---
title: 'VPTPD: Variable-Preconditioned Primal-Dual Methods'
url: https://www.emergentmind.com/topics/variable-preconditioned-transformed-primal-dual-vptpd
type: topic
---

# VPTPD: Variable-Preconditioned Primal-Dual Methods

Searching arXiv for recent papers on Variable-Preconditioned Transformed Primal-Dual methods and closely related formulations.
arxiv_search query="Variable-Preconditioned Transformed Primal-Dual method" max_results=10

arxiv_search query="Transformed Primal-Dual variable preconditioners" max_results=10

Variable-Preconditioned Transformed Primal-Dual (VPTPD) denotes a family of primal-dual algorithms in which transformed saddle-point dynamics are combined with nonstationary preconditioners or variable metrics. Across the recent literature, the name appears in several closely related settings: variable-wise diagonal preconditioning for primal-dual splitting, transformed primal-dual methods with evolving Schur-complement surrogates, structure-preserving JKO solvers for generalized Wasserstein gradient flows, and duality-based solvers for nonlinear PDEs [2301.08468] [2312.12355] [2509.15385] [2510.15351]. The common design pattern is the replacement of fixed scalar step-sizes by operators such as block-diagonal, diagonal, Hessian-derived, Schur-complement, or antisymmetric preconditioners, with the aim of preserving tractable proximal subproblems while improving conditioning, convergence speed, or mesh-independence. This suggests that “VPTPD” is best understood as a methodological family rather than a single canonical recursion.

## 1. Scope, nomenclature, and problem classes

The VPTPD label is attached to several algorithmic variants that share a transformed primal-dual structure but target different mathematical models.

| Variant | Problem class | Characteristic preconditioner |
|---|---|---|
| P-PDS with OVDP [2301.08468] | Nonsmooth convex optimization in split form | Variable-wise diagonal preconditioners from operator-norm bounds |
| TPDv [2312.12355] | Affine constrained optimization in Hilbert spaces | Time-evolving SPD metric operators and Schur-complement proxies |
| VPTPD for generalized Wasserstein gradient flows [2509.15385] | Structure-preserving JKO discretizations | Diagonal Hessian-based primal metric and Schur-complement dual metric |
| DualTPD [2510.15351] | Nonlinear PDEs via Fenchel–Rockafellar duality | Variable block preconditioners for nonlinear and Schur blocks |
| Variable-metric primal-dual Douglas–Rachford [2302.13128] | Saddle-point monotone inclusions | Degenerate block-diagonal varying preconditioners |
| Time-varying preconditioned primal-dual dynamics [2506.00501] | Convex optimization of the form $f(x)+g(Ax)$ | Time-dependent symmetric/skew preconditioners, including antisymmetric regimes |

In the split convex setting, the central problem class is
$$
\min_{x\in\mathbb R^n}\;\sum_{i=1}^Nf_i(x_i)\;+\;\sum_{j=1}^M g_j\Bigl(\sum_{i=1}^NL_{j,i}x_i\Bigr),
$$
with associated dual blocks $y_j$ and linear operators $L_{j,i}$ [2301.08468]. In affine-constrained optimization, the prototype is
$$
\min_{x\in V}\; f(x)\qquad\text{s.t.}\quad Bx=b,
$$
with $f$ continuously differentiable, strongly convex, and Lipschitz differentiable in a variable metric [2312.12355]. In generalized Wasserstein gradient flows, the variational structure arises from the JKO minimization
$$
\rho^{k+1} \in \mathrm{Arg}\,\min_\rho \left\{ \frac{1}{2\tau}d_{\mathcal W_m}^2(\rho,\rho^k)+\mathcal J(\rho)\right\},
$$
which, after Benamou–Brenier reformulation and discretization, becomes an equality-constrained finite-dimensional optimization problem [2509.15385]. In nonlinear PDE applications, Fenchel–Rockafellar duality is used to rewrite energies of the form $\min_{u\in\mathcal V}F(Du)+G(u)$ as nonlinear saddle systems with separated differential and nonlinear blocks [2510.15351].

## 2. Mathematical structure and transformed saddle systems

A recurring feature of VPTPD methods is the conversion of optimization problems into saddle-point systems in which primal and dual variables can be updated under tailored metrics.

