Introduction
For an integer q ≥ 2 q \ge2 q ≥ 2 put
θ q = log ( 1 + 1 q ) log ( 1 − 1 q ) . \theta_q = \log\left(1 + \frac{1}{q}\right)\log\left(1 - \frac{1}{q}\right). θ q = log ( 1 + q 1 ) log ( 1 − q 1 ) .
Hata [3 Cor. 6.3] showed that 1 1 1 , log ( 1 + 1 / q ) \log(1 + 1/q) log ( 1 + 1/ q ) , log ( 1 − 1 / q ) \log(1 - 1/q) log ( 1 − 1/ q ) , θ q \theta_q θ q are linearly independent over Q \mathbb{Q} Q for every integer q ≥ 54 q \ge54 q ≥ 54 , with an effective linear independence measure; in particular θ q ∉ Q \theta_q \notin\mathbb{Q} θ q ∈ / Q for every integer q ≥ 54 q \ge54 q ≥ 54 [3 Thm. 0.1] . Lysov [6 ] found an explicit formula for the common denominator of the type II Hermite–Padé approximants to ( log ( 1 + 1 / z ) , log ( 1 − 1 / z ) , log ( 1 + 1 / z ) log ( 1 − 1 / z ) ) (\log(1 + 1/z), \log(1 - 1/z), \log(1 + 1/z)\log(1 - 1/z)) ( log ( 1 + 1/ z ) , log ( 1 − 1/ z ) , log ( 1 + 1/ z ) log ( 1 − 1/ z )) and deduced that θ q \theta_q θ q is irrational for every q ≥ 33 q \ge33 q ≥ 33 [6 , 7 ] ([6 ] is a preprint; we found no journal version as of 19 November 2026). Calegari, Dimitrov and Tang [1 Thm. C, Thm. 14.0.1] proved, by a new arithmetic holonomy bound, that 1 1 1 , log ( 1 + 1 / m ) \log(1+1/m) log ( 1 + 1/ m ) , log ( 1 + 1 / n ) \log(1+1/n) log ( 1 + 1/ n ) and log ( 1 + 1 / m ) log ( 1 + 1 / n ) \log(1+1/m)\log(1+1/n) log ( 1 + 1/ m ) log ( 1 + 1/ n ) are linearly independent over Q \mathbb{Q} Q whenever m ≠ n m \ne n m = n are integers in Z ∖ { − 1 , 0 } \mathbb{Z}\setminus\{-1,0\} Z ∖ { − 1 , 0 } with ∣ m / n − 1 ∣ < 10 − 6 |m/n-1|<10^{-6} ∣ m / n − 1∣ < 1 0 − 6 . In [1 §15.1.10] they suggest forming “the generating function of the special linear forms obtained from evaluating a regular sequence of functional Hermite–Padé approximants” and reading it through their holonomy bounds. This paper carries out that scheme for Lysov’s forms, and pushes the constant 10 − 6 10^{-6} 1 0 − 6 of [1 Thm. C] .
Dependence on one preprint
Every result in this paper depends on the preprint [1 ] (arXiv:2408.15403v2, 16 September 2024; still the current version on 19 November 2026), which to our knowledge has not been refereed. Theorem 1 and Corollary 2 use only its general holonomy bound [1 Thm. 7.0.1] , which we apply as stated. Theorems 3 and 4 also use the specific construction of [1 §14] for products of two logarithms, together with [1 Thm. 6.0.2, Cor. 9.0.19, Lemma 12.0.1] . That construction contains a step that is asserted rather than proved in [1 ] , the irreducibility of a certain differential operator L a , b \mathcal{L}_{a,b} L a , b , and we inherit it; see § 5 . Each theorem below names exactly what it assumes.
Results
Theorem 1. Assume [1 Thm. 7.0.1] . For every integer q q q with 26 ≤ q ≤ 53 26 \le q \le53 26 ≤ q ≤ 53 , the numbers
1 , log ( 1 + 1 q ) , log ( 1 − 1 q ) , log ( 1 + 1 q ) log ( 1 − 1 q ) 1,\qquad\log\left(1+\frac{1}{q}\right),\qquad\log\left(1-\frac{1}{q}\right),\qquad\log\left(1+\frac{1}{q}\right)\log\left(1-\frac{1}{q}\right) 1 , log ( 1 + q 1 ) , log ( 1 − q 1 ) , log ( 1 + q 1 ) log ( 1 − q 1 ) are linearly independent over Q \mathbb{Q} Q .
With Hata’s theorem [3 ] for q ≥ 54 q \ge54 q ≥ 54 , the same holds for every integer q ≥ 26 q \ge26 q ≥ 26 . Precisely, [3 Cor. 6.3] states: let q ≥ 54 q \ge54 q ≥ 54 be an integer; for every ε > 0 \varepsilon> 0 ε > 0 there is an effectively computable H 1 = H 1 ( ε , q ) H_1=H_1(\varepsilon,q) H 1 = H 1 ( ε , q ) such that
∣ n 0 + n 1 log ( 1 + 1 q ) + n 2 log ( 1 − 1 q ) + n 3 θ q + n 4 Λ ( 1 q ) ∣ ≥ H − σ ( q 2 ) / τ ( q 2 ) − ε \left|n_0+n_1\log\left(1+\frac{1}{q}\right)+n_2\log\left(1-\frac{1}{q}\right)+n_3\theta_q+n_4\Lambda\left(\frac{1}{q}\right)\right|\ge H^{-\sigma(q^2)/\tau(q^2)-\varepsilon} n 0 + n 1 log ( 1 + q 1 ) + n 2 log ( 1 − q 1 ) + n 3 θ q + n 4 Λ ( q 1 ) ≥ H − σ ( q 2 ) / τ ( q 2 ) − ε for all ( n 0 , … , n 4 ) ∈ Z 5 (n_0,\ldots,n_4)\in\mathbb{Z}^5 ( n 0 , … , n 4 ) ∈ Z 5 with H = max 1 ≤ m ≤ 4 ∣ n m ∣ ≥ H 1 H=\max_{1\le m\le4}|n_m|\ge H_1 H = max 1 ≤ m ≤ 4 ∣ n m ∣ ≥ H 1 . Here Λ ( x ) = Li 2 ( ( 1 + x ) / 2 ) − Li 2 ( ( 1 − x ) / 2 ) − log 2 ⋅ log ( ( 1 + x ) / ( 1 − x ) ) \Lambda(x)=\operatorname{Li}_2((1+x)/2)-\operatorname{Li}_2((1-x)/2)-\log2\cdot\log((1+x)/(1-x)) Λ ( x ) = Li 2 (( 1 + x ) /2 ) − Li 2 (( 1 − x ) /2 ) − log 2 ⋅ log (( 1 + x ) / ( 1 − x )) [3 (3.5)] , and σ , τ \sigma,\tau σ , τ are explicit [3 (6.3), (6.4)] ; the bound q ≥ 54 q\ge54 q ≥ 54 comes from Hata’s τ ( k ) > 0 \tau(k)>0 τ ( k ) > 0 for every integer k ≥ 2909 k\ge2909 k ≥ 2909 (and 53 2 < 2909 ≤ 54 2 53^2<2909\le54^2 5 3 2 < 2909 ≤ 5 4 2 ), and σ ( k ) / τ ( k ) → 4 \sigma(k)/\tau(k)\to4 σ ( k ) / τ ( k ) → 4 as k → ∞ k\to\infty k → ∞ . Taking n 4 = 0 n_4=0 n 4 = 0 gives the four-term independence; in fact 1 1 1 , log ( 1 + 1 / q ) \log(1+1/q) log ( 1 + 1/ q ) , log ( 1 − 1 / q ) \log(1-1/q) log ( 1 − 1/ q ) , θ q \theta_q θ q , Λ ( 1 / q ) \Lambda(1/q) Λ ( 1/ q ) are linearly independent over Q \mathbb{Q} Q . Hata states this for integers q q q only: Corollary 6.3 is the case k = q 2 k=q^2 k = q 2 of his Theorem 6.2, the analogous statement at 1 / k 1/\sqrt{k} 1/ k for every integer k ≥ 2909 k\ge2909 k ≥ 2909 .
Since log ( 1 − 1 / q ) = − log ( 1 + 1 / ( q − 1 ) ) \log(1-1/q)=-\log(1+1/(q-1)) log ( 1 − 1/ q ) = − log ( 1 + 1/ ( q − 1 )) , Theorem 1 at q = N + 1 q=N+1 q = N + 1 contains the irrationality of the product of two consecutive logarithms log ( 1 + 1 / N ) log ( 1 + 1 / ( N + 1 ) ) = − θ N + 1 \log(1+1/N)\log(1+1/(N+1))=-\theta_{N+1} log ( 1 + 1/ N ) log ( 1 + 1/ ( N + 1 )) = − θ N + 1 :
Corollary 2. Assume [1 Thm. 7.0.1] . For every integer N N N with 25 ≤ N ≤ 31 25 \le N \le31 25 ≤ N ≤ 31 ,
log ( 1 + 1 N ) log ( 1 + 1 N + 1 ) ∉ Q . \log\left(1+\frac{1}{N}\right)\log\left(1+\frac{1}{N+1}\right) \notin\mathbb{Q}. log ( 1 + N 1 ) log ( 1 + N + 1 1 ) ∈ / Q . These are seven new irrational numbers. By Theorem 1 for 32 ≤ N ≤ 52 32 \le N \le52 32 ≤ N ≤ 52 and Hata’s theorem for N ≥ 53 N \ge53 N ≥ 53 (and, independently, Lysov’s theorem [6 ] for N ≥ 32 N \ge32 N ≥ 32 ), the product is irrational for every N ≥ 25 N \ge25 N ≥ 25 .
The second group of results concerns CDT’s two-parameter family. For m ∈ Z ∖ { − 1 , 0 } m \in\mathbb{Z} \setminus\{-1,0\} m ∈ Z ∖ { − 1 , 0 } write ℓ m = log ( 1 + 1 / m ) \ell_m=\log(1+1/m) ℓ m = log ( 1 + 1/ m ) .
Theorem 3. Assume the following parts of [1 ] as published: Theorem 6.0.2, Corollary 9.0.19, Lemma 12.0.1 and §14, including the irreducibility of the operator L a , b \mathcal{L}_{a,b} L a , b and the reduction in the proof of Lemma 14.3.1. Let m ≠ n m \ne n m = n be integers in Z ∖ { − 1 , 0 } \mathbb{Z} \setminus\{-1,0\} Z ∖ { − 1 , 0 } with ∣ m / n − 1 ∣ < 3 ⋅ 10 − 6 \lvert m/n-1\rvert< 3 \cdot10^{-6} ∣ m / n − 1 ∣ < 3 ⋅ 1 0 − 6 . Then 1 1 1 , ℓ m \ell_m ℓ m , ℓ n \ell_n ℓ n and ℓ m ℓ n \ell_m\ell_n ℓ m ℓ n are linearly independent over Q \mathbb{Q} Q ; in particular ℓ m ℓ n \ell_m\ell_n ℓ m ℓ n is irrational.
Theorem 4. Make the assumptions of Theorem 3. Call a pair ( m , n ) (m,n) ( m , n ) of distinct integers in Z ∖ { − 1 , 0 } \mathbb{Z} \setminus\{-1,0\} Z ∖ { − 1 , 0 } exceptional if 1 1 1 , ℓ m \ell_m ℓ m , ℓ n \ell_n ℓ n , ℓ m ℓ n \ell_m\ell_n ℓ m ℓ n are linearly dependent over Q \mathbb{Q} Q , and take classes of pairs modulo ( m , n ) ∼ ( n , m ) ∼ ( − m − 1 , − n − 1 ) (m,n) \sim(n,m) \sim(-m-1,-n-1) ( m , n ) ∼ ( n , m ) ∼ ( − m − 1 , − n − 1 ) . Then:
at most one exceptional class has a representative with ∣ m / n − 1 ∣ < 5 ⋅ 10 − 5 \lvert m/n-1\rvert< 5 \cdot10^{-5} ∣ m / n − 1 ∣ < 5 ⋅ 1 0 − 5 ;
at most two exceptional classes have a representative with ∣ m / n − 1 ∣ < 10 − 4 \lvert m/n-1\rvert< 10^{-4} ∣ m / n − 1 ∣ < 1 0 − 4 ;
every exceptional class has ∣ m / n − 1 ∣ ≥ 3 ⋅ 10 − 6 \lvert m/n-1\rvert\ge3 \cdot10^{-6} ∣ m / n − 1 ∣ ≥ 3 ⋅ 1 0 − 6 in all its representatives (Theorem 3 );
in the ranges of (i) and (ii), every exceptional class has ∣ m − n ∣ ≥ 2 \lvert m-n\rvert\ge2 ∣ m − n ∣ ≥ 2 , by Hata’s theorem [3 Cor. 6.3] .
The relation ( m , n ) ∼ ( − m − 1 , − n − 1 ) (m,n) \sim(-m-1,-n-1) ( m , n ) ∼ ( − m − 1 , − n − 1 ) comes from ℓ − k = − ℓ k − 1 \ell_{-k}=-\ell_{k-1} ℓ − k = − ℓ k − 1 : the two pairs give the same four numbers up to sign. For 1 ≤ N < M 1 \le N < M 1 ≤ N < M with d = M − N d=M-N d = M − N , the most permissive of the four representatives of { N , M } \{N,M\} { N , M } is ( m , n ) = ( − N − 1 , − M − 1 ) (m,n)=(-N-1,-M-1) ( m , n ) = ( − N − 1 , − M − 1 ) , with ∣ m / n − 1 ∣ = d / ( M + 1 ) \lvert m/n-1\rvert=d/(M+1) ∣ m / n − 1 ∣ = d / ( M + 1 ) , so the pair is covered by Theorem 3 if and only if d < 3 ⋅ 10 − 6 ( M + 1 ) d<3\cdot10^{-6}(M+1) d < 3 ⋅ 1 0 − 6 ( M + 1 ) , and by [1 Thm. C] if and only if d < 10 − 6 ( M + 1 ) d<10^{-6}(M+1) d < 1 0 − 6 ( M + 1 ) . Consecutive pairs (d = 1 d=1 d = 1 ) in either range have q = N + 1 > 3 ⋅ 10 5 q=N+1>3\cdot10^5 q = N + 1 > 3 ⋅ 1 0 5 and are covered by Hata’s theorem. So the new irrational numbers of Theorem 3 are
ℓ N ℓ M , d = M − N ≥ 2 , 10 − 6 ( M + 1 ) ≤ d < 3 ⋅ 10 − 6 ( M + 1 ) , (1) \ell_N\ell_M,\qquad d=M-N\ge2,\qquad10^{-6}(M+1)\le d<3\cdot10^{-6}(M+1), \tag*{(1)} ℓ N ℓ M , d = M − N ≥ 2 , 1 0 − 6 ( M + 1 ) ≤ d < 3 ⋅ 1 0 − 6 ( M + 1 ) , (1)
that is, d ⋅ 10 6 / 3 < M + 1 ≤ d ⋅ 10 6 d\cdot10^6/3<M+1\le d\cdot10^6 d ⋅ 1 0 6 /3 < M + 1 ≤ d ⋅ 1 0 6 : about 2 3 d ⋅ 10 6 \frac{2}{3}d\cdot10^6 3 2 d ⋅ 1 0 6 numbers for each d ≥ 2 d\ge2 d ≥ 2 , infinitely many in all. For example: for d = 2 d=2 d = 2 , 666,664 ≤ N ≤ 1,999,997 666{,}664\le N\le1{,}999{,}997 666 , 664 ≤ N ≤ 1 , 999 , 997 ; for d = 3 d=3 d = 3 , 999,997 ≤ N ≤ 2,999,996 999{,}997\le N\le2{,}999{,}996 999 , 997 ≤ N ≤ 2 , 999 , 996 ; for d = 10 d=10 d = 10 , 3,333,323 ≤ N ≤ 9,999,989 3{,}333{,}323\le N\le9{,}999{,}989 3 , 333 , 323 ≤ N ≤ 9 , 999 , 989 (exact integer computation over all four representatives, code/thm3/f7_newpairs.py). None of them is a known number in disguise: ( N + 1 ) / N (N+1)/N ( N + 1 ) / N is not a perfect power of a rational number, so ℓ N \ell_N ℓ N and ℓ M \ell_M ℓ M are multiplicatively independent and ℓ N ℓ M \ell_N\ell_M ℓ N ℓ M is not of the form c log 2 r c\log^2 r c log 2 r . Theorem 4 names no new number; it is a statement about the family. For example, all but at most one of the numbers log ( 1 + 1 / N ) log ( 1 + 1 / ( N + 2 ) ) \log(1+1/N)\log(1+1/(N+2)) log ( 1 + 1/ N ) log ( 1 + 1/ ( N + 2 )) with 39,998 ≤ N ≤ 666,663 39{,}998 \le N \le666{,}663 39 , 998 ≤ N ≤ 666 , 663 are irrational, and all but at most two with 19,998 ≤ N ≤ 666,663 19{,}998 \le N \le666{,}663 19 , 998 ≤ N ≤ 666 , 663 .
Index conventions
Four parametrisations are in use in the literature and in our sources. We use the following dictionary throughout.
Table 2.
Here ℓ m = log a + 1 a − 1 \ell_m=\log\frac{a+1}{a-1} ℓ m = log a − 1 a + 1 when a = 2 m + 1 a=2m+1 a = 2 m + 1 . In Sections 3–4 the letter n n n is the index of Lysov’s forms ( P j , n , R j , n ) (P_{j,n},R_{j,n}) ( P j , n , R j , n ) ; in Sections 5–6 ( m , n ) (m,n) ( m , n ) is a CDT pair, as in the statements above. Lysov’s paper writes m m m for our q q q , and Zeilberger and Zudilin [10 ] write a = q − 1 a=q-1 a = q − 1 .
What is new
New irrational numbers. The seven products ℓ N ℓ N + 1 \ell_N\ell_{N+1} ℓ N ℓ N + 1 , 25 ≤ N ≤ 31 25 \le N \le31 25 ≤ N ≤ 31 (Corollary 2 ); no method gave their irrationality before. The products (1.1) (Theorem 3 , conditional on [1 §14] ), together with the four-term independence for each of them.
New linear independence at known irrational numbers. The four-term independence of Theorem 1 for 26 ≤ q ≤ 32 26 \le q \le32 26 ≤ q ≤ 32 is new. For 33 ≤ q ≤ 53 33 \le q \le53 33 ≤ q ≤ 53 we know of no explicit statement; ours is the first, and its proof does not use Lysov’s arithmetic. It may follow from Lysov’s construction by the classical type II criterion, provided his refined denominator estimate covers the numerators P 1 , n P_{1,n} P 1 , n , P 2 , n P_{2,n} P 2 , n as well (we have not checked this). Zeilberger and Zudilin [10 p. 8] state that the linear independence of 1 , log ( a / ( a + 1 ) ) , log ( ( a + 1 ) / ( a + 2 ) ) 1,\log(a/(a+1)),\log((a+1)/(a+2)) 1 , log ( a / ( a + 1 )) , log (( a + 1 ) / ( a + 2 )) , known for a ≥ 53 a \ge53 a ≥ 53 by Hata, was “recently improved, though somewhat implicitly, by Volodya Lysov to a ≥ 32 a \ge32 a ≥ 32 ”; that statement concerns the linear forms, not the product.
Nothing new for consecutive pairs in CDT’s range or ours: those are covered by Hata’s theorem.
A family statement (Theorem 4 ), which names no number.
By-products: for every q ≥ 2 q \ge2 q ≥ 2 the characteristic polynomial of Lysov’s recurrence is an explicit quartic (Proposition 3.4 ); all local monodromies of the order-34 operator at its finite nonzero singular points are reflections (Proposition 4.3 ); an explicit form of the factor Φ 6 \Phi_6 Φ 6 of [1 Lemma 14.3.1] and its 2-adic nonvanishing at odd pairs (Section 5.5); a misprinted factor in [1 (14.2.2), (14.2.3)] .
A literature search on 19 November 2026 (arXiv, Hexagon, GitHub, Zenodo, general web search) found no other work on products of two real logarithms of this kind, on the constant 10 − 6 10^{-6} 1 0 − 6 of [1 Thm. C] , or on the Hata–Lysov family since [6 ] . The search is not exhaustive; in particular we could not find an abstract of a 2025 talk by Lysov on this subject.
Status of the results
The proofs are computer-assisted. Every computation that a proof uses is either an exact identity over Q \mathbb{Q} Q (checked with SymPy or python-flint) or a rigorous ball-arithmetic enclosure (Arb, through python-flint). The chain has been checked by AI referees at each decision gate and once as a whole (§ 9 ); it has not yet been read by a human expert, and it is not formalised. In the notation of our working notes (“gate-refereed”: checked by one adversarial AI referee when the result was accepted; “chain-refereed”: covered by the final whole-chain AI referee; “owed”: not yet done), the status labels are:
Theorem 1 and Corollary 2: proved, computer-assisted (exact symbolic identities and Arb certificates), given [1 Thm. 7.0.1] (preprint); gate-refereed once; chain-refereed; human read and Lean owed.
Theorem 3: proof sketch plus certified computation; conditional on CDT as published (Thm. 6.0.2, Cor. 9.0.19, Lemma 12.0.1 and §14, in particular the irreducibility of L a , b \mathcal{L}_{a,b} L a , b and the reduction in the proof of Lemma 14.3.1); Φ 6 \Phi_6 Φ 6 factor of Lemma 14.3.1 via Section 5.5 (proved, exact computer algebra; not gate-refereed); gate-refereed once; chain-refereed; not in Lean, no human read.
Theorem 4: proof sketch plus certified computation; conditional on CDT as published (Thm. 6.0.2, Cor. 9.0.19, Lemma 12.0.1 and §14, in particular the irreducibility of L a , b \mathcal{L}_{a,b} L a , b and the reduction in the proof of Lemma 14.3.1, reused pair by pair); Φ 6 \Phi_6 Φ 6 factor of Lemma 14.3.1 via Section 5.5 (proved, exact computer algebra; not gate-refereed); part (iv) adds Hata’s theorem [3 Cor. 6.3] (checked against the paper); Lemma 6.1 proved, given [1 Lemma 14.2.4] ; Lemma 5.1 (ε ≤ 10 − 4 \varepsilon\le10^{-4} ε ≤ 1 0 − 4 ) and the ray lemma proved; Lemma 6.2 proof sketch (classical); certificates exact (Arb and exact rationals); gate-refereed once; chain-refereed; not in Lean; no human read.
