gem

Thermal Dispersion — GEM Chapter 14

Mathematical models for thermal diffusion and heat content in disordered geomedia, derived from first principles in Mathematical Geoenergy (GEM), Chapter 14 (Thermal Energy: Diffusion and Heat Content).


Overview

Chapter 14 applies Maximum Entropy (MaxEnt) uncertainty quantification to thermal diffusion problems — heat conduction in heterogeneous soils, geothermal heat exchangers, and the ocean heat content (OHC) response to greenhouse forcing.

The central observation is that real geological media are never perfectly uniform: rocks, soil layers, pockets of different composition, and varying moisture content all modulate the local thermal diffusivity. Without detailed knowledge of every spatial variation, the best we can assume is a mean diffusion coefficient D₀. Applying MaxEnt to D then yields an exponential distribution p(D) = (1/D₀)·exp(−D/D₀), which is integrated over the standard Gaussian heat kernel to give a closed-form dispersive thermal kernel.

Two levels of MaxEnt disorder are considered:

  1. Disorder in D only — integrating the Gaussian heat kernel over the MaxEnt prior for D replaces the erfc cumulative response with a pure exponential exp(−x/√(D₀t)) (Eq. 14-1) and eliminates the singular initial transient of the standard solution.

  2. Disorder in D and interface position — a second MaxEnt prior for the spatial uncertainty x (mean x₀) gives the doubly-dispersive response (1/2)/(x₀ + √(D₀t)) (Eq. 14-2), a Lomax-type decay in √t.

The same dispersive kernel reappears in:


Equations

Part 1 — MaxEnt Disorder in D: Dispersive Thermal Kernel

Eq. 14-3 — Standard Gaussian heat kernel (ordered medium)

\[\Delta T(x,t) = \frac{C}{\sqrt{Dt}}\cdot e^{-x^2/(Dt)}\]

Impulse response of the heat equation in a uniform medium. The Gaussian widens with time as √(Dt).


Eq. 14-1 — Dispersive thermal cumulative response (MaxEnt D)

\[T(x,t) = T_1\cdot e^{-x/\sqrt{D_0 t}} + T_0\]

Obtained by integrating the Gaussian Green’s function over the MaxEnt prior p(D) = (1/D₀)·exp(−D/D₀) using the Laplace-Bessel identity (Eq. 20-6 in GEM Ch. 20). The erfc is replaced by a pure exponential — no singularity at t = 0, and the tail is fatter than the Gaussian prediction.


Eq. 14-4 — Dispersive impulse response (MaxEnt D)

\[\Delta T(x,t) = \frac{C}{\sqrt{D_0 t}}\cdot e^{-x/\sqrt{D_0 t}} \left(1 + \frac{x}{\sqrt{D_0 t}}\right)\]

The disordered counterpart of Eq. 14-3. The (1 + u)·e^{−u} envelope (with u = x/√(D₀t)) broadens the peak and increases the late-time tail compared with the Gaussian, consistent with observations in disordered heterogeneous media.


Part 2 — Double MaxEnt: Disorder in D and Interface Position

Eq. 14-2 — Doubly-dispersive thermal response

\[T(x,t) = \frac{1}{2}\cdot\frac{1}{x_0 + \sqrt{D_0 t}}\]

Derived by averaging the dispersive impulse response n(x',t) = exp(−x'/√(D₀t))/(2√(D₀t)) over a MaxEnt distribution of source positions p(x') = (1/x₀)·exp(−x'/x₀):

\[T(t) = \int_0^\infty p(x')\cdot n(x',t)\\,dx' = \frac{1}{2(x_0 + \sqrt{D_0 t})}\]

The result is a Lomax-type decay in the variable √t — slower than exponential and consistent with long thermal tails measured in heat-exchanger experiments.


Part 3 — Smeared Diffusive Response at the Origin

Eq. 14-6 — Smeared diffusive step response

\[\Delta T(t) = \frac{1}{1 + \sqrt{t/\tau}}\]

The fraction of heat remaining at the source after time t, where τ = L²/D₀ and L is the characteristic depth scale. Derived by averaging the dispersive cumulative exp(−x/√(D₀t)) over a MaxEnt depth distribution p(x) = (1/L)·exp(−x/L) and taking the complement:

\[\Delta T(t) = 1 - \int_0^\infty \frac{1}{L}\cdot e^{-x/L}\cdot e^{-x/\sqrt{D_0 t}}\\,dx = 1 - \frac{1}{1 + \sqrt{\tau/t}} = \frac{1}{1 + \sqrt{t/\tau}}\]

Matches impulse-response data from earthen heat-exchanger tests (Witte 2006).


