Timeline for Code that produces plot in V5 doesn't work in later versions
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 9, 2020 at 4:15 | comment | added | J. M.'s missing motivation♦ | Ah, I knew we've discussed Abel/Cesàro summability before... | |
| Jul 8, 2020 at 4:17 | comment | added | Michael E2 |
@xzczd The Plot doesn't work with integrand0, but it does with Plot[Evaluate@integrand0,...]. It must literally numericize the argument (for when that happens, it passes the convergence test).
|
|
| Jul 8, 2020 at 2:56 | comment | added | Michael E2 |
@xzczd Try Block[{\[Tau] = 2}, 10^18*Sqrt[\[Pi]/2] * NIntegrate[integrand0, {y, 0, \[Infinity]}, MaxRecursion -> 12]]. I think tau has to be at least 2. But now I'm not sure how Plot does the graph. I'll work on it...
|
|
| Jul 8, 2020 at 2:42 | history | edited | xzczd♦ | CC BY-SA 4.0 |
deleted 66 characters in body
|
| Jul 8, 2020 at 2:40 | comment | added | xzczd♦ |
Oh… I did suspect the integral diverges, but the moment I see the result of self-implementation of ExtrapolatingOscillatory I forgot everything… I've edited my answer accordingly, thx for pointing out, but…"This is why NIntegrate misses the divergence in Plot: Plot applies N to NIntegrate when it first returns unevaluated, and N[NIntegate[..]] somehow forces the fall-back method " This isn't quite right, because NIntegrate[integrand0 /. τ -> 1, {y, 0, ∞}] evaluates. Or you actually mean NIntegrate[integrand, …?
|
|
| Jul 7, 2020 at 17:50 | comment | added | Michael E2 | @xzczd I made a mistake, but I think we both did. I don't think it's a bug. | |
| Jul 7, 2020 at 17:49 | history | edited | Michael E2 | CC BY-SA 4.0 |
added 5619 characters in body
|
| Jul 7, 2020 at 3:33 | comment | added | xzczd♦ |
I just mean τ_?NumericQ actually helps little, the original T also works in v12.1.1, and the performance is almost the same.
|
|
| Jul 7, 2020 at 3:27 | comment | added | Michael E2 |
@xzczd What's the timing got to do with it? I thought the OP's problem was producing the graph. I get no error messages. (You can ClearSystemCache[] instead of restarting the kernel.)
|
|
| Jul 7, 2020 at 3:23 | comment | added | xzczd♦ |
The way to define T isn't too big a problem here. I've tested both, the timing is similar. (One needs to test with a fresh kernel, the timing reduces significantly after the first run. ) I think the real problem is a bug of NIntegrate, see the update to my answer. BTW the warning nconv is still there using this method. Tested on v12.1.1, Win10.
|
|
| Jul 7, 2020 at 3:15 | history | answered | Michael E2 | CC BY-SA 4.0 |