Butter: Emulsion Dynamics and Texture Design

• Butter is a water-in-oil emulsion produced from cream through phase inversion, involving aeration, coalescence, and flocculation.
• Its processing dynamics, modeled via coupled lattice equations and state diagrams, reveal temperature-dependent routes that determine macro and microstructural qualities.
• Butter research spans food science to engineering, influencing studies in robotic melting detection, nutrient taxation, and even technical applications like computer vision and supergravity.

Searching arXiv for the cited papers to ground the article with current source metadata. arxiv_search query="(Kawaharazuka et al., 2024) OR (Dahl et al., 2023) OR (Nozawa et al., 2024) OR (Nozawa, 21 Jan 2026) OR (Cholakova et al., 2021) OR (Bellincioni et al., 10 Mar 2026) OR (Lin et al., 12 Jul 2025) OR (Banerjee et al., 2016)" max_results=10 sort_by="relevance"

arxiv_search(query="(Kawaharazuka et al., 2024) OR (Dahl et al., 2023) OR (Nozawa et al., 2024) OR (Nozawa, 21 Jan 2026) OR (Cholakova et al., 2021) OR (Bellincioni et al., 10 Mar 2026) OR (Lin et al., 12 Jul 2025) OR (Banerjee et al., 2016)", max_results=10, sort_by="relevance") Butter is the water-in-oil end state of phase inversion from fresh cream, which begins as an oil-in-water emulsion of water and milk fat globules and, under mechanical whipping, passes through a whipped-cream foam stage before inverting to butter (Nozawa et al., 2024). In kitchen settings, butter also exemplifies a continuous state change: a solid pat softens, spreads, liquefies, and becomes fully melted, often with transient phases such as softening edges, pooling, partial film, and complete film (Kawaharazuka et al., 2024). Across recent research, butter is therefore treated not only as a food item with high saturated fat content, but also as a model system for phase inversion, texture design, robotic state recognition, melt-lubricated sliding, and nutritional taxation (Dahl et al., 2023, Bellincioni et al., 10 Mar 2026, Cholakova et al., 2021).

1. Phase inversion from fresh cream to butter

Fresh cream is modeled as an oil-in-water emulsion whose interfaces among water, milk fat globules, and air bubbles undergo interface deformation, aeration, partial or complete coalescence, and flocculation during whipping. In the coupled map lattice formulation proposed for dairy processing, the system is represented on a two-dimensional square lattice with wall boundary conditions and coarse-grained field variables: surface energy $s_{ij}^t$, cohesive energy $c_{ij}^t$, emulsion energy $h_{ij}^t=s_{ij}^t+c_{ij}^t$, and velocity $\mathbf{v}_{ij}^t$. The minimal procedure set consists of whipping $T_w$, coalescence $T_c$, and flocculation $T_f$, intended to retain only the nonlinear maps needed to reproduce the observed inversion routes (Nozawa et al., 2024).

Temperature enters through the threshold $\theta$ of membrane surface activity. High whipping temperature corresponds to smaller $\theta$ and favors rapid coalescence with limited sustained aeration; low whipping temperature corresponds to larger $\theta$ and favors extended aeration with delayed coalescence. The model reproduces two well-known phase inversion routes. At high temperature, the route is viscosity dominance: low-aerated whipped cream $c_{ij}^t$0 deaerated cream $c_{ij}^t$1 soft butter. At low temperature, the route is overrun dominance: highly aerated whipped cream $c_{ij}^t$2 aerated intermediate $c_{ij}^t$3 hard butter. Simulated overrun and viscosity trends are reported as qualitatively consistent with experiments at approximately $c_{ij}^t$4 and $c_{ij}^t$5, respectively (Nozawa et al., 2024).

The mechanical interpretation is explicit. Whipping deforms milk fat globule membranes and raises local surface energy; coalescence converts surface energy into cohesive energy as surface activity drops; flocculation steers flow along emulsion-energy gradients and promotes demulsification. Butter texture is therefore treated as an emergent consequence of coupled aeration, coalescence, and flocculation rather than of a single scalar process variable (Nozawa et al., 2024).

