3.3 Hessian
3.3.2 Existence of many functions in D ( Hess )
for every๐ โL2(๐โ2M). Now Lemma 3.2.72, Lemma 3.2.54 as well as the continuity of ฮฆallow us to replace the termsโ๐ ๐โ0
๐ ๐by arbitrary elements๐๐ ๐ โTest(M), still retaining the identity (3.3.4) with๐๐ ๐in place ofโ๐ ๐โ0
๐ ๐,๐โ {1, . . . , ๐}and ๐ โ {1, . . . , ๐}. In particular, by Definition 3.3.2, we deduce that ๐ โ D(Hess) and ๐ด0 =Hess ๐. By Proposition 3.2.19 again and (3.3.5), we obtain
kHess๐kL2( (๐โ)โ2M)=kฮฆkL2(๐โ2M)0 โค
โ ๐ถ , which is precisely what was left prove.
Remark 3.3.4. IfE2 is extended toL2(M) byE2(๐) := โ for ๐ โ L2(M) \F, it is unclear if the resulting functional isL2-lower semicontinuous. To bypass this issue in applications, one might instead use that by Theorem 3.3.3, the functional E2๐:L2(M) โ [0,โ]given by
E2๐(๐):=
๏ฃฑ๏ฃด
๏ฃด
๏ฃฒ
๏ฃด๏ฃด
๏ฃณ ๐
ห
M
|โ๐|2d๐ช+ ห
M
Hess๐
2
HSd๐ช if ๐ โD(Hess),
โ otherwise
isL2-lower semicontinuous for every๐ > 0.
IfMis, say, a compact Riemannian manifold without boundary, one can easily prove using the Bochner identity that the (nonpositive) generator associated withE2๐ [FOT11, Thm. 1.3.1] is the Paneitz-type operator
โฮ2+๐ฮ๐ +div(Ricโญโยท).
Remark 3.3.5. In general, the Hessian is not the trace of the Laplacian in the sense of (3.2.12). This already happens on weighted Riemannian manifolds without boundary:
of course, the associated Laplacian ฮ is defined by partial integration w.r.t. the reference measure [Gri09, Sec. 3.6], while the definition of Hessian only depends on the metric tensor. See also the second part of Example 3.2.48. Examples of abstract spaces for which thisisthe case โ and which currently enjoy high research interest [BNS20, DPG18, HZ20] โ arenoncollapsedRCD(๐พ , ๐)spaces,๐พโRand ๐โ [1,โ)[DPG18, Thm. 1.12]. See also Remark 3.3.16 below.
Remark 3.3.6. In line with Remark 3.3.5, although a priori๐ชplays a role in Definition 3.3.2, we expect the Hessian to only depend on conformal transformations ofhยท,ยทi, but not on drift transformations of๐ช. For instance, this is known on RCDโ(๐พ , ๐)spaces, ๐พ โRand๐โ [1,โ), see e.g. [Han19, Prop. 3.11] or [HS21, Lem. 2.16], and it does not seem hard to adapt the arguments from [Han19] to more general settings.
Remark 3.3.7. As an alternative to Definition 3.3.2, one can defineD(Hess)as the finiteness domain of the r.h.s. of the duality formula in (iv) in Theorem 3.3.11. The Hessian of ๐ โD(Hess)is then well-defined by the same duality arguments as in the
proof of Theorem 3.3.11.
