Jump to content

Talk:Runge–Kutta methods

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Definition of h

[edit]

"...on the order of h4" That's the first mention of h in the article. Perhaps it should be explained somewhere? --Doradus 03:28, 11 Jul 2004 (UTC)

Oops, my mistake. It's fine as-is. --Doradus 03:29, 11 Jul 2004 (UTC)
Perhaps someone can explain what h is? —Preceding unsigned comment added by 203.122.193.233 (talk) 17:13, 2 December 2010 (UTC)[reply]
changed the section to mention Rcgldr (talk) 01:54, 27 June 2011 (UTC)[reply]
mention of h = Δt was removed from the article. Since the article makes many references to t as time, it would make since to at least mention that h = Δt (a time step) somewhere early on. Even in the cases where t is not time, since the article updates t via , it would still make sense to mention that h = Δt. Rcgldr (talk) 16:46, 1 February 2017 (UTC)[reply]

Great!

[edit]

I just wanted to say that it's articles like this that make wikipedia great! It's completely mindboggling that a free encyclopedia can be this much better than many of the expensive dead-tree ones! I especially like the fact that you put up the equation for Runge Kutta-integration up front, and describe it without using unnecessary mathematics. This gives me the impression that you really want people to understand! Some paper encyclopedias seem more worried about fellow nitpicking scientists than getting the message across, so to speak.

Comparison

[edit]

How much better actually is RK4 than Euler's method? For example, when numerically solving a system of linear ODEs, how much bigger can the step h be to give the same level of accuracy, how do they scale? If someone can post an informative answer, it would be appreciated and make the article more beneficial to those wanting to use these methods. —Preceding unsigned comment added by 128.214.120.79 (talkcontribs)

With RK4, you can extract an error estimate, and use this to cut b when necessary to maintain the error within desired limits. --John Nagle 06:27, 7 January 2007 (UTC)[reply]
(BTW Nagle: You're my hero!)
For the Euler Method, the (global) error grows like O(h), while for (explicit) RK4, it grows like O(h^3). In the real-world, with RK4, that means you can take the cube-root (within a scaling factor(12?)) of the h you would use in the euler method to get the same error bounds, with only 4 times as many function evaluations. (~100x faster for the same level of accuracy using single precision floats.)
...Unfortunately, this only works if your ODE is non-stiff. If you're trying to solve a stiff_equation, then using a higher-order method might even make the situation worse. Depending on your application, you may wish to look at the symplectic integrators, and in particular, the verlet leap-frog, which is almost trivial to implement, and the (global) error grows only linearly wrt time... ~~(anon)06:40, 17 July 2007 (UTC)

How to define K_1, K_2, etc -- with h or without h

[edit]

The following comment states: "the section was changed so the k_i are not defined as h f(...), so they're back to being slopes." The resulting RK4 equations are correct. However, the text below the equations still states: "k_{1} is the increment based on the slope at the beginning of the interval, using y (Euler's method)". This text needs to be changed, as were the equations, to read: "h k_{1} is the increment based on the slope at the beginning of the interval, using y (Euler's method)". Same for k2 - k4. --24.203.150.10 (talk) 22:36, 9 April 2018 (UTC)[reply]

It is confusing: k_i is first defined as an increment (with the h) and then as a slope. Since the figure itself uses the convention that k_i are slopes, it would make sense to only use this convention.129.219.43.222 (talk) 17:31, 15 February 2019 (UTC)[reply]


If the k_i are defined with the h, they cannot be termed "slopes", but increments. Please confirm this.

changed the section so that k_i are termed "deltas" Rcgldr (talk) 01:53, 27 June 2011 (UTC)[reply]
the section was changed so the k_i are not defined as h f(...), so they're back to being slopes. Rcgldr (talk) 16:33, 1 February 2017 (UTC)[reply]

— Preceding unsigned comment added by 77.227.21.43 (talk) 22:04, 13 June 2011 (UTC)[reply]

