Artist Influencer
Kandinsky 1898–1944 — The Structural Story
226 paintings  ·  μ cohesion · hue entropy · CBI · SDI  ·  composition n=13 · improvisation n=28
Period
Series
Other
Size
Cohesion Arc year × μ  ·  dip in improvisation, peak in bauhaus — edge sharpness, not abstraction level
Hue Entropy year × hue entropy  ·  flat across periods — art-historical narrative inverted
Color Independence year × CBI  ·  Paris collapse: color follows structure in late biomorphic work
Composition vs Improvisation SDI × μ  ·  subject colors — named series should separate without title knowledge

Turbulence to Governance

200 paintings · 1898–1944
A structural and color kernel reading of Kandinsky
Russell Parrish · Artist Influencer · artistinfluencer.com  ·  Images: WikiArt CDN · deduplicated
Seven structural and color measurements extracted from 200 Kandinsky paintings across four decades, plotted against time and named series. The visualization tracks how structurally cohesive the gradient field is, how orthogonally partitioned the composition is, how broadly distributed hue is, and how autonomously color operates from luminance.
The standard account of Kandinsky describes a career moving from representation toward abstraction, from spiritual improvisation toward geometric rationalization. The measurements describe something different: a career moving from distributed turbulence toward increasingly governed structural systems. Abstraction level is not what the kernel tracks. Compositional governance is.
Kandinsky's revolution is organizational more than chromatic. The palette does not radically expand as he abstracts. What changes is the structural authority over where elements are placed and how fields are partitioned.
Primary Falsifiable Finding

If the kernel measures what it claims, paintings Kandinsky named "Compositions" (planned, structurally deliberate) and "Improvisations" (spontaneous, intuition-driven) should be distinguishable without reading a title. Either the pixel structure reflects the declared structural intent, or it does not.

They separate. Compositions: mu 0.574, luma_chroma_decoupling 0.020. Improvisations: mu 0.443, luma_chroma_decoupling 0.045. The kernel reads his structural categories from pixel data alone. n=11 Compositions, n=16 Improvisations; directionally clear, small-n.