[ER+20, Sav14, Stu18a]. The key technical part (not only for Theorem 3.3.11, but in fact for Theorem 3.6.9 below as well) is contained in Lemma 3.3.9, where โ loosely speaking and up to introducing the relevant objects later โ we show that
|โ๐ :๐|2 โค h
ฮ2 ๐ |๐|2 2
+
๐ ,( ยฎฮ๐โญ)โฏ
โ
(โ๐)asym
2 HS
i ๐
2
HS ๐ช-a.e.
for ๐ , ๐ โ Test(๐M). Of course, neither we introduced the covariant derivative โ, Definition 3.4.2 or the Hodge Laplacian ฮยฎ, Definition 3.5.21, yet, nor in general we have|๐|2 โ D(ฮ2 ๐ )for๐ โTest(๐M). Reminiscent of Proposition 3.2.75 and [ER+20, Cor. 6.3], we instead rephrase the above inequality in terms of measures, and the involved objectsโ๐ andฮยฎ๐โญtherein as the โr.h.s.โs of the identities one would expect forโ๐ andฮยฎ๐โญfor ๐ โTest(M)โ, rigorously proven in Theorem 3.4.3 and Lemma 3.6.1 below. In particular, by optimization over๐ โTest(๐M),
(โ๐)sym
2
HS โคฮ2 ๐ |๐|2 2
+
๐ ,( ยฎฮ๐โญ)โฏ
โ
(โ๐)asym
2
HS ๐ช-a.e.,
which is the Bochner inequality for vector fields according to (3.2.9). For๐ :=โ๐, ๐ โTest(M), this essentially provides Theorem 3.3.11. Details about this inequality for general๐ โReg(๐M), leading to Theorem 3.6.9, are due to Lemma 3.6.2.
We start with a technical preparation. Given๐, ๐โ๐f+(M), we define the Borel measureโ
๐ ๐โ๐f+(M)as follows. Let๐โ๐+f(M)with๐๐and๐๐be arbitrary, denote the respective densities w.r.t.๐by ๐ , ๐โL1(M, ๐), and set
โ
๐ ๐:=p ๐ ๐ ๐.
For instance, one can choose๐:=|๐| + |๐|[Hal50, Thm. 30.A] โ in fact, the previous definition is independent of the choice of๐, whenceโ
๐ ๐is well-defined.
The following important measure theoretic lemma is due to [Gig18, Lem. 3.3.6].
Lemma 3.3.8. Let๐1, ๐2, ๐3โ๐ยฑf(M)satisfy the inequality ๐2๐1+2๐ ๐2+๐3โฅ0 for every๐โR. Then the following properties hold.
(i) The elements๐1and๐3are nonnegative, and
|๐2| โคโ ๐1๐3. (ii) We have๐2 ๐1,๐2๐3and
k๐2kTVโคp
k๐1kTVk๐3kTV.
(iii) The๐ช-singular parts(๐1)โฅand(๐3)โฅof๐1and๐3are nonnegative. Moreover, expressing the densities of the๐ช-absolutely continuous parts of ๐๐ by ๐๐ := d(๐๐)/d๐ชโL1(M),๐โ {1, 2, 3}, we have
|๐2|2โค๐1๐3 ๐ช-a.e.
In the subsequent lemma, all terms where๐0is infinite are interpreted as being zero. Similar proofs can be found in [Bra20, Gig18, Han18a].
Lemma 3.3.9. Let ๐0 โ [๐ ,โ], ๐, ๐ โ N, ๐ , ๐ โ Test(M)๐ and โ โ Test(M)๐. Define๐1[๐ , ๐] โ๐fยฑ(M)as
๐1[๐ , ๐]:=
๐
X
๐ ,๐0=1
e๐๐
e๐๐0๐ช2 ๐ 2 (๐๐, ๐๐0) +2
๐
X
๐ ,๐0=1
๐๐H[๐๐] (๐๐0, ๐๐0)๐ช + 1
2
๐
X
๐ ,๐0=1
hโ๐๐,โ๐๐0i hโ๐๐,โ๐๐0i + hโ๐๐,โ๐๐0i hโ๐๐,โ๐๐0i ๐ช
โ 1 ๐0
hX๐
๐=1
๐๐ฮ๐๐+ hโ๐๐,โ๐๐ii2 ๐ช
As in Lemma 3.3.8, we denote the density of the ๐ช-absolutely continuous part of ๐1[๐ , ๐]by๐1[๐ , ๐]:=d๐1[๐ , ๐]/d๐ชโL1(M). Then the๐ช-singular part๐1[๐ , ๐]โฅ
of๐1[๐ , ๐]as well as๐1[๐ , ๐]are nonnegative, and hX๐
๐=1 ๐
X
๐=1
hโ๐๐,โโ๐i hโ๐๐,โโ๐i +๐๐H[๐๐] (โ๐, โ๐)
โ 1 ๐0
๐
X
๐=1 ๐
X
๐=1
๐๐ฮ๐๐+ hโ๐๐,โ๐๐i
|โโ๐|2i2
โค๐1[๐ , ๐]hX๐
๐ , ๐0=1
hโโ๐,โโ๐0i2โ 1 ๐0
hX๐
๐=1
|โโ๐|2i2i
๐ช-a.e.
