Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated

Proof of the averaged near-square construction


This proves the whole statement of L251. Throughout, log⁡\log is the natural logarithm, e(u):=e2πiue(u):=e^{2\pi iu}, and

ψ(u):=u−⌊u⌋−12.\psi(u):=u-\lfloor u\rfloor-\tfrac12 .

For an integer C≥1C\ge1 and a real X≥0X\ge0 put

SC(X):=∑C<c≤2Cψ(X/c),YX:=X/C.S_C(X):=\sum_{C<c\le2C}\psi\Bigl(\sqrt{X/c}\Bigr),\qquad Y_X:=\sqrt{X/C}.

1. Two elementary lemmas

Lemma 1 (adjustment). Let c≥1c\ge1, q≥1q\ge1 and m≥0m\ge0 be integers with D:=cq2−m>0D:=cq^2-m>0, and let tt be the largest integer t≥0t\ge0 with ct2<Dct^2<D. Then N:=c(q−t)(q+t)N:=c(q-t)(q+t) satisfies 0<N−m≤c(2t+1)0<N-m\le c(2t+1), 0≤t<q0\le t<q and t<D/ct<\sqrt{D/c}.

Proof. As in the proof of L250: tt exists because c⋅02<Dc\cdot0^2<D and ct2<Dct^2<D bounds tt; then N−m=D−ct2>0N-m=D-ct^2>0, maximality gives c(t+1)2≥Dc(t+1)^2\ge D and hence N−m≤c(2t+1)N-m\le c(2t+1), and ct2<D≤cq2ct^2<D\le cq^2 gives t<qt<q and t<D/ct<\sqrt{D/c}. □\square

Lemma 2 (count). Let C≥1C\ge1 be an integer and 0≤x<x′0\le x<x' real. The number of integer pairs (c,q)(c,q) with C<c≤2CC<c\le2C, q≥1q\ge1 and x<cq2≤x′x<cq^2\le x' equals

∑C<c≤2C(x′/c−x/c)−SC(x′)+SC(x).\sum_{C<c\le2C}\Bigl(\sqrt{x'/c}-\sqrt{x/c}\Bigr)-S_C(x')+S_C(x).

Proof. For fixed cc the admissible qq are the integers in (x/c,x′/c](\sqrt{x/c},\sqrt{x'/c}], all positive because x/c≥0\sqrt{x/c}\ge0. Their number is ⌊x′/c⌋−⌊x/c⌋\lfloor\sqrt{x'/c}\rfloor-\lfloor\sqrt{x/c}\rfloor, and ⌊u⌋=u−ψ(u)−12\lfloor u\rfloor=u-\psi(u)-\frac12. Sum over cc. □\square

2. The two analytic inputs

Vaaler's inequality. Lemma 7.1 of Granville and Ramaré (Mathematika 43 (1996), 73--107), from Vaaler (Bull. Amer. Math. Soc. 12 (1985), 183--216), states for every integer R≥1R\ge1

±ψ0(t)≤12R+2+∑0<∣r∣≤Rar±e(rt),ar±=iαr±βr,\pm\psi_0(t)\le\frac1{2R+2}+\sum_{0<|r|\le R}a_r^{\pm}e(rt), \qquad a_r^{\pm}=i\alpha_r\pm\beta_r,

where ψ0\psi_0 is their sawtooth function (ψ0=ψ\psi_0=\psi off the integers and ψ0=0\psi_0=0 at the integers), and

αr=12π(R+1)(π(1−∣r∣R+1)cot⁡πrR+1+∣r∣r),βr=12R+2(1−∣r∣R+1).\alpha_r=\frac{1}{2\pi(R+1)}\Bigl(\pi\Bigl(1-\frac{|r|}{R+1}\Bigr) \cot\frac{\pi r}{R+1}+\frac{|r|}{r}\Bigr),\qquad \beta_r=\frac{1}{2R+2}\Bigl(1-\frac{|r|}{R+1}\Bigr).

