Find all functions f:Z→Z satisfying: (1) For any positive integer M, there exists an integer k with ∣f(k)∣≥M; (2) For any integers m,n, 2f(m)f(n)−f(m−n)−1 is a perfect square.
Solution
Let α=3+22. All functions satisfying the given conditions are of the form f(n)=21(α2n+α−2n) for some fixed positive integer t.
First, we verify that such f(n) satisfies (⋆). The left-hand side of (⋆) becomes: 2⋅2α2tm+α−2tm⋅2α2tn+α−2tn−2α2t(m−n)+α−2t(m−n)−1=2α2t(m+n)+α−2t(m+n)−1=(2αt(m+n)−α−t(m+n))2. Note that 2αt(m+n)−α−t(m+n)=2(3+22)t(m+n)−(3−22)t(m+n) is an integer, so the above expression is indeed a perfect square.
We now show that these are the only solutions to the functional equation (⋆). Let Am,n be the non-negative integer such that: 2f(m)f(n)−f(m−n)−1=Am,n2.(⋆⋆)
Step 1: All f(n) have the same sign. This follows by setting n=0 in (⋆⋆), which gives (2f(0)−1)f(m)>0. Thus, all f(n) have the same sign as 2f(0)−1.
Step 2:f(n) is an even function. For a fixed t∈Z+, let Mt=max{∣f(t)∣,∣f(−t)∣}. Since f is unbounded, there exists N such that ∣f(N+t)∣>(Mt+1)2. Substituting m=N+t,n=N and m=N,n=N+t into (⋆⋆) yields: 2f(N+t)f(N)−f(t)−1=AN,N+t2,2f(N+t)f(N)−f(−t)−1=AN+t,N2. This implies AN,N+t2>2(Mt+1)2−Mt−1>Mt2, so AN,N+t>Mt; similarly, AN+t,N>Mt. Subtracting these equations gives: f(−t)−f(t)=(AN,N+t−AN+t,N)(AN,N+t+AN+t,N). The left-hand side satisfies ∣f(−t)−f(t)∣<2Mt, while AN,N+t+AN+t,N>2Mt. Analyzing the magnitudes shows that AN,N+t=AN+t,N, so f(t)=f(−t), proving f is even.
Step 3: Determine the general form of f(n). Setting m=n in (⋆⋆) gives: 2f(n)2−f(0)−1=An,n2 i.e., An,n2−2f(n)2=−f(0)−1.(5) Thus, (An,n,f(n)) is a solution to the generalized Pell equation: X2−2Y2=−f(0)−1.(6) We use the theory of solutions to generalized Pell equations. Let Z[2] be the ring of numbers Z=X+2Y with X,Y∈Z. The conjugate of Z is Z′=X−2Y, and (Z1Z2)′=Z1′Z2′. For α=3+22, we have α′=α−1. Solving (6) is equivalent to solving ZZ′=−f(0)−1 in Z[2]. If Z is a solution, then all Zα2s (s∈Z) are also solutions.
Lemma: There exist finitely many fundamental solutions Zi=Xi+2Yi∈Z[2] (i=1,…,t) to (6) such that for any solution (X,Y), there exists i∈{1,…,t} and s∈Z≥0 with X+2Y=Ziα2s.
