Twisted K Ahler-Einstein Metrics in Big Classes: Tam As Darvas and Kewei Zhang

Twisted Kähler–Einstein metrics in big classes
Tamás Darvas and Kewei Zhang
Abstract
We prove existence of twisted Kähler–Einstein metrics in big cohomology classes,
arXiv:2208.08324v2 [math.DG] 30 Aug 2022
using a divisorial stability condition. In particular, when −KX is big, we obtain a

uniform Yau–Tian–Donaldson existence theorem for Kähler–Einstein metrics. To
achieve this, we build up from scratch the theory of Fujita–Odaka type delta in-
variants in the transcendental big setting, using pluripotential theory.
1 Introduction
The study of canonical metrics in Kähler geometry goes back to early work of Calabi [22]
and Yau’s solution of the Calabi conjecture [76]. The different versions of the Yau–Tian–
Donaldson (YTD) conjecture predict that existence of canonical metrics is equivalent
to various algebraic stability conditions of the underlying manifold. We refer to the
textbooks [67, 70] for an introduction to this exciting topic.
In this work we study divisorial stability of transcendental big cohomology classes,
and obtain a uniform Yau–Tian–Donaldson type existence theorem for twisted Kähler–
Einstein (KE) metrics in this setting.
We recall the definition of the Ricci curvature and Kähler–Einstein metrics in the big
context. Let (X, ω) be a compact Kähler manifold, and θ a smooth closed (1, 1)-form
representing a big cohomology class {θ} in H 1,1(X, R). All volumes and measures in this
work are interpreted in the non-pluripolar context of [17] (see Section 2.1).
Given u ∈ E 1 (X, θ) (see Definition 2.1) suppose that the non-pluripolar measure
n c n n n n

θu := (θ + dd u) is absolutely continuous with respect to ω . If log θu /ω is integrable,
¯
then we introduce Ric θu as the following current, with the convention ddc := i∂ ∂/2π:
cθun
Ric θu = Ric ω − dd log n . (1)
ω
Now we define twisted Kähler–Einstein currents/metrics. Let η be a smooth closed
(1, 1)-form representing the cohomology class for which the following decomposition holds:
c1 (X) = {θ} + {η}.
Hodge theory provides f ∈ C ∞ (X) such that Ric ω = θ + η + ddc f . Let ψ be a quasi-
plurisubharmonic function on X and ηψ := η +ddc ψ will be our twisting. We want to find
u ∈ PSH(X, θ), with minimal singularity type, satisfying the ηψ -twisted KE equation:
Ric θu = θu + ηψ .
Using the identity Ric ω = θ + η + ddc f and (1) we reduce the twisted KE equation
to the following complex Monge–Ampère equation:
(θ + ddc u)n = e−u+f −ψ ω n . (2)
1
Immediately we see that e−ψ needs to be integrable for the right hand side to make
sense, i.e., ψ needs to have klt singularity.
Already in the Kähler case it is well known that the above equation is not always solv-
able, with a whole host of algebraic criteria available, under the umbrella of K-stability.
In the Fano case, when {θ} = c1 (X), YTD type existence theorems were proved using
the continuity method/Cheeger–Colding–Tian theory [25, 38, 71, 72], Kähler–Ricci flow
[26] and the variational/non-Archimedean method [9]. Recently the second author gave
a proof using Kähler quantization [77].
In the singular setting of log-Fano pairs, YTD type existence theorems were proved
in [59, 60, 61, 62]. After a resolution of singularities, this case is roughly equivalent with
looking for twisted KE metrics in big and semi-positive classes.
Regarding greater generality, the current algebraic theory collapses in the non-ample
case, with delicate perturbation techniques available to treat only some exceptional (semi-
positive) cases. We circumvent these difficulties in the big case by developing an analytic
approach to divisorial stability via pluripotential theory.
Of course, the original definition of K-stability going back to Tian [68] and extended
by Donaldson [46] is not applicable, as our big class {θ} is not necessarily induced by a
line bundle. However the more recent interpretation of K-stability in terms of divisorial
data, going back to Fujita [49] and Li [58], adapts naturally to our context. Indeed,
based on the Blum–Jonsson interpretation of the Fujita–Odaka delta invariant [14, 21,
50], rooted in the non-Archimedean approach to K-stability (see [9, Definition 7.2]), we
define the twisted delta invariant of our data:
Aψ (E)
δψ ({θ}) := inf . (3)
E Sθ (E)
Here Aψ (E) := AX (E) − ν(ψ, E) and the infimum is taken over all prime divisors E
over X, i.e., E is a prime divisor inside a Kähler manifold Y with π : Y → X being a
proper bimeromorphic map. When X is projective, one can equivalently only consider
projective birational morphisms π, as explained in §6. Here the log discrepancy AX (E)
of E is AX (E) := 1 + coeff E (KY − π ∗ KX ), and ν(ψ, E) denotes the Lelong number of
π ∗ ψ at a very generic point of E (see (13)). The expected Lelong number Sθ (E) of {θ}
along E is defined by
Z τθ (E)
1
Sθ (E) := vol({π ∗ θ} − x{E})dx,
vol({θ}) 0
where τθ (E) := sup{τ ∈ R s.t. {π ∗ θ} − τ {E} is big} is the pseudoeffective threshold.
We emphasize that the volume function vol(·) is understood in the sense of [17], but
it coincides with the algebraic volume in case {θ} is in the Néron–Severi space [17,
Proposition 1.18]. Hence our delta invariant extends the one by Fujita–Odaka to the
transcendental case. When η = 0 and ψ = 0, we use the much simpler notation δ := δ0 .
Definition 1.1. We say (X, {θ}, ηψ ) is ηψ -twisted uniformly K-stable if δψ ({θ}) > 1.
Some remarks are in order, to motivate this definition. In the Fano case, it is known
by [14, Theorem B] that δ(−KX ) > 1 if and only if (X, −KX ) is uniformly K-stable/Ding
stable (as introduced in [19, 42]/[5, 9]). Recent work of Liu–Xu–Zhuang [62] shows that
δ(−KX ) > 1 if and only if (X, −KX ) is K-stable. This condition is further equivalent
with (2) having a unique solution [9, Theorem A]. The main result of this paper is
establishing the historically “harder” direction of this equivalence in the big context, in
the presence of positive twisting current:
2
Theorem 1.2. Suppose that ηψ ≥ 0. If δψ ({θ}) > 1 then (2) has a solution u ∈
PSH(X, θ) with minimal singularity, i.e., Ric θu = θu + ηψ .
We note the following corollary, in the particular case when c1 (X) = {θ} and ψ = 0:
Corollary 1.3. Suppose that −KX is big. If δ(−KX ) > 1 then (2) has a solution
u ∈ PSH(X, θ) with minimal singularity, i.e., θu is a KE metric satisfying Ric θu = θu .
Paraphrased in the YTD framework (as in [14, Theorem B]), the above result says
that uniform K-stability implies existence of a KE metric when −KX is big, perhaps
the first example of a YTD type existence theorem in the big case. When X is Fano,
the argument of Corollary 1.3 gives a novel pluripotential theoretic proof of the classical
(uniform) YTD conjecture, in addition to the ones mentioned above.
One can produce examples with −KX big and δ(−KX ) > 1. Indeed, let V be a del
π
Pezzo surface with klt singularities and let X −
→ V be the minimal resolution of V . Then
−KX = −π ∗ KV + ai Ei , where ai ≥ 0, Ei ’s are π-exceptional curves,
P
and (Ei · Ej ) is
negative definite. Note that −KX is big and −KX = −π ∗ KV +
P
ai Ei is the Zariski
decomposition of −KX . Also, −KX is not nef when the singularities of V are worse than
Du Val. Now if δ(−KV ) > 1, then one can show that δ(−KX ) > 1 as well. Examples of
such V are constructed in [23]. Hence Corollary 1.3 is applicable in this case.
We expect that the twisted KE metric found in Theorem 1.2 is unique, an open
question going back to Berndtsson [11, page 4]. Conversely, if a unique twisted KE
metric exists, in Remark 5.6 we point out that δψ (−KX ) > 1, sketching the argument for
the historically “easier” reverse direction to Theorem 1.2 (see [36, §3] for details).
Compared to previous works, we do not require convexity of the Mabuchi energy. Our
main ingredients are the valuative criteria for integrability [9, 18], (see [16] for an excel-
lent learning source), the Guan–Zhou openness theorem [52], the relative pluripotential
theory developed in [30, 31, 33], and the Ross–Witt Nyström correspondence between
(sub)geodesic rays and test curves introduced in [65], and studied further in [29, 37].
We believe the novelty of our work partly lies in the adaptability and robust nature of
our transcendental techniques. Whenever one can prove convexity of the Ding functional
of certain canonical KE metrics, our methods yield a YTD type existence theorem in
terms of the appropriate delta invariant. To provide evidence, in Section 6 we give a
much simplified proof of the main result of Li–Tian–Wang [61] using our techniques.
That ηψ ≥ 0 guarantees that the twisted Ding functional is convex, as follows from
the main result of [13]. This is used in our proof of Theorem 1.2 in one specific point,
but the rest of the argument works in much more general context, allowing to treat many
different problems in the literature at the same time, as we now point out.
To start, we consider a general quasi-plurisubharmonic (qpsh) function ψ on X that
does not necessarily satisfy ηψ ≥ 0. In addition, let χ be another qpsh function on X with
analytic singularity type, and consider the following more general equation of twisted KE
type, for u ∈ PSH(X, θ) with minimal singularity type:
(θ + ddc u)n = e−u+χ−ψ ω n . (4)
To make sense of this equation, we must assume that X eχ−ψ ω n < ∞, i.e., χ − ψ is
R
klt. Clearly this is more general than (2), as any f ∈ C ∞ (X) is a qpsh function. When
treating canonical metrics, it is natural to consider some type of continuity method. This
is the case here as well, as we consider the following family of equations, for λ > 0:
(θ + ddc u)n = e−λu+χ−ψ ω n . (5)
3
At least for small λ > 0 the Guan–Zhou openness theorem [52] gives X e−λv+χ−ψ ω n < ∞
R
for all v ∈ E 1 (X, θ) (See Theorem 2.2 and Proposition 2.5 below).
The expression on the right hand side of (5) often appears in the literature and, after
adding a constant, it is convenient to introduce the following Radon probability measure:
µ := eχ−ψ ω n . (6)
As in [8], such measures µ are called tame. Following [45], one defines a functional, whose
Euler-Lagrange equation is exactly (5). This is Dµλ : E 1 (X, θ) → R, the λ-Ding functional:
1
Z
λ
Dµ (ϕ) = − log e−λϕ dµ − Iθ (ϕ) for ϕ ∈ E 1 (X, θ), (7)
λ X
where Iθ (·) is the Monge-Ampère energy (see (12)). Our starting point is the following
result that gives a formula for the slope of the λ-Ding functional along subgeodesic rays:
Theorem 1.4. Let (0, ∞) ∋ t 7→ ut ∈ E 1 (X, θ) be a sublinear subgeodesic ray. Then
Dµλ (ut ) Iθ (ut )
Z
lim inf = − lim + sup{τ ∈ R : e−λûτ dµ < ∞}. (8)
t→∞ t t→∞ t X
For the definition of subgeodesic rays {ut }t and their Legendre transforms {ûτ }τ see
Definition 3.1 and (18). Our λ-Ding functional is typically not convex along subgeodesics
and we overcome this difficulty by building on ideas from [37, Section 4] to prove (8). We
refer to [5, Theorem 1.3] and [9, Theorem 5.4] for similar flavour results.
On the heels of the above theorem it is convenient to introduce
λ
Dµλ (ut )
Dµ {ut } := lim inf ,
t→∞ t
and we will call this expression the radial λ-Ding functional of the subgeodesic ray {ut }t .
Next we introduce a more general delta invariant associated to our data:
Aχ,ψ (E)
δµ = δµ ({θ}) := inf , (9)
E Sθ (E)
where Aχ,ψ (E) := AX (E) + ν(χ, E) − ν(ψ, E) and the inf is taken over prime divisors E
in smooth bimeromorphic models over X. Note that Aχ,ψ (E) > 0 by [9, Theorem B.5].
The main technical ingredient in the proof of Theorem 1.2 is the following result,
relating δµ to geodesic semistability of the λ-Ding functionals:
Theorem 1.5. With the notation from above we have
δµ = sup{λ > 0 | Dµλ {ut } ≥ 0 for all sublinear subgeodesic ray ut ∈ E 1 (X, θ)}.
When {θ} is ample and χ = 0, this result is proved in [77] using quantization. See also
[75, Proposition 4.5] for a similar statement in the log Fano setting, and [20, Theorem 3.16]
for a related result about ample classes, formulated using non-Archimedean language.
Theorem 1.5 generalizes these results to big cohomology classes in H 1,1 (X, R), yielding a
universal analytic interpretation of the delta invariant.
Regarding the much more developed theory of KE metrics on manifolds of general type
we note the important work [17], that studies KE equations when KX is big, in which case
a solution always exists. See [10, 48] for synergies between singular KE metrics and the
minimal model program on varieties of general type. Regarding exciting developments
on csck metrics we refer to [20, 43] for connections with divisorial notions of K-stability.
Regarding transcendental K-stability in the Kähler setting, we mention the works [44,
47] exploring a cohomological notion of stability.
4
Future directions. To stay brief, we cannot treat several interesting related questions.
Though the condition ηψ ≥ 0 allows one to conjecture a Bando–Mabuchi uniqueness
theorem [2] for twisted KE metrics [11, page 4], it does come with a slight limitation.
Indeed, when (2) is solvable, a positive klt current will exist in c1 (−KX ). By Nadel
vanishing, this implies that H 2 (X, OX ) = 0, hence R-divisors exhaust the whole space
H 1,1 (X, R). Since neither Theorem 1.4 nor 1.5 uses this numerical condition, we wonder
if ηψ ≥ 0 can be omitted from Theorem 1.2 as well. In this direction, [77, Theorem 2.3]
treats the ample case, suggesting that this is possible.
As pointed out above (3), our definition of δψ is rooted in the non-Archimedean
approach to K–stability. As a result, our treatment has an algebraic/non-Archimedean
interpretation that can be expanded when {θ} is the Chern class of a big line bundle.
We will explore this in the companion paper [36].
Another natural question is to determine the regularity of solutions to (2) in the ample
locus of {θ}. One would hope that the solution is smooth on this open set. The first step
would be to establish this when {θ} is big and nef, akin to [17, Theorem C]. See [79] for
progress on the regularity of conic csc Kähler metrics in big and semipositive classes.
We are curious about connections with the minimal model program. When −KX is
big and nef, X is a crepant resolution of a Fano variety Z with only canonical singularities,
by Kawamata base point freeness. In this case the KE metrics on X are the pullbacks of
KE metrics on Z (we are grateful to Y. Odaka for pointing this out). When −KX is only
big, the above can not hold, but we wonder if one can prove similar flavour theorems.
As is well known, in the (log) Fano case δ > 1 implies that X can not contain any
nontrivial vector fields. See [6] for a related work using Gibbs stability in the big case.
It is desirable to extend our treatment to allow for a non-trivial automorphism group as
well. This should be possible after incorporating ideas from the works [55, 59].
The recent surge of interest in the delta invariant is partially due to its computability
for ample line bundles, as evidenced by [1, 24, 64], just to name a few works. We are
very curious if such computations can also be carried out in case of big line bundles.
Lastly, our approach seems adaptable to the case of canonical KE metrics with pre-
scribed singularity type, as recently studied by Trusiani [73]. If the Ross–Witt Nyström
correspondence could be extended to this context, we suspect that the analogue of The-
orem 1.2 could be established, as the convexity of the corresponding Ding functional is
known in that setting.
Organization. In Section 2 we recall some aspects of finite energy pluripotential theory
and complex singularity exponents. In Section 3 we give a detailed exposition of the Ross–
Witt Nyström correspondence between rays and test curves in the big case. In Section 4
we prove Theorem 1.4 and Theorem 1.5. In Section 5 we prove Theorem 1.2. In Section
6 we give a short proof of the main result of [61].
Acknowledgments. We thank B. Berndtsson, M. Jonsson, Y. Odaka, M. Xia, X.
Zhou and X. Zhu for suggestions on how to improve the presentation of the paper. Upon
noticing our work on the arXiv, R. Dervan and R. Reboulet informed us about their
ongoing independent study of test configurations and non-Archimedean geometry in the
big case, working on a converse to Corollary 1.3.
The first named author was partially supported by an Alfred P. Sloan Fellowship
and National Science Foundation grant DMS–1846942. The second named author is
supported by NSFC grant 12101052 and the Fundamental Research Funds 2021NTST10
and is grateful to G. Tian for numerous inspiring conversations related to this project.
5
2 Preliminaries
2.1 Finite energy pluripotential theory
In this short subsection we recall the basics of finite energy pluripotential theory in big
cohomology classes. For a more thorough treatment we refer to the articles [17], [29], or
the recent textbook [53].
Let (X, ω) be a compact Kähler manifold of dimension n and θ a smooth closed (1, 1)-
form. A function u : X → R ∪ {−∞} is called quasi-plurisubharmonic (qpsh) if locally
u = ρ + ϕ, where ρ is smooth and ϕ is a plurisubharmonic (psh) function. We say that
u is θ-plurisubharmonic (θ-psh) if it is qpsh and θu := θ + ddc u ≥ 0 as currents. We let
PSH(X, θ) denote the space of θ-psh functions on X.
The class {θ} is big if there exists ψ ∈ PSH(X, θ) satisfying θψ ≥ εω for some ε > 0.
We assume that {θ} is big throughout this paper, unless specified otherwise.
Given u, v ∈ PSH(X, θ), u is more singular than v, (notation: u v) if there exists
C ∈ R such that u ≤ v + C. The potential u has the same singularity as v (notation:
u ≃ v) if u v and v u. The classes [u] of this latter equivalence relation are called
singularity types. When {θ} is merely big, all elements of PSH(X, θ) are very singular,
and we distinguish the potential with minimal singularity:
Vθ := sup{u ∈ PSH(X, θ) such that u ≤ 0}. (10)
A function u ∈ PSH(X, θ) is said to have minimal singularity if it has the same singularity
type as Vθ , i.e., [u] = [Vθ ].
We say that [u] is an analytic singularity
P type if it has a representative u ∈ PSH(X, θ)
2
that locally can be written as u = c log( j |fj | ) + g, where c > 0, g is a smooth function
and the fj are a finite set of holomorphic functions. By a fundamental approximation
theorem of Demailly there are plenty of θ-psh functions with analytic singularity type.
The ample locus Amp(θ) of {θ} is defined as the set of points x ∈ X such that there
exists a θ-psh function u with analytic singularity type, which satisfies that θu ≥ εω for
some ε > 0 and that u is smooth around x.
Let θ1 , ..., θn be smooth closed (1, 1)-forms and ϕj ∈ PSH(X, θj ), j = 1, ...n. Following
Bedford-Taylor in the local setting [3, 4], it has been pointed out in [17] that the sequence
of positive measures
1Tj {ϕj >Vθj −k} θmax(ϕ