I can't say if the equations given in the article are wrong, but there: http://search.netscape.com/ns/boomframe.jsp?query=runge+kutta+method&page=1&offset=0&result_url=redir%3Fsrc%3Dwebsearch%26requestId%3D98746ea48585333e%26clickedItemRank%3D1%26userQuery%3Drunge%2Bkutta%2Bmethod%26clickedItemURN%3Dhttp%253A%252F%252Fmathworld.wolfram.com%252FRunge-KuttaMethod.html%26invocationType%3D-%26fromPage%3DnsBrowserRoll%26amp%3BampTest%3D1&remove_url=http%3A%2F%2Fmathworld.wolfram.com%2FRunge-KuttaMethod.html

and in my books, they are a little bit different.

eg.:K3= f(ti+h/2,wi+k2/2) and not f(ti+h/2,wi+k2*h/2)

--anon forgot to sign

You see, you need to multiply by h sooner or later either way. Either you do
k_2 = f ( t_n + h/2, y_n + k_1*h/2)
k_3 = f ( t_n + h/2, y_n + k_2*h/2)
or you do:
k_2 = h*f ( t_n + h/2, y_n + k_1/2)
k_3 = h*f ( t_n + h/2, y_n + k_2/2)
Let me know if this needs more explanation. Oleg Alexandrov 22:34, 31 Mar 2005 (UTC)

--

Oleg is right and I was wrong in my previous comment. My apologies!

--

But is multiplied by h again in the y_{n+1}=y_n+h/6*(...)
On numerical recipes (cap 16.1 equation 16.1.3) is without h.
Maybe is more clear if you try to use a simple function like f(x_n,t_n)=1
Of course you must find something like y_{n+1}=y_n+dt where dt is h
Using RK4 with h/6*(...) you find y_{n+1}=y_n+dt^2
I'm wrong?


-- przemoc —Preceding unsigned comment added by 213.241.34.186 (talk) 21:50, 25 October 2007 (UTC)[reply]

It seems the equations below are wrong. The h's are missing in the arguments of the function f


k_1 = f(t_n, y_n)
k_2 = f(t_n + \tfrac{1}{2}h, y_n + \tfrac{1}{2} k_1)
k_3 = f(t_n + \tfrac{1}{2}h, y_n + \tfrac{1}{2} k_2)
k_4 = f(t_n + h, y_n + k_3)


When compared with numerical recipes, they should be

k_1 = f(t_n, y_n)
k_2 = f(t_n + \tfrac{1}{2}h, y_n + h\tfrac{1}{2} k_1)
k_3 = f(t_n + \tfrac{1}{2}h, y_n + h\tfrac{1}{2} k_2)
k_4 = f(t_n + h, y_n + hk_3)


Otherwise, the corrections of y_n in k_2, k_3 and k_4 are O(1), instead of O(h). P.S.: I think it should be better to number the equations, or at least a block of equations. That would make present and future references to them easier to do. Adriano bb (talk) 18:53, 17 September 2009 (UTC)[reply]

Seems to me you wrote this before noticing my reversal of an edit made by an anonymous user, which introduced the error you are pointing out. Hanche (talk) 21:28, 17 September 2009 (UTC)[reply]

--

It seems to me that, after a lot of discussion, the definitions and explanations of the k_1 ... k_4 parameters in the RK4 section are still wrong. In my opinion they should be:

k_1 is the slope at the beginning of the interval, using y_n , (Euler's method);
k_2 is the slope at the midpoint of the interval, using y_n + 1/2 h k_1 ;
k_3 is again the slope at the midpoint, but now using y_n + 1/2 h k_2 ;
k_4 is the slope at the end of the interval, using y_n + h k_3.

because, if all k's are defined by f(), which is equal to \dot y, then they must be all slopes, and not increments. Moreover these functions f() are evaluated at (y_n + something), not (\dot y + something), and therefore the ″using″ parts are also wrong. To make this ″using″ parts clearer, I suggest rephrasing them as

[...], using y(t_n + 1/2 h) ≈ y_n + 1/2 h k_1 ,

for k_2, and similarly for k_3 and k_4. Joan Solà (talk) 15:10, 28 February 2014 (UTC)[reply]

Starting with RK4

[edit]

I don't quite like the idea of starting with RK4 as it is rather complex, but I thought of starting with a simple two-stage RK method and application thereof, thus elucidating why RK4 is as it is. Does that sound okay? There is a large family of RK methods, we should be aiming to be rather general... Dysprosia 09:49, 15 May 2005 (UTC)[reply]

