Skip to content

A power of a logarithm is read as one of c x^r before its derivative is simplified - #1407

Merged
Rafael-SOWNet merged 1 commit into
masterfrom
logarithm-linears-guard
Sep 17, 2026
Merged

Rafael-SOWNet merged 1 commit into
masterfrom
logarithm-linears-guard

Conversation

@Rafael-SOWNet

@Rafael-SOWNet Rafael-SOWNet commented Sep 17, 2026 •

Copy link
Copy Markdown
Member

Found by re-sampling Rubi's family 7 on today's master: atanh(tanh(a + b x))^4 and acoth(coth(a + b x)), 300 ms and 41 ms answers on 2026-09-16, were timeouts. Bisected to #1379: IsAffineInALogarithmOfAPowerOfX simplified the logarithmic derivative of any single logarithm's argument to see whether it was r/x, and here the argument is a quotient of exponentials whose derivative the simplifier turns over for minutes (past 190 s on the probe) on the way to no.

Change. The argument is read first — products, quotients and powers of x and of constants, however nested (c (d x^m)^n included) — and anything else is not this shape without a derivative being taken. The linear reader in SolveByTheQuotientOfTheLogarithmsLinears gets the same guard: a leaf with a function of x in it is not a linear however it simplifies, so no Simplify is spent on it. Both are declines made cheap; no answer changes.

Measured. atanh(tanh(a + b x))^4 300 ms, acoth(coth(a + b x)) 41 ms, both differentiating back (two rows added to LogarithmOfARationalFunctionByPartsTest, bounded by IntegrationDecline.Guard); #1379's own rows still pass. The 1,774-problem suite: 1706/1774, 0 wrong, 0 timeouts, verdicts identical to master, 89.6 s → 83.5 s. All four suites green (11,513). Family 7 sample: 147 → 150 of 270, the three atanh(tanh)/acoth(coth) timeouts answered (x^2 atanh(tanh(a + b x)) at 18.6 s is past the budget and counted as a timeout still; three others go from a 5 s timeout to a 7–19 s decline, which the sample counts the same way).

Also seen on the same re-sample and not fixed here: 1/sqrt(atanh(tanh(a + b x))), a 306 ms answer at 2aa0006 and unevaluated since #1377 (a846738), where the exponential substitution began taking a symbolic slope and the substitution search no longer reaches u = the logarithm. Noted for a separate look.

Part of #718.

🤖 Generated with Claude Code

https://claude.ai/code/session_012sonx8iAspMiwRwokT1Ura

…is simplified

atanh(tanh(a + b x))^4 was answered in 300 ms and, since #1379, ran past three minutes: the
reader for a power of A + B ln(c x^r) simplified the logarithmic derivative of any single
logarithm's argument to see whether it was r/x, and the argument here is a quotient of
exponentials whose derivative the simplifier turns over for minutes on the way to no. The
argument is read first -- products, quotients and powers of x and of constants, however
nested, c (d x^m)^n -- and anything else is not this shape without a derivative being taken.
The reader of a linear in the quotient-of-linears rule gets the same guard: a leaf with a
function of x in it is not a linear however it simplifies.

Family 7's atanh(tanh) and acoth(coth) rows are answers again, in 300 ms and 41 ms; the
1774-problem suite is unchanged at 1706 with 0 wrong.

Part of #718.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012sonx8iAspMiwRwokT1Ura
@Rafael-SOWNet
Rafael-SOWNet merged commit e780ce9 into master Sep 17, 2026
31 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant