The Unifying Idea

Arithmetic and geometric series are usually taught as separate objects with separate formulas. But they are structurally the same thing viewed through different lenses. The logarithm converts one into the other, and every formula has a twin on the other side.

This note records five observations that all follow from asking: what does this formula become when the underlying structure changes?

1. The Logarithmic Duality

If \(a_1, a_2, a_3, \ldots\) is a geometric series with ratio \(r > 0\), then \(\log a_1, \log a_2, \log a_3, \ldots\) is an arithmetic series with common difference \(\log r\).

Proof. \(a_k = a_1 \cdot r^{k-1}\), so \(\log a_k = \log a_1 + (k-1)\log r\), which is linear in \(k\).

This is why logarithms were invented by Napier in 1614: to convert multiplications (geometric operations) into additions (arithmetic ones). A slide rule is the physical embodiment of this map. In modern terms, \(\log\) is a group isomorphism \((\mathbb{R}^+, \times) \to (\mathbb{R}, +)\), and every arithmetic fact pulls back through it to a geometric fact.

2. The Section Formula and Its Geometric Twin

Given two terms \(t_m\) and \(t_n\) of an arithmetic series, the term \(t_k\) dividing the gap in ratio \((k-m):(n-k)\) is:

\[ t_k = \frac{(n-k),t_m + (k-m),t_n}{n-m} \]

This is a weighted arithmetic mean. Now apply it to \(\log a_m\) and \(\log a_n\) and exponentiate back:

\[ \log a_k = \frac{(n-k)\log a_m + (k-m)\log a_n}{n-m} = \log!\left(a_m^{,n-k} \cdot a_n^{,k-m}\right)^{!\frac{1}{n-m}} \]

\[ \boxed{a_k = \left(a_m^{,n-k} \cdot a_n^{,k-m}\right)^{!\frac{1}{n-m}}} \]

This is a weighted geometric mean - the natural twin of the section formula.

Verification. Given \(a_1 = 2, a_5 = 32\) (series: 2, 4, 8, 16, 32), find \(a_3\):

\[ a_3 = \left(2^{5-3} \cdot 32^{3-1}\right)^{1/4} = (4 \cdot 1024)^{1/4} = 4096^{1/4} = 8 \checkmark \]

The midpoint case is the familiar AM-GM inequality in disguise.

3. The Sum Formula as Middle Term × n

The arithmetic sum formula \(S_n = \frac{n}{2}(t_1 + t_n)\) can be read as:

\[ S_n = \underbrace{\frac{t_1 + t_n}{2}}_{\text{middle term}} \times; n \]

Why every end-pair sums equally. For any offset \(j\):

\[ t_{1+j} + t_{n-j} = (t_1 + jd) + (t_n - jd) = t_1 + t_n \]

The \(d\)-contributions cancel exactly because the two indices are displaced equally in opposite directions. So the sum of \(n/2\) such pairs gives \(S_n = \frac{n}{2}(t_1 + t_n)\).

Why the mean equals the middle term. An arithmetic series is a linear function sampled at uniformly spaced integer indices - a double uniformity. For any linear function, the average value over a symmetric interval equals the value at the midpoint:

\[ \frac{1}{n}\sum_{k=1}^{n}(t_1 + (k-1)d) = t_1 + \frac{n-1}{2}d = t_{\text{mid}} \]

This is a discrete analogue of: \(\frac{1}{b-a}\int_a^b f,dx = f!\left(\frac{a+b}{2}\right)\) for linear \(f\).

Geometric series breaks this because terms grow exponentially in the index - the distribution is skewed, end-pairs don't sum equally, and mean ≠ middle term.

4. RAGAS - Recursive Approach for Geometric and Arithmetic Sequences

The name borrows from rāga (Sanskrit for "to color"), the Indian classical framework where a performance unfolds by recursively elaborating a seed phrase in stages - a process called vistār ("expansion"). A raga also carries a small, fixed set of generating rules (its pakad, or characteristic phrases) that regenerate the same melodic identity at every register and tempo, whether played slow or fast. The equation below has the same two properties:

\[ t_n^{(\ell)} = t_1^{(\ell)} + S_{n-1}^{(\ell+1)} \]

it builds each level from a seed term plus an expansion of the level beneath it, and the operators connecting levels (\(\Delta\) going down, \(S\) going up) are fixed and reused at every floor of the hierarchy - the recursion rule, not the specific numbers, is what carries the structure's identity, in both music and this sequence hierarchy.

