On the Hermitian Geometry of k-Gauduchon Orthogonal Complex Structures
Dissertation
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University
By
Gabriel Khan, B.A., M.S.
Graduate Program in Mathematics
The Ohio State University
2018
Dissertation Committee:
Fangyang Zheng, Advisor Bo Guan King-Yeung Lam John Lafont c Copyright by
Gabriel Khan
2018 Abstract
This work deals with various phenomena relating to complex geometry. We are
particularly interested in non-K¨ahlerHermitian manifolds, and most of the work here
was done to try to understand the geometry of these spaces by understanding the
torsion.
Chapter 1 introduces some background material as well as various equations and
inequalities on Hermitian manifolds. We are focused primarily on the inequalities
that are useful for the analysis that we do later in the thesis. In particular, we focus
on k-Gauduchon complex structures, which were initially defined by Fu, Wang, and
Wu [8].
Chapter 2 discusses the spectral geometry of Hermitian manifolds. In particular,
we estimate the real eigenvalues of ∂∂¯ from below. In doing so, we prove a theorem on
non-self-adjoint drift Laplace operators with L∞ drift. This result is of independent interest, apart from its application to complex geometry. The work in this section is largely based on the Li-Yau estimate [24] as well as an ansatz due to Hamel,
Nadirashvili and Russ [12].
Chapter 3 considers orthogonal complex structures to a given Riemannian metric.
Much of the work in this section is conjectural in nature, but we believe that this
ii is a promising approach to studying Hermitian geometry. We do prove several con- crete results as well. In particular, we show how the moduli space of k-Gauduchon orthogonal complex structures is pre-compact.
iii For Kori
iv Acknowledgments
I am very grateful to Fangyang Zheng for his guidance and for being an excellent advisor. I feel very fortunate to have found an advisor who was willing to help and invest in me. I’m also deeply appreciative that he is always willing to listen and allow me the room to also work on my own projects.
Thanks to Adrian Lam, for his help with the partial differential equations in
Chapter 2. His insights were essential for this project. Further thanks to Bo Guan and John Lafont for their support and for serving on my committee. On a separate note, I would like to thank Jun Zhang for he reaching out and providing me the opportunity to work on his project next year. For this I am truly grateful. I would also like to thank Glen Richard Hall, who was my advisor as an undergraduate. He was very helpful and a great source of problems to work on.
The time spent writing this would have been significantly less enjoyable and more difficult without the help of others. I would like to thank all my friends from Columbus and elsewhere who have supported me through the past six years. I’d also like to thank the creators of the OSU Dissertation LATEX class. This thesis would have been significantly more difficult to write without it.
I owe a lot to my parents, Mizan and Therese. They have been tremendously helpful throughout my life, both in mathematics and otherwise. I would also like to thank Nathan, Nicholas, and Isabel, for being wonderful siblings. Many thanks are
v due to Karen and Micky, for all their help with keeping our lives in order the past few years.
Finally, and most importantly, I would like to thank Kori for her love and strength.
The last few months have been crazy, and I could not have done any of this without her. She is my editor, organizer, and life partner.