1
1 ,Vθ 1 −k)
n
∧ . . . ∧ θmax(ϕ n ,Vθ n −k)
(11)
converges weakly to the so called non-pluripolar product θϕ1 1 ∧ . . . ∧ θϕnn . The resulting
positive Borel measure does not charge pluripolar sets. In the particular case when
ϕ1 = ϕ2 = . . . = ϕn = ϕ and θ1 = ... = θn = θ we will call θϕn the non-pluripolar
Monge-Ampère measure of ϕ, generalizing the usual notion of volume form, when θϕ is a
smooth Kähler form.
As a consequence
R n of Bedford-Taylor theory, the measures in (11)R alln haveR total mass
n
less than X θVθ , in particular, after letting k → ∞ we notice that X θϕ ≤ X θVθ . What
is more, it was proved in [74, Theorem 1.2] that for any u,Rv ∈ PSH(X, R n θ) we have the
n
following monotonicity result for the masses: if v u then X θv ≤ X θu .
Definition R2.1. WeRsay that u ∈ PSH(X, θ) is a full mass potential (notation: u ∈
E(X, θ)) if X θun = X θVnθ =: vol({θ}). Moreover, we say that u ∈ E(X, θ) has finite
energy (notation: u ∈ E (X, θ)) if X |u − Vθ |θun < ∞.
1
R
6
The class E 1 (X, θ) plays a central role in the variational theory of complex Monge–
Ampère equations, as detailed in [8, 17] and later works. Here we only mention that the
Monge–Ampère energy Iθ naturally extends to this space with the usual formula:
n
1 X Z
Iθ (u) = (u − Vθ )θuj ∧ θVn−j , u ∈ E 1 (X, θ). (12)
vol({θ})(n + 1) j=0 X
θ
It is upper semi-continuous (usc) with respect to the L1 topology on PSH(X, θ).

We recall some envelope notions that will be useful in this work. Given any f : X →
[−∞, +∞] the starting point is the envelope Pθ (f ) := usc(sup{v ∈ PSH(X, θ), v ≤ f }),
where usc denotes the upper-semicontinuous regularization. Then, for u, v ∈ PSH(X, θ)
we can introduce the “rooftop envelope” Pθ (u, v) := Pθ (min(u, v)). This allows us to
further introduce envelopes with respect to singularity type [65]:

Pθ [u](v) := usc lim Pθ (u + C, v) .
C→+∞
It is easy to see that Pθ [u](v) depends on the singularity type [u]. When v = Vθ ,
we simply write P [u] := Pθ [u] := Pθ [u](Vθ ) and call this potential the envelope of the
singularity type [u]. It follows from [31, Theorem 3.8] that θPn [u] ≤ 1{P [u]=0} θn . Also, by
[31, Proposition 2.3 and Remark 2.5] we have that X θPn [u] = X θun .
R R
With the help of these envelopes one can define a complete metric on E 1. Indeed, as
pointed out in [32, Theorem 2.10], for u, v ∈ E 1 (X, θ) we have that P (u, v) ∈ E 1 (X, θ)
and the following expression defines a complete metric on E 1 (X, θ) [29, Theorem 1.1]:
d1 (u, v) = Iθ (u) + Iθ (v) − Iθ (P (u, v)).
In addition, d1 -convergence implies L1 -convergence of qpsh potentials [29, Theorem 3.11].
2.2 Complex singularity exponents and openness

To start, we point out several notational differences compared to the closely related work
of Berman–Boucksom–Jonsson [9]. Stemming from our choice ddc := i∂ ∂/2π, ¯ our Ding
functionals (7) are free of factors of two compared to [9, page 4]. Moreover, for any
qpsh function ϕ on X the complex singularity exponent attached to our tame measure
µ := eχ−ψ ω n (recall (6)) is defined to be
Z
−λϕ
cµ [ϕ] := sup λ > 0 : e dµ < ∞ .
X
And our definition of the Lelong number of a qpsh function v at x ∈ X is
ν(v, x) := sup{c > 0 s.t. v(z) − c log |z − x|2 is bounded above near x}. (13)
Given an irreducible analytic subset E ⊂ X, ν(v, E) is equal to the Lelong number of v

at a very general point of E (by [66]).
On the level of potentials, our definition of singularity exponent and Lelong number
differs from the ones used in [9] by a factor of two. However on the level of (1, 1)-currents,
all definitions agree due to our choice of ddc . As an upside, using our terminology, all
the formulas/inequalities we derive in this work contain no extra factors of two.
7
Openness theorems go back to the work of Berndtsson [12, Theorem 4.4] on Demailly–
Kollár’s openness conjecture [41]. Not long after, Guan–Zhou [52] solved the strong
openness conjecture of Demailly [40], which will be used multiple times in this work. The
version below follows from (the proof of) [9, Corollary B.2] and the effective version of
the strong openness theorem [54, Main Theorem (2)] (see also [51, Corollary 1.2]):
TheoremR 2.2. Suppose that λ > 0 and u, uj are qpsh functions on X such that uj ր u
a.e., and X e−λu dµ < ∞. Then there exists j0 such that X e−λuj dµ < ∞ for j ≥ j0 .
R
The above theorem implies that cµ [ϕ] is always positive. The following result (that
follows from [18, 9] and Guan–Zhou openness [52]) gives a more precise valuative char-
acterization of cµ [·].
Theorem 2.3. For any qpsh function ϕ on X one has
Aχ,ψ (E)
cµ [ϕ] = inf , (14)
E ν(ϕ, E)
where Aψ (E) := AX (E) + ν(χ, E) − ν(ψ, E) and the infimum is taken over all prime
divisors E in smooth bimeromorphic models over X.
If on the right hand side of (14) we have ν(ϕ, E) = 0, then we use the convention
1/0 = ∞. No ambiguity will arise from this, as follows from the proof below.
Proof. Let λ ∈ (0, cµ [ϕ]). By [9, Theorem B.5], for small enough ε > 0 we have
ν(χ, E) + AX (E) ≥ (1 + ε)(ν(ψ, E) + λν(ϕ, E)) ≥ ν(ψ, E) + (1 + ε)λν(ϕ, E)
for any prime divisor E over X. Letting ε ց 0 and then λ ր cµ [ϕ], we arrive at
Aχ,ψ (E)
cµ [ϕ] ≤ inf .
E ν(ϕ, E)
Aχ,ψ (E) Aχ,ψ (E)
For the other direction, let λ ∈ 0, inf E ν(ϕ,E) (note that inf E ν(ϕ,E)
is indeed
positive by the previous step). So there is some ε > 0 such that
ν(χ, E) + AX (E) ≥ ν(ψ, E) + (1 + ε)λν(ϕ, E)
for all E. On the other hand, we have X eχ−ψ ω n = X dµ < ∞. Using [9, Theorem B.5]
R R
again, for any small τ > 0 we have that
ν(χ, E) + AX (E) ≥ (1 + τ )ν(ψ, E)
holds for all E. Summarizing, we find that
(1 + τ )(ν(χ, E) + AX (E)) ≥ (1 + τ + τ 2 )ν(ψ, E) + (1 + ε)λν(ϕ, E).
Or equivalently,
1 + τ + τ2 1+ε
ν(χ, E) + AX (E) ≥ ν(ψ, E) + λν(ϕ, E).
1+τ 1+τ
Finally, choosing τ such that τ < ε, we get ν(χ, E)+AX (E) ≥ (1+ε′ )(ν(ψ, E)+λν(ϕ, E))
for some ε′ > 0, which holds for all E. Using [9, Theorem B.5] one more time we obtain
cµ [ϕ] ≥ λ, implying the other direction.
8
Theorem 2.3 has the following consequence.
Lemma 2.4. Let {ψτ }τ ∈I be a family of qpsh functions on X, where I ⊂ R is some

connected open interval. Assume that I ∋ τ 7→ ψτ (x) is concave for any x ∈ X. Then
the map I ∋ τ 7→ 1/cµ [ψτ ] is convex, so in particular it is also continuous.
Proof. For τ1 , τ2 ∈ I and λ ∈ [0, 1], concavity reads ψλτ1 +(1−λ)τ2 ≥ λψτ1 + (1 − λ)ψτ2 . Let
E be any prime divisor over X. Then ν(ψλτ1 +(1−λ)τ2 , E) ≤ λν(ψτ1 , E) + (1 − λ)ν(ψτ2 , E),
so that
ν(ψλτ1 +(1−λ)τ2 , E) λν(ψτ1 , E) (1 − λ)ν(ψτ2 , E)
≤ + .
Aχ,ψ (E) Aχ,ψ (E) Aχ,ψ (E)
Taking sup over all E, by Theorem 2.3, we conclude the result.
Lastly we recall a consequence of Theorem 2.2 and [31, Theorem 1.3]:
Proposition 2.5. cµ [u] = cµ [P [u]] for any u ∈ PSH(X, θ). In particular cµ [v] = cµ [Vθ ]
for any v ∈ E(X, θ).
Proof. Recall that P (Vθ , u + C) ր P [u] a.e. As [P (Vθ , u + C)] = [u] for any C ∈ R, the
previous theorem immediately gives that cµ [u] = cµ [P [u]]. By [31, Theorem 1.3] we have
that P [v] = Vθ for any v ∈ E(X, θ). The last statement immediately follows.
3 The extended Ross–Witt Nyström correspondence