Any sequence satisfies the telescoping identity:

\[ t_n = t_1 + \sum_{k=1}^{n-1}(t_{k+1} - t_k) \]

Define the difference sequence \(\Delta t\) by \(\Delta t(k) = t_{k+1} - t_k\). Then the sum above is exactly \(S_{n-1}\) of \(\Delta t\). So, the \(n\)-th term of any sequence equals its first term plus the partial sum of the sequence one level down. This creates a natural hierarchy:

Example 1. Sequence 1, 5, 11, 19 - differences are 4, 6, 8 (arithmetic, \(d=2\)):

\[ t_n = 1 + S_{n-1}^{\mathrm{arith}} = 1 + \frac{n-1}{2}\Bigl(8 + (n-2) \cdot 2\Bigr) = n^2 + n - 1 \]

Example 2. Sequence 2, 5, 14, 41 - differences are 3, 9, 27 (geometric, \(r=3\)):

\[ t_n = 2 + S_{n-1}^{\mathrm{geom}} = 2 + 3 \cdot \frac{3^{n-1}-1}{2} = \frac{3^n + 1}{2} \]

This is the discrete Fundamental Theorem of Calculus:

\[ f(x) = f(a) + \int_a^x f'(u),du \quad\longleftrightarrow\quad t_n = t_1 + \sum_{k=1}^{n-1}(\Delta t)_k \]

The difference operator \(\Delta\) is discrete differentiation; \(S_{n-1}\) is discrete integration. The sum formula at each level is the integration rule connecting one floor of the hierarchy to the floor above.

Note: Geometric sequences are special - \(\Delta(r^k) = (r-1)r^k\) is again geometric, so they are never annihilated by \(\Delta\). This mirrors the fact that \(e^x\) is its own derivative: the geometric sequence is the discrete eigenfunction of \(\Delta\), just as \(e^x\) is the eigenfunction of \(d/dx\).

5. The Gap Between Discrete Sums and Continuous Integrals

Comparing \(\sum_{k=1}^n f(k)\) with \(\int_0^n f(x),dx\):

The leading term always matches the integral. The discrete sum always carries extra positive correction terms. But shouldn't the continuous integral - the "full area" - be larger?

Why the discrete sum wins for increasing functions. The sum \(\sum_{k=1}^n f(k)\) uses right-endpoint rectangles of width 1. For an increasing function, the right endpoint is the maximum on each unit interval \([k-1, k]\), so each bar overshoots the curve:

\[ \int_{k-1}^{k} f(x),dx \leq f(k) \cdot 1 \]

Summing from \(k=1\) to \(n\):

\[ \int_0^n f(x),dx \leq \sum_{k=1}^{n} f(k) \]

For a decreasing function the inequality flips and the integral wins.

The exact bridge: Euler-Maclaurin formula.

\[ \sum*{k=1}^{n} f(k) = \int_1^n f(x),dx + \frac{f(1)+f(n)}{2} + \sum*{j=1}^{p} \frac{B_{2j}}{(2j)!}\Bigl[f^{(2j-1)}(n) - f^{(2j-1)}(1)\Bigr] + R_p \]

The term \(\frac{f(1)+f(n)}{2}\) is the trapezoid correction - it shaves the triangle off the top of each overshooting bar. The \(B_{2j}\) are Bernoulli numbers (\(B_2 = \frac{1}{6}\), \(B_4 = -\frac{1}{30}\), …), which alternate in sign and encode curvature corrections of successively higher order.

Verification for \(f(k) = k\). Since \(f' = 1\) everywhere, all derivative corrections vanish:

\[ \sum_{k=1}^n k = \int_1^n x,dx + \frac{1+n}{2} = \frac{n^2-1}{2} + \frac{n+1}{2} = \frac{n^2+n}{2} \checkmark \]

The general power-sum formula (Faulhaber) is:

\[ \sum_{k=1}^{n} k^p = \frac{n^{p+1}}{p+1} + \frac{n^p}{2} + \frac{p}{12}n^{p-1} - \cdots \]

The leading term \(\frac{n^{p+1}}{p+1}\) is always \(\int_0^n x^p,dx\). Everything after is Bernoulli correction.

The parallel universe table.

The Euler-Maclaurin formula is the dictionary between the two worlds. The Bernoulli numbers are the exchange rate. As step size shrinks to zero, corrections vanish and the sum collapses into the integral - the passage from series to calculus.

The Meta-Principle

Every observation above followed the same move: take a formula that works in one structured world, ask what it becomes in a parallel world connected by a structure-preserving map, then ask what it generates when applied recursively or taken to a limit.

Transport. Recursion. Limit. These three questions are the engine of mathematical generalization.