2. Macro-texture, micro-structure, and state diagrams

A later multi-scale extension makes the connection between macroscopic textural quality and microscopic structural quality explicit. Overrun is defined from initial and final volumes as $c_{ij}^t$6, with air volume fraction $c_{ij}^t$7. Viscosity is represented in the coupled map lattice by cohesive energy $c_{ij}^t$8, whereas overrun is represented by surface energy $c_{ij}^t$9 (Nozawa, 21 Jan 2026).

Using the Young-Laplace equation, the model derives bubble and grain sizes from macroscopic rheology. With pressures identified as $h_{ij}^t=s_{ij}^t+c_{ij}^t$0 and $h_{ij}^t=s_{ij}^t+c_{ij}^t$1, and with effective interfacial tension $h_{ij}^t=s_{ij}^t+c_{ij}^t$2, the characteristic lengths become

$h_{ij}^t=s_{ij}^t+c_{ij}^t$3

and the density of clad complexes is

$h_{ij}^t=s_{ij}^t+c_{ij}^t$4

The inversion point is the balance point $h_{ij}^t=s_{ij}^t+c_{ij}^t$5, where $h_{ij}^t=s_{ij}^t+c_{ij}^t$6. The paper interprets size selection as a “tug-of-war” between air bubbles and butter grains via their cohesion pressures, while density evolution is described as a “costume change” of clad particles adapting to the current interfacial environment (Nozawa, 21 Jan 2026).

Two complementary state diagrams organize the process. On the macroscopic viscosity-overrun plane, the high-temperature and low-temperature routes appear as two parallel processes of viscosity dominance and overrun dominance. On the microscopic size-density plane, the same routes appear as two orthogonal processes: isodensity/size dominance at high temperature and isosize/density dominance at low temperature (Nozawa, 21 Jan 2026).

Route	Macroscopic path	Microscopic endpoint
High temperature	Viscosity-dominant	Large $h_{ij}^t=s_{ij}^t+c_{ij}^t$7, low $h_{ij}^t=s_{ij}^t+c_{ij}^t$8, uniform grains
Low temperature	Overrun-dominant	Small $h_{ij}^t=s_{ij}^t+c_{ij}^t$9, high $\mathbf{v}_{ij}^t$0, fractal grains

This dual-diagram formulation supports process design by microstructure. The reported guidance is direct: spreadable, creamy-soft butter is associated with larger grains, low composite density, rapidly decreasing air content, and uniform grain structure, whereas firm, fluffy-hard butter is associated with smaller grains, increasing composite density, sustained air content, and fractal grain structure (Nozawa, 21 Jan 2026).

3. Melting, cooking, and dynamical state estimation

Butter melting in a frying pan is treated as a prototypical continuous state change for cooking robots. The discrete label “melted or not” is insufficient because control decisions depend on how far along the process is and on when the transition crosses a practical completion threshold. The proposed recognition pipeline uses a pre-trained vision-LLM in retrieval mode. At time $\mathbf{v}_{ij}^t$1, the robot acquires an image $\mathbf{v}_{ij}^t$2 and computes per-prompt similarities

$\mathbf{v}_{ij}^t$3

A signed, normalized aggregation then yields a scalar trajectory,

$\mathbf{v}_{ij}^t$4

with $\mathbf{v}_{ij}^t$5 for prompts indicating that the change has occurred and $\mathbf{v}_{ij}^t$6 for prompts indicating that it has not. After a 3-second moving average and min-max scaling,

$\mathbf{v}_{ij}^t$7

the trajectory is fitted to

$\mathbf{v}_{ij}^t$8

Prompt weights $\mathbf{v}_{ij}^t$9 are optimized by a genetic algorithm with population size 300 and 300 generations, maximizing $T_w$0, where $T_w$1 is the sigmoid-fitting RMSE. Detection occurs when the smoothed and scaled signal first exceeds $T_w$2 (Kawaharazuka et al., 2024).

The reported butter prompts are dominated by antonyms such as “butter that is not melted in that frying pan” and “not melted butter in a frying pan,” with a smaller retained weight on “melted butter in frying pan.” This composition is said to increase dynamic range and stability because antonyms enter with positive weights but negative signs in the aggregation. In experiments at 10 Hz on a PR2 platform, ImageBind recognized butter melting with high accuracy across equal-weight, single-prompt, and optimized settings, whereas CLIP exhibited larger fluctuations and, under optimized weighting, a tendency toward slightly early detection on the optimization set (Kawaharazuka et al., 2024).