Proof of Lemma: Consider the lattice L:={(a+2b,a−2b)∣a,b∈Z}. Solutions (X,Y) to (6) correspond to points (x,y)=(X+2Y,X−2Y) on the hyperbola xy=−f(0)−1. Multiplying X+2Y by α2 or α−2 generates new solutions, equivalent to moving (x,y) to (α2x,α−2y) or (α−2x,α2y) on the hyperbola. By these transformations, any solution can be mapped to a region where ∣x∣≤α∣f(0)+1∣ and ∣y∣≤α∣f(0)+1∣, corresponding to finitely many lattice points (Xi+2Yi,Xi−2Yi) (i=1,…,t). Thus, any solution can be written as (Xi+2Yi)α2s. □
Fix the fundamental solutions Zi=Xi+2Yi (i=1,…,t) from the lemma, with ZiZi′=−f(0)−1. We may assume no two Zi differ by a power of α2. For each n∈Z, there exist unique in∈{1,…,t} and sn∈Z such that: f(n)=22Zinα2sn−Zin′α−2sn.(7) Let N=maxi{∣Zi∣,∣Zi′∣,∣Zi∣−1,∣Zi′∣−1}. From (7), we have the estimate: N−1α2∣sn∣−N≤∣f(n)∣≤Nα2∣sn∣+N.(8)
Step 4: For positive integers m>n, we have the growth estimate: ∣∣sm+n−∣sm∣∣<∣sn∣+2logαN+2.(9) This follows from substituting −n for n in (⋆⋆), yielding: ∣f(m+n)∣<2f(m)f(−n)=2f(m)f(n).(10) Combining with (8) gives: N−1α2∣sm+n∣−N≤2(Nα2∣sn∣+N)(Nα2∣sm∣+N). Simplifying, we obtain: N−1⋅α2∣sm+n∣≤2N2⋅((α2∣sn∣+1)(α2∣sm∣+1)+1)≤N2⋅α2∣sn∣+2∣sm∣+3 which implies ∣sm+n∣<∣sn∣+∣sm∣+2logαN+2. Similarly, substituting m+n for m in (⋆⋆) gives: 2f(m+n)f(n)>∣f(m)∣,(11) leading to ∣sm+n∣+∣sn∣+2logαN+2>∣sm∣. Combining these inequalities proves (9).
Step 5: Prove that f(0)=1. First, let's explain the basic idea. Suppose for indices m,n we have Zim=Zin and sn has the same sign as sm. Substituting the expressions for f(m) and f(n) from (7) into (⋆⋆), we obtain: Bm,n2=2f(m)f(n)−f(m−n)−1=4(Zimα2sm−Zim′α−2sm)(Zinα2sn−Zin′α−2sn)−22Zim−nα2sm−n−Zim−n′α−2sm−n−1=(2Zimαsm+sn+Zim′α−sm−sn)2−4ZimZim′(αsm−sn+αsn−sm)2−22(Zim−nα2sm−n−Zim−n′α−2sm−n)−1=(2Zimαsm+sn+Zim′α−sm−sn)2+(f(0)+1)4(αsm−sn+αsn−sm)2−22(Zim−nα2sm−n−Zim−n′α−2sm−n)−1.(12) Note that 21(Zimαsm+sn+Zim′α−sm−sn) is an integer greater than (2N)−1α∣sm+sn∣−1−N. Therefore, if sm and sn are very close and both have large absolute values, all terms except this squared term sum to zero. Now we use this property specifically to show f(0)=1. Assume by contradiction that f(0)=1. First, we show that for any i∈{1,…,t}, the sets {221Ziα2s∣s∈Z≥0}∪{221Zi′α2s∣s∈Z≥0} and {41(f(0)+1)α2s} have no common elements. This is because the product of these numbers with their conjugates differs: 221Ziα2s⋅−221Zi′α−2s=8f(0)+1=161(f(0)+1)2, (since f(0)=1). Thus there exists an integer M>2logα(N)+2 such that all numbers of the form 4f(0)+1α2s and 22Ziα2s or 22Zi′α2s (for ∣s∣≥M) are pairwise separated by more than ∣f(0)−1∣+N2. Since f is unbounded, there exist n1<⋯<n2t+1 satisfying: ∣sn1∣>2M,∣sni+1∣>2∣sni∣(i=1,…,2t). Choose L such that ∣sL∣>5⋅22t+1M. Consider: iL+n1,iL+n2,…,iL+n2t+1∈{1,…,t} and the signs of sL+nj. By the pigeonhole principle, there must exist x>y≥1 such that iL+nx=iL+ny=i and sL+nx has the same sign as sL+ny. From inequality (9), we estimate: ∣s(L+nx)−(L+ny)∣=∣snx−ny∣>∣snx−sny∣−(2logα(N)+2)>M. ∣sL+nx∣>∣sL∣−∣snx∣−2logα(N)+2>4⋅2t+1M;∣sL+ny∣>4⋅2t+1M. Returning to our earlier situation, set m=L+nx, n=L+ny, and observe that: 2Ziαsm+sn+Zi′α−sm−sn>N−1α4⋅2t+1M−N is much larger than the square of the remaining terms. Therefore, we must have: (f(0)+1)4(αsm−sn+αsn−sm)2=22(Zim−nα2sm−n−Zim−n′α−2sm−n)+1. But by the definition of M, this equality cannot hold. Contradiction!
