Proof. Let u be a positive solution to the equation
-\frac{\partial u}{\partial t} + \frac{1}{g(u) (1+u_{x} ^{2} )^{3/2} } \frac{\partial ^{2} u}{\partial x^{2} } = 0,
satisfying u(x,0) = f(x), and define the quasilinear operator \overline{P} by
\overline{P} w = -\frac{\partial w}{\partial t} + \frac{1}{g(u) (1+w_{x} ^{2} )^{3/2} } \frac{\partial ^{2} w}{\partial x^{2} } .
For arbitrary \varepsilon , let
v(x,t) = c + \varepsilon + \varepsilon \frac{t}{t+1} + he^{-wx^{2} } ;
for now, h> Me^{R_{0} ^{2} } and w\in (0,1), but further refinements will be made. Is it possible to choose these parameters in such a way that v becomes a strict supersolution? This will be the case if
-\varepsilon \frac{1}{(t+1)^{2} } + \frac{1}{g(u) (1+v_{x} ^{2} )^{3/2} } \frac{\partial ^{2} v}{\partial x^{2} } < 0.
It will be enough if
\frac{1}{g(u) (1+ v_{x} ^{2} )^{3/2} } \frac{\partial ^{2} v}{\partial x^{2} } < \varepsilon \frac{1}{(T+1)^{2} } .
Whenever \frac{d^{2} }{dx^{2} } (he^{-wx^{2} } ) \leq 0, we have
\frac{1}{g(u) (1+ v_{x} ^{2} )^{3/2} } \frac{\partial ^{2} v}{\partial x^{2} } \leq 0 < \varepsilon \frac{1}{(T+1)^{2} } ,
and when \frac{d^{2} }{dx^{2} } (he^{-wx^{2} } ) \geq 0,
\frac{1}{g(u) (1+v_{x} ^{2} )^{3/2} } \frac{\partial ^{2} v}{\partial x^{2} } \leq \frac{1}{g(u) } \frac{d^{2} }{dx^{2} } (he^{-wx^{2} } ) \leq \frac{4}{m} \frac{d^{2} }{dx^{2} } (he^{-wx^{2} } ) ,
remembering that g(u) \geq m/4 = (1/4) \min f . So, it will be enough to make
\frac{4}{m} \frac{d^{2} }{dx^{2} } (he^{-wx^{2} } ) < \varepsilon \frac{1}{(T+1)^{2} } .
We already calculated above that
\frac{d^{2} }{dx^{2} } (he^{-wx^{2} } ) \leq 4w he^{-3/2} ,
which means that we want
4whe^{-3/2} \leq \frac{m}{4} \varepsilon \frac{1}{(T+1)^{2} } .
For concreteness, choose h = 2M e^{R_{0} ^{2} } . Then the above inequality holds if w \in (0,1) and also satisfies
w < \varepsilon \frac{m}{4} \frac{1}{(T+1)^{2} } \frac{1}{8M} e^{3/2 - R_{0} ^{2} } .
With these choices, \overline{P} v <0, so that \overline{P} u - \overline{P} v = Pu -\overline{P} v >0, and this holds at all points in \mathbb{R} \times (0,T). Also, at t=0,
u(x,0) - v(x,0) = f(x) -(c+2\varepsilon + he^{-wx^{2} } ) < -\varepsilon <0.
We now apply the technique explained in Krylov’s book [16], which was also employed earlier, to obtain the bound on \mid u\mid .
Since u \in H_{2+\alpha } (\mathbb{R} \times (0,T) ), there exists a positive number a such that \mid u(x,t) \mid < a for all (x,t) \in \mathbb{R} \times (0,T). For this a and for each R >0, define
v_{R} (x,t) = c + \varepsilon + \varepsilon \frac{t}{t+1} + he^{-wx^{2} } + 2\frac{a}{\cosh R} e^{\lambda t} \cosh x.
At t=0, we have u(x,0) < c+ he^{-wx^{2} } < v_{R} (x,0). What is true at \mid x\mid = R? At these points, we have
\begin{aligned}
u(\pm R,t) - v_{R} (\pm R,t) & = & u(\pm R,t) - \left( c+ \varepsilon + \varepsilon \frac{t}{t+1} + he^{-wR^{2} } \right) - 2ae^{\lambda t} \\
& = & (u(\pm R,t) -ae^{\lambda t} ) - \left( c +\varepsilon + \varepsilon \frac{t}{t+1} + he^{-wR^{2} } + a e^{\lambda t} \right) .
