Forcing FindRoot to return only real solutions

FindRoot documentation reports that if the equation and the initial point are reals, the solutions are searched in the real domain. However, in the following case I get a complex solution

FindRoot[eq, {h, 1.7}]
{h -> -0.990042 - 0.689686 I}


Same for

Assuming[Reals, FindRoot[eq, {h, 1.7}]]


Where eq is the following. What can I do to force Mathematica to return only real solutions?

eq = 0.2888960456513873 \[Sqrt](0.4149486782073932 + (0.5637605604986459 +
h)^2) \[Sqrt](0.4756674436221575 + (0.9900423417575865 +
h)^2) + 2.2076945564517114 h \[Sqrt](0.4149486782073932 +
(0.5637605604986459 +
h)^2) \[Sqrt](0.4756674436221575 + (0.9900423417575865 +
h)^2) + 6.735521430523215 h^2 \[Sqrt](0.4149486782073932 +
(0.5637605604986459 +
h)^2) \[Sqrt](0.4756674436221575 + (0.9900423417575865 +
h)^2) + 10.54153050379155 h^3 \[Sqrt](0.4149486782073932 +
(0.5637605604986459 +
h)^2) \[Sqrt](0.4756674436221575 + (0.9900423417575865 +
h)^2) + 8.93280406600778 h^4 \[Sqrt](0.4149486782073932 +
(0.5637605604986459 +
h)^2) \[Sqrt](0.4756674436221575 + (0.9900423417575865 +
h)^2) + 3.8752891330788843 h^5 \[Sqrt](0.4149486782073932 +
(0.5637605604986459 +
h)^2) \[Sqrt](0.4756674436221575 + (0.9900423417575865 +
h)^2) + 0.666148416209407 h^6 \[Sqrt](0.4149486782073932 +
(0.5637605604986459 +
h)^2) \[Sqrt](0.4756674436221575 + (0.9900423417575865 +
h)^2) + 0.4167441738314279 \[Sqrt](0.4756674436221575 +
(0.9900423417575865 +
h)^2) \[Sqrt](1 - (0.4444444444444444
(-0.022551321792606827 + (0.5637605604986459 +
h)^2)^2)/(0.4149486782073932 + (0.5637605604986459 +
h)^2))

-
NSolve[ eq == 0, h, Reals] yields {} i.e. there are no real solutions. – Artes Jan 23 '13 at 11:27
Same if you Plot[Re[eq], {h, -1, 1}]. – b.gatessucks Jan 23 '13 at 11:31
Thank you. Actually the equation I am using is much longer, it has reals solutions, which I obtained by picking up only the real solutions. To speed up my algorithm I wanted to force findroot itself to return only real solutions – Fabio Dalla Libera Jan 23 '13 at 11:43
With the complete equation I am using, NSolve does not return – Fabio Dalla Libera Jan 23 '13 at 11:45
@FabioDallaLibera You could restrict the region to find solutions, e.g. NSolve[ eq == 0&& -5 < h < 5, h, Reals]. – Artes Jan 23 '13 at 11:51

It will be more convenient to define the function eq this way :

eq[h_] := the formula


Real solutions

For a general technique of using FindRoot the way you would like I recommend to read this post : First positive root. However in order to demonstrate how it works for real numbers we should have a different equation since this one has no real solutions :

NSolve[ eq[h] == 0, h, Reals]

{}


Sometimes it is reasonable to restrict the search for roots to a special region, e.g. if we add an inequality 0 < h < 3 then we don't have to specify the domain Reals because the system understands that h must be real :

NSolve[ eq[h] == 0 && 0 < h < 3, h ]


General remarks on using NSolve you could find here : Solve an equation in R+ since they are valid for Solve as well as for NSolve.

For the function f we can find a global minimum of the real part :

nmin = { h /. #[[2]], #[[1]]}& @ NMinimize[ Re @ eq[h], h]

{-1.83596, 0.100607}


For an idea what we can expect let's plot the real and imaginary parts of f :

Plot[{Re @ eq[h], Im @ eq[h]}, {h, -3, 2}, PlotStyle -> Thick, PlotRange -> {-1, 4},
Epilog -> {Red, PointSize[0.015], Point[nmin]},
PlotLegends -> {Placed["Expressions", {Left, Center}]}]


Even though eq has no real roots we can see that one can expect real solutions e.g. for eq[h] - 3/2 near h == -2 and h == 0, using FindRoot with appropriate starting points we find :

FindRoot[ eq[h] - 3/2, #]& /@ {{h, -2}, {h, 0}}

 {{h -> -2.15793}, {h -> 0.14209}}


For deeper understandig of the behavior of the eq function we should take a look at the complex plane.

Complex solutions

The complex roots can be rewritten as pairs of real numbers :

pts = {Re @ #, Im @ #}& /@ (h /. NSolve[ eq[h] == 0, h])

{{-1.86925, 0.135055}, {-1.86925, -0.135055}, {-1.08654, 0.84989},
{-1.08654, -0.84989}, {-0.990042, 0.689686}, {-0.990042, -0.689686},
{-0.0268207, 0.391104}, {-0.0268207, -0.391104}}


Let's visualize the function eq in the complex plane :

GraphicsRow[
Table[ Show[ ContourPlot @@@ {
{ g[eq[x + I y]], {x, -2.5, 0.5}, {y, -1.5, 1.5},
PlotLabel -> Style[ g[HoldForm @ eq[x + I y]], Blue, 25],
ColorFunction -> "DeepSeaColors", Epilog -> { Darker @ Green, PointSize[0.03],
Point[pts] } },
{ Re @ eq[x + I y] == 0, {x, -2.5, 0.5}, {y, -1.5, 1.5},
ContourStyle -> {Red, Thick}},
{ Im @ eq[x + I y] == 0, {x, -2.5, 0.5}, {y, -1.5, 1.5}, PlotPoints -> 50,
MaxRecursion -> 3, ContourStyle -> {Cyan, Thick}}}],

{g, {Re, Im}} ] ]


The green points denote the complex roots, you can see that there are no real roots. The red lines are solutions to Re[ eq[h] ] == 0 while the cyan lines denote solutions to Im[ eq[h] ] == 0. One can see a that there are some branch cuts on the plots, i.e. the definition of the function eq` is reliable only if we use it in appropriate regions in the complex plane.

-
very nice plots! – chris Jan 23 '13 at 18:39
@chris Thanks. Sometimes plots are crucial to understand what lies beneath. – Artes Jan 24 '13 at 14:19