Each result cited below is universally quantified over the data in its own statement.
For x ∈ X x\in X x ∈ X let S x = { d ( x , k ) : k ∈ K } S_{x}=\{d(x,k):k\in K\} S x = { d ( x , k ) : k ∈ K } , so that D ( x ) = inf S x D(x)=\inf S_{x} D ( x ) = inf S x by Distance from a Point to a Nonempty Subset of a Metric Space ; S x S_{x} S x is nonempty because K K K is, and bounded below by 0 0 0 . For a real a a a we write [ a ] − = max { − a , 0 } [a]_{-}=\max\{-a,0\} [ a ] − = max { − a , 0 } as in Local Slope, Super-Slope, Sub-Slope and Upper Slope Envelope of a Locally Lipschitz Function on a Metric Space , so that [ ψ ( y ) − ψ ( x ) ] − = max { ψ ( x ) − ψ ( y ) , 0 } [\psi(y)-\psi(x)]_{-}=\max\{\psi(x)-\psi(y),0\} [ ψ ( y ) − ψ ( x ) ] − = max { ψ ( x ) − ψ ( y ) , 0 } ; by claim 1 of Elementary Properties of the Maximum of Two Elements , ψ ( x ) − ψ ( y ) ≤ [ ψ ( y ) − ψ ( x ) ] − \psi(x)-\psi(y)\le[\psi(y)-\psi(x)]_{-} ψ ( x ) − ψ ( y ) ≤ [ ψ ( y ) − ψ ( x ) ] − . The slope pair of a locally Lipschitz ψ \psi ψ at x x x used below is always ( ∣ ∇ − ψ ∣ ( x ) , [ ⋅ ] − ) (|\nabla^{-}\psi|(x),[\cdot]_{-}) ( ∣ ∇ − ψ ∣ ( x ) , [ ⋅ ] − ) , in the sense of Elementary Properties of Local Slopes: Order, Negation, Bounds from Difference Quotients, Lipschitz Functions and Squared Distances .
Step 1 (properties of D D D ). By claims 1 to 4 of The Distance to a Set is Nonexpansive (with A = K A=K A = K ), for all x , y ∈ X x,y\in X x , y ∈ X and every k ∈ K k\in K k ∈ K ,
0 ≤ D ( x ) , D ( x ) ≤ d ( x , k ) , D ( x ) ≤ d ( x , y ) + D ( y ) , ∣ D ( x ) − D ( y ) ∣ ≤ d ( x , y ) . (1.1) 0\le D(x),\qquad D(x)\le d(x,k),\qquad D(x)\le d(x,y)+D(y),\qquad |D(x)-D(y)|\le d(x,y). \tag{1.1} 0 ≤ D ( x ) , D ( x ) ≤ d ( x , k ) , D ( x ) ≤ d ( x , y ) + D ( y ) , ∣ D ( x ) − D ( y ) ∣ ≤ d ( x , y ) . ( 1.1 )
Step 2 (points of almost steepest descent of D D D ). We show: if x ∈ X x\in X x ∈ X satisfies 0 < D ( x ) 0<D(x) 0 < D ( x ) , then for all reals ε ′ > 0 \varepsilon'>0 ε ′ > 0 and R > 0 R>0 R > 0 there is y ∈ X y\in X y ∈ X with
0 < d ( x , y ) < R and ( 1 − ε ′ ) d ( x , y ) ≤ D ( x ) − D ( y ) . (2.1) 0<d(x,y)<R\qquad\text{and}\qquad(1-\varepsilon')\,d(x,y)\le D(x)-D(y). \tag{2.1} 0 < d ( x , y ) < R and ( 1 − ε ′ ) d ( x , y ) ≤ D ( x ) − D ( y ) . ( 2.1 )
The choices are made in the order t t t , k k k , y y y . Let t = min { 1 / 2 , R / ( 2 ( 1 + ε ′ ) D ( x ) ) } t=\min\{1/2,\,R/(2(1+\varepsilon')D(x))\} t = min { 1/2 , R / ( 2 ( 1 + ε ′ ) D ( x ))} , so that 0 < t ≤ 1 0<t\le1 0 < t ≤ 1 , and let η = ε ′ t D ( x ) \eta=\varepsilon' t\,D(x) η = ε ′ t D ( x ) , a positive real. By claim 4 of Approximation Property of the Supremum and the Infimum in R \mathbb{R} R , applied to S x S_{x} S x with η \eta η , there is k ∈ K k\in K k ∈ K with d ( x , k ) < D ( x ) + η d(x,k)<D(x)+\eta d ( x , k ) < D ( x ) + η ; by (1.1), 0 < D ( x ) ≤ d ( x , k ) 0<D(x)\le d(x,k) 0 < D ( x ) ≤ d ( x , k ) . By Metric Space with Interpolation Points §interpolation there is an interpolation point y y y of x x x and k k k at parameter t t t : d ( x , y ) ≤ t d ( x , k ) d(x,y)\le t\,d(x,k) d ( x , y ) ≤ t d ( x , k ) and d ( y , k ) ≤ ( 1 − t ) d ( x , k ) d(y,k)\le(1-t)\,d(x,k) d ( y , k ) ≤ ( 1 − t ) d ( x , k ) . The triangle inequality gives d ( x , k ) ≤ d ( x , y ) + d ( y , k ) ≤ d ( x , y ) + ( 1 − t ) d ( x , k ) d(x,k)\le d(x,y)+d(y,k)\le d(x,y)+(1-t)\,d(x,k) d ( x , k ) ≤ d ( x , y ) + d ( y , k ) ≤ d ( x , y ) + ( 1 − t ) d ( x , k ) , so t d ( x , k ) ≤ d ( x , y ) t\,d(x,k)\le d(x,y) t d ( x , k ) ≤ d ( x , y ) , and therefore d ( x , y ) = t d ( x , k ) > 0 d(x,y)=t\,d(x,k)>0 d ( x , y ) = t d ( x , k ) > 0 . Moreover d ( x , y ) < t ( 1 + ε ′ t ) D ( x ) ≤ t ( 1 + ε ′ ) D ( x ) ≤ R / 2 < R d(x,y)<t\,(1+\varepsilon' t)\,D(x)\le t\,(1+\varepsilon')\,D(x)\le R/2<R d ( x , y ) < t ( 1 + ε ′ t ) D ( x ) ≤ t ( 1 + ε ′ ) D ( x ) ≤ R /2 < R . Finally, by (1.1), D ( y ) ≤ d ( y , k ) ≤ ( 1 − t ) d ( x , k ) D(y)\le d(y,k)\le(1-t)\,d(x,k) D ( y ) ≤ d ( y , k ) ≤ ( 1 − t ) d ( x , k ) , so
D ( x ) − D ( y ) ≥ t d ( x , k ) − ( d ( x , k ) − D ( x ) ) > d ( x , y ) − η = d ( x , y ) − ε ′ t D ( x ) ≥ d ( x , y ) − ε ′ t d ( x , k ) = ( 1 − ε ′ ) d ( x , y ) . D(x)-D(y)\ge t\,d(x,k)-\bigl(d(x,k)-D(x)\bigr)>d(x,y)-\eta=d(x,y)-\varepsilon' t\,D(x)\ge d(x,y)-\varepsilon' t\,d(x,k)=(1-\varepsilon')\,d(x,y). D ( x ) − D ( y ) ≥ t d ( x , k ) − ( d ( x , k ) − D ( x ) ) > d ( x , y ) − η = d ( x , y ) − ε ′ t D ( x ) ≥ d ( x , y ) − ε ′ t d ( x , k ) = ( 1 − ε ′ ) d ( x , y ) .