These are the held statement as printed (read on the page image of p. 27 of the held preprint, whose labels the library card follows). The upper sign gives

ψ0(t)≤12R+2+∑0<∣r∣≤R(iαr+βr)e(rt).(V+)\psi_0(t)\le\frac1{2R+2}+\sum_{0<|r|\le R}(i\alpha_r+\beta_r)e(rt). \tag{V+}

As printed, the lower sign is a misprint: as t→0+t\to0^+ its left side tends to 12\frac12, while its right side tends to 12R+2−∑0<∣r∣≤Rβr=1−R2R+2≤0\frac1{2R+2}-\sum_{0<|r|\le R}\beta_r=\frac{1-R}{2R+2}\le0, because the αr\alpha_r are odd in rr and cancel. The correct lower inequality follows from (V+): ψ0\psi_0 is odd, α−r=−αr\alpha_{-r}=-\alpha_r and β−r=βr\beta_{-r}=\beta_r, so (V+) at −t-t, reindexed by r↦−rr\mapsto-r, reads

−ψ0(t)≤12R+2+∑0<∣r∣≤R(−iαr+βr)e(rt).(V-)-\psi_0(t)\le\frac1{2R+2}+\sum_{0<|r|\le R}(-i\alpha_r+\beta_r)e(rt). \tag{V-}

Both (V+) and (V-) also hold with ψ\psi in place of ψ0\psi_0: the two functions differ only at integers tt, where both right sides equal 12R+2+2∑r=1Rβr=12=∣ψ(t)∣\frac1{2R+2}+2\sum_{r=1}^R\beta_r=\frac12=|\psi(t)|.

For the moduli, put x=r/(R+1)∈(0,1)x=r/(R+1)\in(0,1) for 1≤r≤R1\le r\le R. Then 0≤π(1−x)cot⁡(πx)+1≤1/x0\le\pi(1-x)\cot(\pi x)+1\le1/x: the upper bound is πxcot⁡(πx)≤1\pi x\cot(\pi x)\le1, and the lower bound is trivial for x≤12x\le\frac12 and, for x>12x>\frac12, is πycot⁡(πy)≤1\pi y\cot(\pi y)\le1 with y=1−xy=1-x. Hence 0≤αr≤1/(2πr)0\le\alpha_r\le1/(2\pi r), and 0≤βr≤1/(2R+2)0\le\beta_r\le1/(2R+2), so

∣±iαr+βr∣=αr2+βr2≤12π∣r∣+12R+2(0<∣r∣≤R).|\pm i\alpha_r+\beta_r|=\sqrt{\alpha_r^2+\beta_r^2} \le\frac1{2\pi|r|}+\frac1{2R+2}\qquad(0<|r|\le R).

The third-derivative test. Lemma 8.6 of the same paper, with k=2k=2 (statement read on the page image of p. 30 of the held preprint): suppose that on [A,B][A,B] the function ff is four times differentiable with f′′′f''' monotonic and

0<m2≤f′′′(t)≤12Q3/20<m_2\le f'''(t)\le\frac{1}{2Q^{3/2}}

for some integer Q∈[4,N]Q\in[4,N], where NN is an integer and ]A,B]]A,B] contains at most NN integers. Then

∣18N∑A<n≤Be(±f(n))∣4<18Q+3log⁡2Q4πm2NQ3/2.\Bigl|\frac1{8N}\sum_{A<n\le B}e(\pm f(n))\Bigr|^4 <\frac1{8Q}+\frac{3\log^2Q}{4\pi m_2NQ^{3/2}} .

(The printed factor is 3k−k/2/(2πmkN)3k^{-k/2}/(2\pi m_kN) times log⁡kQ/Q2−2−k+1\log^kQ/Q^{2-2^{-k+1}}, which is the displayed term at k=2k=2.)

3. The multiplier sum

Lemma 3. Let C≥4C\ge4 be an integer and X>0X>0 real with

C3/2≤YX≤C3/120.C^{3/2}\le Y_X\le C^3/120 .

Then

∣SC(X)∣≤60 C4/7YX1/7+40 C3/4(log⁡C)3/2.|S_C(X)|\le60\,C^{4/7}Y_X^{1/7}+40\,C^{3/4}(\log C)^{3/2}.

Proof. Write Y=YXY=Y_X. For an integer R≥1R\ge1 and 1≤r≤R1\le r\le R put

Tr:=∑C<c≤2Ce(rX/c),T_r:=\sum_{C<c\le2C}e\Bigl(r\sqrt{X/c}\Bigr),

so that T−r=Tr‾T_{-r}=\overline{T_r} for the sums with negative frequency.

Step 1 (Vaaler). Sum (V+) and (V-) for ψ\psi over the points t=X/ct=\sqrt{X/c}. Since SC(X)S_C(X) is real,

±SC(X)≤C2R+2+∑0<∣r∣≤R∣±iαr+βr∣ ∣Tr∣,\pm S_C(X)\le\frac{C}{2R+2}+\sum_{0<|r|\le R}|\pm i\alpha_r+\beta_r|\,|T_r|,

and the bound on the moduli, with ∣T−r∣=∣Tr∣|T_{-r}|=|T_r| and 1/(R+1)<1/r1/(R+1)<1/r for r≤Rr\le R, gives

∣SC(X)∣≤C2R+2+(1+1π)∑r=1R∣Tr∣r.(1)|S_C(X)|\le\frac{C}{2R+2}+\Bigl(1+\frac1\pi\Bigr)\sum_{r=1}^{R}\frac{|T_r|}{r}. \tag{1}

Step 2 (one frequency). Fix r≥1r\ge1 with rY/C3≤1/30rY/C^3\le1/30. Put h(u):=−rX u−1/2h(u):=-r\sqrt X\,u^{-1/2}, so that e(rX/c)=e(−h(c))e(r\sqrt{X/c})=e(-h(c)) and

h′′′(u)=158 rX u−7/2>0,h'''(u)=\tfrac{15}{8}\,r\sqrt X\,u^{-7/2}>0 ,

which decreases on [C,2C][C,2C]. Since X=YC\sqrt X=Y\sqrt C, its maximum and minimum there are

M2=h′′′(C)=158 rYC3,m2=h′′′(2C)=2−7/2M2.M_2=h'''(C)=\tfrac{15}{8}\,\frac{rY}{C^3},\qquad m_2=h'''(2C)=2^{-7/2}M_2 .

Let Z:=(2M2)−2/3Z:=(2M_2)^{-2/3} and Q:=⌊Z⌋Q:=\lfloor Z\rfloor. The condition rY/C3≤1/30rY/C^3\le1/30 gives 2M2≤182M_2\le\frac18, so Z≥4Z\ge4 and Q≥4Q\ge4; also Q≥Z−1≥34ZQ\ge Z-1\ge\frac34Z. The condition Y≥C3/2Y\ge C^{3/2} gives 2M2≥C−3/22M_2\ge C^{-3/2}, so Q≤Z≤CQ\le Z\le C. From Q≤ZQ\le Z we get h′′′≤M2≤1/(2Q3/2)h'''\le M_2\le1/(2Q^{3/2}). So the third-derivative test applies with [A,B]=[C,2C][A,B]=[C,2C], N=CN=C (the interval ]C,2C]]C,2C] contains exactly CC integers) and the lower sign, and (a+b)1/4≤a1/4+b1/4(a+b)^{1/4}\le a^{1/4}+b^{1/4} gives

∣Tr∣<83/4CQ−1/4+8(34π)1/4C3/4(log⁡Q)1/2 m2−1/4Q−3/8.|T_r|<8^{3/4}CQ^{-1/4} +8\Bigl(\frac{3}{4\pi}\Bigr)^{1/4}C^{3/4}(\log Q)^{1/2}\,m_2^{-1/4}Q^{-3/8}.

Now Q−1/4≤(4/3)1/4(2M2)1/6Q^{-1/4}\le(4/3)^{1/4}(2M_2)^{1/6} and Q−3/8≤(4/3)3/8(2M2)1/4Q^{-3/8}\le(4/3)^{3/8}(2M_2)^{1/4}, while 2M2/m2=29/22M_2/m_2=2^{9/2} and log⁡Q≤log⁡C\log Q\le\log C. Hence

∣Tr∣≤κ1 C(rYC3)1/6+κ2 C3/4(log⁡C)1/2,|T_r|\le\kappa_1\,C\Bigl(\frac{rY}{C^3}\Bigr)^{1/6} +\kappa_2\,C^{3/4}(\log C)^{1/2},

with κ1=83/4(4/3)1/4(15/4)1/6=6.371…<6.4\kappa_1=8^{3/4}(4/3)^{1/4}(15/4)^{1/6}=6.371\ldots<6.4 and κ2=8(3/(4π))1/4(4/3)3/829/8=13.58…<13.6\kappa_2=8(3/(4\pi))^{1/4}(4/3)^{3/8}2^{9/8}=13.58\ldots<13.6.