We give a precise correspondence between finite energy geodesic rays and certain maximal
test curves in the big case. This theory was initiated by Ross and Witt Nyström in [65]
in the Kähler case, and developed further in [29] and [37].
Definition 3.1. A sublinear subgeodesic ray is a subgeodesic (0, ∞) ∋ t 7→ ut ∈ PSH(X, θ)

(notation: {ut }t ) such that ut →L1 u0 := Vθ as t → 0, and there exists C ∈ R such that
ut (x) ≤ Ct for all t ≥ 0, x ∈ X.
Using t-convexity, we obtain some immediate properties of sublinear subgeodesic rays:
Lemma 3.2. Suppose that {ut }t is a sublinear subgeodesic ray. Then the set {ut > −∞}
is the same for any t > 0. In particular, for any x ∈ X the curve t 7→ ut (x) is either
finite and convex on (0, ∞), or equal to −∞ on this interval.
A psh geodesic ray is a sublinear subgeodesic ray that additionally satisfies the fol-
lowing maximality property: for any 0 < a < b, the subgeodesic (0, 1) ∋ t 7→ vta,b :=
ua(1−t)+bt ∈ PSH(X, θ) can be recovered in the following manner:
vta,b := sup ht , t ∈ [0, 1] , (15)

h∈S
where S is the set of subgeodesics (0, 1) ∋ t → ht ∈ PSH(X, θ) such that limtց0 ht ≤ ua

and limtր1 ht ≤ ub .
Remark 3.3. Let {ut }t be a psh geodesic ray. Due to [29, (10)] the map t 7→ supX utt =
supX ut −V
t
θ
= supAmp{θ} ut −V
t
θ
, t > 0 is linear, using an approximation argument via
decreasing geodesic segments with minimal singularity type.
9
Making small tweaks to [65, Definition 5.1], we now give the definition of test curves:
Definition 3.4. A map R ∋ τ 7→ ψτ ∈ PSH(X, θ) is a psh test curve, denoted {ψτ }τ , if
(i) τ 7→ ψτ (x) is concave, decreasing and usc for any x ∈ X.
(ii) ψτ ≡ −∞ for all τ big enough, and ψτ increases a.e. to Vθ as τ → −∞.
Note that this definition is more general than the one in [65] (where the authors
only considered potentials with small unbounded locus), more general than the one in
[29] (where the authors considered only bounded test curves, see below), and more gen-
eral than the one in [37, Section 3.1] (where the authors only consider Kähler classes).
Moreover, condition (ii) allows for the introduction of the following constant:
τψ+ := inf{τ ∈ R : ψτ ≡ −∞} . (16)
We adopt the following convention: psh test curves will always be parametrized by
τ , whereas rays will be parametrized by t. Hence {ψt }t will always refer to some kind of
ray, whereas {φτ }τ will refer to some type of test curve. As pointed out below, rays and
test curves are dual to each other, so one should think of the parameters t and τ to be
dual to each other as well.
Definition 3.5. A psh test curve {ψτ }τ can have the following properties:
(i) {ψτ }τ is maximal if P [ψτ ] = ψτ for any τ ∈ R.
(ii) {ψτ }τ is a finite energy test curve if
Z τ + Z Z
ψ
n n
θψτ − θVθ dτ > −∞ .
−∞ X X
(iii) We say {ψτ }τ is bounded if ψτ = Vθ for all τ negative enough. In this case, one
can introduce the following constant, complementing (16):
τψ− := sup { τ ∈ R : ψτ ≡ Vθ } .
In the above definition, we followed the convention P [−∞] = −∞. Note that bounded
test curves are clearly of finite energy.
We recall the Legendre transform, that will establish the duality between various types
of maximal test curves and geodesic rays. Given a convex function f : [0, +∞) → R, its
Legendre transform is defined as
fˆ(τ ) := inf (f (t) − tτ ) = inf (f (t) − tτ ) , τ ∈ R.
t≥0 t>0
The (inverse) Legendre transform of a decreasing concave function g : R → R ∪ {−∞} is

ǧ(t) := sup(g(τ ) + tτ ) , t ≥ 0.
τ ∈R
There is a sign difference in our choice of Legendre transform compared to the convex
analysis literature, however this choice will be more suitable for our discussion.
We recall that, for every τ ∈ R we have ǧˆ(τ ) ≥ g(τ ) with equality if and only if g
ˇ
is additionally τ -usc. Similarly, fˆ(t) ≤ f (t) for all t ≥ 0 with equality if and only if f
ˇ
is t-lsc. In general, ǧˆ is the τ -usc envelope of g, and fˆ is the t-lsc envelope of f . These
properties are called the involution property of the Legendre transform.
Starting with a psh test curve {ψτ }τ , our goal is to construct a geodesic/subgeodesic
ray by taking the τ -inverse Legendre transform. The first step is the next proposition,
essentially proved in [28]:
10
Proposition 3.6. Suppose {ψτ }τ is a psh test curve. Then supτ (ψτ (x) + tτ ) is usc with
respect to (t, x) ∈ (0, ∞) × X.
Since τψ+ < ∞ and ψτ ≤ Vθ , τ ∈ R, we note that supτ (ψτ + tτ ) ≤ Vθ + tτψ+ for
t ≥ 0. Also, for t = 0 the above proposition may fail, so it is important to only consider
(t, x) ∈ (0, ∞) × X in the statement.
Proof. Let S = {Re s > 0}. We consider the usual complexification of the inverse
Legendre transform:
u(s, z) := sup(ψτ (z) + τ Re s) , (s, z) ∈ S × X .
τ
Also, ut (x) := u(t, x) ≤ Vτ + tτψ+ ≤ tτψ+ , t > 0. Clearly, usc u ∈ PSH(S × X, π ∗ ω), where
usc u is the usc regularization of u on S × X, where π : S × X → X is the natural
projection. It will be enough to prove that usc u = u.
We introduce the set E = {u < usc u} ⊆ S × X. As both u and usc u are R-invariant
in the imaginary direction of S, it follows that E is also R-invariant, i.e., there exists
B ⊆ (0, ∞) × X such that E = B × R.
As E has Monge–Ampère capacity zero, it follows that E has Lebesgue measure zero.
By Fubini’s theorem B ⊆ (0, ∞) × X has Lebesgue measure zero as well. For z ∈ X, we
introduce the z-slices of B:
Bz = B ∩ ((0, ∞) × {z}) .
By Fubini’s theorem again, we have that Bz has Lebesgue measure 0 for all z ∈ X \ F ,
where F ⊆ X is some set of Lebesgue measure 0.
By slightly increasing F , but not its zero Lebesgue measure, we can additionally
assume that ut (z) > −∞ for all t > 0 and z ∈ X \ F (indeed, at least one potential ψτ
is not identically equal to −∞).
Let z ∈ X \ F . We argue that Bz is in fact empty. By our assumptions on F , both
maps t 7→ ut (z) and t 7→ (usc u)(t, z) are locally bounded and convex (hence continuous)
on (0, ∞). As they agree on the dense set (0, ∞) \ Bz , it follows that they have to be the
same, hence Bz = ∅. This allows to conclude that
inf [ut (x) − τ t] = χτ := inf [(usc u)(t, x) − τ t] , τ ∈ R and z ∈ X \ F . (17)
t>0 t>0
By duality of the Legendre transform ψτ (x) = inf t>0 [ut (x) − tτ ] for all x ∈ X and τ ∈ R
(using the τ -usc property of τ 7→ ψτ ). From this and (17) it follows that ψτ = χτ a.e. on
X, for all τ ∈ R. Since both ψτ and χτ are θ-psh (the former by definition, the latter by
Kiselman’s minimum principle [39, Theorem I.7.5]), it follows that ψτ ≡ χτ for all τ ∈ R.
Consequently, applying the τ -Legendre transform to the τ -usc and τ -concave curves
τ 7→ ψτ and τ 7→ χτ , we obtain that t 7→ ut (x) is the t-lsc envelope of t 7→ usc u(t, x) on
(0, ∞) for all x ∈ X. hence, Lemma 3.2 gives that ut (x) = usc u(t, x) on (0, ∞) × X.
Given a sublinear subgeodesic ray {φt }t (psh test curve {ψτ }τ ), we can associate its
(inverse) Legendre transform at x ∈ X as
φ̂τ (x) := inf (φt (x) − tτ ) , τ ∈ R ,
t>0
(18)
ψ̌t (x) := sup(ψτ (x) + tτ ) , t > 0 .
τ ∈R
Our next theorem describes a duality between various types of rays and maximal test
curves, extending various particular cases from [29, 65]:
11
Theorem 3.7. The Legendre transform {ψτ }τ 7→ {ψ̌t }t gives a bijective map with inverse
{φt }t 7→ {φ̂τ }τ between:
(i) psh test curves and sublinear subgeodesic rays,
(ii) maximal psh test curves and psh geodesic rays,
(iii)[29, 65] maximal bounded test curves and geodesic rays with minimal singularity type.
In this case, we additionally have that Vθ + τψ− t ≤ ψ̌t ≤ Vθ + τψ+ t , t ≥ 0 .
Proof. We prove (i). This is essentially [29, Proposition 4.4], where an important par-
ticular case was addressed. Let {ψτ }τ be a psh test curve. Then ψ̌t ∈ PSH(X, θ) for all
t > 0 due to Proposition 3.6. We also see that ψ̌t ≤ Vθ + tτψ+ , and ψ̌t →L1 Vθ as t → 0,
proving that {ψ̌t }t is a sublinear subgeodesic.
For the reverse direction, let {φt }t be a sublinear subgeodesic ray. Then φ̂τ ∈
PSH(X, θ) or φ̂τ ≡ −∞ for any τ ∈ R due to Kiselman’s minimum principle. By prop-
erties of Legendre transforms and Lemma 3.2, we get that τ 7→ φ̂τ (x) is τ -usc, τ -concave
and decreasing. Due to sublinearity of {φt }t we get that φ̂τ ≡ −∞ for τ big enough.
Lastly ψτ ր Vθ a.e. as τ → −∞, since φt →L1 Vθ as t → 0.
We prove (ii). Let τ ∈ R and {ut }t a psh geodesic ray. From [28, Propisition 5.1] (that
only uses the maximum principle (15)) we obtain that ûτ = P [ûτ ](Vθ ) = P [ûτ ](0) = P [ûτ ].
Since {ut }t is sublinear, the curve {ûτ }τ is a maximal psh test curve.
Conversely, let {ψτ }τ be a maximal psh test curve. We will show that the sublinear
subgeodesic {ψ̌t }t is a psh geodesic ray. By elementary translation properties of the
Legendre transform we can assume that τψ+ = 0, in particular {ψ̌t }t is t-decreasing.
Now assume by contradiction that {ψ̌t }t is not a psh geodesic ray. Comparing with
(15), there exists 0 < a < b such that
ψ̌(1−t)a+tb χt := sup ht , t ∈ [0, 1] ,

h∈S
where S is the set of subgeodesics (a, b) ∋ t 7→ ht ∈ PSH(X, θ) satisfying lim ht ≤ ψ̌a

t→a+
and lim ht ≤ ψ̌b . Now let {φt }t be the sublinear subgeodesic such that φt := ψ̌t for
t→b−
t 6∈ (a, b) and φa(1−t)+bt := χt otherwise.
Trivially, ψ̌t ≤ φt ≤ 0, hence by duality, ψτ ≤ φ̂τ ≤ 0 and τψ+ = τφ̂+ = 0. However,
comparing with (18), we claim that φ̂τ ≤ ψτ + τ (a − b) for any τ ∈ R. Since τψ+ = τφ̂+ = 0,
we only need to show this for τ ≤ 0. For such τ we indeed have
inf (φt − tτ ) ≤ φb − bτ = ψ̌b − bτ ≤ inf (ψ̌t − tτ ) + (a − b)τ ,

t∈[a,b] t∈[a,b]
where in the last inequality we used that t 7→ ψ̌t is decreasing.