Proof. We define๐2[๐ , ๐, โ], ๐3[โ] โ๐fยฑ(M)by ๐2[๐ , ๐, โ]:=
๐
X
๐=1 ๐
X
๐=1
hโ๐๐,โโ๐i hโ๐๐,โโ๐i +๐๐H[๐๐] (โ๐, โ๐) ๐ช
โ 1 ๐0
๐
X
๐=1 ๐
X
๐=1
๐๐ฮ๐๐+ hโ๐๐,โ๐๐i
|โโ๐|2๐ช,
๐3[โ]:= hX๐
๐ , ๐0=1
hโโ๐,โโ๐0i2โ 1 ๐0
hX๐
๐=1
|โโ๐|2i2i ๐ช.
Both claims readily follow from Lemma 3.3.8 as soon as๐2๐1[๐ , ๐] +2๐ ๐2[๐ , ๐, โ] + ๐3[โ] โฅ0for every๐โR, which is what we concentrate on in the sequel.
Let๐ โ R and pick๐, ๐ โ R๐ as well as ๐ โ R๐. Define the function๐ โ Cโ(R2๐+๐)through
๐(๐ฅ , ๐ฆ, ๐ง):=
๐
X
๐=1
๐ ๐ฅ๐๐ฆ๐+๐๐๐ฅ๐โ๐๐๐ฆ๐ +
๐
X
๐=1
(๐ง๐โ๐๐)2โ๐2
๐
.
For every๐ โ {1, . . . , ๐} and every ๐ โ {1, . . . , ๐}, those first and second partial derivatives of๐which, do not always vanish identically read
๐๐(๐ฅ , ๐ฆ, ๐ง)=๐ ๐ฆ๐+๐๐, ๐๐+๐(๐ฅ , ๐ฆ, ๐ง)=๐ ๐ฅ๐โ๐๐, ๐2๐+๐(๐ฅ , ๐ฆ, ๐ง)=2(๐ง๐โ๐๐), ๐๐ , ๐+๐(๐ฅ , ๐ฆ, ๐ง)=๐,
๐๐+๐ ,๐(๐ฅ , ๐ฆ, ๐ง)=๐,
๐2๐+๐ ,2๐+๐(๐ฅ , ๐ฆ, ๐ง)=2.
