---
title: 'DeepFingers: Hybrid Operator Learning for VF'
url: https://www.emergentmind.com/topics/deepfingers
type: topic
---

# DeepFingers: Hybrid Operator Learning for VF

Searching arXiv for the DeepFingers paper and the operator-learning papers it builds on.
DeepFingers is a hybrid operator-learning framework for predicting the spatiotemporal evolution of **miscible viscous fingering (VF)** in unstable flow regimes. It was introduced in the context of a broader claim that AI models of hydrodynamically unstable transport can produce **hallucinations**: predictions that appear visually coherent yet are physically unrealistic, including **spurious fluid interfaces** and **reverse diffusion that violate conservation laws** [2604.20372]. The framework combines the branch–trunk structure of **DeepONet** with **FNO** and **U-FNO** components in order to enforce what the paper describes as **“balanced learning across the full spectrum of spatial modes”**, with the specific learned operator
\[
\mathcal{G} : (c_t, \xi) \mapsto c_{t+1}, \qquad \xi=[M,t+1].
\]
Here \(c_t\) and \(c_{t+1}\) are concentration snapshots on a \(64\times128\) grid, and \(M\) is the viscosity ratio [2604.20372].

## 1. Origin and problem setting

DeepFingers was proposed for **miscible viscous fingering**, the instability that occurs when a less viscous fluid displaces a more viscous one in a porous medium or Hele-Shaw geometry [2604.20372]. In the miscible setting, the interface is not a sharp material boundary but a concentration transition zone, and the paper formulates the learning problem in terms of the concentration field \(c(\mathbf{x},t)\), with viscosity depending exponentially on concentration and transport governed by mass and momentum conservation [2604.20372].

The paper emphasizes that VF is difficult both numerically and statistically because the dynamics are **nonlinear, nonstationary, multiscale, and highly sensitive to the viscosity ratio \(M\)** and to perturbations in the initial state [2604.20372]. Small perturbations amplify into narrow fingers, followed by **tip splitting, merging, shielding, and eventually channel formation**. This makes the task more demanding than predicting a low-dimensional diagnostic: the learned object is the full spatiotemporal evolution of concentration snapshots rather than a reduced scalar observable [2604.20372].

A central premise of the work is that unstable flow exposes a failure mode in scientific ML analogous to hallucination in language models. The paper does not define a single scalar hallucination score. Instead, it characterizes hallucinations conceptually as predictions that are **“visually coherent but physically unrealistic”** or that **“violate fundamental physical laws or deviate from expected flow dynamics”** [2604.20372]. This framing is specific to VF because visually plausible fronts can still conceal incorrect interfacial topology, mixing behavior, or transport ordering.

## 2. Hallucination in unstable-flow prediction

The paper identifies concrete hallucination modes in baseline models. In **DAE-LSTM**, early-time predictions contain **spurious yellow patches** of low-viscosity fluid embedded inside black regions of high-viscosity fluid; at later times, fingers become overly smooth, with diffused tips and insufficient splitting [2604.20372]. In **ViT**, late-time predictions generate **nonphysical black islands within yellow regions**, violating expected spatial continuity of the displacement process [2604.20372].

The failures are not treated as merely aesthetic defects. The paper interprets them as conservation and transport violations because they imply nonphysical creation of isolated blobs or interfaces, backward or reverse mixing behavior, or transport trajectories inconsistent with advection–diffusion and monotone trend expectations [2604.20372]. One example given is that, for \(M=13\), baseline models can predict breakthrough too early, or even predict that breakthrough does not decrease monotonically with increasing \(M\), contradicting DNS and the expected physics of stronger instability at larger viscosity contrast [2604.20372].

This analysis leads to a broader methodological claim: hallucinations in unstable flow are not simply generic prediction errors. The paper argues that they arise when the model distorts the physically relevant interaction between low-frequency bulk motion and high-frequency interfacial modes [2604.20372]. A plausible implication is that visually plausible rollouts may be insufficient as a validation criterion for operator learners in instability-dominated PDE systems.

