Kvant Math Problem 194
Take a small example, say $a=3$, $b=7$.
Verified: no
Verdicts: UNKNOWN + UNKNOWN
Solve time: 10m04s
Source on kvant.digital
Problem
Two coprime positive integers $a$ and $b$ are given. It is known that every integer can be represented in the form $ax+by$, where $x$ and $y$ are integers. Consider the set $M$ of integers that can be represented in the form $ax+by$, where $x$ and $y$ are nonnegative integers.
- What is the largest number $c$ that does not belong to the set $M$?
- Prove that of the two numbers $n$ and $c-n$ (where $n$ is any integer), one belongs to $M$ and the other does not.
(In Figure 2, for $a=3$ and $b=7$, the integer points belonging to the set $M$ are shown in red, and those not belonging to it are shown in blue.)
Figure 2
A. A. Kirillov
Exploration
Take a small example, say $a=3$, $b=7$. The numbers representable as $3x+7y$ with $x,y\ge0$ begin
$$0,3,6,7,9,10,12,13,14,\ldots$$
The missing positive integers are
$$1,2,4,5,8,11.$$
The largest missing number is $11=3\cdot7-3-7$.
This suggests the classical Frobenius number $ab-a-b$.
To understand why, consider all integers of the form
$$c-k,\qquad c=ab-a-b.$$
Then
$$c-k=(a-1)(b-1)-1-k.$$
A useful observation is that every integer has a unique residue modulo $a$. Since $b$ is coprime to $a$, the residues
$$0,b,2b,\ldots,(a-1)b$$
are all distinct modulo $a$. Hence every integer $n$ can be written uniquely as
$$n\equiv jb \pmod a, \qquad 0\le j\le a-1.$$
Write
$$n=jb+qa.$$
Then $n\in M$ exactly when $q\ge0$, because $j$ is already nonnegative.
For the special number $c=ab-a-b$,
$$c-n=(a-1-j)b+(-1-q)a.$$
The coefficients differ from those of $n$ in a very symmetric way. If $q\ge0$, then $n\in M$, while the coefficient of $a$ in the displayed representation of $c-n$ equals $-1-q<0$, suggesting that $c-n\notin M$. Conversely, if $q<0$, then perhaps $c-n\in M$.
The delicate point is proving that the displayed representation is the unique representation with $0\le j\le a-1$. Without uniqueness, a negative coefficient in one representation would not exclude membership in $M$.
The core insight is thus the unique decomposition
$$n=jb+qa,\qquad 0\le j\le a-1,$$
and the symmetry
$$c-n=(a-1-j)b+(-1-q)a.$$
Problem Understanding
We are given coprime positive integers $a$ and $b$. Let
$$M={ax+by:x,y\in\mathbb Z_{\ge0}}.$$
We must determine the largest integer not belonging to $M$, and prove a symmetry property: for
$$c=ab-a-b,$$
exactly one of the two integers $n$ and $c-n$ belongs to $M$.
This is a Type C problem, because we must determine an extremal value, namely the largest integer outside $M$, and prove that no larger one exists.
The main difficulty is finding a canonical description of all integers relative to the semigroup generated by $a$ and $b$, and then using it to establish the symmetry between $n$ and $c-n$.
The expected answer is
$$c=ab-a-b.$$
The intuition is that residues modulo $a$ are represented uniquely by $0,b,\ldots,(a-1)b$, and the number $ab-a-b$ sits exactly at the boundary where every larger integer becomes representable.
Proof Architecture
Lemma 1. Every integer $n$ admits a unique representation
$$n=jb+qa, \qquad 0\le j\le a-1,$$
with $q\in\mathbb Z$; uniqueness follows because multiplication by $b$ permutes the residue classes modulo $a$.
Lemma 2. For an integer written as in Lemma 1, $n\in M$ if and only if $q\ge0$; any representation with $0\le j\le a-1$ is canonical, and a nonnegative representation forces $q\ge0$.
Lemma 3. Let
$$c=ab-a-b.$$
If
$$n=jb+qa, \qquad 0\le j\le a-1,$$
then
$$c-n=(a-1-j)b+(-1-q)a.$$
This is a direct computation.
Lemma 4. Exactly one of $n$ and $c-n$ belongs to $M$; by Lemma 2, membership is determined by the sign of $q$, while for $c-n$ it is determined by the sign of $-1-q$.
Lemma 5. The number $c$ does not belong to $M$; applying Lemma 4 to $n=c$ shows that $0\in M$ and $c\notin M$.
Lemma 6. Every integer larger than $c$ belongs to $M$; if $n>c$, then $c-n<0$, and negative integers cannot lie in $M$, so Lemma 4 forces $n\in M$.
The most delicate point is Lemma 2, where one must prove that a canonical representation with $q<0$ cannot be rewritten with nonnegative coefficients.
Solution
Let
$$M={ax+by:x,y\in\mathbb Z_{\ge0}},$$
where $a$ and $b$ are coprime positive integers.
