Letter to the Editor

In my paper [1] I prove

[( )𝑚 ( )𝑚 ( )𝑚 ]
1 2 𝑚−1 1
lim + + ⋅⋅⋅ + = . (1)
𝑚→∞ 𝑚 𝑚 𝑚 𝑒−1
This result can be generalized. For any fixed integer 𝑘.
[( )𝑚 ( )𝑚 )𝑚 ]
𝑒𝑘+1
(
1 2 𝑚+𝑘
lim + + ⋅⋅⋅ + = .
𝑚→∞ 𝑚 𝑚 𝑚 𝑒−1
A simple way to prove this is to observe that, for 𝑘 ≥ 0,
)𝑚 )𝑚 ]
𝑒𝑘+1 − 1
[( ) ( (
𝑚 𝑚 𝑚+1 𝑚+𝑘
lim + + ⋅⋅⋅ + = 1 + 𝑒 + . . . + 𝑒𝑘 = , (2)
𝑚→∞ 𝑚 𝑚 𝑚 𝑒−1
and then sum Equations (1) and (2). A similar argument holds for 𝑘 < 0.

There is also a mistake in my paper, first communicated to the editors by Vito Lampret. Near
the end of the paper I wish to show that
𝑚 [ ] ∑ ∞
∑ 𝐵𝑘 1−𝑘 𝐵𝑘
lim 𝑚 (𝑚(𝑚 − 1) ⋅ ⋅ ⋅ (𝑚 − 𝑘 + 2)) = . (3)
𝑚→∞
𝑘=1
𝑘! 𝑘=1
𝑘!

First, I express the left-hand side of Equation (3) as

𝑚 [ ( )]
∑ 𝐵𝑘 1
lim 1+𝑂 .
𝑚→∞
𝑘=1
𝑘! 𝑚

This is correct. However, the big-𝑂 notation disguises the fact that the implicit constant in
𝑂(1/𝑚) is dependent on 𝑘. Since 𝑘 ranges from 1 to 𝑚 over the sum, the maximum constant for
the 𝑂(1/𝑚) expressions might depend on 𝑚, with the result that the maximum of the 𝑂(1/𝑚)
expressions might not actually be 𝑂(1/𝑚). Thus, a few lines later, it is invalid to make the claim
𝑚 [ ( )] ( )∑ 𝑚
∑ 𝐵𝑘 1 1 𝐵𝑘
lim 𝑂 = lim 𝑂 = 0,
𝑚→∞
𝑘=1
𝑘! 𝑚 𝑚→∞ 𝑚 𝑘=1 𝑘!

even though ∞
∑
𝑘=1 𝐵𝑘 /𝑘! converges.
However, this can be corrected fairly easily. Pick 𝜖 > 0 and find 𝑝 such that 𝑘>𝑝 ∣𝐵𝑘!𝑘 ∣ < 𝜖.
∑
Then write
𝑚 [ ]
∑ 𝐵𝑘 1−𝑘
lim 𝑚 (𝑚(𝑚 − 1) ⋅ ⋅ ⋅ (𝑚 − 𝑘 + 2))
𝑚→∞
𝑘=1
𝑘!
𝑝 [ ]
∑ 𝐵𝑘 1−𝑘 ∑ [ 𝐵𝑘 ]
1−𝑘
= lim 𝑚 (𝑚(𝑚 − 1) ⋅ ⋅ ⋅ (𝑚 − 𝑘 + 2)) + lim 𝑚 (𝑚(𝑚 − 1) ⋅ ⋅ ⋅ (𝑚 − 𝑘 + 2)) .
𝑚→∞
𝑘=1
𝑘! 𝑚→∞
𝑘>𝑝
𝑘!

The sum in the first term to the right of the equals sign has only
∑∞𝑝 terms in it, and so the method
in [1] is valid for this term. Thus the first term is within 𝜖 of 𝑘=1 𝐵𝑘 /𝑘!. Also, it is easy to see
that the second term is within 𝜖 of 0. Equation (3) follows.

1. Michael Z. Spivey, The Euler-Maclaurin formula and sums of powers, this Magazine, 79
(2006), 61–65.