## 3. Spectral-bias explanation

The proposed explanation for these failures is **spectral bias** [2604.20372]. The paper argues that standard neural architectures tend to learn an unbalanced spectrum of modes, and that this imbalance becomes especially damaging in VF because the flow contains both domain-scale structure and fine interfacial detail [2604.20372]. If higher modes are over- or under-represented, the resulting rollouts may invent interfaces, smear tips, suppress tip splitting, or create isolated islands.

For diagnosis, the paper uses **wavelet analysis rather than pure Fourier analysis**, on the grounds that VF is nonstationary and wavelets localize in both scale and time [2604.20372]. The reported spectral signatures are specific. **ViT** underestimates the dominant large-scale mode 1 while overestimating higher modes \(2\!-\!7\), corresponding to exaggerated small-scale fluctuations and nonphysical artifacts [2604.20372]. **DAE-LSTM** exhibits \(M\)-dependent bias; for larger \(M\), higher modes \(3\!-\!7\) fall below DNS, consistent with too few fingers and overly diffused boundaries [2604.20372]. By contrast, **DeepFingers** matches DNS much more closely across all modes, which the paper interprets as evidence of **balanced learning across the full spectrum of spatial modes** [2604.20372].

This spectral interpretation is the immediate rationale for the architecture. DeepFingers is not presented as a model with an explicit spectral regularization term or a formal “spectral debiasing loss.” Rather, the phrase is architectural and interpretive: the model is designed so that Fourier-domain operators capture global nonlocal dependence while U-Net-style multiscale processing restores fine interfacial structure that truncated Fourier modes alone may miss [2604.20372]. This suggests that the framework’s main novelty lies in inductive bias rather than in a new optimization objective.

## 4. Architecture and operator formulation

DeepFingers combines **DeepONet** [1910.03193] with **FNO** [2010.08895] and **U-FNO** components, while preserving a branch–trunk decomposition [2604.20372]. The paper recalls the FNO operator form
\[
\mathcal{G}: A \mapsto B,
\]
with
\[
B = \mathcal{G}(A) = \bigl( \mathcal{Q} \circ \mathcal{L}^{(L)} \circ \cdots \circ \mathcal{L}^{(1)} \circ \mathcal{P} \bigr)(A),
\]
and layer update
\[
\mathcal{L}^{(\ell)}\!\left(z^{(\ell)}\right) = \sigma \!\left(W^{(\ell)} z^{(\ell)} + K^{(\ell)}\!\left(z^{(\ell)}\right)\right),
\]
where
\[
K^{(\ell)}\!\left(z^{(\ell)}\right) = \mathcal{F}^{-1}\!\left( R^{(\ell)} \cdot \mathcal{F}\!\left(z^{(\ell)}\right)\right)
\]
uses learned complex tensors on truncated Fourier modes [2604.20372]. The U-FNO layer is written as
\[
\mathcal{U}^{(\ell)}\!\left(h^{(\ell)}\right) = \sigma \!\left(W^{(\ell)} h^{(\ell)} + U^{(\ell)}\!\left(h^{(\ell)}\right) + K^{(\ell)}\!\left(h^{(\ell)}\right)\right),
\]
with \(U^{(\ell)}\) a U-Net-style multiscale convolutional operator [2604.20372].

DeepONet is reviewed in the paper as learning a nonlinear operator with a branch network for the input function and a trunk network for the query variable, with output
\[
B = \mathcal{G}(A_m)(y) = \sum_{i=1}^q b_i(A_m)\,\varphi_i(y)
\]
for discretized input
\[
A_m = \bigl(A(r_1), \dots, A(r_m)\bigr)
\]
[2604.20372]. DeepFingers retains this branch–trunk logic, but the branch encoder is replaced by an FNO layer and the merged representation is refined with two U-FNO layers [2604.20372].