Part 1. Here Ω = X ∖ K \Omega=X\setminus K Ω = X ∖ K is open, and by The Eikonal Equation on an Open Subset of a Metric Space the eikonal equation ∣ ∇ u ∣ = 1 |\nabla u|=\mathbf{1} ∣∇ u ∣ = 1 in Ω \Omega Ω is H = 0 H=0 H = 0 with H ( x , r , p ) = p − 1 H(x,r,p)=p-1 H ( x , r , p ) = p − 1 . Write D Ω D_{\Omega} D Ω for the restriction of D D D to Ω \Omega Ω .
Step 3 (semicontinuity). Let x ∈ Ω x\in\Omega x ∈ Ω and τ \tau τ a positive real. For every y ∈ Ω y\in\Omega y ∈ Ω with d ( x , y ) < τ d(x,y)<\tau d ( x , y ) < τ , (1.1) gives ∣ D ( y ) − D ( x ) ∣ ≤ d ( x , y ) < τ |D(y)-D(x)|\le d(x,y)<\tau ∣ D ( y ) − D ( x ) ∣ ≤ d ( x , y ) < τ , hence D ( y ) < D ( x ) + τ D(y)<D(x)+\tau D ( y ) < D ( x ) + τ and D ( x ) − τ < D ( y ) D(x)-\tau<D(y) D ( x ) − τ < D ( y ) . So D Ω D_{\Omega} D Ω is upper and lower semicontinuous on Ω \Omega Ω .
Step 4 (Part 1, subsolution). Let ψ 1 ∈ C ‾ ( Ω ) \psi_{1}\in\underline{\mathcal{C}}(\Omega) ψ 1 ∈ C ( Ω ) , let ψ 2 : Ω → R \psi_{2}:\Omega\to\mathbb{R} ψ 2 : Ω → R be locally Lipschitz on Ω \Omega Ω , and let x ∈ Ω x\in\Omega x ∈ Ω be a point at which D Ω − ψ 1 − ψ 2 D_{\Omega}-\psi_{1}-\psi_{2} D Ω − ψ 1 − ψ 2 has a local maximum relative to Ω \Omega Ω , with a radius r 1 > 0 r_{1}>0 r 1 > 0 . For y ∈ Ω y\in\Omega y ∈ Ω with d ( x , y ) < r 1 d(x,y)<r_{1} d ( x , y ) < r 1 we have D ( y ) − ψ 1 ( y ) − ψ 2 ( y ) ≤ D ( x ) − ψ 1 ( x ) − ψ 2 ( x ) D(y)-\psi_{1}(y)-\psi_{2}(y)\le D(x)-\psi_{1}(x)-\psi_{2}(x) D ( y ) − ψ 1 ( y ) − ψ 2 ( y ) ≤ D ( x ) − ψ 1 ( x ) − ψ 2 ( x ) , so by (1.1)
ψ 1 ( x ) − ψ 1 ( y ) ≤ D ( x ) − D ( y ) + ψ 2 ( y ) − ψ 2 ( x ) ≤ d ( x , y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ . \psi_{1}(x)-\psi_{1}(y)\le D(x)-D(y)+\psi_{2}(y)-\psi_{2}(x)\le d(x,y)+|\psi_{2}(y)-\psi_{2}(x)| . ψ 1 ( x ) − ψ 1 ( y ) ≤ D ( x ) − D ( y ) + ψ 2 ( y ) − ψ 2 ( x ) ≤ d ( x , y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣.
