What happens to the sum when we reorder a conditionally convergent series?
Positive terms and negative terms both pile up without end. So choosing which side to draw from first pushes the sum around. Nothing like this happens when the absolute series converges.
For example, think of putting π° coins into a π· piggy bank and taking them out. If both piles of coins are endless, the order alone changes the balance.
n=1βββn(β1)n+1β=ln2(+,+,β)β23βln2
Examples & Exercises
Verify numerically that the (+,+,β) rearrangement's partial sums tend to 23βln2β1.0397.
(Numerical check; rigorous proof omitted.)
23βln2β1.03972
n=150: the rearranged partial sums settle near 1.0397.Show hint
Raising n to 300 shrinks the error to about 0.0025 β the direction of convergence is consistent, not a coincidence.
The general formula is stated as known (idea level): a rearrangement repeating p positive and q negative terms sums to ln2+21βlnqpβ.
For p=2,q=1 this gives 23βln2.
Its proof is not covered here.
Explain at idea level why rearranging an absolutely convergent series never changes its sum. (Idea level; proof omitted.)
If tails can be made arbitrarily small, order stops mattering.Show hint
If βanβ converges absolutely, then for every Ξ΅>0 there is an N with tail βn>Nββ£anββ£<Ξ΅. (Sketch only; rigorous proof omitted)
A rearrangement's partial sums eventually contain the first N terms of the original, so the two partial sums differ only by tail terms β less than 2Ξ΅.
Since Ξ΅ is arbitrary, the two sums agree: when tails can be made arbitrarily small, reordering cannot change the sum.
Prove β£Snββln2β£β€n+11β for the alternating harmonic partial sums.
β£Rnββ£β€an+1β=n+11β
n=100: the error is trapped inside 1/101β0.0099.Show proof
anβ=1/n decreases to 0. Even partial sums increase, odd ones decrease: S2m+2ββS2mβ=a2m+1ββa2m+2β>0 and S2m+3ββS2m+1β=β(a2m+2ββa2m+3β)<0.
Now justify the sandwich.
The even sums (S2mβ) increase and are bounded above by every odd sum, and the odd sums (S2m+1β) decrease and are bounded below by every even sum.
The gap is S2m+1ββS2mβ=a2m+1ββ0, so both subsequences converge to a common limit S (=ln2), with S2mββ€Sβ€S2m+1β for all m.
Hence for every n the limit S=ln2 lies between two consecutive partial sums Snβ and Sn+1β, so β£SβSnββ£β€β£Sn+1ββSnββ£=an+1β=n+11β.
For n=100 this gives β£S100ββln2β£β€1011ββ0.0099 β matching the caption's figure.