\end{aligned}
Since \mid u(x,t) \mid \leq a for all (x,t) \in \mathbb{R} \times (0,T), then in the first term,
u(\pm R,t) -a e^{\lambda t} \leq a -a e^{\lambda t} = a(1-e^{\lambda t} ) \leq 0,
so long as \lambda \geq 0. Therefore, (u-v_{R} )(\pm R,t) \leq 0.
We calculate that
\begin{aligned}
\overline{P} v_{R} & = & -\varepsilon \frac{1}{(t+1)^{2} } - 2 \lambda \frac{a e^{\lambda t} }{\cosh R} \cosh x \\
& + & \frac{1}{g(u) (1+(v_{R} )_{x} )^{3/2} } \left( \frac{d^{2} }{dx^{2} } he^{-wx^{2} } + 2\frac{a}{\cosh R} e^{\lambda t} \cosh x \right) .
\end{aligned}
Grouping the terms,
\frac{1}{g(u) (1+(v_{R} )_{x} ^{2} )^{3/2} } \frac{d^{2} }{dx^{2} } he^{-wx^{2} } \leq 0 < \varepsilon \frac{1}{(t+1)^{2} }
where d^{2} /dx^{2} he^{-wx^{2} } \leq 0, and
\begin{aligned}
\frac{1}{g(u) (1+ (v_{R} )_{x} ^{2} )^{3/2} } \frac{d^{2} }{dx^{2} } he^{-wx^{2} } & \leq & \frac{1}{g(u) } \frac{d^{2} }{dx^{2} } he^{-wx^{2} } \\
& \leq & \frac{4}{m} \frac{d^{2} }{dx^{2} } he^{-wx^{2} } \\
& \leq & \varepsilon \frac{1}{(t+1)^{2} }
\end{aligned}
when d^{2} /dx^{2} he^{-wx^{2} } \geq 0, by the choices of h and w already made. In the other two terms,
\frac{1}{g(u) (1+(v_{R} )_{x} ^{2} )^{3/2} } \leq \frac{4}{m} ,
and so choosing \lambda > 4/m makes
2 \frac{a e^{\lambda t} }{\cosh R} \cosh x \left( -\lambda + \frac{1}{g(u) (1+(v_{R} )_{x} ^{2} )^{3/2} } \right) < 0.
Adding these up, we conclude that \overline{P} v_{R} < 0, and so \overline{P} u - \overline{P} v_{R} = Pu - \overline{P} v_{R} >0, and this is true in all of \mathbb{R} \times (0,T) .
Using the same sign analysis as in the bound on \mid u \mid , with the function v(x,t) playing the role of M, it is possible to show that in the region B_{R} (0) \times (0,T) , the function u-v_{R} cannot attain a maximum away from the parabolic boundary. This implies that for all (x,t) \in B_{R} (0) \times (0,T), that
u(x,t) \leq c+ \varepsilon + \varepsilon \frac{t}{t+1} + he^{-wx^{2} } < c+2\varepsilon + h e^{-wx^{2} } + 2\frac{a}{\cosh R} e^{\lambda t} \cosh x.
Choose (x,t) \in \mathbb{R} \times (0,T). For all R > \mid x\mid , the above inequality holds. Therefore,
u(x,t) \leq c +\varepsilon + \varepsilon \frac{t}{t+1} + he^{-wx^{2} } < c+2\varepsilon + he^{-wx^{2} } .
Since for all \varepsilon > 0, it is possible to find K >0 with he^{-wx^{2} } < \varepsilon for \mid x\mid > K and t\in (0,T), this also shows that there exists K>0 such that for all \mid x\mid > K and t\in (0,T), that
u(x,t) \leq c + 3\varepsilon .
By the same reasoning, we can show that for \mid x\mid large enough, u(x,t) >c-3\varepsilon . Therefore, u \rightarrow c as \mid x\mid \rightarrow \infty . Since the choice of w that works for T also works for all t\in (0,T), there is a single choice of K >0 such that
\mid u(x,t) -c\mid < 3\varepsilon
for all (x,t) \in ((-\infty ,-K) \cup (K, \infty ) ) \times (0,T). This completes the proof of the lemma.