The right-hand side is nonnegative, so claim 3 of Elementary Properties of the Maximum of Two Elements gives [ ψ 1 ( y ) − ψ 1 ( x ) ] − ≤ d ( x , y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ ≤ ( 1 + ε ) d ( x , y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ [\psi_{1}(y)-\psi_{1}(x)]_{-}\le d(x,y)+|\psi_{2}(y)-\psi_{2}(x)|\le(1+\varepsilon)\,d(x,y)+|\psi_{2}(y)-\psi_{2}(x)| [ ψ 1 ( y ) − ψ 1 ( x ) ] − ≤ d ( x , y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ ≤ ( 1 + ε ) d ( x , y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ for every real ε > 0 \varepsilon>0 ε > 0 and every y ∈ Ω y\in\Omega y ∈ Ω with 0 < d ( x , y ) < r 1 0<d(x,y)<r_{1} 0 < d ( x , y ) < r 1 . Hence Elementary Properties of Local Slopes: Order, Negation, Bounds from Difference Quotients, Lipschitz Functions and Squared Distances §upper-bound , applied with ψ = ψ 1 \psi=\psi_{1} ψ = ψ 1 , φ = ψ 2 \varphi=\psi_{2} φ = ψ 2 , c = 1 c=1 c = 1 and r = r 1 r=r_{1} r = r 1 for every ε \varepsilon ε , gives ∣ ∇ − ψ 1 ∣ ( x ) ≤ 1 + ∣ ∇ ψ 2 ∣ ( x ) |\nabla^{-}\psi_{1}|(x)\le1+|\nabla\psi_{2}|(x) ∣ ∇ − ψ 1 ∣ ( x ) ≤ 1 + ∣∇ ψ 2 ∣ ( x ) . Since ψ 1 ∈ C ‾ ( Ω ) \psi_{1}\in\underline{\mathcal{C}}(\Omega) ψ 1 ∈ C ( Ω ) , Test Classes for Slope-Based Viscosity Solutions on a Metric Space §sub-class gives ∣ ∇ ψ 1 ∣ ( x ) = ∣ ∇ − ψ 1 ∣ ( x ) |\nabla\psi_{1}|(x)=|\nabla^{-}\psi_{1}|(x) ∣∇ ψ 1 ∣ ( x ) = ∣ ∇ − ψ 1 ∣ ( x ) , and Elementary Properties of Local Slopes: Order, Negation, Bounds from Difference Quotients, Lipschitz Functions and Squared Distances §order gives ∣ ∇ ψ 2 ∣ ( x ) ≤ ∣ ∇ ψ 2 ∣ ∗ ( x ) |\nabla\psi_{2}|(x)\le|\nabla\psi_{2}|^{*}(x) ∣∇ ψ 2 ∣ ( x ) ≤ ∣∇ ψ 2 ∣ ∗ ( x ) ; so ∣ ∇ ψ 1 ∣ ( x ) − ∣ ∇ ψ 2 ∣ ∗ ( x ) ≤ 1 |\nabla\psi_{1}|(x)-|\nabla\psi_{2}|^{*}(x)\le1 ∣∇ ψ 1 ∣ ( x ) − ∣∇ ψ 2 ∣ ∗ ( x ) ≤ 1 . As also 0 ≤ 1 0\le1 0 ≤ 1 , claim 3 of Elementary Properties of the Maximum of Two Elements gives max { ∣ ∇ ψ 1 ∣ ( x ) − ∣ ∇ ψ 2 ∣ ∗ ( x ) , 0 } ≤ 1 \max\{|\nabla\psi_{1}|(x)-|\nabla\psi_{2}|^{*}(x),0\}\le1 max { ∣∇ ψ 1 ∣ ( x ) − ∣∇ ψ 2 ∣ ∗ ( x ) , 0 } ≤ 1 , that is, H ( x , D ( x ) , max { ∣ ∇ ψ 1 ∣ ( x ) − ∣ ∇ ψ 2 ∣ ∗ ( x ) , 0 } ) ≤ 0 H\bigl(x,D(x),\max\{|\nabla\psi_{1}|(x)-|\nabla\psi_{2}|^{*}(x),0\}\bigr)\le0 H ( x , D ( x ) , max { ∣∇ ψ 1 ∣ ( x ) − ∣∇ ψ 2 ∣ ∗ ( x ) , 0 } ) ≤ 0 . Together with Step 3, D Ω D_{\Omega} D Ω is an s-subsolution of H = 0 H=0 H = 0 in Ω \Omega Ω by Slope-Based Viscosity Subsolutions, Supersolutions and Solutions on a Metric Space §subsolution .
Step 5 (Part 1, supersolution). Let ψ 1 ∈ C ‾ ( Ω ) \psi_{1}\in\overline{\mathcal{C}}(\Omega) ψ 1 ∈ C ( Ω ) , let ψ 2 : Ω → R \psi_{2}:\Omega\to\mathbb{R} ψ 2 : Ω → R be locally Lipschitz on Ω \Omega Ω , and let x ∈ Ω x\in\Omega x ∈ Ω be a point at which D Ω − ψ 1 − ψ 2 D_{\Omega}-\psi_{1}-\psi_{2} D Ω − ψ 1 − ψ 2 has a local minimum relative to Ω \Omega Ω , with a radius r 1 > 0 r_{1}>0 r 1 > 0 . Since Ω \Omega Ω is open , there is a real r 2 > 0 r_{2}>0 r 2 > 0 with B d ( x , r 2 ) ⊆ Ω B_{d}(x,r_{2})\subseteq\Omega B d ( x , r 2 ) ⊆ Ω . No k ∈ K k\in K k ∈ K lies in Ω \Omega Ω , so no k ∈ K k\in K k ∈ K lies in the open ball B d ( x , r 2 ) B_{d}(x,r_{2}) B d ( x , r 2 ) , that is, r 2 ≤ d ( x , k ) r_{2}\le d(x,k) r 2 ≤ d ( x , k ) for every k ∈ K k\in K k ∈ K ; thus r 2 r_{2} r 2 is a lower bound of S x S_{x} S x , and since D ( x ) D(x) D ( x ) is the greatest lower bound of S x S_{x} S x , 0 < r 2 ≤ D ( x ) 0<r_{2}\le D(x) 0 < r 2 ≤ D ( x ) .