vi Vita
July 17, 1990 ...... Born -Washington D.C., USA
2012 ...... B.S. in Mathematics Boston University 2014 ...... M.S. Mathematics The Ohio State University 2012-present ...... Ph.D. Student The Ohio State University
Publications
Research Publications
G. Khan “Hall’s Conjecture on extremal sets for random triangles”. Preprint. (2018)
G. Khan, B. Yang, and F. Zheng. “The Set of All Orthogonal Complex Structures on the Flat 6-Tori”. Adv. Math, 319, 451-471, 2017
G. Khan “The Drift Laplacian and Hermitian Geometry”. AGAG, 53(2), 233-249, 2017
T. Hart, G. Khan, and M. Khan, “Revisiting Toom’s Proof of Bulgarian Solitaire”. Annales des Sciences Math´ematiquesdu Qu´ebec 36(2). (December 2011)
Fields of Study
Major Field: Mathematics
vii Studies in: Complex Geometry Geometric Analysis
viii Table of Contents
Page
Abstract ...... ii
Dedication ...... iv
Acknowledgments ...... v
Vita ...... vii
1. Introduction ...... 1
1.1 Almost complex structures ...... 1 1.2 Complex Structures and the Newlander Nirenberg theorem . . . . . 2 1.3 The Levi-Civita and Chern connections ...... 3 1.3.1 Curvature and torsion ...... 4 1.3.2 Metric Connections and Complex Connections ...... 4 1.4 A crucial lemma ...... 6 1.5 Torsion bounds ...... 7 1.6 k-Gauduchon complex structures ...... 10 1.6.1 Weak k-Gauduchon structures ...... 11 1.7 The W 1,2 estimate on conformal changes ...... 12 1.8 Monotonicity results ...... 13
c 2. The eigenvalues of ...... 16 2.1 The drift Laplacian ...... 16 2.1.1 Previous work ...... 17 2.2 Drift Laplacians with L∞ drift ...... 19 2.3 A general outline of the proof ...... 20 2.4 Step 1: Finding the drift that minimizes the principle eigenvalue . . 21 2.5 Step 2: The a priori Cα estimate and uniform radii estimates . . . 23 2.5.1 Step 2a: Variation estimates ...... 24 2.5.2 Step 2b: Regularity away from the zero locus of ∇u . . . . . 25
ix 2.6 Step 3: The Li-Yau Estimate ...... 25 2.6.1 An estimate using the maximum principle ...... 28 2.6.2 Using the inequality ...... 37 2.7 Step 4: A lower bound on λ ...... 40 2.8 Future work ...... 41 2.9 A concrete example ...... 42
3. Moduli of Orthogonal complex structures ...... 44
3.1 Two dimensional metrics and Teichmuller theory ...... 45 3.2 Higher dimensions ...... 46 3.2.1 Generating orthogonal metrics and canonical metrics . . . . 46 3.3 Flat six dimensional metrics ...... 47 3.3.1 More general phenomena ...... 49 3.4 Obstructions ...... 50 3.5 K¨ahlerand k-Gauduchon complex structures ...... 51 3.6 Stability under deformations ...... 53
Bibliography ...... 55
x Chapter 1: Introduction
This section is dedicated to the basics of complex geometry, with a view towards the theorems that we will prove later. In particular, we are studying how the various curvature tensors and torsion tensors on a complex manifold are related. From this viewpoint, we can use theorems from analysis and Riemannian geometry to under- stand complex geometry.
For this thesis, we will consider a compact Riemannian manifold (M n, g) which admits an integrable complex structure J. We are largely interested in results that can be proven about the Hermitian manifold (M n, g, J) but are independent of the par- ticular choice of J. The results that hold in this direction prove that the Riemannian geometry control various quantities about the Hermitian geometry. Alternatively, we can think of these as being results about the moduli space of orthogonal complex structures. However, before introducing these notions, we will first introduce some of the background.
1.1 Almost complex structures
Complex numbers can be defined by adjoining the real numbers by an element i, which satisfies i2 = −1. Complex geometry generalizes this idea with the notion of an almost complex structure. Given a Riemannian manifold M, an almost complex
1 structure is a smoothly varying endomorphism of TM (denoted J) which satisfies
J 2 = −Id. Intuitively, J corresponds to multiplication by i. We say that an almost complex structure is compatible with, or orthogonal to a metric, if it satisfies g(X,Y ) = g(JX,JY ) for all tangent vectors X and Y . Intuitively, this corresponds to the fact that multiplication by i does not change the norm of a number. In this case, the triple (M, g, J) is called an almost-Hermitian manifold.
1.2 Complex Structures and the Newlander Nirenberg the- orem
In order for a manifold to be genuinely complex, J must satisfy a further condition.
There needs to be be a local biholomorphism between our manifold and Cn. In other words, we want there to be local holomorphic coordinates. Given an arbitrary almost complex structure, it may not be possible to do this. A straightforward computation shows that if J is induced from complex coordinates, then Nijinhuis tensor must
2 vanish. This is a tensor defined by NJ (X,Y ) = −J [X,Y ] + J[X,JY ] + J[JX,Y ] −
[JX,JY ], and it vanishes for any complex structure.
Due to the Newlander-Nirenberg theorem, an almost complex structure is inte- grable if and only if its Nijinhuis tensor vanishes [21]. This theorem was proven for analytic almost complex structures in 1957. In this work, we are interested in complex structures with less regularity, so extensions of the Newlander-Nirenberg theorem are important for our work. The theorem was extended to the C1,α complex structures in [19]. However, it is possible to prove the Newlander-Nirenberg theorem in even weaker regularity. The result that we make use of here was originally proved in [13], which shows the Nijinhuis tensor is well defined and that the Newlander-Nirenberg theorem holds for complex structures that are Cα for α > 1/2.