Good point; I agree. My edit of your previous contribution, in which you introduced the general formula, was perhaps too hasty, but I did not like your second-order example (it evaluates f outside the interval ) and you changed the definition of the k_i (if I remember correctly); I was further emboldened by Oleg's edit comment. One possible outline for the article is to start with Euler's method, then some second-order method which is still easy to understand, like the midpoint rule, then give the general formula for explicit RK, then some examples, including RK4, and then whatever else you'd like to mention. Also keep in mind that some people think that "Runge-Kutta method" means RK4, so it's probably best to mention very early where RK4 is described. But most importantly, don't listen to me and make any improvements to the article that you see fit. -- Jitse Niesen 13:09, 15 May 2005 (UTC)[reply]
Well, call me biased, but I like the article the way it is now. It does refer at the top to numerical ordinary differential equations for general background. Also, indeed, the name "Runge-Kutta method" means most often RK4, as the two-stage method is usually known as midpoint method, or otherwise there also is Heun's method, etc.
But I wonder which two-stage Runge-Kutta Dysprosia is referring to. Maybe there is indeed a way to start even simpler than what is now. However, the section on Explicit Runge-Kutta methods with all those formulas and undertermined coefficients, should indeed stay at the bottom where it is now. Oleg Alexandrov 14:54, 15 May 2005 (UTC)[reply]
(Jitse) I did indeed change the definition for the k_i -- it was the one I was more familiar with. It shouldn't be difficult to push the RK4 example to fit that definition, or vice versa.
(Oleg) We can start with any simple two-stage RK method, and a simple example of solving a simple DE with it 'on paper', so to speak.
Dysprosia 23:05, 15 May 2005 (UTC)[reply]
Dysprosia, thanks for adding the example (and also the examples in Lagrange polynomial and Newton polynomial). Unfortunately, the table does not look very nice on my laptop, where my screen is a mere 1000-and-a-bit pixels wide. It may also be nice to use a nonautonomous equation. Finally, I'm wondering why you are not using the midpoint method. Sorry for not being more constructive; I have little time now. -- Jitse Niesen 14:35, 21 Jun 2005 (UTC)
Sorry, I spaced out the table so it wouldn't look so cluttered and it would line up properly. I have a wide screen. I'll try and fiddle with it later.
I didn't use the midpoint method, so we would have a greater diversity of examples.
Dysprosia 04:49, 22 Jun 2005 (UTC)
I don't see greater diversity as a good thing, in this case, but fair enough. However, please use a decent method as an example, and not a method with c_i outside the interval [0,1]. You can use for instance
             0 |
           2/3 | 2/3
           ----+----------
               | 1/4  3/4
Of course, a plot would be very nice. Jitse Niesen 12:57, 22 Jun 2005 (UTC)
Sorry, my stupidity -- recalculated. Plot later. Dysprosia 12:05, 23 Jun 2005 (UTC)

Local error

[edit]

I removed the last phrase of "The RK4 method is a fourth-order method, meaning that the total accumulated error is on the order of h4, and that it gives the exact integral for polynomials up to the fourth order." because it is not clear for somebody who does not know the subject. To make it clear, it should be explained that methods for solving ODEs can also be used for calculating integrals, that the error in solving ODEs is related to the error in calculating integrals, and that the latter is related to the degree of polynomials that can be integrated exactly. While all this can be done, I think it is not relevant for this article; perhaps it can be added to numerical ordinary differential equations. -- Jitse Niesen 00:03, 8 Jun 2005 (UTC)

implicit and explicit

[edit]

This document mentions that Runge-Kutta methods include both implicit and explicit methods, however it doesn't describe the difference between either of these. This should be clarified.