We verify the hypothesis of Elementary Properties of Local Slopes: Order, Negation, Bounds from Difference Quotients, Lipschitz Functions and Squared Distances §lower-bound for ψ = ψ 1 \psi=\psi_{1} ψ = ψ 1 , φ = ψ 2 \varphi=\psi_{2} φ = ψ 2 and c = 1 c=1 c = 1 . Let ε > 0 \varepsilon>0 ε > 0 and r > 0 r>0 r > 0 be reals. Step 2, applied with ε ′ = ε \varepsilon'=\varepsilon ε ′ = ε and R = min { r , r 1 , r 2 } R=\min\{r,r_{1},r_{2}\} R = min { r , r 1 , r 2 } , gives y ∈ X y\in X y ∈ X with 0 < d ( x , y ) < R 0<d(x,y)<R 0 < d ( x , y ) < R and ( 1 − ε ) d ( x , y ) ≤ D ( x ) − D ( y ) (1-\varepsilon)\,d(x,y)\le D(x)-D(y) ( 1 − ε ) d ( x , y ) ≤ D ( x ) − D ( y ) ; then y ∈ B d ( x , r 2 ) ⊆ Ω y\in B_{d}(x,r_{2})\subseteq\Omega y ∈ B d ( x , r 2 ) ⊆ Ω and d ( x , y ) < r 1 d(x,y)<r_{1} d ( x , y ) < r 1 , so D ( x ) − ψ 1 ( x ) − ψ 2 ( x ) ≤ D ( y ) − ψ 1 ( y ) − ψ 2 ( y ) D(x)-\psi_{1}(x)-\psi_{2}(x)\le D(y)-\psi_{1}(y)-\psi_{2}(y) D ( x ) − ψ 1 ( x ) − ψ 2 ( x ) ≤ D ( y ) − ψ 1 ( y ) − ψ 2 ( y ) , and therefore
[ ψ 1 ( y ) − ψ 1 ( x ) ] − ≥ ψ 1 ( x ) − ψ 1 ( y ) ≥ D ( x ) − D ( y ) − ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ ≥ ( 1 − ε ) d ( x , y ) − ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ . [\psi_{1}(y)-\psi_{1}(x)]_{-}\ge\psi_{1}(x)-\psi_{1}(y)\ge D(x)-D(y)-|\psi_{2}(y)-\psi_{2}(x)|\ge(1-\varepsilon)\,d(x,y)-|\psi_{2}(y)-\psi_{2}(x)| . [ ψ 1 ( y ) − ψ 1 ( x ) ] − ≥ ψ 1 ( x ) − ψ 1 ( y ) ≥ D ( x ) − D ( y ) − ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ ≥ ( 1 − ε ) d ( x , y ) − ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣.
So Elementary Properties of Local Slopes: Order, Negation, Bounds from Difference Quotients, Lipschitz Functions and Squared Distances §lower-bound gives 1 − ∣ ∇ ψ 2 ∣ ( x ) ≤ ∣ ∇ − ψ 1 ∣ ( x ) 1-|\nabla\psi_{2}|(x)\le|\nabla^{-}\psi_{1}|(x) 1 − ∣∇ ψ 2 ∣ ( x ) ≤ ∣ ∇ − ψ 1 ∣ ( x ) . By Elementary Properties of Local Slopes: Order, Negation, Bounds from Difference Quotients, Lipschitz Functions and Squared Distances §order , ∣ ∇ − ψ 1 ∣ ( x ) ≤ ∣ ∇ ψ 1 ∣ ( x ) |\nabla^{-}\psi_{1}|(x)\le|\nabla\psi_{1}|(x) ∣ ∇ − ψ 1 ∣ ( x ) ≤ ∣∇ ψ 1 ∣ ( x ) and ∣ ∇ ψ 2 ∣ ( x ) ≤ ∣ ∇ ψ 2 ∣ ∗ ( x ) |\nabla\psi_{2}|(x)\le|\nabla\psi_{2}|^{*}(x) ∣∇ ψ 2 ∣ ( x ) ≤ ∣∇ ψ 2 ∣ ∗ ( x ) , so 1 ≤ ∣ ∇ ψ 1 ∣ ( x ) + ∣ ∇ ψ 2 ∣ ∗ ( x ) 1\le|\nabla\psi_{1}|(x)+|\nabla\psi_{2}|^{*}(x) 1 ≤ ∣∇ ψ 1 ∣ ( x ) + ∣∇ ψ 2 ∣ ∗ ( x ) , that is, H ( x , D ( x ) , ∣ ∇ ψ 1 ∣ ( x ) + ∣ ∇ ψ 2 ∣ ∗ ( x ) ) ≥ 0 H\bigl(x,D(x),|\nabla\psi_{1}|(x)+|\nabla\psi_{2}|^{*}(x)\bigr)\ge0 H ( x , D ( x ) , ∣∇ ψ 1 ∣ ( x ) + ∣∇ ψ 2 ∣ ∗ ( x ) ) ≥ 0 . (Only local Lipschitz continuity of ψ 1 \psi_{1} ψ 1 was used.) Together with Step 3, D Ω D_{\Omega} D Ω is an s-supersolution of H = 0 H=0 H = 0 in Ω \Omega Ω by Slope-Based Viscosity Subsolutions, Supersolutions and Solutions on a Metric Space §supersolution , and with Step 4 an s-solution by Slope-Based Viscosity Subsolutions, Supersolutions and Solutions on a Metric Space §solution . This proves claim 1.
Part 2. Here Ω = X \Omega=X Ω = X , u ( x ) = a D ( x ) 2 u(x)=a\,D(x)^{2} u ( x ) = a D ( x ) 2 and f ( x ) = b D ( x ) 2 f(x)=b\,D(x)^{2} f ( x ) = b D ( x ) 2 , and by The Discounted Stationary Hopf-Lax Equation on an Open Subset of a Metric Space the equation is H = 0 H=0 H = 0 with H ( x , r , p ) = ρ r + 1 2 p 2 − b D ( x ) 2 H(x,r,p)=\rho\,r+\frac{1}{2}p^{2}-b\,D(x)^{2} H ( x , r , p ) = ρ r + 2 1 p 2 − b D ( x ) 2 , a Hamiltonian on X X X ; by Slope-Based Viscosity Subsolutions, Supersolutions and Solutions on a Metric Space §hamiltonian , H ( x , r , p ) ≤ H ( x , r , p ′ ) H(x,r,p)\le H(x,r,p') H ( x , r , p ) ≤ H ( x , r , p ′ ) whenever p , p ′ ∈ T p,p'\in T p , p ′ ∈ T and p ≤ p ′ p\le p' p ≤ p ′ .