By the maximality of {ψτ }τ , we obtain that φ̂τ ≤ P [ψτ ] = ψτ . As we already pointed
out the reverse inequality above, we conclude that φ̂τ = ψτ . The (inverse) Legendre
transform now gives that ψ̌t = φt , a contradiction. Hence {ψt }t is a psh geodesic ray.
The duality of (iii) is simply [29, Theorem 1.3].
Given a finite energy (sub)geodesic ray {ut}t we know that t → Iθ (ut ) is (convex) linear
[32, Theorem 3.12], allowing to introduce the following radial Monge–Ampère energy:
Iθ (ut )
Iθ {ut } := lim . (19)
t→∞ t
12
Before proving the duality between maximal finite energy test curves and rays, we
point out the following approximation result:
Proposition 3.8. Let {ut }t be a psh geodesic ray. Then there exists a sequence of geodesic
rays {ujt } with minimal singularity type such that ujt ց ut and θûnj → θûnτ weakly for
τ
j → ∞ for all t ≥ 0 and τ ∈ R.
Proof. In the Kähler case, this result is a particular case of [34, Theorem 4.5]. As the
argument in the big case is virtually the same, we will be brief, and only point out how
the sequence of approximating rays {ujt }t is constructed.
We can assume without loss of generality that τû+ = 0. For j ∈ N, τ < 0, set
j
τ τ
ψτ (x) := 1 − max 0, 1 + Vθ + max 0, 1 + ûτ , and ûjτ := P [ψτj ].
j j
We define ûj0 := limτ →0− ûjτ .

Since τ → ûτ is τ -concave, τ -decreasing, and ûτ ≤ Vθ , it is elementary to see that
τ → ψτj is also τ -concave and τ -decreasing. By elementary properties of P [·] we get that
τ → ûjτ is also τ -concave and τ -decreasing (see the proof of [29, Proposition 4.6]).
Arguing the same way as in the proof of [34, Theorem 4.5] we conclude that ujt ց ut ,
ûjτ ց ûτ , and θûnj → θûnτ weakly for j → ∞ for all t ≥ 0, τ ∈ R.
τ
Finally we prove the Ross–Witt Nyström correspondence between maximal finite en-
ergy test curves and finite energy rays.
Theorem 3.9. The Legendre transform {ψτ }τ 7→ {ψ̌t }t gives a bijective map with inverse
{φt }t 7→ {φ̂τ }τ between maximal finite energy test curves and finite energy geodesic rays.
In this case, we additionally have that
τψ+ Z
1
Z Z
Iθ {ψ̌t } = θψnτ − θVnθ dτ + τψ+ . (20)
vol({θ}) −∞ X X
Proof. As previously, we may assume that τψ+ = 0. As a preliminary result, in Proposi-

tion 3.10 below we prove (20) for bounded maximal test curves.
Given a finite energy maximal test curve {ψτ }τ , we know that {ψ̌t }t is a psh geodesic
ray. By Proposition 3.8 above there exists geodesic rays {ψ̌tk }t with minimal singularity
type such that ψ̌tk ց ψ̌t for any t ≥ 0, and X θψnk ց X θψnτ for any τ < τψ+ = τψ+k = 0.
R R
τ
By basic properties of the Monge–Ampère energy we have that Iθ {ψtk } = Iθ (ψ1k ) →
Iθ (ψ1 ) = Iθ {ψt }. By Proposition 3.10 below
Z 0 Z
1
Z
k n n
Iθ {ψ̌t } = θψτk − θVθ dτ .
The R right hand

R side is bounded from below, since {ψτ }τ is a finite energy test curve.
n n
Since X θψτk ց X θψτ , we can take the limit on both the left and right hand side, to
arrive at (20), also implying that {ψ̌t }t is a finite energy geodesic ray.
Conversely, assume that {φt }t is a finite energy geodesic ray, with decreasing ap-
proximating sequence of rays {φkt }t , as detailed above. For similar reasons we have
1
R0
Iθ {φkt } = vol({θ})
R n
θ − X ω n dτ . Since I{φkt } ց Iθ {φt }, the monotone conver-
R
−∞ X φ̂k
τ
gence theorem gives that (20) holds for {φ̂τ }τ , finishing the proof.
13
As promised, to complete the argument of Theorem 3.7, we prove the next proposition,
whose argument can be extracted from [65, Section 6] with additional references to [31].
We recall the precise details here as the results of [65] were proved in the context of
potentials with small unbounded locus.
Proposition 3.10. Suppose that {ψτ }τ is a bounded maximal test curve with τψ+ = 0.
Then Z 0 Z
Iθ (ψ̌t ) 1
Z
n n
= Iθ {ψ̌t } = θψτ − θVθ dτ , t > 0 . (21)
t vol({θ}) −∞ X X
Proof. Without loss of generality we assume that vol({θ}) = 1. For N ∈ Z+ , M ∈ Z and

t > 0, we introduce the following:
ψ̌tN,M := max ψk/2N + tk/2N .

k∈Z
k≤M
It is clear that ψ̌tN,M ∈ PSH(X, θ) has minimal singularity type, since it is a maximum of
a finite collection of θ-psh potentials (indeed, {ψτ }τ is a bounded test curve). Moreover,
we now argue that
t t
Z Z
n N,M +1 N,M
N
θψ ≤ Iθ (ψ̌t ) − Iθ (ψ̌t ) ≤ N θψn N . (22)
2 X (M +1)/2N
2 X M/2
Indeed, by [29, Theorem 2.4(iii)],
Z Z
N,M +1 N,M N,M +1 N,M
ψ̌t − ψ̌t n
θψ̌N,M +1 ≤ Iθ (ψ̌t ) − Iθ (ψ̌t ) ≤ ψ̌tN,M +1 − ψ̌tN,M θψ̌nN,M .
t t
X X
(23)
Clearly ψ̌tN,M +1 ≥ ψ̌tN,M , and using τ -concavity we notice that
n o
Ut := ψ̌tN,M +1 − ψ̌tN,M > 0 ∩ Amp(θ) = ψ(M +1)/2N + 2−N t − ψM/2N > 0 ∩ Amp(θ)

ψ̌tN,M +1 = ψ(M +1)/2N + t(M + 1)/2N , ψ̌tN,M = ψM/2N + tM/2N on Ut .

Since Ut is an open set in the plurifine topology, we have θψn N
= θnN,M +1
U t ψ̌t Ut
(M +1)/2
n n n n
and θψ N Ut = θψ̌N,M Ut . Recall that θψ N and θψ
are supported on the sets
M/2 t M/2 (M +1)/2N
{ψM/2N = 0} and {ψ(M +1)/2N = 0} respectively [31, Theorem 3.8]. Since {ψ(M +1)/2N =
0} ⊆ Ut and {ψ(M +1)/2N = 0} ⊂ {ψM/2N = 0}, applying the above to (23), we get (22).
Fixing N, let M = ⌊2N τψ− ⌋ ∈ Z. Repeated application of (22) gives us
t X t Z
X Z
n N,0 N,M
N
θψ N ≤ Iθ (ψ̌t ) − Iθ (ψ̌t ) ≤ N
θψn N .
M +1≤j≤0
2 X j/2
M ≤j≤−1
2 X j/2
Since M ≤ 2N τψ− we have that ψ̌tN,M = ψM/2N + tM/2N = Vθ + tM/2N . As a result

Iθ (ψ̌tN,M ) = tM/2N , and we obtain
0 Z −1 Z
t t
X Z X Z
n n N,0 n n
θψ N − θVθ ≤ Iθ (ψ̌t ) ≤ θψ N − θVθ .
j=M +1
2N X
j/2
X j=M
2N X
j/2
X
We have Riemann sums on both the left and right of the above inequality. Using
Lemma 3.11 below, and the fact that ψ̌ N,0 ր ψ̌t a.e., it is possible to let N → ∞
and obtain (21), as desired.
14
θψnτ > 0 is a
R
Lemma 3.11. Suppose that {ψτ }τ is a psh test curve. Then τ 7→ X
continuous function for τ ∈ (−∞, τψ+ ).
The proof of this result will be omitted, as is it is exactly the same as [37, Lemma
3.9], that deals with the Kähler case.
Lastly, we recall the maximization {vt }t of a finite energy sublinear subgeodesic ray
{ut }t , which goes back to [29, Proposition 4.6] and is determined by the formula:
v̂τ := P [ûτ ], τ < τû+ , (24)
v̂τ + = limτ րτu+ v̂τ and v̂τ = −∞ for τ > τû+ . By the lemma below {v̂τ }τ is a maximal
û
finite energy test curve and by Theorem 3.7 we immediately have that {vt }t is the smallest
geodesic ray, satisfying vt ≥ ut , t > 0.
Lemma 3.12. {v̂τ }τ constructed above is a maximal psh test curve.
Proof. The argument is very similar to [29, Proposition 4.6], so we will be again brief.
By the definition of P [·] we see that τ → v̂τ is τ -concave. By the proof of [29, Lemma
4.3] we see that τ → vτ (x) is τ -usc for any x ∈ X.
It remains to argue the maximality of {v̂τ }τ . To start,Rwe fix τ1 ∈ (−∞, τû+ ). Since
vτ ր Vθ as τ → −∞, by [31, Theorem 2.1] we have that X θvnτ0 > 0 for some τ0 < τ1 .
There exists α ∈ (0, 1) such that τ1 = ατ0 + (1 − α)τû+ . By τ -concavity, we have that
v̂τ1 ≥ αv̂τ0 + (1 − α)v̂τ + .
û
By [74, Theorem 1.2] and multilinearity of non-pluripolar products we obtain X θv̂nτ1 ≥

R
αn X θv̂nτ0 > 0. As a result, we can apply [29, Lemma 4.7] to conclude that v̂τ1 = P [v̂τ1 ].
R
Lastly, we address maximality in the case τ := τû+ . If s < τ = τû+ , then by the above we
can write P [v̂τ ] ≤ P [v̂s ] = v̂s . Letting s ր τ = τû+ , we obtain that P [v̂τ ] ≤ v̂τ . Since the
reverse inequality is trivial, we get P [vτ ] = vτ , finishing the proof.
Finally, we note the following identity for the radial Monge–Ampère energy (see (19))
of a finite energy subgeodesic and its maximization. It implies that the Legendre trans-
form {ûτ }τ of a finite energy sublinear subgeodesic ray {ut }t is a finite energy test curve
(recall Definition 3.5(ii)). The reverse correspondence however might not be true.
Proposition 3.13. Let {ut }t be a finite energy sublinear subgeodesic ray and {vt }t be its
maximization constructed in (24). Then we have
Z τ + Z
1
Z
û
n
Iθ {vt } = Iθ {ut } = θûτ − n
θVθ dτ + τû+ . (25)
R τ+ R n
1
dτ +τû+ follows from (20) and the fact
R n
Proof. That Iθ {vt } = vol({θ}) û
θû − θ
X Vθ
R n R n R n−∞ X τ
that X θv̂τ = X θP [ûτ ] = X θûτ ([31, Proposition 2.3 and Remark 2.5]).
To finish the proof, we utilize an alternative construction for the maximization {vt }t .
Namely, for k ∈ N, let {vtk }t be the sublinear subgeodesic ray such that vtk = ut for t ≥ k
and [0, k] ∋ t → vtk ∈ E 1(X, θ) is the finite energy geodesic segment joining Vθ and uk .
By the comparison principle we obtain that {vtk }t is indeed a sublinear subgeodesic ray,
moreover vt ≥ vtk ≥ ut . Since the k-limit of {vtk }t is the smallest finite energy geodesic
ray dominating {ut }t , we obtain that vtk ր vt . Hence Iθ {vt } ≥ Iθ {ut } = limk Iθ {vtk } ≥
limk Iθ (v1k ) = Iθ (v1 ) = Iθ {vt }, where we used Iθ {ut } = Iθ {vtk } for all k, t 7→ Iθ (vtk ) is
convex and t 7→ Iθ (vt ) is linear. So Iθ {vt } = Iθ {ut }, finishing the proof.
15
4 Delta invariant and geodesic semistability
In this section we prove Theorem 1.4 and Theorem 1.5. Let µ := eχ−ψ be the tame
measure defined in (6) and fix λ ∈ (0, cµ [Vθ ]).
Our first result gives a precise formula for the slope of the λ-Ding functional (see (7))
along subgeodesic rays, rooted in the ideas of [37, Section 4]:
Theorem 4.1 (Theorem 1.4). Let {ut }t ⊂ E 1 (X, θ) be a sublinear subgeodesic ray. Then
Dµλ (ut )
Z
lim inf = −Iθ {ut } + sup{τ : e−λûτ dµ < ∞} = −Iθ {ut } + sup{τ : cµ [ûτ ] ≥ λ}.
t→∞ t X
Proof. By basic properties of the Legendre transform, supX (ut t −Vθ ) ր τû+ , as t → ∞.
Hence, after adding a t-linear term to ut we can assume that supX (ut t −Vθ ) ր τû+ = 0. In
particular, t 7→ ut is t-decreasing.
Comparing (19) and the definition Dµλ , it is enough to show that
−1
Z Z
−λut
lim inf log e dµ = τD := sup{τ : e−λûτ dµ < ∞}.
t→∞ λt X X
We
R −λVnote that τD > −∞ by Theorem 2.2, since ûτ ր Vθ in L1 as τ → −∞ and
X
e θ −ψ ω n < ∞ by our assumption
R −λû on λ.
Let < τD and C := X e
R τ −λ(u
τ
dµ < ∞. For any t ≥ 0 we have ûτ ≤ ut − τ t, so
t −τ t) −1
R −λu
C ≥ Xe dµ. As a result, lim inf t→∞ λt log X e t dµ ≥ τ , hence
−1
Z
lim inf log e−λut dµ ≥ τD .
t→∞ λt X
We fix p < lim inf t→∞ −1 log X e−λut dµ and ǫ > 0

R
Now we prove the reverse inequality. λt
satisfying p + ε < lim inf t→∞ −1 log X e−λut dµ. We can find t0 > 0 such that
R
λt
Z
e−λut dµ < e−(p+ε)λt , t ≥ t0 .
X
R∞
epλt X e−λut dµ dt < ∞. By Fubini–Tonelli this is equivalent to
R
Hence 0
Z Z ∞
eλ(pt−ut ) dt dµ < ∞ . (26)
X 0
Before proceeding further, we notice that the following estimate holds, using the fact
that ut ≤ Vθ + t supX (ut − Vθ ) ≤ Vθ ≤ 0:
−1
Z
p < lim inf log e−λut dµ ≤ 0.
t→∞ λt X
As a result, ûp = inf(ut − pt) is not identically equal to −∞, since τu+ = 0. Let x ∈ X
such that ûp (x), χ(x) and ψ(x) are finite. By definition of ûp , we can find t0 > 0 so that
ûp (x) + 1 ≥ ut0 (x) − pt0 . Since t 7→ ut (x) is decreasing, we have ûp (x) − p + 1 ≥ ut (x) − pt
for t ∈ [t0 , t0 + 1]. Hence
Z ∞ Z t0 +1
λ(pt−ut (x))+χ(x)−ψ(x)
e dt ≥ eλ(pt−ut (x))+χ(x)−ψ(x) dt
0 t0
Z t0 +1
≥ e−λûp (x) eλ(p−1)+χ(x)−ψ(x) dt ≥ eλ(p−1) e−λûp (x)+χ(x)−ψ(x) .
t0
e−λûp dµ < ∞, hence p ≤ τD , concluding the proof.

