The Verification Venue · one skid, two names
Push Wide, or Step Out
Understeer and oversteer sound like opposites. They are the same event seen at two ends of the car: an axle running out of grip. Front axle first, the nose pushes wide. Rear axle first, the tail steps out. One number tells you which way a car leans, and one surprising move can flip it mid-corner.
The diagnostic is the understeer gradient, K = Wf/Cαf − Wr/Cαr: front axle load divided by front cornering stiffness, minus the same at the rear, in degrees of extra steering per g of cornering. K > 0 and the front gives up first (understeer, push wide). K < 0 and the rear gives up first (oversteer, rotate). K = 0 is neutral. Drive the corner below and watch it decide.
Understeer gradient K
– deg/g
sub-limit tendency
Corner speed vs cap
–
flat-turn cap v = √(μgr)
Lateral load ay
– g
of μ = – available
Front axle circle
–% used
grip spent turning + driving/braking
Rear axle circle
–% used
grip spent turning + driving/braking
Set a corner near its limit, then hit Lift off: the throttle drops to a trailing brake, weight pitches onto the nose, the rear unloads, and a car that was pushing wide snaps into oversteer. That is the trap, live.
Only matters under power: throttle spends the driven axle's circle.
Front-heavy raises K (understeer bias); rear-heavy lowers it.
Dry road ~0.9, wet ~0.6, ice ~0.2, race slicks > 1.2.
A tight hairpin is ~15 m; a sweeping bend ~80 m.
Push it past the cap readout and an axle must let go.
In longitudinal g. Left of centre is braking (weight forward); right is throttle (weight back). This is the axis the folk fix gets wrong.
The weight that does not move
Grip at a tire is set by the load pressing it down, and that load moves around as you turn, brake and accelerate. But nothing is sloshing sideways in the car: the total weight is fixed at 14715 N. What changes is how that fixed weight is shared across the four contact patches, and grip follows load. Two transfers run at once, on two different axes:
Load on each contact patch (live)
Longitudinal transfer moves weight front↔rear (ΔF = m·ax·h/L); lateral transfer moves it left↔right (ΔF = m·ay·h/t). Watch the four boxes fill and drain, always summing back to the same total.
Front left
– N
–
Front right
– N
–
Rear left
– N
–
Rear right
– N
–
The instinctive fix is a trap
Lifting off does not tuck the nose in. It can throw the tail out.
You are in a corner, the front is pushing wide, and every nerve says lift off the throttle. On a front-limited car that seems right: slow down, get grip back. But lifting is not just slowing, it is a longitudinal load-transfer event. Trailing throttle or brushing the brake pitches weight forward, ΔF = m·ax·h/L lands on the front and lifts off the rear. The rear axle, now light, still has to make the same cornering force it did a second ago, so it saturates its shrunken friction circle and lets go. The push becomes a spin. This is lift-off oversteer, and it is famous for catching drivers out in fast bends. It is a different axis from the lateral transfer that governs steady cornering, which is exactly why the intuition misfires. Hit the Lift off button above to see it happen to a car that was understeering a moment before.
So the same car, same corner, same instant can read understeer or oversteer depending on what your right foot just did. That is the tell that these are not two diseases but one, watched from two ends. Here is the honest scope, in four twists the thin explainers skip.
Four honest twists
1. K is the sub-limit ruler, not a guarantee at the edge. The understeer gradient is a linear, quasi-static number: how much extra lock the corner needs per added g, while the tires are still behaving. At the actual limit the verdict is set by which axle saturates its friction circle first, and load transfer can move that under your foot. K tells you the lean; the circle decides the moment.
2. Load transfer is redistribution, not weight sliding across the car. Total weight is constant. Braking does not pour mass onto the front bumper; it changes the share each patch carries, and grip follows load non-linearly. That non-linearity is the whole game.
3. FWD tends to push, RWD tends to power-oversteer, but neither is a law. It is a tendency set by weight distribution, where the drive torque lands and how the tires are loaded. Set the drivetrain and weight above and you can make any layout do either. A front-heavy FWD car will still snap into lift-off oversteer if you lift hard enough.