Method
For Theorem 1 we start from Lysov’s forms R j , n ( q ) = Q n ( q ) f j ( q ) − P j , n ( q ) R_{j,n}(q)=Q_n(q)f_j(q)-P_{j,n}(q) R j , n ( q ) = Q n ( q ) f j ( q ) − P j , n ( q ) , j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 , with ( f 1 , f 2 , f 3 ) = ( log ( 1 + 1 / q ) , log ( 1 − 1 / q ) , ϑ q ) (f_1,f_2,f_3)=(\log(1+1/q),\log(1-1/q),\vartheta_q) ( f 1 , f 2 , f 3 ) = ( log ( 1 + 1/ q ) , log ( 1 − 1/ q ) , ϑ q ) , and their generating functions H j ( y ) = ∑ n R j , n ( q ) y n H_j(y)=\sum_n R_{j,n}(q)y^n H j ( y ) = ∑ n R j , n ( q ) y n . A rational relation r 0 + a 1 f 1 + a 2 f 2 + a 3 f 3 = 0 r_0+a_1f_1+a_2f_2+a_3f_3=0 r 0 + a 1 f 1 + a 2 f 2 + a 3 f 3 = 0 makes − ∑ j a j H j -\sum_j a_jH_j − ∑ j a j H j a power series with rational coefficients of controlled denominators, which continues analytically to C ∖ [ y 2 , ∞ ) \mathbb{C}\setminus[y_2,\infty) C ∖ [ y 2 , ∞ ) for an explicit y 2 y_2 y 2 . The holonomy bound [1 Thm. 7.0.1] , applied with the Koebe map onto C ∖ [ y 2 , ∞ ) \mathbb{C}\setminus[y_2,\infty) C ∖ [ y 2 , ∞ ) and the 14 functions 1 , H , ϑ H , … , ϑ 12 H 1,H,\vartheta H,\ldots,\vartheta^{12}H 1 , H , ϑ H , … , ϑ 12 H (ϑ = y d / d y \vartheta=y\,d/dy ϑ = y d / d y ), gives a contradiction once these 14 functions are linearly independent over C ( y ) \mathbb{C}(y) C ( y ) and q ≥ 26 q\ge26 q ≥ 26 . Their independence follows from the irreducibility of the monodromy of the minimal operator L 13 L_{13} L 13 of the H j H_j H j (order 13), which we certify per q q q from rigorous numerical analytic continuation, using that the local monodromies at the singular points y 2 , y 3 y_2,y_3 y 2 , y 3 are reflections. No arithmetic of the forms beyond the elementary denominator bound lcm ( 1 , … , 4 n ) lcm ( 1 , … , 2 n ) \operatorname{lcm}(1,\ldots,4n)\operatorname{lcm}(1,\ldots,2n) lcm ( 1 , … , 4 n ) lcm ( 1 , … , 2 n ) is needed.
For Theorem 3 we keep CDT’s construction and change only the map: we construct, by optimisation, a polynomial-exponential map ψ \psi ψ of the disc for which h ∘ ψ h\circ\psi h ∘ ψ (h h h the hauptmodul of Γ 0 ( 2 ) \Gamma_0(2) Γ 0 ( 2 ) ) is admissible for every pair with ∣ m / n − 1 ∣ < 3 ⋅ 10 − 6 \lvert m/n-1\rvert<3\cdot10^{-6} ∣ m / n − 1 ∣ < 3 ⋅ 1 0 − 6 at once, and certify the holonomy inequality in its partition form [1 Thm. 6.0.2] by interval arithmetic. For Theorem 4 we put the functions of K = 2 K=2 K = 2 or 3 3 3 hypothetical exceptional pairs into one holonomy bound, after proving that they are jointly independent.
Outline
Section 2 recalls the holonomy bounds. Section 3 treats Lysov’s forms: contact orders, denominators, recurrences and the differential equations. Section 4 proves Theorem 1 and Corollary 2. Sections 5 and 6 prove Theorems 3 and 4. Section 7 lists the computer-assisted steps and how to reproduce them, Section 8 the limits of the method, and Section 9 provenance and the use of AI tools.
The holonomy bounds
We use the notation of [1 §6, §7] . For N ≥ 1 N\ge1 N ≥ 1 write [ 1 , … , N ] = lcm ( 1 , 2 , … , N ) [1,\ldots,N]=\operatorname{lcm}(1,2,\ldots,N) [ 1 , … , N ] = lcm ( 1 , 2 , … , N ) . Let m ≥ 1 m\ge1 m ≥ 1 , let b = ( b i , j ) 1 ≤ i ≤ m , 1 ≤ j ≤ r b=(b_{i,j})_{1\le i\le m,1\le j\le r} b = ( b i , j ) 1 ≤ i ≤ m , 1 ≤ j ≤ r be an array of nonnegative reals whose columns have the form 0 = b 1 , j = ⋯ = b u j , j < b u j + 1 , j = ⋯ = b m , j 0=b_{1,j}=\cdots=b_{u_j,j}<b_{u_j+1,j}=\cdots=b_{m,j} 0 = b 1 , j = ⋯ = b u j , j < b u j + 1 , j = ⋯ = b m , j , and let e ∈ N m e\in\mathbb{N}^m e ∈ N m . A formal power series f i ∈ Q [ [ y ] ] f_i\in\mathbb{Q}[[y]] f i ∈ Q [[ y ]] has denominator type ( b i , ⋅ , e i ) (b_{i,\cdot},e_i) ( b i , ⋅ , e i ) if
f i ( y ) = a i , 0 + ∑ n ≥ 1 a i , n y n n e i [ 1 , … , b i , 1 n ] ⋯ [ 1 , … , b i , r n ] , a i , n ∈ Z f_i(y)=a_{i,0}+\sum_{n\ge1}\frac{a_{i,n}y^n}{n^{e_i}[1,\ldots,b_{i,1}n]\cdots[1,\ldots,b_{i,r}n]},\qquad a_{i,n}\in\mathbb{Z} f i ( y ) = a i , 0 + n ≥ 1 ∑ n e i [ 1 , … , b i , 1 n ] ⋯ [ 1 , … , b i , r n ] a i , n y n , a i , n ∈ Z
Let σ i = ∑ j b i , j \sigma_i=\sum_j b_{i,j} σ i = ∑ j b i , j be the row sums (nondecreasing in i i i , by the shape of the columns), and put
τ ( b ) = 1 m 2 ∑ i = 1 m ( 2 i − 1 ) σ i , τ ( b ; e ) = τ ( b ) + τ # ( e ) , \tau(b)=\frac{1}{m^2}\sum_{i=1}^{m}(2i-1)\sigma_i,\qquad\tau(b;e)=\tau(b)+\tau^\#(e), τ ( b ) = m 2 1 i = 1 ∑ m ( 2 i − 1 ) σ i , τ ( b ; e ) = τ ( b ) + τ # ( e ) ,
with τ # \tau^\# τ # as in [1 Thm. 6.0.2, Def. 6.0.1] ; τ # ( 0 ) = 0 \tau^\#(0)=0 τ # ( 0 ) = 0 , and τ # ( e ) \tau^\#(e) τ # ( e ) depends only on m m m , ∑ i e i \sum_i e_i ∑ i e i and max i e i \max_i e_i max i e i .
For a holomorphic map φ \varphi φ from a neighbourhood of the closed unit disc D ‾ \overline{\mathbb{D}} D to C \mathbb{C} C with φ ( 0 ) = 0 \varphi(0)=0 φ ( 0 ) = 0 put
T ( φ ) = ∬ T 2 log ∣ φ ( z ) − φ ( w ) ∣ d μ H a a r ( z ) d μ H a a r ( w ) . T(\varphi)=\iint_{\mathbb{T}^2}\log\lvert\varphi(z)-\varphi(w)\rvert\,d\mu_{\mathrm{Haar}}(z)\,d\mu_{\mathrm{Haar}}(w). T ( φ ) = ∬ T 2 log ∣ φ ( z ) − φ ( w )∣ d μ Haar ( z ) d μ Haar ( w ) .
Theorem 2.1 ([1 Thm. 7.0.1] ). Let f 1 , … , f m ∈ Q [ [ y ] ] f_1,\ldots,f_m\in\mathbb{Q}[[y]] f 1 , … , f m ∈ Q [[ y ]] be linearly independent over Q ( y ) \mathbb{Q}(y) Q ( y ) , of denominator types given by ( b , e ) (b,e) ( b , e ) , and let φ : ( D , 0 ) → ( C , 0 ) \varphi:(\mathbb{D},0)\to(\mathbb{C},0) φ : ( D , 0 ) → ( C , 0 ) be holomorphic with log ∣ φ ′ ( 0 ) ∣ > max { σ m , τ ( b ; e ) } \log\lvert\varphi'(0)\rvert>\max\{\sigma_m,\tau(b;e)\} log ∣ φ ′ ( 0 )∣ > max { σ m , τ ( b ; e )} , such that every f i ( φ ( z ) ) f_i(\varphi(z)) f i ( φ ( z )) is the germ of a meromorphic function on ∣ z ∣ < 1 |z|<1 ∣ z ∣ < 1 . Then
m ≤ T ( φ ) log ∣ φ ′ ( 0 ) ∣ − τ ( b ; e ) . m \le\frac{T(\varphi)}{\log|\varphi'(0)|-\tau(b;e)}. m ≤ log ∣ φ ′ ( 0 ) ∣ − τ ( b ; e ) T ( φ ) .
If all f i f_i f i are known to be holonomic, the hypothesis log ∣ φ ′ ( 0 ) ∣ > max { σ m , τ } \log|\varphi'(0)|>\max\{\sigma_m,\tau\} log ∣ φ ′ ( 0 ) ∣ > max { σ m , τ } may be relaxed to log ∣ φ ′ ( 0 ) ∣ > τ \log|\varphi'(0)|>\tau log ∣ φ ′ ( 0 ) ∣ > τ .
(In [1 ] the map is defined on D \mathbb{D} D and T T T is defined by a limit; we only use maps holomorphic on a neighbourhood of D \mathbb{D} D .)
Theorem 2.2 ([1 Thm. 6.0.2] , partition form). Let f 1 , … , f m f_1,\ldots,f_m f 1 , … , f m and ( b , e ) (b,e) ( b , e ) be as in Theorem 2.1, all f i f_i f i holonomic. Let φ 0 , … , φ l \varphi_0,\ldots,\varphi_l φ 0 , … , φ l be holomorphic maps ( D , 0 ) → ( C , 0 ) (\mathbb{D},0)\to(\mathbb{C},0) ( D , 0 ) → ( C , 0 ) with ∣ φ 0 ′ ( 0 ) ∣ < ⋯ < ∣ φ l ′ ( 0 ) ∣ |\varphi_0'(0)|<\cdots<|\varphi_l'(0)| ∣ φ 0 ′ ( 0 ) ∣ < ⋯ < ∣ φ l ′ ( 0 ) ∣ and log ∣ φ l ′ ( 0 ) ∣ > τ ( b ; e ) \log|\varphi_l'(0)|>\tau(b;e) log ∣ φ l ′ ( 0 ) ∣ > τ ( b ; e ) , such that every f i ( φ k ( z ) ) f_i(\varphi_k(z)) f i ( φ k ( z )) is meromorphic on ∣ z ∣ < 1 |z|<1 ∣ z ∣ < 1 . Let 0 = ϑ 0 < ϑ 1 < ⋯ < ϑ l < ϑ l + 1 = m 0=\vartheta_0<\vartheta_1<\cdots<\vartheta_l<\vartheta_{l+1}=m 0 = ϑ 0 < ϑ 1 < ⋯ < ϑ l < ϑ l + 1 = m , and let g : [ 0 , 1 ) → R ∪ { − ∞ } g:[0,1)\to\mathbb{R}\cup\{-\infty\} g : [ 0 , 1 ) → R ∪ { − ∞ } be g ( t ) = log ∣ φ k ( e 2 π i ( m t − ϑ k ) / ( ϑ k + 1 − ϑ k ) ) ∣ g(t)=\log|\varphi_k(e^{2\pi i(mt-\vartheta_k)/(\vartheta_{k+1}-\vartheta_k)})| g ( t ) = log ∣ φ k ( e 2 π i ( m t − ϑ k ) / ( ϑ k + 1 − ϑ k ) ) ∣ for t ∈ [ ϑ k / m , ϑ k + 1 / m ) t\in[\vartheta_k/m,\vartheta_{k+1}/m) t ∈ [ ϑ k / m , ϑ k + 1 / m ) . Then
m ≤ ∫ 0 1 ∫ 0 1 max ( g ( s ) , g ( t ) ) d s d t + 1 m ∑ k = 1 l ϑ k 2 log ∣ φ k ′ ( 0 ) φ k − 1 ′ ( 0 ) ∣ log ∣ φ l ′ ( 0 ) ∣ − τ ( b ; e ) . m \le\frac{\int_0^1\int_0^1\max(g(s),g(t))\,ds\,dt+\frac{1}{m}\sum_{k=1}^{l}\vartheta_k^2\log\left|\frac{\varphi_k'(0)}{\varphi_{k-1}'(0)}\right|}{\log|\varphi_l'(0)|-\tau(b;e)}. m ≤ log ∣ φ l ′ ( 0 ) ∣ − τ ( b ; e ) ∫ 0 1 ∫ 0 1 max ( g ( s ) , g ( t )) d s d t + m 1 ∑ k = 1 l ϑ k 2 log φ k − 1 ′ ( 0 ) φ k ′ ( 0 ) .
With l = 0 l=0 l = 0 (the trivial partition ) the numerator is the rearrangement integral R ( φ ) = ∬ max ( g ( s ) , g ( t ) ) d s d t R(\varphi)=\iint\max(g(s),g(t))\,ds\,dt R ( φ ) = ∬ max ( g ( s ) , g ( t )) d s d t with g ( s ) = log ∣ φ ( e 2 π i s ) ∣ g(s)=\log|\varphi(e^{2\pi is})| g ( s ) = log ∣ φ ( e 2 π i s ) ∣ . The double integral depends only on the distribution of g g g , which for a partition is the mixture ∑ k ϑ k + 1 − ϑ k m ⋅ ( law of log ∣ φ k ∣ on T ) \sum_k\frac{\vartheta_{k+1}-\vartheta_k}{m}\cdot(\text{law of }\log|\varphi_k|\text{ on }\mathbb{T}) ∑ k m ϑ k + 1 − ϑ k ⋅ ( law of log ∣ φ k ∣ on T ) . We use Theorem 2.1 in Section 4 and Theorem 2.2 in Sections 5 and 6.
Lemma 2.3. If φ \varphi φ is univalent on a neighbourhood of D \mathbb{D} D and φ ( 0 ) = 0 \varphi(0)=0 φ ( 0 ) = 0 , then T ( φ ) = log ∣ φ ′ ( 0 ) ∣ T(\varphi)=\log|\varphi'(0)| T ( φ ) = log ∣ φ ′ ( 0 ) ∣ .
Proof. The function ( z , w ) ↦ log ∣ ( φ ( z ) − φ ( w ) ) / ( z − w ) ∣ (z,w)\mapsto\log|(\varphi(z)-\varphi(w))/(z-w)| ( z , w ) ↦ log ∣ ( φ ( z ) − φ ( w )) / ( z − w ) ∣ is pluriharmonic on a neighbourhood of D 2 \mathbb{D}^2 D 2 (the quotient is holomorphic and zero-free there, by univalence), so its mean over T 2 \mathbb{T}^2 T 2 is its value log ∣ φ ′ ( 0 ) ∣ \log|\varphi'(0)| log ∣ φ ′ ( 0 ) ∣ at ( 0 , 0 ) (0,0) ( 0 , 0 ) ; and ∬ T 2 log ∣ z − w ∣ = 0 \iint_{\mathbb{T}^2}\log|z-w|=0 ∬ T 2 log ∣ z − w ∣ = 0 . □
Lysov's forms
Definitions
For integers n ≥ 0 n\ge0 n ≥ 0 and l l l put
c n , l = ∑ k ( − 1 ) n + l ( n + l l ) ( n l − k ) ( 2 n + k 2 k ) ( n + k k ) ( 4 n + 2 k 2 n + k ) , c_{n,l}=\sum_k(-1)^{n+l}\binom{n+l}{l}\binom{n}{l-k}\binom{2n+k}{2k}\binom{n+k}{k}\binom{4n+2k}{2n+k}, c n , l = k ∑ ( − 1 ) n + l ( l n + l ) ( l − k n ) ( 2 k 2 n + k ) ( k n + k ) ( 2 n + k 4 n + 2 k ) ,
the sum over 0 ≤ k ≤ 2 n 0\le k\le2n 0 ≤ k ≤ 2 n with 0 ≤ l − k ≤ n 0\le l-k\le n 0 ≤ l − k ≤ n (so c n , l = 0 c_{n,l}=0 c n , l = 0 unless 0 ≤ l ≤ 3 n 0\le l\le3n 0 ≤ l ≤ 3 n ), and
Q n ( z ) = ∑ l = 0 3 n c n , l z 2 l ∈ Z [ z ] , Q_n(z)=\sum_{l=0}^{3n}c_{n,l}z^{2l}\in\mathbb{Z}[z], Q n ( z ) = l = 0 ∑ 3 n c n , l z 2 l ∈ Z [ z ] ,
an even polynomial of degree 6 n 6n 6 n (Lysov’s Q 6 n Q_{6n} Q 6 n , [6 formula (21) and Lemma 8] ; we use the formula as a definition). Let f 1 ( z ) = log ( 1 + 1 / z ) f_1(z)=\log(1+1/z) f 1 ( z ) = log ( 1 + 1/ z ) , f 2 ( z ) = log ( 1 − 1 / z ) f_2(z)=\log(1-1/z) f 2 ( z ) = log ( 1 − 1/ z ) and f 3 = f 1 f 2 f_3=f_1f_2 f 3 = f 1 f 2 , as Laurent series at z = ∞ z=\infty z = ∞ . Let P j , n P_{j,n} P j , n be the polynomial part of Q n f j Q_nf_j Q n f j , and R j , n = Q n f j − P j , n = O ( z − 1 ) R_{j,n}=Q_nf_j-P_{j,n}=O(z^{-1}) R j , n = Q n f j − P j , n = O ( z − 1 ) . Since Q n Q_n Q n is even and f 2 ( z ) = f 1 ( − z ) f_2(z)=f_1(-z) f 2 ( z ) = f 1 ( − z ) , we have P 2 , n ( z ) = P 1 , n ( − z ) P_{2,n}(z)=P_{1,n}(-z) P 2 , n ( z ) = P 1 , n ( − z ) . Explicitly, with A N = ∑ i = 1 N ( − 1 ) i + 1 / i A_N=\sum_{i=1}^{N}(-1)^{i+1}/i A N = ∑ i = 1 N ( − 1 ) i + 1 / i ,
P 1 , n ( z ) = ∑ l c n , l ∑ m = 1 2 l ( − 1 ) m + 1 m z 2 l − m , P 3 , n ( z ) = ∑ l c n , l ∑ m = 1 l ( − A 2 m − 1 m ) z 2 ( l − m ) , (2) \begin{aligned}
P_{1,n}(z)&=\sum_l c_{n,l}\sum_{m=1}^{2l}\frac{(-1)^{m+1}}{m}z^{2l-m},\\
P_{3,n}(z)&=\sum_l c_{n,l}\sum_{m=1}^{l}\left(-\frac{A_{2m-1}}{m}\right)z^{2(l-m)}, \tag*{(2)}
\end{aligned} P 1 , n ( z ) P 3 , n ( z ) = l ∑ c n , l m = 1 ∑ 2 l m ( − 1 ) m + 1 z 2 l − m , = l ∑ c n , l m = 1 ∑ l ( − m A 2 m − 1 ) z 2 ( l − m ) , (2)
because [ z − 2 m ] f 3 = − A 2 m − 1 / m [z^{-2m}]f_3=-A_{2m-1}/m [ z − 2 m ] f 3 = − A 2 m − 1 / m and the odd negative powers of z z z do not occur in f 3 f_3 f 3 .
The functions are Cauchy transforms: for ∣ z ∣ > 1 |z|>1 ∣ z ∣ > 1 ,
f 1 ( z ) = ∫ − 1 0 d x z − x , f 2 ( z ) = − ∫ 0 1 d x z − x , f 3 ( z ) = ∫ − 1 1 ω ( x ) d x z − x . f_1(z)=\int_{-1}^{0}\frac{dx}{z-x},\qquad
f_2(z)=-\int_{0}^{1}\frac{dx}{z-x},\qquad
f_3(z)=\int_{-1}^{1}\frac{\omega(x)\,dx}{z-x}. f 1 ( z ) = ∫ − 1 0 z − x d x , f 2 ( z ) = − ∫ 0 1 z − x d x , f 3 ( z ) = ∫ − 1 1 z − x ω ( x ) d x .
with ω \omega ω odd and ω ( x ) = − log ( 1 + 1 / x ) \omega(x)=-\log(1+1/x) ω ( x ) = − log ( 1 + 1/ x ) on ( 0 , 1 ) (0,1) ( 0 , 1 ) . Write μ j \mu_j μ j for the three (signed) measures; their total variations are 1 , 1 , 4 log 2 1,1,4\log2 1 , 1 , 4 log 2 . Then R j , n ( z ) = ∫ Q n ( x ) d μ j ( x ) / ( z − x ) R_{j,n}(z)=\int Q_n(x)\,d\mu_j(x)/(z-x) R j , n ( z ) = ∫ Q n ( x ) d μ j ( x ) / ( z − x ) .
Contact orders
Lemma 3.1 (Lysov). For all n ≥ 0 n\ge0 n ≥ 0 and j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 : ∫ x k Q n ( x ) d μ j ( x ) = 0 \int x^kQ_n(x)\,d\mu_j(x)=0 ∫ x k Q n ( x ) d μ j ( x ) = 0 for 0 ≤ k < 2 n 0\le k<2n 0 ≤ k < 2 n . Equivalently R j , n ( z ) = O ( z − 2 n − 1 ) R_{j,n}(z)=O(z^{-2n-1}) R j , n ( z ) = O ( z − 2 n − 1 ) as z → ∞ z\to\infty z → ∞ .
This is [6 Lemmas 7 and 8] . Since [6 ] is an unrefereed preprint we give the argument in full; it follows Lysov’s.