Step 6 (the constant a a a ). Since 0 ≤ ρ 2 0\le\rho^{2} 0 ≤ ρ 2 and 0 ≤ b 0\le b 0 ≤ b , the real ρ 2 + 8 b \rho^{2}+8b ρ 2 + 8 b is nonnegative, and by Existence and Uniqueness of the Nonnegative Square Root s = ρ 2 + 8 b s=\sqrt{\rho^{2}+8b} s = ρ 2 + 8 b is the unique real with 0 ≤ s 0\le s 0 ≤ s and s 2 = ρ 2 + 8 b s^{2}=\rho^{2}+8b s 2 = ρ 2 + 8 b . As ρ 2 ≤ s 2 \rho^{2}\le s^{2} ρ 2 ≤ s 2 with 0 ≤ ρ 0\le\rho 0 ≤ ρ and 0 ≤ s 0\le s 0 ≤ s , claim 2 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field gives ρ ≤ s \rho\le s ρ ≤ s , so a = 1 4 ( s − ρ ) ≥ 0 a=\frac{1}{4}(s-\rho)\ge0 a = 4 1 ( s − ρ ) ≥ 0 . Moreover ρ + 2 a = 1 2 ( s + ρ ) \rho+2a=\frac{1}{2}(s+\rho) ρ + 2 a = 2 1 ( s + ρ ) , so
ρ a + 2 a 2 = a ( ρ + 2 a ) = 1 8 ( s − ρ ) ( s + ρ ) = 1 8 ( s 2 − ρ 2 ) = b . \rho a+2a^{2}=a(\rho+2a)=\tfrac{1}{8}(s-\rho)(s+\rho)=\tfrac{1}{8}(s^{2}-\rho^{2})=b . ρ a + 2 a 2 = a ( ρ + 2 a ) = 8 1 ( s − ρ ) ( s + ρ ) = 8 1 ( s 2 − ρ 2 ) = b .
Consequently, for every x ∈ X x\in X x ∈ X ,
H ( x , u ( x ) , 2 a D ( x ) ) = ρ a D ( x ) 2 + 2 a 2 D ( x ) 2 − b D ( x ) 2 = 0. (6.1) H\bigl(x,u(x),2aD(x)\bigr)=\rho a\,D(x)^{2}+2a^{2}D(x)^{2}-b\,D(x)^{2}=0 . \tag{6.1} H ( x , u ( x ) , 2 a D ( x ) ) = ρ a D ( x ) 2 + 2 a 2 D ( x ) 2 − b D ( x ) 2 = 0. ( 6.1 )
Step 7 (two estimates for u u u ). For x , y ∈ X x,y\in X x , y ∈ X we have u ( x ) − u ( y ) = a ( D ( x ) − D ( y ) ) ( D ( x ) + D ( y ) ) u(x)-u(y)=a\bigl(D(x)-D(y)\bigr)\bigl(D(x)+D(y)\bigr) u ( x ) − u ( y ) = a ( D ( x ) − D ( y ) ) ( D ( x ) + D ( y ) ) , and by (1.1) D ( x ) − D ( y ) ≤ d ( x , y ) D(x)-D(y)\le d(x,y) D ( x ) − D ( y ) ≤ d ( x , y ) and D ( x ) + D ( y ) ≤ 2 D ( x ) + d ( x , y ) D(x)+D(y)\le2D(x)+d(x,y) D ( x ) + D ( y ) ≤ 2 D ( x ) + d ( x , y ) , both right-hand sides being nonnegative. If D ( y ) ≤ D ( x ) D(y)\le D(x) D ( y ) ≤ D ( x ) , multiplying these bounds for nonnegative factors gives ( D ( x ) − D ( y ) ) ( D ( x ) + D ( y ) ) ≤ d ( x , y ) ( 2 D ( x ) + d ( x , y ) ) \bigl(D(x)-D(y)\bigr)\bigl(D(x)+D(y)\bigr)\le d(x,y)\bigl(2D(x)+d(x,y)\bigr) ( D ( x ) − D ( y ) ) ( D ( x ) + D ( y ) ) ≤ d ( x , y ) ( 2 D ( x ) + d ( x , y ) ) ; if D ( x ) < D ( y ) D(x)<D(y) D ( x ) < D ( y ) the left-hand side is negative. Since 0 ≤ a 0\le a 0 ≤ a ,
u ( x ) − u ( y ) ≤ a d ( x , y ) ( 2 D ( x ) + d ( x , y ) ) and, symmetrically, ∣ u ( x ) − u ( y ) ∣ ≤ a d ( x , y ) ( 2 D ( x ) + d ( x , y ) ) , (7.1) u(x)-u(y)\le a\,d(x,y)\bigl(2D(x)+d(x,y)\bigr)\qquad\text{and, symmetrically,}\qquad|u(x)-u(y)|\le a\,d(x,y)\bigl(2D(x)+d(x,y)\bigr), \tag{7.1} u ( x ) − u ( y ) ≤ a d ( x , y ) ( 2 D ( x ) + d ( x , y ) ) and, symmetrically, ∣ u ( x ) − u ( y ) ∣ ≤ a d ( x , y ) ( 2 D ( x ) + d ( x , y ) ) , ( 7.1 )
the second because ∣ u ( x ) − u ( y ) ∣ = a ∣ D ( x ) − D ( y ) ∣ ( D ( x ) + D ( y ) ) |u(x)-u(y)|=a\,|D(x)-D(y)|\,\bigl(D(x)+D(y)\bigr) ∣ u ( x ) − u ( y ) ∣ = a ∣ D ( x ) − D ( y ) ∣ ( D ( x ) + D ( y ) ) and (1.1). Now let x ∈ X x\in X x ∈ X and τ \tau τ a positive real, and let δ 0 = min { 1 , τ / ( a ( 2 D ( x ) + 1 ) + 1 ) } \delta_{0}=\min\{1,\tau/(a(2D(x)+1)+1)\} δ 0 = min { 1 , τ / ( a ( 2 D ( x ) + 1 ) + 1 )} . For y ∈ X y\in X y ∈ X with d ( x , y ) < δ 0 d(x,y)<\delta_{0} d ( x , y ) < δ 0 , (7.1) gives ∣ u ( x ) − u ( y ) ∣ ≤ a ( 2 D ( x ) + 1 ) d ( x , y ) ≤ a ( 2 D ( x ) + 1 ) δ 0 < τ |u(x)-u(y)|\le a(2D(x)+1)\,d(x,y)\le a(2D(x)+1)\,\delta_{0}<\tau ∣ u ( x ) − u ( y ) ∣ ≤ a ( 2 D ( x ) + 1 ) d ( x , y ) ≤ a ( 2 D ( x ) + 1 ) δ 0 < τ . Hence u ( y ) < u ( x ) + τ u(y)<u(x)+\tau u ( y ) < u ( x ) + τ and u ( x ) − τ < u ( y ) u(x)-\tau<u(y) u ( x ) − τ < u ( y ) , so u u u is upper and lower semicontinuous on X X X .
