Simulated Lap Times: How Close Do They Get

Lap simulation is excellent at the question engineers actually ask and poor at the one enthusiasts ask. Predicting whether change A is faster than change B is reliable, because the errors are common to both cases and cancel. Predicting an absolute lap time is not, because it depends on a tyre model that cannot be derived from physics and has to be measured, and on a driver who is assumed rather than simulated.

Why the difference case is the easy one

A simulation carries systematic errors: an imperfect tyre model, an idealised driver, a smooth road where the real one is bumpy. When the same model runs twice with one parameter changed, those errors appear in both runs and largely cancel in the difference. The prediction that a 20 kg weight reduction is worth a certain amount, or that a wing angle change costs a certain amount of straight line speed, is therefore trustworthy even when neither absolute time is.

This is why simulation is used the way it is: to rank options before anything is built, to narrow a setup window before a test, and to decide which of twenty ideas deserve the two days of track time available. Those are all difference questions.

The tyre model is the limit

Everything above the tyre row can be measured or calculated directly. The tyre cannot. Its behaviour is captured by an empirical model fitted to rig measurements, and it has to represent grip as a function of slip, vertical load, camber, temperature, pressure and wear at the same time. Friction also falls as vertical load rises, which means the model has to get load sensitivity right or every prediction involving weight transfer drifts.

Outside the range where the rig data was taken, the model is extrapolating. A tyre measured at moderate loads and asked about a corner with 900 kg of downforce on top of the car's weight is being asked a question it has no data for.

The size of each error, roughly

The individual uncertainties are small and they compound. A useful way to see why absolute predictions drift is to attach a figure to each one and ask what a lap time built on all of them is worth.

The two tyre rows dominate, and together they can account for 12 s on their own, roughly 3 per cent of a 6:45 lap. Every row above them is smaller and every one of them cancels cleanly in a comparison. That is the whole argument for using simulation on differences and distrusting it on absolutes, in one table.

The driver the simulation uses does not exist

Most lap simulations solve for the fastest possible lap given the vehicle model, which produces a driver who brakes exactly at the limit every time, never lifts early and never makes an error. Real record laps are driven by people, and over the Nordschleife's roughly 70 corners a consistent tenth per corner is seven seconds.

That gap is not a defect in the simulation, it is the definition of the two quantities. One is the fastest lap the car could theoretically produce, the other is the fastest lap somebody actually produced. They should not match, and a simulation tuned until it does match has been fitted to a driver rather than to a car.

How the gap is managed in practice

• Correlate, then use differences. Run the simulation against a real session, note the offset, and from then on trust the deltas rather than the absolutes.
• Re-correlate after any large change. A new tyre or a substantially different aerodynamic package moves the model outside its fitted range, and the previous offset no longer applies.
• Feed real data back. Measured speed and acceleration traces from a run are what the tyre model gets refitted against.
• Treat published simulated times with the same caution as any unverified figure. A simulated lap has no notary, no calibrated timing and no documented car, which are exactly the three things a Nürburgring record procedure requires.

Questions readers ask

How accurate are simulated lap times?

Good at predicting differences, weaker at absolutes. Systematic errors cancel when the same model runs twice with one parameter changed, so the value of a modification is reliable. The absolute time depends on a tyre model fitted to measured data and on an idealised driver.

Why is the tyre the hardest part to model?

Because it cannot be derived from first principles. Its behaviour is an empirical fit to rig measurements covering slip, load, camber, temperature, pressure and wear at once, and friction falls as vertical load rises, so load sensitivity has to be captured correctly or every weight transfer prediction drifts.

Why is a simulated lap faster than a real one?

Because the simulated driver brakes exactly at the limit every time and never makes an error. Over roughly 70 corners a real driver losing a consistent tenth per corner is seven seconds behind, and that is a competent lap rather than a poor one.

Can a simulation replace track testing?

For narrowing options, largely yes, which is what it is used for. For an absolute figure, no. A simulated time carries none of the verification a real record does: no notary, no calibrated timing, no documented vehicle condition.

What is correlation in this context?

Running the model against a real session, measuring the offset between prediction and reality, and then trusting the differences rather than the absolute numbers. It has to be repeated whenever a change moves the model outside the range its tyre data covers.

Sources

• Vehicle Physics Pro, tyre model documentation, on empirical tyre models, slip curves and load sensitivity.
• Tire friction overview, on why friction falls as vertical load rises and what that does to a model.
• Nürburgring, record drives, on the verification a real record carries and a simulation does not.
• Racelogic, velocity measurement accuracy, on the measured data against which models are correlated.