Murnau 1898–1909 mu 0.379 · HE 0.703 · ops 0.424 Figurative and expressionist. Already high hue entropy: the Murnau palette is vibrant before abstraction arrives. mu is the lowest of the career, not the highest: soft chromatic color patches produce no dominant connected gradient component despite representational subject matter. What anchors mu is edge sharpness, not subject presence. n=64.
Improvisation 1910–1921 mu 0.485 · HE 0.752↑ · ops 0.341↓ Turbulent abstraction and Russian interlude. HE peaks here, palette breadth at its widest. mu rises from Murnau as some forms assert harder edges, but orthogonal_partition_score drops to its career low: the composition is not yet geometrically ordered. gradient_floor_85 is highest of any period (0.315), the whole field is activated, not just edge zones. luma_chroma_decoupling peaks (0.034): color and luminance most decoupled. SDI is essentially flat (0.261 vs 0.254 Murnau): Kandinsky reorganizes internal governance without radically relocating structural mass. n=62.
Bauhaus 1922–1933 mu 0.511↑ · HE 0.588↓ · ops 0.498↑ Hard geometry produces the sharpest structural transition. mu continues rising (+0.026 from Improvisation). ops spikes to 0.498, the Bauhaus geometric system imposing orthogonal partitioning that no other period matches. HE drops sharply to 0.588 with the highest within-period variance (SD ±0.172): some paintings are spectrally wide, others tightly constrained. focal_dominance peaks (0.436). The period of geometric rationalization is the most structurally consolidated. n=42.
Paris 1934–1944 mu 0.503 · HE 0.608 · CBI 0.059↓ Biomorphic abstraction. ops holds high (0.503), mu holds (0.503). HE partially recovers. CBI collapses to career low (0.059): color in the Paris work follows luminance contours at edges. chroma_focality spikes (0.422 vs 0.259 in Improvisation) and edge_color_coupling rises: color is precisely placed on biomorphic outlines, not free. The period that appears most playful is the most chromatically authoritarian by measurement; structural cohesion itself remains stable. n=32.
Top: mu (structural cohesion). Middle: orthogonal_partition_score (geometric ordering). Bottom: HE (palette breadth). Improvisation marked ↑ for HE peak and ops nadir; Bauhaus marked ↑ for governance consolidation.
Murnau
0.379↓
0.424
0.703
Improvisation
0.485
0.341↓
0.752↑
Bauhaus
0.511↑
0.498↑
0.588↓
Paris
0.503
0.503
0.608
focal_count (career's strongest correlate with year, r = −0.327): compositions simplify as governance increases: 14.3 → 15.6 → 10.2↓ → 11.8
CBI (color autonomy at edges, r = −0.246 with year): declines across the career, collapsing in Paris: 0.103 → 0.105 → 0.098 → 0.059↓
Composition mu 0.574 · lcd 0.020 Planned, structurally deliberate works. Higher mu than Improvisations (+0.131), lower luma_chroma_decoupling (−0.026), fewer focal events, higher focal_dominance. Color and luminance are more coupled. The planned works are more structurally resolved and more chromatically obedient. n=11.
Improvisation mu 0.443 · lcd 0.045 Spontaneous, intuition-driven works. Lower mu, higher gradient_floor_85 (whole field activated), higher luma_chroma_decoupling (color and luminance more separated). SDI 0.262 vs Composition 0.239: structural mass more dispersed. The spontaneous works are more structurally turbulent across every structural metric measured. n=16.
Landscape mu 0.379 · HE 0.703 Murnau and pre-Munich figurative work. Low mu despite representational subject matter: soft chromatic color patches produce dispersed gradient fields even when depicting recognizable scenes. HE already high: the Murnau palette breadth predates abstraction. n=49.
Portrait mu 0.504 · CBI 0.103 Small sample (n=10) but highest mu of any named subject category. Figurative contours provide hard edges that anchor the gradient field in a way Murnau landscapes do not. Edge sharpness, not subject presence, is what drives cohesion.
Structural
mu (μ)
Cohesion: largest connected edge component as fraction of total edge mass. In Kandinsky, mu tracks edge sharpness not subject matter. Soft color patches = low mu. Hard geometric boundaries = high mu.
Geometric
ops
Orthogonal partition score: degree to which the gradient field is organized along horizontal and vertical axes. Career correlation r = +0.308. Bauhaus geometric system produces highest ops of any period.
Color
HE
Hue entropy: angular breadth of color use across the hue wheel. Peaks in Improvisation period (0.752). Contracts in Bauhaus (0.588, SD ±0.172, highest within-period variance). Not flat: between-period variance is 9x within-period variance.
Color
CBI
Chromatic Boundary Independence: color drawing edges where luminance does not. Declines across the career (r = −0.246). Paris collapse to 0.059 driven by chroma_focality spike and edge_color_coupling rise: color precisely placed, not free.
Scroll wheelZoom toward cursor, each panel independent
Click + dragPan, each panel independent
Double-clickReset zoom and pan
HoverThumbnail, title, year, all metrics
Period togglesIsolate periods; stats bar updates
Series togglesIsolate Composition / Improvisation / other
Fixed / rescaleRescale axes to visible range
MetricWhat it measures
mu Structural cohesion: how clearly demarcated regions are by edge sharpness. High = crisp geometric boundaries. Low = soft chromatic fields. In Kandinsky, mu tracks edge sharpness not subject matter; soft color patches produce low mu regardless of whether the image is representational or abstract. Career low in Murnau (0.379) despite figurative subjects. Career high trend in Bauhaus (0.511).
orthogonal_partition Geometric ordering along horizontal and vertical axes. Drops in Improvisation (0.341, career low), spikes in Bauhaus (0.498). The strongest positive career correlation in the dataset (r = +0.308). Bauhaus hard geometry imposes orthogonal partitioning that no other period approaches.
hue_entropy Palette breadth. Peaks in Improvisation (0.752). Contracts in Bauhaus (0.588) with the highest within-period SD (±0.172): geometric discipline constrains some paintings tightly while others remain spectrally wide. Several Circles (1926) has HE 0.896 (career near-high) in the career's lowest-mean period.
CBI Color autonomy at edges. Note: Improvisation CBI (0.105) may understate color freedom because gradient_floor_85 is highest here (0.315): the activated field suppresses the quiet luminance zones CBI needs to detect autonomous color edges. Use luma_chroma_decoupling as a complementary read for the Improvisation period.
luma_chroma_decoupling Field-wide separation between luminance and chroma structure. Peaks in Murnau (0.051) and Improvisation (0.034). Career low in Paris (0.031). The sharpest Composition/Improvisation differentiator: Comp 0.020 vs Improv 0.045. Color and luminance are more decoupled in the spontaneous works.
focal_count Number of major gradient clusters. Career's strongest year correlation (r = −0.327). Falls from 15.6 (Improvisation) to 10.2 (Bauhaus). Compositions average fewer focal events (14.4) than Improvisations (15.9): planned works concentrate structural energy; spontaneous works distribute it.
