Skip to content

A resonance whose sign is symbolic is written in j = -a/b in place of i - #1864

Merged
Rafael-SOWNet merged 2 commits into
masterfrom
an-exponential-resonant-with-a-symbolic-frequency
Oct 13, 2026
Merged

Rafael-SOWNet merged 2 commits into
masterfrom
an-exponential-resonant-with-a-symbolic-frequency

Conversation

@Rafael-SOWNet

@Rafael-SOWNet Rafael-SOWNet commented Oct 13, 2026 •

Copy link
Copy Markdown
Member

Part of #718. Reopens #1852's decline with an answer.

e^(x/n) sin(x sqrt(-1/n^2)) is resonant: 1/n + i sqrt(-1/n^2) is zero when n > 0, and 1/n - i sqrt(-1/n^2) is zero when n < 0. #1852 declined it because neither sign could be chosen. The rule for a polynomial times an exponential times a trigonometric function now writes Euler's identity in j = -a/b instead of i when it cannot show that either term vanishes. Here a is the exponential's rate and b is the frequency. Where the case is resonant at all, j^2 = -1 holds whichever sign n has, and a + j b is zero by construction, so one answer holds on both sides. If b evaluates to zero, the case is still declined.

A logarithm's substitution produces the same integrand, so Rubi's 4.7.5 rows such as x^m sin(a + ln(k x^n) sqrt(-(1 + m)^2/n^2))^p are answered too.

Sizes, measured on master fb527ae, on 2.5.0 and on this branch:

Integrand 2.5.0 master fb527ae branch
exp(x/n)*sin(x*sqrt((-1)/n^2)) 149 chars, no value anywhere unevaluated 110 chars
exp(x/n)*sin(a + x*sqrt((-1)/n^2)) 157 chars, no value anywhere unevaluated 254 chars
sin(a + ln(k*x^n)*sqrt((-1)/n^2)) unevaluated unevaluated 339 chars
x^m*sin(a + ln(k*x^n)*sqrt(-(1 + m)^2/n^2)) unevaluated unevaluated 465 chars
x^m*cos(a + ln(k*x^n)*sqrt(-1/4*(1 + m)^2/n^2))^2 unevaluated unevaluated 1,027 chars
x^2*sin(a + ln(k*x^n)*sqrt((-1)/n^2))^3 unevaluated unevaluated 1,873 chars
x^m*sin(a + ln(k*x^2)*sqrt(-1/36*(1 + m)^2))^3 unevaluated unevaluated 2,049 chars

The test differentiates each of these back at n = 3 and at n = -3.

Rubi corpus, compared with master fb527ae:

  • 4.7.5 and all of 2.3 (684 problems): 592 → 617 solved, with 0 wrong on either side.
  • The 25 that moved all went from unevaluated to solved, and each was re-run on its own.
  • Family 4 sample: 999 → 1001.
  • Every other family is unchanged.
  • 4.7.5 alone: 121 → 146 of 160. The 14 that remain are all secants and cosecants, a separate seam: 13 unevaluated and 1 timeout.

The allocation gate passed. The suite passed: 15,293 passed, 0 failed.

The #1852 entry in BREAKING-CHANGES.md said the resonance "is declined now". It is amended to the answer, and so is the 4.7.5 entry's sentence that left a symbolic b alone.

🤖 Generated with Claude Code

https://claude.ai/code/session_012sonx8iAspMiwRwokT1Ura

Rafael-SOWNet and others added 2 commits October 13, 2026 01:19
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012sonx8iAspMiwRwokT1Ura
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012sonx8iAspMiwRwokT1Ura
@Rafael-SOWNet Rafael-SOWNet added this to the 2.6.0 milestone Oct 13, 2026
@Rafael-SOWNet
Rafael-SOWNet merged commit 8bdfcfe into master Oct 13, 2026
34 checks passed
@Rafael-SOWNet
Rafael-SOWNet deleted the an-exponential-resonant-with-a-symbolic-frequency branch October 13, 2026 02:47
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