Step 3 (choice of RR). Let R:=⌈C3/7Y−1/7⌉R:=\lceil C^{3/7}Y^{-1/7}\rceil. Since Y≤C3Y\le C^3, the number C3/7Y−1/7C^{3/7}Y^{-1/7} is at least 11, so C3/7Y−1/7≤R≤2C3/7Y−1/7C^{3/7}Y^{-1/7}\le R\le2C^{3/7}Y^{-1/7}. Every r≤Rr\le R satisfies

rYC3≤2Y6/7C18/7≤130,\frac{rY}{C^3}\le\frac{2Y^{6/7}}{C^{18/7}}\le\frac1{30},

because Y≤C3/120Y\le C^3/120 and 607/6=118.7…<12060^{7/6}=118.7\ldots<120; so Step 2 applies to every r≤Rr\le R. Using ∑r≤Rr−5/6≤1+6(R1/6−1)≤6R1/6\sum_{r\le R}r^{-5/6}\le1+6(R^{1/6}-1)\le6R^{1/6} and ∑r≤Rr−1≤1+log⁡R\sum_{r\le R}r^{-1}\le1+\log R in (1),

∣SC(X)∣≤C2R+6κ1(1+1π)C1/2Y1/6R1/6+κ2(1+1π)C3/4(log⁡C)1/2(1+log⁡R).|S_C(X)|\le\frac{C}{2R} +6\kappa_1\Bigl(1+\frac1\pi\Bigr)C^{1/2}Y^{1/6}R^{1/6} +\kappa_2\Bigl(1+\frac1\pi\Bigr)C^{3/4}(\log C)^{1/2}(1+\log R).

The first term is at most 12C4/7Y1/7\frac12C^{4/7}Y^{1/7}. In the second, R1/6≤21/6C1/14Y−1/42R^{1/6}\le2^{1/6}C^{1/14}Y^{-1/42}, so the term is at most 6⋅6.4⋅1.3184⋅1.1225 C4/7Y1/7<56.9 C4/7Y1/76\cdot6.4\cdot1.3184\cdot1.1225\,C^{4/7}Y^{1/7}<56.9\,C^{4/7}Y^{1/7}. In the third, R≤2C3/7R\le2C^{3/7} and C≥4C\ge4 give 1+log⁡R≤1+log⁡2+37log⁡C≤2log⁡C1+\log R\le1+\log2+\frac37\log C\le2\log C, so the term is at most 2⋅13.6⋅1.3184 C3/4(log⁡C)3/2<35.9 C3/4(log⁡C)3/22\cdot13.6\cdot1.3184\,C^{3/4}(\log C)^{3/2}<35.9\,C^{3/4}(\log C)^{3/2}. Adding the three bounds proves the lemma. □\square

4. Proof of the statement

Let K≥64K\ge64 and 0≤m≤2−15K28/110\le m\le2^{-15}K^{28/11} be integers. If m≤K5/2/128m\le K^{5/2}/128, then L250 supplies the required integer. Assume from now on that K5/2/128<m≤2−15K28/11K^{5/2}/128<m\le2^{-15}K^{28/11}. This range is nonempty only if K1/22>215/128=28K^{1/22}>2^{15}/128=2^8, so

K>2176,log⁡K>176log⁡2>121.9.K>2^{176},\qquad \log K>176\log2>121.9 .

Put

C:=⌈16mK2⌉,E:=⌊K2128C⌋,Y:=Ym=m/C.C:=\Bigl\lceil\frac{16m}{K^2}\Bigr\rceil,\qquad E:=\Bigl\lfloor\frac{K^2}{128C}\Bigr\rfloor,\qquad Y:=Y_m=\sqrt{m/C}.

(F1) Size of CC. C≥16m/K2>K/8>285C\ge16m/K^2>\sqrt K/8>2^{85}. Also C<16m/K2+1≤2−11K6/11+1≤2−10K6/11C<16m/K^2+1\le2^{-11}K^{6/11}+1\le2^{-10}K^{6/11}, since 2−11K6/11>2−11296≥12^{-11}K^{6/11}>2^{-11}2^{96}\ge1.