Step 8 (Part 2, subsolution). Let ψ 1 ∈ C ‾ ( X ) \psi_{1}\in\underline{\mathcal{C}}(X) ψ 1 ∈ C ( X ) , let ψ 2 : X → R \psi_{2}:X\to\mathbb{R} ψ 2 : X → R be locally Lipschitz on X X X , and let x ∈ X x\in X x ∈ X be a point at which u − ψ 1 − ψ 2 u-\psi_{1}-\psi_{2} u − ψ 1 − ψ 2 has a local maximum relative to X X X , with a radius r 1 > 0 r_{1}>0 r 1 > 0 . For y ∈ X y\in X y ∈ X with d ( x , y ) < r 1 d(x,y)<r_{1} d ( x , y ) < r 1 , u ( y ) − ψ 1 ( y ) − ψ 2 ( y ) ≤ u ( x ) − ψ 1 ( x ) − ψ 2 ( x ) u(y)-\psi_{1}(y)-\psi_{2}(y)\le u(x)-\psi_{1}(x)-\psi_{2}(x) u ( y ) − ψ 1 ( y ) − ψ 2 ( y ) ≤ u ( x ) − ψ 1 ( x ) − ψ 2 ( x ) , so by (7.1)
ψ 1 ( x ) − ψ 1 ( y ) ≤ u ( x ) − u ( y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ ≤ ( 2 a D ( x ) + a d ( x , y ) ) d ( x , y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ , \psi_{1}(x)-\psi_{1}(y)\le u(x)-u(y)+|\psi_{2}(y)-\psi_{2}(x)|\le\bigl(2aD(x)+a\,d(x,y)\bigr)\,d(x,y)+|\psi_{2}(y)-\psi_{2}(x)| , ψ 1 ( x ) − ψ 1 ( y ) ≤ u ( x ) − u ( y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ ≤ ( 2 a D ( x ) + a d ( x , y ) ) d ( x , y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ ,
and since the right-hand side is nonnegative, claim 3 of Elementary Properties of the Maximum of Two Elements bounds [ ψ 1 ( y ) − ψ 1 ( x ) ] − [\psi_{1}(y)-\psi_{1}(x)]_{-} [ ψ 1 ( y ) − ψ 1 ( x ) ] − by the same quantity. Given a real ε > 0 \varepsilon>0 ε > 0 , let r = min { r 1 , ε / ( a + 1 ) } r=\min\{r_{1},\varepsilon/(a+1)\} r = min { r 1 , ε / ( a + 1 )} ; for y ∈ X y\in X y ∈ X with 0 < d ( x , y ) < r 0<d(x,y)<r 0 < d ( x , y ) < r we have a d ( x , y ) ≤ ε a\,d(x,y)\le\varepsilon a d ( x , y ) ≤ ε , hence [ ψ 1 ( y ) − ψ 1 ( x ) ] − ≤ ( 2 a D ( x ) + ε ) d ( x , y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ [\psi_{1}(y)-\psi_{1}(x)]_{-}\le\bigl(2aD(x)+\varepsilon\bigr)\,d(x,y)+|\psi_{2}(y)-\psi_{2}(x)| [ ψ 1 ( y ) − ψ 1 ( x ) ] − ≤ ( 2 a D ( x ) + ε ) d ( x , y ) + ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ . So Elementary Properties of Local Slopes: Order, Negation, Bounds from Difference Quotients, Lipschitz Functions and Squared Distances §upper-bound , with ψ = ψ 1 \psi=\psi_{1} ψ = ψ 1 , φ = ψ 2 \varphi=\psi_{2} φ = ψ 2 and the nonnegative real c = 2 a D ( x ) c=2aD(x) c = 2 a D ( x ) , gives ∣ ∇ − ψ 1 ∣ ( x ) ≤ 2 a D ( x ) + ∣ ∇ ψ 2 ∣ ( x ) |\nabla^{-}\psi_{1}|(x)\le2aD(x)+|\nabla\psi_{2}|(x) ∣ ∇ − ψ 1 ∣ ( x ) ≤ 2 a D ( x ) + ∣∇ ψ 2 ∣ ( x ) . By Test Classes for Slope-Based Viscosity Solutions on a Metric Space §sub-class , ∣ ∇ ψ 1 ∣ ( x ) = ∣ ∇ − ψ 1 ∣ ( x ) |\nabla\psi_{1}|(x)=|\nabla^{-}\psi_{1}|(x) ∣∇ ψ 1 ∣ ( x ) = ∣ ∇ − ψ 1 ∣ ( x ) , and by Elementary Properties of Local Slopes: Order, Negation, Bounds from Difference Quotients, Lipschitz Functions and Squared Distances §order , ∣ ∇ ψ 2 ∣ ( x ) ≤ ∣ ∇ ψ 2 ∣ ∗ ( x ) |\nabla\psi_{2}|(x)\le|\nabla\psi_{2}|^{*}(x) ∣∇ ψ 2 ∣ ( x ) ≤ ∣∇ ψ 2 ∣ ∗ ( x ) ; so ∣ ∇ ψ 1 ∣ ( x ) − ∣ ∇ ψ 2 ∣ ∗ ( x ) ≤ 2 a D ( x ) |\nabla\psi_{1}|(x)-|\nabla\psi_{2}|^{*}(x)\le2aD(x) ∣∇ ψ 1 ∣ ( x ) − ∣∇ ψ 2 ∣ ∗ ( x ) ≤ 2 a D ( x ) , and as 0 ≤ 2 a D ( x ) 0\le2aD(x) 0 ≤ 2 a D ( x ) , claim 3 of Elementary Properties of the Maximum of Two Elements gives q = max { ∣ ∇ ψ 1 ∣ ( x ) − ∣ ∇ ψ 2 ∣ ∗ ( x ) , 0 } ≤ 2 a D ( x ) q=\max\{|\nabla\psi_{1}|(x)-|\nabla\psi_{2}|^{*}(x),0\}\le2aD(x) q = max { ∣∇ ψ 1 ∣ ( x ) − ∣∇ ψ 2 ∣ ∗ ( x ) , 0 } ≤ 2 a D ( x ) . Both lie in T T T , so monotonicity of H H H and (6.1) give H ( x , u ( x ) , q ) ≤ H ( x , u ( x ) , 2 a D ( x ) ) = 0 H(x,u(x),q)\le H\bigl(x,u(x),2aD(x)\bigr)=0 H ( x , u ( x ) , q ) ≤ H ( x , u ( x ) , 2 a D ( x ) ) = 0 . With Step 7, u u u is an s-subsolution of H = 0 H=0 H = 0 in X X X by Slope-Based Viscosity Subsolutions, Supersolutions and Solutions on a Metric Space §subsolution .
Step 9 (Part 2, supersolution). Let ψ 1 ∈ C ‾ ( X ) \psi_{1}\in\overline{\mathcal{C}}(X) ψ 1 ∈ C ( X ) , let ψ 2 : X → R \psi_{2}:X\to\mathbb{R} ψ 2 : X → R be locally Lipschitz on X X X , let x ∈ X x\in X x ∈ X be a point at which u − ψ 1 − ψ 2 u-\psi_{1}-\psi_{2} u − ψ 1 − ψ 2 has a local minimum relative to X X X , with a radius r 1 > 0 r_{1}>0 r 1 > 0 , and let q = ∣ ∇ ψ 1 ∣ ( x ) + ∣ ∇ ψ 2 ∣ ∗ ( x ) ∈ T q=|\nabla\psi_{1}|(x)+|\nabla\psi_{2}|^{*}(x)\in T q = ∣∇ ψ 1 ∣ ( x ) + ∣∇ ψ 2 ∣ ∗ ( x ) ∈ T . If D ( x ) = 0 D(x)=0 D ( x ) = 0 , then u ( x ) = 0 u(x)=0 u ( x ) = 0 and H ( x , u ( x ) , q ) = 1 2 q 2 ≥ 0 H(x,u(x),q)=\frac{1}{2}q^{2}\ge0 H ( x , u ( x ) , q ) = 2 1 q 2 ≥ 0 .
Suppose 0 < D ( x ) 0<D(x) 0 < D ( x ) . We verify the hypothesis of Elementary Properties of Local Slopes: Order, Negation, Bounds from Difference Quotients, Lipschitz Functions and Squared Distances §lower-bound for ψ = ψ 1 \psi=\psi_{1} ψ = ψ 1 , φ = ψ 2 \varphi=\psi_{2} φ = ψ 2 and c = 2 a D ( x ) c=2aD(x) c = 2 a D ( x ) . Let ε > 0 \varepsilon>0 ε > 0 and r > 0 r>0 r > 0 be reals; choose, in this order, ε ′ = min { 1 , ε / ( 2 ( 2 a D ( x ) + 1 ) ) } \varepsilon'=\min\{1,\varepsilon/(2(2aD(x)+1))\} ε ′ = min { 1 , ε / ( 2 ( 2 a D ( x ) + 1 ))} and R = min { r , r 1 , D ( x ) , ε / ( 2 ( a + 1 ) ) } R=\min\{r,r_{1},D(x),\varepsilon/(2(a+1))\} R = min { r , r 1 , D ( x ) , ε / ( 2 ( a + 1 ))} , and let y y y be given by Step 2 for these ε ′ \varepsilon' ε ′ and R R R , with h = d ( x , y ) h=d(x,y) h = d ( x , y ) . Then 0 < h < R 0<h<R 0 < h < R and 0 ≤ ( 1 − ε ′ ) h ≤ D ( x ) − D ( y ) 0\le(1-\varepsilon')h\le D(x)-D(y) 0 ≤ ( 1 − ε ′ ) h ≤ D ( x ) − D ( y ) , while (1.1) gives 2 D ( x ) − h ≤ D ( x ) + D ( y ) 2D(x)-h\le D(x)+D(y) 2 D ( x ) − h ≤ D ( x ) + D ( y ) with 0 < D ( x ) < 2 D ( x ) − h 0<D(x)<2D(x)-h 0 < D ( x ) < 2 D ( x ) − h . Multiplying these lower bounds of nonnegative factors and using 0 ≤ a 0\le a 0 ≤ a ,
u ( x ) − u ( y ) = a ( D ( x ) − D ( y ) ) ( D ( x ) + D ( y ) ) ≥ a ( 1 − ε ′ ) h ( 2 D ( x ) − h ) ≥ h ( 2 a D ( x ) − 2 a D ( x ) ε ′ − a h ) . u(x)-u(y)=a\bigl(D(x)-D(y)\bigr)\bigl(D(x)+D(y)\bigr)\ge a(1-\varepsilon')h\,(2D(x)-h)\ge h\bigl(2aD(x)-2aD(x)\,\varepsilon'-a\,h\bigr). u ( x ) − u ( y ) = a ( D ( x ) − D ( y ) ) ( D ( x ) + D ( y ) ) ≥ a ( 1 − ε ′ ) h ( 2 D ( x ) − h ) ≥ h ( 2 a D ( x ) − 2 a D ( x ) ε ′ − a h ) .