Murnau: low mu despite figures. Toggle Murnau alone and find Autumn in Murnau (mu 0.062), the career's lowest structural cohesion in a figurative painting. The soft color patches produce no dominant gradient anchor. Compare against The Singer (mu 0.957, 1903) in the same period. The range is enormous. Subject matter is not the driver.
Improvisation: turbulence and palette peak. Toggle series = Improvisation alongside series = Composition within the same period. Improvisations scatter higher in SDI, lower in mu, higher in luma_chroma_decoupling. The spontaneous works are genuinely more structurally dispersed. The kernel reads what Kandinsky intended as structural categories.
Bauhaus: governance consolidation. Find Crossing (1928): ops 0.951, the career high for geometric ordering. Then find Several Circles (1926): HE 0.896 in the career's lowest-mean HE period. High geometric structure does not require narrow palette. The two signals are independent. Bauhaus focal_dominance 0.436, the career high: fewer elements, each carrying more compositional authority.
Paris: apparent freedom, measured constraint. CBI collapses to 0.059. Find the 1944 Composition (CBI 0.006): the lowest chromatic boundary independence in the dataset. Color is precisely placed on biomorphic outlines, not free. The period that looks most playful is the most chromatically authoritarian by measurement.
The Composition/Improvisation separation is the finding that earns the most confidence. Kandinsky named these series as structural categories with explicit intent: Compositions are planned and deliberate; Improvisations are spontaneous and intuitive. Every structural metric separates them in the predicted direction without title information. Compositions: mu 0.574, focal_count 14.4, luma_chroma_decoupling 0.020. Improvisations: mu 0.443, focal_count 15.9, luma_chroma_decoupling 0.045. The spontaneous works are genuinely more structurally distributed. The planned works are more cohesive and more chromatically obedient. The kernel reads Kandinsky's own structural intention from pixel data alone.
The Murnau period inverts the expected baseline. The figurative landscapes have the lowest mu of the career (0.379), lower than the abstract Improvisation period, lower than the Bauhaus geometric period, lower than Paris. Autumn in Murnau (1908) has mu 0.062, the career's lowest structural cohesion in a non-abstract painting. The mechanism is specific: Kandinsky's Murnau brushwork uses soft chromatic color patches with diffuse boundaries between color areas. Soft boundaries produce no dominant connected gradient component. Edge sharpness drives mu; subject presence does not. The conclusion cuts against intuition: Kandinsky's figurative work is structurally less resolved than his abstract work, because abstraction eventually imposed harder edges.
The Bauhaus period is where governance consolidates. orthogonal_partition_score rises to 0.498 (career correlation r = +0.308, the strongest directional signal in the dataset). focal_dominance spikes to 0.436 (career high): fewer compositional elements, each carrying more structural authority. Focal_count falls to 10.2, the lowest of any period. HE contracts to 0.588 but with the highest within-period variance (SD ±0.172): geometric discipline constrains some paintings tightly while others remain spectrally wide. Several Circles (1926) has HE 0.896 in the career's lowest-mean period; high geometric structure does not require narrow palette. The Bauhaus system governs composition, not necessarily color.
The Paris CBI collapse is the most counterintuitive finding. CBI falls to 0.059, the career low, in the period of apparently playful biomorphic freedom. The mechanism is measurable: chroma_focality spikes to 0.422 (vs 0.259 in the Improvisation period) and edge_color_coupling rises to 0.401. The biomorphic forms concentrate chroma into specific zones and bind it precisely to structural outlines. Color is not free in the Paris work; it is placed with chromatic authority onto carefully drawn boundaries. The period that appears most liberated is the most structurally obedient by chromatic measurement.
One methodological note on the Improvisation period CBI. gradient_floor_85 is highest here (0.315 vs 0.278 in Murnau): the whole field is activated, not just edge zones. This suppresses CBI by removing the quiet luminance zones the metric needs to detect color-autonomous edges. Improvisation CBI (0.105) likely understates color freedom in this period. luma_chroma_decoupling (0.034 for Improvisations, 0.045 for the named Improvisation series specifically) is the better color autonomy signal when the gradient field is globally activated. The Improvisations have the most decoupled color-luminance relationship of any named series; CBI alone does not fully capture it.
Four career correlations point in the same direction simultaneously. focal_count falls (r = −0.327): fewer compositional events. orthogonal_partition_score rises (r = +0.308): more geometric ordering. focal_dominance rises (r = +0.291): each remaining event carries more authority. CBI falls (r = −0.246): color becomes more structurally obedient. SDI remains essentially flat across all four periods, a null finding worth stating. Kandinsky reorganizes internal governance without radically relocating structural mass. The energy does not move to the edges or concentrate at the center; it is redistributed among fewer, more authoritative elements. Together these describe a career not moving from representation to abstraction but from distributed energetic turbulence toward increasingly concentrated compositional governance. Abstraction did not remove structure from Kandinsky's work. It relocated structure: from representational subject edges toward geometric systems, focal hierarchies, and precise chromatic placement. The Improvisation period is the most structurally free. The Paris period is the most chromatically authoritarian. Abstraction did not liberate Kandinsky's compositions. It gave him new instruments to govern them.
Structural + color kernel · artistinfluencer.com · ORCID 0009-0008-9781-7995 abstraction did not liberate; it governed
Dataset and limitations. 200 paintings from WikiArt CDN, deduplicated by image hash. Period boundaries follow standard Kandinsky periodization; subject tags pre-assigned from title text before kernel computation. Composition (n=11) and Improvisation (n=16) counts are small; the falsifiable finding is directionally clear but should be treated with appropriate caution. The 1914–1921 Russian period is folded into Improvisation; it may cluster closer to early Bauhaus structurally; insufficient n to separate. CBI is suppressed in the Improvisation period by elevated gradient_floor_85 (globally activated field); luma_chroma_decoupling is the preferred color autonomy signal for that period. ci_cbi_ratio is unreliable for Paris period (denominator instability when CBI approaches zero); excluded from analysis. Pale grounds and geometric whitespace in Bauhaus and Paris paintings inflate high_key_fraction relative to earlier periods; this is a medium characteristic, not a finding. All metrics at thumbnail resolution (200px long edge); mu and structural metrics are resolution-robust.