Step 6: Complete the proof. With f(0)=1, (5) simplifies to: 2f(n)2−2=An,n2. Thus, An,n is even, and we rewrite the equation as: f(n)2−2(2An,n)2=1. The solutions to this standard Pell equation are: f(n)=2αtn+α−tn,tn∈Z≥0. Substituting n=0 into (★) gives: Am,02=2f(m)f(0)−f(m)−1=f(m)−1=(2(2+1)tm−(2−1)tm)2, implying tn must be even. Let tn=2sn. Substituting into (10) and (11), we obtain for m>n: 21(α2sm+α−2sm)(α2sn+α−2sn)>21(α2sm+n+α−2sm+n), 21(α2sm+n+α−2sm+n)(α2sn+α−2sn)>21(α2sm+α−2sm), which implies: sm+sn≥sm+nandsm+n+sn≥sm.(13) The following shows that for any ℓ∈N, we have sℓ=ℓs1. Fix ℓ, and let D:=max{2,s1,…,sℓ}.
Consider any positive integer N>ℓ satisfying sN>5D. From the given conditions, for i=−1,0,…,ℓ−1, we have sN+i>4D. For i=0,…,ℓ−1, taking m=N+i and n=N−1 in (12) yields: BN+i,N−12=21(α2sN+i+α−2sN+i)(α2sN−1+α−2sN−1)−21(α2si+1−α−2si+1)−1=(2αsN+i+sN−1−α−sN+i−sN−1)2+2(α2sN+ℓ−2sN−1+α2sN−1−2sN+ℓ)−(α2si+1+α−2si+1) Since 2αsN+i+sN−1−α−sN+i+sN−1 is an integer greater than α6D−1, while the remaining terms are less than α2D+1, it must hold that: α2sN+ℓ−2sN−1+α2sN−1−2sN+ℓ=α2si+1+α−2si+1. This implies ∣sN+i−sN−1∣=si+1. Since f is unbounded, choose N>ℓ such that sN>6D and sN>sN−1. From the above discussion, we have: ∣sN+1−sN∣=s1,sN−sN−1=s1,∣sN+1−sN−1∣=s2, so either s2=0 or s2=2s1. If s2=0, we can use sN>6D and (13) to inductively prove that sN+u+2=sN+u for all u∈N. This contradicts the unboundedness of f.
Therefore, s2=2s1, which implies sN+1=sN−1+2s1. Next, we examine the conditions that sN+2 must satisfy: ∣sN+2−sN∣=s2=2s1,∣sN+2−sN+1∣=s1. Thus sN+2=sN+2s1=sN−1+3s1, which gives s3=∣sN+2−sN−1∣=3s1. Continuing this recursion, we obtain sℓ=ℓs1. This proves that all solutions f satisfying the functional equation are of the form: f(n)=21(α2ns1+α−2ns1).
Want a route through all this instead of an archive? The track
puts 2,000 problems in a working order, from AMC 10 level to the IMO shortlist.
Source: MathNet,
licensed CC-BY-4.0.
Statement and solution reproduced as published; topic and difficulty added by this site.