4. The friction "circle" is really an ellipse. Drawing it as a circle assumes a tire's grip limit is equal sideways and lengthwise. Real tires are usually a bit stronger one way, so the true envelope is a friction ellipse; the circle is its equal-μ special case. We draw the circle here because the story (spend grip one way, lose it the other) is the same, just rounder.
So how do you fix each one
Understeer — front pushed wide
Ease the steering a touch (a saturated front tire makes less grip, not more, so more lock does nothing) and gently reduce speed by lifting or trailing so the front regains its circle. Gently: a hard lift is how you turn a push into a spin. Then feed the corner back in.
Oversteer — rear stepped out
Look and steer where you want to go: countersteer into the slide, correction roughly proportional to how far the rear has stepped out. Modulate the throttle to reload the rear. Do not stab the brake, which unloads the rear further and makes it worse.
The two, side by side
| Understeer | Oversteer | |
|---|---|---|
| Which axle runs out of grip first | Front | Rear |
| What the car does | Pushes wide, nose plows straight | Rotates, tail swings out |
| Understeer gradient K | K > 0 | K < 0 |
| Common tendency | FWD, front-heavy, power-on (FWD) | RWD power-on, lift-off, rear-heavy |
| The instinct | Add more lock (does not work) | Freeze / brake (makes it worse) |
| The fix | Ease lock, gently slow, regain front grip | Countersteer, modulate throttle, look ahead |
The check: every number recomputed in front of you
Nothing here is stored. For your current settings the page recomputes the whole corner from the same equations the offline verifier uses, live:
The offline gate recomputes all of it, two independent ways where it can: node research/understeer-vs-oversteer/verify-understeer-vs-oversteer.mjs. Free choices & scope. Fixed vehicle: m = 1500 kg, wheelbase L = 2.6 m, track t = 1.55 m, CG height h = 0.55 m. The tire is a representative saturating model (cornering stiffness Cα(Fz) = 0.34·Fz − 1.8e−5·Fz2, and a mild peak-μ load sensitivity); the signs and orderings are the robust, sourced physics, the exact degrees and newtons are illustrative. It is a linear, quasi-static bicycle model: linear tires below the limit, no aero/downforce, no banking, no limited-slip diff or roll-stiffness tuning. v = √(μgr) is the flat, unbanked, no-aero cap; downforce and banking both raise it.
What is exactly true here, and what is a model
Exactly true (the sourced physics). Understeer and oversteer are one phenomenon: an axle exceeding its grip. The steady-state understeer gradient is K = Wf/Cαf − Wr/Cαr with the sign convention K > 0 understeer, K = 0 neutral, K < 0 oversteer (Gillespie, Ch. 6; Milliken & Milliken). Every tire lives inside a friction circle Fx2 + Fy2 ≤ (μFz)2, so grip spent braking or driving is grip not available to corner. On a flat, unbanked turn the speed cap is v = √(μgr). Load transfer is ΔF = m·ax·h/L front↔rear and ΔF = m·ay·h/t left↔right, and it redistributes a constant total weight. Lifting off mid-corner is a longitudinal transfer that can cause lift-off oversteer.
A model, not a measurement (the numbers). Cornering stiffness and peak grip are given a representative load-sensitivity so the sandbox can show why a heavier axle tends to give up first; real tires vary widely. The lateral demand is split front/rear by the geometry of a bicycle model (zero yaw moment), braking uses a fixed front bias, and throttle loads only the driven axle. These are honest simplifications: they get the directions right (which axle, which way it flips) without claiming to be any specific car.
Named simplifications. Linear tires below the limit; quasi-static (no transient yaw dynamics); no aerodynamic downforce; no road banking; no limited-slip differential or anti-roll tuning; and the friction circle is the equal-μ idealization of the friction ellipse. None of these change the reversal at the heart of the page.