DUT-8: Tunable Disorder in a Nickel MOF

• DUT-8 is a nickel-based metal–organic framework with a square-grid pillared architecture and two-up/two-down disorder rule.
• Guest adsorption reversibly shifts its correlated disorder state, offering nonlinear diffuse X-ray scattering signatures for readout.
• Its configurational manifold, modeled via a six-vertex system, underpins its emerging role as a solid-state reservoir computer.

DUT-8 is a nickel-based metal–organic framework (MOF) in which correlated structural disorder can be tuned reversibly by chemistry. In recent work, it has been used as a proof-of-concept system for solid-state reservoir computing, because guest adsorption shifts the framework between distinct correlated-disorder states whose diffuse X-ray scattering signatures provide nonlinear, information-rich readouts (Greenbaum et al., 20 Feb 2026). In this setting, DUT-8 is notable not because it contains ordinary defects, but because it realizes a highly degenerate, topologically constrained configurational manifold whose state can be modulated by external chemical perturbation.

1. Framework architecture and the local disorder rule

DUT-8 was originally introduced as a framework built from Ni$_2$-carboxylate paddlewheel units linked into columns by dabco and interconnected laterally by 2,6-naphthalene dicarboxylate (ndc) linkers. The columns lie on a square grid, so the underlying topology is that of a pillared square lattice. The decisive geometric feature is the asymmetry of the ndc linker: each ndc connection shifts one neighboring dabco-linked column relative to the other either “up” or “down” along the column axis (Greenbaum et al., 20 Feb 2026).

For an isolated pair of columns, the up/down choice is energetically equivalent. Across the extended lattice, however, these local choices are globally constrained. Around each square channel, the framework must return to the same height after one circuit, and therefore the sequence of shifts around every plaquette must contain exactly two up steps and two down steps. This two-up/two-down condition is the central local rule governing DUT-8 disorder.

That rule leaves exactly six allowed channel configurations. In the “jigsaw tile” representation adopted from earlier work, four of the six correspond to cyclic permutations of an up-up-down-down pattern and have $C_{2h}$-type geometry, while two correspond to up-down-up-down and have $D_{2d}$-type geometry. Any chemically valid DUT-8 configuration can therefore be represented as a compatible tiling of these six motifs. The resulting disorder is not random in the defect sense; it is correlated, topologically constrained disorder in the relative vertical positions of the framework columns.

2. Configurational manifold and six-vertex interpretation

Because many distinct global arrangements satisfy the same local two-up/two-down rule, DUT-8 possesses a highly degenerate configurational manifold rather than a small set of defective variants of one parent crystal structure. The system is explicitly placed in the context of the six-vertex model, a classic model of constrained disorder and frustration (Greenbaum et al., 20 Feb 2026). This identifies DUT-8 as a crystalline solid with an extensive number of structurally distinct but topologically allowed configurations.

The paper parameterizes this configurational landscape by two coarse-grained descriptors, $(\phi,\eta)$. They are defined as

$0 \le \phi \le 1,\qquad 0 \le \eta \le 1,$

where $\phi$ measures the tendency for neighboring square channels to alternate in their up/down pattern, and $\eta$ is the fraction of channels of the up-down-up-down type, i.e. the $D_{2d}$-symmetric grayscale jigsaw tiles. These coordinates obey the emergent constraint

$\phi \ge \eta.$

The allowed states therefore occupy a triangular region in $(\phi,\eta)$-space. Ordered states lie on the boundary of this triangular map, whereas the interior corresponds to disordered states. The significance of this construction is twofold. First, it shows that DUT-8 disorder is long-range correlated because local choices are coupled by lattice-wide compatibility conditions. Second, it explains why the framework offers many legal internal states for computation: violations of the two-up/two-down rule would create a mismatch across whole columns of carboxylate–paddlewheel connections and incur a very large energy penalty, so the system is effectively confined to the allowed manifold.

3. Responsive disorder and guest-induced disorder–disorder transitions

The defining concept introduced in the recent reservoir-computing work is responsive disorder: correlated disorder with many accessible configurations, where the actual disorder state can be shifted systematically by an external perturbation (Greenbaum et al., 20 Feb 2026). In DUT-8, the perturbation is guest adsorption. Earlier experiments had already shown that DUT-8 crystallites are not frozen into one disorder realization during synthesis; instead, exposure to different guest molecules causes reversible disorder–disorder transitions across the configurational landscape.

The proposed mechanism is that different guests prefer pore geometries of different symmetry, so host–guest interactions bias the balance between local channel types. In this language, the guest acts like a tunable chemical field that steers the framework toward one region of its configurational manifold. The specific transition highlighted is the reversible switching associated with DCM and DMF guests. States with larger $C_{2h}$0 mimic the effect of incorporating DCM into the pores, whereas lower-$C_{2h}$1 states correspond to replacing DCM by DMF.

This transition is not an order–disorder transformation into a unique endpoint. It is a disorder–disorder transition, in which one correlated disordered ensemble is transformed into another. That distinction is central to the physics of DUT-8. A common misconception is to treat the material as a flexible MOF with a few ordered polymorphs plus defects. The present description instead emphasizes a framework whose accessible states are distributed across a constrained, many-body disorder manifold and can be reversibly traversed by solvent exchange.

4. Diffuse scattering as a probe of correlated disorder

The experimentally and conceptually central observable in DUT-8 is diffuse X-ray scattering. The recent study uses X-ray powder diffraction, but focuses not on sharp Bragg peaks alone; the key information lies in the diffuse scattering produced by correlated disorder (Greenbaum et al., 20 Feb 2026). Bragg diffraction mainly reflects the average periodic structure, whereas DUT-8’s essential state information resides in how local motifs are arranged and correlated across the lattice.

