,1,2$
when
$j\equiv i(\text{mod}\,3)$
. When
$j\equiv 0(\text{mod}\,3)$
, then (6.9*) with
$\overline{t}_{j}$
in place of
$\overline{t}$
yields
so that
$\unicode[STIX]{x1D70C}(\text{}\underline{t}_{j},\overline{t}_{j})\geqslant 1-c_{j}^{\ast }>1-\unicode[STIX]{x1D713}>0$
. So if again
${\mathcal{P}}$
is defined by (7.3), with
${\mathcal{P}}(\text{}\underline{t}_{j},\overline{t}_{j})$
or
${\mathcal{P}}_{\unicode[STIX]{x1D716}}(\text{}\underline{t}_{j},\overline{t}_{j})$
, then again (7.1), (7.2) hold.
This takes care of the case when
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}<0<\overline{\unicode[STIX]{x1D711}}_{3}$
. Otherwise
$\text{}\underline{\unicode[STIX]{x1D711}}_{1}=0=\overline{\unicode[STIX]{x1D711}}_{3}$
, and
$\text{}\underline{\unicode[STIX]{x1D711}}_{i}=0=\overline{\unicode[STIX]{x1D711}}_{i}$
for
$i=1,2,3$
. But as is well known, this is true for almost every
$\boldsymbol{\unicode[STIX]{x1D709}}\in \mathbb{R}^{2}$
in the sense of Lebesgue measure. With a little effort one could construct a proper
$3$
-system
${\mathcal{P}}$
with each
$\text{}\underline{\unicode[STIX]{x1D711}}_{i}=\overline{\unicode[STIX]{x1D711}}_{i}=0$
. The proof of Theorem 1.1 is thus complete.
As for the remark below the Theorem, it is best possible since all the quantities
$\text{}\underline{\unicode[STIX]{x1D711}}_{i},\overline{\unicode[STIX]{x1D711}}_{i}$
are completely described. But one really should prove that
$\boldsymbol{\unicode[STIX]{x1D709}}=(\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2})$
with
$1,\unicode[STIX]{x1D709}_{1},\unicode[STIX]{x1D709}_{2}$
linearly independent over
$\mathbb{Q}$
with
$\overline{\unicode[STIX]{x1D711}}_{3}=1$
exist. We invite the reader to construct a proper
$3$
-system
${\mathcal{P}}$
with this property.
A. Appendix.
With the notation from [Reference Schmidt and Summerer4], equation (8.2) can be proved as follows. We have
It follows that on the one hand
and hence
On the other hand we get
and hence
The combination of these two estimates gives
$$\begin{eqnarray}\displaystyle 2\tilde{\unicode[STIX]{x1D712}}_{3}(m_{j}^{\ast })+\tilde{\unicode[STIX]{x1D712}}_{1}(m_{j}^{\ast }) & = & \displaystyle -\tilde{\unicode[STIX]{x1D712}}_{1}(a_{j}^{\ast })\biggl(3+2\tilde{\unicode[STIX]{x1D712}}_{3}(m_{j}^{\ast })+4\tilde{\unicode[STIX]{x1D712}}_{1}(m_{j}^{\ast })-3\frac{b_{j}^{\ast }-a_{j}^{\ast }}{m_{j}^{\ast }}\biggr)\nonumber\\ \displaystyle & & \displaystyle \displaystyle +\,\frac{3}{4}\frac{b_{j}^{\ast }-a_{j}^{\ast }}{m_{j}^{\ast }}+O(1/m_{j}^{\ast })\nonumber\\ \displaystyle & = & \displaystyle -\tilde{\unicode[STIX]{x1D712}}_{1}(a_{j}^{\ast })(3+2\tilde{\unicode[STIX]{x1D712}}_{3}(m_{j}^{\ast })+4\tilde{\unicode[STIX]{x1D712}}_{1}(m_{j}^{\ast }))\nonumber\\ \displaystyle & & \displaystyle \displaystyle +\,3\frac{b_{j}^{\ast }-a_{j}^{\ast }}{m_{j}^{\ast }}\biggl(\frac{1}{4}+\tilde{\unicode[STIX]{x1D712}}_{1}(a_{j}^{\ast })\biggr)+O(1/m_{j}^{\ast }).\nonumber\end{eqnarray}$$
Taking into account (7.4) the assertion in equation (8.2) follows.