2 We focus on complex manifolds, so we will only consider integrable complex struc- tures. This is a non-trivial restriction, many familiar manifolds do not admit orthog- onal complex structures. For instance, the only sphere admitting a complex structure orthogonal to its round metric is S2. It is an open question whether any S6 admits a complex structure, but there are no complex structure which is close to being or- thogonal to the round metric [23]. Furthermore, no sphere other than S2 or S6 admit almost complex structures.
One particular class of complex structures that have been studied in depth are
K¨ahlercomplex structures. We say a complex structure is K¨ahlerif its K¨ahlerform
ω = g(J·, ·) is closed. We are primarily interested in the non-K¨ahlercase, as many of our results are already known or vacuous in the K¨ahler setting.
1.3 The Levi-Civita and Chern connections
Given a smooth manifold, there is canonical way to differentiate functions, but there is no way to differentiate vector fields without introducing a connection. A general affine connection on a manifold is a bilinear map ∇ satisfying the following conditions:
∇ : C∞(M, TM) × C∞(M, TM) → C∞(M, TM)
(X,Y ) 7→ ∇X Y
such that for all smooth functions f ∈ C∞M and all vector fields X,Y on M:
∞ • ∇fX Y = f∇X Y (∇ is C (M)-linear in the first variable);
• ∇X fY = df(X)Y + f∇X Y (∇ satisfies the product rule).
3 1.3.1 Curvature and torsion
Given an arbitrary connection ∇, it is possible to define the torsion and curvature.
If we let X and Y be smooth vector fields on our manifold, the torsion tensor τ is defined by
∇ τ (X,Y ) = ∇X Y − ∇Y X − [X,Y ].
Note that τ takes in two vector fields and produces a third. It is tensorial in X and Y , and measures how much the derivatives ∇X ∇Y fail to commute when acting on functions. The torsion can be interpreted as how much a frame must twist as it moves along auto-parallel curves.
We can also define the curvature of the connection. This is a tensor that takes three vector fields and produces a fourth. This measures how non-flat the connection is. If we let X,Y and Z be smooth vector fields, the curvature tensor R is defined in the following way:
∇ R (X,Y,Z) = ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ]Z
1.3.2 Metric Connections and Complex Connections
There is no a priori relation between a Riemannian metric and an affine connection.
In order to be able to theorems about the metric using connections, we really want to consider metric connections. A connection ∇ is said to be metric if ∇g = 0. This gives a natural compatibility condition between connections and Riemannian metrics.
There are reasons to study non-metric connections, but all of the connections that we consider here will be metric.
4 We can also define what it means for a complex structure and a connection to
be compatible. We say that a connection is complex if it satisfies ∇J = 0. Given a
Hermitian manifold, there are several canonical connections. We will primarily focus
on two of these, the Levi-Civita connection and the Chern connection.
The Levi-Civita connection is the unique metric connection which is torsion-free.
We denote this connection as ∇L.C.. It is the main connection used in Riemannian geometry.
The Chern connection is the unique connection which is metric, complex, and whose torsion is of type (2, 0) with respect to J. ∇h. This connection can be defined
h L.C. 1 using the following formula: g(∇X Y,Z) = g(∇X Y,Z) − 2 dω(JX,Y,Z) [2]. These two connections are the same if and only if the complex structure is K¨ahler,
in which case we have ∇L.C.J = 0. Since the Chern connection is metric, it admits
the same autoparallel curves as the Levi-Civita connection and will differ from the
∇h l Levi-Civita only by a contorsion tensor. Here, we denote τijk ≡ gklτij . We also Pn j denote the torsion one form ηi as j=1 Tij. It is worth noting that the Lie form, which is usually defined as θ = Jδω, is directly related to η. By computing in a
unitary frame, we have that θ = −2ηi − 2ηi.
It is straightforward to see that the torsion of the Chern connection controls
∇L.C.J. This observation will be crucial later on, and so we denote it as a lemma.