Proof. Put ∂ = 1 x d d x \partial=\frac{1}{x}\frac{d}{dx} ∂ = x 1 d x d and D = d d x ∘ 1 x \mathcal{D}=\frac{d}{dx}\circ\frac{1}{x} D = d x d ∘ x 1 , so that ∂ x m = m x m − 2 \partial x^m=mx^{m-2} ∂ x m = m x m − 2 , D x m = ( m − 1 ) x m − 2 \mathcal{D}x^m=(m-1)x^{m-2} D x m = ( m − 1 ) x m − 2 , both operators commute with x ↦ − x x\mapsto-x x ↦ − x , and for smooth f , g f,g f , g on ( 0 , 1 ] (0,1] ( 0 , 1 ]
∫ ε 1 f ∂ g = [ f g x ] ε 1 − ∫ ε 1 ( D f ) g , ∫ ε 1 f D g = [ f g x ] ε 1 − ∫ ε 1 ( ∂ f ) g . \int_{\varepsilon}^{1}f\,\partial g=\left[\frac{fg}{x}\right]_{\varepsilon}^{1}-\int_{\varepsilon}^{1}(\mathcal{D}f)g,\qquad
\int_{\varepsilon}^{1}f\,\mathcal{D}g=\left[\frac{fg}{x}\right]_{\varepsilon}^{1}-\int_{\varepsilon}^{1}(\partial f)g. ∫ ε 1 f ∂ g = [ x f g ] ε 1 − ∫ ε 1 ( D f ) g , ∫ ε 1 f D g = [ x f g ] ε 1 − ∫ ε 1 ( ∂ f ) g .
Rodrigues form. We have
Q n = κ n ∂ n [ ( x 2 − 1 ) n x 2 n U ~ ] , κ n = 1 ( 2 n ) ! ( n ! ) 2 , Q_n=\kappa_n\partial^n\left[(x^2-1)^nx^{2n}\widetilde U\right],\qquad
\kappa_n=\frac{1}{(2n)!(n!)^2}, Q n = κ n ∂ n [ ( x 2 − 1 ) n x 2 n U ] , κ n = ( 2 n )! ( n ! ) 2 1 ,
U ~ = ∂ n [ x 2 n U ] , U = D 2 n [ ( x 2 − 1 ) 2 n x 4 n ] . \widetilde U=\partial^n[x^{2n}U],\qquad
U=\mathcal{D}^{2n}[(x^2-1)^{2n}x^{4n}]. U = ∂ n [ x 2 n U ] , U = D 2 n [( x 2 − 1 ) 2 n x 4 n ] .
In the variable w = x 2 w=x^2 w = x 2 (∂ = 2 d / d w \partial=2\,d/dw ∂ = 2 d / d w , D 2 n = x ∂ 2 n x − 1 \mathcal{D}^{2n}=x\,\partial^{2n}x^{-1} D 2 n = x ∂ 2 n x − 1 , D = d / d w D=d/dw D = d / d w ) this is the pair of identities
Q n ( w ) = 1 n ! D n [ w n ( w − 1 ) n V n ( w ) ] , Q_n(w)=\frac{1}{n!}D^n\left[w^n(w-1)^nV_n(w)\right], Q n ( w ) = n ! 1 D n [ w n ( w − 1 ) n V n ( w ) ] ,
V n ( w ) = 16 n ( 2 n ) ! n ! D n [ w n + 1 / 2 D 2 n [ w 2 n − 1 / 2 ( w − 1 ) 2 n ] ] . V_n(w)=\frac{16^n}{(2n)!\,n!}D^n\left[w^{n+1/2}D^{2n}\left[w^{2n-1/2}(w-1)^{2n}\right]\right]. V n ( w ) = ( 2 n )! n ! 1 6 n D n [ w n + 1/2 D 2 n [ w 2 n − 1/2 ( w − 1 ) 2 n ] ] .
with V n ( w ) = ∑ k ( − 1 ) k B k w k V_n(w)=\sum_k(-1)^kB_kw^k V n ( w ) = ∑ k ( − 1 ) k B k w k , B k = ( 2 n + k 2 k ) ( n + k k ) ( 4 n + 2 k 2 n + k ) B_k=\binom{2n+k}{2k}\binom{n+k}{k}\binom{4n+2k}{2n+k} B k = ( 2 k 2 n + k ) ( k n + k ) ( 2 n + k 4 n + 2 k ) . For (i), 1 n ! D n [ w n + k ( w − 1 ) n ] = ∑ j ( n j ) ( − 1 ) n − j ( n + k + j n + k ) w k + j \frac{1}{n!}\mathcal{D}^n[w^{n+k}(w-1)^n]=\sum_j\binom{n}{j}(-1)^{n-j}\binom{n+k+j}{n+k}w^{k+j} n ! 1 D n [ w n + k ( w − 1 ) n ] = ∑ j ( j n ) ( − 1 ) n − j ( n + k n + k + j ) w k + j ; multiplying by ( − 1 ) k B k (-1)^kB_k ( − 1 ) k B k and collecting w l w^l w l gives ∑ k + j = l ( − 1 ) n + l ( n + l n + l − k ) ( n j ) B k = c n , l \sum_{k+j=l}(-1)^{n+l}\binom{n+l}{n+l-k}\binom{n}{j}B_k=c_{n,l} ∑ k + j = l ( − 1 ) n + l ( n + l − k n + l ) ( j n ) B k = c n , l . For (ii), the coefficient of w k w^k w k on the right is 16 n ( 2 n ) ! n ! ( 2 n k ) ( − 1 ) k Γ ( 2 n + k + 1 2 ) Γ ( k + 1 2 ) ( n + k ) ! k ! \frac{16^n}{(2n)!n!}\binom{2n}{k}(-1)^k\frac{\Gamma(2n+k+\frac{1}{2})}{\Gamma(k+\frac{1}{2})}\frac{(n+k)!}{k!} ( 2 n )! n ! 1 6 n ( k 2 n ) ( − 1 ) k Γ ( k + 2 1 ) Γ ( 2 n + k + 2 1 ) k ! ( n + k )! , which equals ( − 1 ) k B k (-1)^kB_k ( − 1 ) k B k by the Legendre duplication formula for 0 ≤ k ≤ 2 n 0\leq k\leq2n 0 ≤ k ≤ 2 n (the ratio simplifies to 1; checked symbolically in i1_contact.py), and both sides vanish for k > 2 n k>2n k > 2 n . The factors U U U , x 2 n U x^{2n}U x 2 n U , U ~ \widetilde U U are even polynomials, and x 2 n U x^{2n}U x 2 n U and ( x 2 − 1 ) n x 2 n U ~ (x^2-1)^nx^{2n}\widetilde U ( x 2 − 1 ) n x 2 n U vanish to order ≥ 2 n \geq2n ≥ 2 n at 0.
Moments. By the Cauchy-transform representations, [ z − s − 1 ] R j , n = ∫ x s Q n d μ j [z^{-s-1}]R_{j,n}=\int x^sQ_n\,d\mu_j [ z − s − 1 ] R j , n = ∫ x s Q n d μ j . (For f 3 f_3 f 3 this is the identity [ z − 2 m ] f 3 = − A 2 m − 1 / m = − 2 ∫ 0 1 x 2 m − 1 log ( 1 + 1 / x ) d x [z^{-2m}]f_3=-A_{2m-1}/m=-2\int_0^1x^{2m-1}\log(1+1/x)\,dx [ z − 2 m ] f 3 = − A 2 m − 1 / m = − 2 ∫ 0 1 x 2 m − 1 log ( 1 + 1/ x ) d x , which follows from ∫ 0 1 x N − 1 log ( 1 + x ) d x = A N / N \int_0^1x^{N-1}\log(1+x)\,dx=A_N/N ∫ 0 1 x N − 1 log ( 1 + x ) d x = A N / N for even N N N and ∫ 0 1 x N − 1 ( − log x ) d x = 1 / N 2 \int_0^1x^{N-1}(-\log x)\,dx=1/N^2 ∫ 0 1 x N − 1 ( − log x ) d x = 1/ N 2 ; odd negative powers vanish on both sides.) Since Q n Q_n Q n is even, the lemma is equivalent to
∫ 0 1 x s Q n d x = 0 ( 0 ≤ s ≤ 2 n − 1 ) , \int_0^1x^sQ_n\,dx=0\qquad(0\leq s\leq2n-1), ∫ 0 1 x s Q n d x = 0 ( 0 ≤ s ≤ 2 n − 1 ) ,
(O1)
∫ 0 1 x s Q n ℓ d x = 0 ( s odd , s ≤ 2 n − 1 ) , \int_0^1x^sQ_n\ell\,dx=0\qquad(s\text{ odd},\ s\leq2n-1), ∫ 0 1 x s Q n ℓ d x = 0 ( s odd , s ≤ 2 n − 1 ) ,
(O2)
where ℓ ( x ) = log ( 1 + 1 / x ) \ell(x)=\log(1+1/x) ℓ ( x ) = log ( 1 + 1/ x ) . Let h = x s h=x^s h = x s or h = x s ℓ h=x^s\ell h = x s ℓ accordingly.
Step 1. Integrating by parts n n n times against D \mathcal{D} D gives ∫ 0 1 Q n h = ± κ n ∫ 0 1 ( x 2 − 1 ) n x 2 n U ~ ⋅ r d x \int_0^1Q_nh=\pm\kappa_n\int_0^1(x^2-1)^nx^{2n}\widetilde U\cdot r\,dx ∫ 0 1 Q n h = ± κ n ∫ 0 1 ( x 2 − 1 ) n x 2 n U ⋅ r d x with r = D n h r=\mathcal{D}^nh r = D n h . The boundary terms vanish: D n − 1 − j [ ( x 2 − 1 ) n x 2 n U ~ ] \mathcal{D}^{n-1-j}[(x^2-1)^nx^{2n}\widetilde U] D n − 1 − j [( x 2 − 1 ) n x 2 n U ] is divisible by ( x 2 − 1 ) j + 1 (x^2-1)^{j+1} ( x 2 − 1 ) j + 1 and vanishes to order 2 j + 2 2j+2 2 j + 2 at 0, while D j h = O ( x − 2 j ∣ log x ∣ ) \mathcal{D}^jh=O(x^{-2j}|\log x|) D j h = O ( x − 2 j ∣ log x ∣ ) at 0.
Step 2. If h = x s h=x^s h = x s with s s s odd, then r = 0 r=0 r = 0 (D n \mathcal{D}^n D n kills odd polynomials of degree ≤ 2 n − 1 \leq2n-1 ≤ 2 n − 1 ). If h = x s h=x^s h = x s with s s s even, r = c x s − 2 n r=cx^{s-2n} r = c x s − 2 n . If h = x s ℓ h=x^s\ell h = x s ℓ , then by induction D j [ x s ℓ ] = D j [ x s ] ℓ + ( rational ) \mathcal{D}^j[x^s\ell]=\mathcal{D}^j[x^s]\ell+(\text{rational}) D j [ x s ℓ ] = D j [ x s ] ℓ + ( rational ) , since ℓ ′ = − 1 / ( x ( x + 1 ) ) \ell'=-1/(x(x+1)) ℓ ′ = − 1/ ( x ( x + 1 )) ; as D n x s = 0 \mathcal{D}^nx^s=0 D n x s = 0 , r r r is rational with poles only at 0 (order ≤ 2 n \leq2n ≤ 2 n ) and − 1 -1 − 1 (order ≤ n \leq n ≤ n ), and r = O ( x − 2 ) r=O(x^{-2}) r = O ( x − 2 ) at ∞ \infty ∞ . Its odd part is D n [ x s ℓ e v e n ] \mathcal{D}^n[x^s\ell_{\mathrm{even}}] D n [ x s ℓ even ] with ℓ e v e n = 1 2 log ( 1 − x − 2 ) = − ∑ j ≥ 1 x − 2 j / ( 2 j ) \ell_{\mathrm{even}}=\frac{1}{2}\log(1-x^{-2})=-\sum_{j\geq1}x^{-2j}/(2j) ℓ even = 2 1 log ( 1 − x − 2 ) = − ∑ j ≥ 1 x − 2 j / ( 2 j ) ; here D n x s − 2 j = 0 \mathcal{D}^nx^{s-2j}=0 D n x s − 2 j = 0 for 1 ≤ j ≤ ( s − 1 ) / 2 1\leq j\leq(s-1)/2 1 ≤ j ≤ ( s − 1 ) /2 , and the remaining powers are ≤ x − 2 n − 1 \leq x^{-2n-1} ≤ x − 2 n − 1 , so the odd part of r r r is O ( x − 2 n − 1 ) O(x^{-2n-1}) O ( x − 2 n − 1 ) . Put B = x 2 n ( x 2 − 1 ) n r B=x^{2n}(x^2-1)^nr B = x 2 n ( x 2 − 1 ) n r . In both nontrivial cases B B B is a polynomial divisible by ( x − 1 ) n (x-1)^n ( x − 1 ) n , of degree ≤ 4 n − 2 \leq4n-2 ≤ 4 n − 2 , whose odd part x 2 n ( x 2 − 1 ) n r o d d x^{2n}(x^2-1)^nr_{\mathrm{odd}} x 2 n ( x 2 − 1 ) n r odd has degree ≤ 2 n − 1 \leq2n-1 ≤ 2 n − 1 ; and ∫ 0 1 Q n h = ± κ n ∫ 0 1 U ~ B d x \int_0^1Q_nh=\pm\kappa_n\int_0^1\widetilde UB\,dx ∫ 0 1 Q n h = ± κ n ∫ 0 1 U B d x .
Step 3. Integrating by parts n n n times against D \mathcal{D} D (U ~ = D n [ x 2 n U ] \widetilde U=\mathcal{D}^n[x^{2n}U] U = D n [ x 2 n U ] ) gives ∫ 0 1 U ~ B = ± ∫ 0 1 x 2 n U D n B \int_0^1\widetilde UB=\pm\int_0^1x^{2n}U\mathcal{D}^nB ∫ 0 1 U B = ± ∫ 0 1 x 2 n U D n B ; the boundary terms vanish because D j B \mathcal{D}^jB D j B is divisible by ( x − 1 ) n − j (x-1)^{n-j} ( x − 1 ) n − j and is O ( x − 2 j ) O(x^{-2j}) O ( x − 2 j ) , while D n − 1 − j [ x 2 n U ] \mathcal{D}^{n-1-j}[x^{2n}U] D n − 1 − j [ x 2 n U ] vanishes to order 2 j + 2 2j+2 2 j + 2 at 0.
Step 4. Integrating by parts 2 n 2n 2 n times against D \mathcal{D} D (U = D 2 n E U=\mathcal{D}^{2n}E U = D 2 n E , E = ( x 2 − 1 ) 2 n x 4 n E=(x^2-1)^{2n}x^{4n} E = ( x 2 − 1 ) 2 n x 4 n ) gives ∫ 0 1 x 2 n U D n B = ± ∫ 0 1 D 2 n [ x 2 n D n B ] ⋅ E \int_0^1x^{2n}U\mathcal{D}^nB=\pm\int_0^1\mathcal{D}^{2n}[x^{2n}\mathcal{D}^nB]\cdot E ∫ 0 1 x 2 n U D n B = ± ∫ 0 1 D 2 n [ x 2 n D n B ] ⋅ E ; the boundary terms vanish because D 2 n − 1 − j E \mathcal{D}^{2n-1-j}E D 2 n − 1 − j E is divisible by ( x 2 − 1 ) j + 1 (x^2-1)^{j+1} ( x 2 − 1 ) j + 1 and vanishes to order 2 j + 2 2j+2 2 j + 2 at 0, while D j \mathcal{D}^j D j of a polynomial is O ( x − 2 j ) O(x^{-2j}) O ( x − 2 j ) . Step 5. x 2 n D n B x^{2n}\mathcal{D}^{n}B x 2 n D n B is a polynomial of degree ≤ 4 n − 2 \le4n-2 ≤ 4 n − 2 whose odd part x 2 n D n [ B o d d ] x^{2n}\mathcal{D}^{n}[B_{\mathrm{odd}}] x 2 n D n [ B odd ] vanishes (deg B o d d ≤ 2 n − 1 B_{\mathrm{odd}}\le2n-1 B odd ≤ 2 n − 1 ). An even polynomial of degree ≤ 4 n − 2 \le4n-2 ≤ 4 n − 2 is killed by ∂ 2 n \partial^{2n} ∂ 2 n , so every integral is 0. As sanity checks, i1_contact.py verifies (i) and (ii) exactly for small n n n , the moments of f 3 f_3 f 3 for m ≤ 30 m\le30 m ≤ 30 , and the contact orders exactly for n ≤ 24 n\le24 n ≤ 24 . □
Corollary 3.2. ∣ Q n ( x ) ∣ ≤ ( 2 n + 1 ) 2 20 n |Q_n(x)|\le(2n+1)^{2^{20n}} ∣ Q n ( x ) ∣ ≤ ( 2 n + 1 ) 2 20 n on [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] , and for every integer q ≥ 2 q\ge2 q ≥ 2 and j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 ,
∣ R j , n ( q ) ∣ ≤ ( 2 n + 1 ) ( 2 20 q 2 ) n ∣ μ j ∣ q − 1 . |R_{j,n}(q)|\le(2n+1)\left(\frac{2^{20}}{q^2}\right)^n\frac{|\mu_j|}{q-1}. ∣ R j , n ( q ) ∣ ≤ ( 2 n + 1 ) ( q 2 2 20 ) n q − 1 ∣ μ j ∣ .
Proof. We have
∑ l ∣ c n , l ∣ ≤ ∑ k max l ( n + l l ) 2 n ( 2 n + k 2 k ) ( n + k k ) ( 4 n + 2 k 2 n + k ) ≤ ( 2 n + 1 ) 2 4 n + n + 4 n + 3 n + 8 n . \begin{aligned}
\sum_l |c_{n,l}|\le\sum_k\max_l \binom{n+l}{l}2^n\binom{2n+k}{2k}\binom{n+k}{k}\binom{4n+2k}{2n+k} \\
&\le(2n+1)2^{4n+n+4n+3n+8n}.
\end{aligned} l ∑ ∣ c n , l ∣ ≤ k ∑ l max ( l n + l ) 2 n ( 2 k 2 n + k ) ( k n + k ) ( 2 n + k 4 n + 2 k ) ≤ ( 2 n + 1 ) 2 4 n + n + 4 n + 3 n + 8 n .
By Lemma 3.1, R j , n ( q ) = q − 2 n ∫ Q n ( x ) x 2 n d μ j ( x ) / ( q − x ) R_{j,n}(q)=q^{-2n}\int Q_n(x)x^{2n}\,d\mu_j(x)/(q-x) R j , n ( q ) = q − 2 n ∫ Q n ( x ) x 2 n d μ j ( x ) / ( q − x ) . □
Denominators
The following lemma is [6 Lemma 9] at prime-power level; we give the proof in full, so that [6 ] is used only for the definition of c n , l c_{n,l} c n , l .
Lemma 3.3. For every n ≥ 1 n\ge1 n ≥ 1 and j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 , the coefficients of P j , n P_{j,n} P j , n have denominators dividing [ 1 , … , 4 n ] [ 1 , … , 2 n ] [1,\ldots,4n][1,\ldots,2n] [ 1 , … , 4 n ] [ 1 , … , 2 n ] .
Proof. Write D k = [ 1 , … , k ] D_k=[1,\ldots,k] D k = [ 1 , … , k ] and fix a prime p p p . We use Kummer’s theorem in the form
v p ( ( A B ) ) = ∑ i ≥ 1 ( ⌊ A p i ⌋ − ⌊ B p i ⌋ − ⌊ A − B p i ⌋ ) , v_p\left(\binom{A}{B}\right)=\sum_{i\ge1}\left(\left\lfloor\frac{A}{p^i}\right\rfloor-\left\lfloor\frac{B}{p^i}\right\rfloor-\left\lfloor\frac{A-B}{p^i}\right\rfloor\right), v p ( ( B A ) ) = i ≥ 1 ∑ ( ⌊ p i A ⌋ − ⌊ p i B ⌋ − ⌊ p i A − B ⌋ ) ,
each term being 0 or 1. In every nonzero summand of c n , l c_{n,l} c n , l we have l − n ≤ k ≤ 2 n l-n\le k\le2n l − n ≤ k ≤ 2 n .
P 1 P_1 P 1 and P 2 P_2 P 2 . By (3.1) the denominators of P 1 , n P_{1,n} P 1 , n divide the numbers m ≤ 2 l ≤ 6 n m\le2l\le6n m ≤ 2 l ≤ 6 n , with coefficient c n , l c_{n,l} c n , l . If p ≤ 2 n p\le2n p ≤ 2 n , then v p ( m ) ≤ ⌊ log p 6 n ⌋ ≤ ⌊ log p 4 n ⌋ + 1 ≤ v p ( D 4 n D 2 n ) v_p(m)\le\lfloor\log_p 6n\rfloor\le\lfloor\log_p 4n\rfloor+1\le v_p(D_{4n}D_{2n}) v p ( m ) ≤ ⌊ log p 6 n ⌋ ≤ ⌊ log p 4 n ⌋ + 1 ≤ v p ( D 4 n D 2 n ) . If 2 n < p ≤ 4 n 2n<p\le4n 2 n < p ≤ 4 n , then p 2 > 6 n p^2>6n p 2 > 6 n , so v p ( m ) ≤ 1 = v p ( D 4 n ) v_p(m)\le1=v_p(D_{4n}) v p ( m ) ≤ 1 = v p ( D 4 n ) . If p > 4 n p>4n p > 4 n and p ∣ m p\mid m p ∣ m , then 4 n < p < 2 l 4n<p<2l 4 n < p < 2 l (p p p is odd). Then for every summand of c n , l c_{n,l} c n , l : 2 n + k ≤ 4 n < p 2n+k\le4n<p 2 n + k ≤ 4 n < p and p < 2 l ≤ 2 n + 2 k ≤ 4 n + 2 k ≤ 8 n < 2 p p<2l\le2n+2k\le4n+2k\le8n<2p p < 2 l ≤ 2 n + 2 k ≤ 4 n + 2 k ≤ 8 n < 2 p , so ⌊ ( 4 n + 2 k ) / p ⌋ − 2 ⌊ ( 2 n + k ) / p ⌋ = 1 \lfloor(4n+2k)/p\rfloor-2\lfloor(2n+k)/p\rfloor=1 ⌊( 4 n + 2 k ) / p ⌋ − 2 ⌊( 2 n + k ) / p ⌋ = 1 and p ∣ ( 4 n + 2 k 2 n + k ) p\mid\binom{4n+2k}{2n+k} p ∣ ( 2 n + k 4 n + 2 k ) . Hence p ∣ c n , l p\mid c_{n,l} p ∣ c n , l , and c n , l / m c_{n,l}/m c n , l / m is p p p -integral. (The signs ( − 1 ) m + 1 (-1)^{m+1} ( − 1 ) m + 1 play no role.) P 2 , n ( z ) = P 1 , n ( − z ) P_{2,n}(z)=P_{1,n}(-z) P 2 , n ( z ) = P 1 , n ( − z ) .
