RobotGuide
Math foundations/The Mathematics of Contact: Hertz Theory

Foundations T23 / 27 · Contact & Convex Geometry

The Mathematics of Contact: Hertz Theory

Press a finger into a desk and the force is not a linear function of indentation — it goes as the 3/2 power of indentation. Starting from the part elasticity can fully prove, this article pushes the power law out inch by inch.

about 45 min4 theorems20 proof steps4 self-test questions

One-line takeaway

When two elastic quadratic surfaces touch, F = (4/3)E*√R δ3/2. The nonlinearity is not caused by material nonlinearity — it is the geometric consequence of the contact area growing on its own as indentation deepens.

The reading→force calibration curve of a tactile sensor, the contact stiffness at a dexterous fingertip, the stiffness parameters of contact constraints in a simulation engine — all of them rest on this power law. Without knowing its origin, when the calibration fit breaks you do not know what to suspect: material nonlinearity, or a broken assumption.

Why this tool exists

Compress a spring and force is proportional to compression: F = kδ. Press a finger, a silicone pad, or a tire and it is not: soft at first, stiffer the deeper you press. The engineer's first reflex is material nonlinearity — swap in a more expensive constitutive model. For hyperelastic materials like silicone that is part of the story; but for a steel ball on a steel plate or a fingertip on a desk, the material works well within its linear range and the force–deformation relation is still a curve.

The real mechanism is geometry: contact does not happen on a fixed surface. The deeper you press, the larger the contact circle and the more material joins in carrying the load. The stiffness k = dF/dδ is not a constant; it grows together with the contact area. So the problem becomes two questions: how big is the contact area, and how is pressure distributed over it? The two depend on each other — the pressure distribution determines deformation, deformation determines the contact area — and must be solved simultaneously.

In 1882, the 22-year-old Heinrich Hertz solved this problem. His assumptions remain the default starting point of tactile and contact modeling today: small deformation, linear elasticity, frictionless, no adhesion. This article walks his reasoning in full — including one important methodology: in a linear elastic problem, guess a solution + verify + uniqueness is a complete proof. This move is more common than you would think.

Definitions

Definition 1.1 Effective modulus E*

When two elastic bodies touch, their compliances add: 1/E* = (1−ν₁²)/E₁ + (1−ν₂²)/E₂, where E is Young's modulus and ν is Poisson's ratio. For a rigid body pressing an elastic body, E* is just the elastic body's own E/(1−ν²).

Why define it this wayBoth bodies deform, and the total compliance is the sum of the two — the series-springs logic. The (1−ν²) in the denominator comes from this: pressing a point forces the material not only to deform downward but also to squeeze sideways; the lateral effect captured by Poisson's ratio makes the surface softer.

Definition 1.2 Effective radius of curvature R

The shapes of two quadratic surfaces near the contact point can be folded into 1/R = 1/R₁ + 1/R₂ (convex positive, concave negative). For a sphere on a plane, R is just the sphere's radius.

Why define it this wayThe contact problem only cares about how the two surfaces approach each other as a function of position; this second-order quantity can be folded into an equivalent sphere pressing an equivalent plane. A general 3D surface has two principal curvatures and yields an elliptical contact patch; circular contact (sphere-on-plane, cylinders pressed axis-to-axis) needs only one R.

Definition 1.3 Approach δ and contact radius a

δ is the approach distance of the two far-field rigid bodies (not the local deformation of the surface); a is the radius of the contact circle. The output of Hertz theory is exactly the relation among the three quantities F, δ, a.

Why define it this wayδ and a are quantities that must be solved for, not given — precisely why the contact problem is harder than elasticity on a fixed boundary: the boundary (the contact zone) itself changes with load; this is a so-called free-boundary problem.

Theorems and proofs

Lemma 1.1 The Boussinesq solution (point load)

When a concentrated normal force P acts on the surface of an elastic half-space (z≤0, surface z=0), the normal surface displacement at distance ρ from the point of application is u(ρ) = P/(πE*ρ).

Proof

The full proof requires potential theory of elasticity (Galerkin vectors or Love functions), beyond this article's equipment — declared at the boundary. Here we only verify dimensional and superposition consistency: u ∝ P/E*·(1/ρ) is the only combination that satisfies doubling force doubles displacement (linearity), has dimension of length, and respects isotropic symmetry. It is the input to every integral below, but remember: it is cited, not proven.

