Diffusion on Curved Spaces
The animation above shows a sealed rigid container whose walls curve inward and outward. Inside, a swarm of particles begins concentrated near a single point and diffuses outward. The container is a compact region of bounded by a curved wall (a deformed ball) and the shape of the wall determines how the swarm spreads. Near the poles of the container, where the wall curves sharply, the particles are reflected at steeper angles than on the flatter equatorial band. The gold-to-blue colour encodes the heat kernel value evaluated at each particle's position: gold is high concentration, deep blue is low.
That formula is the heat kernel in flat space , up to normalisation. The interior of the container is flat and the wall enters only through reflection. On a curved manifold the formula is an approximation and the true kernel acquires corrections at every order in with coefficients that are curvature invariants. The first such correction appears at order and is , with the scalar curvature.
This post derives that result from first principles. We begin with the classical heat equation on and its Gaussian fundamental solution (§1). We then build the Riemannian machinery (metric tensor, covariant gradient, divergence and Laplace–Beltrami operator) from the ground up (§2). The heat equation on a Riemannian manifold has its own fundamental solution, derived through the parametrix construction (§3). We close with the spectral representation of the heat kernel and Weyl's law (§4).
Sign convention. The Laplace–Beltrami operator is defined as and is negative semi-definite on functions. The heat equation is . The probabilistic convention is and uses the generator of Brownian motion. The two conventions differ by the substitution throughout and their kernels are related by .
1. Heat Equation on
1.1. Derivation from conservation
Let be the concentration of a diffusing substance. Fick's first law [1] states that diffusive flux opposes the concentration gradient. With the diffusion coefficient set to 1:
Conservation of mass over any smooth bounded region requires
where is the outward unit normal to . Applying the divergence theorem to the right side of [eq:conservation] and substituting [eq:fick]:
Since is arbitrary and all integrands are continuous, we conclude:
1.2. Solution via the Fourier transform
For the Fourier transform and its inverse are
Differentiation in becomes multiplication in :
so applying [eq:fourier-deriv] twice in each coordinate and summing:
Taking the Fourier transform of [eq:heat-flat] in and applying [eq:fourier-laplacian]:
This is a first-order linear ODE in for each fixed , with solution
By the convolution theorem this product in frequency space corresponds to convolution in physical space:
1.3. Gaussian heat kernel
The inverse Fourier transform of factors over coordinates and reduces to a one-dimensional integral. Completing the square:
where and the contour shift is justified by Cauchy's theorem. Multiplying over all coordinates gives
By [eq:conv-soln] the solution to [eq:heat-flat] is or equivalently:
The visualisation below shows in one dimension for several values of : the bell flattens and widens as .
The kernel [eq:flat-kernel] satisfies, for all and :
- Positivity: .
- Mass conservation: .
- Symmetry: .
- Semigroup: .
- Approximate identity: for all , uniformly on compact sets.
Proof. (1) is immediate from the definition.
Proof of (2). Substitute to obtain . Then
where the last step uses the standard Gaussian integral , a product of one-dimensional integrals [2].
(3) is immediate from .
Proof of (4). Write . The prefactors contribute . The combined exponent is
Expanding and collecting powers of :
The coefficient of is , and the optimal point in is
Completing the square the combined exponent equals
where the residual is independent of (verified by expanding both sides and cancelling). The Gaussian integral over now factors:
Combining the prefactors and residual:
(5) follows from (2) and the standard theory of approximate identities [2, Ch. 5]. □
The semigroup property has a direct visual interpretation. Fix the total time and vary the split . The blue kernel and the gold kernel change shape continuously while their convolution (the muted silver curve ) stays constant.
2. Riemannian Setup
2.1. Metric tensor and volume form
A Riemannian metric on a smooth -manifold is a smooth, symmetric, positive-definite covariant 2-tensor field . In local coordinates it reads
with Einstein summation throughout. The matrix is symmetric positive definite at every point [3, Ch. 2].
Write for the determinant of the metric matrix. The Riemannian volume form is the natural measure on :
This is independent of the chart: under a coordinate change the Jacobian determinant from the transformation of exactly cancels the factor from . The inner product on functions is
2.2. Gradient and divergence
The gradient of is the unique vector field characterised by
In local coordinates: , where .
