Biot Number Framework for Heat Conduction

• Biot Number Framework is a rigorous asymptotic analysis for heat conduction with Robin boundary conditions, quantifying surface heat transfer relative to internal conduction.
• It refines the classical lumped-capacitance model by incorporating higher-order corrections and explicit error estimates to accurately predict average temperatures.
• The framework employs streamlined elliptic solves to compute key functionals like φ, providing enhanced thermal analysis for heterogeneous and complex domains.

The Biot number framework provides a rigorous asymptotic structure for analyzing heat conduction in solid bodies with convection (Robin) boundary conditions, focusing on the "dunking" problem: a body at initial temperature $T_{\text i}$ rapidly exposed to a thermal environment at $T_\infty$. The crucial dimensionless parameter is the Biot number ($\operatorname{Bi}$), expressing the ratio of surface heat transfer to internal thermal conduction. This framework yields systematic higher-order corrections, explicit error estimates, and principled refinements to the classical "lumped-capacitance" approximation, enabling accurate characterization of average temperatures and error bounds across a broad class of domains and heterogeneous materials (Kaneko et al., 2024).

1. Mathematical Structure of the Dunking Problem

The framework considers a body occupying a Lipschitz domain $\Omega \subset \mathbb{R}^d$ ($d=1,2,3$) with boundary $\partial \Omega$. The temperature $T(x, t)$ evolves under

$$\rho(x) c(x)\, \partial_t T = \nabla \cdot [k(x) \nabla T] \quad \text{in } \Omega, \quad t>0,$$

subject to Robin (Newton’s law of cooling) boundary conditions,

$$k(x) \partial_n T + h(x) [T - T_\infty] = 0 \quad \text{on } \partial \Omega,$$

and uniform initial temperature $T(x,0)=T_{\text i}$.

Non-dimensionalization with respect to a length scale $T_\infty$0 and reference quantities leads to

$T_\infty$1

$T_\infty$2

with initial condition $T_\infty$3, where $T_\infty$4 is the normalized temperature, $T_\infty$5, $T_\infty$6, and $T_\infty$7 is the extrinsic Biot number. The intrinsic Biot number is defined as $T_\infty$8 with $T_\infty$9 and surface-to-volume ratio $\operatorname{Bi}$0.

2. Physical Interpretation and Significance of the Biot Number

The Biot number quantifies the relative difficulty of heat transfer at the boundary compared to the rate of internal conduction. Specifically, $\operatorname{Bi}$1 signifies that conduction is much faster than boundary transfer, justifying the nearly isothermal, "lumped-capacitance" regime. In the opposite limit, large $\operatorname{Bi}$2 implies substantial internal gradients.

3. Lumped-Capacitance and Higher-Order Approximations

The principal outputs of interest are the domain-average temperature $\operatorname{Bi}$3 and boundary-average temperature $\operatorname{Bi}$4. By separation of variables, the solution admits an eigenexpansion: $\operatorname{Bi}$5 where $\operatorname{Bi}$6 and $\operatorname{Bi}$7 solve the associated elliptic eigenproblem.

For small $\operatorname{Bi}$8, the leading eigenvalue satisfies $\operatorname{Bi}$9, where $\Omega \subset \mathbb{R}^d$0 is a domain- and property-dependent functional. Neglecting $\Omega \subset \mathbb{R}^d$1 yields the standard lumped-capacitance model,

$\Omega \subset \mathbb{R}^d$2

with explicit solution $\Omega \subset \mathbb{R}^d$3.

A second-order Padé-type improvement uses a rational approximation $\Omega \subset \mathbb{R}^d$4, giving

$\Omega \subset \mathbb{R}^d$5

4. Error Estimators and Functional Outputs

A distinguishing feature of the framework is the derivation of rigorous error estimators for both first- and second-order models. For slow time $\Omega \subset \mathbb{R}^d$6, the maximum absolute error in the lumped estimator $\Omega \subset \mathbb{R}^d$7 is

$\Omega \subset \mathbb{R}^d$8

with a universal bound for all $\Omega \subset \mathbb{R}^d$9: $d=1,2,3$0

For $d=1,2,3$1, the error is $d=1,2,3$2 with the explicit estimator

$d=1,2,3$3

where $d=1,2,3$4 and $d=1,2,3$5 are functionals involving the solution $d=1,2,3$6 to a sensitivity elliptic problem.

The key quantity $d=1,2,3$7—a single elliptic (Helmholtz-type) solve plus quadratic post-processing—fully characterizes first-order error and the second-order Padé correction.

5. Conductive Length Scale and Domain Effects

The functional $d=1,2,3$8 admits a physical interpretation as an effective "conduction length" $d=1,2,3$9, governing the amplification of the standard length scale $\partial \Omega$0 to account for nonuniformities. In canonical domains with constant coefficients ($\partial \Omega$1), values are $\partial \Omega$2 (interval), $\partial \Omega$3 (disk), and $\partial \Omega$4 (sphere). For slender or geometrically complex shapes, $\partial \Omega$5 (e.g., sharp-cornered triangles with small aspect ratio $\partial \Omega$6 have $\partial \Omega$7), indicating severe internal gradients even when $\partial \Omega$8.

Monotonicity and invariance properties include: $\partial \Omega$9 always; $T(x, t)$0 is invariant under translation, rotation, and scale; and $T(x, t)$1 is nonincreasing in $T(x, t)$2. For heterogeneous materials ($T(x, t)$3), explicit bounds in terms of $T(x, t)$4 and $T(x, t)$5-variance are available (Proposition 5.4).

6. Refined Lumped-Capacitance Criterion

The classical lumped-capacitance criterion, Biot number $T(x, t)$6, presumes $T(x, t)$7. The refined small-Biot condition emerging from this framework is $T(x, t)$8. Thus, for domains with $T(x, t)$9, the classical threshold is insufficient. The improved criterion dictates $$\rho(x) c(x)\, \partial_t T = \nabla \cdot [k(x) \nabla T] \quad \text{in } \Omega, \quad t>0,$$0 for uniformly small internal gradients, with direct implications for thermal system design and simulation.

7. Practical Implementation and Numerical Validation

Evaluation of $$\rho(x) c(x)\, \partial_t T = \nabla \cdot [k(x) \nabla T] \quad \text{in } \Omega, \quad t>0,$$1 proceeds via a single sparse elliptic solve (for $$\rho(x) c(x)\, \partial_t T = \nabla \cdot [k(x) \nabla T] \quad \text{in } \Omega, \quad t>0,$$2) on the specific domain and material parameters. The marginal effort to compute additional functionals $$\rho(x) c(x)\, \partial_t T = \nabla \cdot [k(x) \nabla T] \quad \text{in } \Omega, \quad t>0,$$3, $$\rho(x) c(x)\, \partial_t T = \nabla \cdot [k(x) \nabla T] \quad \text{in } \Omega, \quad t>0,$$4 is negligible. For $$\rho(x) c(x)\, \partial_t T = \nabla \cdot [k(x) \nabla T] \quad \text{in } \Omega, \quad t>0,$$5, the first-order lumped model $$\rho(x) c(x)\, \partial_t T = \nabla \cdot [k(x) \nabla T] \quad \text{in } \Omega, \quad t>0,$$6 provides $$\rho(x) c(x)\, \partial_t T = \nabla \cdot [k(x) \nabla T] \quad \text{in } \Omega, \quad t>0,$$7 to within $$\rho(x) c(x)\, \partial_t T = \nabla \cdot [k(x) \nabla T] \quad \text{in } \Omega, \quad t>0,$$85% absolute error for times up to $$\rho(x) c(x)\, \partial_t T = \nabla \cdot [k(x) \nabla T] \quad \text{in } \Omega, \quad t>0,$$9; for $$k(x) \partial_n T + h(x) [T - T_\infty] = 0 \quad \text{on } \partial \Omega,$$0, the second-order $$k(x) \partial_n T + h(x) [T - T_\infty] = 0 \quad \text{on } \partial \Omega,$$1 reduces the error to $$k(x) \partial_n T + h(x) [T - T_\infty] = 0 \quad \text{on } \partial \Omega,$$2. These accuracy guarantees are supported by adaptive finite-element computations on domains spanning sharp triangles, finned blocks, and gear-like boundaries, as well as various heterogeneous media. The bounds are sharp for small Biot and conservative for larger $$k(x) \partial_n T + h(x) [T - T_\infty] = 0 \quad \text{on } \partial \Omega,$$3, demonstrating the framework's reliability and generality (Kaneko et al., 2024).