Let A ∈ SO ( 3 ) A \in \SO\p{3} A ∈ SO ( 3 ) , so that ∣ det A ∣ = 1 \abs{\det{A}} = 1 ∣ det A ∣ = 1 , A − 1 = A ⊤ A^{-1} = A^\trans A − 1 = A ⊤ , and A A A is an automorphism of R 3 \R^3 R 3 . In particular, A − 1 x ⋅ ξ = x ⋅ A ξ A^{-1}x \cdot \xi = x \cdot A\xi A − 1 x ⋅ ξ = x ⋅ A ξ , so by change of variables,
∫ S 2 e i x ⋅ ξ d σ ( x ) = ∣ det A − 1 ∣ ∫ S 2 e i A − 1 x ⋅ ξ = ∫ S 2 e i x ⋅ A ξ d x . \int_{S^2} e^{ix \cdot \xi} \,\diff\sigma\p{x}
= \abs{\det{A^{-1}}} \int_{S^2} e^{iA^{-1}x \cdot \xi}
= \int_{S^2} e^{ix \cdot A\xi} \,\diff{x}. ∫ S 2 e i x ⋅ ξ d σ ( x ) = ∣ ∣ det A − 1 ∣ ∣ ∫ S 2 e i A − 1 x ⋅ ξ = ∫ S 2 e i x ⋅ A ξ d x .
In other words, this quantity is invariant under rotations, so it only depends on ∣ ξ ∣ \abs{\xi} ∣ ξ ∣ . Hence, it suffices to calculate the integral when ξ = ( 0 , 0 , ∣ ξ ∣ ) \xi = \p{0, 0, \abs{\xi}} ξ = ( 0 , 0 , ∣ ξ ∣ ) . By using spherical coordinates, we have
∫ S 2 e i x ⋅ ξ d σ ( x ) = ∫ S 2 cos ( ∣ ξ ∣ x 3 ) + i sin ( ∣ ξ ∣ x 3 ) d σ ( x ) = ∫ 0 π ∫ 0 2 π ( cos ( ∣ ξ ∣ cos φ ) + i sin ( ∣ ξ ∣ cos φ ) ) sin φ d θ d φ = 2 π ∫ − 1 1 cos ( ∣ ξ ∣ u ) + i sin ( ∣ ξ ∣ u ) d u ( u = cos φ ) = 4 π sin ∣ ξ ∣ ∣ ξ ∣ . \begin{aligned}
\int_{S^2} e^{ix \cdot \xi} \,\diff\sigma\p{x}
&= \int_{S^2} \cos\p{\abs{\xi}x_3} + i\sin\p{\abs{\xi}x_3} \,\diff\sigma\p{x} \\
&= \int_0^\pi \int_0^{2\pi} \p{\cos\p{\abs{\xi}\cos\phi} + i\sin\p{\abs{\xi}\cos\phi}} \sin\phi \,\diff\theta \,\diff\phi \\
&= 2\pi \int_{-1}^1 \cos\p{\abs{\xi}u} + i\sin\p{\abs{\xi}u} \,\diff{u}
&& (u = \cos\phi) \\
&= \frac{4\pi\sin{\abs{\xi}}}{\abs{\xi}}.
