= Solution
In an <electrical insulator>, <lattice> vibrations usually carry most heat. A small <temperature gradient> produces an imbalance between <phonons> moving in opposite directions. In an isotropic kinetic approximation,
$$
\boxed{\kappa_{\rm ph}\simeq\frac13c_Vv_s\ell_{\rm tr}
=\frac13c_Vv_s^2\tau_{\rm tr},}
$$
where $c_V$ is <heat capacity> per unit <volume>, $v_s$ an appropriate acoustic speed and $\ell_{\rm tr}$ the transport <mean free path>. The factor one third is the angular average of the squared direction cosine. With several branches and frequency-dependent scattering, replace this estimate by the corresponding weighted mode integral. A short lifetime matters for conductivity only insofar as it relaxes heat current.
For a clean three-dimensional <crystal> at $T\ll\Theta_D$, the <Debye heat capacity> varies as $T^3$. If diffuse boundary scattering fixes $\ell_{\rm tr}$ near a specimen dimension, then \b[$\kappa_{\rm ph}\propto T^3$]. The size dependence is a test of that boundary-limited regime. Specular surfaces, defects and small specimens can change the simple law; at ballistic scales a local conductivity description itself needs care.
As <temperature> rises, more <phonons> are excited, but phonon-phonon scattering shortens their transport path. Normal processes conserve total <crystal momentum>, whereas an <Umklapp process> satisfies $\mathbf q_1+\mathbf q_2-\mathbf q_3=\mathbf G\ne0$ with reciprocal vector $\mathbf G$ and can relax directed heat flow. Low-temperature <Umklapp scattering> is strongly suppressed; its increasing importance often produces a conductivity maximum followed by a decline. Isotope and impurity scattering lower or broaden that maximum.
At sufficiently high <temperature>, $c_V$ approaches its classical constant while a typical anharmonic Umklapp rate grows approximately as $T$. Under that further approximation,
$$
\boxed{\ell_{\rm tr}\propto T^{-1},\qquad
\kappa_{\rm ph}\propto T^{-1}.}
$$
Thus a clean crystalline insulator commonly shows a low-temperature rise, a peak, then a high-temperature fall. This is not a universal curve for every insulator: glasses, strong disorder, multiple scattering mechanisms, radiative heat transfer and structural transitions can give different behavior.
A steady-state bar experiment supplies a concrete test. Inject known heating power, measure the small <temperature> difference between two separated sensors and infer $\kappa=PL/(A\Delta T)$ after correcting for heat leaks. Keep the gradient small enough that a single representative <temperature> is meaningful, calibrate sensor contact effects and subtract radiative/background losses. Repeat with different widths or isotope/defect content to distinguish boundary limitation from intrinsic scattering. A power-dependent result may reflect a changing <temperature> profile rather than a new microscopic law.
Back to article page
