Spectral norm

1. The H1H^1 calculation (concrete)

Let {φj}j≥0\{\varphi_j\}_{j\ge0}​ be an orthonormal L2L^2-basis of eigenfunctions of −Δ-\Delta with eigenvalues λj≥0\lambda_j\ge0:

−Δφj=λjφj,⟨φj,φℓ⟩L2=δjℓ.-\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

∥(1−Δ)1/2w∥L22=∑j(1+λj) ∣wj∣2,\|(1-\Delta)^{1/2}w\|_{L^2}^2 =\sum_j \big(1+\lambda_j\big)\,|w_j|^2,

because (1−Δ)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 (1−Δ)1/2(1-\Delta)^{1/2}.

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

c  ∥(1−Δ)1/2w∥L2≤∥w∥H1≤C  ∥(1−Δ)1/2w∥L2.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

∥w∥H12≃∑j(1+λj) ∣wj∣2,\|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 ∥w∥H12\|w\|_{H^1}^2​ is ∥w∥L22+∥∇w∥L22\|w\|_{L^2}^2+\|\nabla w\|_{L^2}^2​. Using the eigen-expansion one checks ∥∇w∥L22=∑jλj∣wj∣2\|\nabla w\|_{L^2}^2=\sum_j\lambda_j|w_j|^2. Hence ∥w∥H12=∑j(1+λj)∣wj∣2\|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 k≥0k\ge0. By functional calculus

(1−Δ)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,

∥(1−Δ)k/2w∥L22=∑j(1+λj)k∣wj∣2.\|(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 (1−Δ)k/2w(1-\Delta)^{k/2}w.

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

What remains is to explain why the graph norm ∥(1−Δ)k/2w∥L2\|(1-\Delta)^{k/2}w\|_{L^2}​ is equivalent to the standard Sobolev norm ∥w∥Hk\|w\|_{H^k}​. On a compact manifold this is standard:

  • The operator 1−Δ1-\Delta is a positive, elliptic, self-adjoint pseudodifferential operator of order 2. Its (fractional) powers (1−Δ)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 w∈C∞(S)w\in C^\infty(S), c  ∥(1−Δ)k/2w∥L2≤∥w∥Hk≤C  (∥w∥L2+∥(1−Δ)k/2w∥L2).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 (1−Δ)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:

∥w∥Hk2  ≃  ∑j≥0(1+λj)k ∣wj∣2,\|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.