
Why is the Out[74] not {1 - a, {x -> -1}}?
Maximize[{Sqrt[x^2] - a, -1 <= x <= 1}, x]
{1 - a, {x -> -1}}

Why is the Out[74] not {1 - a, {x -> -1}}?
Maximize[{Sqrt[x^2] - a, -1 <= x <= 1}, x]
{1 - a, {x -> -1}}
I suspect the reason this fails is because Mathematica does not know the derivative of Abs[x]. The reason behind this is Abs[x] is nowhere differentiable in the complex plane and Derivative and D take complex derivatives.
In Wolfram Alpha, we had to handle Abs ourselves assuming real variables since most people would want that. If you run WolframAlpha["derivative of Abs[x]"], you'll see the step-by-step solution specifies x is a real variable.
Since Derivative is not a protected symbol, we can add custom SubValues:
Derivative[1][Abs] = Sign;
Derivative[2][Abs] = Derivative[1][Sign] = DiracDelta;
Now Mathematica knows the (real) derivatives of these functions:
D[Abs[x], x]
(* Sign[x] *)
D[Abs[x], {x, 2}]
(* DiracDelta[x] *)
D[Sign[x], x]
(* DiracDelta[x] *)
Unfortunately this does not fix your Maximize issue:
Maximize[{Abs[x] - a, -1 <= x <= 1}, x]
(* Maximize[{-a + Abs[x], -1 <= x <= 1}, x] *)
Maximize[{(Abs[x] - a), -1 <= x <= 1}, {x, a}]and see what you get. It is not a but. – ciao Jun 19 '14 at 04:251 - a? It's not clear to me why MMA would return a symbolic maximum in the other instances and not here. – mfvonh Jun 19 '14 at 04:47Out[1]= {[Infinity], {x -> Indeterminate, a -> Indeterminate}}
– fptan Jun 19 '14 at 09:14Out[2]= Maximize[{-a + Abs[x], -1 <= x <= 1}, x]
– fptan Jun 19 '14 at 09:15Out[3]= Maximize[{-a + Abs[x], -1 <= x <= 1}, x, Reals]
– fptan Jun 19 '14 at 09:16