Talk:Bernoulli's method

Please clarify: How a sequence of real numbers can have a complex root as a limit?

@D.Lazard This is a question that deserves a longer answer. One thing to note is that the coefficients of the polynomial don't need to be real. The methods works with either case so it could find a dominant complex root (see Henrici p155). If everything is real, then finding dominant complex conjugates requires extra steps after computing the sequence. McNamee & Pan in chapter 10.2 describe these steps (they cite Jennings too) and Henrici also explains it in chapter 7.5 and it a little involved. The short version requires setting up a system of equations to solve for the angle and radius by essentially using four consecutive terms in the sequence to compute two determinants and pluging them into three equations.

I'm not sure if adding these would really benefit the article because if multiple roots are desired, then generalized approaches like QD or Aitken could compute these. Is there something that could be said to clarify things but without prompting more questions? Or would it be better to describe how to find conjugate pairs? Basilelp (talk) 04:25, 24 April 2025 (UTC)

Writing a good Wikipedia article requires to respect the WP:principle of least astonishment. The tag {{clarification needed}} means simply that the principle is not respected, since everybody may ask the same question. Here are several points, where the article deserves to be improved. It is important to say what is known for the 3 last items, since they rely on questions that evey competent reader cac ask himself.

• Readers must know what is an absolute value, and not using this term may confuse readers. So, "magnitude" must be replaced everywhere with "absolute value, as I did in the lead.
• It must be explicitely said that, if the coefficients are real and the root of largest absolute values is not real, then the algorithm does not work since the conjugate root has the same absolute value.
• If there are several roots of maximal absolute value, does the sequence has a limit (never, sometimes of always)? If it has a limit, how is it related to the roots (in general or when the maximal roots are complex conjugate)?
• Same questions in the case of multiple roots.

By the way, I'll add a sentence explaining why the given initial conditions provide nonzero c_i's. D.Lazard (talk) 14:09, 24 April 2025 (UTC)

Thinking on the method, I can conclude that

• The method converges always if there is a unique root of maximal absolute value, and this does not depend whether there are multiple roots or whether the root of maximal value is multiple or not.
• If one starts with other initial conditions that the ones of the article, the method converges almost always to the root of maximal modulus, but may sometimes converge to another root.
• If there are several roots of maximal modulus, I believe that the method never converges (I so not know a proof), but some information can be deduced from the limit behavior of the sequence of the quotients.

The proof of the first item relies on the fact that all solutions of the recurrence relation are linear combinations of the ⁠$r_{i}^{m}$⁠ and ⁠$m^{\alpha }r_{i}^{m}$⁠ with ⁠$\alpha <k$⁠ if ⁠$r_{i}$⁠ has a multiplicity ⁠$k>1$⁠. Otherwise, the proof proceeds as in the article.

The proof that ⁠$c_{1}\neq 0$⁠ with the given initial conditions can be done by remarking that the matrix that relates the ⁠$c_{i}$⁠ and the initial conditions is a nonsingular Vandermonde matrix or confluent Vandermonde matrix. With the initial conditions of the article, Cramer's rule expresses the ⁠$c_{i}$⁠ as a quotient of two such matrices.

I have not the time for detailing the proof in the article, but I am quite sure that the proofs exists in the litterature. D.Lazard (talk) 09:39, 25 April 2025 (UTC)

With regards to the initial premise of a polynomial with real coefficients having a single dominant complex root, I do not think this is the case. Either, it would have to have complex coefficents or the root would need to be part of a complex conjugate. I didn't want to try and write a proof for this, but I did spend a long time trying to find a counter-example polynomial with only real coefficents and a single dominant root and got nowhere. I suspect there is a proof somewhere. Basilelp (talk) 18:35, 25 April 2025 (UTC)

Found it. "The complex conjugate root theorem states that if the coefficients of a polynomial are real, then the non-real roots appear in pairs of the form (a + ib, a – ib)." See Geometrical properties of polynomial roots.

So no, using a polynomial with real coefficients will not create a sequence that converges on a complex number because it can't. If there was a single dominant root then the coefficients would have at least one complex number and the method would be able to find that. Basilelp (talk) 22:20, 25 April 2025 (UTC)

Discussion on Hatnote Disambiguation

"hatnote: hatnotes are for disambiguating between similar titles — difficult to be confused between a method, a differential equation and a principle"

I would have hoped this was the case but if you plug "Bernoulli's Method" into Google or Bing, it will come back with Bernoulli's Differential Equation and Bernoulli's Principle. Of the dozens of videos on YouTube called "Bernoulli's Method" only one of them actually covers the rootfinding algorithm and the rest explain the principle or differential equation. Basilelp (talk) 03:00, 26 April 2025 (UTC)

Just to add to this, if you look at Bernoulli's principle they have a hatnote for Bernoulli differential equation, as well as Bernoulli's Theorem of Probability, so there is lots of confusion. Basilelp (talk) 04:12, 26 April 2025 (UTC)