Positivity and Conservation in Cell-Count ODEs

Why Positivity and Conservation Matter

A cell-count model should not predict a negative number of cells from a biologically admissible initial state. Its equations should also account correctly for how cells move between compartments: a transfer within the modeled population changes the counts in individual compartments, but a one-for-one transfer does not create or remove cells from the total. These are structural properties of the model, not conclusions that should depend on a particular numerical simulation.

Here, a compartment is a cell state or population, and \(x_i(t)\) denotes its count at time \(t\). The count may be an expected or scaled population quantity in a deterministic model; the mathematical requirement is that it remain nonnegative. Positivity concerns each component separately. A total-population balance concerns the sum of components and must distinguish internal transitions from division, death, and exchange with populations outside the model.

Boundary Condition for Nonnegative Cell Counts

Write a compartment model as \(\dot{x}=f(t,x)\), where \(x=(x_1,\ldots,x_m)\), \(\dot{x}\) is the vector of time derivatives, and \(f_i(t,x)\) is the net rate of change of compartment \(i\). Assume that \(f\) is continuous in time and locally Lipschitz in the state, so solutions are locally well defined and unique. A standard sufficient condition for nonnegativity is that, whenever \(x\) is nonnegative and \(x_i=0\), the rate into that component satisfies \(f_i(t,x)\geq 0\). This is called a boundary-inward, or quasi-positivity, condition.

The condition is checked one component at a time, on the boundary where that component is zero. It does not require every derivative to be nonnegative everywhere. A compartment with a positive count can decrease because of death or transfer out; the key is that, at zero count, the model must not continue removing cells from it at a negative rate.

For example, suppose compartment \(i\) receives a nonnegative input \(u_i(t,x)\) and loses cells at a rate \(\lambda_i(t,x)x_i\), where \(u_i\geq 0\) and \(\lambda_i\geq 0\). At \(x_i=0\), its derivative is \(u_i\geq 0\). A transfer out of a compartment is similarly safe when its rate vanishes if the source count is zero. These conditions apply even if the rates depend on time or on other nonnegative compartment counts.

Why the Boundary Condition Works

The boundary condition rules out a trajectory crossing from a nonnegative count to a negative one. For an intuitive first-contact argument, suppose a solution starts in the nonnegative orthant and consider the first time a component reaches zero from above. At that boundary point, a strictly negative derivative would make the component decrease through zero. The condition \(f_i\geq 0\) rules out that outward crossing.

For a more precise justification, local Lipschitz continuity bounds how much a component’s derivative can change when its own count is made slightly positive. In a bounded neighborhood of the solution, let \(L\) be a Lipschitz constant in that component. If the other coordinates are held fixed and \(x_i\geq 0\), the boundary condition gives

Thus, while the state is nonnegative, the component cannot decrease faster than proportionally to its own count. Multiplying by \(e^{Lt}\) gives

Consequently, if \(x_i\) starts nonnegative, it cannot become negative while the state remains in that neighborhood. Applying this argument at a possible first boundary contact to each component establishes the stated invariance result. This justification guarantees nonnegativity only while the solution exists; it does not by itself guarantee that counts remain bounded or that a solution exists for all future time.

Summing Compartments: What Cancels?

Define the total modeled cell count as \(N(t)=\sum_{i=1}^m x_i(t)\). Its rate of change is the sum of the compartment rates. An internal transfer that removes one cell from a source and adds one cell to a destination contributes equal and opposite terms, so it cancels from \(\dot{N}\). This cancellation is an accounting identity; it does not imply that the individual compartments are constant.

A compact way to write the accounting is to represent each internal transfer \(r\) by a vector \(s_r\). Its entries describe the changes to the compartments per unit of transfer flux \(q_r\), so its contribution to the ODE is \(s_rq_r\). For a one-for-one transfer, one compartment loses one cell and another gains one cell, and therefore the entries of \(s_r\) sum to zero. Summing the ODE over compartments removes that transfer term.

Here \(u_i\) collects contributions that are not represented as internal transfers, such as external input, cell division, or death. The formula makes the condition for cancellation explicit: an internal process disappears from the total balance only if its net change in the sum of compartment counts is zero. In particular, a process described as differentiation does not automatically conserve cell number. Its stoichiometry—the number of cells removed and added per event—determines its contribution.

Branching, Division, and the Meaning of “Internal”

A process can occur entirely within the modeled system and still change the total cell count. For example, if one progenitor is removed and two progenitor daughters are added per division event, the net contribution is one additional cell per event. Calling the event “internal” does not make it cancel. The same accounting applies to differentiation that produces multiple descendants: count the cells removed and added, rather than assuming one-for-one replacement.

Model Checks, Numerical Diagnostics, and Limits

A useful model check is to write each compartment equation as a list of inflows, outflows, and production or loss terms, then evaluate its derivative on the boundary where that compartment is zero. An outflow proportional to its source count vanishes there; a constant outflow does not. Next, sum the equations and confirm that each one-for-one internal flux appears once with a minus sign and once with a plus sign. Any uncancelled term should have a biological interpretation, such as branching, division, death, or exchange with an unmodeled population.

Numerical solutions can show small negative counts even when the continuous ODE preserves nonnegativity. A time-stepping method may take a step across a boundary, especially when steps are too large or the system has fast and slow processes. Check whether the negative values are tiny relative to the relevant count scale and whether they shrink when tolerances are tightened or the step size is reduced. A substantial negative value, or one that persists under tighter settings, calls for checking the equations, units, initial conditions, and solver configuration. Silently replacing all negative values by zero can conceal an error and alter mass balance.

For computation, one alternative is to use a numerical method or step-size control that better respects the nonnegative region. Another is to transform variables, for example by representing a positive state using its logarithm. These approaches have trade-offs: a log transformation cannot represent an exact zero directly, and numerical safeguards cannot repair an ODE whose boundary condition is wrong. The most direct safeguard is to formulate the model so that its continuous-time dynamics have the required boundary behavior, then verify that the numerical solution is sufficiently accurate.

Positivity does not mean that counts are bounded, approach an equilibrium, or match observed data. Nor does cancellation of internal fluxes imply conservation when division, death, or branching changes the number of cells. The total-balance calculation is for the compartments included in \(N\); adding or removing a modeled compartment changes which processes count as internal. These checks establish mathematical consistency of the stated model, not the biological validity of its assumptions.