Iterative versus Non-iterative

[edit]

If I'm not mistaken, there are iterative approaches to calculating the internal stages of the Runge-Kutta method. For instance, for a 3rd order Runga-Kutta iterative approach, a foward Euler step is taken from t to (t + dt/3), then a leapfrog step is taken from t to (t + dt/2) using forcing at (t + dt/3). Then a final leapfrog step is taken from t to (t + dt) using forcing at (t + dt/2). Someone should make a more formal mention of this proceedure on this page. I would, but I'm still currently looking up journal articles to fully understand how this plays out in the expansion to yeild 3rd order accuracy. If I'm not mistaken, similar proceedures can be adopted to perform the time integral at higher orders. —Preceding unsigned comment added by 64.241.37.140 (talkcontribs)

The Usage-section

[edit]

The Usage section in the article as of now is quite voluminous and I also think it violates WP:NOT#HOWTO. --Berland 18:33, 28 June 2007 (UTC)[reply]

I agree it's too long, but to me it's more of an example than a how to. More seriously though, it seems to follow a different notation from the rest of the article and so is more likely to confuse than clarify. Is an example really needed here or is the article clear without it?--RDBury (talk) 05:39, 21 November 2007 (UTC)[reply]

I agree with RDBury, the change of notation is rather confusing, the section should be either "translated" to the notation defined above or either erased. Maybe the translation from tableau to equations - in proper notation - would be enough?

Jose Brox

Pronunciation?

[edit]

How is 'Runge' pronounced? I'm not wild about IPA phonetics, but a "rhymes with" would be awfully useful. :) —Preceding unsigned comment added by Myself248 (talkcontribs) 23:18, 24 November 2007 (UTC)[reply]

The 'u' is (approximately) pronounced like the 'oo' in 'look', the 'ng' is as in 'sing', and the 'e' is as in 'the'. I don't know a word which rhymes with 'Runge'. -- Jitse Niesen (talk) 16:45, 25 November 2007 (UTC)[reply]

Not to teach subject matter ?!

[edit]

I suppose the banner saying that purpose of the wikipedia is only to present facts is put there by some sort of robot, but it is unfortunate. The wiki disavows being a reference that meets scholarly standards. So an emphasis on merely "presenting facts" seems rather silly. Doing so would result in some rather barren pages for mathematics.

It would be useful to have an example of stability analysis. From an article on the scholarpedia, I see that this involves an equation with matrices in it. For persons who are not at the top of their game in matrix notation, an example would clarify whether a notation means the number 1 or the identity matrix, the absolute value of something or a determinant, etc.

It is interesting that the vector b, which appears at the bottom of the table is written horizontaly and denoted as b_transpose. This suggests that, in another way of looking at things, b is naturally thought about as a column vector. It would be useful to comment upon this.

In the spirit of knowing that certain approximations are bad, it would be useful to comment on whether pronouncing "Runge" to rhyme with "lunge" , as mamy USA instructions do, is a good one.

Tashiro 15:22, 1 December 2007 (UTC)[reply]

Advisory excluding examples from Wiki Articles should be removed ==

[edit]

Whatever the Wiki is, it is presumably trying to present HUMAN useful information, and an example is often used in mathematical articles to better communicate what is meant by what was more abstractly said. The author's example should stay, and the editor's advisory should go. —Preceding unsigned comment added by 76.191.132.120 (talk) 20:06, 22 January 2008 (UTC)[reply]

Leaves something to be desired

[edit]

At the time of writing the article might be clear as crystal to someone who already knows what the Runge-Kutta method is trying to achieve, but to someone who is pretty much a beginner like me I am left wondering: OK the method produces a series of yk's but what do these represent? In other words what is the meaning of the output of the method? Feraudyh (talk) 14:00, 23 April 2008 (UTC)[reply]

y_n is equal to y(t_n): it is the value of the solution to the differential equation at the point t_n. The aim with any numerical method for differential equations is to approximate the exact solution as closely as possible. Hope this helps. Tweet7 (talk) 21:53, 23 November 2009 (UTC)[reply]

PDEs

[edit]

Runge-Kutta can be used for PDEs as well (I believe mostly parabolic) should a section on using Runge-Kutta for PDEs exist on this page? What is everyones thoughts on this? Im just a mere math undergrad with little sophistication when in comes to numerical analysis. --User:MATThematical

RK methods can be applied to well-posed initial boundary value problems (whether parabolic or hyperbolic), however, the bare application to those methods isn't terribly informative and the analysis of the limitations of those applications is quite in depth. There are literally whole bookshelves worth of material devoted to understanding just that. -- Grigjd3 — Preceding unsigned comment added by Grigjd3 (talkcontribs) 13:23, 27 May 2011 (UTC)[reply]