Lemma 1. There is a C1 estimate on the complex structure in terms of the torsion.
|∇L.C.J| ≤ 3|T ∇h |
To see this, note that we have the following formula:
1 1 − dω(JX,Y,Z) = (τ − τ + τ ) 2 2 JXYZ YZJX ZJXY
5 This characterization gives us an equivalent way of defining the K¨ahlercondition.
A complex structure is K¨ahlerif and only if ∇L.C.J ≡ 0.
For the rest of this thesis, we will use T to denote T ∇h to avoid overly stacking
indices.
1.4 A crucial lemma
Much of the work for this thesis owes its roots to Bo Yang and Fangyang Zheng’s
work [27]. By working in a unitary frame on a Hermitian manifold, they derived four
formulas relating the Chern curvature, the torsion of the Chern connection, and the
Riemannian curvature. For clarity, we recall the results of this lemma here. This
lemma was originally proved in [27]
n Theorem. (Lemma 7) Let (M , g) be a Hermitian manifold and let p ∈ M. Let {ei} be a unitary frame near p such that θ|p = 0. Furthermore, let R···· be components of
h the Riemannian curvature (in terms of the unitary frame), R···· be components of the
· Chern curvature and T·· be the components of the Chern torsion tensor. Then, at the point p we have:
k h h 2Tij,¯l = Rj¯lik¯ − Ri¯ljk¯
l l r l r Tij,k = Rijk¯l − TriTjk + TrjTik
l k r r k j l i l j k i 2Rijkl = Tij,k¯ − Tij,¯l + 2TijTkl + TriTrl + TrjTrk − TriTrk − TrjTrl
h j i r r j i l k Rk¯li¯j = Rk¯li¯j − Tik,¯l − Tjl,k¯ + TikTjl − TrkTrl − TriTrj
where r is summed through and h,i = ei(h) and h,¯i =e ¯i(h).
6 This lemma was invaluable to the rest of the work in this thesis, as it provided
ways of computing basic estimates. In particular, we are able to calculate various
different bounds on the torsion in terms of the curvature tensors.
1.5 Torsion bounds
In this section, we discuss ways to control the torsion. These are the initial
estimates that follow directly from the preceding lemma. In particular, we show that
we can bound the C1 norm of the torsion by two norms, which both measure the
“difference” between the Hermitian and Riemannian curvature. This section was originally published in [16].
To begin, we define the two norms that measure how much the metric fails to be
h 2 K¨ahler. This first norm, which we denote as kR − Rk?, is defined in the following way:
h 2 X h 2 X 2 kR − Rk? = |Ri¯jk¯l − Ri¯jk¯l| + 2 |Rijk¯¯l| i,j,k,l i,j,k,l h Recalling that RXY Z¯W¯ = 0 by Gray’s theorem [11], this can be thought of as a norm of the differences of Riemannian and Hermitian curvatures.
h 2 The second norm is denoted kR − Rk??, and is defined as follows:
h 2 X 2 kR − Rk?? = |Rijk¯l| i,j,k,l h Since RXYZW¯ = 0 [27], this can also be thought of as a norm. Using these two norms, we can measure how much a Hermitian metric fails to be K¨ahlerian.
Theorem. The following inequalities hold point-wise:
2 h 2 h ||T || ≤ |R − R|? and ||η|| ≤ |R − R|?
7 Proof. From equation 41 of [27],
1 h h X k 2 Y X (R + R ) − R ¯ ¯ = (|T | + 2Re(T T )) 2 XXY¯ Y¯ Y YX¯ X¯ XYY X XY kY kX k
We also have that RXY X¯Y¯ = RXXY¯ Y¯ − RXYY¯ X¯ . Therefore, we can rewrite the left- hand side of the above equation:
X k 2 Y X (|TXY | + 2Re(TkY TkX )) k
1 h h = (R + R ) − R ¯ ¯ − R ¯ ¯ 2 XXY¯ Y¯ Y YX¯ X¯ XXY Y XY XY 1 h 1 h = (R − R ¯ ¯ ) + (R − R ¯ ¯ ) − R ¯ ¯ 2 XXY¯ Y¯ XXY Y 2 Y YX¯ X¯ Y YXX XY XY
Pn X Y We want to gain a better understanding of k=1 2Re(TkX TkY ). We choose a
P i unitary frame and denote i Tki = ηk. Using this, we have the following equation
" !# X X i j X i X j Re(TkiTkj) = Re Tki Tkj i j i j 2 = |ηk|
That is to say,
X X i j X i 2 2 2Re(TkiTkj) = 2| Tki| = 2|ηk| i j i Thus, we have the following:
X X 1 h 1 h (R − R ¯ ¯) + (R − R ¯ ¯) − R ¯¯ 2 i¯ij¯j iijj 2 j¯ji¯i jjii ijij i j X X X k 2 i j = (|Tij| + 2Re(TkiTkj)) k i j = ||T ||2 + 2||η||2
8 2 h This then immediately proves that ||T || ≤ kR − Rk?.