Part 4 — Convolution Box Model

Eq. 14-5 — Convolution for thermal box model

\[\text{Response}(t) = \text{Input}(t) \otimes \text{Transfer}(t) = \int_0^t \text{Input}(\tau)\cdot\text{Transfer}(t-\tau)\\,d\tau\]

Models a thermal input coupled through a dispersive transfer function. Used to construct the modulated CPU and geothermal transient responses (Eqs. 14-7, 14-8).


Part 5 — Ocean Heat Content Model

Eq. 14-9 — Depth-integrated heat content

\[I(t) = \int_0^\infty \Delta T(t\mid x)\cdot e^{-x/L}\\,dx\]

The dispersive impulse response ΔT(t|x) (Eq. 14-4) is convolved with the exponential depth weight exp(−x/L), modelling the total heat absorbed by an ocean layer of characteristic depth L.


Eq. 14-10 — Ocean heat content response

\[I(t) = \frac{\sqrt{D_0 t}/L + 2}{\bigl(\sqrt{D_0 t}/L + 1\bigr)^2}\]

Closed-form result of integrating the dispersive impulse response ΔT(t|x) = (1/σ)·exp(−x/σ)·(1+x/σ) (Eq. 14-4, σ = √(D₀t)) over the exponential depth weight exp(−x/L). Setting s = √(D₀t)/L:

\[I(t) = \frac{s + 2}{(s+1)^2}, \qquad s = \frac{\sqrt{D_0 t}}{L}\]

As s → 0 (early time): I → 2 (maximum, pulse at the surface). As s → ∞ (late time): I ~ 1/s → 0 (heat diffuses to great depths).


Eq. 14-12 / 14-13 — Convolution with linear forcing

\[F(t) = k\cdot(t - t_0)\] \[\frac{R(t)}{k} = \tfrac{2L}{3}(D_0 t)^{3/2} - L^2 D_0 t + 2L^3\sqrt{D_0 t} - 2L^4\ln\Bigl(\tfrac{\sqrt{D_0 t}+L}{L}\Bigr)\]

The convolution R(t) = F(t) ⊗ I(t) gives the accumulated OHC under a linearly growing greenhouse forcing, with t₀ ≈ 1960 as the onset year.


Repository Files

File Purpose
thermal_dispersion_symbolic.py Symbolic derivation of all Chapter 14 equations using SymPy
thermal_dispersion_numerical.py Numerical implementation, validation, and composite figure
thermal_dispersion_model_output.png Output figure (6 panels) generated by thermal_dispersion_numerical.py

Usage

Install dependencies (from models/requirements.txt):

pip install -r ../requirements.txt

Run symbolic derivation (all assertions print ✓):

python thermal_dispersion_symbolic.py

Run numerical model and generate figure:

MPLBACKEND=Agg python thermal_dispersion_numerical.py

Key Physical Insights

  1. Disorder transforms erfc → exp: introducing MaxEnt uncertainty in the diffusion coefficient D (Eq. 14-1) replaces the complementary error function profile with a pure exponential exp(−x/√(D₀t)). The initial singularity of the standard kernel vanishes and the late-time tail is heavier.

  2. Double MaxEnt gives Lomax-in-√t: adding MaxEnt uncertainty in the interface position x gives a Lomax-type 1/(x₀ + √(D₀t)) response (Eq. 14-2) — the same functional form that appears in lake sizes (Ch. 15) and contamination spreading (Ch. 20), demonstrating the universality of dispersive MaxEnt.

  3. Geothermal heat-exchanger evidence: the smeared response 1/(1 + √(t/τ)) (Eq. 14-6) matches borehole heat-exchanger measurements with a single parameter τ, far more naturally than the standard erfc formula because geological soil is inherently disordered.

  4. Ocean heat content as a single-parameter model: Eq. 14-10 reproduces the observed OHC increase at multiple ocean depths with a single effective diffusion coefficient D₀ ≈ 2–3 cm²/s. No ocean-circulation model or parameter tuning beyond D₀ and the layer thickness L is needed.

  5. Same dispersive kernel across scales: exp(−x/√(D₀t)) / (2√(D₀t)) appears in geothermal heat diffusion (Ch. 14), battery discharge (Ch. 18), and contaminant transport (Ch. 20), confirming that MaxEnt disorder in D is a universal mechanism across geophysical and engineering systems.


References