R
By this and (26) we obtain that X
16
Motivated by the above result, we introduce
Dµλ (ut )
Dµλ {ut } := lim inf ,
t→∞ t
and we will call this expression the radial λ-Ding functional of the finite energy sublinear
subgeodesic ray {ut }t .
Remark 4.2. In the Fano case, when the geodesic ray {ut }t is induced by a filtration of
the section ring, D λ{ut } actually coincides with the βλ -invariant introduced in [75, §4].
Next, we move on to prove Theorem 1.5. We start with the following preliminary
estimate regarding the δµ invariant defined in (9):
Lemma 4.3. We have cµ [Vθ ] ≥ δµ .
Proof. Let π : Y → X be a smooth bimeromorphic model over X and E be a prime
divisor in Y . Recall that τθ (E) = sup{τ > 0 s.t. {π ∗ θ} − τ {E} is a big class} is the
pseudo-effective threshold of E. Equivalently, one has
τθ (E) = sup ν(u, E). (27)

u∈PSH(X,θ)
Let hE be a smooth hermitian metric on OY (E), with canonical section SE and

Θ(hE ) := −ddc log hE . We notice that
[Vπ∗ θ−tΘ(hE ) ] = [Vπ∗ θ − t log |SE |2hE ] = [Vθ ◦ π − t log |SE |2hE ], t ∈ [0, ν(Vθ , E)].
This follows from the fact that the map U 7→ U − t log |SE |2hE gives a monotone increasing
bijection between PSH(Y, π ∗ θ) and PSH(Y, π ∗ θ − tΘ(hE )) for t ∈ [0, ν(Vθ , E)], hence
preserves potentials of minimal singularity type.
As a result, using [74, Theorem 1.2], for t ∈ [0, ν(Vθ , E)] we get that
Z
vol({π θ} − t{E}) = (π ∗ θ − tΘ(hE ) + ddc Vπ∗ θ−tΘ(hE ) )n
∗
ZY Z
∗ c 2 n
= (π θ − tΘ(hE ) + dd (Vπ∗ θ − t log |SE |hE )) = θVnθ ,
Y X
where in the last line we used Θ(hE ) + ddc log |SE |2hE = 0 on Y \ E. From (27) we get
that ν(Vθ , E) ≤ τθ (E). Using this and the above identity we obtain that
Z τθ (E)
1
Sθ (E) := vol({π ∗ θ} − x{E})dx ≥ ν(Vθ , E). (28)
vol({θ}) 0
Using Theorem 2.3, this allows to conclude: cµ [Vθ ] = inf E ν(Vθ ,E)
≥ inf E Sθ (E)
= δµ .
π
Let E ⊂ Y − → X be a prime divisor over X. We shall make use of the following
bounded test curve naturally associated to E, going back to [65]:


 Vθ , τ ≤ 0;
v E ,

0 < τ < τθ (E);
τ
ûE
τ := E
(29)


 limτ րτθ (E) vτ , τ = τθ (E);
−∞, τ > τθ (E),

17

where vτE is given by vτE := sup ϕ ∈ PSH(X, θ)ν(ϕ, E) ≥ τ, ϕ ≤ 0 . Using basic

properties of Lelong numbers, one can show that vτE ∈ PSH(X, θ) and that
ν(vτE , E) ≥ τ. (30)
The test curve {ûE
τ }τ is actually maximal, but we do not need this in what follows.
Let {uE }
t t be the finite energy (sub)geodesic ray associated to the test curve {ûE
τ }τ .
We then have the following observation.
Lemma 4.4. We have Iθ {uE
t } = Sθ (E).
Proof. Let hE be a smooth hermitian metric on O(E), with canonical section SE and
Θ(hE ) := −ddc log hE . Using (30) we notice the following identity of singularity types:
[π ∗ vEτ − τ log |SE |2hE ] = [Vπ∗ θ−τ Θ(hE ) ], τ ∈ [0, τθ (E)].
This follows from the fact that the map U 7→ U −τ log |SE |2hE gives a monotone increasing
bijection between {w ∈ PSH(Y, π ∗ θ), ν(w, E) ≥ τ } and PSH(Y, π ∗ θ − τ Θ(hE )), hence
preserves qpsh potentials of minimal singularity type in each set.
Since Θ(hE ) + ddc log |SE |2hE = 0 on Y \ E, we have that
(π ∗ θ − τ Θ(hE ) + ddc (π ∗ vEτ − τ log |SE |2hE ))n = (π ∗ θ + ddc π ∗ vEτ )n .

As a result, using [74, Theorem 1.2], we notice that for τ < τθ (E) we have
Z Z
n
θûEτ = π ∗ θπn∗ vτE = vol({π ∗ θ} − τ {E}). (31)
X Y
That Iθ {uE
t } = Sθ (E), follows after comparing (1) and (25).
Next we introduce a new valuative invariant attached to certain test curves. Given
a sublinear subgeodesic ray {ut }t ⊂ E 1 (X, θ), it follows from (25) that the Legendre
transform {ûτ }τ of {ut }t is a finite energy test curve. For any prime divisor E over X,
we define the expected Lelong number ν({ûτ }τ , E) of the test curve {ûτ }τ along E to be
(
ν(ûIθ {ut } , E), if Iθ {ut } < τû+ ,
ν({ûτ }τ , E) :=
ν(Vθ , E), if Iθ {ut } = τû+ .
The following estimate for the expected Lelong number will play a key role, and should
be compared with [50, Lemma 2.2].
Proposition 4.5. For any sublinear subgeodesic ray {ut }t ⊂ E 1 (X, θ) and any prime
divisor E over X we have
ν({ûτ }τ , E) ≤ Sθ (E).
Proof. Put f (τ ) := ν(ûτ , E) for τ ∈ (−∞, τû+ ). Then it is clear that f (τ ) is a non-
negative convex non-decreasing function defined on (−∞, τû+ ). So in particular we may
set f (−∞) := limτ ց−∞ f (τ ). When Iθ {ut } = τû+ , the desired inequality is ν({ûτ }, E) =
ν(Vθ , E) ≤ Sθ (E), which is (28).
So in what follows we can assume that Iθ {ut } < τû+ . We first consider the case when
f (−∞) = f (Iθ {ut }). This means that f (τ ) = a ∈ R, for τ ∈ (−∞, Iθ {ut }]. Namely,
ν(ûτ , E) ≡ a for τ ∈ (−∞, Iθ {ut }]. So by [74, Theorem 1.2] and (31) we have
Z Z
n
θûτ ≤ θvnaE = vol({π ∗ θ} − a{E}) for τ ∈ (−∞, Iθ {ut }],
X X
18
where vaE is defined below (29). Thus, by (25)
Z Iθ {ut }
1
Iθ {ut } ≤ vol({π θ} − a{E}) − vol({θ}) dτ + τû+ ,
∗
vol({θ}) −∞
which forces that vol({π ∗ θ} − a{E}) = vol({θ}) since {ûτ }τ has finite energy. So we have
Z τθ (E) Z a
1 ∗ 1
Sθ (E) = vol({π θ} − x{E})dx ≥ vol({π ∗ θ} − a{E})dx = a,
vol({θ}) 0 vol({θ}) 0
what we aimed to prove. Thus, we can assume that f (−∞) < f (Iθ {ut }). Put
f (Iθ {ut }) − f (Iθ {ut } − h)
b := f−′ (Iθ {ut }) := lim+ ,
h→0 h
which has to be a finite positive number by the convexity of f . Now we put
(
0, τ ∈ (−∞, Iθ {ut } − b−1 f (Iθ {ut })),
g(τ ) :=
b(τ − Iθ {ut }) + f (Iθ {ut }), τ ∈ [Iθ {ut } − b−1 f (Iθ {ut }), τû+ ].
Due to convexity, we have f (τ ) ≥ g(τ ), τ ∈ (−∞, τû+ ). By [74, Theorem 1.2] and (31),
Z Z
n
θûτ ≤ θvnE = vol({π ∗ θ} − g(τ ){E}) for τ ∈ (−∞, τû+ ).
g(τ )
X X
Thus using (25) we have that

τû+
1
Z
Iθ {ut } ≤ vol({π θ} − g(τ ){E}) − vol({θ}) dτ + τû+
∗
vol({θ}) −∞
τû+
1
Z
≤ vol({π θ} − g(τ ){E}) − vol({θ}) dτ + τû+
∗
vol({θ}) Iθ {ut }−b−1 f (Iθ {ut })
τθ (E)
1
Z
≤ vol({π θ} − x{E})dx + Iθ {ut } − b−1 f (Iθ {ut })
∗
vol({θ})b 0
= b−1 (Sθ (E) − f (Iθ {ut}) + Iθ {ut },
Using the above inequality, since b > 0, we finally arrive at ν({ûτ }τ , E) = ν(ûIθ {ut } , E) =
f (Iθ {ut }) ≤ Sθ (E), finishing the proof.
Inspired by the above, for any sublinear subgeodesic ray {ut }t ⊂ E 1 (X, θ) we put
(
cµ [ûIθ {ut } ], if Iθ {ut } < τû+ ,
cµ {ûτ } :=
cµ [Vθ ], if Iθ {ut } = τû+ ,
which is called the expected complex singularity exponent of {ûτ }τ .

We show the following analytic characterization of the δ-invariant that should be
compared with Fujita–Odaka’s basis-divisor formulation [50].
Theorem 4.6. We have
δµ = inf cµ {ûτ },
{ut }
where the inf is over all sublinear subgeodesic rays {ut }t ⊂ E 1 (X, θ).
19
This result implies that δµ > 0. Indeed, we have δµ ≥ inf u∈PSH(X,θ) cµ [u]. By strong
openness [52] the tame measure µ has Lp density for some p > 1. So Hölder’s inequality
and [41, 69] implies that cµ [u] ≥ p−1
p ω
c n [u] > α for some uniform α > 0.
Proof. Given {ut}t , if Iθ {ut } = τû+ , then by Lemma 4.3 cµ {ûτ } = cµ [Vθ ] ≥ δµ . If
Iθ {ut } < τû+ , then Theorem 2.3 and Proposition 4.5 imply that

cµ {ûτ } = cµ [ûIθ {ut } ] = inf ≥ inf = δµ .
E ν(ûIθ {ut } , E) E Sθ (E)
So we obtain that inf {ut } cµ {ûτ } ≥ δµ .

