Totally Convex Functionals

• Totally convex functionals are convex functionals that maintain convexity along every interpolation in quadratic Wasserstein spaces, ensuring robust analytic properties.
• They facilitate a full convex-analysis framework including Fenchel conjugacy, subdifferentials, and Moreau–Yosida regularization through a Lagrangian lift to Hilbert spaces.
• This approach transforms optimal transport problems into Hilbert-space settings, providing unique insights into cyclic monotonicity and the structure of optimal couplings.

Totally convex functionals form a class of convex-analytic objects whose precise meaning depends on the ambient space. In the most direct recent arXiv treatment, the term refers to functionals $\phi:\mathcal P_2(H)\to \mathbb R\cup\{+\infty\}$ on the quadratic Wasserstein space over a separable Hilbert space $H$, where total convexity means convexity along every interpolation induced by every coupling, not merely along optimal transport geodesics; equivalently, the Lagrangian lift $\hat\phi=\phi\circ\iota$ to $L^2(Q,\mathbb M;H)$ is an ordinary convex functional on a Hilbert space (Pinzi et al., 1 Sep 2025). In that setting total convexity supplies a full convex-analysis dictionary—Fenchel conjugacy, biconjugation, subdifferentials, cyclic monotonicity, Moreau–Yosida regularization, and Brenier-type optimal transport structure for laws of random measures (Pinzi et al., 1 Sep 2025).

1. Definition and conceptual position

Let $H$ be a separable Hilbert space and $\mathcal P_2(H)$ the quadratic Wasserstein space. The recent optimal-transport treatment defines a functional $\phi:\mathcal P_2(H)\to\mathbb R\cup\{+\infty\}$ to be totally convex if for every coupling $\boldsymbol\mu\in\mathcal P_2(H\times H)$ and every $t\in[0,1]$,

$$\phi(\mu_t)\le (1-t)\phi(\mu_0)+t\phi(\mu_1), \qquad \mu_t:=(\pi_t^{12})_\sharp \boldsymbol\mu, \qquad \pi_t^{12}(x_1,x_2):=(1-t)x_1+t x_2.$$

The same source defines $H$0 to be totally $H$1-convex if

$H$2

is totally convex, equivalently if the lifted functional $H$3 is $H$4-convex on $H$5 (Pinzi et al., 1 Sep 2025).

This notion is stronger than ordinary Wasserstein geodesic convexity. Geodesic convexity only requires convexity along at least one optimal interpolation between two measures, whereas total convexity requires convexity along every interpolation induced by every coupling. The same paper states that, if $H$6, every continuous geodesically convex functional on $H$7 is already totally convex (Pinzi et al., 1 Sep 2025). This places total convexity at the point where Wasserstein geometry begins to recover Hilbert-space convex analysis.

A central structural ingredient is the maximal-correlation pairing

$H$8

which satisfies

$H$9

This identity is the analogue of the Euclidean polarization formula and is the basis for the conjugacy theory of totally convex functionals (Pinzi et al., 1 Sep 2025).

2. Fenchel theory and total subdifferentials

The corresponding conjugate is the Kantorovich–Legendre–Fenchel transform

$\hat\phi=\phi\circ\iota$0

Under a mild linear lower bound, $\hat\phi=\phi\circ\iota$1 is proper, totally convex, and lower semicontinuous; moreover, $\hat\phi=\phi\circ\iota$2 is the largest totally convex lower semicontinuous functional below $\hat\phi=\phi\circ\iota$3. The paper gives the equivalences

$\hat\phi=\phi\circ\iota$4

and also the support-type representation

$\hat\phi=\phi\circ\iota$5

for a suitable $\hat\phi=\phi\circ\iota$6 (Pinzi et al., 1 Sep 2025).