P 3 P_3 P 3 . By (3.1) the term c n , l A 2 m − 1 / m c_{n,l}A_{2m-1}/m c n , l A 2 m − 1 / m (m ≤ l ≤ 3 n m\le l\le3n m ≤ l ≤ 3 n ) has v p ≥ v p ( c n , l ) − e 1 − e 2 v_p\ge v_p(c_{n,l})-e_1-e_2 v p ≥ v p ( c n , l ) − e 1 − e 2 with e 1 = v p ( m ) e_1=v_p(m) e 1 = v p ( m ) and e 2 = ⌊ log p ( 2 m − 1 ) ⌋ e_2=\lfloor\log_p(2m-1)\rfloor e 2 = ⌊ log p ( 2 m − 1 )⌋ . Now e 1 ≤ ⌊ log p 2 n ⌋ + δ 1 e_1\le\lfloor\log_p 2n\rfloor+\delta_1 e 1 ≤ ⌊ log p 2 n ⌋ + δ 1 , where δ 1 = 1 \delta_1=1 δ 1 = 1 if some p a ∣ m p^a\mid m p a ∣ m with p a ∈ ( 2 n , 3 n ] p^a\in(2n,3n] p a ∈ ( 2 n , 3 n ] and δ 1 = 0 \delta_1=0 δ 1 = 0 otherwise (a higher excess is impossible since 2 p a > 3 n 2p^a>3n 2 p a > 3 n ); and e 2 ≤ ⌊ log p 4 n ⌋ + δ 2 e_2\le\lfloor\log_p 4n\rfloor+\delta_2 e 2 ≤ ⌊ log p 4 n ⌋ + δ 2 , where δ 2 = 1 \delta_2=1 δ 2 = 1 if some p b ∈ ( 4 n , 2 m − 1 ] p^b\in(4n,2m-1] p b ∈ ( 4 n , 2 m − 1 ] and δ 2 = 0 \delta_2=0 δ 2 = 0 otherwise (again 2 p b > 6 n 2p^b>6n 2 p b > 6 n ). If δ 1 + δ 2 > 0 \delta_1+\delta_2>0 δ 1 + δ 2 > 0 then m > 2 n m>2n m > 2 n , so l > 2 n l>2n l > 2 n and k ≥ l − n > n k\ge l-n>n k ≥ l − n > n in every summand. If δ 1 = 1 \delta_1=1 δ 1 = 1 : n < p a n<p^a n < p a , k ≤ 2 n < p a k\le2n<p^a k ≤ 2 n < p a and n + k ≥ l ≥ m ≥ p a n+k\ge l\ge m\ge p^a n + k ≥ l ≥ m ≥ p a , so p ∣ ( n + k k ) p\mid\binom{n+k}{k} p ∣ ( k n + k ) . If δ 2 = 1 \delta_2=1 δ 2 = 1 : 2 n + k ≤ 4 n < p b < 2 m ≤ 2 l ≤ 2 n + 2 k ≤ 4 n + 2 k 2n+k\le4n<p^b<2m\le2l\le2n+2k\le4n+2k 2 n + k ≤ 4 n < p b < 2 m ≤ 2 l ≤ 2 n + 2 k ≤ 4 n + 2 k , so p ∣ ( 4 n + 2 k 2 n + k ) p\mid\binom{4n+2k}{2n+k} p ∣ ( 2 n + k 4 n + 2 k ) . These are different binomial factors, so v p ( c n , l ) ≥ δ 1 + δ 2 v_p(c_{n,l})\ge\delta_1+\delta_2 v p ( c n , l ) ≥ δ 1 + δ 2 , and the term has v p ≥ − v p ( D 2 n D 4 n ) v_p\ge-v_p(D_{2n}D_{4n}) v p ≥ − v p ( D 2 n D 4 n ) . □
(An exact check of the divisibility for n ≤ 120 n \le120 n ≤ 120 , and independently for selected n ≤ 81 n \le81 n ≤ 81 , agrees.)
The recurrence
Proposition 3.4. There are polynomials p 0 , … , p 4 ∈ Z [ n , w ] p_0,\ldots,p_4 \in\mathbb{Z}[n,w] p 0 , … , p 4 ∈ Z [ n , w ] , of degree 34 in n n n and of degrees 3, 4, 5, 6, 3 in w w w , such that for every integer n ≥ 0 n \ge0 n ≥ 0 and every z ∈ C z \in\mathbb{C} z ∈ C , with w = z 2 w=z^2 w = z 2 ,
∑ i = 0 4 p i ( n , z 2 ) X n + i ( z ) = 0 for X ∈ { Q , P 1 , P 2 , P 3 } , \sum_{i=0}^{4} p_i(n,z^2)X_{n+i}(z)=0 \qquad\text{for } X \in\{Q,P_1,P_2,P_3\}, i = 0 ∑ 4 p i ( n , z 2 ) X n + i ( z ) = 0 for X ∈ { Q , P 1 , P 2 , P 3 } , hence also for X = R j X=R_j X = R j . Moreover
∑ j = 0 4 lc n ( p j ) ( w ) λ j = 2 59 ( 8 w + 1 ) ( 9 w − 1 ) ( 25 w − 16 ) ⋅ χ w ( − λ ) , \sum_{j=0}^{4} \operatorname{lc}_{n}(p_j)(w)\lambda^j
=2^{59}(8w+1)(9w-1)(25w-16)\cdot\chi_w(-\lambda), j = 0 ∑ 4 lc n ( p j ) ( w ) λ j = 2 59 ( 8 w + 1 ) ( 9 w − 1 ) ( 25 w − 16 ) ⋅ χ w ( − λ ) , where χ w ( λ ) = λ 4 + ( 16384 w 3 − 20480 w 2 + 5248 w − 64 ) λ 3 + ( 135168 w 2 + 159744 w + 1536 ) λ 2 + ( 294912 w − 16384 ) λ + 65536 \chi_w(\lambda)=\lambda^4+(16384w^3-20480w^2+5248w-64)\lambda^3+(135168w^2+159744w+1536)\lambda^2+(294912w-16384)\lambda+65536 χ w ( λ ) = λ 4 + ( 16384 w 3 − 20480 w 2 + 5248 w − 64 ) λ 3 + ( 135168 w 2 + 159744 w + 1536 ) λ 2 + ( 294912 w − 16384 ) λ + 65536 .
Proof (computer-assisted, exact). Two relations for c n , l c_{n,l} c n , l . There are explicit polynomials a 0 , a 1 , a 2 a_0,a_1,a_2 a 0 , a 1 , a 2 and α , β a , i \alpha,\beta_{a,i} α , β a , i with
∑ m = 0 2 a m ( n , l ) c n , l + m = 0 , α ( n ) c n + 1 , l = ∑ a = 0 3 ∑ i = 0 3 β a , i ( n ) ( l − i ) a c n , l − i , \sum_{m=0}^{2}a_m(n,l)c_{n,l+m}=0,\qquad
\alpha(n)c_{n+1,l}=\sum_{a=0}^{3}\sum_{i=0}^{3}\beta_{a,i}(n)(l-i)^a c_{n,l-i}, m = 0 ∑ 2 a m ( n , l ) c n , l + m = 0 , α ( n ) c n + 1 , l = a = 0 ∑ 3 i = 0 ∑ 3 β a , i ( n ) ( l − i ) a c n , l − i , a 0 = − ( l − 3 n ) ( l + n + 1 ) ( l + n + 2 ) ( 2 l + 2 n + 1 ) a_0=-(l-3n)(l+n+1)(l+n+2)(2l+2n+1) a 0 = − ( l − 3 n ) ( l + n + 1 ) ( l + n + 2 ) ( 2 l + 2 n + 1 ) , a 2 = − ( l + 2 ) 3 ( 2 l + 3 ) a_2=-(l+2)^3(2l+3) a 2 = − ( l + 2 ) 3 ( 2 l + 3 ) , α = − ( n + 1 ) 3 ( 2 n + 1 ) 2 ( 4 n + 1 ) ( 4 n + 3 ) \alpha=-(n+1)^3(2n+1)^2(4n+1)(4n+3) α = − ( n + 1 ) 3 ( 2 n + 1 ) 2 ( 4 n + 1 ) ( 4 n + 3 ) (the others are in runs/i1_relations.json). They are proved by creative telescoping in k k k : with explicit certificates G 1 , G 2 G_1,G_2 G 1 , G 2 (hypergeometric terms given by factorial formulas), the summand identities
∑ m a m F ( n , l + m , k ) = G 1 ( n , l , k + 1 ) − G 1 ( n , l , k ) \sum_m a_mF(n,l+m,k)=G_1(n,l,k+1)-G_1(n,l,k) m ∑ a m F ( n , l + m , k ) = G 1 ( n , l , k + 1 ) − G 1 ( n , l , k ) and its analogue hold at every integer point of the relevant regions. Over a common factorial base each reduces to one polynomial identity in Q [ n , l , k ] \mathbb{Q}[n,l,k] Q [ n , l , k ] , which is identically zero; at k = − 1 k=-1 k = − 1 the certificates vanish (factor k 2 k^2 k 2 ), they vanish for k > 2 n + 2 k>2n+2 k > 2 n + 2 , so the sums over k k k telescope; the finitely many boundary cases are checked directly (i1_certs.py).
Order-4 operator. In generating-function form the relations say D n Q n = 0 D_nQ_n=0 D n Q n = 0 , with D n D_n D n an operator of order 4 in w d / d w w\,d/dw w d / d w whose leading coefficient − 2 ( w − 1 ) 2 -2(w-1)^2 − 2 ( w − 1 ) 2 does not depend on n n n , and α ( n ) Q n + 1 = E n Q n \alpha(n)Q_{n+1}=E_nQ_n α ( n ) Q n + 1 = E n Q n with E n E_n E n of order 3. Reducing E n + i ⋯ E n E_{n+i}\cdots E_n E n + i ⋯ E n modulo D n D_n D n on the right expresses Q n + i Q_{n+i} Q n + i , i = 0 , … , 4 i=0,\ldots,4 i = 0 , … , 4 , through Q n , … , ( w d / d w ) 3 Q n Q_n,\ldots,(w\,d/dw)^3Q_n Q n , … , ( w d / d w ) 3 Q n ; the cofactor vector of the resulting 4 × 5 4\times5 4 × 5 matrix, made primitive, is ( p 0 , … , p 4 ) (p_0,\ldots,p_4) ( p 0 , … , p 4 ) , and ∑ i p i Q n + i = 0 \sum_i p_iQ_{n+i}=0 ∑ i p i Q n + i = 0 is re-verified exactly (i1_opalg.py).
The numerators. ∑ i p i ( n , z 2 ) P j , n + i = − ∑ i p i ( n , z 2 ) R j , n + i \sum_i p_i(n,z^2)P_{j,n+i}=-\sum_i p_i(n,z^2)R_{j,n+i} ∑ i p i ( n , z 2 ) P j , n + i = − ∑ i p i ( n , z 2 ) R j , n + i . The left side is a polynomial, and by Lemma 3.1 the right side is O ( z 5 − 2 n ) O(z^{5-2n}) O ( z 5 − 2 n ) ; so both vanish for n ≥ 3 n\ge3 n ≥ 3 , and n = 0 , 1 , 2 n=0,1,2 n = 0 , 1 , 2 are checked directly. The characteristic polynomial identity is exact (i1_extras.py). □
For q ≥ 2 q \ge2 q ≥ 2 the prefactor 2 59 ( 8 w + 1 ) ( 9 w − 1 ) ( 25 w − 16 ) 2^{59}(8w+1)(9w-1)(25w-16) 2 59 ( 8 w + 1 ) ( 9 w − 1 ) ( 25 w − 16 ) does not vanish at w = q 2 w=q^2 w = q 2 , so the characteristic roots of the recurrence are the roots λ \lambda λ of χ q 2 ( − λ ) \chi_{q^2}(-\lambda) χ q 2 ( − λ ) . For 2 ≤ q ≤ 60 2 \le q \le60 2 ≤ q ≤ 60 they are real, positive and distinct, and 1 / λ 1 < 2 − 20 q 2 1/\lambda_1 < 2^{-20q^2} 1/ λ 1 < 2 − 20 q 2 (Arb root isolation, fourterm/check_y1.py); we order them λ 1 > λ 2 > λ 3 > λ 4 > 0 \lambda_1 > \lambda_2 > \lambda_3 > \lambda_4 > 0 λ 1 > λ 2 > λ 3 > λ 4 > 0 and put y k = 1 / λ k y_k=1/\lambda_k y k = 1/ λ k , so that 0 < y 1 < y 2 < y 3 < y 4 0<y_1<y_2<y_3<y_4 0 < y 1 < y 2 < y 3 < y 4 . For example, for q = 26 q=26 q = 26 : y 1 ≈ 1.98 ⋅ 10 − 13 y_1 \approx1.98\cdot10^{-13} y 1 ≈ 1.98 ⋅ 1 0 − 13 , y 2 ≈ 156.37 y_2\approx156.37 y 2 ≈ 156.37 , y 3 ≈ 182.38 y_3\approx182.38 y 3 ≈ 182.38 , y 4 ≈ 2703 y_4\approx2703 y 4 ≈ 2703 . (For every q ≥ 2 q\ge2 q ≥ 2 the roots are distinct, since the discriminant of c 34 c_{34} c 34 below does not vanish at w = q 2 w=q^2 w = q 2 ; we use their reality and order only for 26 ≤ q ≤ 53 26\le q\le53 26 ≤ q ≤ 53 .)
The differential equations
Fix an integer q q q with 2 ≤ q ≤ 60 2\le q\le60 2 ≤ q ≤ 60 and put
H j ( y ) = ∑ n > 0 R j , n ( q ) y n ( j = 1 , 2 , 3 ) , H a = ∑ j = 1 3 a j H j ( a ∈ C 3 ) . H_j(y)=\sum_{n>0}R_{j,n}(q)y^n\quad(j=1,2,3),\qquad H_a=\sum_{j=1}^{3}a_jH_j\quad(a\in\mathbb{C}^3). H j ( y ) = n > 0 ∑ R j , n ( q ) y n ( j = 1 , 2 , 3 ) , H a = j = 1 ∑ 3 a j H j ( a ∈ C 3 ) .
By Corollary 3.2, each H j H_j H j is holomorphic on ∣ y ∣ < 2 − 20 q 2 |y|<2^{-20q^2} ∣ y ∣ < 2 − 20 q 2 , a disc containing 0 0 0 and y 1 y_1 y 1 . Let ϑ = y d / d y \vartheta=y\,d/dy ϑ = y d / d y and
L = ∑ j = 0 4 y 4 − j p j ( ϑ − j , q 2 ) , L=\sum_{j=0}^{4}y^{4-j}p_j(\vartheta-j,q^2), L = j = 0 ∑ 4 y 4 − j p j ( ϑ − j , q 2 ) ,
a differential operator of order 34 34 34 . The coefficient of y N + 4 y^{N+4} y N + 4 in L ∑ X n y n L\sum X_ny^n L ∑ X n y n is ∑ j p j ( N ) X N + j \sum_jp_j(N)X_{N+j} ∑ j p j ( N ) X N + j , so by Proposition 3.4 , L H j = E j LH_j=E_j L H j = E j with E j E_j E j a polynomial of degree ≤ 3 \le3 ≤ 3 . The ϑ 34 \vartheta^{34} ϑ 34 -coefficient of L L L is c 34 ( y ) = ∑ j y 4 − j c n ( p j ) c_{34}(y)=\sum_jy^{4-j}c_n(p_j) c 34 ( y ) = ∑ j y 4 − j c n ( p j ) , which is a constant times ( 8 w + 1 ) ( 9 w − 1 ) ( 25 w − 16 ) y 4 χ w ( − 1 / y ) (8w+1)(9w-1)(25w-16)y^4\chi_w(-1/y) ( 8 w + 1 ) ( 9 w − 1 ) ( 25 w − 16 ) y 4 χ w ( − 1/ y ) ; its roots are y 1 , … , y 4 y_1,\ldots,y_4 y 1 , … , y 4 , and c 34 ( 0 ) ≠ 0 c_{34}(0)\ne0 c 34 ( 0 ) = 0 . Hence the singular points of L L L are 0 , y 1 , … , y 4 , ∞ 0,y_1,\ldots,y_4,\infty 0 , y 1 , … , y 4 , ∞ , and L L L has no apparent singularities.
Proposition 3.5. For every a ∈ C 3 a\in\mathbb{C}^3 a ∈ C 3 , H a H_a H a continues analytically to a single-valued holomorphic function on C ∖ [ y 2 , ∞ ) \mathbb{C}\setminus[y_2,\infty) C ∖ [ y 2 , ∞ ) .
Proof. H a H_a H a is holomorphic on the disc ∣ y ∣ < 2 − 20 q 2 |y|<2^{-20q^2} ∣ y ∣ < 2 − 20 q 2 , which contains 0 0 0 and y 1 y_1 y 1 , and satisfies L H a = E a LH_a=E_a L H a = E a ; solutions of a linear differential equation continue along every path avoiding its singular points. The fundamental group of C ∖ ( [ y 2 , ∞ ) ∪ { 0 , y 1 } ) \mathbb{C}\setminus([y_2,\infty)\cup\{0,y_1\}) C ∖ ([ y 2 , ∞ ) ∪ { 0 , y 1 }) is generated by loops around 0 0 0 and y 1 y_1 y 1 that lie in that disc, where H a H_a H a is single-valued; so the continuation is single-valued on C ∖ ( [ y 2 , ∞ ) ∪ { 0 , y 1 } ) \mathbb{C}\setminus([y_2,\infty)\cup\{0,y_1\}) C ∖ ([ y 2 , ∞ ) ∪ { 0 , y 1 }) and, being holomorphic near 0 0 0 and y 1 y_1 y 1 , extends holomorphically to C ∖ [ y 2 , ∞ ) \mathbb{C}\setminus[y_2,\infty) C ∖ [ y 2 , ∞ ) . □
The minimal operator
For each q q q with 26 ≤ q ≤ 53 26\le q\le53 26 ≤ q ≤ 53 we computed an operator of recurrence order 25 and n n n -degree 13, common to the four sequences Q , P 1 , P 2 , P 3 Q,P_1,P_2,P_3 Q , P 1 , P 2 , P 3 at z = q z=q z = q (modular nullspaces, Chinese remaindering and rational reconstruction; checked exactly for n ≤ 60 n\le60 n ≤ 60 ). Its associated differential operator L 13 = ∑ j = 0 25 y 25 − j p ~ j ( ϑ − j ) L_{13}=\sum_{j=0}^{25}y^{25-j}\widetilde{p}_j(\vartheta-j) L 13 = ∑ j = 0 25 y 25 − j p j ( ϑ − j ) has order 13. Its leading coefficient has degree 25: its roots are y 1 , … , y 4 y_1,\ldots,y_4 y 1 , … , y 4 and 21 apparent singular points.
Proposition 3.6. For each q q q with 26 ≤ q ≤ 53 26 \le q \le53 26 ≤ q ≤ 53 the order-25 recurrence holds for all n ≥ 0 n \ge0 n ≥ 0 on Q , P 1 , P 2 , P 3 Q,P_1,P_2,P_3 Q , P 1 , P 2 , P 3 at z = q z=q z = q . Hence L 13 H a = E a ′ L_{13}H_a=E'_a L 13 H a = E a ′ with E a ′ E'_a E a ′ a polynomial, for every a ∈ C 3 a \in\mathbb{C}^3 a ∈ C 3 .
Proof (computer-assisted, exact). Fraction-free right division in Z [ n ] [ S ] \mathbb{Z}[n][S] Z [ n ] [ S ] (S S S the shift) gives μ ( n ) ⋅ Rec 25 = K ~ ⋅ L 4 \mu(n)\cdot\operatorname{Rec}_{25}=\widetilde{K}\cdot L_4 μ ( n ) ⋅ Rec 25 = K ⋅ L 4 with zero remainder, where L 4 = ∑ i p i ( n , q 2 ) S i L_4=\sum_i p_i(n,q^2)S^i L 4 = ∑ i p i ( n , q 2 ) S i and μ ( n ) \mu(n) μ ( n ) is a nonzero integer times ∏ t = 0 21 p 4 ( n + t , q 2 ) \prod_{t=0}^{21}p_4(n+t,q^2) ∏ t = 0 21 p 4 ( n + t , q 2 ) . The polynomial p 4 ( ⋅ , q 2 ) p_4(\cdot,q^2) p 4 ( ⋅ , q 2 ) has no root at a nonnegative integer, so μ ( n ) ≠ 0 \mu(n)\ne0 μ ( n ) = 0 for n ≥ 0 n\ge0 n ≥ 0 , and Proposition 3.4 gives the claim (i1_rec25.py). □
Proof of Theorem 1
Fix q q q with 26 ≤ q ≤ 53 26 \le q \le53 26 ≤ q ≤ 53 . Write f = ( f 1 ( q ) , f 2 ( q ) , f 3 ( q ) ) f=(f_1(q),f_2(q),f_3(q)) f = ( f 1 ( q ) , f 2 ( q ) , f 3 ( q )) , P n = ( P 1 , n ( q ) , P 2 , n ( q ) , P 3 , n ( q ) ) P_n=(P_{1,n}(q),P_{2,n}(q),P_{3,n}(q)) P n = ( P 1 , n ( q ) , P 2 , n ( q ) , P 3 , n ( q )) and v n = ( Q n ( q ) , P n ) ∈ Q 4 v_n=(Q_n(q),P_n)\in\mathbb{Q}^4 v n = ( Q n ( q ) , P n ) ∈ Q 4 .
Reduction
Suppose r 0 + a 1 f 1 + a 2 f 2 + a 3 f 3 = 0 r_0+a_1f_1+a_2f_2+a_3f_3=0 r 0 + a 1 f 1 + a 2 f 2 + a 3 f 3 = 0 with ( r 0 , a ) ∈ Z 4 ∖ { 0 } (r_0,a)\in\mathbb{Z}^4\setminus\{0\} ( r 0 , a ) ∈ Z 4 ∖ { 0 } . Then a ≠ 0 a\ne0 a = 0 , and
G ( y ) : = ∑ n ≥ 0 ( r 0 Q n ( q ) + a ⋅ P n ) y n = ∑ n ( Q n ( q ) ( r 0 + a ⋅ f ) − a ⋅ R n ) y n = − H a ( y ) , G(y):=\sum_{n\ge0}(r_0Q_n(q)+a\cdot P_n)y^n=\sum_n(Q_n(q)(r_0+a\cdot f)-a\cdot R_n)y^n=-H_a(y), G ( y ) := n ≥ 0 ∑ ( r 0 Q n ( q ) + a ⋅ P n ) y n = n ∑ ( Q n ( q ) ( r 0 + a ⋅ f ) − a ⋅ R n ) y n = − H a ( y ) ,
where R n = ( R 1 , n ( q ) , R 2 , n ( q ) , R 3 , n ( q ) ) R_n=(R_{1,n}(q),R_{2,n}(q),R_{3,n}(q)) R n = ( R 1 , n ( q ) , R 2 , n ( q ) , R 3 , n ( q )) . By Lemma 3.3 (and Q n ( q ) ∈ Z Q_n(q)\in\mathbb{Z} Q n ( q ) ∈ Z ), G ∈ Q [ [ y ] ] G\in\mathbb{Q[[y]]} G ∈ Q [[ y ]] has constant term r 0 ∈ Z r_0\in\mathbb{Z} r 0 ∈ Z and denominator type [ 1 , … , 4 n ] [ 1 , … , 2 n ] [1,\ldots,4n][1,\ldots,2n] [ 1 , … , 4 n ] [ 1 , … , 2 n ] ; so do ϑ j G \vartheta^jG ϑ j G for all j j j (the coefficients are multiplied by n j n^j n j ). By Proposition 3.5 , G G G continues to C ∖ [ y 2 , ∞ ) \mathbb{C}\setminus[y_2,\infty) C ∖ [ y 2 , ∞ ) .