h This theorem shows that |R −R|? measures how much a metric fails to be K¨ahler
h by directly bounding the torsion. There are many terms in kR − Rk? that are not
h needed to control the torsion, but we define |R − R|? as such in view of the next
lemma.
Lemma 2. Let ∇c00 be the derivative with respect to the Chern connection of a (0, 1)
c00 h vector. Then the following inequality holds: k∇ T k ≤ kR − Rk?
Proof. We use the first equation of Lemma 7 and the following Bianchi identity:
Ri¯jk¯l − Rk¯ji¯l = Ri¯jk¯l + R¯jki¯l
= −Rki¯j¯l = Rik¯j¯l
Combining these we find that:
j h h e¯l(Tik) = (Ri¯jk¯l − Ri¯jk¯l) − (Rk¯ji¯l − Rk¯ji¯l) − Rik¯j¯l
We can also bound the derivative with respect to a (1, 0) vector.
Lemma 3. The following inequality holds where ∇c0 is the derivative with respect to
the Chern connection of a (1, 0) vector:
c0 h h |∇ T | ≤ C(n)kR − Rk? + kR − Rk??
This is an immediate consequence of the equation from Theorem 4 and the fol-
lowing equation from [27]:
l l r l r Tij,k = Rijk¯l + TrjTik − TriTjk
9 Using a straightforward computation, we can relate the derivatives of the torsion
tensor with respect to the Levi-Civita connection to the derivatives of the torsion
tensor with respect to the Chern connection and a quadratic expression in torsion.
Therefore, we also have the following result:
Theorem. Let T be the torsion tensor and ∇T the derivative of the torsion tensor with respect to the Levi-Civita connection. Then there exists a constant C0(n) that grows linearly in n such that the following inequality holds:
0 h h ||∇T || ≤ C (n)kR − Rk? + kR − Rk??
1.6 k-Gauduchon complex structures
A K¨ahlerform ω = g(J·, ·) is said to be k-Gauduchon if i∂∂¯(ωk) ∧ ωn−k−1 = 0.
This condition was introduced by Fu, Wang, and Wu in [8]. When k = n − 1, this is the condition ∂∂ω¯ n−1 which is more commonly known as Gauduchon, or standard
Gauduchon. A result due to Gauduchon shows that for any complex structure J, and compatible metric g, there is another metricg ˜ conformal to g for whichg ˜(J·, ·)
is Gauduchon. However, we will initially focus on k-Gauduchon for other values of k.
A result due to Gauduchon [9] shows that for any integrable complex structure,
the Lie form θ satisfies the following identity:
2n − 2 |θ|2 + 2δθ = S − 2hW (ω), ωi 2n − 1
In this equation, S is the scalar curvature, and W is the Weyl curvature, viewed
as a map of two forms. In light of this, we define the Gauduchon curvature G as
follows: 2n − 2 G = S − hW (ω), ωi 2n − 1 10 We now substitute the Gauduchon curvature into the k-Gauduchon condition. We
consider the following:
¯ k n−k−1 αk = i∂∂(ω ) ∧ ω
Using a slight generalization of the calculation in [20], we can solve for the right hand side to obtain the following.
n 2kω 2 2 α = −(n − 2)η ,¯ +(2k − 1)|η| + (n − k − 1)|T | k n(n − 1)(n − 2) i i
Substituting in the Gauduchon curvature, we obtain:
2kωn α = −(n − 2)G + (2k − n)|η|2 + (n − k − 1)|T |2 k n(n − 1)(n − 2)
¯ k n−k−1 n If the metric is k-Gauduchon, then 0 = i∂∂(ω ) ∧ ω . For n − 1 > k > 2 , this shows the following:
(n − 2)G = (2k − n)|η|2 + (n − k − 1)|T |2
Since (2k − n), (n − k − 1) > 0, this gives us point-wise bounds on torsion in terms
of G alone.
