Towards the end of their paper, Belady and Lehman propose a simple cost model.
They present 3 activities:
Activity P is programming new features.
Activity A is antiregressive, it is improving the system & programmers.
Activity C is complexity, it is keeping the system running & fixing bugs
When taken as time rates, all three as measured in # of programmers to assign from budget B such that:
B = P + A + C, i.e.: you must divide the budget wholly between the 3 activities.
Their observations lead them to this qualitative model between the variables:
A should be linearly related to P.
Under-spending in A leads to growth in C.
C grows linearly with P and time (so quadratically related to total manpower).
Equilibrium between the three variables is possible but the short-term benefits of spending less in A and more in P lead C to grow monotonically in practice.