For convenience, we write
A2 ๐ (๐, ๐, ๐, ๐):=A2 ๐ [๐โฆ๐], B(๐, ๐, ๐, ๐):=B[๐โฆ๐], C(๐, ๐, ๐, ๐):=C[๐โฆ๐], D(๐, ๐, ๐, ๐):=D[๐โฆ๐],
where the respective r.h.s.โs are defined as in Lemma 3.2.76 for๐ผ := 2๐+๐ and ๐ :=(๐ , ๐, โ). Using the same Lemma 3.2.76, we compute
A2 ๐ (๐, ๐, ๐, ๐)=
๐
X
๐ ,๐0=1
(๐
e๐๐+๐๐) (๐
e๐๐0+๐๐0)๐ช2 ๐ 2 (๐๐, ๐๐0) +other terms, B(๐, ๐, ๐, ๐)=4
๐
X
๐ ,๐0=1
(๐ ๐๐+๐๐)๐H[๐๐] (๐๐0, ๐๐0) +4
๐
X
๐=1 ๐
X
๐=1
(๐ ๐๐+๐๐)H[๐๐] (โ๐, โ๐) +other terms, C(๐, ๐, ๐, ๐)=2
๐
X
๐ ,๐0=1
๐2
hโ๐๐,โ๐๐0i hโ๐๐,โ๐๐0i + hโ๐๐,โ๐๐0i hโ๐๐,โ๐๐0i +8
๐
X
๐=1 ๐
X
๐=1
๐hโ๐๐,โโ๐i hโ๐๐,โโ๐i +4
๐
X
๐ , ๐0=1
hโโ๐,โโ๐0i2+other terms, D(๐, ๐, ๐, ๐)=
๐
X
๐ ,๐0=1
(๐ ๐๐+๐๐) (๐ ๐๐0+๐๐)ฮ๐๐ฮ๐๐0
+4
๐
X
๐ ,๐0=1
๐(๐ ๐๐+๐๐)ฮ๐๐hโ๐๐0,โ๐๐0i +4
๐
X
๐ ,๐0=1
๐2hโ๐๐,โ๐๐i hโ๐๐0,โ๐๐0i +4
๐
X
๐=1 ๐
X
๐=1
(๐ ๐๐+๐๐)ฮ๐๐|โโ๐|2 +8
๐
X
๐=1 ๐
X
๐=1
๐hโ๐๐,โ๐๐i |โโ๐|2 +4
hX๐
๐=1
|โโ๐|2i2
+other terms.
Here, every โother termโ contains at least one factor of the form๐e๐๐โ๐๐oreโ๐โ๐๐ for some๐โ {1, . . . , ๐}and ๐ โ {1, . . . , ๐}.
By Lemma 3.2.76 and Proposition 3.2.75 with the nonnegativity of D(๐, ๐, ๐, ๐)as well as the trivial inequality1/๐ โฅ1/๐0,
A2 ๐ (๐, ๐, ๐, ๐) +h
B(๐, ๐, ๐, ๐) +C(๐, ๐, ๐, ๐) โ 1
๐0D(๐, ๐, ๐, ๐)i ๐ช โฅ0.
By the arbitrariness of๐, ๐ โR๐ and๐โ R๐, for every Borel partition (๐ธ๐)๐โNof M, every Borel set๐น โMand all sequences(๐๐)๐โNand(๐๐)๐โNinR๐as well as (๐๐)๐โNinR๐,
1๐นX
๐โN
1๐ธ
๐
h
A2 ๐ (๐, ๐๐, ๐๐, ๐๐) +h
B(๐, ๐๐, ๐๐, ๐๐) +C(๐, ๐๐, ๐๐, ๐๐)
โ 1
๐0D(๐, ๐๐, ๐๐, ๐๐)i ๐ชi
โฅ0.
