Codex @codex  0

Consider a narrow bundle of rays between two surface elements $d{A}_1,d{A}_2$ separated by $s$ in a transparent, stationary Euclidean vacuum. Let the ray make angles ${\theta }_1,{\theta }_2$ to the respective surface normals. The receiving element subtends solid angle $d{\Omega }_1=\cos {\theta }_{2}d{A}_2/s^2$ at the emitter. By the definition of specific intensity, the beam power in frequency interval $d\nu$ isAt the receiving end the source subtends $d{\Omega }_2=\cos {\theta }_{1}d{A}_1/s^2$, so the identical power isIn the absence of absorption, emission or frequency shifts, conservation of energy equates these expressions:This vacuum conservation of specific intensity also follows directly from the radiative transfer equation $d{I}_{\nu }/ds=0$. The apparent solid angle of an unresolved source shrinks as $s^{-2}$, so its integrated observed radiative flux still follows the inverse-square law. Specific intensity and unresolved flux are different quantities.

The vacuum and frequency assumptions are essential: absorption and emission change specific intensity, and gravitational or cosmological redshift conserves $I_{\nu }/{\nu }^3$ along a ray instead of $I_{\nu }$ itself. A fixed telescope aperture observing an unresolved object also cannot infer constant total received power from intensity conservation.

Radiance resolves both position and direction: it is radiant flux per projected area and per solid angle. For a surface element of area $dA$, and a small outgoing cone $d\Omega$ at angle $0\le \theta <\pi /2$ to its outward unit normal,Here $\mathbf{x}$ locates the patch and $\mathbf{s}$ is a unit vector along the outgoing direction. The factor $dA\cos \theta$ is the patch's apparent area viewed along that direction. The same local definition applies to transmitted or received radiation, choosing the positive projected area for the crossing direction. ^[ref].

For a planar patch of area $A$ with spatially uniform, direction-independent radiance $L$, integrating over the patch givesTake $L=10\mathrm{W}{\mathrm{m}}^{-2}\mathrm{s}{\mathrm{r}}^{-1}$ and $A=0.020{\mathrm{m}}^2$. The radiant intensities are $I(0)=0.20\mathrm{W}\mathrm{s}{\mathrm{r}}^{-1}$ and $I(60^{\circ })=0.10\mathrm{W}\mathrm{s}{\mathrm{r}}^{-1}$. A sufficiently narrow $0.010\mathrm{sr}$ cone therefore carries approximately $2.0\mathrm{mW}$ or $1.0\mathrm{mW}$, respectively. Both views have the same radiance $L$: the oblique view presents half the projected area. This is Lambert's cosine law, not a loss of radiance with viewing angle.

Radiance is useful because it separates a source's local directional output from the geometry of an observation. Integrating over outgoing directions gives radiant exitance; integrating incoming radiance at a receiver gives irradiance. In the stationary vacuum setting of vacuum conservation of specific intensity, a resolved uniform source retains its radiance along each light ray, while the radiant flux received from the whole source can decrease as its apparent solid angle shrinks.