Suppose a doubly infinite sequence of real numbers
has the property that
Show that if this sequence is bounded (i.e., if there exists a number such that for all ), then has the same value for all .
Solutions — 3
Solution 1
Assume that the sequence is not constant. Define a function by the equation
or equivalently,
where and are the largest and the smallest, respectively, of the three numbers .
Note that if for any one value of , then the original sequence would be constant. Thus by our assumption, we must have for all . It is also trivial that is unbounded if the function is unbounded. We define to be the closed interval of values between the numbers and . Unlike the usual meaning, we do not insist that , so is always an interval of length .
Claim 1: is monotonically decreasing.
By translation invariance of the indices, it suffices to show that . Bearing in mind that is an arithmetic average of , and , we see that if , then . Alternatively, if , then either or both and are on the same side of , with being strictly closer than to .
By the above considerations and (3), we deduce the following. If , then . If instead , then either , in which case , or , in which case . This shows that with equality iff .
Claim 2: We cannot have for two consecutive values of .
As in Claim 1, we reduce to the statement that if , then . Suppose . As noted above, we then have , which implies that unless . But if , then our assumption that forces , contradicting the initial assumption that is non-constant. Claim 2 now follows.
Claim 3: , .
In view of Claims 1 and 2, it suffices to show that whenever . As usual, we may assume that . The function is invariant under translations of the sequence, and under multiplication by . So we may assume that , and that and are both positive. Since
it follows that
If , then by (4), we have and
and so , as required.
If instead , then and
By Claim 1, we conclude that , as required.
Finally, if , then and
and so , as required. We have proved Claim 3.
With Claim 3 in hand, the unboundedness of for negative follows easily.
Solution 2
Let , so we have for all , and in particular for . The unique solution is for , where and are the three roots of the equation , and depend on , and . Since every is real, is real and . Thus , . We assume that is non-constant, and so .
Let be the argument function. Call good if . Writing , we have whenever is good.
Now , so
Thus the number "hops more than a full circuit around the unit circle" as ranges over a set of the form
and every contains a good number (because ). We can pick arbitrarily large good numbers , and for each good we have . But if were bounded, and hence bounded, then would be bounded for all . This is impossible, since and so if , then , and (where log indicates a logarithm to any preferred base).
Solution 3
This solution is similar to Solution 2, except that we take the more obvious approach of using the given recurrence relation directly. The solution is for , where and are the three roots of the equation , and depend on , and . Since every is real, is real and . Thus for ; we call this Formula 1, and we need to show that it is also valid for . Assume that is non-constant, and so .
Suppose that . The sequence , where , satisfies the same recurrence relation as , so it can be written in the same form: for ; we call this Formula 2. To show that Formula 1 extends to , we need to show that and .
Let . Comparing Formulae 1 and 2, we see that
Taking half the difference of two such equations (for and ), we get
and so
Now takes on different values for , and these values do not differ by , so it follows that . Now (5) tells us that , i.e. , and so Formula 1 is valid for .
Since is arbitrary, our solution for is also valid for . The value of is the same as in Solution 2. Thus as before we see that there are infinitely many good values of for which . But , so as in Solution 2, boundedness is impossible.