Let ϕ ( w ) = 4 y 2 w / ( 1 + w ) 2 \phi(w)=4y_2w/(1+w)^2 ϕ ( w ) = 4 y 2 w / ( 1 + w ) 2 , the Koebe map of D \mathbb{D} D onto C ∖ [ y 2 , ∞ ) \mathbb{C}\setminus[y_2,\infty) C ∖ [ y 2 , ∞ ) , and ϕ r ( w ) = ϕ ( r w ) \phi_r(w)=\phi(rw) ϕ r ( w ) = ϕ ( r w ) for 0 < r < 1 0<r<1 0 < r < 1 . Consider the m = 14 m=14 m = 14 functions 1 , G , ϑ G , … , ϑ 12 G 1,G,\vartheta G,\ldots,\vartheta^{12}G 1 , G , ϑG , … , ϑ 12 G , with type rows ( 0 , 0 ) (0,0) ( 0 , 0 ) for the constant 1 1 1 and ( 4 , 2 ) (4,2) ( 4 , 2 ) for the others, e = 0 e=0 e = 0 . Then σ i = 6 \sigma_i=6 σ i = 6 for i ≥ 2 i\ge2 i ≥ 2 and τ = 1 196 ∑ i = 2 14 ( 2 i − 1 ) ⋅ 6 = 6 ⋅ 195 196 \tau=\frac{1}{196}\sum_{i=2}^{14}(2i-1)\cdot6=6\cdot\frac{195}{196} τ = 196 1 ∑ i = 2 14 ( 2 i − 1 ) ⋅ 6 = 6 ⋅ 196 195 .
All pullbacks by ϕ r \phi_r ϕ r are holomorphic on a neighbourhood of D ‾ \overline{\mathbb{D}} D , and ϕ r \phi_r ϕ r is univalent, so T ( ϕ r ) = log ∣ ϕ r ′ ( 0 ) ∣ = L q + log r T(\phi_r)=\log|\phi'_r(0)|=L_q+\log r T ( ϕ r ) = log ∣ ϕ r ′ ( 0 ) ∣ = L q + log r by Lemma 2.3 , where
L q = log ( 4 y 2 ) . L_q=\log(4y_2). L q = log ( 4 y 2 ) .
If the 14 functions are linearly independent over Q ( y ) \mathbb{Q}(y) Q ( y ) , Theorem 2.1 gives 14 ≤ ( L q + log r ) / ( L q + log r − τ ) 14\le(L_q+\log r)/(L_q+\log r-\tau) 14 ≤ ( L q + log r ) / ( L q + log r − τ ) for every r < 1 r<1 r < 1 close to 1, hence 14 ≤ L q / ( L q − τ ) 14\le L_q/(L_q-\tau) 14 ≤ L q / ( L q − τ ) , that is
M 14 ( q ) : = 13 14 L q − 6 ⋅ 195 196 ≤ 0. M_{14}(q):=\frac{13}{14}L_q-6\cdot\frac{195}{196}\le0. M 14 ( q ) := 14 13 L q − 6 ⋅ 196 195 ≤ 0.
(The condition L q > max { 6 , τ } = 6 L_q>\max\{6,\tau\}=6 L q > max { 6 , τ } = 6 of Theorem 2.1 holds.) But M 14 ( q ) > 0 M_{14}(q)>0 M 14 ( q ) > 0 for every q q q with 26 ≤ q ≤ 53 26\le q\le53 26 ≤ q ≤ 53 ; the Arb values are in Table 1, for example M 14 ( 26 ) = 0.0092251638 ± 5 ⋅ 10 − 12 M_{14}(26)=0.0092251638\pm5\cdot10^{-12} M 14 ( 26 ) = 0.0092251638 ± 5 ⋅ 1 0 − 12 and M 14 ( 53 ) = 1.3688032 ± 3 ⋅ 10 − 8 M_{14}(53)=1.3688032\pm3\cdot10^{-8} M 14 ( 53 ) = 1.3688032 ± 3 ⋅ 1 0 − 8 (budget_4t.py; y 2 y_2 y 2 is a root of the exact characteristic polynomial of Proposition 3.4 ). This contradiction proves Theorem 1, given:
Proposition 4.1. Let a ∈ C 3 ∖ { 0 } a\in\mathbb{C}^3\setminus\{0\} a ∈ C 3 ∖ { 0 } . The 14 functions 1 , H a , ϑ H a , … , ϑ 12 H a 1,H_a,\vartheta H_a,\ldots,\vartheta^{12}H_a 1 , H a , ϑ H a , … , ϑ 12 H a are linearly independent over C ( y ) \mathbb{C}(y) C ( y ) , hence over Q ( y ) \mathbb{Q}(y) Q ( y ) .
The rest of this section proves Proposition 4.1 . For a = ( 0 , 0 , 1 ) a=(0,0,1) a = ( 0 , 0 , 1 ) it is the case used for Corollary 2 , which we also certified separately (Remark 4.6).
An independence lemma
Fix a base point y ∗ ∈ ( y 1 , y 2 ) y_* \in(y_1,y_2) y ∗ ∈ ( y 1 , y 2 ) which is a regular point of L 13 L_{13} L 13 , and consider germs at y ∗ y_* y ∗ ; γ \gamma γ runs over closed paths at y ∗ y_* y ∗ avoiding all singular points of L 13 L_{13} L 13 . Let
U a = span C { ( γ − 1 ) H a } , S = Sol ( L 13 ) ( dim S = 13 ) . U_a = \operatorname{span}_{\mathbb{C}}\{(\gamma-1)H_a\}, \qquad S = \operatorname{Sol}(L_{13}) \;(\dim S = 13). U a = span C {( γ − 1 ) H a } , S = Sol ( L 13 ) ( dim S = 13 ) .
By Proposition 3.6 , U a ⊆ S U_a \subseteq S U a ⊆ S . U a U_a U a is invariant under monodromy, since δ ( γ − 1 ) H a = ( δ γ − 1 ) H a − ( δ − 1 ) H a \delta(\gamma-1)H_a=(\delta\gamma-1)H_a-(\delta-1)H_a δ ( γ − 1 ) H a = ( δ γ − 1 ) H a − ( δ − 1 ) H a .
Lemma 4.2. If dim U a = u \dim U_a=u dim U a = u , then 1 , H a , ϑ H a , … , ϑ u − 1 H a 1,H_a,\vartheta H_a,\ldots,\vartheta^{u-1}H_a 1 , H a , ϑ H a , … , ϑ u − 1 H a are linearly independent over C ( y ) \mathbb{C}(y) C ( y ) .
Proof. Suppose c 0 + K H a = 0 c_0+KH_a=0 c 0 + K H a = 0 with K = ∑ j < u c j + 1 ϑ j K=\sum_{j<u}c_{j+1}\vartheta^j K = ∑ j < u c j + 1 ϑ j , c i ∈ C ( y ) c_i\in\mathbb{C}(y) c i ∈ C ( y ) . Continuation along γ \gamma γ (perturbed off the poles of the c i c_i c i ) gives c 0 + K ( γ H a ) = 0 c_0+K(\gamma H_a)=0 c 0 + K ( γ H a ) = 0 , so K K K kills ( γ − 1 ) H a (\gamma-1)H_a ( γ − 1 ) H a for every γ \gamma γ , hence the u u u -dimensional space U a U_a U a . A nonzero operator of order ≤ u − 1 \leq u-1 ≤ u − 1 cannot do that, so K = 0 K=0 K = 0 , and then c 0 = 0 c_0=0 c 0 = 0 .
So Proposition 4.1 follows from U a = S U_a=S U a = S , which in turn follows from U a ≠ 0 U_a\ne0 U a = 0 (Lemma 4.4 ) and the irreducibility of S S S under the monodromy (Proposition 4.5 ).
Local exponents
Proposition 4.3. For every integer q ≥ 2 q\geq2 q ≥ 2 , each of the four roots y k y_k y k of c 34 c_{34} c 34 is a regular singular point of L L L with exponents 0 , 1 , … , 32 0,1,\ldots,32 0 , 1 , … , 32 and 1 2 \frac{1}{2} 2 1 . Every local solution of L f = E Lf=E L f = E (E E E holomorphic at y k y_k y k ) has the form h 0 ( t ) + c t 1 / 2 h ( t ) h_0(t)+ct^{1/2}h(t) h 0 ( t ) + c t 1/2 h ( t ) , t = y − y k t=y-y_k t = y − y k , with h 0 , h h_0,h h 0 , h holomorphic and h ( 0 ) ≠ 0 h(0)\ne0 h ( 0 ) = 0 . In particular the local monodromy of the affine solution space of L f = E Lf=E L f = E at y k y_k y k is a reflection, and a solution is single-valued around y k y_k y k if and only if it is holomorphic there.
Proof (computer-assisted, exact). Let c 34 , c 33 c_{34},c_{33} c 34 , c 33 be the ϑ 34 \vartheta^{34} ϑ 34 - and ϑ 33 \vartheta^{33} ϑ 33 -coefficients of L L L . In d / d y d/dy d / d y form, a 34 = y 34 c 34 a_{34}=y^{34}c_{34} a 34 = y 34 c 34 has a simple zero at y k y_k y k (the discriminant of c 34 c_{34} c 34 is a constant times w 3 ( w − 1 ) 2 ( 8 w + 1 ) 8 ( 9 w − 1 ) 8 ( 25 w − 16 ) 8 ≠ 0 w^3(w-1)^2(8w+1)^8(9w-1)^8(25w-16)^8\ne0 w 3 ( w − 1 ) 2 ( 8 w + 1 ) 8 ( 9 w − 1 ) 8 ( 25 w − 16 ) 8 = 0 ) and a 33 ( y k ) = y k 33 c 33 ( y k ) a_{33}(y_k)=y_k^{33}c_{33}(y_k) a 33 ( y k ) = y k 33 c 33 ( y k ) . So y k y_k y k is regular singular with exponents 0 , … , 32 0,\ldots,32 0 , … , 32 and β k = 33 − c 33 ( y k ) / ( y k c 34 ′ ( y k ) ) \beta_k=33-c_{33}(y_k)/(y_kc_{34}'(y_k)) β k = 33 − c 33 ( y k ) / ( y k c 34 ′ ( y k )) . The pseudo-remainder of 2 c 33 − 65 y c 34 ′ 2c_{33}-65yc'_{34} 2 c 33 − 65 y c 34 ′ by c 34 c_{34} c 34 in Z [ w ] [ y ] \mathbb{Z}[w][y] Z [ w ] [ y ] is identically zero, so β k = 1 2 \beta_k=\frac{1}{2} β k = 2 1 for all k k k and all q q q (cross-check: Res y ( c 34 , ( 33 − b ) y c 34 ′ − c 33 ) \operatorname{Res}_y(c_{34},(33-b)yc'_{34}-c_{33}) Res y ( c 34 , ( 33 − b ) y c 34 ′ − c 33 ) is a constant times w 3 ( w − 1 ) 2 ( 8 w + 1 ) 10 ( 9 w − 1 ) 10 ( 25 w − 16 ) 10 ( 2 b − 1 ) 4 w^3(w-1)^2(8w+1)^{10}(9w-1)^{10}(25w-16)^{10}(2b-1)^4 w 3 ( w − 1 ) 2 ( 8 w + 1 ) 10 ( 9 w − 1 ) 10 ( 25 w − 16 ) 10 ( 2 b − 1 ) 4 ). In the power-series recursion at y k y_k y k the pivot vanishes only at exponent 1 2 \frac{1}{2} 2 1 , never at an integer, which gives 33 holomorphic homogeneous solutions, a holomorphic particular solution and one solution t 1 / 2 h ( t ) t^{1/2}h(t) t 1/2 h ( t ) (fourterm/exponents.py).
These statements concern L L L , not L 13 L_{13} L 13 (which has its own exponents and apparent points); they are transferred to S S S only after S ⊆ Sol ( L ) S\subseteq\operatorname{Sol}(L) S ⊆ Sol ( L ) has been certified (step (a) below).
No combination is single-valued
Lemma 4.4. For every a ∈ C 3 ∖ { 0 } a\in\mathbb{C}^3\setminus\{0\} a ∈ C 3 ∖ { 0 } , U a ≠ 0 U_a\ne0 U a = 0 .
Proof. Suppose U a = 0 U_a=0 U a = 0 , so that H a H_a H a extends to a single-valued holomorphic function on C ∖ { 0 , y 1 , … , y 4 } \mathbb{C}\setminus\{0,y_1,\ldots,y_4\} C ∖ { 0 , y 1 , … , y 4 } .
It is holomorphic at 0 0 0 and y 1 y_1 y 1 (where it equals its power series) and at y 2 , y 3 , y 4 y_2,y_3,y_4 y 2 , y 3 , y 4 by Proposition 4.3 . So H a H_a H a is entire.
E a E_a E a is a polynomial of degree ≤ 3 \le3 ≤ 3 , so H a H_a H a is killed by L ′ = ϑ ( ϑ − 1 ) ( ϑ − 2 ) ( ϑ − 3 ) L L'=\vartheta(\vartheta-1)(\vartheta-2)(\vartheta-3)L L ′ = ϑ ( ϑ − 1 ) ( ϑ − 2 ) ( ϑ − 3 ) L . With u = 1 / y u=1/y u = 1/ y , the ϑ u \vartheta_u ϑ u -leading coefficient of L ′ L' L ′ at u = 0 u=0 u = 0 is ± lc ( p 0 ) ≠ 0 \pm\operatorname{lc}(p_0)\ne0 ± lc ( p 0 ) = 0 , so ∞ \infty ∞ is a regular singular point. A single-valued solution of a Fuchsian equation near ∞ \infty ∞ has moderate growth, hence is meromorphic at ∞ \infty ∞ : H a H_a H a is a polynomial.
L y D = ∑ j p j ( D − j ) y D + 4 − j Ly^D=\sum_j p_j(D-j)y^{D+4-j} L y D = ∑ j p j ( D − j ) y D + 4 − j , so if H a ≢ 0 H_a\not\equiv0 H a ≡ 0 has degree D ≥ 0 D\ge0 D ≥ 0 , the coefficient of y D + 4 y^{D+4} y D + 4 in L H a LH_a L H a is p 0 ( D , q 2 ) p_0(D,q^2) p 0 ( D , q 2 ) times the leading coefficient of H a H_a H a ; p 0 ( ⋅ , q 2 ) p_0(\mathord\cdot,q^2) p 0 ( ⋅ , q 2 ) has no nonnegative integer root for 2 ≤ q ≤ 60 2\le q\le60 2 ≤ q ≤ 60 . Since deg E a ≤ 3 < D + 4 \deg E_a\le3<D+4 deg E a ≤ 3 < D + 4 , H a ≡ 0 H_a\equiv0 H a ≡ 0 , that is, ( a ⋅ f ) Q n − a ⋅ P n = 0 (a\mathord\cdot f)Q_n-a\mathord\cdot P_n=0 ( a ⋅ f ) Q n − a ⋅ P n = 0 for all n n n .
The Casoratian det [ v 0 , v 1 , v 2 , v 3 ] \det[v_0,v_1,v_2,v_3] det [ v 0 , v 1 , v 2 , v 3 ] equals, as a function of z = q z=q z = q ,
− z 291060 ( 7427354003333880 z 6 − 4691165040913890 z 4 − 141637476029121 z 2 + 63093459176677 ) -\frac{z}{291060}\left(7427354003333880z^6-4691165040913890z^4-141637476029121z^2+63093459176677\right) − 291060 z ( 7427354003333880 z 6 − 4691165040913890 z 4 − 141637476029121 z 2 + 63093459176677 )
which has no integer root other than 0. So v 0 , … , v 3 v_0,\ldots,v_3 v 0 , … , v 3 are independent and ( a ⋅ f , − a ) = 0 (a\mathord\cdot f,-a)=0 ( a ⋅ f , − a ) = 0 , i.e. a = 0 a=0 a = 0 .
Irreducibility of the monodromy
The data. For each q q q , the script mono13.py computes rigorous enclosures of the monodromy of g = y 5 H 3 g=y^5H_3 g = y 5 H 3 (multiplication by y 5 y^5 y 5 is equivariant, and shifts the exponents of L 13 L_{13} L 13 at 0 to make the Taylor recursion well-conditioned) on the 14-dimensional affine module spanned by g g g and y 5 S y^5S y 5 S . Coordinates: the 13 scaled Taylor coefficients at the base point y ∗ = round ( 0.42 y 2 ) y_*=\operatorname{round}(0.42y_2) y ∗ = round ( 0.42 y 2 ) and an affine coordinate. The generators are: A A A , the circle ∣ y ∣ = y ∗ |y|=y_* ∣ y ∣ = y ∗ (which encloses 0 0 0 , y 1 y_1 y 1 and some apparent points); 2 and 3, loops around y 2 y_2 y 2 and y 3 y_3 y 3 conjugated by paths that avoid all 21 apparent points. The computation:
initial data at b = 2 − 21 b=2^{-21} b = 2 − 21 from the power series of g g g , with exact Q n ( q ) Q_n(q) Q n ( q ) , P 3 , n ( q ) P_{3,n}(q) P 3 , n ( q ) and the tail bound of Corollary 3.2;
Taylor steps c ↦ c + h c\mapsto c+h c ↦ c + h along the paths, ∣ h ∣ ≤ 0.2 δ |h|\le0.2\delta ∣ h ∣ ≤ 0.2 δ (δ \delta δ the distance to the nearest singular point of L 13 L_{13} L 13 ): coefficients of an approximate solution by the recurrence at a midpoint, the residual L 13 y ^ − E L_{13}\hat y-E L 13 y ^ − E computed in balls, and the error y − y ^ y-\hat y y − y ^ bounded by a Cauchy majorant whose constants max ∣ A i / A 13 ∣ \max|A_i/A_{13}| max ∣ A i / A 13 ∣ on a circle are enclosed by arc subdivision using factored forms of the coefficients and certified root balls of the leading coefficient; the step matrix is enclosed and applied as a ball product;
working precision 2200 bits, truncation target 1700 bits per step; all radii of the generator matrices stay below 10 − 308 10^{-308} 1 0 − 308 (typically 10 − 400 10^{-400} 1 0 − 400 ).
The output is three 14 × 14 ball matrices g A , g 2 , g 3 g_A,g_2,g_3 g A , g 2 , g 3 whose last row is exactly ( 0 , … , 0 , 1 ) (0,\ldots,0,1) ( 0 , … , 0 , 1 ) , and the coordinate vector v v v of g g g . As a check, g A v = v g_Av=v g A v = v holds within the balls (H 3 H_3 H 3 is holomorphic inside ∣ y ∣ < y 2 |y| < y_2 ∣ y ∣ < y 2 ). The 13 × 13 13 \times13 13 × 13 blocks are the actions on S S S .
The reflection criterion. Let u = ( g 2 − 1 ) v ∈ S u=(g_2-1)v\in S u = ( g 2 − 1 ) v ∈ S .
(a) If the 13 vectors g w u g_wu g w u (w w w words in g A , g 2 , g 3 g_A,g_2,g_3 g A , g 2 , g 3 ) are linearly independent, then S = span { g w u } S=\operatorname{span}\{g_wu\} S = span { g w u } . Each g w u g_wu g w u is a continuation of ( γ 2 − 1 ) H 3 (\gamma_2-1)H_3 ( γ 2 − 1 ) H 3 , a solution of the homogeneous equation L f = 0 Lf=0 L f = 0 , so S ⊆ Sol ( L ) S\subseteq\operatorname{Sol}(L) S ⊆ Sol ( L ) , and loops act on S S S through their classes in π 1 ( C ∖ { 0 , y 1 , … , y 4 } ) \pi_1(\mathbb{C}\setminus\{0,y_1,\ldots,y_4\}) π 1 ( C ∖ { 0 , y 1 , … , y 4 }) . In particular g 2 g_2 g 2 acts on the affine module C H 3 ⊕ S \mathbb{C}H_3\oplus S C H 3 ⊕ S as a conjugate of the local loop at y 2 y_2 y 2 , which is a reflection by Proposition 4.3. Its image is C u \mathbb{C}u C u , so g 2 ∣ S = 1 + u ⊗ ϕ g_2|_S=1+u\otimes\phi g 2 ∣ S = 1 + u ⊗ ϕ for some ϕ ∈ S ∗ \phi\in S^* ϕ ∈ S ∗ .
(b) The row ψ = row i 0 \psi=\text{row }i_0 ψ = row i 0 of ( g 2 − 1 ) ∣ S (g_2-1)|_S ( g 2 − 1 ) ∣ S equals u i 0 ϕ u_{i_0}\phi u i 0 ϕ ; one entry ball excludes 0, so ϕ ≠ 0 \phi\ne0 ϕ = 0 and u i 0 ≠ 0 u_{i_0}\ne0 u i 0 = 0 .
(c) If moreover the 13 covectors ψ g w \psi g_w ψ g w are linearly independent, S S S is irreducible under ⟨ g A , g 2 , g 3 ⟩ \langle g_A,g_2,g_3\rangle ⟨ g A , g 2 , g 3 ⟩ : for an invariant subspace W ≠ 0 W\ne0 W = 0 , either ϕ ( W ) ≠ 0 \phi(W)\ne0 ϕ ( W ) = 0 , and then u = ( g 2 x − x ) / ϕ ( x ) ∈ W u=(g_2x-x)/\phi(x)\in W u = ( g 2 x − x ) / ϕ ( x ) ∈ W for some x ∈ W x\in W x ∈ W , so W ⊇ span { g w u } = S W\supseteq\operatorname{span}\{g_wu\}=S W ⊇ span { g w u } = S ; or ϕ ( W ) = 0 \phi(W)=0 ϕ ( W ) = 0 , and then W ⊆ ⋂ w ker ( ϕ g w ) = 0 W\subseteq\bigcap_w\ker(\phi g_w)=0 W ⊆ ⋂ w ker ( ϕ g w ) = 0 .