Error term

[edit]

I see lots of literature that says the error term is O(h^5) for a single step, and O(h^4) globally, but the error term derivation is never given. It should be pretty straightforward using the Taylor series. Would be great if someone could add it (I'd do it myself, but the derivation would probably constitute original research, plus I could make a mistake :P It must be in a text book that someone has somewhere). --Numsgil (talk) 08:14, 24 May 2008 (UTC)[reply]

Hyphenated name, not endashed

[edit]

The current name of this page (and others that are linked from it) includes an endash, but it should be a hyphen. This page needs to be moved to its hyphenated version. Among other things, it would make it easier for people to link to it. --TedPavlic | talk 19:48, 6 August 2008 (UTC)[reply]

Why exactly should it be a hyphen? According to WP:DASH, one of the roles of en dashes is to indicate disjunction, with as one of its main applications being to serve as "a substitute for some uses of and, to or versus for marking a relationship involving independent elements in certain compound expressions". Runge–Kutta methods are methods due to Messrs. Runge and Kutta, being independent individuals. Still according to WP:DASH: "When naming an article, a hyphen is not used as a substitute for an en dash that properly belongs in the title". As Runge-Kutta methods with a hyphen redirects to Runge–Kutta methods with an en dash, linking is easy enough.  --Lambiam 22:37, 12 August 2008 (UTC)[reply]
To throw more wood into the fire: It should neither be hyphenated nor en-dashed. (Although, you can argue for the hyphenation; the en-dash is entirely wrong.) I cite Brian Garner on this issue: proper nouns used in a phrasal adjective should not be hyphenated. — Preceding unsigned comment added by 138.246.2.239 (talk) 11:53, 14 October 2014 (UTC)[reply]

Defining aij for a given order??

[edit]

Is there no general way to define aij for a given order p? The article mentions "there are also accompanying requirements," but doesn't state in general what they are. If there is no general solution, then I think that needs to be stated in the article. If there is one or more, then I think at least one of them should be shown. Erikmartin (talk) 14:19, 7 August 2008 (UTC)[reply]

There is no general definition of aij. The coefficients simply are what the person who derived the tableau decided that they should be.
What I wonder about though, is how the tableaus are derived? How can one determine of which order a tableau is? I think this information is very essential and should be in the article! --Kri (talk) 10:20, 19 October 2009 (UTC)[reply]
There are entire books and papers dedicated to this. In general it is rather complex to determine or even guess the order just by looking at the table. But with enough mangling it can be done. Usually do it is done in reverse. The limits on table coeffients is done so the method will have maximal or some specific order (or adhere to some additional properties, i.e. be embedded, be explicit, be parallel, be singly diagonal, have some coefficients equal to 1.0, etc), and then parameters are choosed to preserve the order and then maximize some other properties (like numbers of zeros in a table, or a stability of the method), while keeping it correct. 81.6.34.246 (talk) 00:57, 31 December 2018 (UTC)[reply]

Crank-Nicolson Example

[edit]

The final Butcher tableau in this article (added in this [1] edit) looks to be incorrect. I believe it should be

Correcting my own LaTeX typos. --92.242.168.176 (talk) 16:43, 15 March 2010 (UTC)\\[reply]

I'm not so sure about this. Crank Nicolson is a 1-stage method. I'll look it up later tonight and find out and correct it with the relevent sources :) (Oliver 31/10/2011 14:54)

definition of k as a slope

[edit]

Is there a problem with the defnition of k_i and the corresponding description of k_i? For example k_1 = h*f(t_n,y_n), and f(...) is dy/dt. so h*f(t_n,y_n) is infact "the difference in y over the interval t -> t+h, based on the slope at the beginning of the interval" and not "the slope at the beginning of the interval" as in the current text. The same for definitions/descriptions of k_2, k_3 and k_4. —Preceding unsigned comment added by 143.117.13.21 (talk) 16:04, 2 June 2010 (UTC)[reply]