Since $\gcd(a,b)=1$, the numbers
$$0,b,2b,\ldots,(a-1)b$$
represent all residue classes modulo $a$ exactly once. Hence for every integer $n$ there exists a unique pair $(j,q)$ such that
$$n=jb+qa, \qquad 0\le j\le a-1, \qquad q\in\mathbb Z.$$
We first characterize membership in $M$.
Suppose
$$n=jb+qa, \qquad 0\le j\le a-1.$$
If $q\ge0$, then
$$n=qa+jb,$$
and both coefficients are nonnegative, so $n\in M$.
Conversely, assume $n\in M$. Then
$$n=xa+yb$$
for some $x,y\ge0$. Since
$$n=jb+qa,$$
we obtain
$$(y-j)b=(q-x)a.$$
Because $\gcd(a,b)=1$, there exists an integer $t$ such that
$$y-j=ta, \qquad q-x=tb.$$
Hence
$$y=j+ta.$$
Since $0\le j\le a-1$ and $y\ge0$, the integer $t$ cannot be negative. Indeed, if $t\le-1$, then
$$y=j+ta\le(a-1)-a=-1,$$
contrary to $y\ge0$.
Therefore $t\ge0$, and
$$q=x+tb\ge0.$$
We have proved
$$n\in M \quad\Longleftrightarrow\quad q\ge0.$$
Now define
$$c=ab-a-b.$$
Let
$$n=jb+qa, \qquad 0\le j\le a-1.$$
Then
$$\begin{aligned} c-n &=(ab-a-b)-(jb+qa)\ &=(a-1-j)b+(-1-q)a. \end{aligned}$$
Since
$$0\le a-1-j\le a-1,$$
this is the canonical representation of $c-n$.
Applying the criterion already proved, we obtain
$$n\in M \quad\Longleftrightarrow\quad q\ge0,$$
and
$$c-n\in M \quad\Longleftrightarrow\quad -1-q\ge0.$$
The inequalities
$$q\ge0, \qquad -1-q\ge0$$
cannot hold simultaneously, and one of them must hold because $q$ is an integer.
Hence exactly one of the numbers $n$ and $c-n$ belongs to $M$.
This proves part 2.
To determine the largest integer outside $M$, substitute $n=c$ into the statement just proved. Since
$$c-c=0\in M,$$
the number $c$ itself does not belong to $M$.
Now let $n>c$. Then
$$c-n<0.$$
Every element of $M$ is nonnegative, so $c-n\notin M$. By the property proved above, exactly one of $n$ and $c-n$ belongs to $M$, therefore $n\in M$.
Thus every integer greater than $c$ lies in $M$, while $c$ does not. Consequently $c$ is the largest integer not belonging to $M$.
Therefore
$$\boxed{c=ab-a-b}.$$
Equality is attained by the number $ab-a-b$, and every integer larger than it belongs to $M$.
Verification of Key Steps
The first delicate step is the characterization of $M$ by the sign of $q$.
Assume
$$n=jb+qa, \qquad 0\le j\le a-1,$$
and also
$$n=xa+yb, \qquad x,y\ge0.$$
From
$$(y-j)b=(q-x)a$$
we deduce
$$y-j=ta.$$
If one forgets to use $0\le j\le a-1$, one might incorrectly allow negative $t$. The bound on $j$ gives
$$y=j+ta<0$$
whenever $t\le-1$, contradicting $y\ge0$. This forces $t\ge0$ and hence $q=x+tb\ge0$.
The second delicate step is verifying that the expression for $c-n$ is canonical. We obtain
$$c-n=(a-1-j)b+(-1-q)a.$$
Since
$$0\le j\le a-1,$$
we have
$$0\le a-1-j\le a-1.$$
Without checking this range, one could not apply the membership criterion to $c-n$.
The third delicate step is the exclusivity argument. For an integer $q$, either $q\ge0$ or $q\le-1$. In the first case,
$$-1-q<0.$$
In the second case,
$$-1-q\ge0.$$
Thus exactly one of the two coefficients governing membership is nonnegative. There is no exceptional value of $q$ where both conditions fail or both hold.
Alternative Approaches
A classical geometric approach considers lattice points $(x,y)$ with $x,y\ge0$ and the lines
$$ax+by=n.$$
For each residue class modulo $a$, one identifies the smallest representable number and counts the nonrepresentable integers below $ab$. The boundary point corresponding to $(b-1,a-1)$ leads to the value $ab-a-b$. The symmetry arises from reflecting lattice points inside the rectangle
$$0\le x\le b-1,\qquad 0\le y\le a-1.$$
Another approach works entirely modulo $a$. For each residue class choose the least element of $M$ in that class, namely
$$0,b,2b,\ldots,(a-1)b.$$
An integer in the residue class of $jb$ is representable precisely when it is at least $jb$. This immediately yields the largest nonrepresentable integer as
$$\max_{0\le j\le a-1}\bigl(jb-a\bigr)=ab-a-b.$$
The canonical-representation method used in the main proof is preferable because it simultaneously yields the extremal value and the symmetry statement in a single calculation.