The architecture specified in the paper contains five stages:

| Component | Input | Role |
|---|---|---|
| Branch network | \(c_t\) | Single FNO layer with \(L=1\) |
| Trunk network | \(\xi=[M,t+1]\) | Fully connected conditioning network |
| Merge | branch + trunk outputs | Pointwise combination into \(C_{\text{out}}=64\) channels |
| Refinement | merged state | Two U-FNO layers |
| Projection | refined state | Map to single concentration channel |

The merge operation is
\[
h^{(0)} = b_{out} \odot t_{out},
\]
followed by
\[
h^{(1)} = \sigma \!\left(
K^{(1)}(h^{(0)})
+ U^{(1)}(h^{(0)}) + W^{(1)} h^{(0)}
\right),
\]
\[
h^{(2)} = \sigma \!\left(
K^{(2)}(h^{(1)})
+ U^{(2)}(h^{(1)}) + W^{(2)} h^{(1)}
\right),
\]
and final projection
\[
c_{t+1} = \mathcal{S}\!\left(h^{(2)}\right)
\]
[2604.20372].

The learned operator is therefore explicitly
\[
\mathcal{G} : (c_t, \xi) \mapsto c_{t+1}, \qquad \xi=[M,t+1].
\]
Inputs and outputs both have shape \((1,64,128)\), and the merged hidden state has shape \((64,64,128)\) [2604.20372]. Time and viscosity ratio enter only through the trunk, while the concentration field enters through the branch FNO [2604.20372]. This means DeepFingers does not learn a continuous-in-time field solution directly; it learns a **successor operator** between concentration snapshots [2604.20372].

## 5. Training regime and evaluation methodology

The training data are generated by **DNS simulations** of VF over a range of viscosity ratios \(M\), with concentration maps discretized on a \(64\times128\) grid [2604.20372]. Prediction is performed **auto-regressively in time**: the model is initialized from a prescribed initial condition and rolled out by repeatedly feeding \(c_t\) and \([M,t+1]\) to produce \(c_{t+1}\) [2604.20372]. The visible text does not provide an explicit loss formula, optimizer, learning-rate schedule, epoch count, or train/validation split, and it does not describe any conservation-law regularizer or PDE residual penalty [2604.20372]. Its “physics-based” character derives from operator design and physics-based diagnostics rather than from constrained training.

The evaluation emphasizes physically meaningful behavior rather than pixelwise image error [2604.20372]. The paper argues that because VF is intrinsically unstable, exact pixel alignment with DNS is not the appropriate standard. Instead it uses:

- **wavelet spectral energies by mode**
- **domain-averaged concentration** \(\bar c(t)\)
- **breakthrough concentration at the outlet** \(\bar c_{\text{out}}(t)\)
- **degree of mixing** \(\sigma_c^2\)
- **rate of mixing** \(\epsilon_c\)

[2604.20372]

The principal viscosity-ratio regimes explicitly discussed are
\[
M=13,\quad M=28,\quad M=55,
\]
ranging from weak to strong instability [2604.20372]. The baselines are **DAE-LSTM**, **ViT**, and, in a separate mitigation experiment, a modified ViT with cross-attention inspired by CViT [2604.20372]. The paper notes that cross-attention removes the black-island artifact, but still yields short and diffused fingers for \(M=13\), which suggests that temporal conditioning alone does not resolve the multiscale instability problem [2604.20372].