A distinct but related line of work analyzes butter on a hot, slightly inclined pan as a melt-lubricated sliding problem. A thin liquid film of thickness $T_w$3 forms beneath the solid, and the system evolves through a self-regulating feedback between melt-layer thickness, sliding velocity $T_w$4, and heat transfer. The conduction-limited Stefan condition is

$T_w$5

with $T_w$6 the melt rate per unit area. Combined normal and tangential force balances determine a unique steady triplet $T_w$7. In the thin-film limit,

$T_w$8

and

$T_w$9

The reported consequences are that increasing $T_c$0 thickens the film and increases $T_c$1, increasing $T_c$2 increases $T_c$3 and decreases $T_c$4, and increasing $T_c$5 slows $T_c$6 while thickening $T_c$7. The same framework is stated to apply to butter when butter’s thermophysical properties are supplied; the experimentally validated 2D theory predicts film thicknesses of approximately $T_c$8–$T_c$9 and melt rates of order $T_f$0 in analogous systems (Bellincioni et al., 10 Mar 2026).

4. Cocoa butter and low-energy nano-fragmentation

Cocoa butter appears in the literature as a natural triglyceride oil that can undergo the “cold-burst” process, a cooling-heating cycle intended to generate nanoemulsions or nanoparticles without high mechanical energy. The mechanism begins with rapid cooling, which produces a metastable $T_f$1 polymorph; subsequent $T_f$2 reorganization is accompanied by local volume shrinkage, nanovoid formation, and negative pressure inside the frozen particle. If wetting conditions are favorable, surfactant solution is drawn into the nanoporous network, the particle swells, and fragmentation follows (Cholakova et al., 2021).

The paper distinguishes three coupled mechanisms. First, wetting-controlled imbibition into nanopores requires a low three-phase contact angle at the air-water-solid lipid interface, with intensive bursting reported for $T_f$3. Second, in mixed-triglyceride oils such as cocoa butter, partial internal melting introduces a frozen-oil/melted-oil/surfactant-solution contact line; lowering the corresponding angle $T_f$4 improves dewetting of liquid oil from the still-frozen matrix and assists droplet ejection. Third, micelle-driven osmotic amplification requires non-spherical or supramolecular aggregates in the external aqueous phase; when these penetrate the pores and reorganize into many smaller spherical micelles, the internal micelle number concentration rises sharply and draws in additional water (Cholakova et al., 2021).

For cocoa butter specifically, the reported melting onset is approximately $T_f$5, complete melting occurs by approximately $T_f$6, the melting peak is approximately $T_f$7, and the freezing peak extends from approximately $T_f$8 down to approximately $T_f$9. Under a standard protocol of cooling to approximately $\theta$0–$\theta$1 and then heating at $\theta$2, only partial bursting was observed and many micrometer drops remained. When cocoa butter was fully crystallized by dispersing it in 50 wt% ethylene glycol, storing it at $\theta$3 overnight, and then heating, the bursting became “much more pronounced,” comparable to GEL01. The paper explicitly notes that numeric DLS size distributions were not reported for cocoa butter, even though mean diameters of about $\theta$4–$\theta$5 were achieved with some oils in the broader study (Cholakova et al., 2021).

The practical optimization rules are narrow and quantitative. Strong bursting requires $\theta$6, equilibrium surface tension below approximately $\theta$7 with rapidly decreasing dynamic surface tension, and non-spherical aggregates in the aqueous phase. Representative values reported on frozen coconut-oil substrates include $\theta$8 for 1.5 wt% Tween 20 alone and $\theta$9 for 1.5 wt% Tween 20 plus 0.5 wt% monoolein; in an ionic system, LAS+SLES with 30 mM CaCl$\theta$0 gave $\theta$1 and a reduced $\theta$2, with strong bursting (Cholakova et al., 2021).

5. Consumption, pricing, and the Danish saturated fat tax

In nutrition policy, butter was one of the foods targeted by Denmark’s saturated fat tax. The tax was introduced in October 2011 and repealed in January 2013. It was set at 16 DKK per kilogram of saturated fat for foods exceeding 2.3% saturated fat; with Denmark’s 25% VAT, the effective levy was 20 DKK per kilogram of saturated fat. Products subject to the tax included meats, dairy, oils, margarine, butter, and blended butter products, whereas drinking milk and milk-based yogurts were explicitly excluded. The cited study combines butter and butter blends because it found no significant differences between them (Dahl et al., 2023).

