Can surds upon surds be simplified?

[Mentallic]

For some surds inside of surds, they can be converted into a more simple form:

\sqrt{a+b\sqrt{c}}=e+f\sqrt{g}

Such as: \sqrt{11-6\sqrt{2}}=3-\sqrt{2}

However, there are some that cannot be simplified into this form (as far as I know).

Such as: \sqrt{3+\sqrt{7}}

I am curious to know if there is fast method in realizing whether these types of equations can be simplified.
My only way of knowing so far is as follows:

To see if \sqrt{31+12\sqrt{3}} can be simplified, first I let it be equal to some general simplified form:

\sqrt{31+12\sqrt{3}}=a+b\sqrt{3}

squaring both sides:

a^2+3b^2+2\sqrt{3}ab=31+12\sqrt{3}

Therefore, a^2+3b^2=31 (1) and
2\sqrt{3}ab=12\sqrt{3} (2)

Making a or b the subject in (2)
b=\frac{6}{a}

Substituting into (1)

a^2+3(\frac{36}{a^2})=31

a^4-31a^2+108=0

Now we have a quadratic in a^2. I will now know from the quadratic formula if this expression can be simplified or not by looking at the discriminant. If it is a perfect square, then it can be simplified, else, it cannot be.

 

[Gib Z]

I think the discriminant is actually the quickest way, unfortunately. However, I think you might be interested in this anyway: http://www.cybertester.com/data/denest.pdf

 

[Mentallic]

Thanks for the link

For simplicities sake, I have concluded from reading through the site that to quickly check if such questions as I have posed can be denested (this is the term used), I can shorten the procedure by checking:

Given \sqrt{A \pm B} where A,B all reals, A+B>0

Checking to see if \sqrt{A^2-B^2} is a rational number will give me the indication whether to pursue the simplified answer.

Of course there is a wider array of problems, but I am happy with the progress made for the moment.