Proposition 4.5. For every q q q with 26 ≤ q ≤ 53 26\le q\le53 26 ≤ q ≤ 53 , the 13 × 13 13\times13 13 × 13 determinants of (a) and (c) exclude 0. Hence S S S is irreducible under the monodromy group of L 13 L_{13} L 13 .
Proof (computer-assisted, Arb). The words were chosen greedily on midpoints (a heuristic that does not affect rigour); at every q q q both lists are ∅ \varnothing ∅ , A A A , A 2 A^2 A 2 , A 3 A^3 A 3 , A 4 A^4 A 4 , 3 A 3 3A^3 3 A 3 , A 5 A^5 A 5 , A 3 A 3 A^3A^3 A 3 A 3 , A 6 A^6 A 6 , A 2 3 A 3 A^2 3A^3 A 2 3 A 3 , A 7 A^7 A 7 , A 3 3 A 3 A^3 3A^3 A 3 3 A 3 , A 8 A^8 A 8 (in the notation of the certificate files). The determinants are evaluated in Arb at 2200 bits from the ball entries (fourterm/irred.py); their imaginary parts are listed in Table 1 , with relative radii at most 4 ⋅ 10 − 30 4\cdot10^{-30} 4 ⋅ 1 0 − 30 ; the real parts are balls around 0 of radius at most 2 ⋅ 10 − 219 2\cdot10^{-219} 2 ⋅ 1 0 − 219 . Irreducibility under the subgroup ⟨ g A , g 2 , g 3 ⟩ \langle g_A,g_2,g_3\rangle ⟨ g A , g 2 , g 3 ⟩ implies irreducibility under the whole monodromy group. □
Table 1. Certificates for Theorem 1: the least m m m with M m ( q ) > 0 M_m(q)>0 M m ( q ) > 0 , the margin M 14 ( q ) M_{14}(q) M 14 ( q ) (midpoint of an Arb ball of radius ≤ 5 ⋅ 10 − 8 \le5\cdot10^{-8} ≤ 5 ⋅ 1 0 − 8 , truncated to six decimals), y 2 y_2 y 2 , the base point y ∗ y_* y ∗ , and the midpoints of the two 13 × 13 13\times13 13 × 13 determinants of Proposition 4.5 (code/thm1_cor2/fourterm/summary_4t.json). The jump between q = 37 q=37 q = 37 and q = 38 q=38 q = 38 is a change of scaled coordinates (base point and scaling), not a failure.
Table 1.
Proof of Proposition 4.1, Theorem 1 and Corollary 2. By Lemma 4.4, U a U_a U a is a nonzero monodromy-invariant subspace of S S S , so U a = S U_a=S U a = S by Proposition 4.5; then Lemma 4.2 with u = 13 u=13 u = 13 gives Proposition 4.1, and the reduction above gives Theorem 1. Corollary 2 is the case 26 ≤ q ≤ 32 26\le q\le32 26 ≤ q ≤ 32 of Theorem 1: if θ q = u / v \theta_q=u/v θ q = u / v with u , v ∈ Z u,v\in\mathbb{Z} u , v ∈ Z , v ≠ 0 v\ne0 v = 0 , then ( r 0 , a ) = ( − u , 0 , 0 , v ) (r_0,a)=(-u,0,0,v) ( r 0 , a ) = ( − u , 0 , 0 , v ) is a nonzero relation. □
Remark 4.6 (Independent certificate for Corollary 2). Corollary 2 was first proved by a different final step: for 26 ≤ q ≤ 32 26\le q\le32 26 ≤ q ≤ 32 the 14 continuations of H 3 H_3 H 3 along the words ∅ \varnothing ∅ , 2, 3, A 2 A^2 A 2 , A 3 A^3 A 3 , A 2 2 A^2 2 A 2 2 , A 2 3 A^2 3 A 2 3 , A 3 2 A^3 2 A 3 2 , A 3 3 A^3 3 A 3 3 , A 4 2 A^4 2 A 4 2 , A 4 3 A^4 3 A 4 3 , A 5 2 A^5 2 A 5 2 , A 6 2 A^6 2 A 6 2 , A 7 2 A^7 2 A 7 2 have a nonzero 14 × 14 14\times14 14 × 14 Arb determinant (cert det.py), for example
[ 257906186217.156323475542228360 ± 3.7 ⋅ 10 − 19 ] i ( q = 26 ) , [257906186217.156323475542228360\pm3.7\cdot10^{-19}]i\qquad(q=26), [ 257906186217.156323475542228360 ± 3.7 ⋅ 1 0 − 19 ] i ( q = 26 ) ,
[ 992625.090850805422458198252898 ± 3.9 ⋅ 10 − 25 ] i ( q = 32 ) . [992625.090850805422458198252898\pm3.9\cdot10^{-25}]i\qquad(q=32). [ 992625.090850805422458198252898 ± 3.9 ⋅ 1 0 − 25 ] i ( q = 32 ) .
So dim { γ H 3 } = 14 \dim\{\gamma H_3\}=14 dim { γ H 3 } = 14 , and the elementary analogue of Lemma 4.2 gives the independence of 1 , H 3 , … , ϑ 12 H 3 1,H_3,\ldots,\vartheta^{12}H_3 1 , H 3 , … , ϑ 12 H 3 . (The passage from coordinate vectors to germs uses E ≠ 0 E\ne0 E = 0 .) Independent reruns at q = 26 q=26 q = 26 and q = 32 q=32 q = 32 with initial data that do not use Lemma 3.1 (base b = 2 − 22 b=2^{-22} b = 2 − 22 , the crude bound ( 2 n + 1 ) 2 20 n ∣ μ 3 ∣ / ( q − 1 ) (2n+1)2^{20n}|\mu_3|/(q-1) ( 2 n + 1 ) 2 20 n ∣ μ 3 ∣/ ( q − 1 ) ) and a different discretisation reproduce these determinants to all 30 printed digits, and give the same irreducibility certificates.
Remark 4.7. Smaller holonomy budgets suffice for larger q q q (m min m_{\min} m m i n in Table 1), but the proof uses m = 14 m=14 m = 14 at every q q q . No case analysis on the relation vector and no input from linear forms in logarithms is needed: only a ≠ 0 a\ne0 a = 0 is used.
Widening CDT's range: proof of Theorem 3
CDT's construction
We recall [1 §14] . For an odd integer a a a with ∣ a ∣ ≥ 3 |a|\ge3 ∣ a ∣ ≥ 3 let
A ( a , x ) = ( 1 − 2 a x + x 2 ) − 1 / 2 = ∑ k u k ( a ) x k A(a,x)=(1-2ax+x^2)^{-1/2}=\sum_k u_k(a)x^k A ( a , x ) = ( 1 − 2 a x + x 2 ) − 1/2 = k ∑ u k ( a ) x k
and
H ( a , x ) = A ( a , x ) ∫ 0 x A ( a , t ) d t = ∑ k v k ( a ) x k , H(a,x)=A(a,x)\int_0^x A(a,t)\,\mathrm{d}t=\sum_k v_k(a)x^k, H ( a , x ) = A ( a , x ) ∫ 0 x A ( a , t ) d t = k ∑ v k ( a ) x k ,
so that v k ( a ) / u k ( a ) → 1 2 log a + 1 a − 1 v_k(a)/u_k(a)\to\frac{1}{2}\log\frac{a+1}{a-1} v k ( a ) / u k ( a ) → 2 1 log a − 1 a + 1 . For distinct odd a = 2 m + 1 a=2m+1 a = 2 m + 1 , b = 2 n + 1 b=2n+1 b = 2 n + 1 put λ a = 1 2 ℓ m \lambda_a=\frac{1}{2}\ell_m λ a = 2 1 ℓ m , λ b = 1 2 ℓ n \lambda_b=\frac{1}{2}\ell_n λ b = 2 1 ℓ n , and suppose r 0 + r a λ a + r b λ b + r a b λ a λ b = 0 r_0+r_a\lambda_a+r_b\lambda_b+r_{ab}\lambda_a\lambda_b=0 r 0 + r a λ a + r b λ b + r ab λ a λ b = 0 with r ∈ Z 4 r\in\mathbb{Z}^4 r ∈ Z 4 , ( r a , r b , r a b ) ≠ 0 (r_a,r_b,r_{ab})\ne0 ( r a , r b , r ab ) = 0 . CDT form
P ( x ) = ∑ k ( r 0 u k ( a ) u k ( b ) + r a v k ( a ) u k ( b ) + r b u k ( a ) v k ( b ) + r a b v k ( a ) v k ( b ) ) x k P(x)=\sum_k\left(r_0u_k(a)u_k(b)+r_av_k(a)u_k(b)+r_bu_k(a)v_k(b)+r_{ab}v_k(a)v_k(b)\right)x^k P ( x ) = k ∑ ( r 0 u k ( a ) u k ( b ) + r a v k ( a ) u k ( b ) + r b u k ( a ) v k ( b ) + r ab v k ( a ) v k ( b ) ) x k
(a combination of Hadamard products) and symmetrise: G ( y ) = P ( x ) + P ( x / ( x − 1 ) ) G(y)=P(x)+P(x/(x-1)) G ( y ) = P ( x ) + P ( x / ( x − 1 )) with y = x 2 / ( x − 1 ) y=x^2/(x-1) y = x 2 / ( x − 1 ) [1 Def. 14.1.9] . By [1 Prop. 14.1.11] , G G G has denominator type [ 1 , … , 2 k ] 2 [1,\ldots,2k]^2 [ 1 , … , 2 k ] 2 , satisfies a differential equation L a , b L_{a,b} L a , b whose singular points are 0 0 0 , 4 4 4 , ∞ \infty ∞ and the four points Z 4 ( a , b ) Z_4(a,b) Z 4 ( a , b ) , the y y y -images of the four products x = ( a ± a 2 − 1 ) ( b ± b 2 − 1 ) x=(a\pm\sqrt{a^2-1})(b\pm\sqrt{b^2-1}) x = ( a ± a 2 − 1 ) ( b ± b 2 − 1 ) (plus apparent singularities at the roots of a polynomial R 10 R_{10} R 10 ), and G G G is overconvergent at the image α \alpha α of the smallest product. With the seven “pure” functions B 1 , … , B 7 B_1,\ldots,B_7 B 1 , … , B 7 of [1 §10] and ten functions built from G G G (G G G , G ′ G' G ′ , G ′ ′ G'' G ′′ , G ′ ′ ′ G''' G ′′′ and six primitives), [1 Lemma 14.3.1] gives m = 17 m=17 m = 17 functions, linearly independent over Q ( y ) \mathbb{Q}(y) Q ( y ) , with
σ = ( 0 , 2 , 2 , 4 , … , 4 ) , e = ( 0 , 0 , 1 ; 0 , 0 , 0 , 0 , 0 , 0 ; 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1 ) , \sigma=(0,2,2,4,\ldots,4),\qquad e=(0,0,1;0,0,0,0,0,0;1,1,1,1,1,1,1,1), σ = ( 0 , 2 , 2 , 4 , … , 4 ) , e = ( 0 , 0 , 1 ; 0 , 0 , 0 , 0 , 0 , 0 ; 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1 ) ,
τ = 1136 289 + 78419 242760 = 1032659 242760 = 4.2538 … \tau=\frac{1136}{289}+\frac{78419}{242760}=\frac{1032659}{242760}=4.2538\ldots τ = 289 1136 + 242760 78419 = 242760 1032659 = 4.2538 …
[1 (14.5.2)–(14.5.4)] . The order of the functions is CDT's: B 1 , B 2 , B 3 B_1,B_2,B_3 B 1 , B 2 , B 3 ; B 4 , B 5 , G , G ′ , G ′ ′ , G ′ ′ ′ B_4,B_5,G,G',G'',G''' B 4 , B 5 , G , G ′ , G ′′ , G ′′′ ; B 6 , B 7 B_6,B_7 B 6 , B 7 and the six primitives. The map is φ = h ∘ ψ \varphi=h\circ\psi φ = h ∘ ψ , with
h ( q ) = − 256 q ∏ k ≥ 1 ( 1 + q k ) 24 = − 256 ( η ( 2 τ ) η ( τ ) ) 24 , h(q)=-256q\prod_{k\ge1}(1+q^k)^{24}=-256\left(\frac{\eta(2\tau)}{\eta(\tau)}\right)^{24}, h ( q ) = − 256 q k ≥ 1 ∏ ( 1 + q k ) 24 = − 256 ( η ( τ ) η ( 2 τ ) ) 24 ,
the hauptmodul of Γ 0 ( 2 ) \Gamma_0(2) Γ 0 ( 2 ) in the cusp coordinate q = e 2 π i τ q=e^{2\pi i\tau} q = e 2 π i τ , and ψ : D → D \psi:\mathbb{D}\to\mathbb{D} ψ : D → D holomorphic with ψ − 1 ( 0 ) = { 0 } \psi^{-1}(0)=\{0\} ψ − 1 ( 0 ) = { 0 } ; admissibility of φ \varphi φ is checked through [1 Cor. 9.0.19] . The constant 10 − 6 10^{-6} 1 0 − 6 of [1 Thm. C] enters only through [1 Lemma 14.2.4] (which needs ∣ a / b ∓ 1 ∣ < 1 2 |a/b\mp1|<\frac{1}{2} ∣ a / b ∓ 1∣ < 2 1 ), [1 Lemma 14.2.6] (which needs 0 < ∣ a / b − 1 ∣ < 10 − 3 0<|a/b-1|<10^{-3} 0 < ∣ a / b − 1∣ < 1 0 − 3 ), and the location of the singularities, [1 §14.4] , which determines the admissible maps. We change only the map.
Location of the singular values
Lemma 5.1. Let 0 < ε ≤ 10 − 4 0<\varepsilon\le10^{-4} 0 < ε ≤ 1 0 − 4 and let m ≠ n m\ne n m = n be integers in Z ∖ { − 1 , 0 } \mathbb{Z}\setminus\{-1,0\} Z ∖ { − 1 , 0 } with ∣ m / n − 1 ∣ < ε |m/n-1|<\varepsilon ∣ m / n − 1∣ < ε ; put a = 2 m + 1 a=2m+1 a = 2 m + 1 , b = 2 n + 1 b=2n+1 b = 2 n + 1 . Then the three genuine singular values (the points of Z 4 ( a , b ) Z_4(a,b) Z 4 ( a , b ) other than α \alpha α ) are real with ∣ y ∣ ≥ Y 1 : = 1 / ε − 6 |y|\ge Y_1:=1/\varepsilon-6 ∣ y ∣ ≥ Y 1 := 1/ ε − 6 , and α \alpha α is real with ∣ α ∣ ≤ Y 0 : = ε 4 / 16 |\alpha|\le Y_0:=\varepsilon^4/16 ∣ α ∣ ≤ Y 0 := ε 4 /16 .
Proof. Let A = ∣ a ∣ A=|a| A = ∣ a ∣ , B = ∣ b ∣ B=|b| B = ∣ b ∣ , u = a + sgn ( a ) A 2 − 1 u=a+\operatorname{sgn}(a)\sqrt{A^2-1} u = a + sgn ( a ) A 2 − 1 , v v v likewise; the products are x ∈ { u v , u / v , v / u , 1 / ( u v ) } x\in\{uv,u/v,v/u,1/(uv)\} x ∈ { uv , u / v , v / u , 1/ ( uv )} , all real since A , B ≥ 3 A,B\ge3 A , B ≥ 3 , and y = x 2 / ( x − 1 ) = ( x − 1 ) + 2 + 1 / ( x − 1 ) y=x^2/(x-1)=(x-1)+2+1/(x-1) y = x 2 / ( x − 1 ) = ( x − 1 ) + 2 + 1/ ( x − 1 ) . By the mean value theorem for arccosh \operatorname{arccosh} arccosh , δ : = ∣ log ∣ u ∣ − log ∣ v ∣ ∣ ≤ ∣ A − B ∣ / min ( A , B ) 2 − 1 \delta:=|\log|u|-\log|v||\le|A-B|/\sqrt{\min(A,B)^2-1} δ := ∣ log ∣ u ∣ − log ∣ v ∣∣ ≤ ∣ A − B ∣/ min ( A , B ) 2 − 1 . For positive pairs ∣ A − B ∣ / min ( A , B ) < ε / ( 1 − ε ) |A-B|/\min(A,B)<\varepsilon/(1-\varepsilon) ∣ A − B ∣/ min ( A , B ) < ε / ( 1 − ε ) , for negative pairs ≤ 2 ε ∣ n ∣ / ( 2 ∣ n ∣ ( 1 − ε ) − 1 ) ≤ ε / ( 1 − 1.5 ε ) \le2\varepsilon|n|/(2|n|(1-\varepsilon)-1)\le\varepsilon/(1-1.5\varepsilon) ≤ 2 ε ∣ n ∣/ ( 2∣ n ∣ ( 1 − ε ) − 1 ) ≤ ε / ( 1 − 1.5 ε ) , using ∣ n ∣ ≥ 1 / ε |n|\ge1/\varepsilon ∣ n ∣ ≥ 1/ ε ; and min ( A , B ) ≥ 2 / ε − 3 \min(A,B)\ge2/\varepsilon-3 min ( A , B ) ≥ 2/ ε − 3 , so the square-root factor is at most 1 + ε 2 / 7 1+\varepsilon^2/7 1 + ε 2 /7 . Hence δ ≤ ε ( 1 + 1.6 ε ) \delta\le\varepsilon(1+1.6\varepsilon) δ ≤ ε ( 1 + 1.6 ε ) , and x ∈ { u / v , v / u } x\in\{u/v,v/u\} x ∈ { u / v , v / u } has ∣ x − 1 ∣ ≤ e δ − 1 ≤ ε ( 1 + 3 ε ) |x-1|\le e^\delta-1\le\varepsilon(1+3\varepsilon) ∣ x − 1∣ ≤ e δ − 1 ≤ ε ( 1 + 3 ε ) , so ∣ y ∣ ≥ ( 1 − 3 ε ) / ε − 2 − 2 ε > 1 / ε − 6 |y|\ge(1-3\varepsilon)/\varepsilon-2-2\varepsilon>1/\varepsilon-6 ∣ y ∣ ≥ ( 1 − 3 ε ) / ε − 2 − 2 ε > 1/ ε − 6 . For x = u v x=uv x = uv , ∣ x ∣ ≥ ( 2 min ( A , B ) − 1 ) 2 |x|\ge(2\min(A,B)-1)^2 ∣ x ∣ ≥ ( 2 min ( A , B ) − 1 ) 2 gives y > 1 / ε y>1/\varepsilon y > 1/ ε ; for x = 1 / ( u v ) x=1/(uv) x = 1/ ( uv ) , ∣ α ∣ ≤ 2 / ∣ u v ∣ 2 < ε 4 / 16 |\alpha|\le2/|uv|^2<\varepsilon^4/16 ∣ α ∣ ≤ 2/∣ uv ∣ 2 < ε 4 /16 .
The constant 6 cannot be dropped entirely: CDT’s “∣ y ∣ > 1 / ε |y|>1/\varepsilon ∣ y ∣ > 1/ ε ” fails by O ( 1 ) O(1) O ( 1 ) . At ( m , n ) = ( − 10 6 , − 10 6 − 1 ) (m,n)=(-10^6,-10^6-1) ( m , n ) = ( − 1 0 6 , − 1 0 6 − 1 ) the smallest genuine singular value has ∣ y ∣ = 10 6 − 1.50 |y|=10^6-1.50 ∣ y ∣ = 1 0 6 − 1.50 ; at ε = 3 ⋅ 10 − 6 \varepsilon=3\cdot10^{-6} ε = 3 ⋅ 1 0 − 6 the worst case is 1 / ε − 1.83 1/\varepsilon-1.83 1/ ε − 1.83 , at ( m , n ) = ( − 333333 , − 333334 ) (m,n)=(-333333,-333334) ( m , n ) = ( − 333333 , − 333334 ) . This does not affect [1 Thm. C] , whose map has room.
Admissibility of one map for the whole range
Let φ = h ∘ ψ \varphi=h\circ\psi φ = h ∘ ψ with ψ \psi ψ entire, ψ − 1 ( 0 ) = { 0 } \psi^{-1}(0)=\{0\} ψ − 1 ( 0 ) = { 0 } , and suppose
(a) ∣ ψ ∣ < 1 |\psi|<1 ∣ ψ ∣ < 1 on T \mathbb{T} T ;
(b) ∣ φ ∣ > 2 Y 0 |\varphi|>2Y_0 ∣ φ ∣ > 2 Y 0 on T \mathbb{T} T ;
(c′) ∣ φ ∣ < Y 1 |\varphi|<Y_1 ∣ φ ∣ < Y 1 on T \mathbb{T} T .
By (a) and the maximum principle ψ ( D ‾ ) ⊂ D \psi(\overline{\mathbb{D}})\subset\mathbb{D} ψ ( D ) ⊂ D , so φ \varphi φ is holomorphic on a neighbourhood of D ‾ \overline{\mathbb{D}} D . By (c′), φ ( D ) ⊂ D ( 0 , Y 1 ) \varphi(\mathbb{D})\subset D(0,Y_1) φ ( D ) ⊂ D ( 0 , Y 1 ) , which by Lemma 5.1 misses every genuine singular value of every pair in the range. Since φ \varphi φ vanishes only at 0, simply, and ∣ φ ∣ > 2 Y 0 |\varphi|>2Y_0 ∣ φ ∣ > 2 Y 0 on T \mathbb{T} T , the argument principle shows that every point of U = D ( 0 , 2 Y 0 ) U=D(0,2Y_0) U = D ( 0 , 2 Y 0 ) has exactly one preimage in D \mathbb{D} D ; this is the univalent leaf over U ∋ α U\ni\alpha U ∋ α required by [1 Cor. 9.0.19] , with Y 0 = { α } Y^0=\{\alpha\} Y 0 = { α } and Y 1 Y^1 Y 1 the three genuine singular values. ([1 Cor. 9.0.19] leaves U U U free, subject to being a contractible neighbourhood of 0 in C ∖ { 4 } \mathbb{C}\setminus\{4\} C ∖ { 4 } containing α \alpha α on which G G G is holomorphic; G G G is holomorphic on its principal disc, which contains D ( 0 , 2 Y 0 ) D(0,2Y_0) D ( 0 , 2 Y 0 ) .) So the 17 pullbacks f i ∘ φ f_i\circ\varphi f i ∘ φ are meromorphic on D \mathbb{D} D , for every pair with ∣ m / n − 1 ∣ < ε |m/n-1|<\varepsilon ∣ m / n − 1∣ < ε at once.
