Loss of differentiability from square-root hyperbolic source data ($R_x+R{R}_y=0,S_x+S{S}_y=2$)

For $U_x+A(U)U_y=(1,\sqrt{v}{)}^T$ with $A=\left(\begin{array}{cc}u& 2\\ 2v& u\end{array}\right)$ and $v>0$, the variables $R=u-2\sqrt{v}$ and $S=u+2\sqrt{v}$ decouple into an unforced and a uniformly forced Inviscid Burgers equation. Smooth data for $u,v$ need not be smooth data for these variables when $v$ vanishes. For example $u(0,y)=2y$, $v(0,y)=y^2$ gives $R_0=4\min (y,0)$ and $S_0=4\max (y,0)$. Their derivative corners travel along different characteristic families. In the positive quadrant, $R=0$ and $S=2x$ below $y=x^2$, whereas $S=2(x+2x^2+2y)/(1+4x)$ above. Reconstructing $u=(R+S)/2$, $v=(S-R{)}^2/16$ gives a continuous piecewise classical continuation whose $y$ derivative jumps at the parabola. Treating $\sqrt{y^2}$ as $y$ for all real $y$ would miss this obstruction. The degeneracy at zero density and the regularity class are essential parts of the Cauchy problem.