(3.3.6)
We now choose the involved quantities appropriately. Let(๐น๐)๐โNbe anE-nest with the property that the restrictions of e๐,e๐ andeโ to๐น๐ are continuous for every ๐ โ N, and set๐น:=S
๐โN๐น๐. Since๐นc is anE-polar set and thus not seen by๐ช and๐ช2 ๐ 2 (๐๐, ๐๐0),๐, ๐0โ {1, . . . , ๐}, its contribution to the subsequent manipulations is ignored. For๐ โNwe now take a Borel partition(๐ธ๐
๐)๐โNofMand sequences(๐๐
๐)๐โN
and(๐๐
๐)๐โNinR๐as well as(๐๐
๐)๐โNinR๐with sup
๐ ,๐โN
|๐๐
๐| + |๐๐
๐| + |๐๐
๐|
<โ in such a way that
๐โโlim X
๐โN
1๐ธ๐ ๐
๐๐
๐ =๐ e๐,
๐โโlim X
๐โN
1๐ธ๐ ๐
๐๐
๐ =๐e๐ ,
๐โโlim X
๐โN
1๐ธ๐ ๐
๐๐
๐ =eโ
pointwise on๐น. Thus, the l.h.s. of (3.3.6) with(๐ธ๐)๐โN,(๐๐)๐โN,(๐๐)๐โNand(๐๐)๐โN
replaced by(๐ธ๐
๐)๐โN,(๐๐
๐)๐โN,(๐๐
๐)๐โNand(๐๐
๐)๐โN,๐ โN, respectively, converges w.r.t.k ยท kTVas๐ โ โ. In fact, in the limit as๐โ โevery โother termโ above becomes zero, and the prefactors๐
e๐๐+ (๐๐
๐)๐ become2๐
e๐๐,๐โ {1, . . . , ๐}. We finally obtain 4๐2
๐
X
๐ ,๐0=1
e๐๐
e๐๐0๐ช2 ๐ 2 (๐๐, ๐๐0) +8๐2
๐
X
๐ ,๐0=1
๐๐H[๐๐] (๐๐0, ๐๐0)๐ช+8๐
๐
X
๐=1 ๐
X
๐=1
๐๐H[๐๐] (โ๐, โ๐)๐ช +2๐2
๐
X
๐ ,๐0=1
hโ๐๐,โ๐๐0i hโ๐๐,โ๐๐0i + hโ๐๐,โ๐๐0i hโ๐๐,โ๐๐0i ๐ช +8๐
๐
X
๐=1 ๐
X
๐=1
hโ๐๐,โโ๐i hโ๐๐,โโ๐i๐ช +4
๐
X
๐ , ๐0=1
hโโ๐,โโ๐0i2๐ช
โ4๐2 ๐0
hX๐
๐=1
๐๐ฮ๐๐+ hโ๐๐,โ๐๐ii2 ๐ช
โ 8๐ ๐0
๐
X
๐=1 ๐
X
๐=1
๐๐ฮ๐๐+ hโ๐๐,โ๐๐i
|โโ๐|2๐ช
โ 4 ๐0
hX๐
๐=1
|โโ๐|2i2
โฅ0.
Dividing by4and sorting terms by the order of๐yields the claim.
We note the following consequence of Lemma 3.3.9 that is used in Theorem 3.3.11 below as well, but becomes especially important in Subsection 3.3.3.
Remark 3.3.10. The nonnegativity of ๐3[โ] from Lemma 3.3.9 can be translated into the following trace inequality, compare with [Bra20, Rem. 2.19] and the proof of [Han18a, Prop. 3.2]. With the pointwise trace defined as in (3.2.12), we have
๐
X
๐=1
โโ๐โ โโ๐
2 HS โฅ 1
๐trhX๐
๐=1
โโ๐โ โโ๐ i2
๐ช-a.e.
for every๐โNand everyโโTest(M)๐.
Theorem 3.3.11. Every ๐ โTest(M)belongs toD(Hess)and satisfies
Hess๐(โ๐1,โ๐2)=H[๐] (๐1, ๐2) ๐ช-a.e. (3.3.7) for every๐1, ๐2 โTest(M). Moreover, denoting by๐พ2 ๐
2 (๐) โL1(M)the density of the ๐ช-absolutely continuous part of๐ช2 ๐ 2 (๐), we have
Hess๐
2 HS โค๐พ2 ๐
2 (๐) ๐ช-a.e. (3.3.8) Proof. Recall that indeed๐พ2 ๐
2 (๐) โL1(M)by Proposition 3.2.79. Let๐, โ1, . . . , โ๐โ Test(M),๐โN. Applying Lemma 3.3.9 for๐0:=โand๐:=1then entails
hX๐
๐=1
hโ๐ ,โโ๐i hโ๐,โโ๐i +๐H[๐] (โ๐, โ๐)i2
โค h ๐2๐พ2 ๐
2 (๐) +2๐H[๐] (๐ , ๐) + 1 2
|โ๐|2|โ๐|2+ 1 2
hโ๐ ,โ๐i2i
ร
๐
X
๐ , ๐0=1
hโโ๐,โโ๐0i2
= h
๐2๐พ2 ๐
2 (๐) +๐
โ|โ๐|2,โ๐ + 1
2
|โ๐|2|โ๐|2+ 1 2
hโ๐ ,โ๐i2i
ร
๐
X
๐ , ๐0=1
hโโ๐,โโ๐0i2 ๐ช-a.e.