Here 2 a D ( x ) ε ′ ≤ ε 2 ⋅ 2 a D ( x ) 2 a D ( x ) + 1 ≤ ε 2 2aD(x)\,\varepsilon'\le\frac{\varepsilon}{2}\cdot\frac{2aD(x)}{2aD(x)+1}\le\frac{\varepsilon}{2} 2 a D ( x ) ε ′ ≤ 2 ε ⋅ 2 a D ( x ) + 1 2 a D ( x ) ≤ 2 ε and a h ≤ a a + 1 ⋅ ε 2 ≤ ε 2 a\,h\le\frac{a}{a+1}\cdot\frac{\varepsilon}{2}\le\frac{\varepsilon}{2} a h ≤ a + 1 a ⋅ 2 ε ≤ 2 ε , so u ( x ) − u ( y ) ≥ ( 2 a D ( x ) − ε ) h u(x)-u(y)\ge\bigl(2aD(x)-\varepsilon\bigr)h u ( x ) − u ( y ) ≥ ( 2 a D ( x ) − ε ) h . Since d ( x , y ) < r 1 d(x,y)<r_{1} d ( x , y ) < r 1 , u ( x ) − ψ 1 ( x ) − ψ 2 ( x ) ≤ u ( y ) − ψ 1 ( y ) − ψ 2 ( y ) u(x)-\psi_{1}(x)-\psi_{2}(x)\le u(y)-\psi_{1}(y)-\psi_{2}(y) u ( x ) − ψ 1 ( x ) − ψ 2 ( x ) ≤ u ( y ) − ψ 1 ( y ) − ψ 2 ( y ) , and therefore
[ ψ 1 ( y ) − ψ 1 ( x ) ] − ≥ ψ 1 ( x ) − ψ 1 ( y ) ≥ u ( x ) − u ( y ) − ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ ≥ ( 2 a D ( x ) − ε ) d ( x , y ) − ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ , [\psi_{1}(y)-\psi_{1}(x)]_{-}\ge\psi_{1}(x)-\psi_{1}(y)\ge u(x)-u(y)-|\psi_{2}(y)-\psi_{2}(x)|\ge\bigl(2aD(x)-\varepsilon\bigr)\,d(x,y)-|\psi_{2}(y)-\psi_{2}(x)| , [ ψ 1 ( y ) − ψ 1 ( x ) ] − ≥ ψ 1 ( x ) − ψ 1 ( y ) ≥ u ( x ) − u ( y ) − ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ ≥ ( 2 a D ( x ) − ε ) d ( x , y ) − ∣ ψ 2 ( y ) − ψ 2 ( x ) ∣ ,
with 0 < d ( x , y ) < r 0<d(x,y)<r 0 < d ( x , y ) < r . So Elementary Properties of Local Slopes: Order, Negation, Bounds from Difference Quotients, Lipschitz Functions and Squared Distances §lower-bound gives 2 a D ( x ) − ∣ ∇ ψ 2 ∣ ( x ) ≤ ∣ ∇ − ψ 1 ∣ ( x ) 2aD(x)-|\nabla\psi_{2}|(x)\le|\nabla^{-}\psi_{1}|(x) 2 a D ( x ) − ∣∇ ψ 2 ∣ ( x ) ≤ ∣ ∇ − ψ 1 ∣ ( x ) , and Elementary Properties of Local Slopes: Order, Negation, Bounds from Difference Quotients, Lipschitz Functions and Squared Distances §order gives ∣ ∇ − ψ 1 ∣ ( x ) ≤ ∣ ∇ ψ 1 ∣ ( x ) |\nabla^{-}\psi_{1}|(x)\le|\nabla\psi_{1}|(x) ∣ ∇ − ψ 1 ∣ ( x ) ≤ ∣∇ ψ 1 ∣ ( x ) and ∣ ∇ ψ 2 ∣ ( x ) ≤ ∣ ∇ ψ 2 ∣ ∗ ( x ) |\nabla\psi_{2}|(x)\le|\nabla\psi_{2}|^{*}(x) ∣∇ ψ 2 ∣ ( x ) ≤ ∣∇ ψ 2 ∣ ∗ ( x ) ; hence 2 a D ( x ) ≤ q 2aD(x)\le q 2 a D ( x ) ≤ q . Both lie in T T T , so monotonicity of H H H and (6.1) give H ( x , u ( x ) , q ) ≥ H ( x , u ( x ) , 2 a D ( x ) ) = 0 H(x,u(x),q)\ge H\bigl(x,u(x),2aD(x)\bigr)=0 H ( x , u ( x ) , q ) ≥ H ( x , u ( x ) , 2 a D ( x ) ) = 0 . (Only local Lipschitz continuity of ψ 1 \psi_{1} ψ 1 was used.)
With Step 7, u u u is an s-supersolution of H = 0 H=0 H = 0 in X X X by Slope-Based Viscosity Subsolutions, Supersolutions and Solutions on a Metric Space §supersolution , and with Step 8 an s-solution by Slope-Based Viscosity Subsolutions, Supersolutions and Solutions on a Metric Space §solution . This proves claim 2.