NoteThe honesty principle of the principles layer: cited lemmas must be marked as cited. What this article actually proves in full is the Hertz law itself — the conclusion after superposing Boussinesq solutions.

Theorem 1.2 Center displacement under a uniform circular load

With uniform pressure p on a disk of radius a, the surface displacement at the center is u(0) = 2pa/E*.

Proof

Slice the disk into thin rings of radius s and width ds. The load on a ring is dP = p·(2πs ds).

u(0) = ∫ dP/(πE* s) = ∫₀^a p·2πs·ds/(πE* s) = (2p/E*)∫₀^a ds = 2pa/E*

The first step uses Lemma 1.1 and superposition (linear elasticity lets the solutions of point loads add directly); every point of the ring is at distance s from the center, so dP/(πE*s) can be written down directly. ∎

NoteThis ring-integral move is the workhorse of the article. Note that the s in numerator and denominator cancels — the singularity of the point-load solution as ρ→0 is exactly canceled by the ring's area element s ds; this is why every integral below converges.

Theorem 1.3 Uniqueness in elastostatics

Given all boundary conditions on a linear elastic body (given forces or given displacements), the solution is unique. That is: if two solutions share identical boundary conditions, they are identical everywhere.

Proof

Let the two solutions be (u₁, σ₁) and (u₂, σ₂); consider the difference w = u₁ − u₂ and the stress difference σ = σ₁ − σ₂. By linearity, (w, σ) also solves the elastic equations, and on the boundary: on traction parts σ·n = 0, on displacement parts w = 0.

Compute the strain energy of the w field: U = (1/2)∫_V σ:ε dV, where ε is the strain tensor of w. Using equilibrium (the divergence of σ is zero) and integration by parts: ∫_V σ:ε dV = ∫_S (σ·n)·w dS.

Every boundary part has either σ·n = 0 (traction boundary) or w = 0 (displacement boundary), so the integral is zero: U = 0.

The strain energy density is a positive definite quadratic form in ε (the definition of material stability), so U = 0 ⟹ ε ≡ 0 ⟹ w is a rigid-body displacement; w = 0 on the displacement boundary pins down the rigid motion, hence w ≡ 0 and the two solutions coincide. ∎

NoteNot a single step is skipped, and it is the key to the methodology: with uniqueness, the contact problem turns from solving equations into guess + verify. Hertz's original reasoning is exactly this path. Positive definiteness is not formalism — it corresponds to material instability (buckling), where uniqueness genuinely fails, and that class of elastostatic problems needs its own theory.

Theorem 1.4 The Hertz law

Under the Hertz assumptions (small deformation, linear elasticity, frictionless, no adhesion), a sphere of radius R pressed on a plane: the pressure distribution inside the contact circle is p(r) = p₀√(1 − r²/a²), the contact radius is a² = Rδ, and the total force is F = (4/3)E*√R δ3/2.

Proof

Strategy (borrowing Theorem 1.3): guess a pressure distribution, verify that the surface displacement it produces inside the contact circle is exactly the parabola δ − r²/(2R) (i.e. conforming to the rigid sphere, no interpenetration, no sticking outside the boundary), then invoke uniqueness to assert this is the solution.

Step one: with a rigid sphere indented by δ into a plane, the gap between sphere and plane is g(r) = δ − r²/(2R) (paraxial expansion of the sphere equation: R − √(R²−r²) ≈ r²/(2R)). Contact requires the surface displacement u(r) = g(r) to hold for r ≤ a, and the pressure to vanish for r > a.

Step two: guess the distribution. At the contact edge r = a the gap slope is nonzero, the pressure must taper continuously to zero, and the solution should be fixed by geometric self-similarity — try p(r) = p₀√(1 − r²/a²) (the simplest elliptical profile that guarantees zero edge pressure and a symmetric distribution; Hertz's original guess).

Step three: verify the shape condition. Substitute the guessed p into the ring integral (a generalization of Theorem 1.2):

u(r) = (1/πE*)∫₀^a p(s)·s·[∫₀2π dθ/√(r²+s²−2rs·cosθ)] ds

The inner angular integral has a closed form: ∫₀2π dθ/√(r²+s²−2rs·cosθ) = 4K(k)/(r+s), where K is the complete elliptic integral of the first kind and k² = 4rs/(r+s)². The integral after substituting p(s) is a classical elliptic-integral identity (Hertz 1882; modern treatment in Johnson, Contact Mechanics, Chapter 3). The result:

u(r) = (πp₀/(4E*·a))·(2a² − r²), for r ≤ a

The verification at the center point can be done independently: u(0) = (1/πE*)∫p(s)·s·(2π/s)ds = (2/E*)∫₀^a p₀√(1−s²/a²)ds = (2p₀/E*)·(πa/4) = πp₀a/(2E*), consistent with the expression above at r=0. The closed form for general r involves elliptic integrals; this article verifies it numerically (see the measured numbers below: maximum deviation 0.04% over 12 radius points).

