← Visual companionIn-app Blueprint viewLean 4 source indexPDFDependency graph

Erratum: Deep Equilibrium Models for Poisson Imaging Inverse Problems via Mirror Descent

2 Poisson Problem Setup, Counterexample, and Shifted Burg Geometry

Definition 12 Box constraint, KL divergence, and problem data
✓

Let \(a {\gt} 0\), \(\xi {\gt} 0\), \(y \in \mathbb {R}^m_{\ge 0}\) (zero counts allowed), \(A \in \mathbb {R}^{m \times n}_{\ge 0}\) with every row nonzero, and \(R_\theta : \mathcal{E}_n \to \mathbb {R}\) of class \(C^2\) on an open neighborhood \(U_R \supseteq [0, a]^n\). Set \(C := [0, a]^n\),

\[ \mathrm{KL}(y, Ax) := \sum _{i=1}^m d_{\mathrm{KL}}(y_i, (Ax)_i), \qquad F(x) := \sum _{i=1}^m \bigl(y_i \log \tfrac {y_i}{(Ax)_i} + (Ax)_i - y_i\bigr) + R_\theta (x), \]

with \(0 \log (0/u) := 0\) (even for \(u = 0\)), and \(\Psi (x) := \mathrm{KL}(y, Ax) + R_\theta (x) + \iota _{[0, a]^n}(x)\). Define the open positive-observation set \(U_+ := \{ x \in \mathcal{E}_n \mid \forall i,\; y_i {\gt} 0 \implies (Ax)_i {\gt} 0\} \) and the boundary singular set

\[ \mathcal{S} := \{ x \in [0, a]^n \mid \exists i \in \{ 1, \dots , m\} ,\; y_i {\gt} 0 \text{ and } (Ax)_i = 0\} . \]
Lemma 13 Effective domain of \(\Psi \) and singular boundary
✓

We have \(\mathrm{dom}(\Psi ) = [0, a]^n \cap U_+\), \(\overline{\mathrm{dom}(\Psi )} = [0, a]^n\), and \(\overline{\mathrm{dom}(\Psi )} \setminus \mathrm{dom}(\Psi ) = \mathcal{S}\).

Proof ▶

Membership \((Ax)_i {\gt} 0\) for every \(i\) with \(y_i {\gt} 0\) characterizes finiteness of \(\mathrm{KL}(y, Ax)\), while \(\iota _{[0,a]^n}(x) {\lt} +\infty \) characterizes \(x \in [0, a]^n\). Approximating any \(x \in [0, a]^n\) along convex combinations \((1 - t) x + t (a, \dots , a)\) stays in \([0, a]^n \cap U_+\) for \(t \in (0, 1]\), proving \(\overline{\mathrm{dom}(\Psi )} = [0, a]^n\) and \(\overline{\mathrm{dom}(\Psi )} \setminus \mathrm{dom}(\Psi ) = \mathcal{S}\).

Definition 14 Classical and shifted Burg entropies and Bregman divergence
✓

The classical Burg entropy on \(\mathcal{E}_n\) is \(h(x) := -\sum _{j=1}^n \log x_j\) on \(\mathbb {R}^n_{{\gt}0}\) (and \(+\infty \) otherwise), with \((\nabla h(x))_j = -1/x_j\). For shift \(\xi {\gt} 0\), the shifted Burg entropy and Bregman divergence are

\[ h_\xi (x) := \sum _{j=1}^n \bigl(-\log (x_j + \xi )\bigr), \qquad (\nabla h_\xi (x))_j = -\frac{1}{x_j + \xi }, \]
\[ D_{h_\xi }(u, x) := h_\xi (u) - h_\xi (x) - \langle \nabla h_\xi (x), u - x \rangle . \]
Example 15 Counterexample to Lipschitz continuity of \(\nabla h\) on \(L_0\) — Example 1
✓

With \(A = \begin{pmatrix} 1 & 1 \\ 0 & 1 \end{pmatrix}\), \(R_\theta \equiv 0\), \(a \ge 1\), \(y \in \mathbb {R}^2_{{\gt}0}\), and \(x^0 = (0, 1)\), we have \(\Psi (x^0) {\lt} +\infty \) and every \(x_t := (t, 1)\) for \(t \in (0, 1]\) lies in \(L_0(x^0) \cap (0, a]^n\) with \(t \mapsto \Psi (x_t)\) bounded above on \((0, 1]\). Yet \(\| \nabla h(x_t) - \nabla h(x_1)\| = |1 - 1/t| \to +\infty \) while \(\| x_t - x_1\| = 1 - t \le 1\), so \(\nabla h\) is not Lipschitz continuous on \(\mathrm{dom}(h) \cap L_0(x^0)\).