(F2) Size of YY. C≥16m/K2C\ge16m/K^2 gives Y≤K/4Y\le K/4. Since K≥64K\ge64, m>K5/2/128≥K2/16m>K^{5/2}/128\ge K^2/16, so 16m+K2≤32m16m+K^2\le32m and m/C>m/(16m/K2+1)=mK2/(16m+K2)≥K2/32m/C>m/(16m/K^2+1)=mK^2/(16m+K^2)\ge K^2/32; thus Y>K/(42)Y>K/(4\sqrt2).

(F3) Size of EE. E≤K2/(128C)<K2/16≤mE\le K^2/(128C)<K^2/16\le m, and E≥K2/(128C)−1≥K2/(256C)E\ge K^2/(128C)-1\ge K^2/(256C) because C≤K≤K2/256C\le K\le K^2/256.

So for X∈{m,m+E}X\in\{m,m+E\} we have m≤X≤2mm\le X\le2m and K/(42)<YX≤2 Y≤K/(22)K/(4\sqrt2)<Y_X\le\sqrt2\,Y\le K/(2\sqrt2).

Lemma 3 applies to X=mX=m and X=m+EX=m+E. C≥4C\ge4 by (F1). By (F1), C3/2≤2−15K9/11<K/(42)<YXC^{3/2}\le2^{-15}K^{9/11}<K/(4\sqrt2)<Y_X. By (F1), C3>K3/2/512C^3>K^{3/2}/512, and K1/2>288K^{1/2}>2^{88} gives K3/2/(512⋅120)≥K/(22)≥YXK^{3/2}/(512\cdot120)\ge K/(2\sqrt2)\ge Y_X, so YX≤C3/120Y_X\le C^3/120. Since YX1/7≤(K/(22))1/7<0.863 K1/7Y_X^{1/7}\le(K/(2\sqrt2))^{1/7}<0.863\,K^{1/7},

∣SC(m)∣+∣SC(m+E)∣≤103.6 C4/7K1/7+80 C3/4(log⁡C)3/2.(2)|S_C(m)|+|S_C(m+E)|\le103.6\,C^{4/7}K^{1/7}+80\,C^{3/4}(\log C)^{3/2}. \tag{2}

The main term. For C<c≤2CC<c\le2C,

m+Ec−mc=Ec (m+E+m)≥E2C (2+1)m,\sqrt{\frac{m+E}{c}}-\sqrt{\frac mc} =\frac{E}{\sqrt c\,(\sqrt{m+E}+\sqrt m)} \ge\frac{E}{\sqrt{2C}\,(\sqrt2+1)\sqrt m},

using c≤2Cc\le2C and E≤mE\le m. There are exactly CC such cc, so the sum in Lemma 2 (with x=mx=m, x′=m+Ex'=m+E) is at least

EC(2+2)m=E(2+2) Y≥K2/(256C)(2+2)K/4=K64(2+2) C>K219 C.(3)\frac{E\sqrt C}{(2+\sqrt2)\sqrt m}=\frac{E}{(2+\sqrt2)\,Y} \ge\frac{K^2/(256C)}{(2+\sqrt2)K/4}=\frac{K}{64(2+\sqrt2)\,C}>\frac{K}{219\,C}. \tag{3}

The error is smaller. Each term of (2) is below K/(438C)K/(438C).

  • First term. By (F1), C≤2−10K6/11C\le2^{-10}K^{6/11}, so C11/7≤2−110/7K6/7C^{11/7}\le2^{-110/7}K^{6/7}. Since 2110/7=5.37…×104>438⋅103.6=45376.82^{110/7}=5.37\ldots\times10^4>438\cdot103.6=45376.8, this gives 103.6 C4/7K1/7=103.6 C11/7K1/7/C<K/(438C)103.6\,C^{4/7}K^{1/7}=103.6\,C^{11/7}K^{1/7}/C<K/(438C).
  • Second term. By (F1), C7/4≤2−35/2K21/22C^{7/4}\le2^{-35/2}K^{21/22} and log⁡C<611log⁡K\log C<\frac6{11}\log K. So $80\cdot438,C^{7/4}(\log C)^{3/2}\le35040\cdot2^{-35/2} (6/11)^{3/2}K^{21/22}(\log K)^{3/2}<0.077,K^{21/22}(\log K)^{3/2}$. The function (log⁡K)3/2K−1/22(\log K)^{3/2}K^{-1/22} decreases for log⁡K>33\log K>33, and at K=2176K=2^{176} it equals 1347.6…/256<5.31347.6\ldots/256<5.3; so 0.077(log⁡K)3/2<0.41 K1/22<K1/220.077(\log K)^{3/2}<0.41\,K^{1/22}<K^{1/22}, and 80 C3/4(log⁡C)3/2<K/(438C)80\,C^{3/4}(\log C)^{3/2}<K/(438C).