To see the reverse direction, let E be any prime divisor over X and let {ûE τ }τ be
the bounded test curve constructed in (29). Then by Lemma 4.4 we have Iθ {uE t } =
Sθ (E), and τû+E = τθ (E). If Iθ {uE t } = τ +
ûE
, i.e., S θ (E) = τθ (E), then û E
τ ∈ E(X, θ)
for all τ ∈ (−∞, τθ (E)) in light of (25). So [32, Theorem 1.1] and (30) imply that
ν(Vθ , E) = ν(ûEτ , E) ≥ τ for all τ ∈ (−∞, τθ (E)), and hence ν(Vθ , E) = τθ (E) = Sθ (E).
So by Theorem 2.3 again,
cµ {ûE
τ } = cµ [Vθ ] ≤ = .
ν(Vθ , E) Sθ (E)
+
If Iθ {uE E
t } < τûE , i.e., Sθ (E) < τθ (E), then from (30) we also have ν(ûSθ (E) , E) ≥ Sθ (E)
(this is actually an equality, in view of Proposition 4.5). The we derive that

cµ {ûE E E
τ } = cµ [ûIθ {uE } ] = cµ [ûSθ (E) ] ≤ E
≤ ,
t ν(ûSθ (E) , E) Sθ (E)
by Lemma 4.4 and Theorem 2.3. This implies the reverse direction.
Now we are in the position to prove the following key result.
Theorem 4.7 (=Theorem 1.5). One has
δµ = sup{λ > 0|Dµλ{ut } ≥ 0 for all sublinear subgeodesic ray {ut }t ⊂ E 1(X, θ)}. (32)
Proof. Assume that D λ {ut } ≥ 0 holds for any sublinear subgeodesic ray {ut }Rt ⊂ E 1 (X, θ).
Consider the associated test curve {ûτ }τ . Set for simplicity β := sup{τ : X e−λûτ dµ <
∞}. Then by Theorem 4.1, β − Iθ {ut } = Dθλ {ut } ≥ 0, so that β ≥ Iθ {ut }. This means
that for any τ < Iθ {ut }, cµ [ûτ ] ≥ λ.
If Iθ {ûτ } = τû+ , then cµ {ûτ } = cµ [Vθ ] ≥ cµ [ûτ ] ≥ λ. If Iθ {ûτ } < τû+ , then by Lemma
2.4 we also have cµ {ûτ } = cµ [ûIθ {ut } ] ≥ λ. Taking inf over all such {ut }t then yields
δµ ≥ λ by the previous theorem. Hence, the right hand side of (32) is dominated by δµ .
It remains to argue that for any λ ∈ (0, δµ ), Dµλ {ut } ≥ 0 holds for any sublinear
subgeodesic ray {ut }t ⊂ E 1 (X, θ). To start we note by Proposition 2.5 and Lemma 4.3
that Dµλ (v) is finite for any v ∈ E 1 (X, θ).
Consider the associated test curve {ûτ }τ and put again β := sup{τ : X e−λûτ dµ <
R
∞}. RFirst, if Iθ {ut } = τû+ , then in view of (25) ûτ has full mass for any τ < τû+ , so
that X e−λûτ dµ < ∞ for all τ ∈ (−∞, τû+ ) by Proposition 2.5 and Lemma 4.3. Thus by
Theorem 4.1, Dµλ {ut } = β − Iθ {ut } = τû+ − τû+ = 0, as desired. Next, if Iθ {ut } < τû+ ,
then we can apply Theorem 4.6 to conclude that cµ [ûIθ {ut } ] ≥ δµ > λ, so that β ≥ Iθ {ut }.
Thus Dµλ {ut } = β − Iθ {ut } ≥ Iθ {ut } − Iθ {ut } = 0, finishing the proof.
20
As a consequence of Theorem 4.7 we note the following practical result.
Proposition 4.8. For any λ ∈ (0, δµ ), then there exists ε > 0 such that
Dµλ {wt } ≥ −εIθ {wt }. (33)
for all finite energy geodesic rays {wt }t ⊂ R1 (X, θ) with supX wt = 0. In particular, if
{wt }t is also d1 -unit speed, then Dµλ {wt } ≥ ε.
Proof. Consider Tian’s α-invariant [69] adapted to our setting:
Z
−α(u−supX u)
αµ := sup α > 0 : sup e dµ < ∞ ,
u∈PSH(X,θ) X
which is positive. Indeed, pick A > 0 sufficiently large such that θ + Aω is Kähler, so that
PSH(X, θ) ⊂ PSH(X, θ + Aω). The positivity of αµ follows from [8, Proposition 1.4].
Fix η ∈ (λ, δµ ) and α ∈ (0, min{λ, αµ }). By Hölder’s inequality, for all w ∈ E 1 (X, θ)
we have
1 λ−α η−λ
Z Z Z
−λw −ηw
− log e dµ ≥ − log e dµ − log e−αw dµ.
λ X λ(η − α) X λ(η − α) X
Therefore, one can find C > 0 such that for all w ∈ E 1 (X, θ) with supX w = 0 we have
η(λ − α) η α(η − λ)
Dµλ (w) ≥ Dµ (w) − Iθ (w) − C. (34)
λ(η − α) λ(η − α)
Plutting wt in the above inequality, dividing with t and letting t → ∞ we find that
η(λ − α) η α(η − λ)
Dµλ {wt } ≥ Dµ {wt } − Iθ {wt }.
λ(η − α) λ(η − α)
So we conclude (33) by Theorem 4.7. In case {wt }t is unit speed, we have that 1 =
d1 (Vθ , w1 ) = −Iθ (w1 ). By linearity of t →
7 Iθ (wt ) the last statement also follows.
5 Kähler–Einstein metrics with positive twist

In this section we return to the assumptions of the introduction, that c1 (X) admits a
decomposition c1 (X) = {θ}+{η}, where η is a smooth representative of a pseudoeffective
class and ψ ∈ PSH(X, η). In particular, −KX is bigR (thus X is projective). Further
assume that η + ddc ψ ∈ {η} has klt singularities, i.e., X e−ψ ω n < ∞.
As pointed out in the introduction, the twisted KE equation Ric(θu ) = θu +ηψ reduces
to the scalar equation
θun = ef −u−ψ ω n , (35)
with f ∈ C ∞ (X) a Ricci potential satisfying Ric ω = θ + η + ddc f. So we choose our tame
measure µ to be
µ := ef −ψ ω n .
The above equation does not always admit a solution. Setting up a continuity method,
we are led to study the solvability of the following more general KE type equations:
θun = e−λu dµ (36)
21
for λ > 0. To make sense of the above equation, we need to further assume that cµ [Vθ ] > λ,
a condition that will later follow from our algebraic existence criteria. Due to Proposition
2.5 we have cµ [v] > λ for any v ∈ E 1 (X, θ), hence the λ-twisted Ding energy corresponding
to (36) can be defined as follows:
1
Z
λ
Dµ (u) := − log e−λu dµ − Iθ (u) for u ∈ E 1 (X, θ).
λ X
We say Dµλ is coercive/proper if there exists ε > 0 and C > 0,
Dµλ(u) ≥ ε(sup u − Iθ (u)) − C

X
holds for all u ∈ E 1 (θ). As is well known (cf. [7, 8]), properness of the twisted Ding
energy Dµλ implies that (36) admits a solution:
Proposition 5.1. If Dµλ is proper then Dµλ has a global minimizer in E 1 (X, θ). Any such
minimizer is a solution to (36), up to an additive constant.
Proof. Let ui ∈ E 1 (X, θ) with supX ui = 0 be a minimizing sequence:
lim Dµλ (ui ) = inf Dµλ (u).

i→∞ u∈E 1 (X,θ)
The properness condition implies that we can find C > 0 such that Iθ (ui ) ≥ −C for all
i. Then up to a subsequence one can find u∞ ∈ E 1 (X, θ) such R that ui → Ru∞ in L1 . Note
λ 1 1 1 −λϕ
that Dµ is L -lsc on E (X, θ), as Iθ is L -usc and the term X e dµ = X ef −λϕ−ψ ω n is
L1 -continuous (by openness [12, 52] and [41, Theorem 0.2(2)]). So we find that u∞ is a
minimizer of Dµ1 .
We now proceed to show that u∞ solves (35). Indeed, for any f ∈ C 0 (X, R), we have
−1
Z
λ λ
Dµ (u∞ ) ≤ D (Pθ (u∞ + tf )) ≤ log e−λ(u∞ +tf ) dµ − Iθ (Pθ (u∞ + tf ))
λ X
for all t ∈ R. So the expression −1 log X e−λ(u∞ +tf ) dµ − Iθ (Pθ (u∞ + tf )) is minimized at
R
λ
t = 0. By [7, Lemma 4.2] we can take derivaties and conclude that
f e−λu∞ dµ f θun∞
R R
X X
R =
X
e−λu∞ dµ vol({θ})
n
θu
Re
−λu∞ dµ
for any f ∈ C 0 (X, R). This implies that e−λu∞ dµ
= ∞
vol({θ})
, finishing the argument.
X
We arrive at the main result of this section:

Theorem 5.2. Suppose that cµ [Vθ ] > 1 and that Dµ1 is not proper. Then δµ ≤ 1.
A direct consequence is the following:
Corollary 5.3 (=Theorem 1.2). If δµ > 1 then Dµ1 is proper. In particular, (35) admits
a solution u ∈ PSH(X, θ) with minimal singularity type.
Proof. From Lemma 4.3 we obtain that cµ [Vθ ] > 1. By the previous result Dµ1 is proper.
Thus, by Proposition 5.1, (35) admits a solution u ∈ E 1 (X, θ). That u has minimal
singularity type follows from Kolodzej’s estimate (see [17, Theorem B]), as the right
hand side of (35) has Lp density for some p > 1 by Theorem 2.2.
22
Proof of Theorem 5.2. Since −KX is big under our conditions, we have that X is Moishe-
zon [56], hence also projective [63]. Also, the condition cµ [Vθ ] > 1 implies that cµ [v] > 1
for any v ∈ E 1 (X, θ) (Proposition 2.5). As a result, given a finite energy geodesic t 7→ vt ,
both conditions of [13, Theorem 0.1] are satisfied, and we obtain that t 7→ Dµ1 (vt ) is
convex.
Fixing β ∈ (0, 1), we introduce the following auxiliary Ding energy, for u ∈ E 1 (X, θ):
Z
1
Dβ (u) := Dµ (u) − β(sup u − Iθ (u)) = − log e−ϕ dµ − (1 − β)Iθ (ϕ) − β sup u
X X X
Notice that Dβ is still convex along finite energy geodesic segments t 7→ ut emanating
from Vθ , as t 7→ supX ut = supX (ut − Vθ ) is linear (see Remark 3.3 and [29, (13)]).
Moreover, both Dµ1 and Dβ are L1 -lsc, as shown in the proof of Proposition 5.1.
Since Dµ1 is not proper, there exists ϕj ∈ E 1 (X, θ) with supX ϕj = 0 satisfying
1
Dµ1 (ϕj ) ≤ d1 (Vθ , ϕj ) − j.
j
We claim that d1 (Vθ , ϕj ) → ∞. Indeed, suppose that d1 (Vθ , ϕj ) = −Iθ (ϕj ) ≤ C for all j.
Since Iθ is L1 -usc, after possibly passing to a subsequence, one can find ϕ∞ ∈ E 1 (X, θ)
with ϕj → ϕ∞ in the L1 topology. Using that Dµ1 is L1 -lsc, we deduce −∞ < Dµ1 (ϕ∞ ) ≤
lim inf j Dµ1 (ϕj ) = −∞, a contradiction. This proves the claim.
As a result, for some ε > 0 there exists ϕj ∈ E 1 (X, θ) such that supX ϕj = 0,
Dβ (ϕj )
≤ −ε and d1 (Vθ , ϕj ) → ∞.
d1 (Vθ , ϕj )
Let [0, d1 (Vθ , ϕj )] ∋ t 7→ ujt ∈ E 1 (X, Vθ ) be the unit speed finite energy geodesic
joining Vθ , ϕj . Due to convexity of Dβ along such geodesic segments we have that
Dβ (ujt )
≤ −ε, d1 (Vθ , ujt ) = −Iθ (ujt ) = t for all t ∈ [0, d1(Vθ , ϕj )].
t
Since supX (ujt − Vθ ) = supX ujt = 0, using a diagonal argument, after possibly picking
a subsequence, there exists a finite energy sublinear subgeodesic ray {ut }t such that
ujt (x) → ut (x) with respect to L1loc ((0, ∞) × X).
By Fubini’s theorem we obtain that ujt converges to ut with respect to L1 (X), for all
t ∈ (0, ∞) \ E, where E is a set of Lebesgue measure zero. Thus
Dβ (ut )
≤ −ε, sup ut = 0, 0 ≥ Iθ (ut ) ≥ −t, for all t ∈ (0, ∞) \ E, (37)
t X
as Dβ is L1 -lsc, supX is L1 -continuous and Iθ is L1 -usc.