The certificate
We take ψ ( z ) = z exp ( ∑ k = 0 96 c k z k ) \psi(z)=z\exp\left(\sum_{k=0}^{96}c_kz^k\right) ψ ( z ) = z exp ( ∑ k = 0 96 c k z k ) with dyadic coefficients c k ∈ 2 − 40 Z c_k\in2^{-40}\mathbb{Z} c k ∈ 2 − 40 Z (the certified object; stored in code/thm3/runs/cert3_p4_C_eps3.0e-06_eps3.0e-06_l3.json), found by an admissibility-aware L-BFGS optimisation of L − T / 17 L-T/17 L − T /17 (started from lune-shaped maps and from CDT’s map). Then L = log ∣ φ ′ ( 0 ) ∣ = log 256 + c 0 L=\log|\varphi'(0)|=\log256+c_0 L = log ∣ φ ′ ( 0 ) ∣ = log 256 + c 0 exactly. We apply Theorem 2.2 with the four maps φ k ( z ) = φ ( r k z ) \varphi_k(z)=\varphi(r_kz) φ k ( z ) = φ ( r k z ) , where
( r 0 , r 1 , r 2 , r 3 ) = ( 1 2048 , 99 256 , 215 256 , 1 ) , ( ϑ 1 , ϑ 2 , ϑ 3 ) = ( 61 200 , 231 200 , 1109 500 ) ; (r_0,r_1,r_2,r_3)=\left(\frac{1}{2048},\frac{99}{256},\frac{215}{256},1\right),\qquad(\vartheta_1,\vartheta_2,\vartheta_3)=\left(\frac{61}{200},\frac{231}{200},\frac{1109}{500}\right); ( r 0 , r 1 , r 2 , r 3 ) = ( 2048 1 , 256 99 , 256 215 , 1 ) , ( ϑ 1 , ϑ 2 , ϑ 3 ) = ( 200 61 , 200 231 , 500 1109 ) ;
the φ k \varphi_k φ k are restrictions of φ \varphi φ , so they are admissible.
Certification (cert3.py, cert3d.py). The circle is cut into dyadic intervals, adaptively. On each interval, an Arb value of log φ \log\varphi log φ at the midpoint (via η \eta η -quotients) and a rigorous bound for ∣ d log φ / d ν ∣ |d\log\varphi/d\nu| ∣ d log φ / d ν ∣ , namely ( 1 + 24 ∑ k k ρ k / ( 1 − ρ k ) ) ⋅ sup ∣ 1 + z g ′ ( z ) ∣ (1+24\sum_k k\rho^k/(1-\rho^k))\cdot\sup|1+zg'(z)| ( 1 + 24 ∑ k k ρ k / ( 1 − ρ k )) ⋅ sup ∣1 + z g ′ ( z ) ∣ with ρ ≥ sup ∣ ψ ∣ \rho\ge\sup|\psi| ρ ≥ sup ∣ ψ ∣ on the interval and g = ∑ c k z k g=\sum c_k z^k g = ∑ c k z k , give an upper step function for log ∣ φ ∣ \log|\varphi| log ∣ φ ∣ , hence an exact rational upper bound for the rearrangement integral of the mixture, and decide (a), (b), (c′). The results at ε = 3 ⋅ 10 − 6 \varepsilon=3\cdot10^{-6} ε = 3 ⋅ 1 0 − 6 :
(a), (b), (c′) hold; max T log ∣ φ ∣ ≤ 13801311154345 / 2 40 = 12.55222 < log Y 1 = 12.71688 \max_T\log|\varphi|\le13801311154345/2^{40}=12.55222<\log Y_1=12.71688 max T log ∣ φ ∣ ≤ 13801311154345/ 2 40 = 12.55222 < log Y 1 = 12.71688 ;
L = 4.7983738127618231 … L=4.7983738127618231\ldots L = 4.7983738127618231 … , mixture integral ≤ 9.08327 \le9.08327 ≤ 9.08327 , penalty 1 17 ∑ ϑ k 2 log ( r k / r k − 1 ) = 2.51411 … / 17 \frac{1}{17}\sum\vartheta_k^2\log(r_k/r_{k-1})=2.51411\ldots/17 17 1 ∑ ϑ k 2 log ( r k / r k − 1 ) = 2.51411 … /17 , so that the numerator is [ 9.23115527305872 ± 1.6 ⋅ 10 − 15 ] [9.23115527305872\pm1.6\cdot10^{-15}] [ 9.23115527305872 ± 1.6 ⋅ 1 0 − 15 ] ;
hence 17 ≤ numerator / ( L − τ ) ≤ 16.95200 17\le\text{numerator}/(L-\tau)\le16.95200 17 ≤ numerator / ( L − τ ) ≤ 16.95200 would be needed: contradiction, with margin L − numerator / 17 − τ ≥ 0.0015378 L-\text{numerator}/17-\tau\ge0.0015378 L − numerator /17 − τ ≥ 0.0015378 .
The margin is thin (0.048 in the function count) but certified, and all slack of the step functions is in the safe direction. Certificates for ε = 2 ⋅ 10 − 6 \varepsilon=2\cdot10^{-6} ε = 2 ⋅ 1 0 − 6 and 2.5 ⋅ 10 − 6 2.5\cdot10^{-6} 2.5 ⋅ 1 0 − 6 (trivial partition; margins ≥ 0.01138 \ge0.01138 ≥ 0.01138 and ≥ 0.00131 \ge0.00131 ≥ 0.00131 ) and a certificate for CDT’s own map at 10 − 6 10^{-6} 1 0 − 6 (margin ≥ 0.0308 \ge0.0308 ≥ 0.0308 ) serve as crosschecks; only the 3 ⋅ 10 − 6 3\cdot10^{-6} 3 ⋅ 1 0 − 6 certificate is needed, since Lemma 5.1 at ε = 3 ⋅ 10 − 6 \varepsilon=3\cdot10^{-6} ε = 3 ⋅ 1 0 − 6 covers every pair below it. An independent non-rigorous sampling with a q-product instead of η \eta η gives numerator ≈ 9.2297 \approx9.2297 ≈ 9.2297 .
The epsilon-dependent steps of CDT's chain
5.5. 5.5. 5.5. The ε \varepsilon ε -dependent steps of CDT’s chain. [1 Lemma 14.2.4] needs ∣ a / b ∓ 1 ∣ < 1 2 |a/b\mp1|<\frac{1}{2} ∣ a / b ∓ 1∣ < 2 1 , and [1 Lemma 14.2.6] needs 0 < ∣ a / b − 1 ∣ < 10 − 3 0<|a/b-1|<10^{-3} 0 < ∣ a / b − 1∣ < 1 0 − 3 ; here ∣ a / b − 1 ∣ < ε ( 1 + 2 ε ) ≤ 10 − 5 |a/b-1|<\varepsilon(1+2\varepsilon)\le10^{-5} ∣ a / b − 1∣ < ε ( 1 + 2 ε ) ≤ 1 0 − 5 . The proof of Lemma 14.2.6 uses only b − a = 2 ( n − m ) ∈ Z ∖ { 0 } b-a=2(n-m)\in\mathbb{Z}\setminus\{0\} b − a = 2 ( n − m ) ∈ Z ∖ { 0 } , so ∣ m − n ∣ ≥ 2 |m-n|\ge2 ∣ m − n ∣ ≥ 2 changes nothing. [1 Lemma 14.3.1] needs a 10 × 10 10\times10 10 × 10 determinant det ( a , b ) \det(a,b) det ( a , b ) to be nonzero. We recomputed it exactly from the operator of [1 Prop. 14.1.11] and the reduction [1 (14.3.2)–(14.3.5)] , with numerators S k S_k S k of the shape [1 (14.3.6)] (one for each basis vector of the unknown Q Q Q ), and proved (Proposition 5.2 below)
det ( a , b ) ( a − b ) 86 = κ a 6 b 6 ( a + b ) 24 Q 1 4 Q 2 4 Q 3 4 Q 4 2 Q 5 2 Φ 4 Φ 6 Φ 14 3 Φ 26 Φ 79 3 , \det(a,b)(a-b)^{86}=\kappa a^6b^6(a+b)^{24}Q_1^4Q_2^4Q_3^4Q_4^2Q_5^2\Phi_4\Phi_6\Phi_{14}^3\Phi_{26}\Phi_{79}^3, det ( a , b ) ( a − b ) 86 = κ a 6 b 6 ( a + b ) 24 Q 1 4 Q 2 4 Q 3 4 Q 4 2 Q 5 2 Φ 4 Φ 6 Φ 14 3 Φ 26 Φ 79 3 ,
with κ = 645700815 / 37778931862957161709568 \kappa=645700815/37778931862957161709568 κ = 645700815/37778931862957161709568 , Q 1 = 4 a 2 − 4 a b + 1 Q_1=4a^2-4ab+1 Q 1 = 4 a 2 − 4 ab + 1 , Q 2 = 4 b 2 − 4 a b + 1 Q_2=4b^2-4ab+1 Q 2 = 4 b 2 − 4 ab + 1 , Q 3 = 4 a 2 − 4 a b + 4 b 2 − 3 Q_3=4a^2-4ab+4b^2-3 Q 3 = 4 a 2 − 4 ab + 4 b 2 − 3 , Q 4 = 16 a 2 − 40 a b + 16 b 2 + 9 Q_4=16a^2-40ab+16b^2+9 Q 4 = 16 a 2 − 40 ab + 16 b 2 + 9 , Q 5 = 80 a 2 − 128 a b + 80 b 2 − 45 Q_5=80a^2-128ab+80b^2-45 Q 5 = 80 a 2 − 128 ab + 80 b 2 − 45 , and Φ d \Phi_d Φ d irreducible of a a a -degree d d d . The factors are handled as follows.
a a a , b b b , a ± b a\pm b a ± b are nonzero, and the Q i Q_i Q i are odd for odd a a a , b b b .
Φ 4 = 80 a 4 − 88 a 3 b + 24 a 2 b 2 − 399 a 2 − 88 a b 3 + 786 a b + 80 b 4 − 399 b 2 + 9 \Phi_4=80a^4-88a^3b+24a^2b^2-399a^2-88ab^3+786ab+80b^4-399b^2+9 Φ 4 = 80 a 4 − 88 a 3 b + 24 a 2 b 2 − 399 a 2 − 88 a b 3 + 786 ab + 80 b 4 − 399 b 2 + 9 is odd for odd a a a , b b b , hence nonzero. (CDT’s Lemma 14.2.6 asserts Res ( R 10 , y ( 4 − y ) ) ≠ 0 \operatorname{Res}(R_{10},y(4-y))\ne0 Res ( R 10 , y ( 4 − y )) = 0 , but its proof treats only Φ 14 \Phi_{14} Φ 14 and Φ 79 \Phi_{79} Φ 79 .)
For Φ 14 \Phi_{14} Φ 14 and Φ 79 \Phi_{79} Φ 79 we use the argument in the proof of CDT’s Lemma 14.2.6, which uses only ∣ b − a ∣ ≥ 1 |b-a|\ge1 ∣ b − a ∣ ≥ 1 and ∣ a / b − 1 ∣ ≤ 10 − 3 |a/b-1|\le10^{-3} ∣ a / b − 1∣ ≤ 1 0 − 3 ; a larger ∣ b − a ∣ |b-a| ∣ b − a ∣ only shrinks the error terms.
Φ 26 \Phi_{26} Φ 26 : Lemma 14.2.4 gives Res ( R 4 , R 10 ) ≠ 0 \operatorname{Res}(R_4,R_{10})\ne0 Res ( R 4 , R 10 ) = 0 by irreducibility and division, for ∣ a / b − 1 ∣ < 1 2 |a/b-1|<\frac{1}{2} ∣ a / b − 1∣ < 2 1 .
Φ 6 \Phi_6 Φ 6 divides none of Δ y ( R 10 ) \Delta_y(R_{10}) Δ y ( R 10 ) , Res y ( R 10 , R 4 ) \operatorname{Res}_y(R_{10},R_4) Res y ( R 10 , R 4 ) , Res y ( R 10 , y ( y − 4 ) ) \operatorname{Res}_y(R_{10},y(y-4)) Res y ( R 10 , y ( y − 4 )) , contrary to the text of [1 ] , so neither lemma covers it. Explicitly,
Φ 6 ( a , b ) = 59290 a 6 b 2 − 185493 a 5 b 3 + 249865 a 4 b 4 − 185493 a 3 b 5 + 59290 a 2 b 6 + 8491 a 5 b − 65541 a 4 b 2 + 119686 a 3 b 3 − 65541 a 2 b 4 + 8491 a b 5 − 6090 a 4 + 15225 a 3 b − 21000 a 2 b 2 + 15225 a b 3 − 6090 b 4 + 975 a 2 − 4260 a b + 975 b 2 + 1755 , \begin{aligned}
\Phi_6(a,b)={}&59290a^6b^2-185493a^5b^3+249865a^4b^4-185493a^3b^5+59290a^2b^6\\
&+8491a^5b-65541a^4b^2+119686a^3b^3-65541a^2b^4+8491ab^5\\
&-6090a^4+15225a^3b-21000a^2b^2+15225ab^3-6090b^4\\
&+975a^2-4260ab+975b^2+1755,
\end{aligned} Φ 6 ( a , b ) = 59290 a 6 b 2 − 185493 a 5 b 3 + 249865 a 4 b 4 − 185493 a 3 b 5 + 59290 a 2 b 6 + 8491 a 5 b − 65541 a 4 b 2 + 119686 a 3 b 3 − 65541 a 2 b 4 + 8491 a b 5 − 6090 a 4 + 15225 a 3 b − 21000 a 2 b 2 + 15225 a b 3 − 6090 b 4 + 975 a 2 − 4260 ab + 975 b 2 + 1755 ,
and Φ 6 ( a , b ) ≡ 16 ( m o d 32 ) \Phi_6(a,b)\equiv16\pmod{32} Φ 6 ( a , b ) ≡ 16 ( mod 32 ) for all odd a , b a,b a , b (check over the 16 × 16 16\times16 16 × 16 odd residue classes mod 32 32 32 ). So Φ 6 ( a , b ) ≠ 0 \Phi_6(a,b)\ne0 Φ 6 ( a , b ) = 0 at every odd pair. (Also ∣ Φ 6 ( a , b ) ∣ / ∣ a ∣ 8 ≥ 1864 |\Phi_6(a,b)|/|a|^8\ge1864 ∣ Φ 6 ( a , b ) ∣/∣ a ∣ 8 ≥ 1864 whenever a ≠ b a\ne b a = b are integers with ∣ a / b − 1 ∣ < 1 / 20 |a/b-1|<1/20 ∣ a / b − 1∣ < 1/20 , so Φ 6 \Phi_6 Φ 6 has no integer zeros near the diagonal.)
Hence, given CDT’s set-up, [1 ] holds for all odd a ≠ b a\ne b a = b with ∣ a / b − 1 ∣ < 10 − 3 |a/b-1|<10^{-3} ∣ a / b − 1∣ < 1 0 − 3 , with no extra hypothesis. The set of irreducible factors agrees with CDT’s printed list; the multiplicities differ, which does not matter. A misprint in [1 ] : the factor “1 + 4 b 2 − 2 a b 1+4b^2-2ab 1 + 4 b 2 − 2 ab ”, printed in both (14.2.2) and (14.2.3), should read 1 + 4 b 2 − 4 a b 1+4b^2-4ab 1 + 4 b 2 − 4 ab .
Proposition 5.2. Given the shape [1 (14.2.1)] of the operator of [1 Prop. 14.1.11] and the reduction [1 (14.3.2)–(14.3.5)] , the displayed identity for det ( a , b ) ( a − b ) 86 \det(a,b)(a-b)^{86} det ( a , b ) ( a − b ) 86 holds in Q ( a , b ) \mathbb{Q}(a,b) Q ( a , b ) , hence at every pair of integers a ≠ b a\ne b a = b .
Proof (computer-assisted, exact). The operator. CDT print only R 4 R_4 R 4 , R 10 R_{10} R 10 and c 3 / c 4 c_3/c_4 c 3 / c 4 of their operator L = ∑ i ≤ 4 c i D i \mathcal{L}=\sum_{i\le4}c_iD^i L = ∑ i ≤ 4 c i D i . We interpolate a candidate L ~ ∈ Q [ a , b , y ] [ D ] \widetilde{\mathcal{L}}\in\mathbb{Q}[a,b,y][D] L ∈ Q [ a , b , y ] [ D ] with c 4 = ( y − 4 ) 2 y 3 R 4 R 10 c_4=(y-4)^2y^3R_4R_{10} c 4 = ( y − 4 ) 2 y 3 R 4 R 10 and verify it: the coefficients of y k y^k y k , k < 63 k<63 k < 63 , of L ~ ( G A ) \widetilde{\mathcal{L}}(G_A) L ( G A ) vanish identically in Q [ a , b ] \mathbb{Q}[a,b] Q [ a , b ] , where G A = ∑ n P n ( a ) P n ( b ) p n ( y ) G_A=\sum_nP_n(a)P_n(b)p_n(y) G A = ∑ n P n ( a ) P n ( b ) p n ( y ) (P n P_n P n Legendre polynomials, p n ( x + w ) = x n + w n p_n(x+w)=x^n+w^n p n ( x + w ) = x n + w n for w = x / ( x − 1 ) w=x/(x-1) w = x / ( x − 1 ) ), and these 63 conditions on the 63 unknowns of CDT’s ansatz have rank 63 (already at ( a , b ) = ( 7 , 5 ) (a,b)=(7,5) ( a , b ) = ( 7 , 5 ) ). So the ansatz has a unique solution over Q ( a , b ) \mathbb{Q}(a,b) Q ( a , b ) , and CDT’s operator is L ~ \widetilde{\mathcal{L}} L ; also c 3 / c 4 c_3/c_4 c 3 / c 4 agrees symbolically with CDT’s printed formula (f1_Lsym.py).
An a priori degree bound. With L ~ \widetilde{\mathcal{L}} L , the 10 numerators S k = y 4 R 10 4 ∑ i a i y i S_k=y^4R_{10}^4\sum_i a_i y^i S k = y 4 R 10 4 ∑ i a i y i obtained from b 3 = ( y − 4 ) 2 R 4 y k / R 10 b_3=(y-4)^2R_4y^k/R_{10} b 3 = ( y − 4 ) 2 R 4 y k / R 10 , k = 0 , … , 9 k=0,\ldots,9 k = 0 , … , 9 (CDT’s (14.3.6) is the case of a general Q = ∑ k q k y k Q=\sum_kq_ky^k Q = ∑ k q k y k , with one numerator S 42 S_{42} S 42 ), are explicit in Q [ a , b , y ] \mathbb{Q}[a,b,y] Q [ a , b , y ] . In the variables t = a − b t=a-b t = a − b and b b b , the leading coefficient of R 10 R_{10} R 10 is − 32 t 4 -32t^4 − 32 t 4 , and the remainders of S k S_k S k modulo R 10 R_{10} R 10 give det ( a , b ) = det M ′ ( t , b ) / t 200 \det(a,b)=\det M'(t,b)/t^{200} det ( a , b ) = det M ′ ( t , b ) / t 200 with M ′ M' M ′ a polynomial matrix. Summing the largest entry degrees column by column gives deg b det M ′ ≤ 690 \deg_b\det M'\le690 deg b det M ′ ≤ 690 (and deg t det M ′ ≤ 685 \deg_t\det M'\le685 deg t det M ′ ≤ 685 ), using the entries only (f1_detsym.py).
The identity. It is equivalent to det M ′ ( t , b ) − κ t 114 Π ( t + b , b ) = 0 \det M'(t,b)-\kappa t^{114}\Pi(t+b,b)=0 det M ′ ( t , b ) − κ t 114 Π ( t + b , b ) = 0 , where κ Π \kappa\Pi κ Π is the right side; this polynomial has b b b -degree at most max ( 690 , deg Π ) = 690 \max(690,\deg\Pi)=690 max ( 690 , deg Π ) = 690 . For each of the 692 integers b 0 = − 346 , … , 345 b_0=-346,\ldots,345 b 0 = − 346 , … , 345 we computed det M ′ ( t , b 0 ) ∈ Q [ t ] \det M'(t,b_0)\in\mathbb{Q}[t] det M ′ ( t , b 0 ) ∈ Q [ t ] exactly (fraction-free elimination, every division checked exact) and found it equal to κ t 114 Π ( t + b 0 , b 0 ) \kappa t^{114}\Pi(t+b_0,b_0) κ t 114 Π ( t + b 0 , b 0 ) (`f1_ identity.py, about 35 minutes). A polynomial of b b b -degree ≤ 690 \le690 ≤ 690 that vanishes on 692 lines b = b 0 b=b_0 b = b 0 is zero. □
An earlier version of this computation interpolated det ( a , b 0 ) ( a − b 0 ) 86 \det(a,b_0)(a-b_0)^{86} det ( a , b 0 ) ( a − b 0 ) 86 on ten lines from 420 values each, which gives the right answer (it is a consequence of Proposition 5.2) but, by the bound above, did not by itself prove it (code/thm3/f1/).
Proof of Theorem 3. A relation for a pair with ∣ m / n − 1 ∣ < 3 ⋅ 10 − 6 |m/n-1|<3\cdot10^{-6} ∣ m / n − 1∣ < 3 ⋅ 1 0 − 6 gives, by [1 Prop. 14.1.11, Lemma 14.3.1] and Section 5.5, 17 functions as in § 5.1 , linearly independent over Q ( y ) \mathbb{Q}(y) Q ( y ) and holonomic. The map of § 5.4 is admissible for them, and Theorem 2.2 gives 17 ≤ 16.952 17\le16.952 17 ≤ 16.952 , a contradiction. □
Pooling pairs: proof of Theorem 4
Suppose there are K K K pairwise inequivalent exceptional classes with representatives ( m k , n k ) (m_k,n_k) ( m k , n k ) in the range. Each gives a G G G -function G k G_k G k and its ten associated functions (the G G G -family of the pair); the pure functions B 1 , … , B 7 B_1,\ldots,B_7 B 1 , … , B 7 do not depend on the pair. The idea is to put all 7 + 10 K 7+10K 7 + 10 K functions into one holonomy bound.
Joint independence
Lemma 6.1. Assume [1 Lemma 14.2.4] . Let ( a , b ) (a,b) ( a , b ) , ( c , d ) (c,d) ( c , d ) be integer pairs with ∣ a ∣ , ∣ b ∣ , ∣ c ∣ , ∣ d ∣ ≥ 2 |a|,|b|,|c|,|d|\ge2 ∣ a ∣ , ∣ b ∣ , ∣ c ∣ , ∣ d ∣ ≥ 2 , a ≠ b a\ne b a = b , c ≠ d c\ne d c = d , ∣ a / b − 1 ∣ < 1 2 |a/b-1|<\frac{1}{2} ∣ a / b − 1∣ < 2 1 , ∣ c / d − 1 ∣ < 1 2 |c/d-1|<\frac{1}{2} ∣ c / d − 1∣ < 2 1 . If Z 4 ( a , b ) ∩ Z 4 ( c , d ) ≠ ∅ Z_4(a,b)\cap Z_4(c,d)\ne\varnothing Z 4 ( a , b ) ∩ Z 4 ( c , d ) = ∅ , then { c , d } = { a , b } \{c,d\}=\{a,b\} { c , d } = { a , b } or { c , d } = { − a , − b } \{c,d\}=\{-a,-b\} { c , d } = { − a , − b } . Moreover 0 , 4 ∉ Z 4 ( a , b ) 0,4\notin Z_4(a,b) 0 , 4 ∈ / Z 4 ( a , b ) .
