The root solving method Newton Raphson converges quickly to the estimated root value but requires a 'close' enough initial guess to converge. I have read that an initial value is often chosen by use of the bisection method, where it iterates until a low level of tolerance and it is fed as an initial guess into Newton's. However, the bisection method requires a change of signs along the function.

My question is what other method could you use to feed into Newton's if the function is never negative over its domain?

  • $\begingroup$ Newton's method may run slowly if the function does not have non zero slope at a root. $\endgroup$
    – copper.hat
    May 29, 2020 at 1:12
  • $\begingroup$ If the function is never negative over its domain (assuming "nice" function), then it has no roots. The only possible zero would be at a horizontal tangent touching the x axis. $\endgroup$ May 29, 2020 at 2:59
  • $\begingroup$ I am afraid that the last sentence is almost killing the question. $\endgroup$ May 29, 2020 at 4:34
  • $\begingroup$ @ClaudeLeibovici A function can have roots but not necessarily be negative, a root is defined by f(x)=0; where 0 is not negative. So it still can have roots $\endgroup$
    – notMyName
    May 29, 2020 at 15:23

1 Answer 1


You describe a situation where your function $f$ is defined on $\Omega \subseteq \mathbb{R}$ and is nonnegative, i.e., $f : \Omega \rightarrow [0,\infty)$.

As you correctly observe, you cannot hope to apply the bisection method to $f$ in order to narrow the search for an initial guess.

However, there are at least two options. Any root of $f$ is necessarily a global minimum, of $f$, hence a root of $f'$.

  1. If $f$ is at least three times differentiable and $f'''$ is continuous, then you can hope to apply Newton's method to the equation $f'(x) = 0$ and have quadratic convergence. In this case, you may be able to detect a sign change of $f'$ and apply bisection to narrow the search interval. This procedure will give you a list of candidates for roots of $f$, but you will of course have to verify them one by one.
  2. Alternatively, you can attempt to locate a minimum of $f$ using a dedicated algorithm, such as the golden section search. Again, you will have to verify if the candidates are in fact roots of $f$. Restrictions apply to the application of the golden section search and while the convergence to a local minimum is assured it is only linear.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.