mjhjmtu

There do exist general denesting algorithms employing Galois theory, but for the simple case of quadratic algebraic numbers one can employ a simple rule that I discovered as a teenager.

Simple Denesting Rule $rm colorbluesubtract out sqrtnorm:, then colorbrowndivide out sqrttrace $

Recall $rm: w = a + bsqrtn: $ has norm $rm =: w:cdot: w' = (a + bsqrtn) cdot: (a - bsqrtn) =: a^2 - n, b^2 $

and, $ $ furthermore, $rm w:$ has $ $ trace $rm: =: w+w' = (a + bsqrtn) + (a - bsqrtn): =: 2a$

Here $:61-24sqrt5:$ has norm $= 29^2.:$ $rm, colorbluesubtracting out sqrtnorm = 29 $ yields $ 32-24sqrt5:$

and this has $rm sqrttrace: =: 8, thus, colorbrowndividing it out, $ of this yields the sqrt: $,pm( 4,-,3sqrt5).$

For many further examples see my prior posts on denesting.

There are the following identities.
$$sqrta+sqrtb=sqrtfraca+sqrta^2-b2+sqrtfraca-sqrta^2-b2$$ and
$$sqrta-sqrtb=sqrtfraca+sqrta^2-b2-sqrtfraca-sqrta^2-b2,$$
where all numbers under radicals are non-negatives.

For example:
$$sqrt5+2sqrt6=sqrt5+sqrt24=sqrtfrac5+sqrt5^2-242+sqrtfrac5-sqrt5^2-242=sqrt3+sqrt2.$$
This is interesting, when $a$ and $b$ are rationals and $a^2-b$ is a square of a rational number.

The first identity is true because
$$left(sqrtfraca+sqrta^2-b2+sqrtfraca-sqrta^2-b2right)^2=$$
$$=fraca+sqrta^2-b2+fraca-sqrta^2-b2+2sqrtfraca+sqrta^2-b2cdotsqrtfraca-sqrta^2-b2=a+sqrtb.$$

You can derive a formula for $sqrta+bsqrtc$. You will have to assume that $sqrta+bsqrtc$ can be rewritten as the sum of two surds (radicands). So $$sqrta+bsqrtc=sqrtd+sqrte$$

Squaring both sides yields $$a+bsqrtc=d+e+2sqrtde$$

From that, we can see that $a=d+e$ so $e=a-d$ and $bsqrtc=2sqrtderightarrow b^2c=4de$.

Substituting $e$ with $a-d$ gives $b^2c=4d(a-d)$. So $b^2c=4ad-4d^2$. Rearranging the terms gives us $4d^2-4ad+b^2c=0$

Using the Quadratic Equation, we have $$d=frac apmsqrta^2-b^2c2$$

And since $a=d+e$, $e$ is the conjugate of $d$. So $e=frac a-sqrta^2-b^2c2$ and $d=frac a+sqrta^2-b^2c2.,$ Thus

$$sqrta+bsqrtc,=, sqrtfrac a+sqrta^2-b^2c2 +sqrtfrac a-sqrta^2-b^2c2$$

(I will assume $b$ is not a square, since otherwise it would not be a nested radical.)

A nested radical can be denested if and only if there exist $u,vinmathbbN$ such that the nested radical is of the form $sqrtu^2+vpm2usqrtv$ in which case it is also equal to $|upmsqrtv|$.

It's not hard to show that those expressions are equal, which means that all nested radicals of that form can indeed be denested.

For the other way, let's consider the following equality where $a,b,c,d,einmathbbN$:
$$sqrtapmsqrtb=cpm dsqrte$$
(Note that we can also write e.g. $sqrt3-2sqrt2$ in that form as $sqrt3-sqrt8$) If we square both sides, we get:
$$apmsqrtb=c^2+ed^2pm2cdsqrte$$
This suggests we pick $u=c$ and $v=ed^2$. Then $apmsqrtb=u^2+vpm2usqrtv$ as claimed.

This isn't quite the end of the story since preferably we'd also like to know that $a$ actually corresponds to $u^2+v$ and $b$ to $4u^2v$, i.e. that you can't have two equal nested radicals of different forms. Depending on how deep you want to go, you can either take that for granted for now or observe that $apmsqrtb$ satisfies the polynomial relations $(x-a)^2-b=0$ and $(x-u^2-v)^2-4u^2v=0$ and it follows from that that the two polynomials must be equal (using e.g. the concept of minimal polynomials or explicitly dividing the one by the other with remainder taking into account that $apmsqrtb$ will satisfy no linear relation) which gives us our correspondence.

One way of approaching this problem is by viewing it as a zero of an equation. Let me explain. Let's say you want to compute $sqrtx_0$ where $x_0$ is a zero of some quadratic polynomial of the form $x^2-bx+1$. Now, one way to go is to note that if you have a zero of $x^2+ax+1$, then it will still be a zero if you multiply it with $x^2-ax+1$ which equals
$$x^4 + (2-a^2) x^2 + 1$$
Now the idea is to work backwards. So, in particular, if you can find you can find an $a$ such that $b=a^2-2$, then you can conclude that the square root of you polynomial is equal to one of the zeros of the polynomials $x^2-ax+1$ or $x^2+ax+1$. It is usually not too hard to find out which. If you found out which, you can rewrite your square root accordingly to the desired form :)

To conclude, one of the tricks is to find the right form of your polynomials such that you end up with something useful. This method will however require some puzzling.

Edit applying this method to the example $sqrt5+2sqrt6$, you will find that the polynomial you need (thus the one for which you want to calculate the square root of a zero) is $x^2-10x+1$. Then according to the above method (which you derive on the go), your $a=sqrt12$ and then you just need to solve $x^2-ax+1=0$ which is the only possibility since for the other one, filling in a positive number will yield a positive number. Solving this equation by completing the square is not too difficult. It turns out that the zeros lie at around 0.5 and 3. Hence, it is not difficult to note you need the larger zero which turns out to be exactly given by $sqrt2+sqrt3$. Does that make sense?