By Lemma 2, (2) and (3), the number of pairs (c,q)(c,q) with C<c≤2CC<c\le2C, q≥1q\ge1 and m<cq2≤m+Em<cq^2\le m+E exceeds K/(219C)−2K/(438C)=0K/(219C)-2K/(438C)=0; being an integer, it is at least 11. Take such a pair, and apply Lemma 1 with D=cq2−m∈(0,E]D=cq^2-m\in(0,E]. It gives tt with t<E/ct<\sqrt{E/c} and N=c(q−t)(q+t)N=c(q-t)(q+t) with

0<N−m≤c(2t+1)<2cE+c≤22C⋅K2128C+2C=K4+2C≤K,0<N-m\le c(2t+1)<2\sqrt{cE}+c\le2\sqrt{2C\cdot\frac{K^2}{128C}}+2C =\frac K4+2C\le K,

since C≤2−10K6/11≤3K/8C\le2^{-10}K^{6/11}\le3K/8. The three factors are positive integers at most KK:

  • 1≤c≤2C≤K1\le c\le2C\le K;
  • q−t≥1q-t\ge1, because t<qt<q;
  • q≤(m+E)/c<Ym+E≤K/(22)<0.36Kq\le\sqrt{(m+E)/c}<Y_{m+E}\le K/(2\sqrt2)<0.36K and t<E/C≤K/(82 C)<0.09Kt<\sqrt{E/C}\le K/(8\sqrt2\,C)<0.09K, so q+t<Kq+t<K.

So N∈(m,m+K]N\in(m,m+K] is a product of three positive integers at most KK, and each prime factor of NN is at most KK. This proves the first sentence of the statement.

The bound for k(n)k(n). Let n≥1n\ge1 and K:=max⁡(64,⌈(215n)11/28⌉)K:=\max(64,\lceil(2^{15}n)^{11/28}\rceil), so that n≤2−15K28/11n\le2^{-15}K^{28/11}. If k(n)≥Kk(n)\ge K, some integers 0≤m≤n0\le m\le n and k≥Kk\ge K have each of m+1,…,m+km+1,\ldots,m+k divisible by a prime greater than kk; then each of m+1,…,m+Km+1,\ldots,m+K has a prime factor greater than KK, contradicting the first part. Hence k(n)≤K−1<max⁡(64,(215n)11/28)k(n)\le K-1<\max(64,(2^{15}n)^{11/28}). The argument uses only m≥0m\ge0, so it applies unchanged when mm is required to be positive. For the last sentence, let k≥64k\ge64: by the first part, for every integer 0≤n≤2−15k28/110\le n\le2^{-15}k^{28/11} the block n+1,…,n+kn+1,\ldots,n+k contains an integer with no prime factor greater than kk, so nk>2−15k28/11n_k>2^{-15}k^{28/11}. □\square

Where the exponent comes from

The proof uses the third-derivative test only through Lemma 3. With Y≍KY\asymp K fixed by the requirement q+t≤Kq+t\le K, the main term is of order K/CK/C and Lemma 3's leading error is of order C4/7K1/7C^{4/7}K^{1/7}; they balance at C≍K6/11C\asymp K^{6/11}, which gives m≍CK2≍K28/11m\asymp CK^2\asymp K^{28/11}. The second error term, of order C3/4(log⁡C)3/2C^{3/4}(\log C)^{3/2}, is smaller by a power of KK in that range, but only by K1/22K^{1/22} against (log⁡K)3/2(\log K)^{3/2}, so it needs KK large. The first error term forces a small constant such as 2−152^{-15}; with it the averaged range is nonempty only for K>2176K>2^{176}, where both error terms are under control, and below that L250 alone covers the statement.