Proof Strategy for Rearrangement

Two Claims, Not One

The Riemann Rearrangement Theorem shows that a conditionally convergent series can be ordered to approach a prescribed real number. The key proof lesson is that this kind of construction must establish two logically separate claims. First, the proposed ordering must be a permutation: every original term occurs exactly once. Second, the partial sums in that ordering must converge to the claimed value. A convincing argument for one claim does not supply the other.

This separation is useful even when the order is specified in advance. For example, a sequence of partial sums might approach a target while the construction has omitted infinitely many terms. Conversely, a list might be a valid permutation without its partial sums converging. The proof strategy is therefore to establish the indexing facts first and then analyze the sums.

The first three steps are about the index map, not about the values of the terms. The last step is about the values and their order. Keeping these tasks distinct makes it easier to locate a gap: an argument about small terms or small overshoots may establish convergence, but it does not show that every index was used.

A Fair Interleaving Principle

Many rearrangement constructions divide the original indices into several disjoint lists, or queues, and then take terms from those queues according to a schedule. A schedule is useful only if it does not neglect a queue forever. The diagonal schedule below gives a systematic way to interleave countably many queues while ensuring that every position in every queue is eventually reached.

Proof. A pair \((j,r)\) appears exactly once in the diagonal listing: it lies on the diagonal with sum \(j+r\), and within that diagonal its position is fixed by the tie rule. If the pair is used, its output is the single index \(q_{j,r}\). Since the queues are disjoint and each queue lists its members without repetition, no output index can occur twice.

To prove coverage, let \(n\) be any positive integer. Because the sets \(Q_j\) partition the positive integers, \(n\) belongs to a unique queue \(Q_j\). It has a finite position \(r\) in that queue, so \(n=q_{j,r}\). The pair \((j,r)\) appears after finitely many pairs, and it is not skipped because \(q_{j,r}\) is defined. Thus \(n\) occurs in the output list. Every positive integer occurs exactly once, so the list is a permutation. \(\square\)

The schedule need not serve each queue at equal intervals. What matters is the fairness built into the diagonal order: every fixed pair \((j,r)\) is reached after finitely many steps. If a proof uses a different schedule, it must verify the corresponding fairness property rather than assume it.

A Finite-Initial-Segment Test for Coverage

For constructions not naturally expressed as queues, it is often convenient to track how far the list has progressed through the original indices. The following criterion turns eventual coverage of finite initial segments into a concise proof of surjectivity.

Proof. The hypothesis that no value occurs more than once says that \(\pi\) is injective. To show that it is onto, take any positive integer \(n\). Apply the coverage hypothesis with \(N=n\). There is a finite \(K_n\) by which every index from \(1\) through \(n\) has appeared, so in particular \(n=\pi(k)\) for some \(k\leq K_n\). Since this holds for every \(n\), \(\pi\) is onto. It is both injective and onto, and therefore is a permutation. \(\square\)

A useful way to apply the criterion is to find a stage bound: after how many selections can we guarantee that every index at most \(N\) has appeared? The bound need not be sharp. It needs only to be finite for each fixed \(N\). This is often easier than describing the exact position at which each individual index appears.

Worked Examples: Auditing the Index Construction

Why Convergence Is Not Enough

Suppose a proposed ordering produces partial sums converging to \(L\). That statement concerns only the terms that were actually selected. It does not imply that the list contains every original index. For example, listing only the even-indexed terms of a convergent series may produce a convergent series of partial sums, but it is a subseries, not a rearrangement of the original series. A proof of convergence cannot repair a failure of coverage.

The converse distinction matters too. Once the index sequence is shown to be a permutation, convergence does not follow automatically for a conditionally convergent series. In that setting the order affects the partial sums; the Riemann Rearrangement Theorem supplies a specific construction and a separate convergence argument. For an absolutely convergent series, the Rearrangement Invariance Theorem gives the sum after the permutation property is established.

In constructions involving positive and negative terms, zeros, or several other categories, keep an explicit record of what is still unused. If the construction takes terms in blocks, check that every block is finite and that the successive blocks exhaust each category. If there are infinitely many categories, use a fair schedule or provide another argument that every fixed category and every fixed position within it is eventually reached. These checks address the indexing claim; estimates on block endpoints address the limit claim.