1.6.1 Weak k-Gauduchon structures
We can extend the notion of k-Gauduchon to C1 complex structures, using the
previous identity. We say that a complex structure is weakly k-Gauduchon if (n −
2)G = (2k − n)|η|2 + (n − k − 1)|T |2. This definition is equivalent to the standard
one for twice-differentiable complex structure, but has the advantage of extending the
definition to once-differentiable complex structures.
11 For weakly k-Gauduchon complex structures, the same estimate on the torsion applies. As such, they are uniformly C1 by the earlier lemma.
1.7 The W 1,2 estimate on conformal changes
Under a conformal changeg ˜ = eug of the metric, the Lie form θ satisfies the
following equation.
2 −2u 2 2 X |θe| = e {|θ| + (2n − 1)|∇u| + 2(2n − 1) (θiui) i We calculate the new scalar curvature under the conformal transformation.
S˜ = e−2u S + 2(2n − 1)4u − (2n − 2)(2n − 1)k∇u|2
Since we know that there is a unique unit volume metric in any conformal class that makes the complex structure Gauduchon (ie. δθ = 0), we can try to estimate this conformal change, which provides a natural W 1,2 estimate on the conformal factor.
−2u 2 2 X e {|θ| + (2n − 1)|∇u| + 2(2n − 1) (θiui)} i = 2n − 2 e−2u S + 2(2n − 1)4u − (2n − 2)(2n − 1)k∇u|2 − 2e−2uhW ω, ωi 2n − 1 We rewrite this in the following way.
(2n − 1) {|θ|2 + (2n − 1)|∇u|2 − 4(2n − 1)kθk2 − k∇uk} 2 ≤ 2n − 2 S + 2(2n − 1)4u − (2n − 2)(2n − 1)k∇u|2 + 2kW k 2n − 1
12 Simplifying this, we obtain W 1,2 bounds on u in terms of R, n, and W , which yields a bound on the total scalar curvature of the Gauduchon metric.
1.8 Monotonicity results
As a separate application of Lemma 7, we explore various monotonicity formu- las that compare the curvature of the Chern connection and the curvature of the underlying Riemannian manifold.
In some sense, the Hermitian curvature is greater than the Riemannian curvature on the same manifold, regardless of the choice of complex structure. It is well known that the Chern scalar curvature dominates the Riemannian scalar curvature [27], but there are other monotonicity formulae as well. We have not yet been able to use these formulas to control the geometry in any meaningful way, so at this point we have not found good applications for this work.
One example of such a monotonicity formula is the following. Rh We consider Bh(X,Y ) = XXY¯ Y¯ , which is also known as the holomorphic bisec- |X|2|Y |2 tional curvature.
We also consider its Riemannian counterpart:
aR ¯ ¯ + (1 − a)R ¯ ¯ B (X,Y ) = XXY Y XYY X a |X|2|Y |2
This curvature was discussed in [27] as a possible notion of Riemannian bisectional curvature. We generally denote B(X,Y ) = B0(X,Y ) for conciseness.
13 Using the results from Lemma 7, we have the following sequence of equalities.
X X X 1 Bh(e , e ) − B(e , e ) = (Rh + Rh ) − (R − 2R ) i j i j 2 eie¯iej e¯j ej e¯j eie¯i eie¯iej e¯j eiej e¯ie¯j i j6=i i,j i6=j n X X X k 2 i j = |Tij| + 2Re(TkiTkj) i i6=j k=1 n n X X X k 2 X X X i j = |Tij| + 2Re(TkiTkj) i i6=j k=1 k=1 i i6=j n n X X X k 2 X X i 2 X i 2 = |Tij| + (| Tki| − |Tki| ) i i6=j k=1 k=1 i i n n n X X X k 2 X 2 X X i 2 = |Tij| + |ηk| − |Tki| i j k=1 k=1 k=1 i n n X X X k 2 X X i 2 2 = |Tij| − |Tki| + |η| i j k=1 k=1 i n n X X X k 2 X X k 2 2 = |Tij| − |Tik| + |η| i j k=1 i=1 k X X X k 2 2 = |Tij| + |η| i j k6=j
Here, the fifth line follows from skew symmetry of the torsion and the second equality follows from Lemma 7.