Let lj ∈ (0, ∞) \ E be a sequence converging to some t ∈ (0, ∞). Due to t-convexity
of {ut }t , and the fact that supX ulj = 0, we must have that ulj L1 -converges to ut . As a
result, we have that (37) must hold for all t ∈ (0, ∞).
In addition, due to [29, Theorem 3.11], we see that ut → Vθ in L1 as t → 0, as we
have d1 (ut , Vθ ) = −Iθ (ut ) → 0 as t → 0.
By the definition of Dβ , Theorem 4.1 implies that
Dβ {ut } = −(1 − β)Iθ {ut } + sup{τ ∈ R : cµ [ûτ ] ≥ 1} ≤ −ε.
23
Now let {vt }t ⊂ E 1 (X, θ) be the maximized ray of {ut }t (see (24)). Note that v̂τ =
P [ûτ ], hence by Proposition 2.5 we have that cµ [v̂τ ] = cµ [ûτ ] for all τ < 0. Since supX vt =
0 and Iθ {vt } = Iθ {ut } (by (25)), we obtain that
Dβ {vt } = −(1 − β)Iθ {vt } + sup{τ ∈ R : cµ [v̂τ ] ≥ 1} = Dβ {ut } ≤ −ε. (38)
We conclude that {vt }t can not be a trivial ray, meaning that Iθ {vt } < 0. Indeed, suppose
otherwise that Iθ {vt } = 0 (i.e., vt = Vθ , t ≥ 0). We obtain that v̂τ = Vθ for all τ < 0. So
cµ [vτ ] = cµ [Vθ ] > 1 for all τ < 0, which is absurd in light of (38).
Let {wtβ }t be the unit speed reparamaterization of {vt }t (i.e., Iθ {wtβ } = −1). We still
have that Dβ {wtβ } ≤ 0. By the definition of Dβ , we obtain that Dµ1 {wtβ } ≤ β. Since β
can be arbitrary small, we conclude that δµ ≤ 1 by Proposition 4.8.
The (partial) converse. In this paragraph we give a (partial) converse for Corollary
5.3, and point out the role of uniqueness in a full converse. First we show the converse
to Proposition 5.1, following [7, Theorem 6.6].
Proposition 5.4. If u ∈ E 1 (X, θ) is a solution to (35) then u minimizes Dµ1 .
Proof. Given u, v ∈ E 1 (X, θ), let [0, 1] ∋ t 7→ ut ∈ E 1(X, θ) be the geodesic segment
connecting u0 = u and u1 = v, constructed in [29]. If u satisfies θun = e−u µ then
cµ [Vθ ] > 1 by openness [12, 52], so Dµ1 is well defined on E 1 (X, θ) by Proposition 2.5. We
introduce
f (t) := Dµ1 (ut ), t ∈ [0, 1].
By d1 -continuity of the Ding energy and [13, Theorem 0.1], f is continuous and convex.
We want to show that Dµ1 (u) ≤ Dµ1 (v). Note that u has minimal singularity by [17].
We can assume that v also has minimal singularity type, since such potentials are d1 -dense
in E 1 (X, θ). By the above considerations, it is enough to argue that

d
f (t) ≥ 0.
dt t=0+
After adding a constant to v, we may further assume that u > v on X. Consider
ut − u
g(t) := for t ∈ (0, 1].
t
Since u, v have minimal singularity type, t → ut is uniformly t-Lipschitz. As a result,
limt→0 g(t) =: u̇+
0 is uniformly bounded and well defined away from a pluripolar set. We
notice that (by [29, Theorem 2.4(iii)])
Iθ (ut ) − Iθ (u) 1 1
Z Z
n
≤ g(t)θu = g(t)e−u dµ.
t vol(θ) X vol(θ) X
Letting t → 0 by monotone convergence we get

d 1
Z
Iθ (ut ) ≤ u̇+
0e
−u
dµ. (39)
dt + t=0 vol(θ) X
On the other hand, using dominated convergence we conclude that

R −u R −u
e dµ − e dµ u − ut eu−ut − 1 −u
t
Z Z
lim X X
= lim e dµ = (−u̇+
0 )e
−u
dµ. (40)
tց0 t tց0 X t u − ut X
Putting together (39) and (40), we conclude that dtd t=0+ Dµ1 (ut ) ≥ 0, finishing the proof.

24
We record a partial converse to Corollary 5.3:
Proposition 5.5. Assume that (35) is solvable. Then δµ ≥ 1.
Proof. As follows from the previous result (and its argument), Dµ1 is well defined and
bounded from below on E 1(X, θ). This gives δµ ≥ 1 in light of Theorem 1.5.
Remark 5.6. If the twisted KE metric found in Corollary 5.3 is unique, as conjectured
in [11, page 4], then a full converse is possible: δµ > 1 if and only if there exists a unique
twisted KE metric. Indeed, if a unique twisted KE metric exists then the conditions of the
properness/existence principle of the first named author and Y. Rubinstein [35, Theorem
3.7] (c.f. [27, Theorem 4.7]) are satisfied, implying that the Ding functional Dµ1 is proper.
A ‘thermodynamic’ argument (as in [8, Proposition 4.11] and [78, §3]) then gives that
Dµ1+ε is proper as well, for ε > 0 small (see [36, §3] for precise details). Hence, Theorem
1.5 implies that δµ > 1.
6 The case of log-Fano pairs

A log Fano pair (Z, ∆) consists of the following data: Z is a normal projective variety
and ∆ an effective Weil divisor on Z such that −KZ − ∆ is an ample Q-Cartier divisor
and that (Z, ∆) has klt singularities. In what follows we give a simplified proof of the
following result due to Li–Tian–Wang [61], using the techniques of our paper:
Theorem 6.1. The log Fano pair (Z, ∆) admits a KE metric if it is uniformly K-stable.
Put for simplicity L := −KZ − ∆. Then the pair (Z, ∆) being uniformly K-stable
means exactly that δ∆ (L) > 1 (cf. [14]), where
AZ,∆ (E)
δ∆ (L) := inf .
E SL (E)
π
Here the infimum is over all prime divisors E ⊂ Y − → Z over Z, where Y is a smooth
projective manifold and π is a projective birational morphism. The log discrepancy is
AZ,∆ (E) := 1 + ordE (KY + π ∗ L) and the expected vanishing order SL (E) is given by
τL (E)
1
Z
SL (E) := vol(π ∗ L − xE)dx,
vol(L) 0
with τL (E) := sup{x > 0 : π ∗ L − xE big} being the pseudoeffective threshold. The
volume function vol(·) here is understood in the sense of algebraic geometry (cf. [57]).
To prove that (Z, ∆) admits a KE metric, we use the variational framework in [9, 61].
The goal is to argue that the corresponding Ding functional D defined in [61, (40)] for the
pair (Z, ∆) is proper. Suppose on the contrary that D is not proper. With this assumption
Li–Tian–Wang were able to construct a unit-speed geodesic ray {ϕt }t in E 1(Z, L) such
t)
that D{ϕt } := limt→∞ D(ϕ t
≤ 0 (see [61, §4.1.1]). To derive a contradiction out of this,
the Li–Tian–Wang argument takes four more involved steps (see [61, §4.2-4.5]).
Relying on the results in this paper we can obtain a contradiction immediately. Indeed,
σ
take a log resolution X −→ Z of the pair (Z, ∆), so in particular X is a smooth birational
model over Z. We take our big class {θ} to be c1 (σ ∗ L) and let µ = eχ−ψ ω n be the tame
measure on X that is associated to the log Fano pair (Z, ∆) (see [8, Lemma 3.2(i)], with
25
χ := ψ + , ψ := ψ − and ω n := dV for some Kähler form ω on X). Then we consider the
following algebraic δ-invariant:
Aχ,ψ (E)
δ̃µ (σ ∗ L) := inf ,
E Sσ∗ L (E)
where the inf is over all prime divisors in smooth birational models over X. Note that this
invariant is potentially different from the δµ invariant defined in (9), as the latter involves
inf over all smooth bimeromorphic models over X. However due to [9, Theorem B.7] and
the non-pluripolar volume being the same as the algebraic volume in the Nerón–Severi
space (see [15] and [17, Proposition 1.18]), we find that all the previous arguments in this
paper hold without change when restricting to divisors in smooth birational models over
X. Hence, Theorem 1.5 (or simply Theorem 4.6) gives that
δµ ({θ}) = δ̃µ (σ ∗ L).
Moreover, as any prime divisor E over X is a prime divisor over Z, one has Aχ,ψ (E) =
AZ,∆ (E) and Sσ∗ L (E) = SL (E). So we find that δµ ({θ}) ≥ δ∆ (L). (They are actually
equal, as by Hironaka’s theorem any prime divisor over Z can be viewed as a prime divisor
over X as well.) Note further that we can also choose the background data θ ∈ c1 (σ ∗ L)
and µ appropriately so that
D(ϕ) = Dµ1 (σ ∗ ϕ) for any ϕ ∈ E 1 (Z, L).
Now, (Z, ∆) being uniformly K-stable implies that δµ ({θ}) ≥ δ∆ (L) > 1. So we derive
from Proposition 4.8 that D{ϕt } = Dµ1 {σ ∗ ϕt } ≥ ε > 0 for some fixed ε > 0 as {σ ∗ ϕt }t is
a unit-speed geodesic ray in E 1 (X, θ). This immediately gives us a contradiction, hence
proving Theorem 6.1.
References
[1] H. Abban and Z. Zhuang. K-stability of Fano varieties via admissible flags. Forum Math. Pi, 10:Pa-
per No. e15, 2022.
[2] S. Bando and T. Mabuchi. Uniqueness of Einstein Kähler metrics modulo connected group actions,
Algebraic geometry, Sendai, 1985. Volume 10, Adv. Stud. Pure Math. Pages 11–40. North-Holland,
Amsterdam, 1987.
[3] E. Bedford and B. A. Taylor. A new capacity for plurisubharmonic functions. Acta Mathematica,
149:1–40, 1982.
[4] E. Bedford and B. A. Taylor. The dirichlet problem for a complex Monge–Ampère equation. Inven-
tiones mathematicae, 37(1):1–44, 1976.
[5] R. J. Berman. K-polystability of Q-Fano varieties admitting Kähler–Einstein metrics. Inventiones
mathematicae, 203(3):973–1025, 2016.
[6] R. J. Berman. Measure preserving holomorphic vector fields, invariant anti-canonical divisors and
Gibbs stability. Anal. Math., 48(2):347–375, 2022.
[7] R. J. Berman, S. Boucksom, V. Guedj, and A. Zeriahi. A variational approach to complex monge-
ampère equations. Publications mathématiques de l’IHÉS :1–67, 2013.
[8] R. J. Berman, S. Boucksom, P. Eyssidieux, V. Guedj, and A. Zeriahi. Kähler-Einstein metrics and
the Kähler-Ricci flow on log Fano varieties. Journal für die reine und angewandte Mathematik
(Crelles Journal), 751:27–89, 2019.
[9] R. J. Berman, S. Boucksom, and M. Jonsson. A variational approach to the Yau-Tian-Donaldson
conjecture. Journal of the American Mathematical Society, 34(3):605–652, 2021.
26
[10] R. J. Berman and H. Guenancia. Kähler-Einstein metrics on stable varieties and log canonical pairs.
Geometric and Functional Analysis, 24(6):1683–1730, 2014.
[11] B. Berndtsson. A Brunn-Minkowski type inequality for Fano manifolds and some uniqueness theo-
rems in Kähler geometry. Inventiones mathematicae, 200(1):149–200, 2015.
[12] B. Berndtsson. The openness conjecture and complex Brunn-Minkowski inequalities, Complex ge-
ometry and dynamics. Volume 10, Abel Symp. Pages 29–44. Springer, Cham, 2015.
[13] B. Berndtsson and M. Păun. Bergman kernels and the pseudoeffectivity of relative canonical bun-
dles. Duke Mathematical Journal, 145(2):341–378, 2008.
[14] H. Blum and M. Jonsson. Thresholds, valuations, and K-stability. Adv. in Math., 365:107062, 57,
2020.
[15] S. Boucksom. On the volume of a line bundle. International Journal of Mathematics, 13(10):1043–
1063, 2002.
[16] S. Boucksom. Singularities of plurisubharmonic functions and multiplier ideals. http://sebastien.
boucksom.perso.math.cnrs.fr/notes/L2.pdf, 2017.
[17] S. Boucksom, P. Eyssidieux, V. Guedj, and A. Zeriahi. Monge–Ampère equations in big cohomology
classes. Acta mathematica, 205(2):199–262, 2010.
[18] S. Boucksom, C. Favre, and M. Jonsson. Valuations and plurisubharmonic singularities. Publications
of the Research Institute for Mathematical Sciences, 44(2):449–494, 2008.
[19] S. Boucksom, T. Hisamoto, and M. Jonsson. Uniform K-stability, Duistermaat–Heckman measures
and singularities of pairs. Annales de l’Institut Fourier, 67(2):743–841, 2017.
[20] S. Boucksom and M. Jonsson. A non-Archimedean approach to K-stability, II: divisorial stability
and openness, 2022. arXiv:2206.09492.
[21] S. Boucksom and M. Jonsson. A non-Archimedean approach to K-stability, 2018. arXiv:1805.11160.
[22] E. Calabi. On Kähler manifolds with vanishing canonical class. Algebraic geometry and topology. A
symposium in honor of S. Lefschetz. Volume 12. 1957, pages 78–89.
[23] I. Cheltsov, J. Park, and C. Shramov. Delta invariants of singular del Pezzo surfaces. Journal of
Geometric Analysis, 31(3):2354–2382, 2021.
[24] I. Cheltsov and K. Zhang. Delta invariants of smooth cubic surfaces. European Journal of Mathe-
matics, 5(3):729–762, 2019.
[25] X. Chen, S. Donaldson, and S. Sun. Kähler–Einstein metrics on Fano manifolds. I-III. Journal of
the American Mathematical Society, 28(1):183–197, 199–234, 235–278, 2015.
[26] X. Chen, S. Sun, and B. Wang. Kähler-Ricci flow, Kähler-Einstein metric, and K-stability. Geometry
& Topology, 22(6):3145–3173, 2018.
[27] T. Darvas. Geometric pluripotential theory on Kähler manifolds. Advances in complex geometry.
Volume 735. Amer. Math. Soc., 2019, pages 1–104.
[28] T. Darvas. Weak geodesic rays in the space of Kähler potentials and the class E(X, ω0 ). Journal of
the Institute of Mathematics of Jussieu, 16(4):837–858, 2017.
[29] T. Darvas, E. Di Nezza, and C. H. Lu. L1 metric geometry of big cohomology classes. Annales de
l’Institut Fourier, 68(7):3053–3086, 2018.
[30] T. Darvas, E. Di Nezza, and C. H. Lu. Log-concavity of volume and complex Monge-Ampère
equations with prescribed singularity. Mathematische Annalen, 379(1-2):95–132, 2021.
[31] T. Darvas, E. Di Nezza, and C. H. Lu. Monotonicity of non-pluripolar products and complex
Monge–Ampère equations with prescribed singularity. Analysis & PDE, 11(8):2049–2087, 2018.
[32] T. Darvas, E. Di Nezza, and C. H. Lu. On the singularity type of full mass currents in big cohomology
classes. Compositio Mathematica, 154(2):380–409, 2018.
[33] T. Darvas, E. Di Nezza, and C. H. Lu. The metric geometry of singularity types. Journal für die
reine und angewandte Mathematik (Crelles Journal), 2021(771):137–170, 2021.
[34] T. Darvas and C. H. Lu. Geodesic stability, the space of rays and uniform convexity in Mabuchi
geometry. Geometry & Topology, 24(4):1907–1967, 2020.
[35] T. Darvas and Y. Rubinstein. Tian’s properness conjectures and Finsler geometry of the space of
Kähler metrics. Journal of the American Mathematical Society, 30(2):347–387, 2017.
[36] T. Darvas, M. Xia, and K. Zhang. Uniform Ding stability in big classes, 2022. preprint in prepara-
tion.
27
[37] T. Darvas and M. Xia. The closures of test configurations and algebraic singularity types. Advances
in Mathematics, 397:Paper No. 108198, 56, 2022.
[38] V. Datar and G. Székelyhidi. Kähler–Einstein metrics along the smooth continuity method. Geo-
metric and Functional Analysis, 26(4):975–1010, 2016.
[39] J.-P. Demailly. Complex analytic and differential geometry. https://www-fourier.ujf-grenoble.
fr/~demailly/manuscripts/agbook.pdf, 2012.
[40] J.-P. Demailly. Multiplier ideal sheaves and analytic methods in algebraic geometry, School on
Vanishing Theorems and Effective Results in Algebraic Geometry (Trieste, 2000). Volume 6, ICTP
Lect. Notes, pages 1–148. Abdus Salam Int. Cent. Theoret. Phys., Trieste, 2001.
[41] J.-P. Demailly and J. Kollár. Semi-continuity of complex singularity exponents and Kähler-Einstein
metrics on Fano orbifolds. Annales Scientifiques de l’École Normale Supérieure (4), 34(4):525–556,
2001.
[42] R. Dervan. Uniform stability of twisted constant scalar curvature Kähler metrics. International
Mathematics Research Notices, 2016(15):4728–4783, 2016.
[43] R. Dervan and E. Legendre. Valuative stability of polarised varieties, 2020. arXiv:2010.04023, to
appear in Mathematische Annalen.
[44] R. Dervan and J. Ross. K-stability for Kähler manifolds. Math. Res. Lett., 24(3):689–739, 2017.
issn: 1073-2780.
[45] W. Y. Ding. Remarks on the existence problem of positive Kähler-Einstein metrics. Mathematische
Annalen, 282(3):463–471, 1988.
[46] S. Donaldson. Scalar curvature and stability of toric varieties. Journal of Differential Geometry,
62(2):289–349, 2002.
[47] Z. S. Dyrefelt. K-semistability of cscK manifolds with transcendental cohomology class. The Journal
of Geometric Analysis:1–34, 2016.
[48] P. Eyssidieux, V. Guedj, and A. Zeriahi. Singular Kähler–Einstein metrics. Journal of the American
Mathematical Society, 22(3):607–639, 2009.
[49] K. Fujita. A valuative criterion for uniform K-stability of Q-Fano varieties. Journal für die reine
und angewandte Mathematik, 2019(751):309–338, 2019.
[50] K. Fujita and Y. Odaka. On the K-stability of Fano varieties and anticanonical divisors. Tohoku
Mathematical Journal, 70(4):511–521, 2018.
[51] Q. Guan, Z. Li, and X. Zhou. Estimation of weighted l2 norm related to demailly’s strong openness
conjecture, 2016. arXiv: 1603.05733 [math.CV].
[52] Q. Guan and X. Zhou. A proof of Demailly’s strong openness conjecture. Annals of mathematics
(2), 182(2):605–616, 2015.
[53] V. Guedj and A. Zeriahi. Degenerate complex Monge-Ampère equations, volume 26 of EMS Tracts
in Mathematics. European Mathematical Society (EMS), Zürich, 2017. isbn: 978-3-03719-167-5.
[54] P. H. Hiep. The weighted log canonical threshold. C. R. Math. Acad. Sci. Paris, 352(4):283–288,
2014.
[55] T. Hisamoto. Stability and coercivity for toric polarizations. 2016. arXiv:1610.07998.
[56] S. Ji and B. Shiffman. Properties of compact complex manifolds carrying closed positive currents.
J. Geom. Anal., 3(1):37–61, 1993.
[57] R. Lazarsfeld. Positivity in algebraic geometry. I, volume 48 of Ergebnisse der Mathematik und ihrer
Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics. Springer-Verlag, Berlin, 2004.
Classical setting: line bundles and linear series.
[58] C. Li. K-semistability is equivariant volume minimization. Duke Mathematical Journal, 166(16):3147–
3218, 2017.
[59] C. Li. G-uniform stability and Kähler-Einstein metrics on Fano varieties. Inventiones Mathematicae,
227(2):661–744, 2022.
[60] C. Li, G. Tian, and F. Wang. On the Yau-Tian-Donaldson conjecture for singular Fano varieties.
Communications on Pure and Applied Mathematics, 74(8):1748–1800, 2021.
[61] C. Li, G. Tian, and F. Wang. The uniform version of Yau-Tian-Donaldson conjecture for singular
Fano varieties, 2021. arXiv:1903.01215.
28
[62] Y. Liu, C. Xu, and Z. Zhuang. Finite generation for valuations computing stability thresholds and
applications to K-stability. Annals of Mathematics (2), 196(2):507–566, 2022.
[63] B. G. Moishezon. On n-dimensional compact complex manifolds having n algebraically independent
meromorphic functions. I. Izv. Akad. Nauk SSSR Ser. Mat., 30:133–174, 1966.
[64] J. Park and J. Won. K-stability of smooth del Pezzo surfaces. Mathematische Annalen, 372(3-
4):1239–1276, 2018.
[65] J. Ross and D. Witt Nyström. Analytic test configurations and geodesic rays. Journal of Symplectic
Geometry, 12(1):125–169, 2014.
[66] Y. T. Siu. Analyticity of sets associated to Lelong numbers and the extension of closed positive
currents. Inventiones Mathematicae, 27:53–156, 1974. issn: 0020-9910.
[67] G. Székelyhidi. An introduction to extremal Kähler metrics, volume 152 of Graduate Studies in
Mathematics. American Mathematical Society, Providence, RI, 2014.
[68] G. Tian. Kähler–Einstein metrics with positive scalar curvature. Inventiones Mathematicae, 130(1):1–
37, 1997.
[69] G. Tian. On Kähler–Einstein metrics on certain Kähler manifolds with c1 (M ) > 0. Inventiones
mathematicae, 89(2):225–246, 1987.
[70] G. Tian. Canonical metrics in Kähler geometry. Lectures in Mathematics ETH Zürich. Birkhäuser
Verlag, Basel, 2000. Notes taken by Meike Akveld.
[71] G. Tian. K-stability and Kähler-Einstein metrics. Communications on Pure and Applied Mathe-
matics, 68(7):1085–1156, 2015.
[72] G. Tian and F. Wang. On the existence of conic Kähler-Einstein metrics. Advances in Mathematics,
375, 2020.
[73] A. Trusiani. Kähler-Einstein metrics with prescribed singularities on Fano manifolds, preprint, 2020.
arXiv:2006.09130.
[74] D. Witt Nyström. Monotonicity of non-pluripolar Monge–Ampère masses. Indiana University Math-
ematics Journal, 68(2):579–591, 2019.
[75] C. Xu and Z. Zhuang. On positivity of the CM line bundle on K-moduli spaces. Annals of Mathe-
matics (2), 192(3):1005–1068, 2020.
[76] S.-T. Yau. On the Ricci curvature of a compact Kähler manifold and the complex Monge–Ampère
equation, I. Communications on pure and applied mathematics, 31(3):339–411, 1978.
[77] K. Zhang. A quantization proof of the uniform Yau-Tian-Donaldson conjecture. arXiv:2102.02438,
to appear in Journal of the European Mathematical Society.
[78] K. Zhang. Continuity of delta invariants and twisted Kähler–Einstein metrics. Advances in Mathe-
matics, 388:107888, 2021.
[79] K. Zheng. Singular scalar curvature equations, 2022. arXiv: 2205.14726 [math.DG].
Department of Mathematics, University of Maryland