Proof. Z 4 ( a , b ) Z_4(a,b) Z 4 ( a , b ) is the root set of the quartic R 4 ( a , b ; y ) R_4(a,b;y) R 4 ( a , b ; y ) of [1 (14.2.1)] , which is irreducible over Q \mathbb{Q} Q by [1 Lemma 14.2.4] , with constant term 1 and leading coefficient 4 ( a − b ) 2 4(a-b)^2 4 ( a − b ) 2 . Two irreducible quartics with a common root and constant term 1 are equal. With σ = a 2 + b 2 \sigma=a^2+b^2 σ = a 2 + b 2 and π = a b \pi=ab π = ab the coefficients of y 4 y^4 y 4 and y 3 y^3 y 3 are c 4 = 4 σ − 8 π c_4=4\sigma-8\pi c 4 = 4 σ − 8 π and c 3 = 32 π 2 − 16 π σ + 16 π − 8 σ c_3=32\pi^2-16\pi\sigma+16\pi-8\sigma c 3 = 32 π 2 − 16 π σ + 16 π − 8 σ , and c 3 = − 2 c 4 − 4 π c 4 c_3=-2c_4-4\pi c_4 c 3 = − 2 c 4 − 4 π c 4 identically. So R 4 R_4 R 4 determines π \pi π and σ \sigma σ , hence ( a + b ) 2 (a+b)^2 ( a + b ) 2 and ( a − b ) 2 (a-b)^2 ( a − b ) 2 , hence { a , b } \{a,b\} { a , b } up to a common sign. Finally R 4 ( 0 ) = 1 R_4(0)=1 R 4 ( 0 ) = 1 and R 4 ( 4 ) = ( 16 a 2 − 40 a b + 16 b 2 + 9 ) 2 R_4(4)=(16a^2-40ab+16b^2+9)^2 R 4 ( 4 ) = ( 16 a 2 − 40 ab + 16 b 2 + 9 ) 2 , which is odd (code/thm4/lemmaD.py). □
Lemma 6.2. *Every branch of every function in the G G G -family of ( a , b ) (a,b) ( a , b ) is holomorphic on C ∖ ( { 0 , 4 } ∪ Z 4 ( a , b ) ) \mathbb{C}\setminus(\{0,4\}\cup Z_4(a,b)) C ∖ ({ 0 , 4 } ∪ Z 4 ( a , b )) , and every branch of B 1 , … , B 7 B_1,\ldots,B_7 B 1 , … , B 7 is holomorphic on C ∖ { 0 , 4 } \mathbb{C}\setminus\{0,4\} C ∖ { 0 , 4 } .
Proof sketch. P ( x ) P(x) P ( x ) is a combination of Hadamard products f ⋆ g f\star g f ⋆ g with f ∈ { A ( a , ⋅ ) , H ( a , ⋅ ) } f\in\{A(a,\cdot),H(a,\cdot)\} f ∈ { A ( a , ⋅ ) , H ( a , ⋅ )} and g ∈ { A ( b , ⋅ ) , H ( b , ⋅ ) } g\in\{A(b,\cdot),H(b,\cdot)\} g ∈ { A ( b , ⋅ ) , H ( b , ⋅ )} , each factor continuable along every path avoiding its two singular points. By Hadamard’s multiplication theorem in its global form (deform the contour in f ⋆ g ( x ) = 1 2 π i ∮ f ( t ) g ( x / t ) d t / t f\star g(x)=\frac{1}{2\pi i}\oint f(t)g(x/t)\,\mathrm{d}t/t f ⋆ g ( x ) = 2 π i 1 ∮ f ( t ) g ( x / t ) d t / t ; [2 ] for the star-domain form, and [5 Thm. A] (preprint) for the path-wise form: if f f f continues along every path avoiding a closed discrete set S S S and g g g along every path avoiding T T T , then f ⋆ g f\star g f ⋆ g continues along every path avoiding { 0 } ∪ S ⋅ T \{0\} \cup S \cdot T { 0 } ∪ S ⋅ T , f ⋆ g f \star g f ⋆ g continues along every path avoiding 0 0 0 and the four products. In y y y , the two roots of x 2 − y x + y = 0 x^2-yx+y=0 x 2 − y x + y = 0 are distinct off y ∈ { 0 , 4 } y \in\{0,4\} y ∈ { 0 , 4 } and avoid { 0 } ∪ { products } ∪ { p / ( p − 1 ) } \{0\} \cup\{\text{products}\} \cup\{p/(p-1)\} { 0 } ∪ { products } ∪ { p / ( p − 1 )} exactly off { 0 } ∪ Z 4 \{0\} \cup Z_4 { 0 } ∪ Z 4 , so G = P ( x ) + P ( x / ( x − 1 ) ) G=P(x)+P(x/(x-1)) G = P ( x ) + P ( x / ( x − 1 )) continues along every path in C ∖ ( { 0 , 4 } ∪ Z 4 ) \mathbb{C}\setminus(\{0,4\}\cup Z_4) C ∖ ({ 0 , 4 } ∪ Z 4 ) ; derivatives and primitives inherit this. The B i B_i B i are functions on P 1 ∖ { 0 , 4 , ∞ } \mathbb{P}^1\setminus\{0,4,\infty\} P 1 ∖ { 0 , 4 , ∞ } [1 §10] .
Lemma 6.3 (cross-pair independence). Make the assumptions of Theorem 3. Let ( a k , b k ) (a_k,b_k) ( a k , b k ) , k = 1 , … , K k=1,\ldots,K k = 1 , … , K , be odd pairs with 0 < ∣ a k / b k − 1 ∣ < 10 − 3 0<|a_k/b_k-1|<10^{-3} 0 < ∣ a k / b k − 1∣ < 1 0 − 3 , pairwise inequivalent ({ a j , b j } ≠ ± { a k , b k } \{a_j,b_j\}\ne\pm\{a_k,b_k\} { a j , b j } = ± { a k , b k } for j ≠ k j\ne k j = k ), each with a relation vector as above. Then B 1 , … , B 7 B_1,\ldots,B_7 B 1 , … , B 7 and the K K K G G G -families are linearly independent over C ( y ) \mathbb{C}(y) C ( y ) .
Proof sketch. Suppose ∑ i S i B i + ∑ k ∑ l R k l F k l = 0 \sum_i S_iB_i+\sum_k\sum_l R_{kl}F_{kl}=0 ∑ i S i B i + ∑ k ∑ l R k l F k l = 0 with S i , R k l ∈ C ( y ) S_i,R_{kl}\in\mathbb{C}(y) S i , R k l ∈ C ( y ) , F k l F_{kl} F k l the G G G -family of pair k k k . Fix k k k . CDT’s proof of Lemma 14.3.1 (following Lemma 12.1.1) uses a loop γ k \gamma_k γ k , a conjugate of a small loop around a point s ∈ Z 4 ( a k , b k ) s\in Z_4(a_k,b_k) s ∈ Z 4 ( a k , b k ) , with G k ≠ γ k G k G_k\ne\gamma_kG_k G k = γ k G k . By Lemma 6.1, s ∉ { 0 , 4 } ∪ Z 4 ( a j , b j ) s\notin\{0,4\}\cup Z_4(a_j,b_j) s ∈ / { 0 , 4 } ∪ Z 4 ( a j , b j ) for j ≠ k j\ne k j = k , so by Lemma 6.2 every branch of every F j l F_{jl} F j l (j ≠ k j\ne k j = k ) and of every B i B_i B i is holomorphic at s s s and fixed by γ k \gamma_k γ k . Applying 1 − γ k 1-\gamma_k 1 − γ k leaves ∑ l R k l ( 1 − γ k ) F k l = 0 \sum_lR_{kl}(1-\gamma_k)F_{kl}=0 ∑ l R k l ( 1 − γ k ) F k l = 0 , which is exactly the relation reached in CDT’s single-pair proof (the varied primitives ( 1 − γ k ) ∫ y e ( G k − T ) (1-\gamma_k)\int y^e(G_k-T) ( 1 − γ k ) ∫ y e ( G k − T ) are primitives of y e ( 1 − γ k ) G k y^e(1-\gamma_k)G_k y e ( 1 − γ k ) G k , and CDT’s argument does not depend on the choice of primitive). CDT’s argument, with the determinant of Section 5.5, gives R k l = 0 R_{kl}=0 R k l = 0 . Doing this for every k k k leaves ∑ S i B i = 0 \sum S_iB_i=0 ∑ S i B i = 0 , so S i = 0 S_i=0 S i = 0 [1 Lemma 12.0.1] .
Pairs sharing an entry, ( a , b ) (a,b) ( a , b ) and ( a , c ) (a,c) ( a , c ) , are inequivalent, so Lemma 6.3 covers them. As a falsification test (not used in the proof), the exact rank modulo two primes p > 2008 p>2008 p > 2008 of the union of the series, truncated at y 1000 y^{1000} y 1000 , with polynomial coefficients of degree ≤ D \le D ≤ D , is full for 30 instances (K = 1 , 2 , 3 K=1,2,3 K = 1 , 2 , 3 , including shared entries, pairs in the range and degenerate relation vectors), while duplicated pairs give exactly the expected nullity 10 ( D + 1 ) 10(D+1) 10 ( D + 1 ) (code/thm4/xrank.py, runs/xrank.out).
The pooled holonomy inequality
By Lemma 6.3 we have m = 7 + 10 K m=7+10K m = 7 + 10 K functions, holonomic and linearly independent over Q ( y ) \mathbb{Q}(y) Q ( y ) , with σ = ( 0 , 2 , 2 , 4 , 4 , 4 , 4 ∣ 4 10 K ) \sigma=(0,2,2,4,4,4,4\mid4^{10K}) σ = ( 0 , 2 , 2 , 4 , 4 , 4 , 4 ∣ 4 10 K ) and e = ( 0 , 0 , 1 , 0 , 0 , 1 , 1 ∣ ( 0 4 , 1 6 ) K ) e=(0,0,1,0,0,1,1\mid(0^4,1^6)^K) e = ( 0 , 0 , 1 , 0 , 0 , 1 , 1 ∣ ( 0 4 , 1 6 ) K ) (now in the order B 1 , … , B 7 B_1,\ldots,B_7 B 1 , … , B 7 , then for each pair G , G ′ , G ′ ′ , G ′ ′ ′ G,G',G'',G''' G , G ′ , G ′′ , G ′′′ and the six primitives; for K = 1 K=1 K = 1 this is CDT’s list reordered within the block σ i = 4 \sigma_i=4 σ i = 4 , which changes neither τ ( b ) \tau(b) τ ( b ) nor τ ♯ ( e ) \tau^\sharp(e) τ ♯ ( e ) ). Evaluating τ ♯ \tau^\sharp τ ♯ of [1 (6.0.6)] at the explicit point ξ = 169 / 100 \xi=169/100 ξ = 169/100 (an upper bound for the minimum, hence a valid τ \tau τ ) gives
τ 2 = 1106977681 262702440 = 4.2138081 … , τ 3 = 29601442477 7082066992 = 4.1797744 … \tau_2=\frac{1106977681}{262702440}=4.2138081\ldots,\qquad\tau_3=\frac{29601442477}{7082066992}=4.1797744\ldots τ 2 = 262702440 1106977681 = 4.2138081 … , τ 3 = 7082066992 29601442477 = 4.1797744 …
for K = 2 K=2 K = 2 and K = 3 K=3 K = 3 (tau_pool.py; CDT’s τ \tau τ for K = 1 K=1 K = 1 is reproduced by the same code).
For the maps of this section the disc condition (c′) fails, and we use instead:
Lemma 6.4 (ray lemma). Let φ \varphi φ be holomorphic on a neighbourhood of D ‾ \overline{\mathbb{D}} D and Γ = ( − ∞ , − Y 1 ] ∪ [ Y 1 , ∞ ) \Gamma= (-\infty,-Y_1] \cup[Y_1,\infty) Γ = ( − ∞ , − Y 1 ] ∪ [ Y 1 , ∞ ) . If φ ( T ) ∩ Γ = ∅ \varphi(\mathbb{T}) \cap\Gamma= \varnothing φ ( T ) ∩ Γ = ∅ , then φ ( D ‾ ) ∩ Γ = ∅ \varphi(\overline{\mathbb{D}}) \cap\Gamma= \varnothing φ ( D ) ∩ Γ = ∅ .
Proof. For y 0 ∈ Γ y_0 \in\Gamma y 0 ∈ Γ the number of zeros of φ − y 0 \varphi-y_0 φ − y 0 in D \mathbb{D} D is the winding number of φ ( T ) \varphi(\mathbb{T}) φ ( T ) about y 0 y_0 y 0 . It is constant on each half-line of Γ \Gamma Γ (connected subsets of C ∖ φ ( T ) \mathbb{C} \setminus\varphi(\mathbb{T}) C ∖ φ ( T ) ), and 0 for ∣ y 0 ∣ > max T ∣ φ ∣ |y_0| > \max_{\mathbb{T}}|\varphi| ∣ y 0 ∣ > max T ∣ φ ∣ . □
With Lemma 5.1 (real singular values), (a), (b) and “φ ( T ) ∩ Γ = ∅ \varphi(\mathbb{T}) \cap\Gamma= \varnothing φ ( T ) ∩ Γ = ∅ ” make φ \varphi φ admissible for all pairs in the range simultaneously. The certificate decides φ ( s ) ∉ Γ \varphi(s) \notin\Gamma φ ( s ) ∈ / Γ on each interval either by ∣ φ ∣ < Y 1 |\varphi| < Y_1 ∣ φ ∣ < Y 1 or by a lower bound for ∣ sin arg φ ∣ |\sin\arg\varphi| ∣ sin arg φ ∣ . The results (cert_pool.py, trivial partition, δ max = 0.002 \delta_{\max} = 0.002 δ m a x = 0.002 ):
Table 1 (PDF p. 24).
The 5 ⋅ 10 − 5 5 \cdot10^{-5} 5 ⋅ 1 0 − 5 map was re-optimised for L − T / 27 L-T/27 L − T /27 (reopt.py); the 10 − 5 10^{-5} 1 0 − 5 map is an earlier optimised map dilated by r = 1 − 2 − 14 r = 1 - 2^{-14} r = 1 − 2 − 14 (the undilated map crosses the ray [ Y 1 , ∞ ) [Y_1,\infty) [ Y 1 , ∞ ) near 1 / ε 1/\varepsilon 1/ ε , which the certificate detects); the other two are stored optimised maps. Only the rows 5 ⋅ 10 − 5 5 \cdot10^{-5} 5 ⋅ 1 0 − 5 and 10 − 4 10^{-4} 1 0 − 4 are needed. Lemma 5.1 is proved for ε ≤ 10 − 4 \varepsilon\le10^{-4} ε ≤ 1 0 − 4 ; at ε = 5 ⋅ 10 − 5 \varepsilon= 5 \cdot10^{-5} ε = 5 ⋅ 1 0 − 5 the map has little room (φ ( T ) \varphi(\mathbb{T}) φ ( T ) crosses the negative axis at ≈ − 19988 \approx-19988 ≈ − 19988 , inside Y 1 = 19994 Y_1 = 19994 Y 1 = 19994 ), and an exhaustive high-precision check of Lemma 5.1 on a boundary window found the closest genuine value at Y 1 + 4.25 Y_1 + 4.25 Y 1 + 4.25 .
Proof of Theorem 4. Suppose there were K = 2 K = 2 K = 2 (resp. 3 3 3 ) inequivalent exceptional classes with representatives in ∣ m / n − 1 ∣ < 5 ⋅ 10 − 5 |m/n - 1| < 5 \cdot10^{-5} ∣ m / n − 1∣ < 5 ⋅ 1 0 − 5 (resp. 10 − 4 10^{-4} 1 0 − 4 ). CDT’s construction gives a G G G -function for each, and Lemma 6.3 (applicable since ∣ a / b − 1 ∣ < 10 − 3 |a/b - 1| < 10^{-3} ∣ a / b − 1∣ < 1 0 − 3 ) gives 7 + 10 K 7 + 10K 7 + 10 K independent functions. The certified map is admissible for all of them, and Theorem 2.2 gives 27 ≤ 25.885 27 \le25.885 27 ≤ 25.885 (resp. 37 ≤ 33.169 37 \le33.169 37 ≤ 33.169 ), a contradiction. This proves (i) and (ii); (iii) is Theorem 3 . For (iv): a consecutive pair ( N , N + 1 ) (N,N+1) ( N , N + 1 ) in these ranges has 1 / ( N + 2 ) < 10 − 4 1/(N+2) < 10^{-4} 1/ ( N + 2 ) < 1 0 − 4 , so q = N + 1 ≥ 10 4 > 54 q = N + 1 \ge10^4 > 54 q = N + 1 ≥ 1 0 4 > 54 , and Hata’s theorem gives the four-term independence. □
What is computer-assisted, and how to reproduce it
The folder code/ of the accompanying repository https://github.com/telaviv/log-products-irrationality contains every script and certificate output used above, in one directory per result; code/README.md maps each number of this paper to a script and an output file. Software: Python 3.12, python-flint 0.9.0 (FLINT/Arb ball arithmetic), SymPy 1.14, mpmath 1.3, NumPy and SciPy (optimisation only). The computer-assisted steps are:
Table 1 (PDF p. 25).
The map optimisations (lib-opt.py, lib-range.py, reopt.py) and the falsifier are not part of any proof: they produce candidates that the certificates then check. The referees (§ 9 ) re-ran the main certificates bit for bit on separate copies (among them the budget and irreducibility certificates of Theorem 1 , the rank certificate at q = 29 q = 29 q = 29 , the 3 ⋅ 10 − 6 3 \cdot10^{-6} 3 ⋅ 1 0 − 6 map certificate and the 5 ⋅ 10 − 5 5 \cdot10^{-5} 5 ⋅ 1 0 − 5 pooled certificate), and re-ran the continuation step independently, with different initial data and discretisation, at q = 26 , 32 q = 26, 32 q = 26 , 32 (Remark 4.6; outputs in code/thm1_cor2/runs/referee_rerun/) and at q = 40 , 53 q = 40, 53 q = 40 , 53 (the gate referee’s reruns of Theorem 1 , which reproduce both determinants of Proposition 4.5 to all 30 printed digits; their outputs are not included in code/).
What the computer establishes is, in each case, a finite statement: an identity of explicit polynomials, a determinant ball that excludes 0, or an exact rational upper bound for an integral. What it does not establish: the hand arguments (Lemmas 3.3, 4.2 , 4.4 , the reflection criterion, Lemmas 5.1 , 6.1 , 6.4 , the reductions), the theorems of [1 ] we use, and the correctness of python-flint’s Arb routines (in particular acb.modular eta and matrix determinants), which we take on trust. Arb [4 ] is now part of FLINT [8 ] ; python-flint [9 ] 0.9.0, as used, is linked against FLINT 3.6.0.
Limits
q ≤ 25 q \le25 q ≤ 25 is out of reach of this route. At q = 25 q = 25 q = 25 , M 14 ( 25 ) ≈ − 0.0665 < 0 M_{14}(25) \approx-0.0665 < 0 M 14 ( 25 ) ≈ − 0.0665 < 0 , and 14 is the largest budget available for this module (order 13 plus the constant). Neither the rank nor the irreducibility can help; a new family of forms, a better denominator type or a better map would be needed.
10 − 5 10^{-5} 1 0 − 5 is not reached. For Theorem 3 the best range margin we found at 4 ⋅ 10 − 6 4 \cdot10^{-6} 4 ⋅ 1 0 − 6 is − 0.013 -0.013 − 0.013 (partition form), and even for a single pair the best margin at 10 − 5 10^{-5} 1 0 − 5 is − 0.0086 -0.0086 − 0.0086 [numerical, local optima]. We have no certified ceiling: the only rigorous ceiling we know is far from tight. This agrees with [1 Remark 14.0.4] , which says that the constant could be improved but probably not beyond about 10 − 4 10^{-4} 1 0 − 4 .
Pooling stops near 6 ⋅ 10 − 5 6 \cdot10^{-5} 6 ⋅ 1 0 − 5 for K = 2 K = 2 K = 2 on the maps we explored (margins − 0.009 -0.009 − 0.009 at 6 ⋅ 10 − 5 6 \cdot10^{-5} 6 ⋅ 1 0 − 5 , − 0.031 -0.031 − 0.031 at 7 ⋅ 10 − 5 7 \cdot10^{-5} 7 ⋅ 1 0 − 5 , − 0.069 -0.069 − 0.069 at 10 − 4 10^{-4} 1 0 − 4 , numerical).
The main inherited assumption, the irreducibility of L a , b L_{a,b} L a , b (asserted in CDT’s template Lemma 12.1.1 and implicit in Lemma 14.3.1), is open. Its local system is the push-forward of a rank-2 system along the double cover x ↦ y x \mapsto y x ↦ y ; by Mackey’s criterion it would follow from the irreducibility of the rank-2 system and its non-isomorphism with its conjugate under x ↦ x / ( x − 1 ) x \mapsto x/(x - 1) x ↦ x / ( x − 1 ) [heuristic].
Provenance and use of AI tools
Provenance
The mathematical inputs are: Lysov’s explicit formula for Q n Q_n Q n [6 ] , used as a definition (his Lemmas 7–9 are re-proved above); the holonomy bounds and the construction for products of two logarithms of Calegari, Dimitrov and Tang [1 ] , used as stated; Hata’s theorem [3 ] , used only for the extension to q ≥ 54 q \ge54 q ≥ 54 and for Theorem 4(iv). The scheme of Section 4 is the one suggested in [1 §15.1.10] . The recurrence of Proposition 3.4 was found by modular linear algebra and then proved by creative telescoping; the minimal operators were found the same way. Apart from the preprints cited above ([1 ] , [6 ] and, for Lemma 6.2, [5 ] ), nothing in this paper depends on unpublished work by others.
This work was carried out with substantial assistance from AI systems (Anthropic’s Claude models), which were used to search the literature, to develop the proofs, to write the scripts and produce the certificates, to referee the arguments, and to draft this text. The referee reports were also produced by Claude models, working adversarially on separate copies of the files; they are not a substitute for a human expert’s reading. The author directed the project and is responsible for its content. None of the results is verified in Lean or another proof assistant.
References