Yes, this problem is the result of a recent edit in which someone moved the hs outside of the ks, to increase consistency across the article. The end result is mathematically identical, but the definition of k1-k4 as slopes is now incorrect. Should h be moved back into the ks, or should section on slope be modified? 76.212.130.54 (talk) 18:39, 20 June 2011 (UTC)[reply]

changed the section so that k_i are termed "deltas" Rcgldr (talk) 01:49, 27 June 2011 (UTC)[reply]
the section was changed so the k_i are not defined as h f(...), so they're back to being slopes. Rcgldr (talk) 16:35, 1 February 2017 (UTC)[reply]

Merson

[edit]

I was surprised to find no mention of Runge Kutta Merson, which, adding a fifth evaluation to RK4, enables a better interval to be computed. In cases where stability varies over the range of integration, the cost of the extra evaluation is more than balanced by the reduction in the number of evaluations required. I regret my maths is not up to the high standard set in this article, but I found RKM so useful that I am mentioning it here hoping that someone better qualified will put it in. John Wheater (talk) 11:14, 3 September 2010 (UTC).[reply]

Runge-Kutta Nystrom

[edit]

I understand there exists a class of integrators referred to as Runge-Kutta Nystrom integrators designed to allow one to integrate second order ODEs using the method of lines. My personal literature search has only revealed to me some of the most recent findings about such systems at all. If anyone is knowledgeable about such methods, it would be nice to see an explanation of them and perhaps a couple of useful examples (for instance, ERKNs at various orders). — Preceding unsigned comment added by Grigjd3 (talkcontribs) 13:29, 27 May 2011 (UTC)[reply]

Yes, it would be nice to see a page on RK-Nystrom methods, and yes, you're correct they're used to integrate 2nd-order ODEs. Hairer, Nørsett & Wanner (1993) have a section on numerical methods for second-order ODEs, II.14, pages 283--301, that includes a couple examples that I would be willing to describe. Where would an appropriate place be to add them? Should I add them to this page or a new page? MathRocks2012 (talk) 21:14, 6 August 2013 (UTC)[reply]

Doubts on the derivation of Runge-Kutta fourth order method

[edit]

I have some major doubts on the derivation of Runge-Kutta fourth order method as it is presented now, even if I found this explanation elsewhere (for example, here) and even if, at first, this sounds right, now I am wondering why you are allowed to take the coefficients up to shouldn't they be expanded at least up to to obtain a expression? Some clarifications are in order. -- CristianCantoro (talk) 14:18, 15 March 2013 (UTC) In particular, more detailed calculation (using Butcher tables) lead to more complicated systems of equations (see pag. 5 of this) with the same result. Unfortunately I wasn't able to find an explicit derivation using Butcher tables with all the complete calculations. -- CristianCantoro (talk) 14:20, 15 March 2013 (UTC)[reply]

I agree with the previous statements. Butcher tableaus were invented to aid with the derivation of order conditions using rooted trees. We could replace this discussion to define order conditions for any fourth-order method, and then simply remark that one such solution is given by the coefficients provided in the classical Runge-Kutta sheme. MathRocks2012 (talk) 19:08, 9 August 2013 (UTC)[reply]

I fully support the point that the derivation is misleading and should be taken down. The terms in must not appear as after being only multiplied by h. has enough terms to be at least eligible to give , but then again it has more terms than the declared would require. So the math seems to be just wrong here, though it does lead to the correct result. 195.70.199.20 (talk) 15:37, 18 September 2020 (UTC)[reply]

Parsing error

[edit]

There is a parsing error being thrown in Derivation of the Runge–Kutta fourth-order method. Sizeofint (talk) 11:35, 13 December 2015 (UTC)[reply]

Assessment comment

[edit]

The comment(s) below were originally left at Talk:Runge–Kutta methods/Comments, and are posted here for posterity. Following several discussions in past years, these subpages are now deprecated. The comments may be irrelevant or outdated; if so, please feel free to remove this section.

Add implicit methods, order conditions, order barriers, more methods. Clean up example. -- Jitse Niesen (talk) 11:50, 28 April 2007 (UTC) The error for this method should be included as well--Cronholm144 16:19, 20 May 2007 (UTC)[reply]

Last edited at 16:19, 20 May 2007 (UTC). Substituted at 02:34, 5 May 2016 (UTC)

Picture

[edit]

The picture of the classical method is not consistent with the text. It's the same confusion as earlier mentioned, about the definition of the k's with or without h. In the text, the k's are y-values, and in the picture they are slopes. Madyno (talk) 13:49, 2 March 2019 (UTC)[reply]

The first image has writing in German

[edit]

--2.231.51.224 (talk) 14:07, 2 January 2020 (UTC)[reply]

Butcher Tableau for the trapezoidal rule?