The total subdifferential is not merely a subset of $\hat\phi=\phi\circ\iota$7. It is a family of plans $\hat\phi=\phi\circ\iota$8. A plan $\hat\phi=\phi\circ\iota$9 belongs to $L^2(Q,\mathbb M;H)$0 if $L^2(Q,\mathbb M;H)$1 and, for every $L^2(Q,\mathbb M;H)$2 and every $L^2(Q,\mathbb M;H)$3,

$L^2(Q,\mathbb M;H)$4

This refines the $L^2(Q,\mathbb M;H)$5-subdifferential

$L^2(Q,\mathbb M;H)$6

because

$L^2(Q,\mathbb M;H)$7

Thus the total subdifferential selects both a dual measure $L^2(Q,\mathbb M;H)$8 and an optimal coupling between $L^2(Q,\mathbb M;H)$9 and $H$0 (Pinzi et al., 1 Sep 2025).

The subdifferential theory inherits the usual monotone-operator structure. The lifted graph $H$1 coincides with the ordinary convex subdifferential $H$2 on $H$3. Consequently, $H$4 is totally cyclically monotone; if $H$5 is totally convex and lower semicontinuous, then $H$6 is maximal totally monotone; and there is a unique deterministic minimal section $H$7, written

$H$8

The paper defines $H$9 to be $\mathcal P_2(H)$0-differentiable at $\mathcal P_2(H)$1 when $\mathcal P_2(H)$2 is a singleton (Pinzi et al., 1 Sep 2025).

3. Lagrangian lifting to a Hilbert space

The decisive construction is the law map

$\mathcal P_2(H)$3

with $\mathcal P_2(H)$4 nonatomic. This map is surjective and $\mathcal P_2(H)$5-Lipschitz. It also satisfies

$\mathcal P_2(H)$6

together with the sharper identities

$\mathcal P_2(H)$7

and

$\mathcal P_2(H)$8

In this language, total convexity is exactly ordinary convexity of $\mathcal P_2(H)$9 (Pinzi et al., 1 Sep 2025).

This lifting turns a non-Hilbertian Wasserstein problem into Hilbert-space convex analysis. Conjugation commutes with lifting: $\phi:\mathcal P_2(H)\to\mathbb R\cup\{+\infty\}$0 Moreau–Yosida regularization also commutes with lifting: $\phi:\mathcal P_2(H)\to\mathbb R\cup\{+\infty\}$1 The same paper uses these identities to import Fenchel–Moreau theory, Rockafellar-type cyclic monotonicity, Yosida regularization, and differentiability theory from $\phi:\mathcal P_2(H)\to\mathbb R\cup\{+\infty\}$2 back to $\phi:\mathcal P_2(H)\to\mathbb R\cup\{+\infty\}$3 (Pinzi et al., 1 Sep 2025).

This suggests that total convexity is the Wasserstein notion precisely calibrated to preserve linear convex analysis after passage through $\phi:\mathcal P_2(H)\to\mathbb R\cup\{+\infty\}$4. It is not an auxiliary strengthening but the condition under which the Wasserstein problem becomes genuinely Hilbertian.

4. Optimal transport for laws of random measures

The main application is quadratic optimal transport on

$\phi:\mathcal P_2(H)\to\mathbb R\cup\{+\infty\}$5

the space of laws of random measures. For $\phi:\mathcal P_2(H)\to\mathbb R\cup\{+\infty\}$6, the transport cost is

$\phi:\mathcal P_2(H)\to\mathbb R\cup\{+\infty\}$7

and there is a lifted maximal-correlation identity

$\phi:\mathcal P_2(H)\to\mathbb R\cup\{+\infty\}$8

Optimal Kantorovich potentials can then be chosen in the form of a totally convex proper lower semicontinuous functional $\phi:\mathcal P_2(H)\to\mathbb R\cup\{+\infty\}$9 and its conjugate $\boldsymbol\mu\in\mathcal P_2(H\times H)$0, satisfying

$\boldsymbol\mu\in\mathcal P_2(H\times H)$1

on the support of an optimal coupling (Pinzi et al., 1 Sep 2025).

Optimality admits the expected cyclic-monotonicity and subdifferential characterizations. In particular, optimal random couplings are concentrated on $\boldsymbol\mu\in\mathcal P_2(H\times H)$2, and this is the random-measure analogue of the statement that optimal quadratic couplings in Euclidean space are concentrated on the graph of a convex subdifferential (Pinzi et al., 1 Sep 2025).

