This function carries out a minimization of the function `f`

using a Newton-type algorithm. See the references for details.

```
nlm(f, p, …, hessian = FALSE, typsize = rep(1, length(p)),
fscale = 1, print.level = 0, ndigit = 12, gradtol = 1e-6,
stepmax = max(1000 * sqrt(sum((p/typsize)^2)), 1000),
steptol = 1e-6, iterlim = 100, check.analyticals = TRUE)
```

f

the function to be minimized, returning a single numeric
value. This should be a function with first argument a vector of
the length of `p`

followed by any other arguments specified by
the `…`

argument.

If the function value has an attribute called `gradient`

or
both `gradient`

and `hessian`

attributes, these will be
used in the calculation of updated parameter values. Otherwise,
numerical derivatives are used. `deriv`

returns a
function with suitable `gradient`

attribute and optionally a
`hessian`

attribute.

p

starting parameter values for the minimization.

…

additional arguments to be passed to `f`

.

hessian

if `TRUE`

, the hessian of `f`

at the minimum is returned.

typsize

an estimate of the size of each parameter at the minimum.

fscale

an estimate of the size of `f`

at the minimum.

print.level

this argument determines the level of printing
which is done during the minimization process. The default
value of `0`

means that no printing occurs, a value of `1`

means that initial and final details are printed and a value
of 2 means that full tracing information is printed.

ndigit

the number of significant digits in the function `f`

.

gradtol

a positive scalar giving the tolerance at which the
scaled gradient is considered close enough to zero to
terminate the algorithm. The scaled gradient is a
measure of the relative change in `f`

in each direction
`p[i]`

divided by the relative change in `p[i]`

.

stepmax

a positive scalar which gives the maximum allowable
scaled step length. `stepmax`

is used to prevent steps which
would cause the optimization function to overflow, to prevent the
algorithm from leaving the area of interest in parameter space, or to
detect divergence in the algorithm. `stepmax`

would be chosen
small enough to prevent the first two of these occurrences, but should
be larger than any anticipated reasonable step.

steptol

A positive scalar providing the minimum allowable relative step length.

iterlim

a positive integer specifying the maximum number of iterations to be performed before the program is terminated.

check.analyticals

a logical scalar specifying whether the analytic gradients and Hessians, if they are supplied, should be checked against numerical derivatives at the initial parameter values. This can help detect incorrectly formulated gradients or Hessians.

A list containing the following components:

the value of the estimated minimum of `f`

.

the point at which the minimum value of
`f`

is obtained.

the gradient at the estimated minimum of `f`

.

the hessian at the estimated minimum of `f`

(if
requested).

an integer indicating why the optimization process terminated.

- 1:
relative gradient is close to zero, current iterate is probably solution.

- 2:
successive iterates within tolerance, current iterate is probably solution.

- 3:
last global step failed to locate a point lower than

`estimate`

. Either`estimate`

is an approximate local minimum of the function or`steptol`

is too small.- 4:
iteration limit exceeded.

- 5:
maximum step size

`stepmax`

exceeded five consecutive times. Either the function is unbounded below, becomes asymptotic to a finite value from above in some direction or`stepmax`

is too small.

the number of iterations performed.

Note that arguments after `…`

must be matched exactly.

If a gradient or hessian is supplied but evaluates to the wrong mode
or length, it will be ignored if `check.analyticals = TRUE`

(the
default) with a warning. The hessian is not even checked unless the
gradient is present and passes the sanity checks.

The C code for the “perturbed” cholesky, `choldc()`

has
had a bug in all R versions before 3.4.1.

From the three methods available in the original source, we always use method “1” which is line search.

The functions supplied should always return finite (including not
`NA`

and not `NaN`

) values: for the function value itself
non-finite values are replaced by the maximum positive value with a warning.

Dennis, J. E. and Schnabel, R. B. (1983).
*Numerical Methods for Unconstrained Optimization and Nonlinear
Equations*.
Prentice-Hall, Englewood Cliffs, NJ.

Schnabel, R. B., Koontz, J. E. and Weiss, B. E. (1985).
A modular system of algorithms for unconstrained minimization.
*ACM Transactions on Mathematical Software*, **11**, 419--440.
10.1145/6187.6192.

`constrOptim`

for constrained optimization,
`optimize`

for one-dimensional
minimization and `uniroot`

for root finding.
`deriv`

to calculate analytical derivatives.

For nonlinear regression, `nls`

may be better.

```
# NOT RUN {
f <- function(x) sum((x-1:length(x))^2)
nlm(f, c(10,10))
nlm(f, c(10,10), print.level = 2)
utils::str(nlm(f, c(5), hessian = TRUE))
f <- function(x, a) sum((x-a)^2)
nlm(f, c(10,10), a = c(3,5))
f <- function(x, a)
{
res <- sum((x-a)^2)
attr(res, "gradient") <- 2*(x-a)
res
}
nlm(f, c(10,10), a = c(3,5))
## more examples, including the use of derivatives.
# }
# NOT RUN {
demo(nlm)
# }
```

Run the code above in your browser using DataCamp Workspace