Totient sections and tail differences
Put 𝜑(0) =0 and ℕ ={0,1,2,…}. At base two,
𝜑(4𝑛) =2𝜑(2𝑛) and 𝜑(4𝑛 +2) =𝜑(2𝑛 +1) leave five
independent sequences among the seven sections through level two. More
generally, the sections through level 𝑒 ≥1 have the basis 𝜑(𝑛), 𝜑(2𝑛) and 𝜑(2𝑗𝑛 +𝑟) with 1 ≤𝑗 ≤𝑒 and odd 0 <𝑟 <2𝑗, of size 2𝑒 +1. The all-base theorem in the short
paper gives rank 𝑘𝑒 +1 and integral
coordinates. Its retained positive residues satisfy 𝑘 ∤𝑟; coprimality with 𝑘 is unnecessary.
The coefficient relations are finite linear identities. The series
question asks whether a particular infinite weighted sum can be
rational. Put
𝑆:=∑𝑛≥1𝜑(𝑛)2𝑛=12+14+28+216+432+⋯=1.3676308019850223…
The series converges
absolutely because 𝜑(𝑛) ≤𝑛. Erdős Problem #249 asks whether 𝑆 is irrational, which remains open. The
comparisons below test proposed passages from coefficient structure to
the value of this sum. Each retains specified properties while making
the corresponding sum rational. It therefore defeats an argument using
only those properties, not every argument that might also use the
arithmetic of the totient.
Throughout, a cofinal condition means that a witness exists beyond
each threshold. It need not hold at every sufficiently large index. Each
comparison sequence is used only for the properties it preserves;
agreement with a fixed totient prefix, for example, imposes no
multiplicativity condition on the continuation.
Remark 1.1 (Earlier coefficient results). Erdős and Graham
record the irrationality question for 𝑆 [10]. An integer sequence is 𝑘-regular when the ℤ-module generated by its
sections 𝑛 ↦𝑐(𝑘𝑗𝑛 +𝑟), 𝑗 ≥0 and 0 ≤𝑟 <𝑘𝑗, is finitely generated.
Coons proved that 𝜑 is not
𝑘-regular for any 𝑘 ≥2 [2]. Martin’s Theorem 1 implies that the
functions 𝑛 ↦𝜑(𝑎𝑖𝑛 +𝑏𝑖)
are linearly independent for pairwise nonproportional affine forms with
positive integer slopes [4]. The exact ranks in §10.8, together with
the all-base normal form in the short paper, identify the retained
sections and every relation among them. The independence itself is not
claimed as new. The CRT–Dirichlet separation method is already present
in Yazdani’s Theorem 2 and proof, which explicitly credit Shallit [30]; our evaluation matrix
uses one such separation for each column. Yazdani also proves
non-automaticity of 𝜑(𝑛)mod𝑚 for every 𝑚 ≥3 in every
base [30]. Allouche,
Shallit and Yassawi give a survey account [31].
To test rationality, we remove the finite prefix, which becomes an
integer after multiplication by 2𝑁. Write 2𝑁𝑆 =Φ𝑁 +𝑅𝑁 with Φ𝑁 :=∑𝑁𝑛=1𝜑(𝑛)2𝑁−𝑛 ∈ℕ and 𝑅𝑁 :=∑𝑗≥0𝜑(𝑁 +1 +𝑗)/2𝑗+1 ∈[0,∞) (Erdos249257.TotientTailPeriodKiller.totientTail,
two_pow_mul_totient_series_eq).
If 𝑆 =𝑎/(2𝑐𝑣) with 𝑐 ≥0 and 𝑣 ≥1 odd, Euler’s theorem gives ℎ =𝜑(𝑣) >0 and 𝑁0 =𝑐 such that 𝑅𝑁+ℎ −𝑅𝑁 ∈ℤ for every 𝑁 ≥𝑁0 (Erdos249257.TotientTailPeriodKiller.eventual_period_of_not_irrational).
Indeed,
𝑅𝑁+ℎ−𝑅𝑁≡2𝑁(2ℎ−1)𝑆(mod)𝑍,
and the right side is integral for
those choices.
Thus it suffices, for every positive shift ℎ, to find arbitrarily large 𝑁 for which 𝑅𝑁+ℎ −𝑅𝑁 is not an integer. A finite
test is obtained by truncating this difference. Throughout, 𝑎mod𝑚 denotes the least nonnegative
residue, and an unqualified index in a sum is a nonnegative integer. For
ℎ,𝑁,𝐿 ∈ℕ, write
𝐷(ℎ,𝑁,𝐿):=∑𝑗<𝐿(𝜑(𝑁+ℎ+1+𝑗)−𝜑(𝑁+1+𝑗))2𝐿−1−𝑗∈ℤ,
the
depth-𝐿 truncation of 2𝐿(𝑅𝑁+ℎ −𝑅𝑁), and
C(ℎ,𝑁,𝐿):⟺(𝑁+ℎ+𝐿+2)<𝐷(ℎ,𝑁,𝐿)mod2𝐿<2𝐿−(𝑁+ℎ+𝐿+2),
a decidable condition (Erdos249257.TotientTailPeriodKiller.windowDiscrepancy,
Erdos249257.TotientTailPeriodKiller.certifiedKill).
The omitted tails satisfy the exact identity
2𝐿(𝑅𝑁+ℎ−𝑅𝑁)−𝐷(ℎ,𝑁,𝐿)=𝑅𝑁+ℎ+𝐿−𝑅𝑁+𝐿.
Both tails are nonnegative, so their difference is bounded in absolute
value by the larger bound 𝐵 =𝑁 +ℎ +𝐿 +2, not the sum of their bounds.
The two strict inequalities in C(ℎ,𝑁,𝐿) keep the resulting error interval away from every
multiple of 2𝐿; hence 𝑅𝑁+ℎ −𝑅𝑁 ∉ℤ. Conversely,
a fixed noninteger difference has a fixed positive distance 𝛿 from ℤ. Since 2𝐿𝛿 >2𝐵 eventually, some depth
detects it. Completeness concerns detection of an existing difference,
not existence of the differences required below (Erdos249257.TotientTailPeriodKiller.exists_certifiedKill_iff_tail_diff_notMem_int).
The resulting quantified condition is the following.
Definition 1.2 (The quantified residue condition).
Sep:⟺∀ℎ≥1 ∀𝑁0 ∃𝑁≥𝑁0 ∃𝐿,C(ℎ,𝑁,𝐿).
Sep ⟺𝑆 ∉ℚ (irrational_totient_series_iff_certificate_supply).
The equivalence identifies the missing quantified input; it supplies no
certificate by itself.
One successful (ℎ,𝑁,𝐿) excludes
reduced denominators dividing 2𝑁(2ℎ −1). To exclude an arbitrary
rational value, its odd denominator first determines a shift and its
power of two determines a threshold; only then may we choose the witness
𝑁,𝐿. This is the order in Sep. The depth must grow with
𝑁, since success requires 2(𝑁 +ℎ +𝐿 +2) <2𝐿. A large enough depth
makes the interval test possible, but does not make the actual residue
lie inside it. A failed test at one depth likewise does not prove
integrality. Erdos249257.TotientTailPeriodKiller.certifiedKill_depth_floor
Unconditional results
Reduced coefficients