Alternating signs can make a series converge even when the magnitudes alone have an infinite sum. The alternating-series test explains when successive positive and negative contributions settle toward a limit, and it gives a practical error bound for partial sums.
Check the magnitudes, not just the signs
Write an alternating series as where The alternating-series test guarantees convergence if decreases eventually and
The shrinking magnitudes make successive partial sums overshoot and undershoot by progressively smaller amounts. Their gap tends to zero, forcing both sides toward the same limit. Alternation alone is insufficient: has nonshrinking terms and divergent partial sums.
If decrease begins only after a finite number of terms, convergence is unaffected. However, an error bound for a particular partial sum requires the decreasing-tail condition to hold from the relevant point onward.
Example: the alternating harmonic series
For the magnitudes decrease to zero. Therefore the series converges. The corresponding absolute-value series diverges.
This is conditional convergence: the signed series converges, but the series of magnitudes does not. By contrast, converges absolutely because is a convergent p-series.
Absolute convergence always implies convergence of the original series. Conditional convergence requires two separate conclusions: convergence with signs and divergence after taking absolute values. Do not label a series conditional merely because its terms alternate.
The next omitted term bounds the error
For a decreasing-to-zero alternating series, if includes the first N terms, then
The error has the sign of the next omitted term. The true sum lies between two consecutive partial sums. This gives both an accuracy guarantee and a direction for the approximation.
The index is easy to mishandle. If the series begins at equals zero, “N terms” ends at index so identify the next actual term rather than blindly inserting N into a formula.
Example: choose enough terms for a tolerance
For the alternating harmonic series, the error after N terms is at most To guarantee error less than 0.01, require which means Thus is sufficient.
For error at most 0.01, is sufficient because equality is allowed. The distinction between “less than” and “at most” changes the smallest guaranteed term count.
A bound is not an exact error. The actual error may be substantially smaller, but the guarantee must follow from the inequality you can prove.
Example: certify a numerical estimate
For the first three terms sum to The next term is so the true sum lies between and
The three-term estimate is high because the next omitted term is negative. The maximum error from this test is You can report an interval rather than an unjustified string of decimal digits.
Check monotonicity honestly
A magnitude formula may approach zero without decreasing everywhere. Use a derivative of the related positive function or compare with to establish eventual decrease. If the conditions fail, the test is inconclusive, not a proof of divergence.
Try it yourself
Classify and bound the error after 400 terms.
Answer: The magnitudes decrease to zero, so the alternating series converges. Its absolute-value p-series has and diverges, so convergence is conditional. The error is at most and the next term is positive, making the 400-term sum an underestimate.
What to remember
Verify decrease and a zero magnitude limit. Test absolute values separately, and use the next omitted term for the alternating error bound and its direction.