Proof of the coordinate formula. We need components satisfying for all vectors . Setting gives , whence . □
The divergence of is the smooth function characterised by
where is the Lie derivative. In local coordinates:
Proof of the coordinate formula. By Cartan's formula applied to the closed top-degree form (for which ), we get . In coordinates, expanding and taking the exterior derivative yields . Comparing with [eq:div-def-eq] gives [eq:div-coords] [4, Ch. 16]. □
2.3. Laplace–Beltrami operator
The Laplace–Beltrami operator on is
Substituting [eq:grad-def-eq] into [eq:div-coords] gives the coordinate formula:
On with the Euclidean metric, , , and [eq:lbo-coords] reduces to the standard Laplacian .
is the negative -gradient of the Dirichlet energy
Explicitly, for any :
Proof. Differentiating under the integral and using the symmetry of :
It remains to show that the last integral equals . Apply the Leibniz rule for the Riemannian divergence to the vector field :
Integrating over and applying the Riemannian divergence theorem (boundary term vanishes because ):
Rearranging yields . This is [eq:first-variation] [3, Ch. 2]. □
2.4. Self-adjointness and Green's identity
For on a compact Riemannian manifold without boundary:
Proof. Apply the Leibniz rule to :
Integrating over and applying the divergence theorem (no boundary since is closed):
Rearranging:
Applying the same argument with and exchanged (the gradient inner product is symmetric):
The two left-hand sides are therefore equal. This gives all three expressions in [eq:green] [5, Ch. 1]. □
Taking confirms is negative semi-definite: .
2.5. Examples
Round 2-sphere. In polar coordinates on :
The eigenfunctions of are the spherical harmonics with eigenvalues , multiplicity [5, Ch. II §4].
Flat torus. On with , the eigenfunctions are the Fourier modes for , eigenvalues , and the heat kernel is
By Poisson summation this sum equals with the flat Gaussian [eq:flat-kernel]. As the lattice images with are exponentially suppressed near the diagonal.
3. Diffusion on a Curved Manifold
3.1. Heat equation on
Let be a smooth Riemannian manifold. The heat equation on is
For compact without boundary, a smooth solution for follows from the spectral theory of §4 via the heat semigroup . For complete non-compact Dodziuk [6] constructed the heat kernel by exhausting with compact domains and proved uniqueness of bounded solutions when the Ricci curvature is bounded below. Li–Yau [7] proved Gaussian bounds for the kernel under the same hypothesis.
3.2. Heat kernel
The heat kernel of is a smooth function such that
solves [eq:heat-curved] for every . It satisfies:
- for all and .
- (symmetry from self-adjointness).
- (equality for compact ).
- (semigroup).
- for all [6].
3.3. Geodesic distance as the natural substitute
On the kernel [eq:flat-kernel] depends on and only through the Euclidean distance . On a Riemannian manifold the canonical substitute is the geodesic distance . The heat kernel is not simply because curvature introduces corrections at every order in . That expression is the leading term. The parametrix ansatz [8] is
where the transport coefficients are smooth functions determined by substituting [eq:parametrix] into the heat equation and matching powers of .
3.4. Transport equations and the short-time expansion
We derive the coefficients by inserting [eq:parametrix] into and reading off each power of .
Setup. Fix and work in normal coordinates centred at (provided by the exponential map ). In these coordinates [3, Ch. 5]:
- and ,
- to second order,
- the volume density satisfies .
Set . The exponential factor is then . Write the ansatz as where .
Action of the heat operator. Using the product formula for with :
We evaluate two quantities. The eikonal identity for geodesic distance:
The identity follows because is the unit radial vector and gives . The Laplacian of : writing and applying the Leibniz rule,
Since :
Assembling [eq:product-laplacian], subtracting from , and using [eq:laplacian-phi]:
Collecting terms at order and using :
(with ). Multiplying through by turns the left side into . The -th transport ODE follows:
Coefficient . Set in [eq:transport-ODE]:
Hence . The normalisation condition as requires the leading coefficient to equal 1 at . Since we get and
In particular as in flat space.
Coefficient on the diagonal. Integrate [eq:transport-ODE] for from to :
As both sides vanish. L'Hôpital gives:
It remains to compute at . In normal coordinates has the expansion [8, Ch. 4]:
Since , applying to [eq:theta-expand] at and using (contraction of the Ricci tensor):
Substituting [eq:delta-theta] into [eq:u1-lhopital]:
Short-time diagonal expansion. Combining [eq:u0] and [eq:u1-diag] in [eq:parametrix] evaluated at [9]:
Return to the opening animation. Particles are coloured by , the flat-space kernel. The interior of the container is flat () and the curved wall changes the spread only through reflection. On a curved manifold equation [eq:heat-expansion] quantifies the correction: at a point where the diagonal kernel exceeds the flat baseline and the particle density stays more concentrated because there is less room to spread. Where (saddle regions) the kernel is below the flat value and particles disperse faster.
The higher coefficients are universal polynomials in the curvature tensor and its covariant derivatives. The next term is [10]:
3.5. Convergence of the parametrix
The parametrix [eq:parametrix] approximates the true heat kernel. It satisfies with a nonzero right side. To construct the true kernel, one writes where the correction solves a Volterra integral equation involving . For compact the correction is smooth and . Hence [eq:heat-expansion] is asymptotically exact [11, Ch. 3].
4. Spectral Perspective
4.1. Spectral theorem for compact manifolds
Let be compact and connected without boundary. The operator extends to a self-adjoint unbounded operator on with domain . Its spectrum is a discrete sequence
and the corresponding eigenfunctions are smooth and form a complete orthonormal basis for :
The proof uses compactness of the Sobolev embedding (Rellich–Kondrachov). This forces the resolvent of to be compact. The spectral theorem for compact self-adjoint operators then gives the discrete spectrum; elliptic regularity implies smoothness of eigenfunctions [12, Ch. 8].