Proof ▶

Direct evaluation gives \(A x_t = (t + 1, 1)\) componentwise \(\ge 1\), keeping \(\mathrm{KL}(y, A x_t)\) continuous on \([0, 1]\), hence bounded by its maximum. Meanwhile \(\nabla h(x_t)_0 - \nabla h(x_1)_0 = 1 - 1/t\) grows arbitrarily large as \(t \downarrow 0\) whereas \(\| x_t - x_1\| \le 1\).

Lemma 16 Strong convexity and Lipschitz gradient of \(h_\xi \) on \({[0, a]^n}\)
✓

For all \(u, x \in [0, a]^n\),

\[ D_{h_\xi }(u, x) \ge \frac{1}{2(a + \xi )^2} \| u - x\| ^2, \qquad \| \nabla h_\xi (u) - \nabla h_\xi (x)\| \le \frac{1}{\xi ^2} \| u - x\| . \]
Proof ▶

By the 1D integral/mean-value identity for \(s \mapsto -\log (s + \xi )\) on \([0, a]\), each coordinate Bregman divergence satisfies \(-\log (\frac{u_j + \xi }{x_j + \xi }) + \frac{u_j - x_j}{x_j + \xi } \ge \frac{(u_j - x_j)^2}{2(a + \xi )^2}\), and \(|-(u_j + \xi )^{-1} + (x_j + \xi )^{-1}| \le \frac{|u_j - x_j|}{\xi ^2}\).

Definition 17 Mirror Descent subproblem and closed-form update \(T_\tau \)
✓

For step size \(\tau {\gt} 0\) and \(x \in [0, a]^n\), the shifted-Burg Mirror Descent subproblem objective is

\[ \Phi _{x, \tau }(z) := \iota _{[0, a]^n}(z) + \langle z - x, \nabla F(x) \rangle + \frac{1}{\tau } D_{h_\xi }(z, x). \]

We define the operator \(T_\tau \) coordinatewise by

\[ (T_\tau (x))_j := \begin{cases} \Pi _{[0, a]}\! \left(\dfrac {x_j + \xi }{1 + \tau (x_j + \xi ) \partial _j F(x)} - \xi \right) & \text{if } 1 + \tau (x_j + \xi ) \partial _j F(x) {\gt} 0, \\[6pt] a & \text{otherwise}. \end{cases} \]
Theorem 18 Variational equivalence \((1) \iff (4) \iff (5)\) and continuity of \(T_\tau \)
✓

For any \(\tau {\gt} 0\) and \(x \in [0, a]^n\) satisfying \(1 + \tau (x_j + \xi ) \partial _j F(x) {\gt} 0\) for all \(j\):

  1. \(T_\tau (x)\) satisfies the optimality inclusion (5):

    \[ -\nabla F(x) - \frac{1}{\tau }\bigl(\nabla h_\xi (T_\tau (x)) - \nabla h_\xi (x)\bigr) \in \partial \iota _{[0, a]^n}(T_\tau (x)) = N_{[0, a]^n}(T_\tau (x)); \]
  2. \(T_\tau (x)\) is the unique global minimizer (1) of \(\Phi _{x, \tau }\) over \(\mathcal{E}_n\), and conversely any global minimizer of \(\Phi _{x, \tau }\) coincides with \(T_\tau (x)\) (4).

Furthermore, for every \(\tau \in \mathbb {R}\), \(T_\tau \) is continuous on \([0, a]^n \cap U_+\).

Proof ▶

The coordinate scalar inclusion for \(N_{[0, a]}(y_j)\) follows by inspecting the three cases \((x_j + \xi )/d_j - \xi \le 0\), \(\ge a\), and \(\in (0, a)\). The three-point Bregman identity together with Lemma 16 makes \(\Phi _{x, \tau }(z) - \Phi _{x, \tau }(T_\tau (x)) {\gt} 0\) for all \(z \ne T_\tau (x)\). Clamping the denominator below by \(\delta = \frac{\xi }{2(a + \xi )}\) agrees identically with \(T_\tau \) on \([0, a]^n \cap U_+\), establishing continuity even when denominators approach or cross zero.