Scales and D3

TanStack Charts uses D3 as an explicit algorithm layer:

• Your application imports and configures the D3 modules required by a chart.
• D3 supplies battle-tested scales, array transforms, shape interpolation, time utilities, and spatial algorithms.
• TanStack Charts supplies the typed grammar, responsive ranges, guide layout, scene compilation, rendering, reconciliation, and lifecycle.

There is no second scale or transform language to learn. There is also no hidden D3 umbrella import.

Direct dependency ownership#

If application source imports a d3-* module, declare that module and its matching TypeScript package directly:

pnpm add d3-array d3-scale d3-shape
pnpm add -D @types/d3-array @types/d3-scale @types/d3-shape

Omit any module the application does not import. A bar chart that only imports d3-scale should not install or bundle shape, force, geo, zoom, or hierarchy code.

This rule also applies when definitions live in framework component source. The adapter mounts a definition; it does not own the D3 imports used to author it.

Capability map#

Use the official D3 pages as the API reference for each algorithm. TanStack Charts documentation only describes how its output crosses the chart boundary.

Positional scales are required#

Every ChartSpec declares x and y:

import { scaleLinear, scaleUtc } from 'd3-scale'

const spec = {
  marks,
  x: { scale: scaleUtc, nice: true },
  y: { scale: scaleLinear, nice: true },
}

Use null only when no mark materializes that dimension:

import { defineChart, frame } from '@tanstack/charts'

const borderOnlyChart = defineChart({
  marks: [frame()],
  x: null,
  y: null,
})

A mark with x values needs an x scale. A mark with y values needs a y scale. The scale factory chooses the mapping; materialized mark channels supply its domain.

Factory domains come from marks#

Pass the D3 factory itself when the domain should cover the rendered data:

import { scaleLinear, scaleUtc } from 'd3-scale'

const chart = defineChart({
  marks: [lineY(rows, { x: 'date', y: 'value' })],
  x: { scale: scaleUtc, nice: true },
  y: { scale: scaleLinear, nice: true },
})

Continuous factories use the finite extent of their channels. Band and point factories use distinct values in first-seen order. Bars and areas include zero when they use an implicit zero baseline. Empty channels retain the factory's native D3 domain.

Return a scale from a zero-argument factory when it needs configuration before domain inference:

const x = {
  scale: () => scaleBand<string>().padding(0.16),
}

Use the axis nice option because nicening must happen after inference.

Fixed domains remain application semantics#

Pass a scale instance when the domain must not follow the rendered marks:

const normalizedY = {
  scale: scaleLinear().domain([0, 1]),
}

const windowedX = {
  scale: scaleUtc().domain([windowStart, windowEnd]),
}

Be equally deliberate with:

• Whether a log scale is valid for all values
• Whether time is local or UTC
• Which categories exist when some are filtered out
• Whether multiple facets share a domain
• Whether a color domain must remain stable across sessions

Responsive ranges belong to TanStack Charts#

Do not assign pixel ranges to positional scales used by the chart:

const xScale = scaleUtc
const yScale = scaleLinear

For each scene, TanStack Charts:

1. Creates a factory scale or copies a configured instance.
2. Calculates the plot rectangle after guide measurement.
3. Assigns the current responsive pixel range to the copy.
4. Uses the copy for marks, ticks, grids, and interaction points.

The source scale is never mutated. This makes one module-level definition safe across container resizes, server rendering, multiple mounted hosts, and facets.

The reverse axis option reverses the responsive range without changing the domain:

const y = {
  scale: scaleLinear,
  reverse: true,
}

Categorical scales and bandwidth#

Pass a D3 band-scale factory for categorical positions:

import { scaleBand } from 'd3-scale'

const categoryScale = () =>
  scaleBand<string>().paddingInner(0.12).paddingOuter(0.06)

TanStack Charts applies the plot range, reads the scale bandwidth, and treats the mapped value as the center of the band for mark and interaction coordinates. Bars use the primary bandwidth by default.

For grouped bars, pass a second band scale as the mark’s groupScale. Its range is assigned within the primary band. Grouping is explicit because z alone cannot decide whether a chart should overlap, stack, dodge, or only color its rows.