In the computational workflow, DUT-8 configurations were generated by a loop-move Monte Carlo algorithm on a $C_{2h}$2 supercell with periodic boundary conditions. Loop moves were chosen because they preserve the physical constraints and ensure ergodicity within the allowed manifold. Each configuration was decorated using the coarse-grained DUT-8 structural representation from earlier work, and the corresponding powder X-ray diffraction pattern was calculated with TOPAS.

For the reservoir readout, the analysis selected the diffuse-dominated reciprocal-space window

$C_{2h}$3

rebinned it at $C_{2h}$4 intervals, and normalized the resulting intensities. This yields a 27-channel readout vector for each configuration. Operationally, the mapping is: a disorder state in DUT-8 produces an X-ray diffraction pattern; a diffuse-scattering window is extracted; that window is discretized into 27 intensity channels; and those intensities are used as the observable state.

The mapping from $C_{2h}$5 to scattering is nonlinear because diffuse intensity depends on spatial correlations rather than simply on motif counts. This is the physical basis for separability in machine-learning tasks. Different regions of the triangular disorder map produce recognizably different diffraction patterns, and intermediate correlated-disorder states do not reduce to simple linear interpolation between limiting ordered patterns.

5. DUT-8 as a solid-state reservoir computer

Within the reservoir-computing framework, DUT-8 serves as a fixed nonlinear dynamical system whose internal degrees of freedom are not trained; only a simple output layer is trained (Greenbaum et al., 20 Feb 2026). In the static setting, the input is a point in $C_{2h}$6-space, the physical reservoir is the correlated disorder state of DUT-8, the readout is the 27-channel diffuse-scattering signature, and the trained output is a single linear model.

For classification, the study sampled 3,000 Monte Carlo configurations across the configurational landscape. Using the 27-channel scattering signatures as features, it trained a single-layer linear support vector classifier on standard target functions defined on the $C_{2h}$7 inputs. One explicit example is an AND-like task with output $C_{2h}$8 only when

$C_{2h}$9

The reported result is that classification accuracy is consistently better than direct linear classification on the raw $D_{2d}$0 coordinates and is comparable to the previously reported formose reaction reservoir.

For dynamic tasks, the study introduced a minimal effective guest-bias model through the Monte Carlo Hamiltonian

$D_{2d}$1

with $D_{2d}$2 an externally imposed effective chemical field and $D_{2d}$3 the fraction of $D_{2d}$4-type channels. This coupling is described as a chemical Zeeman term. The field is driven sinusoidally,

$D_{2d}$5

and Monte Carlo sampling at each time step uses

$D_{2d}$6

The previous configuration is carried forward to the next time step, so the state evolves continuously rather than being reinitialized.

This dynamics yields a fading memory. Structural reorganization proceeds through collective rearrangements of loops of dabco-linked columns, and the set of loops that can move depends on the current state. Consequently, two times with the same instantaneous applied field can correspond to different DUT-8 configurations if their prior trajectories differ. Using the diffuse-scattering readouts, the study trained a single regression layer for time-series transformations, following the methodology previously used for artificial spin-vortex ice reservoirs. It reports successful learning of symmetric and asymmetric target functions and emphasizes the sawtooth function as evidence of fading memory. The dynamic training protocol used 200 time steps, and performance was reported in terms of mean-squared deviations, numerically comparable to those of the vortex-spin-ice reservoir.

6. Broader significance, materials context, and limitations

DUT-8 occupies a distinctive position in the literature on MOFs and correlated disorder. It is highlighted as the most prominent MOF in which experimentally characterized disorder–disorder transitions have already been demonstrated, and its six-vertex connection links it directly to classic frustrated models (Greenbaum et al., 20 Feb 2026). The broader comparison is to other materials where diffuse scattering reveals correlated disorder, including Prussian blue analogues and disordered rocksalts. The recent work further suggests that analogous responsive disorder may exist in other classes of materials, including liquid-crystal assemblies, hydrogen-bonded networks, and orbital-molecule liquids.

The computational interpretation of DUT-8 follows directly from its physics. Nonlinearity arises from the complex mapping between guest bias or disorder coordinates and diffuse-scattering signatures. Memory arises from history-dependent loop rearrangements during guest-driven disorder–disorder transitions. Separability arises because low-dimensional inputs are expanded into a richer scattering feature space. A plausible implication is that other solids with externally tunable correlated disorder may also support reservoir-style information processing if their disorder can be read out by a sufficiently sensitive probe.

The present demonstration nonetheless remains a proof of concept. The chemical-field Hamiltonian is explicitly a minimal effective model, not a microscopic host–guest energy model. The workflow is based on simulated disorder states and calculated diffraction patterns rather than in situ experimental reservoir operation. Future work is therefore directed toward direct experimental scattering readouts and toward practical issues including guest-exchange kinetics, strain effects, and particle size. The study also suggests exploring other MOF chemistries and alternative input/output choices, such as mechanical stress as input and infrared spectroscopy as readout.

In this synthesis, DUT-8 is not merely a flexible porous solid. It is a material whose square-grid, pillared architecture and ndc-induced up/down shifts generate a six-vertex-type manifold of correlated disordered states, whose position within that manifold can be shifted reversibly by guest chemistry, and whose disorder can be interrogated by diffuse scattering. That combination makes DUT-8 both a subject of condensed-matter and MOF physics and a model system for chemically programmable, solid-state physical computing (Greenbaum et al., 20 Feb 2026).