Implementing Initial Conditions and Baseline States

Initial Conditions Are Part of the Model

An ODE model does not define a trajectory from its rate equations alone. It also needs a state at a specified starting time. In a QSP model, that state may represent a biological baseline before an intervention, or a state at the start of a simulation after some earlier history. Choosing it carelessly can make a mathematically valid simulation describe the wrong biological situation.

The key distinction is between the state at the start and the rate of change at the start. A state can be a plausible measured baseline without being a steady state of the model. Conversely, if it is intended to be a steady baseline under specified conditions, the model equations must imply zero derivative there. Evaluating the right-hand side at time zero makes this distinction explicit.

Here \(x\) denotes the model state vector, \(u\) the input, and \(\theta\) the parameters. An initial condition is not a parameter or an input: it is the starting value of a state. As established in the earlier material on mapping biological states to indexed vectors and separating parameters, inputs, and model state, the order of the state components supplied to the model must match the order expected by its right-hand-side function.

When Is a Baseline a Steady State?

For a baseline to be steady, the model’s net rate of change must be zero for every state component under the baseline conditions. For a time-independent model with constant baseline input \(u^*\) and parameters \(\theta\), this means solving the vector equation

Each component equation represents a balance of its incoming and outgoing processes. A component can have substantial production and removal at steady state; what must vanish is their net contribution to the derivative. The earlier material on balance equations and initial conditions that satisfy conservation constraints is also relevant: a candidate state must obey any applicable conservation constraints as well as the rate equations.

A zero derivative at one time is a local consistency check, not automatically proof that a state stays steady. For a genuinely constant solution, the right-hand side must remain zero as the state, input, and parameters remain at their baseline values. If an input changes after the start, the same state will generally no longer be an equilibrium under the new conditions.

Use the Time-Zero Derivative to Check Consistency

The most direct check is to substitute the proposed state and the relevant initial inputs into the model’s right-hand side. For a vector state, calculate each component, retaining its units. A nonzero component means that the corresponding state begins changing at that rate; it does not, by itself, mean the initial condition is invalid. It means only that the initial condition is not a steady state under the conditions used in the calculation.

A short justification follows from the definition of a derivative. For a small positive time increment \(\Delta t\), the initial rate gives the first-order change

This approximation explains why the derivative is a useful diagnostic: a large derivative relative to the state’s scale predicts a noticeable early change, while a zero derivative removes the first-order change at that instant. The approximation is local; it does not claim that the derivative remains constant as the state evolves.

What Changes When the System Is Perturbed?

A perturbation may change an input or parameter while leaving the state continuous at the start of the perturbed simulation. In that case, initialize the perturbed simulation with the baseline state, but evaluate its derivative using the perturbed conditions. The state records what is present at the start; the changed right-hand side determines how it begins to move. Do not replace the baseline state with a new equilibrium unless the simulation is explicitly intended to start after the system has already equilibrated under the new conditions.

For a sudden change in an input at \(t_0\), it is useful to distinguish the conditions immediately before and after the change. If the model specifies no instantaneous jump in a state, then \(x(t_0)\) is the same on both sides. The initial derivative for the post-change trajectory uses the post-change input. This is a statement about setting the initial state and evaluating the equations, not a general treatment of dose events or other discontinuous-input implementation.

Practical Checks and Common Failure Modes

A reliable initial-condition check makes the intended baseline explicit, evaluates the model function with the matching state ordering and conditions, and inspects the resulting derivative component by component. When a baseline is claimed to be steady, investigate any nonzero component rather than dismissing it as a solver issue. The cause may be a state value drawn from a different condition, a parameter or input assigned incorrectly, or an error in the equations or state indexing.

A computed derivative need not be exactly zero in finite-precision arithmetic. Compare its magnitude with the scale and units of that derivative, and consider whether the values used were rounded. A small number is not automatically negligible: its practical importance depends on the state scale and the time horizon. Conversely, do not compare a concentration derivative directly with a concentration; the former has units of concentration per time.

• Wrong reasoning: “The baseline concentration was measured, so it must be a steady state.” Correction: A measured state may be changing. Test steadiness by evaluating the model derivative under the stated conditions.
• Wrong reasoning: “The derivative is nonzero, so the initial condition is invalid.” Correction: A nonzero derivative is consistent with a valid starting state for a transient simulation; it contradicts only a claim that the state is steady under those conditions.
• Wrong reasoning: “The perturbed system should start at its new equilibrium.” Correction: If the perturbation begins from the old baseline, retain that state unless the model setup specifically assumes prior equilibration under the new conditions.
• Wrong reasoning: “One component has zero derivative, so the whole model is at steady state.” Correction: A vector is steady only when every state derivative is zero.

There are several ways to obtain an initial state. Solving the equilibrium equations directly is appropriate when a steady baseline is intended and the equations can be solved reliably. Integrating a model under baseline conditions until it settles can also suggest a baseline, but the resulting state should still be checked against the equations and any conservation constraints. Using measured or externally specified baseline values can better represent a particular system, but those values need not define an equilibrium. The right choice depends on what the simulation is meant to represent; none of these approaches makes a transient starting state interchangeable with a steady state.

The derivative check has limits. It tests the equations at the supplied state and conditions; it cannot establish that the biological model is correct, that the baseline measurements are accurate, or that a computed equilibrium is unique or biologically meaningful. A zero derivative can also occur at an equilibrium that is unstable. The check is therefore a consistency test, not a complete validation of a model or baseline.