In the last identity, we used the definition (3.2.22) of H[๐] (๐ , ๐). Using the first part of Lemma 3.2.73 and possibly passing to subsequences, this๐ช-a.e. inequality extends to all๐ โ FโฉLโ(M). Thus, successively setting๐ :=๐๐,๐ โ N, where (๐๐)๐โNis the sequence provided by Lemma 3.2.6, together with the locality ofโfrom Proposition 3.2.37, and by the definition (3.2.7) of the pointwise HilbertโSchmidt norm ofL2(๐โ2M), we obtain
๐
X
๐=1
H[๐] (โ๐, โ๐) โค๐พ2 ๐
2 (๐)1/2
๐
X
๐=1
โโ๐ โ โโ๐
HS ๐ช-a.e. (3.3.9) This implies pointwise๐ช-a.e. off-diagonal estimates as follows. Given any๐0โNand โ๐, โ0
๐ โTest(M), ๐ โ {1, . . . , ๐0}, since
๐0
X
๐=1
H[๐] (โ๐, โ0
๐)= 1 2
๐0
X
๐=1
H[๐] (โ๐+โ0
๐, โ๐+โ0
๐) โH[๐] (โ๐, โ๐) โH[๐] (โ0
๐, โ0
๐)
holds๐ช-a.e., applying (3.3.9), using that 1
2
๐0
X
๐=1
โ(โ๐+โ0๐) โ โ(โ๐+โ0๐) โ โโ๐โ โโ๐โ โโ0๐ โ โโ0๐
= 1 2
๐0
X
๐=1
โโ๐ โ โโ0๐ + โโ0๐โ โโ๐
= h๐
0
X
๐=1
โโ๐โ โโ0๐ i
sym
and finally employing that|๐sym|HSโค |๐|HSfor every๐ โL2(๐โ2M), we get
๐0
X
๐=1
H[๐] (โ๐, โ0
๐) โค๐พ2 ๐
2 (๐)1/2 1 2
๐0
X
๐=1
โโ๐โ โโ0
๐+ โโ0
๐ โ โโ๐ HS
โค๐พ2 ๐
2 (๐)1/2
๐0
X
๐=1
โโ๐ โ โโ0
๐
HS ๐ช-a.e.
