Spectral norm

1. The H1H^1 calculation (concrete)

Let {φj}j0\{\varphi_j\}_{j\ge0}​ be an orthonormal L2L^2-basis of eigenfunctions of Δ-\Delta with eigenvalues λj0\lambda_j\ge0:

Δφj=λjφj,φj,φL2=δj.-\Delta\varphi_j=\lambda_j\varphi_j,\qquad\langle\varphi_j,\varphi_\ell\rangle_{L^2}=\delta_{j\ell}.

Expand w=jwjφjw=\sum_j w_j\varphi_j​ (convergence in L2L^2). Then

(1Δ)1/2wL22=j(1+λj)wj2,\|(1-\Delta)^{1/2}w\|_{L^2}^2 =\sum_j \big(1+\lambda_j\big)\,|w_j|^2,

because (1Δ)1/2φj=(1+λj)1/2φj(1-\Delta)^{1/2}\varphi_j=(1+\lambda_j)^{1/2}\varphi_j​ and Parseval / Plancherel give the squared L2L^2-norm as the sum of squared coefficients.

So the spectral sum with weight (1+λj)(1+\lambda_j) is exactly the squared L2L^2-norm of (1Δ)1/2(1-\Delta)^{1/2}.

Now the usual Sobolev H1H^1 norm on a compact manifold is equivalent to the graph norm of (1Δ)1/2(1-\Delta)^{1/2}; concretely there exist constants c,C>0c,C>0 depending only on the manifold so that

c  (1Δ)1/2wL2wH1C  (1Δ)1/2wL2.c\;\|(1-\Delta)^{1/2}w\|_{L^2}\le \|w\|_{H^1}\le C\;\|(1-\Delta)^{1/2}w\|_{L^2}.

Combining this with the identity above yields

wH12j(1+λj)wj2,\|w\|_{H^1}^2 \simeq \sum_j (1+\lambda_j)\,|w_j|^2,

which is the H1H^1 instance of the claim.

(Why is the equivalence true? On a compact manifold wH12\|w\|_{H^1}^2​ is wL22+wL22\|w\|_{L^2}^2+\|\nabla w\|_{L^2}^2​. Using the eigen-expansion one checks wL22=jλjwj2\|\nabla w\|_{L^2}^2=\sum_j\lambda_j|w_j|^2. Hence wH12=j(1+λj)wj2\|w\|_{H^1}^2=\sum_j(1+\lambda_j)|w_j|^2 up to the choice of normalization and conventions — so in fact for H1H^1 one often gets equality (no constants) when you identify norms appropriately.)

2. The general HkH^k (spectral functional calculus)

Exactly the same spectral computation works for any real k0k\ge0. By functional calculus

(1Δ)k/2φj=(1+λj)k/2φj,(1-\Delta)^{k/2}\varphi_j=(1+\lambda_j)^{k/2}\varphi_j,

so for w=jwjφjw=\sum_j w_j\varphi_j,

(1Δ)k/2wL22=j(1+λj)kwj2.\|(1-\Delta)^{k/2}w\|_{L^2}^2 =\sum_j (1+\lambda_j)^{k}|w_j|^2.

Thus the right-hand spectral sum is exactly the squared L2L^2-norm of (1Δ)k/2w(1-\Delta)^{k/2}w.

3. Equivalence of (1Δ)k/2(1-\Delta)^{k/2}-graph norm and the usual Sobolev HkH^k norm

What remains is to explain why the graph norm (1Δ)k/2wL2\|(1-\Delta)^{k/2}w\|_{L^2}​ is equivalent to the standard Sobolev norm wHk\|w\|_{H^k}​. On a compact manifold this is standard:

  • The operator 1Δ1-\Delta is a positive, elliptic, self-adjoint pseudodifferential operator of order 2. Its (fractional) powers (1Δ)k/2(1-\Delta)^{k/2} are elliptic pseudo-differential operators of order k.
  • Elliptic regularity (or basic pseudodifferential calculus) implies that the graph norm of any positive elliptic operator of order k is equivalent to the standard Sobolev HkH^k-norm. Concretely, there exist constants c,C>0c,C>0 (dependent only on the geometry of SS and kk) such that for all wC(S)w\in C^\infty(S), c  (1Δ)k/2wL2wHkC  (wL2+(1Δ)k/2wL2).c\;\|(1-\Delta)^{k/2}w\|_{L^2} \le \|w\|_{H^k} \le C\;\big(\|w\|_{L^2}+\|(1-\Delta)^{k/2}w\|_{L^2}\big). On a compact manifold the two-sided inequality can be tightened (by adjusting constants) to show full equivalence (no extra low-order term needed if you use (1Δ)k/2(1-\Delta)^{k/2} itself as the defining operator).

If you prefer an elementary (but longer) route: cover the manifold with finitely many coordinate charts, use a partition of unity, and compare the local Euclidean Sobolev norms defined by derivatives up to order k with the global spectral norm. The finiteness of the cover and standard elliptic estimates give the equivalence constants.

4. Conclusion

Putting items (2) and (3) together yields the desired equivalence:

wHk2    j0(1+λj)kwj2,\|w\|_{H^k}^2 \;\simeq\; \sum_{j\ge0} (1+\lambda_j)^k\,|w_j|^2,

where the symbol \simeq means equality up to multiplicative constants depending only on kk and the Riemannian manifold SS.