Identification relies on a difference-in-differences design comparing Danish households with Northern German households in Schleswig-Holstein and Hamburg, using quarterly GfK Consumer Scan data from 2009 to 2014. For butter consumption, a Wald test on pre-tax trends fails to reject parallel trends ($\theta$3), but for butter expenditure and price the reported pre-trend $\theta$4-values are $\theta$5, which the authors flag as a concern. The main estimand is the average treatment effect on the treated, implemented with the doubly robust DiD estimator of Callaway and Sant’Anna (2021) (Dahl et al., 2023).

Measure	Tax period	Post-tax period
Consumption	$\theta$6 g, $\theta$7	$\theta$8 g, $\theta$9
Expenditure	$\theta$0 Euro cents, $\theta$1	$\theta$2 Euro cents, $\theta$3
Price	$\theta$4 Euro cents per 100 g, $\theta$5	$\theta$6 Euro cents per 100 g, $\theta$7
Package size	$\theta$8 g, $\theta$9	$c_{ij}^t$00 g, $c_{ij}^t$01

During the taxed period, butter prices increased by 23.44% relative to the Danish pre-tax mean, and expenditure increased by 25.20%, but purchased quantity showed no statistically significant change. The paper interprets this as inelastic demand: households absorbed higher prices with higher spending rather than reducing purchased quantities. After repeal, butter consumption rose by 16.05% and expenditure by 16.92% relative to the pre-tax reference, while prices remained 3.22% above the pre-tax mean. Package sizes declined significantly only after repeal, by 10.11% relative to pre-tax (Dahl et al., 2023).

Border effects were substantial. Among Danish households in bordering regions, the average share of butter purchased abroad rose from about 2% in pre-tax quarters to roughly 8% during the tax period and remained elevated post-repeal. A random-effects model indicated significant increases in the predicted share of butter purchased abroad for households within approximately 70 km of the German border during the tax period, while no significant price spillover to Northern German households was detected. The study also reports that butter expenditure increased for both low- and high-income households during the tax period, with post-repeal consumption increases largely driven by high-income households. The butter-specific pattern is therefore presented as a case in which retail prices and expenditures rose substantially without reducing consumption, while border-adjacent households shifted purchases abroad (Dahl et al., 2023).

6. Homonymous and extended technical uses

In technical literature unrelated to dairy science, “Butter” is also used as a model name in computer vision. The detector introduced in “Butter: Frequency Consistency and Hierarchical Fusion for Autonomous Driving Object Detection” is a lightweight, real-time 2D object detector for autonomous driving that targets cross-scale feature consistency and semantic gaps in multi-level fusion. Its neck combines Frequency-Adaptive Feature Consistency Enhancement (FAFCE), which performs contextual low-frequency damping, displacement-guided resampling, and contextual high-frequency amplification, with Progressive Hierarchical Feature Fusion Network (PHFFNet), which performs progressive fusion with Context-Aware Spatial Fusion. Reported model complexity is 5.4M parameters and approximately 31 GFLOPs at $c_{ij}^t$02 input, with mAP@50 values of 94.4 on KITTI, 53.7 on BDD100K, and 53.2 on Cityscapes; an appendix reports an increase in small-object AP from 41.8 to 43.9 when FAFCE is used (Lin et al., 12 Jul 2025).

A separate homonymous usage appears through the surname Butter in supergravity. Butter’s 2013 construction of a new higher-derivative invariant in 4D $c_{ij}^t$03 superconformal gravity, the TLog multiplet, is described as the missing ingredient needed to reproduce the five-dimensional supersymmetric Weyl-squared Lagrangian under dimensional reduction and thereby resolve a longstanding mismatch between macroscopic and microscopic entropy for a class of 4D non-BPS extremal black holes. In the conventions quoted, the corrected macroscopic entropy is

$c_{ij}^t$04

which matches the first subleading expansion of

$c_{ij}^t$05

This usage is terminologically unrelated to butter as a food, but it is a persistent source of lexical overlap in arXiv indexing and citation practice (Banerjee et al., 2016).