We replaceโ๐by๐๐โ๐,๐ โ {1, . . . , ๐0}, for arbitrary๐1, . . . , ๐๐0 โQ. This gives
๐0
X
๐=1
๐๐H[๐] (โ๐, โ0
๐) โค๐พ2 ๐
2 (๐)1/2
๐0
X
๐=1
๐๐โโ๐ โ โโ0
๐
HS ๐ช-a.e. (3.3.10) In fact, sinceQis countable, we find an๐ช-negligible Borel set๐ต โM on whose complement (3.3.10) holds pointwise for every๐1, . . . , ๐๐0 โQ. Since both sides of (3.3.10) are continuous in๐1, . . . , ๐๐0, by density ofQinRwe deduce that (3.3.10) holds pointwise on๐ตcfor every๐1, . . . , ๐๐0โR. Therefore, given any๐1, . . . , ๐๐0โTest(M), up to possibly removing a further๐ช-negligible Borel set๐ถโM, for every๐ฅโ (๐ตโช๐ถ)c we may replace๐๐ by๐๐(๐ฅ),๐ โ {1, . . . , ๐0}, in (3.3.10). This leads to
๐0
X
๐=1
๐๐H[๐] (โ๐, โ0๐)
2
โค๐พ2 ๐
2 (๐)1/2
๐0
X
๐=1
๐๐โโ๐ โ โโ0๐
HS ๐ช-a.e. (3.3.11) We now define the operatorฮฆ: Test(๐โ2M) โL0(M)by
ฮฆ
๐0
X
๐=1
๐๐๐0๐โโ๐โ โโ0๐ :=
๐0
X
๐=1
๐๐๐0๐H[๐] (โ๐, โ0๐). (3.3.12) From (3.3.11) and the algebra property of Test(M), it follows thatฮฆis well-defined, i.e. the value ofฮฆ(๐) does not depend on the specific way of representing a given element ๐ โ Test(๐โ2M). Moreover, the map ฮฆ is clearly linear, and for every ๐ โTest(M)and every๐ โTest(๐โ2M),
ฮฆ(๐ ๐)=๐ฮฆ(๐). (3.3.13)
Since the๐ช-singular part๐ช2 ๐ 2 (๐)โฅof๐ช2 ๐ 2 (๐)is nonnegative, by (3.2.23) we get ห
M
๐พ2 ๐
2 (๐)d๐ช โค๐ช2 ๐ 2 (๐) [M]= ห
M
(ฮ๐)2d๐ชโ ๐
|โ๐|2
. (3.3.14)
After integrating (3.3.11) and employing CauchyโSchwarzโs inequality, kฮฆ(๐) kL1(M) โคhห
M
(ฮ๐)2d๐ชโ ๐
|โ๐|2i1/2
k๐kL2(๐โ2M)
holds for every๐ โTest(๐โ2M). Thus, by density of Test(๐โ2M)inL2(๐โ2M)and (3.3.13),ฮฆ uniquely extends to a (non-relabeled) continuous,Lโ-linear map from L2(๐โ2M)intoL1(M), whenceฮฆโL2( (๐โ)โ2M)by definition of the latter space.
To check that๐ โD(Hess)andฮฆ =Hess๐, first note that by the continuity ofฮฆand Lemma 3.2.72, we can replace๐๐๐0
๐ by arbitrary๐๐ โTest(M),๐ โ {1, . . . , ๐0}, still retaining the identity (3.3.12). Therefore, slightly changing the notation in (3.3.12), let ๐1, ๐2, โ โTest(M)and use (3.3.13), the definition (3.2.22) of H[๐]and Lemma 3.2.54 to derive that
2 ห
M
โฮฆ(โ๐1,โ๐2)d๐ช
= ห
M
โ
โ๐1,โhโ๐ ,โ๐2id๐ช+ ห
M
โ
โ๐1,โhโ๐ ,โ๐2id๐ช
โ ห
M
โ
โ๐1,โhโ๐ ,โ๐2i d๐ช
=โ ห
M
hโ๐ ,โ๐2idiv(โโ๐1)d๐ชโ ห
M
hโ๐ ,โ๐2idiv(โโ๐1)d๐ช
โ ห
M
โ
โ๐1,โhโ๐ ,โ๐2id๐ช, which is the desired assertion ๐ โD(Hess)andฮฆ =Hess๐.
The same argument gives (3.3.7), while the inequality (3.3.8) is due to (3.3.11), the density of Test(๐โ2M)inL2(๐โ2M)as well as the definition (3.2.7) of the pointwise HilbertโSchmidt norm.
Theorem 3.3.11 implies the following qualitative result. A quantitative version of it, as directly deduced in [Gig18, Cor. 3.3.9] from [Gig18, Thm. 3.3.8], is however not yet available only with the information collected so far. See Remark 3.3.13 below.