\end{aligned} ∫ S 2 e i x ⋅ ξ d σ ( x ) = ∫ S 2 cos ( ∣ ξ ∣ x 3 ) + i sin ( ∣ ξ ∣ x 3 ) d σ ( x ) = ∫ 0 π ∫ 0 2 π ( cos ( ∣ ξ ∣ cos φ ) + i sin ( ∣ ξ ∣ cos φ ) ) sin φ d θ d φ = 2 π ∫ − 1 1 cos ( ∣ ξ ∣ u ) + i sin ( ∣ ξ ∣ u ) d u = ∣ ξ ∣ 4 π sin ∣ ξ ∣ . ( u = cos φ )
Define
L ( f ) = ∫ S 2 ∫ S 2 f ( x + y ) d σ ( x ) d σ ( y ) L\p{f} = \int_{S^2} \int_{S^2} f\p{x + y} \,\diff\sigma\p{x} \,\diff\sigma\p{y} L ( f ) = ∫ S 2 ∫ S 2 f ( x + y ) d σ ( x ) d σ ( y )
for f ∈ C c ∞ ( R 3 ) f \in C^\infty_c\p{\R^3} f ∈ C c ∞ ( R 3 ) . Since C c ∞ ( R 3 ) C^\infty_c\p{\R^3} C c ∞ ( R 3 ) is dense in L 2 ( R 3 ) L^2\p{\R^3} L 2 ( R 3 ) with respect to the L 2 L^2 L 2 norm, it suffices to show that there exists a constant C > 0 C > 0 C > 0 so that ∣ L ( f ) ∣ ≤ C ∥ f ∥ L 2 \abs{L\p{f}} \leq C\norm{f}_{L^2} ∣ L ( f ) ∣ ≤ C ∥ f ∥ L 2 for any f ∈ C c ∞ ( R 3 ) f \in C^\infty_c\p{\R^3} f ∈ C c ∞ ( R 3 ) . The derivatives of such an f f f are compactly supported as well, so f f f is in the Schwartz space. In particular, we may apply the inverse Fourier transform and a change of variables to get
f ( x ) = ∫ R 3 e 2 π i x ⋅ ξ f ^ ( ξ ) d ξ = ∫ 0 ∞ r 2 ∫ S 2 e 2 π i x ⋅ r ξ f ^ ( r ξ ) d σ ( ξ ) d r . f\p{x}
= \int_{\R^3} e^{2\pi ix \cdot \xi}\hat{f}\p{\xi} \,\diff\xi
= \int_0^\infty r^2 \int_{S^2} e^{2\pi ix \cdot r\xi}\hat{f}\p{r\xi} \,\diff\sigma\p{\xi} \,\diff{r}. f ( x ) = ∫ R 3 e 2 πi x ⋅ ξ f ^ ( ξ ) d ξ = ∫ 0 ∞ r 2 ∫ S 2 e 2 πi x ⋅ r ξ f ^ ( r ξ ) d σ ( ξ ) d r .
Notice that by integrating by parts, we see that ∥ f ^ ∥ L ∞ ( r S 2 ) \norm{\hat{f}}_{L^\infty\p{rS^2}} ∥ ∥ f ^ ∥ ∥ L ∞ ( r S 2 ) decays faster than any power of r r r , since f ( n ) f^{\p{n}} f ( n ) has compact support. Thus, the integrand is absolutely convergent, and so we may apply Fubini's theorem in the following:
L ( f ) = ∫ S 2 ∫ S 2 f ( x + y ) d σ ( x ) d σ ( y ) = ∫ S 2 ∫ S 2 ∫ 0 ∞ r 2 ∫ S 2 e 2 π i ( x + y ) ⋅ r ξ f ^ ( r ξ ) d σ ( ξ ) d r d σ ( x ) d σ ( y ) = ∫ 0 ∞ ∫ S 2 r 2 f ^ ( r ξ ) ( ∫ S 2 e i x ⋅ 2 π r ξ d σ ( x ) ) ( ∫ S 2 e i y ⋅ 2 π r ξ d σ ( y ) ) d σ ( ξ ) d r = ∫ 0 ∞ ∫ S 2 r 2 f ^ ( r ξ ) ( 4 π sin ∣ 2 π r ξ ∣ ∣ 2 π r ξ ∣ ) 2 d σ ( ξ ) d r = 4 ∫ R 3 f ^ ( ξ ) ( sin 2 π ∣ ξ ∣ ∣ ξ ∣ ) 2 d ξ . \begin{aligned}
L\p{f}
&= \int_{S^2} \int_{S^2} f\p{x + y} \,\diff\sigma\p{x} \,\diff\sigma\p{y} \\
&= \int_{S^2} \int_{S^2} \int_0^\infty r^2 \int_{S^2} e^{2\pi i\p{x+y} \cdot r\xi}\hat{f}\p{r\xi} \,\diff\sigma\p{\xi} \,\diff{r} \,\diff\sigma\p{x} \,\diff\sigma\p{y} \\
&= \int_0^\infty \int_{S^2} r^2\hat{f}\p{r\xi} \p{\int_{S^2} e^{ix \cdot 2\pi r\xi} \,\diff\sigma\p{x}} \p{\int_{S^2} e^{iy \cdot 2\pi r\xi} \,\diff\sigma\p{y}} \,\diff\sigma\p{\xi} \,\diff{r} \\
&= \int_0^\infty \int_{S^2} r^2\hat{f}\p{r\xi} \p{\frac{4\pi\sin\abs{2\pi r\xi}}{\abs{2\pi r\xi}}}^2 \,\diff\sigma\p{\xi} \,\diff{r} \\
&= 4 \int_{\R^3} \hat{f}\p{\xi} \p{\frac{\sin 2\pi \abs{\xi}}{\abs{\xi}}}^2 \,\diff{\xi}.