tdarvas@umd.edu
School of Mathematical Sciences, Beijing Normal University
kwzhang@bnu.edu.cn
29

Twisted K Ahler-Einstein Metrics in Big Classes: Tam As Darvas and Kewei Zhang

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Twisted K Ahler-Einstein Metrics in Big Classes: Tam As Darvas and Kewei Zhang

Uploaded by

Copyright:

Available Formats

Twisted Kähler–Einstein metrics in big classes

Tamás Darvas and Kewei Zhang

using a divisorial stability condition. In particular, when −KX is big, we obtain a

Vθ := sup{u ∈ PSH(X, θ) such that u ≤ 0}. (10)

1Tj {ϕj >Vθj −k} θmax(ϕ

It is upper semi-continuous (usc) with respect to the L1 topology on PSH(X, θ).

d1 (u, v) = Iθ (u) + Iθ (v) − Iθ (P (u, v)).

In addition, d1 -convergence implies L1 -convergence of qpsh potentials [29, Theorem 3.11].

2.2 Complex singularity exponents and openness

And our definition of the Lelong number of a qpsh function v at x ∈ X is

Given an irreducible analytic subset E ⊂ X, ν(v, E) is equal to the Lelong number of v

ν(χ, E) + AX (E) ≥ (1 + ε)(ν(ψ, E) + λν(ϕ, E)) ≥ ν(ψ, E) + (1 + ε)λν(ϕ, E)

ν(χ, E) + AX (E) ≥ ν(ψ, E) + (1 + ε)λν(ϕ, E)

again, for any small τ > 0 we have that

ν(χ, E) + AX (E) ≥ (1 + τ )ν(ψ, E)

holds for all E. Summarizing, we find that

(1 + τ )(ν(χ, E) + AX (E)) ≥ (1 + τ + τ 2 )ν(ψ, E) + (1 + ε)λν(ϕ, E).

Lemma 2.4. Let {ψτ }τ ∈I be a family of qpsh functions on X, where I ⊂ R is some

3 The extended Ross–Witt Nyström correspondence

Definition 3.1. A sublinear subgeodesic ray is a subgeodesic (0, ∞) ∋ t 7→ ut ∈ PSH(X, θ)

Using t-convexity, we obtain some immediate properties of sublinear subgeodesic rays:

vta,b := sup ht , t ∈ [0, 1] , (15)

where S is the set of subgeodesics (0, 1) ∋ t → ht ∈ PSH(X, θ) such that limtց0 ht ≤ ua

The (inverse) Legendre transform of a decreasing concave function g : R → R ∪ {−∞} is

ψ̌(1−t)a+tb χt := sup ht , t ∈ [0, 1] ,

where S is the set of subgeodesics (a, b) ∋ t 7→ ht ∈ PSH(X, θ) satisfying lim ht ≤ ψ̌a

inf (φt − tτ ) ≤ φb − bτ = ψ̌b − bτ ≤ inf (ψ̌t − tτ ) + (a − b)τ ,

where in the last inequality we used that t 7→ ψ̌t is decreasing.

We define ûj0 := limτ →0− ûjτ .

Proof. As previously, we may assume that τψ+ = 0. As a preliminary result, in Proposi-

The R right hand

Proof. Without loss of generality we assume that vol({θ}) = 1. For N ∈ Z+ , M ∈ Z and

ψ̌tN,M +1 = ψ(M +1)/2N + t(M + 1)/2N , ψ̌tN,M = ψM/2N + tM/2N on Ut .

Since M ≤ 2N τψ− we have that ψ̌tN,M = ψM/2N + tM/2N = Vθ + tM/2N . As a result

By [74, Theorem 1.2] and multilinearity of non-pluripolar products we obtain X θv̂nτ1 ≥

We fix p < lim inf t→∞ −1 log X e−λut dµ and ǫ > 0

e−λûp dµ < ∞, hence p ≤ τD , concluding the proof.

τθ (E) = sup ν(u, E). (27)

Let hE be a smooth hermitian metric on OY (E), with canonical section SE and

(π ∗ θ − τ Θ(hE ) + ddc (π ∗ vEτ − τ log |SE |2hE ))n = (π ∗ θ + ddc π ∗ vEτ )n .

Thus using (25) we have that

which is called the expected complex singularity exponent of {ûτ }τ .

Aχ,ψ (E) Aχ,ψ (E)

So we obtain that inf {ut } cµ {ûτ } ≥ δµ .

Aχ,ψ (E) Aχ,ψ (E)

Theorem 4.7 (=Theorem 1.5). One has

Dµλ {wt } ≥ −εIθ {wt }. (33)

5 Kähler–Einstein metrics with positive twist

θun = e−λu dµ (36)

We say Dµλ is coercive/proper if there exists ε > 0 and C > 0,

Dµλ(u) ≥ ε(sup u − Iθ (u)) − C

lim Dµλ (ui ) = inf Dµλ (u).

We arrive at the main result of this section:

as Dβ is L1 -lsc, supX is L1 -continuous and Iθ is L1 -usc.

Dβ {ut } = −(1 − β)Iθ {ut } + sup{τ ∈ R : cµ [ûτ ] ≥ 1} ≤ −ε.

On the other hand, using dominated convergence we conclude that

Proposition 5.5. Assume that (35) is solvable. Then δµ ≥ 1.

6 The case of log-Fano pairs

δµ ({θ}) = δ̃µ (σ ∗ L).

D(ϕ) = Dµ1 (σ ∗ ϕ) for any ϕ ∈ E 1 (Z, L).

Department of Mathematics, University of Maryland

You might also like