The Monge problem is solved under a regularity hypothesis on the source law. A measure $\boldsymbol\mu\in\mathcal P_2(H\times H)$3 is called super-regular when it vanishes on random exceptional sets and is concentrated on $\boldsymbol\mu\in\mathcal P_2(H\times H)$4. For super-regular $\boldsymbol\mu\in\mathcal P_2(H\times H)$5 and arbitrary $\boldsymbol\mu\in\mathcal P_2(H\times H)$6, the paper proves existence and uniqueness of the strict Monge solution, induced by the minimal section of an optimal totally convex potential: $\boldsymbol\mu\in\mathcal P_2(H\times H)$7 The source class is nontrivial: the paper shows that it includes laws with full support in $\boldsymbol\mu\in\mathcal P_2(H\times H)$8 obtained as pushforwards of nondegenerate Gaussian measures on $\boldsymbol\mu\in\mathcal P_2(H\times H)$9, and in finite dimension the class of super-regular measures is dense (Pinzi et al., 1 Sep 2025).

5. Relation to ordinary convexity and stronger curvature notions

Ordinary convexity is much weaker. A useful background result states that if $t\in[0,1]$0 is a convex subset of a real vector space, then a radially lower semicontinuous function

$t\in[0,1]$1

is convex if and only if for all $t\in[0,1]$2 there exists $t\in[0,1]$3 such that

$t\in[0,1]$4

Thus, under radial lower semicontinuity, one admissible interior interpolation point on each segment already forces full convexity (Leonetti, 2017). By contrast, total convexity on $t\in[0,1]$5 requires convexity along every interpolation induced by every coupling, and the corresponding calculus is correspondingly stronger (Pinzi et al., 1 Sep 2025).

A second, different usage of the phrase appears in Banach-space optimization. There, total convexity typically means that the functional bends away from its tangent planes in a quantitatively positive way, often formulated through positivity of

$t\in[0,1]$6

for $t\in[0,1]$7 (Leonetti, 2017). This is not the same definition as the Wasserstein-space one, even though both are stronger than plain convexity. This suggests that the term must always be read relative to its ambient geometry.

The distinction becomes operational in approximation theory. A recent reconstruction theorem for convex Lipschitz functionals on compact convex subsets of a separable Hilbert space produces explicit finite max-affine reconstructions

$t\in[0,1]$8

which preserve convexity and the same Lipschitz constant, but the paper explicitly does not study strict, strong, uniform, or total convexity, and notes that max-affine functions typically have flat regions or affine faces (Kratsios, 8 May 2026). This suggests that preserving ordinary convexity is compatible with finitely computable max-affine structure, whereas preserving stronger curvature properties generally requires additional structure.

6. Adjacent literatures and terminological boundaries

A recurrent source of confusion is that many papers study functionals on convex objects or convex functionals without studying total convexity. One geometric line studies valuations on the space

$t\in[0,1]$9

with max/min additivity, monotone decreasing order behavior, rigid-motion invariance, and $$\phi(\mu_t)\le (1-t)\phi(\mu_0)+t\phi(\mu_1), \qquad \mu_t:=(\pi_t^{12})_\sharp \boldsymbol\mu, \qquad \pi_t^{12}(x_1,x_2):=(1-t)x_1+t x_2.$$0-continuity. Its main theorems classify simple and homogeneous valuations, but the paper explicitly states that it does not characterize convex, strictly convex, or totally convex functionals in the optimization sense (Cavallina et al., 2015).