Theorem 4. In any unitary frame, the following equation holds,
X X h X X X k 2 X X i 2 X X i 2 B (ei, ej) − B0(ei, ej) = |Tij| + | Tki| + |Tki| i j6=i i j6=i k6=i,j k i k i
Note that we also have the following.
n X Bh(e , e ) − B (e , e ) = |T k |2 + 2Re(T ei T ei ) i i 0 i i eiei kei kei k=1
14 We can simplify that as follows.
n X Bh(e , e ) − B(e , e ) = 2|T ei |2 i i i i kei k=1
Currently, we have not found a use for these estimates in our geometric analysis.
If it is possible to prove strong monotonicity formulas, it will be possible to compare the Hermitian and Riemannian geometries. However, scalar curvature and similar scalar quantities do not seem strong enough to effectively control the geometry, so stronger monotonicity results are needed.
15 c Chapter 2: The eigenvalues of
Given a metric connection ∇T on a Riemannian manifold, there is an associated
drift Laplace-type operator ∆T := ∆ − g(T, ∇·) where T is the torsion one-form.
Equivalently, ∆T is the trace of the covariant derivative of d. On a general Hermitian
manifold, this is intricately related with the complex geometry.
c If we let {ek} be a basis of an orthonormal frame and ∇k be the covariant Chern
derivative in the ek direction, this operator can be defined explicitly as
c X c c ∆ = − ∇k∇k k .
Equivalently, we have that ∆c = ∆ +ξ(∇·) where ξ is the Lie form of the complex
structure and ∆ is the Laplace-Beltrami operator. As such, for a Hermitian manifold,
∆c is exactly ∇ξ. When the complex structure is not balanced (dωn−1 6= 0), ξ is non-
zero so ∆c acts on functions as a drift-Laplacian. Furthermore, whenever the complex structure is not conformally balanced, the operator is not self-adjoint.
2.1 The drift Laplacian
In this section, we are interested in trying to estimate the principle eigenvalue of
∆c. In order to do this for an arbitrary J orthogonal to a given metric, we must assume minimal regularity on the drift.
16 The drift Laplacian is a natural operator that appears in physical applications. It
has been studied in various settings, including on a Riemannian manifold [7] and for
smooth domains in Euclidean space [12]
The associated heat equation has been studied and the drift term acts as convec-
tion (i.e. stirring). Often, though not always, stirring speeds up the diffusion process.
Therefore, we might expect to be able to derive lower bounds on the real part of the
spectrum for the drift Laplacian. Such an estimate would provide a lower bound on
the exponential decay of the heat equation.
The following theorem provides an example of such bounds. If the drift-Laplacian
is non-self-adjoint, the statement applies only to the real eigenvalues.Given a non-
selfadjoint operator ∆+v(∇·) on Riemannian manifold, the spectrum of this operator is not necessarily real, but are isolated points in a conic neighborhood of the real axis.
We expect that there are positive lower bounds on the real part of any eigenvalue, but we are currently only able to derive lower bounds for real eigenvalues.
2.1.1 Previous work
Conjecture 5. Given a compact Hermitian manifold (M n, h), there exists a constant
C depending only on the Riemannian geometry such that if u = λu then λ ≥ C.
This would mirror the case for the Laplace-Beltrami operator, where the spectrum can be estimated using the dimension, diameter, and Ricci curvature.
In [16], we proved a similar result on the spectrum using a C1 bound on the drift.
Theorem 6. Let (M n, g) be a compact Riemannian manifold without boundary, sat-
isfying Ric M ≥ −(n − 1)k and v be a one-form on M. Suppose u ∈ C3(M) is a
solution to the equation ∆u + v(∇u) = λu with λ real. Let kvk be the C0 norm of
17 2 v and k∇vk be the C0 norm of ∇v (as a two tensor). Let D = 2nd where d is the
1 2 2 2 diameter of M and E = 2n ((16n − 32n + 5)kvk + 2(n − 1) k + 2(n − 1)k∇vk) Then we have: √ 1 (1 + 1 + 4DE)2 − DE λ ≥ √ D exp(1 + 1 + 4DE)
We then translated this estimate on the eigenvalue on the drift Laplacian into an estimate on the Laplacian on a Hermitian manifold.
Theorem 7. Suppose that (M n, h) is a compact, Hermitian manifold. Then there exists a uniform C > 0 such that any real eigenvalue λ of the Chern connection
Laplacian satisfies: