The Equilibrium Is the Initialization: Lazy Identity Collapse in Physics-Structured Deep Equilibrium Reasoning

Joyjeet Singh

cs.LGJul 13, 2026

The Equilibrium Is the Initialization: Lazy Identity Collapse in Physics-Structured Deep Equilibrium Reasoning

Authors: Joyjeet Singh

Organizations: Independent Researcher

Abstract

Deep equilibrium models promise input-adaptive implicit computation: harder problems should demand more solver iterations, and the solved equilibrium should encode the result of genuine iterative inference. We report a cautionary study of a port-Hamiltonian DEQ with a learned initialization on two reasoning tasks -- ProofWriter entailment over frozen DeBERTa embeddings and a BFS-verified graph-reachability benchmark -- in which the implicit computation is a silent no-op. Across tasks, seeds, and controlled ablation arms, the solved equilibrium equals the solver's start point to numerical precision, and bypassing the solver entirely changes test accuracy by +0.00 percentage points in 18 of 19 training runs. Controlled interventions falsify the tempting explanation: removing the anchoring term reproduces every result, and retraining with noise-decoupled starts yields a solver that converges to the noisy start while the decoder learns to ignore it. The single escaping run diverges instead ( $\|h^{*}-z_0\|=171$ ), producing a co-adapted noise channel whose removal improves accuracy. Iteration counts are uncorrelated with ground-truth difficulty ( $r=0.009$ ), and the full apparatus never outperforms a two-layer MLP on either task. We trace the mechanism to gradient starvation along two distinct routes, show that the standard zeroing ablation is confounded and gives wildly seed-dependent answers where the correct substitution test gives a stable zero, and distill a four-test diagnostic protocol for auditing claimed implicit computation. All experiments run on a single free Colab GPU; code, raw logs, and analysis scripts are released.

Explore similar work

Sep 15, 2026stat.ML

Certified Inference and Training for Deep Equilibrium Networks: A Continuation Framework with Polynomial Complexity Guarantees

We develop a certified continuation framework for equilibrium computation and for training deep equilibrium networks (DEQs), with training formulated as interpolation to accuracy

2^{-b}

. For inference, compact input homotopy selects a unique branch from a supplied start root, and a rounded Newton tracker follows it under certified boundary, conditioning, derivative, and tube-radius bounds. For training, we augment local-plus-low-rank recurrence with programmable dormant bilinear rank-one channels. Loaded Tikhonov solves diagnose a failed interpolation pass without spectral decomposition; an output-preserving repair aligned with the pass residual supplies the required direction. Training requires certified gate realization and column stability on each pass region, well-posed inference, and finite-update error budgets. With polynomial geometric, encoding, precision, and complete backend budgets, both certified inference and training have bit cost

O(\operatorname{poly}(L+b))

, where

L

is the encoded instance length. The trainer uses

O(b+\ell)

passes and reserve channels from an initial residual bounded by

2^\ell

. These guarantees concern a certified promise class. Lean 4 verifies the quantitative core and concrete inference backend; numerical comparisons illustrate the loaded mechanism.

Alex Borisevich

Graph Neural Network Inference Finite-Sample Certificates

Ask PDF

May 20, 2026cs.LG

Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning

Scaling test-time compute by iteratively updating a latent state has emerged as a powerful paradigm for reasoning. Yet the internal mechanisms that enable these iterative models to generalize beyond memorized patterns remain unclear. We hypothesize that generalizable reasoning arises from learning task-conditioned attractors: latent dynamical systems whose stable fixed points correspond to valid solutions. We formalize this process through Equilibrium Reasoners (EqR), which enable test-time scaling without external verifiers or task-specific priors. EqR scales internal dynamics along two axes: depth, by running more iterations, and breadth, by aggregating stochastic trajectories from multiple initializations. Empirically, gains from test-time scaling are tightly coupled with stronger convergence toward solution-aligned attractors. This attractor perspective allows neural networks to adaptively allocate test-time compute based on task difficulty. While simple cases converge within 1 to 5 iteration steps, harder cases benefit from massive test-time scaling. By unrolling up to the equivalent of 40,000 layers, scalable latent reasoning boosts accuracy from 2.6% for feedforward models to over 99% on Sudoku-Extreme. These results suggest that learned attractor landscapes provide a useful mechanistic lens for understanding scalable reasoning in iterative latent models.

Benhao Huang, Zhengyang Geng, Zico Kolter

Test-Time Scaling Large Language Model Scale Attractors Latent States

Ask PDF

Jun 25, 2026cs.LG

Retrieval-Warmed Energy-Based Reasoning: A Five-Arm Ablation Methodology for Diffusion-as-Inference on Structured Reasoning Tasks

Warm-started diffusion samplers accelerate iterative inference, but it is rarely clear which part of the pipeline carries the gain. We study \textbf{retrieval-warmed energy-based reasoning (RW-EBR)} -- an IRED energy-based diffusion model \cite{du2024ired} augmented with a Modern Hopfield trajectory memory -- and contribute a \textbf{five-arm ablation methodology} (oracle, best-constant, per-query-random, shuffled, aligned) that separates three confounded effects: class-prior bias shift, stochastic warm-starting, and graph-aligned value reuse. The diagnostic decomposition is adapted from LLM-RAG evaluation \cite{ru2024ragchecker}. On \textbf{connectivity-2} (Erdős--Rényi all-pairs reachability), the aligned-vs-shuffled-oracle swing reaches \textbf{

+35

,pp} balanced accuracy on a fixed 1{,}000-graph validation-set diagnostic, with value distribution and retrieval mechanics fixed, only per-graph alignment destroyed, while per-query random initialisation falls below cold -- per-graph alignment, not bias shift or stochasticity, dominates. Yet the \emph{deployable} cold-prediction pipeline misses the acceptance gate at stored-value quality. The same diagnostic logic, stopped at the key-quality screen, applied to \textbf{Sudoku} with a task-specific key encoder produces a clean negative at a \emph{different} component -- key quality, under the current setup. The decomposition names the first blocking component on each task. The setting -- graph reachability refined by an iterative diffusion sampler, with explainability of failure modes as the lens -- places the work within structured and spatio-temporal reasoning.

Libo Sun, Po-Wei Harn, Zewei Zhang +2

Inference-Time Ablation Diffusion Models Single-Message Shuffle Model

Ask PDF