\end{aligned} L ( f ) = ∫ S 2 ∫ S 2 f ( x + y ) d σ ( x ) d σ ( y ) = ∫ S 2 ∫ S 2 ∫ 0 ∞ r 2 ∫ S 2 e 2 πi ( x + y ) ⋅ r ξ f ^ ( r ξ ) d σ ( ξ ) d r d σ ( x ) d σ ( y ) = ∫ 0 ∞ ∫ S 2 r 2 f ^ ( r ξ ) ( ∫ S 2 e i x ⋅ 2 π r ξ d σ ( x ) ) ( ∫ S 2 e i y ⋅ 2 π r ξ d σ ( y ) ) d σ ( ξ ) d r = ∫ 0 ∞ ∫ S 2 r 2 f ^ ( r ξ ) ( ∣ 2 π r ξ ∣ 4 π sin ∣ 2 π r ξ ∣ ) 2 d σ ( ξ ) d r = 4 ∫ R 3 f ^ ( ξ ) ( ∣ ξ ∣ sin 2 π ∣ ξ ∣ ) 2 d ξ .
The last equality comes from a change of variables. By Plancherel, ∥ f ^ ∥ L 2 = ∥ f ∥ L 2 \norm{\hat{f}}_{L^2} = \norm{f}_{L^2} ∥ ∥ f ^ ∥ ∥ L 2 = ∥ f ∥ L 2 . If we set h ( ξ ) = ( sin 2 π ∣ ξ ∣ ∣ ξ ∣ ) 2 h\p{\xi} = \p{\frac{\sin 2\pi \abs{\xi}}{\abs{\xi}}}^2 h ( ξ ) = ( ∣ ξ ∣ s i n 2 π ∣ ξ ∣ ) 2 , then
lim ξ → 0 h ( ξ ) = 2 π , \lim_{\xi\to0} h\p{\xi} = 2\pi, ξ → 0 lim h ( ξ ) = 2 π ,
so h h h is bounded near the origin. On the other hand,
∣ h ( ξ ) 2 ∣ ≤ 1 ∣ ξ ∣ 4 , \abs{h\p{\xi}^2}
\leq \frac{1}{\abs{\xi}^4}, ∣ ∣ h ( ξ ) 2 ∣ ∣ ≤ ∣ ξ ∣ 4 1 ,
so h ∈ L 2 ( R 3 ) h \in L^2\p{\R^3} h ∈ L 2 ( R 3 ) . Thus, by Cauchy-Schwarz,
∣ L ( f ) ∣ ≤ 4 ∥ h ∥ L 2 ∥ f ^ ∥ L 2 = 4 ∥ h ∥ L 2 ∥ f ∥ L 2 , \abs{L\p{f}}
\leq 4\norm{h}_{L^2}\norm{\hat{f}}_{L^2}
= 4\norm{h}_{L^2}\norm{f}_{L^2}, ∣ L ( f ) ∣ ≤ 4 ∥ h ∥ L 2 ∥ ∥ f ^ ∥ ∥ L 2 = 4 ∥ h ∥ L 2 ∥ f ∥ L 2 ,
which completes the proof.