For split convex problems, introducing dual variables $y_j$ yields the convex-concave saddle formulation
$$
\min_{x}\max_{y}\;\sum_{i=1}^N f_i(x_i)\;+\;\sum_{j=1}^M\langle L_{j,i}x_i,y_j\rangle \;-\;\sum_{j=1}^M g_j^*(y_j),
$$
which is then attacked by preconditioned primal-dual splitting [2301.08468]. In affine-constrained settings, the Lagrangian
$$
\mathcal L(x,y)=f(x)+\langle Bx-b,\;y\rangle_Q
$$
leads to the KKT system
$$
\begin{bmatrix} \nabla\mathcal F_h & B^T \\ B & 0 \end{bmatrix}
\begin{bmatrix}u^*\\ p^*\end{bmatrix}
=
\begin{bmatrix}0\\ b\end{bmatrix},
$$
and the transformed primal-dual construction uses an SPD operator $I_u\approx \nabla^2\mathcal F_h$ to rewrite the system so that the dual block acquires a Schur-complement term $-B\,I_u^{-1}B^T$ [2312.12355] [2509.15385].

The same transformation principle appears in the nonlinear PDE literature. In the TPDv framework, the continuous transformed flow is written with time-dependent SPD operators
$$
M(t)\colon V\to V,\qquad Q(t)\colon Q\to Q,
$$
and an approximation $\widetilde S(t)\approx S(t)$ where $S(t)=B\,M(t)^{-1}B^T$ [2312.12355]. In DualTPD, after applying the triangular transform
$$
E=\begin{pmatrix}I&0\\-D^T\,M_\sigma^{-1}&I\end{pmatrix},
$$
the nonlinear saddle system becomes an upper-block form with $A=\nabla F^*(\sigma)$ and elliptic Schur complement $S=D^T\,A^{-1}D$ [2510.15351].

A related but distinct continuous-time formulation considers the monotone inclusion associated with
$$
\min f(x)+g(Ax)
$$
and replaces the un-preconditioned Arrow–Hurwicz flow by
$$
P(t)\,\dot z(t)\;+\;M\,z(t)\;\ni\;0,
$$
where $P(t)$ is a time-varying $2\times2$ block preconditioner with symmetric and skew parts [2506.00501]. This places VPTPD in direct contact with continuous-time primal-dual dynamics and with the preconditioning logic behind Chambolle–Pock-type schemes.

## 3. Variable preconditioners and metric design

The defining feature of VPTPD is the explicit design of nonconstant preconditioners from operator norms, local Hessians, Schur complements, or problem-specific block structure.

In the OVDP construction for preconditioned primal-dual splitting, one defines
$$
\gamma_{j,i}\ge\|L_{j,i}\|_{\mathrm{op}}
\quad\text{and chooses}\quad
P_i = \frac{1}{\,\sum_{j=1}^M\gamma_{j,i}^{\,2-\theta}\,}\,I_{n_i},
\qquad
D_j = \frac{1}{\,\sum_{i=1}^N\gamma_{j,i}^{\,\theta}\,}\,I_{m_j},
$$
for $\theta\in[0,2]$ [2301.08468]. The three stated special cases are $\theta=1$ (OVDP2), $\theta=0$ (OVDP1), and $\theta=2$ (OVDP3). The construction uses only upper bounds on operator norms, not explicit matrix entries, and because it is variable-wise rather than element-wise, it “keeps any proximity operator analytically computable” [2301.08468].

In TPDv for affine-constrained optimization, the preconditioners are evolving metric operators $M_k$ and $Q_k$, together with a Schur-complement proxy $\widetilde S_k\approx B\,M_k^{-1}B^T$. The discrete update
$$
Q_{k+1}
=\frac{1}{1+\alpha_k\gamma_k}Q_k
+\frac{\alpha_k\gamma_k}{1+\alpha_k\gamma_k}\widetilde S_k
$$
shows that the dual metric is an interpolation of the previous preconditioner and a new Schur-complement approximation [2312.12355].

In the generalized Wasserstein setting, the paper defines a regularized smooth objective
$$
\hat H(u)=\frac{1}{2\tau}\sum_i \frac{\|m_i\|^2}{M(\rho_i)+r}\Delta V+\mathcal J_h(u)
$$
and sets
$$
I_u^k=\mathrm{diag}(\nabla^2\hat H(u^k)),
\qquad
I_p^k=B\,(I_u^k)^{-1}B^T
$$
[2509.15385]. The stated rationale is that $I_u^k$ “approximates the inverse Hessian so as to reduce condition-number, yet remains diagonal for cheap inversion and separability,” while $I_p^k$ is “its associated Schur-complement to precondition the dual” [2509.15385].