Color scales#

Omitting color.scale uses the chart theme’s ordinal palette for categorical group values:

const chart = defineChart({
  marks: [lineY(rows, { x: 'date', y: 'value', z: 'region' })],
  x: { scale: xScale },
  y: { scale: yScale },
})

Use a configured D3 color scale for semantic stability:

import { scaleOrdinal } from 'd3-scale'

const regionColor = scaleOrdinal(
  ['North', 'South', 'West'],
  ['#2563eb', '#f97316', '#10b981'],
)

const chart = defineChart({
  marks: [lineY(rows, { x: 'date', y: 'value', z: 'region' })],
  x: { scale: xScale },
  y: { scale: yScale },
  color: {
    scale: regionColor,
    legend: colorLegend({ label: 'Region' }),
  },
})

The color scale is copied before use. Unlike positional scales, its range is semantic color output and remains the range you configured.

Use a factory when a custom color mapping should infer its domain:

const color = {
  scale: () => scaleOrdinal<string, string>().range(['#2563eb', '#f97316']),
}

Radius scales#

dot treats r as pixels unless rScale is supplied:

import { scaleSqrt } from 'd3-scale'

dot(rows, {
  x: 'revenue',
  y: 'retention',
  r: 'accounts',
  rScale: {
    scale: () => scaleSqrt().range([3, 24]),
  },
})

The radius factory infers [0, maximum] from r. A configured scale instance still keeps its explicit domain. D3 owns the radius mapping; the chart owns dot geometry and rendering.

Curves#

Straight lines and areas do not need d3-shape. Opt into a curve only when the design requires it:

import { curveMonotoneX } from 'd3-shape'
import { d3Curve, lineY } from '@tanstack/charts'

lineY(rows, {
  x: 'date',
  y: 'value',
  curve: d3Curve(curveMonotoneX),
})

d3Curve adapts a D3 curve factory to the small line-and-area curve contract. Importing it is explicit so a straight chart does not need the shape path.

Horizontal areaX marks use the separate d3AreaXCurve bridge from @tanstack/charts/d3/area-x.

Transforms produce rows#

D3 transforms do not need a TanStack wrapper:

import { bin, max } from 'd3-array'

const upper = max(values) ?? 1
const histogram = bin()
  .domain([0, upper])
  .thresholds(20)(values)
  .map((items, index) => ({
    id: index,
    x1: items.x0 ?? 0,
    x2: items.x1 ?? upper,
    count: items.length,
    items,
  }))

Pass histogram to rect, barY, lineY, dot, or a custom mark according to the desired geometry. Keep substantial transforms in ordinary functions beside the definition and memoize them through application reactivity when necessary.

The same rule applies to stacks, pies, hierarchies, force layouts, and server-prepared intervals: preserve the useful output as typed rows, then map it through mark channels. A responsive geographic projection instead belongs in geoShape's projection factory because its pixel range depends on the final plot bounds.

Pixel-to-value inversion#

Brush, cursor, crop, and zoom gestures are application-owned. To invert a scene coordinate:

const scene = host.getScene()
const interactiveX = sourceXScale
  .copy()
  .range([scene.chart.x, scene.chart.x + scene.chart.width])

const selectedDate = interactiveX.invert(pointerX)

For a normal continuous y axis, use the reversed chart range:

const interactiveY = sourceYScale
  .copy()
  .range([scene.chart.y + scene.chart.height, scene.chart.y])

const selectedValue = interactiveY.invert(pointerY)

Apply the application’s precision policy after inversion. For example, round a day-based selection with a D3 time interval or round a currency threshold to the supported increment. Pixels do not imply semantic precision.

When the application owns the gesture, disable the native nearest-point focus strategy if the two interactions would conflict. See Interactions and Selections.

Log-scale example#

import { scaleLinear, scaleLog } from 'd3-scale'
import { defineChart, dot } from '@tanstack/charts'

interface Measurement {
  id: string
  input: number
  output: number
}

const rows: readonly Measurement[] = [
  { id: 'a', input: 1, output: 12 },
  { id: 'b', input: 10, output: 31 },
  { id: 'c', input: 100, output: 49 },
  { id: 'd', input: 1_000, output: 72 },
  { id: 'e', input: 10_000, output: 88 },
]

const logChart = defineChart({
  marks: [
    dot(rows, {
      x: 'input',
      y: 'output',
      key: 'id',
      r: 4,
      fill: '#2563eb',
    }),
  ],
  x: {
    scale: scaleLog().domain([1, 10_000]),
    label: 'Input',
    grid: true,
  },
  y: {
    scale: scaleLinear().domain([0, 100]),
    label: 'Output',
    grid: true,
  },
})

This source imports d3-scale directly, so install d3-scale and @types/d3-scale as direct dependencies.

For chart-side scale, guide, and color types, see Scales, Guides, and Color Reference.