4.2. Heat kernel as a spectral series
The heat semigroup acts on by
which converges in for all and in for (exponential decay of dominates polynomial growth of ). The integral kernel of is
and [eq:kernel-rep] with this kernel reproduces [eq:semigroup-def].
On the interval with Dirichlet boundary conditions, and . The figure below animates the solution : faded traces are individual modes, bright gold is the sum. Mode (eigenvalue ) decays 49 times faster than mode .
4.3. Weyl's law
Let . Then
where is the volume of the unit ball in . Equivalently, as .
Proof via the heat trace. Define the heat trace
The two expressions for are equal: substituting the spectral series [eq:spectral-kernel] into the integral gives by orthonormality.
Step 1 (Heat trace asymptotics). Integrating the diagonal expansion [eq:heat-expansion] over :
where is the average scalar curvature. The leading term is
Step 2 (Laplace–Stieltjes transform). Regard as a right-continuous step function on . Since , the Lebesgue–Stieltjes measure is the sum of unit point masses at each , and
This is the Laplace–Stieltjes transform of .
Step 3 (Karamata's Tauberian theorem). If is a non-negative measure on and its Laplace transform satisfies as , then as . Chavel applies it to the heat trace in this way and quotes the theorem from Feller (Vol. II, 2nd ed., 1971, p. 446) [5, Ch. VI §4, p. 155].
Apply this with , and . Then
Recognising and :
Hence as in [eq:weyl].
Step 4 (Equivalent eigenvalue asymptotics). Inverting: if for large , then
□The figure below plots the exact trace (gold) against the Weyl leading term (blue) for the interval . The animated cursor sweeps from large to small: agreement is excellent near but breaks down at large , where only the mode remains.
4.4. Geometry encoded in the spectrum
The heat trace expansion [eq:trace-expansion] shows that the first two coefficients determine the volume and total scalar curvature . For a compact surface (), the Gauss–Bonnet theorem gives where . Hence the second heat invariant encodes the Euler characteristic.
The higher heat invariants are integrals of local curvature polynomials. Their computation underlies the heat-kernel proof of the Atiyah–Singer index theorem [14, 15]: for an elliptic operator on a compact manifold, the alternating heat trace is constant in and equals the Fredholm index ; expanding via the parametrix and integrating expresses the index as an integral of characteristic classes.
References
- A. Fick, "Über Diffusion," Annalen der Physik 170(1) (1855), 59–86. doi:10.1002/andp.18551700105
- E. M. Stein and R. Shakarchi, Fourier Analysis: An Introduction, Princeton University Press, 2003. Princeton UP
- J. M. Lee, Introduction to Riemannian Manifolds, 2nd ed., Springer GTM 176, 2018. doi:10.1007/978-3-319-91755-9
- J. M. Lee, Introduction to Smooth Manifolds, 2nd ed., Springer GTM 218, 2013. doi:10.1007/978-1-4419-9982-5
- I. Chavel, Eigenvalues in Riemannian Geometry, Academic Press, 1984. ScienceDirect
- J. Dodziuk, "Maximum principle for parabolic inequalities and the heat flow on open manifolds," Indiana Univ. Math. J. 32(5) (1983), 703–716. JSTOR
- P. Li and S.-T. Yau, "On the parabolic kernel of the Schrödinger operator," Acta Math. 156 (1986), 153–201. doi:10.1007/BF02399203
- M. Berger, P. Gauduchon, and E. Mazet, Le spectre d'une variété riemannienne, Springer LNM 194, 1971. doi:10.1007/BFb0064643
- S. Minakshisundaram and Å. Pleijel, "Some properties of the eigenfunctions of the Laplace operator on Riemannian manifolds," Canad. J. Math. 1 (1949), 242–256. doi:10.4153/CJM-1949-021-5
- P. B. Gilkey, Invariance Theory, the Heat Equation, and the Atiyah–Singer Index Theorem, 2nd ed., CRC Press, 1995. doi:10.1201/9780203749791
- S. Rosenberg, The Laplacian on a Riemannian Manifold, Cambridge University Press, 1997. doi:10.1017/CBO9780511623783
- M. E. Taylor, Partial Differential Equations II, Springer Applied Math. Sciences 116, 1996. doi:10.1007/978-1-4757-4187-2
- H. Weyl, "Über die asymptotische Verteilung der Eigenwerte," Nachrichten von der Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse (1911), 110–117. EuDML
- M. F. Atiyah, R. Bott, and V. K. Patodi, "On the heat equation and the index theorem," Inventiones Mathematicae 19 (1973), 279–330. doi:10.1007/BF01425417
- E. Getzler, "A short proof of the local Atiyah–Singer index theorem," Topology 25(1) (1986), 111–117. doi:10.1016/0040-9383(86)90008-X