In DualTPD, the nonlinear and elliptic blocks are both preconditioned:
$$
M_\sigma(\sigma_k)\approx\nabla F^*(\sigma_k),
\qquad
M_u(\sigma_k)\approx D^T\,M_\sigma(\sigma_k)^{-1}D.
$$
The nonlinear block may use a mass matrix inverse or Jacobian inverse, and the Schur block may be approximated by geometric multigrid or PCG with AMG/Hiptmair–Xu auxiliary-space [2510.15351].

The continuous-time preconditioned dynamics literature adds a further design axis: asymptotically antisymmetric preconditioners. There, convergence analysis requires uniform positive definiteness of the symmetric part $S(t)$ and, for the general problem, the asymptotic condition
$$
B(t)+\gamma(t)\to 0,
\qquad
\int_0^\infty\!\frac{(B(t)+\gamma(t))^2}{\,B(t)-\gamma(t)\,}\,dt<+\infty
$$
as stated in the source [2506.00501]. In the exact-antisymmetric case, one takes $\gamma(t)=-B(t)$.

## 4. Iterative schemes, proximal steps, and transformed updates

The update rules vary with the application, but they retain a recognizable primal-dual structure with variable metrics.

For block-split nonsmooth optimization, the preconditioned primal-dual splitting updates are
$$
x^{t+1} = \mathrm{prox}_{P^{-1},\,f}\bigl(x^t - P\,L^*\,y^t\bigr),\qquad
y^{t+1} = \mathrm{prox}_{D^{-1},\,g^*}\Bigl(y^t + D\,L\bigl(2x^{t+1}-x^t\bigr)\Bigr),
$$
or, blockwise,
$$
x_i^{t+1} = \mathrm{prox}_{P_i^{-1},\,f_i}\Bigl(x_i^t - P_i\sum_{j}L_{j,i}^*\,y_j^t\Bigr),\qquad
y_j^{t+1} = \mathrm{prox}_{D_j^{-1},\,g_j^*}\Bigl(y_j^t + D_j\sum_{i}L_{j,i}(2x_i^{t+1}-x_i^t)\Bigr)
$$
[2301.08468]. The implementation computes $u_i^t=\sum_j L_{j,i}^*y_j^t$ and $v_j^t=\sum_i L_{j,i}(2x_i^{t+1}-x_i^t)$ before the two prox evaluations.

For affine constraints with a transformed primal-dual step, the TPDv discretization performs four operations per iteration: a slave-step for $x$, an update of the dual-metric preconditioner $Q$, a dual update, and an over-relaxation for $x$ [2312.12355]. The generalized Wasserstein VPTPD method modifies this further into a semi-implicit-explicit proximal iteration:
$$
x^{k+1} = \mathrm{prox}_{\tau g}^{\,I_u^k}\Bigl(x^k - \tau\,(I_u^k)^{-1}[\nabla f(x^k)+B^T y^k]\Bigr),
$$
$$
y^{k+1} = y^k + \sigma\,(I_p^k)^{-1}[B(2x^{k+1}-x^k) - b - B\,(I_u^k)^{-1}[\nabla f(x^{k+1})+B^T y^k]],
$$
with optional over-relaxations
$$
\bar x^{k+1}=(1+\kappa_1)x^{k+1}-\kappa_1x^k,\qquad
\bar y^{k+1}=(1+\kappa_2)y^{k+1}-\kappa_2y^k
$$
[2509.15385].

A central subroutine in that setting is the generalized proximal operator
$$
\mathrm{prox}_f^M(v):=\mathrm{Arg}\,\min_u \left\{\frac12\|u-v\|_M^2+f(u)\right\},
\qquad
\|w\|_M^2=w^T M w,
$$
which becomes cellwise separable for $f(u)=(\lambda/2\tau)\Phi_h(u)$ [2509.15385]. The resulting scalar equation $L(\rho)=0$ is solved by Newton’s method,
$$
\rho^{t+1} = \rho^t - \frac{L(\rho^t)}{L'(\rho^t)},
$$
with an interval bracketing the unique root and a tailored initial guess; the paper states “global Newton convergence and bound-preservation” [2509.15385].

