Delayed negative feedback in red-cell production ($u^{\prime }=\frac{1}{1+u(t-d{)}^n}-\delta u$)

The decreasing production function and increasing linear loss have one positive constant equilibrium $u_{\ast }$. It scales as ${\delta }^{-1/(n+1)}$ for small loss and ${\delta }^{-1}$ for large loss. With $a=n{u}_{\ast }^{n-1}/(1+u_{\ast }^n{)}^2>0$, the linear characteristic equation is $\sigma +\delta +a{e}^{-\sigma d}=0$. It has no real nonnegative root, and zero is never a root. Thus loss of stability from the zero-delay stable state must occur through a nonzero imaginary conjugate pair.

If $a>\delta$, the first crossing has $\omega =\sqrt{a^2-{\delta }^2}$ and $d_c=\arccos (-\delta /a)/\omega$. Differentiation of the characteristic equation gives $\mathrm{Re}{\sigma }^{\prime }(d_c)={\omega }^2/[(1+\delta d_c{)}^2+{\omega }^2{d}_c^2]>0$, proving transversality. If $a\le \delta$, the magnitude inequality $|\sigma +\delta |=a|e^{-\sigma d}|$ rules out a root in the closed right half-plane. The model needs an initial history on the whole delay interval. The delay represents a production/maturation lag rather than instantaneous production controlled by the current cell count.