Friday, February 27, 2026

Relation of Polar and Fenchel transforms

Relation of Polar and Fenchel transforms

Polar and Fenchel Transforms of a Gauge

Let X,YX, Y be a dual pair with bilinear pairing ⟨x,y⟩\langle x, y \rangle.
Let C⊂XC \subset X be a nonempty convex set containing 00.

We define two main objects:

  • The gauge function (Minkowski functional):
    γC(x):=inf⁡{t>0:x∈tC}. \gamma_C(x) := \inf\{t > 0 : x \in tC\}.

  • The indicator function:
    δC∘(y)={0,y∈C∘,+∞,y∉C∘. \delta_{C^\circ}(y) = \begin{cases} 0, & y \in C^\circ, \\ +\infty, & y \notin C^\circ. \end{cases}

We also define three fundamental transformations:

  • The polar transform of a set:
    C∘:={y∈Y:⟨x,y⟩≤1;∀x∈C}. C^\circ := \{ y \in Y : \langle x, y \rangle \le 1 ; \forall x \in C\}.

  • The polar transform of a gauge:
    k∘(y):=inf⁡{μ≥0:⟨x,y⟩≤μk(x);∀x∈X}. k^\circ(y) := \inf\{ \mu \ge 0 : \langle x, y \rangle \le \mu k(x) ; \forall x \in X\}.

  • The Fenchel transform (Fenchel conjugate) of a convex function:
    f∗(y):=sup⁡{⟨x,y⟩−f(x):x∈X}. f^*(y) := \sup\{ \langle x, y \rangle - f(x) : x \in X \}.

In this note, we prove the following results:

  1. Polar transform of a gauge: (γC)∘=γC∘(\gamma_C)^\circ = \gamma_{C^\circ}.

  2. Fenchel transform of a gauge: (γC)∗=δC∘(\gamma_C)^* = \delta_{C^\circ}.


NOTE. There is a more general definition of the polar transform on the class of nonnegative convex functions vanishing at 00:

k∘(y):=inf⁡{μ≥0:⟨x,y⟩≤1+μk(x);∀x∈X}. k^\circ(y) := \inf\{ \mu \ge 0 : \langle x, y \rangle \le 1 + \mu k(x) ; \forall x \in X \}.

On this class of functions, the following diagram commutes (see Convex Analysis - Rockafellar - Corollary 15.5.1 - P.138):

γC⇢γC∘↓↓δC∘⇢δC \gamma_C \dashrightarrow \gamma_{C^\circ} \\ \downarrow \qquad\quad \downarrow \\ \delta_{C^\circ} \dashrightarrow \delta_{C}

where ⇢\dashrightarrow denotes the Polar transform and ↓\downarrow denotes the Fenchel transform.


1. Polar of a Gauge

Fix μ>0\mu > 0. The condition

⟨x,y⟩≤μ,γC(x)∀x \langle x, y \rangle \le \mu, \gamma_C(x) \quad \forall x

is equivalent to checking it only on CC.

If x∈Cx \in C, then γC(x)≤1\gamma_C(x) \le 1, hence

⟨x,y⟩≤μ∀x∈C. \langle x, y \rangle \le \mu \quad \forall x \in C.

This is equivalent to

⟨x,μ−1y⟩≤1∀x∈C, \langle x, \mu^{-1} y \rangle \le 1 \quad \forall x \in C,

i.e. μ−1y∈C∘\mu^{-1} y \in C^\circ, or equivalently y∈μC∘y \in \mu C^\circ.

Therefore

(γC)∘(y)=inf⁡{μ≥0:y∈μC∘}=γC∘(y). (\gamma_C)^\circ(y) =\inf\{ \mu \ge 0 : y \in \mu C^\circ \} =\gamma_{C^\circ}(y).


2. Fenchel Conjugate of a Gauge

Normalize any xx as

x=tc,t:=γC(x),c∈C. x = t c, \quad t := \gamma_C(x), \quad c \in C.

Then

⟨x,y⟩−γC(x)t(⟨c,y⟩−1). \langle x, y \rangle - \gamma_C(x) t(\langle c, y \rangle - 1).

Hence

(γC)∗(y)=sup⁡t≥0,;c∈Ct(⟨c,y⟩−1). (\gamma_C)^*(y) =\sup_{t \ge 0,; c \in C} t(\langle c, y \rangle - 1).

Case analysis

Case 1: y∉C∘y \notin C^\circ

Then there exists c∈Cc \in C such that ⟨c,y⟩>1\langle c, y \rangle > 1.

Letting t→+∞t \to +\infty gives +∞+\infty.

Case 2: y∈C∘y \in C^\circ

Then

⟨c,y⟩≤1∀c∈C. \langle c, y \rangle \le 1 \quad \forall c \in C.

Hence t(⟨c,y⟩−1)≤0t(\langle c, y \rangle - 1) \le 0 for all t≥0t \ge 0.

Taking t=0t = 0 yields value 00, so the supremum equals 00.


Conclusion

We obtain the compact formula

(γC)∗(y)={0,y∈C∘,+∞,y∉C∘, (\gamma_C)^*(y) = \begin{cases} 0, & y \in C^\circ, \\ +\infty, & y \notin C^\circ, \end{cases}

or equivalently

(γC)∗=δC∘. (\gamma_C)^* = \delta_{C^\circ}.

Popular Posts