The Douglas–Rachford variant proceeds from the non-stationary proximal-point iteration with block-diagonal preconditioner
$$
\Delta_k:=\mathrm{diag}(P_k^{-1},Q_k)
$$
and factorizes the step into two half-resolvents. Its explicit VPTPD iteration computes
$$
x^k = \mathrm{prox}^{P_k}_f(p^k),\qquad y^k = \mathrm{prox}^{Q_k^{-1}}_{g^*}(q^k),
$$
forms reflected points $r_x^k=2x^k-p^k$ and $r_y^k=2y^k-q^k$, solves
$$
y'=(I+Q_k K P_k K^T)^{-1}(r_y^k+Q_k K r_x^k),\qquad x'=r_x^k-P_k K^T y',
$$
and then updates $p^{k+1}=p^k+x'-x^k$, $q^{k+1}=q^k+y'-y^k$ [2302.13128].

DualTPD for nonlinear PDEs uses a two-stage correction:
$$
\delta u_k=M_{u_k}^{-1}\bigl(r_k^u-D^T M_{\sigma_k}^{-1}r_k^\sigma\bigr),\qquad
\delta\sigma_k=M_{\sigma_k}^{-1}\bigl(r_k^\sigma+D\,\delta u_k\bigr),
$$
followed by
$$
\sigma_{k+1}=\sigma_k-\alpha_k\delta\sigma_k,\qquad
u_{k+1}=u_k-\alpha_k\delta u_k
$$
[2510.15351].

## 5. Convergence theory and stability results

The convergence theory attached to VPTPD depends on the specific formulation, but several recurring themes are visible: norm-control of the preconditioned coupling, Lyapunov decay, monotonicity under transformed dynamics, and variable-metric fixed-point arguments.

For OVDP in preconditioned primal-dual splitting, the critical quantity is
$$
D^{1/2}LP^{1/2},
$$
and the P–PDS convergence theorem requires
$$
\|D^{1/2}LP^{1/2}\|_{\mathrm{op}}<1.
$$
The paper proves that the OVDP choice of $P_i$ and $D_j$ satisfies the convergence condition and states: “Let $f_i,g_j$ be proper closed convex, and let $P=\mathrm{diag}(P_i)$, $D=\mathrm{diag}(D_j)$ be chosen by OVDP as above. Then the sequence $(x^t,y^t)$ generated by the P–PDS iterations converges to a saddle point of the convex–concave problem” [2301.08468].

For TPDv in Hilbert spaces, the analysis is Lyapunov-based. The continuous flow is shown to be exponentially stable under assumptions including $f\in\mathcal S_{\mu_{f,M(t)},L_{f,M(t)}}$ with $\mu_{f,M(t)}>1/2$, and the discrete method has linear convergence. The paper states that there exists $\rho_k\in(0,1)$ such that
$$
\mathcal E_{k+1}\le \rho_k\,\mathcal E_k,
$$
and for the canonical choice of $\alpha_k$ gives the explicit bound
$$
\mathcal E_{k+1}\le \Bigl[1-\tfrac12\min\{\tfrac{\alpha_k\gamma_k}{1+\alpha_k\gamma_k},\;\alpha_k\beta_k\}\Bigr]\mathcal E_k
$$
[2312.12355].

For generalized Wasserstein gradient flows, the VPTPD iteration combines proximal splitting with Hessian-derived metrics. Under convexity of $\Phi_h$, strong convexity of $\hat H$, Lipschitz continuity of $\nabla\mathcal J_h$ with respect to the $I_u$ metric, full-row-rank $B$, and step-size bounds, the iteration “converges globally at a linear (geometric) rate”:
$$
\|x^k-x^*\|_{I_u}+\|y^k-y^*\|_{I_p}\le C\rho^k
$$
for some $\rho\in(0,1)$ [2509.15385]. The same paper also proves “existence and uniqueness of bounded solutions for the resulting generalized proximal operator,” together with a “convergent and bound-preserving Newton solver” [2509.15385].

DualTPD emphasizes strong monotonicity and mesh independence. Under the assumptions on $F^*$ and the inf–sup stability of $D$, the preconditioned nonlinear saddle map remains strongly monotone; a Lyapunov energy
$$
V(\sigma,u)=\tfrac12\|\sigma-\sigma^*\|_{M_\sigma}^2+\tfrac12\|u-u^*\|_{M_u}^2
$$
is shown to decrease, and the contraction factor is bounded away from $1$ independently of mesh size, so “the iteration count is mesh-independent” [2510.15351].