[edit]

The article gives the following Butcher Tableau for the trapezoidal rule.

Is that correct? I did't think that the trapezoidal rule is an embedded method, so it shouldn't have the row. -The-erinaceous-one (talk) 21:26, 14 August 2020 (UTC)[reply]

Autonomous (time invariant) equations

[edit]

I have school-level maths but was able to understand RK4. Thanks to those who wrote the text so carefully. I am working with autonomous differential equations. The one sentence I do not understand is about them, currently written as:

In many practical applications the function f {\displaystyle f} f is independent of t {\displaystyle t} t (so called autonomous system, or time-invariant system, especially in physics), and their increments are not computed at all and not passed to function f {\displaystyle f} f, with only the final formula for t n + 1 {\displaystyle t_{n+1}} t_{n+1} used.

Meaning of "their increments"? "With only the final formula . . . used"? Is this saying that only k4 is used?

If anyone could clarify this, I'd be grateful. — Preceding unsigned comment added by KPD674 (talkcontribs) 00:55, 18 February 2021 (UTC)[reply]

KPD674, I think that the point that sentence is making is that, if f() doesn't depend on t explicitly, then it's not necessary to calculate t_n+h/2 when calculating k_2, etc. This strikes me as a statement of the obvious. Given that it's unsourced I wonder whether it would be better to remove that sentence entirely. Best, Wham2001 (talk) 07:54, 18 February 2021 (UTC)[reply]

Kutta-Merson method

[edit]

Would there be any objections to adding the Kutta-Merson method to the section on adaptive RK methods? See, e.g., https://encyclopediaofmath.org/wiki/Kutta-Merson method. My impression is that it's less well-known than RKF45, and seems to have a tendency to overestimate errors in some circumstances, but it's also relatively straightforward and uses the traditional RK4 for the the fourth-order method.

If I don't hear any objections I'll likely add it later this week. DrPippy (talk) 14:19, 3 November 2021 (UTC)[reply]

Still used today?

[edit]

Have these methods been overtaken? Are other methods used today by default? Are they just of interest for teaching purposes? Are they still used if computing power is limited? Are they used as part of more modern methods? I think we should be told. Burraron (talk) Burraron (talk) 10:22, 5 April 2024 (UTC)[reply]

A starting point for the level of someone asking this kind of question would be to look at SciPy solve_ivp. It can implement just about anything for you that was available half a century ago.
It is unfortuante to say, but yours is a case of the failure of the secondary education system (and likely tertiary educatio nas well) to teach young people about the sheer depth and breadth of knowledge available in the world. Not even the most advanced degrees of any subject will begin to prepare anyone for serious learning required to tackle the explosion of research in the computer-backed information age. Even those on the cutting edge always feel like they are a step behind.
Modern methods are so sophisticated that RK4 could realistically be considered kindergarten level. Even most of the freeware on SciPy even is massively outdated. For example, a cursory search will find you that RK10 existed in 1978. Yet unless you are directly researching algorithm design or are trying to invent new mathematics to avoid conventional data structures, it is probably best to just stick to using a software package that handles integration for you. Given the rapid development of this subject in light of the failure of climate models to actually work (don't tell Wikipedia), you could easily spend the rest of your life researching this topic with its growth continuing to outpace you. 174.130.197.109 (talk) 20:38, 19 July 2024 (UTC)[reply]

I am pretty sure that anybody making a space-object trajectory simulator software today would use Runge-Kutta methods, at some phase in its development, to give decent long-extrapolation accuracy with less computing power. If there is anything that superceedes it for such use, I don't know that trend. Past uses deserve to be preserved, imo, if only to make re-development handy for newcomers. I think there may be some methods based on Keplerian ellipsoidal orbits, but their applicability always needs to be challenged by comparison with results of other methods. — Preceding unsigned comment added by Jim Swenson (talkcontribs) 23:25, 11 April 2024 (UTC)[reply]