Another line studies increasing convex functionals on lattices of measurable or continuous functions. On countable product spaces, increasing convex functionals under marginal constraints admit measure-valued dual representations that subsume transport and martingale-transport duality (Bartl et al., 2015). On $$\phi(\mu_t)\le (1-t)\phi(\mu_0)+t\phi(\mu_1), \qquad \mu_t:=(\pi_t^{12})_\sharp \boldsymbol\mu, \qquad \pi_t^{12}(x_1,x_2):=(1-t)x_1+t x_2.$$1 with $$\phi(\mu_t)\le (1-t)\phi(\mu_0)+t\phi(\mu_1), \qquad \mu_t:=(\pi_t^{12})_\sharp \boldsymbol\mu, \qquad \pi_t^{12}(x_1,x_2):=(1-t)x_1+t x_2.$$2 Polish, upward continuity yields a $$\phi(\mu_t)\le (1-t)\phi(\mu_0)+t\phi(\mu_1), \qquad \mu_t:=(\pi_t^{12})_\sharp \boldsymbol\mu, \qquad \pi_t^{12}(x_1,x_2):=(1-t)x_1+t x_2.$$3-additive measure representation for convex increasing functionals, while downward continuity is characterized by weak$$\phi(\mu_t)\le (1-t)\phi(\mu_0)+t\phi(\mu_1), \qquad \mu_t:=(\pi_t^{12})_\sharp \boldsymbol\mu, \qquad \pi_t^{12}(x_1,x_2):=(1-t)x_1+t x_2.$$4 compactness of dual sublevels and dual attainment (Delbaen, 2022). These are duality and representation theorems, not theories of total convexity.

Related but distinct are robust convex integral functionals on $$\phi(\mu_t)\le (1-t)\phi(\mu_0)+t\phi(\mu_1), \qquad \mu_t:=(\pi_t^{12})_\sharp \boldsymbol\mu, \qquad \pi_t^{12}(x_1,x_2):=(1-t)x_1+t x_2.$$5, where the main issues are elimination of singular dual measures, Mackey continuity, Lebesgue properties, and exact $$\phi(\mu_t)\le (1-t)\phi(\mu_0)+t\phi(\mu_1), \qquad \mu_t:=(\pi_t^{12})_\sharp \boldsymbol\mu, \qquad \pi_t^{12}(x_1,x_2):=(1-t)x_1+t x_2.$$6-conjugacy (Owari, 2013); logarithmic means of two convex functionals, where the central objects are weighted arithmetic, harmonic, geometric, and logarithmic means in $$\phi(\mu_t)\le (1-t)\phi(\mu_0)+t\phi(\mu_1), \qquad \mu_t:=(\pi_t^{12})_\sharp \boldsymbol\mu, \qquad \pi_t^{12}(x_1,x_2):=(1-t)x_1+t x_2.$$7 (Raïssouli et al., 2020); and convex integral functionals of càdlàg processes, where the emphasis is on interchange rules, conjugates, and subdifferentials on nondecomposable path spaces (Perkkiö et al., 2018). None of these papers develops total convexity as such.

The same caution applies in geometric analysis. For area-measure integral functionals on convex bodies, Brunn–Minkowski-type concavity implies monotonicity and, in the fully mixed setting, characterizes mixed volumes (Colesanti et al., 2016). In shape optimization, different additions—Minkowski, Blaschke, or a functional-specific $$\phi(\mu_t)\le (1-t)\phi(\mu_0)+t\phi(\mu_1), \qquad \mu_t:=(\pi_t^{12})_\sharp \boldsymbol\mu, \qquad \pi_t^{12}(x_1,x_2):=(1-t)x_1+t x_2.$$8-addition—lead to different concavity notions for shape functionals such as volume, capacity, torsional rigidity, and $$\phi(\mu_t)\le (1-t)\phi(\mu_0)+t\phi(\mu_1), \qquad \mu_t:=(\pi_t^{12})_\sharp \boldsymbol\mu, \qquad \pi_t^{12}(x_1,x_2):=(1-t)x_1+t x_2.$$9 (Bucur et al., 2011). On metric spaces of convex bodies, continuity, affine invariance, and local Lipschitz behavior are developed through Hausdorff and Banach–Mazur metrics rather than through any theory of total convexity (Zong, 2014). A further terminological variant appears in the study of measurable linear functionals under convex, i.e. logarithmically concave, measures on $H$00, where “convex” refers to the measure rather than the functional (Fufaev, 2016).

This suggests a broad taxonomy. “Totally convex functional” is a precise term only after fixing the ambient structure: in $H$01 it means convexity along arbitrary couplings and Hilbertian lift-convexity (Pinzi et al., 1 Sep 2025); in Banach-space optimization it commonly refers to quantitative separation from tangent planes (Leonetti, 2017); and in several neighboring literatures the phrase is simply absent.