Step four: fix the constants. Rewrite the verified result: u(r) = (πp₀a/(2E*)) − (πp₀/(4E*a))·r². Require u(r) = δ − r²/(2R) and match term by term — r² coefficient: πp₀/(4E*a) = 1/(2R), i.e. πp₀ = 2E*a/R; constant term: δ = πp₀a/(2E*). Substituting the former: δ = (2E*a/R)·a/(2E*) = a²/R, i.e. a² = Rδ.

Step five: integrate the total force and eliminate:

F = ∫₀^a 2πr·p₀√(1−r²/a²) dr = 2πp₀·(a²/3)·[antiderivative of −(1−r²/a²)3/2·(3/2) evaluated from 0 to a] = (2/3)πa²p₀

From δ = πp₀a/(2E*) we get p₀ = 2E*δ/(πa) = 2E*√δ/(π√R); substituting: F = (2/3)πa²·2E*√δ/(π√R) = (4/3)E*·a²·√δ/√R = (4/3)E*·Rδ·√δ/√R = (4/3)E*√R·δ3/2. ∎

NoteRecap of the proof structure: lemma (point load, cited) + ring integral (full) + uniqueness (full) + verification (full at the center point, closed form cited for general points plus numerical verification). This verify-style proof is fully legitimate in linear elastic problems — because of Theorem 1.3. Note that p(r) hits zero at the edge r=a with a vertical tangent: the pressure distribution is singular at the contact edge, a foreshadowing of the edge-slips-first discussion (T15).

Work it by hand

A silicone fingertip simplified as a sphere: R = 8 mm, E* = 0.3 MPa, indentation δ = 1 mm. Find the contact radius, the peak pressure, and the total normal force.

  1. Contact radius: a = √(Rδ) = √(8 mm × 1 mm) = 2.83 mm.
  2. Total force: F = (4/3)E*√R·δ3/2 = (4/3)×0.3 MPa×√8 mm×(1 mm)1.5 = (4/3)×0.3×2.83 = 1.13 N.
  3. Peak pressure: p₀ = 2E*δ/(πa) = 2×0.3×1/(π×2.83) = 0.0675 MPa = 67.5 kPa.
  4. Check whether the assumptions hold: δ/R = 1/8 = 12.5%, outside the comfort zone of small deformation (empirically <10%), and silicone is hyperelastic with real adhesion — this example deliberately straddles the line: a real fingertip's force would be slightly below the Hertz prediction (large deformation flattens the contact zone) while adhesion raises the force near zero load. The directions are known; the quantitative answer needs numerics or measurement.
Answer

a ≈ 2.83 mm, F ≈ 1.13 N, p₀ ≈ 67.5 kPa. And you know the credibility boundary of all three numbers: a 12.5% indentation ratio is already at the edge of the assumptions — knowing how to use the formula and knowing when the formula fails are two different things.

Run the numbers

Where the numbers come from

Nondimensionalized (E*=1, R=1); ring integrals of the Hertz pressure distribution compute the surface displacement (complete elliptic integrals via AGM iteration, 400 panels × 24-point Gauss):

  • Shape condition: over 12 radius points inside the contact zone, the maximum relative deviation of u(r) from δ − r²/(2R) is 0.04%
  • Power-law exponent of the total force: a six-point log-log fit over δ ∈ [0.01, 0.4] gives slope 1.5000 (theory 1.5)
  • Point-by-point comparison with the closed form F=(4/3)E*√R δ3/2: relative difference ≤ 0.001%

Where this shows up

The object hereThe counterpart in RLUsed in
F ∝ δ3/2 and contact stiffness k ∝ √δThe physical origin of the reading→force calibration curve of tactile sensors; the mechanism of getting harder the more you pressused in →
a² = Rδ (the contact zone grows with load)Why soft-fingertip grasping is stable: the true contact area is far larger than the nominal areaused in →
Edge pressure singularity p(r) ∝ √(a−r)The mechanical setup for slip starting at the contact edge (T15 presliding)used in →
The uniqueness theorem and verify-style proofsThe legitimacy basis of guess-verify iterations in contact solvers inside simulation enginesused in →

Boundaries: where this article stops

Where this article stops: the potential-theory proof of the Boussinesq solution, the contact ellipse for general 3D surfaces, large deformation (δ/R ≳ 10%), adhesive contact (JKR/DMT theory — the real situation of a silicone fingertip), and viscoelastic hysteresis — none covered. Honest statement: the real calibration curve of a silicone fingertip stacks a hyperelastic constitutive law and adhesion on top of Hertz — Hertz gives the baseline before the first correction term appears, not the endpoint.