The variable-metric Douglas–Rachford paper establishes weak convergence under bounded, self-adjoint, positive semidefinite, convergent preconditioners $P_k,Q_k$, summable metric variation, and single-valued resolvents; the proof uses quasi-Fejér arguments, Opial’s lemma, and the variable-metric proximal-point theorem [2302.13128]. The continuous-time preconditioned dynamics paper, by contrast, derives “non-ergodic decay of the primal–dual gap” under asymptotically antisymmetric preconditioners, while warning that “in the symmetric case, the non-ergodic gap need not converge” [2506.00501].

## 6. Applications, empirical behavior, and practical interpretation

The application range of VPTPD is unusually broad. In the OVDP paper, the three demonstration problems are “mixed noise removal of hyperspectral images, hyperspectral unmixing, and graph signal recovery” [2301.08468]. For hyperspectral mixed-noise removal with SSTV regularization, “OVDP2 typically cut the required iterations/time by 3×–10× compared to unpreconditioned PDS.” For hyperspectral unmixing in CLSUnSAL, “OVDP1/2 reduced runtime roughly 5× over scalar PDS and 10×–20× over ASP/PDP.” For graph-signal recovery with GTV sampling, “OVDP2 gave the fastest convergence in both iterations and CPU time” [2301.08468].

In generalized Wasserstein gradient flows, the tested simulations range “from 1D to 3D settings” and include “saturation equation, Cahn–Hilliard, droplet wetting.” The reported gains are: “Reduced iteration counts by up to 99 % over basic PDJKO and 70–90 % over PrePDJKO,” “Achieved 4–20× CPU–time speedups, especially in higher dimensions,” “Preserved energy dissipation, mass conservation and bounds exactly,” and “Adaptive step-size further accelerated convergence by 20–30 %” [2509.15385].

In the TPDv paper, two nonlinear PDEs are used: Darcy–Forchheimer flow and a nonlinear Maxwell problem. For Darcy–Forchheimer, the excerpted table for refinement levels $h=1/64$–$1/512$ states that “iteration counts remain constant; CPU time grows $\approx O(N)$.” For the nonlinear Maxwell problem, “VPTPD requires 30–50 outer iterations, each costing 2–3 V-cycles, vs. 70–100+ iterations in the other methods. CPU times are 2–5× faster in large-scale tests” [2312.12355].

DualTPD extends this PDE performance picture. For the $p$-Laplacian with $p=1.5,4,10$, the reported behavior is “iteration counts $\approx5$–$25$ independent of $h\in[1/32,1/512]$, CPU time $\sim\mathcal O(N)$,” and “for $p<2$, DualTPD-J exhibits superlinear (Newton-like) behavior from zero initial data.” For nonlinear $H(\curl)$ problems, including ferromagnetism and the $p$-curl analogue, the reported results are “15–18 DualTPD iterations, $\sim9$–10 PCG inner steps, iteration count independent of mesh $h\in\{1/4,1/8,1/16,1/32,1/64\}$, CPU time growing linearly with DoF.” The paper further states that, compared with “a ‘plain’ primal–dual (Chambolle–Pock) solver or the primal TPD method, DualTPD is 2–5× faster and has far fewer iterations” [2510.15351].

The continuous-time preconditioned dynamics experiments support the asymptotically antisymmetric design principle. In quadratic minimization, linearly constrained quadratic programs, and LASSO, “the antisymmetric variant decays orders of magnitude faster than either the symmetric preconditioning or the un-preconditioned flow,” and “antisymmetric again gives the fastest primal–dual gap decay” [2506.00501].

A common misconception is to equate VPTPD with one specific update formula. The literature instead shows a stable conceptual core—transformed primal-dual structure, variable preconditioning, and preserved tractability of the primal and dual subproblems—but multiple realizations adapted to split convex models, JKO schemes, Douglas–Rachford factorizations, and nonlinear PDE discretizations [2301.08468] [2312.12355] [2509.15385] [2510.15351]. A plausible implication is that future developments will continue to differ mainly in how they choose the metric operators and in which structural invariants—such as bound-preservation, energy dissipation, or mesh-independence—they enforce.

Source: https://www.emergentmind.com/topics/variable-preconditioned-transformed-primal-dual-vptpd