The congruence conjecture for three Egyptian fractions of numerator five

What Lean checks

An ordered lexicographically first solution uniquely maximizes the last denominator for every admissible positive k outside residue class 1. The proof uses the elementary global comparison and a kernel-reduced exhaustive certificate below 100, with proved denominator bounds; no published comparison theorem is a hypothesis.

Theorem. Let kk be a positive integer for which

5k=1x+1y+1z,1≤x<y<z,\frac5k=\frac1x+\frac1y+\frac1z,\qquad 1\le x\lt y\lt z,

has a solution. If k≢1(mod5)k\not\equiv1\pmod5, the lexicographically first solution uniquely maximizes zz. This proves David Batista's conjecture that every term of A398581 is 11 modulo 55. Batista records the case k≡4(mod5)k\equiv4\pmod5 and gives a table through k=1500k=1500. OEIS A398581

Proof using the best two-term underapproximation theorem

For k>5k>5, write 5/k=p/q5/k=p/q in lowest terms. Set

m=⌊q/p⌋+1,h=pm−q,B=qm,y=⌊B/h⌋+1,d=hy−B.m=\lfloor q/p\rfloor+1,\qquad h=pm-q,\qquad B=qm,\qquad y=\lfloor B/h\rfloor+1,\qquad d=hy-B.

These are the first two strict greedy steps. The error after taking 1/m1/m is h/Bh/B, and the error after taking 1/m+1/y1/m+1/y is d/(By)d/(By).

kmod5k\bmod5 pp qq hh
0 1 k/5k/5 1
2 5 kk 3
3 5 kk 2
4 5 kk 1

The error is always a unit fraction in these four cases. For h=1h=1 we have d=1d=1. For h=2h=2 we have d=1d=1 or d=2∣Bd=2\mid B. For h=3h=3 the identity q=5m−3q=5m-3 gives B≡2m2(mod3)B\equiv2m^2\pmod3, so B≡0B\equiv0 or 2(mod3)2\pmod3. Thus d=3∣Bd=3\mid B or d=1d=1. In every case z=By/dz=By/d is integral.

The denominators are strictly increasing. The first greedy remainder is at most 1/[m(m−1)]1/[m(m-1)], which gives y>my>m since m≥2m\ge2. The second remainder is at most 1/[y(y−1)]1/[y(y-1)], so z≥y(y−1)>yz\ge y(y-1)>y. The triple is lexicographically first because any admissible first denominator is at least mm and, with that choice, any admissible second denominator is at least yy.

Hùng Việt Chu's Theorem 1.3 says that the strict greedy pair gives the unique largest two-unit-fraction sum strictly below p/qp/q whenever the least positive hh with p∣q+hp\mid q+h is at most three, apart from a possible tie at 10/1710/17. Our hh is exactly that integer. The table covers every required residue, and 5/k=10/175/k=10/17 is impossible for integral kk. Every competing triple has remaining fraction 1/z′1/z'. Its first two terms sum to at most 1/m+1/y1/m+1/y, with strict inequality for a different ordered pair. Hence z′≤zz'\le z, with equality only for the same triple. Chu, Theorem 1.3

For k=1,2k=1,2, even 1+1/2+1/31+1/2+1/3 is too small. For k=3,4,5k=3,4,5 the optimal triples are respectively (1,2,6)(1,2,6), (1,5,20)(1,5,20) and (2,3,6)(2,3,6). The first denominator is forced in each case. For k=4k=4, starting at two gives at most 1/2+1/3+1/4<5/41/2+1/3+1/4<5/4. Once the first denominator is fixed, zz decreases strictly as yy increases. This completes the proof.

Independent elementary comparison

The following argument replaces Chu's theorem for k≥100k\ge100. The finite cases below 100 can be enumerated with the accompanying verifier. Retain the preceding notation for the greedy triple and put z0=By/dz_0=By/d. Since d≤hd\le h, we have z0≥(qm)2/h2z_0\ge (qm)^2/h^2.

For a competing first denominator x>mx>m, write t=px−qt=px-q and b=qxb=qx. If the second denominator is vv and the third is ww, then e=tv−b≥1e=tv-b\ge1 and

w=b(b+e)te≤F(x),F(x)=qx(qx+1)px−q.w=\frac{b(b+e)}{t e}\le F(x),\qquad F(x)=\frac{qx(qx+1)}{px-q}.

Also x<3q/px<3q/p. The function FF is convex for x>q/px>q/p. Indeed, on substituting x=(t+q)/px=(t+q)/p, it is a linear function of tt plus a positive constant times 1/t1/t.

Here m,q≥20m,q\ge20. For m+2≤x<3q/pm+2\le x<3q/p it suffices by convexity to bound the two endpoints. At the first,

F(m+2)≤q2(m+3)2h+2p<(qm)2h2, F(m+2)\le\frac{q^2(m+3)^2}{h+2p}<\frac{(qm)^2}{h^2},

since (m+3)/m<6/5(m+3)/m<6/5 and the table gives (h+2p)/h2≥13/9>36/25(h+2p)/h^2\ge13/9>36/25. At the other endpoint x=3q/p<3mx=3q/p<3m, the numerator is at most 12(qm)212(qm)^2 and the denominator is 2q>12h22q>12h^2. The last inequality follows from q≥100q\ge100 when p=5p=5, and from q≥20q\ge20, h=1h=1 when p=1p=1.

For x=m+1x=m+1 and h≤2h\le2, the same estimate works with numerator at most q2(m+2)2q^2(m+2)^2 and (h+p)/h2≥7/4(h+p)/h^2\ge7/4. If h=3h=3, then p=5p=5, t=8t=8, and b=(5x−8)x≡0,4,5(mod8)b=(5x-8)x\equiv0,4,5\pmod8. Thus e=8v−b≥3e=8v-b\ge3, giving

w≤b(b+3)24≤q2(m+4)224<(qm)29. w\le\frac{b(b+3)}{24}\le\frac{q^2(m+4)^2}{24}<\frac{(qm)^2}{9}.

Here (m+4)/m≤6/5(m+4)/m\le6/5. This excludes every larger first denominator. At x=mx=m the third denominator decreases strictly as the second increases.

The finite enumeration ranges over k/5<x<3k/5k/5\lt x<3k/5 and kx/(5x−k)<y<2kx/(5x−k)kx/(5x-k)\lt y<2kx/(5x-k) with y>xy>x. It calculates z=kxy/((5x−k)y−kx)z=kxy/((5x-k)y-kx) and checks divisibility and z>yz>y. These bounds follow directly from x<y<zx\lt y\lt z, so the enumeration covers every solution. The verifier also compares against Batista's full published table through 1500. Batista's table

Prior work and scope

Chu proved the general best-two-term theorem in 2023–2024. The additional step here proves that his optimizing pair actually extends to an exact, strictly ordered three-term representation in all four required classes. A November 18, 2026 search by sequence number, author, and equivalent Egyptian-fraction formulations located no earlier proof specifically stating Batista's congruence. Historical priority remains unverified. The example 5/11=1/3+1/9+1/99=1/4+1/5+1/2205/11=1/3+1/9+1/99=1/4+1/5+1/220 in Batista's entry illustrates the behavior in the remaining class. Batista · Chu

Lean verification

The full theorem is formalized as Egyptian.batista_admissible_congruence in EgyptianTheorem.lean. It follows the independent elementary comparison and proves completeness of its kernel-checked finite enumeration. It has no pair-optimality or published theorem premise. See verification status and reproduction.