Here you go: https://math.stackexchange.com/a/3393408/633100
I'll actually Paste the same answer here (I tried sharing link alone but apparently answers have to be longer than 30 characters)
$d(xy)=(x+dx)(y+dy)-xy \ $ [from Definition of a "differential"]
$\ \ \ \ \ \ \ \ \ =ydx + xdy + dxdy $
Here what people normally do is, "$dxdy$ is too small and negligible". But that is SO BAD.
Instead, go back to definition of the derivative.
$\lim_{ h -> 0 }\frac{f(x+h)+f(x)}{h}$
Divide the first equation we got by $dx$
$(xy)' = y + x(y') + dy $
$dy -> 0$ by definition of a derivative/a differential.
Another way of looking at this would be,
Integrate both sides and all the terms will become bigger values except this $dxdy$ will alone have to be integrated twice to give it a good value.
for any quantity z, dz is defined as the infinitesimally small change in that quantity. ie approaches 0. These are generally used in combination with one or more 'differentials' and so we normally don't cancel of these to equal 0. for instance, $\frac{dv}{dt}$ is not 0/0. But with $dxdy$ if you integrate the whole equation, all the terms of the equation move from being differentials to regular terms but except this last one. There are no other differentials in the equation which makes it legal to cancel the last one to 0.
At least that's how I see it.