Self-test: 4 questions

All hand-computation and «find the flaw in this proof» questions — nothing to memorize. Stuck? Go back to the section.

01Two steel balls of the same material pressed against each other versus a steel ball pressed on a rigid plane: which contact is softer? By how much?

The one where both bodies deform is softer: 1/E* = 2(1−ν²)/E, the effective modulus is halved. For steel ν≈0.3, E* = E/(2×0.91) ≈ 0.55E, twice as soft as a single ball on a rigid body (E*≈1.1E). The larger the contact area, the larger the displacement — each elastic body contributes one share of compliance.

02Where does the 3/2 power of δ in the Hertz law come from? Explain the mechanism in one sentence.

The contact area grows linearly with indentation (a²=Rδ): double the indentation, double the load-bearing area, which is equivalent to a spring stiffness proportional to √δ — integrating, F = ∫k dδ ∝ δ3/2. The nonlinearity comes entirely from geometry; the material is linear elastic.

03Why is the pressure at the contact edge called singular even though it tends to zero?

p(r) ∝ √(1−r²/a²) hits zero at r→a with a vertical slope; its gradient diverges at the edge. The strain gradient in the edge material is enormous — exactly the mechanical root of local slip starting at the edge under tangential load (the Cattaneo-Mindlin mechanism of T15).

04At 12.5% indentation of a silicone fingertip, is the Hertz prediction still trustworthy? How should it be handled?

The direction is trustworthy, the numbers are not: large deformation makes the actual force lower, adhesion makes it higher near zero load. The engineering treatment: use Hertz as the skeleton of the shape function and calibrate the magnitude against measurements — one reason unified tactile representation still encodes separate entry points per sensing principle and material (see C9).

Up next

The full series · 27 articles

T1The Matrix Is a Ledger for Rotation≈ 25 minT2From Slope to Gradient: The First Line of Every Robot Learning Algorithm≈ 25 minT3Vectors and forces: the physical reading of the dot productAbout 20 minT4Two Rulers and a Better Rulerabout 25 minT5The Ledger of One Motor Revolution: Radians, Speed, and Powerabout 20 minT6Complex multiplication is rotation: the one-dimensional version of quaternionsAbout 20 minT7Bases and coordinates: why robots put everything in matrices45 minT8Matrix Multiplication: Chaining Transformations, Not a Table OperationAbout 40 minT9The Jacobian: Rotate a Joint One Degree — How Far Does the End Effector Move?About 45 minT10Eigenvectors: The Directions a Matrix Dare Not Touchabout 45 minT11SVD and the Pseudoinverse: When There Is No Exact Solution, Find the Nearest Oneabout 50 minT12Taylor expansion: putting a straight line to work in place of a curveAbout 40 minT13Probability spaces: why «random» needs axioms50 minT14Expectation: A Rigorous Construction of «Average»About 55 minT15Conditional Expectation: What the Vertical Bar Really Means≈ 60 minT16Four inequalities: the quantitative answer to «how many samples are enough»45 minT17Norms and fixed points: where «it always converges» comes from55 minT18Markov Chains: When the Time Average Equals the Ensemble Averageabout 55 minT19The Mathematical Construction of the MDP: What One Axiom Is Worthabout 50 minT20Dynamic Programming: The Accounting for Two Iterations≈ 50 minT21Optimization: why constraints cannot rely on penalties aloneAbout 55 minT22Stochastic Approximation: Every Term in the Learning-Rate Formulaabout 55 minT23The Mathematics of Contact: Hertz Theoryabout 45 minT24Friction and the Friction Cone: The Geometry of Constraintsabout 40 minT25Force Closure and Grasp Metrics: ε Is a Linear ProgramAbout 50 minT26Contact Kinematics: Why Rolling Is Not «Sliding Along»≈ 45 minT27Slip detection: from «I felt it» to «I computed it»About 50 min