## 6. Reported performance, qualitative behavior, and limitations

The core aggregate RMSE table against DNS is:

| Model | \(\bar c\) RMSE | \(\bar c_{\text{out}}\) RMSE | \(\sigma_c^2\) RMSE |
|---|---:|---:|---:|
| DeepFingers | 0.00410 | 0.00966 | 0.00327 |
| ViT | 0.11463 | 0.02389 | 0.00523 |
| DAE-LSTM | 0.05334 | 0.03254 | 0.00385 |

For the mixing-rate metric, the paper reports:
- **DeepFingers**: \(\epsilon_c\) RMSE \(= 2.67\times 10^{-7}\)
- **ViT**: \(9.10\times10^{-7}\)
- **DAE-LSTM**: \(8.66\times10^{-7}\)

[2604.20372]

These numbers show DeepFingers performing best on every reported global metric [2604.20372]. Qualitatively, it reproduces **short slowly advancing fingers** at low \(M\), **accelerated growth** and **more frequent tip splitting** at higher \(M\), and at \(M=55\) even the onset of **channeling** [2604.20372]. The paper explicitly states that it captures **tip splitting, finger merging, and channel formation** [2604.20372].

The evidence for reduced hallucination is both visual and metric-based. DeepFingers avoids the **yellow patches** found in DAE-LSTM and the **black islands** found in ViT [2604.20372]. Spectrally, it tracks DNS mode redistribution more faithfully [2604.20372]. In global transport diagnostics, it preserves the expected monotonic relationship between instability strength and breakthrough behavior [2604.20372]. In uncertainty tests over varying initial conditions, its predicted PDFs for \(\bar c\), \(\bar c_{\text{out}}\), \(\sigma_c^2\), and \(\epsilon_c\) remain close to DNS [2604.20372].

The paper is also explicit about limitations. It does not provide a rigorous quantitative hallucination score, nor an explicit conservation-penalty term in training [2604.20372]. The governing PDEs and some metric definitions are deferred to supplementary material [2604.20372]. The framework is trained and evaluated on a fixed \(64\times128\) grid for **2D concentration-field evolution**, and generalization to different geometries, resolutions, or full coupled velocity–pressure prediction is not shown [2604.20372]. Because the rollout is autoregressive, some accumulation of rollout error remains possible [2604.20372]. Computational cost is not reported in the visible text, though the architecture is acknowledged to be more complex than a plain ViT or a simple latent-space recurrent model [2604.20372].

## 7. Position within operator learning and scientific machine learning

DeepFingers occupies a specific position within operator learning for PDE surrogates. Relative to plain **FNO**, it adds a **DeepONet-style trunk** for conditioning on regime and rollout time, and relative to plain **DeepONet**, it adds explicit Fourier-domain and multiscale convolutional operators in the branch and refinement stages [2604.20372]. A plausible implication is that the framework is best understood not as a generic sequence predictor, but as a regime-conditioned successor operator specialized to unstable, multiscale transport.

Its broader scientific claim is that unstable flow reveals a failure mode of AI models that ordinary image realism or pointwise error metrics may miss [2604.20372]. The paper’s central mechanistic thesis is that **hallucinations in unstable flow emerge when model inductive bias distorts the physical spectrum of interacting scales** [2604.20372]. In that sense, DeepFingers is both a predictive architecture and a proposed diagnostic perspective on scientific ML.

This perspective also clarifies a common misconception. DeepFingers is not described as a physics-informed neural network in the sense of enforcing PDE residuals during training. The visible text contains **no explicit conservation-law regularizer, PDE residual loss, or mass-balance constraint** [2604.20372]. Its physical plausibility comes instead from operator design, regime conditioning, and evaluation by physics-based observables. Another potential misconception is that the model solves hallucination through temporal conditioning alone; the cross-attention ViT experiment is presented precisely to argue that temporal conditioning helps but does not fully solve the multiscale instability problem [2604.20372].

In summary, DeepFingers is a hybrid operator learner for unstable porous-media transport that predicts
\[
(c_t,M,t+1)\mapsto c_{t+1}
\]
by combining **DeepONet-style conditioning**, **FNO-based global spectral operators**, and **U-FNO multiscale refinement** [2604.20372]. Its significance lies less in a new loss function than in the claim that physical plausibility in unstable flow depends on balancing low- and high-frequency spatial modes, and that failures to do so produce hallucinations even when predicted flow fields look qualitatively plausible [2604.20372].

Source: https://www.emergentmind.com/topics/deepfingers