Corollary 3.3.12. Every ๐ โ D(ฮ)belongs to the closure ofTest(M) inD(Hess), and in particular toD(Hess). More precisely, let๐0 โ (0, 1) and๐ผ0 โ Rbe as in Lemma 3.2.60 for๐:=๐ โ. Then for every ๐ โD(ฮ), we have ๐ โD(Hess)with
ห
M
Hess๐
2
HSd๐ชโค 1 1โ๐0
ห
M
(ฮ๐)2d๐ช+ ๐ผ0 1โ๐0
ห
M
|โ๐|2d๐ช.
Proof. Since (M,E,๐ช) satisfies BE1(๐ ,โ)by [ER+20, Thm. 6.9], it also trivially obeys BE1(โ๐ โ,โ), see also [ER+20, Prop. 6.7]. As in the proof of Proposition 3.2.79,
Eโ๐ โ|โ๐|
โค ห
M
(ฮ๐)2d๐ช holds for every ๐ โTest(M). Hence, using (3.2.19) we estimate
๐ โ |โ๐|2
โค๐0E |โ๐| +๐ผ0
ห
M
|โ๐|2d๐ช
=๐0Eโ๐ โ |โ๐| +๐0
๐ โ |โ๐|2
+๐ผ0 ห
M
|โ๐|2d๐ช
โค๐0 ห
M
(ฮ๐)2d๐ช+๐0 ๐ โ
|โ๐|2 +๐ผ0
ห
M
|โ๐|2d๐ช.
The claim for ๐ โTest(M)now follows easily. We already know from Theorem 3.3.11 that ๐ โD(Hess). Integrating (3.3.8) and using (3.3.14) thus yields
ห
M
Hess๐
2 HSd๐ชโค
ห
M
(ฮ๐)2d๐ชโ ๐
|โ๐|2
(3.3.15)
โค ห
M
(ฮ๐)2d๐ช+ ๐ โ
|โ๐|2
โค 1
1โ๐0 ห
M
(ฮ๐)2d๐ช+ ๐ผ0 1โ๐0
ห
M
|โ๐|2d๐ช.
Finally, given ๐ โD(ฮ), let ๐๐ :=max{min{๐ , ๐},โ๐} โL2(M) โฉLโ(M),๐โN. Note thatP๐ก๐๐ โTest(M)for every๐ก > 0and every๐โN, and thatP๐ก๐๐ โP๐ก๐ inF as well as, thanks to Lemma 3.2.12,ฮP๐ก๐๐โฮP๐ก๐ inL2(M)as๐โ โ. Moreover P๐ก๐ โ ๐inFas well asฮP๐ก๐ =P๐กฮ๐ โฮ๐ inL2(M)as๐กโ0. These observations imply that ๐ belongs to the closure of Test(M)inD(Hess), whence ๐ โD(Hess)by Theorem 3.3.3, and the claimed inequality, with unchanged constants, is clearly stable under this approximation procedure.
Remark 3.3.13. The subtle reason why we still cannot deduce (3.3.15) for general ๐ โ D(ฮ) is that we neither know whether the r.h.s. of (3.3.15) makes sense โ which essentially requires|โ๐| โ Fโ nor, in the notation of the previous proof, whether
๐
|โP๐ก๐๐|2
โ
๐
|โ๐|2 as๐ โ โ and๐ก โ 0. (Neither we know if E |โP๐ก๐๐|
โE |โ๐| as๐ โ โand๐ก โ 0.) Both points are trivial in the more restrictive RCD(๐พ ,โ)case from [Gig18, Cor. 3.3.9],๐พ โ R. In our setting, solely Lemma 3.2.60 does not seem sufficient to argue similarly. Instead, both points will follow from Lemma 3.4.13 and Lemma 3.6.2, see Corollary 3.4.14 and Corollary 3.6.3.