Skip to main content

Compute Engine Changelog

0.54.0 2026-02-26​

  • New expr.polynomialCoefficients() method: Returns the coefficients of a polynomial expression in descending order of degree, or undefined if the expression is not a polynomial. Auto-detects the variable when the expression has exactly one unknown. Subsumes isPolynomial (check !== undefined) and degree computation (length - 1).

  • polynomialCoefficients() now accepts an array of variables: Pass ['x', 'y'] to verify the expression is polynomial in all listed variables. Coefficients are decomposed by the first variable.

  • New expr.polynomialRoots() method: Returns the roots of a polynomial expression, or undefined if not a polynomial. Handles degree 3+ polynomials with rational roots via the Rational Root Theorem.

  • New Polynomial CAS function: Constructs a polynomial from a coefficient list (descending order) and a variable. Inverse of CoefficientList: Polynomial([1, 0, 2, 1], x) evaluates to x³ + 2x + 1.

  • Improved Factor for degree 3+ polynomials: Factor now uses the Rational Root Theorem to factor polynomials with integer coefficients and rational roots. Previously only handled degree ≤ 2.

  • Improved Factor with content extraction: Factor now extracts the GCD of integer coefficients before applying other strategies. For example, Factor(6x² + 12x + 6, x) now produces 6(x+1)².

  • New PartialFraction CAS function: Decomposes rational expressions into partial fractions. Supports distinct and repeated linear factors, irreducible quadratic factors, and improper fractions (polynomial division performed first). Example: PartialFraction(1/((x+1)(x+2)), x) → 1/(x+1) - 1/(x+2).

  • New Apart CAS function: Alias for PartialFraction.

  • New PolynomialRoots CAS function: Returns the roots of a polynomial as a set. Example: PolynomialRoots(x² - 5x + 6, x) → {2, 3}.

  • New Discriminant CAS function: Returns the discriminant of a polynomial of degree 2, 3, or 4. Supports symbolic coefficients. Example: Discriminant(x² - 5x + 6, x) → 1.

  • simplify() auto-decomposes partial fractions: When a Divide expression has a denominator already in factored form (product or power) and the decomposition is simpler, simplify() automatically applies partial fraction decomposition.

  • Breaking: CoefficientList now returns descending order: The CAS function CoefficientList now returns coefficients from highest to lowest degree (e.g., [1, 0, 2, 1] for x^3 + 2x + 1), matching the new polynomialCoefficients() method and common external conventions. Previously it returned ascending order.

  • expr.match() now accepts string patterns with auto-wildcarding: Pass a LaTeX string like 'ax^2+bx+c' and single-character symbols are automatically treated as wildcards. Results use clean unprefixed keys ({a: 3, b: 2, c: 5}) with self-matches filtered out. useVariations and matchMissingTerms default to true for string patterns.

  • expr.match() now accepts MathJSON arrays directly: Pass a raw MathJSON pattern like ['Add', '_a', '_b'] without calling ce.box() first.

  • New matchMissingTerms option for match(): When enabled, expressions with fewer operands than the pattern can still match by treating missing terms as identity elements (0 for Add, 1 for Multiply). For example, 3x^2+5 matches the pattern ax^2+bx+c with b = 0. Enabled by default for string patterns.

  • Non-strict parsing: implicit superscript for letter+digit: In non-strict mode, a single letter immediately followed by a digit 2–9 is parsed as an exponent: x2 + y2 → x^2 + y^2. Handles common copy-paste from web pages. Only digits 2–9, only single ASCII letters, and only when adjacent (no space).

0.53.1 2026-02-25​

  • timeLimit now reliably interrupts long-running evaluations: Factorial, Sum, Product, Loop, and Reduce all respect the timeLimit property and throw CancellationError when the deadline is exceeded. Previously, generators yielded too infrequently (every 1,000–50,000 iterations), allowing a single gen.next() call to block for longer than the timeout. All generators now yield every iteration. The Factorial handler no longer silently swallows CancellationError, and withDeadline/withDeadlineAsync now use try/finally to always reset the engine deadline.

  • Fixed GPU compilation of Sum, Product, Loop, and Function: These constructs no longer leak JavaScript-specific syntax (IIFEs, let, while, arrow functions, { re, im } objects) into GLSL/WGSL output. Sum/Product with small constant bounds are unrolled inline; larger ranges emit native for loops. Loop emits a GPU for loop with int/i32 index. Function (lambda) now throws a clear error for GPU targets. Block-level Declare statements infer vec2/vec2f type from subsequent complex-valued assignments.

  • Added GLSL/WGSL compilation for Heaviside, Sinc, FresnelC, FresnelS, BesselJ: These five special functions now compile to GPU shader targets. FresnelC/FresnelS use a three-region rational Chebyshev approximation (ported from Cephes/scipy) with a shared _gpu_polevl helper. BesselJ uses power series, Hankel asymptotic, and Miller's backward recurrence depending on the argument range. Both GLSL and WGSL preambles are emitted on demand.

  • Fixed GLSL/WGSL block expression compilation: Block expressions (produced by \coloneq / semicolon blocks) now emit valid GPU shader code instead of JavaScript syntax. Variable declarations use float x (GLSL) or var x: f32 (WGSL) instead of let x, and blocks are emitted as plain statements instead of JavaScript IIFEs. compileFunction correctly formats multi-statement bodies.

  • Fixed \; in \text{where} clauses: Visual spacing commands like \;, \,, \quad, etc. between comma-separated bindings in where-clauses are now correctly skipped instead of being parsed as HorizontalSpacing expressions wrapped in InvisibleOperator.

  • Fixed require() returning empty exports on Node 22+ (#292): Because the package sets "type": "module", Node treated the UMD .js files as ESM, breaking the UMD factory pattern. The UMD builds now use a .cjs extension so Node always treats them as CommonJS.

0.53.0 2026-02-21​

Runtime and Scoping​

  • True lexical scoping for Function expressions: Functions now capture their defining scope and resolve free variables from that scope chain (not the call site), with a fresh child scope on each call.

  • BigOp scope pollution fixed: Sum, Product, and other big operators now only declare their index variable locally. Other names are declared in the enclosing scope via noAutoDeclare.

  • Closure capture for nested functions: Returned functions now correctly capture outer parameters across multiple nesting levels.

  • EvalContext.values removed: Symbol values now live only in BoxedValueDefinition.value. The per-frame shadow map and withArguments option were removed.

  • forget() now resets values set by assume(): forget('x') now clears values introduced by assume('x = ...') (value reset to undefined), in addition to clearing assumptions.

Expressions and Equality​

  • expand() now returns the input expression instead of null: Both the free function and internal expand()/expandAll() now return the original expression when no expansion is possible.

  • New .toRational() method: Returns [numerator, denominator] integers for rational expressions, or null otherwise.

  • New .factors() method: Returns multiplicative factors as a flat array by decomposing Multiply and Negate structurally.

  • .is() now tries expansion: After structural comparison, .is() expands both sides before numeric fallback, catching forms like (x+1)^2 and x^2+2x+1.

  • .is() is now symmetric: a.is(b) === b.is(a) now holds across all expression types.

LaTeX Parsing​

  • Parse \mleft/\mright delimiters: Alternative delimiters from the mleftright package are now treated like \left/\right.

  • Parse \color in math mode: \color{...} is now recognized in math mode; the color argument is consumed so the following math parses normally.

  • Parse : and \colon as infix operators: Outside quantifier contexts, a bare :/\colon now parses as Colon (e.g. f:[a,b]\to\R), without affecting := assignment or quantifier syntax.

  • Parse \dfrac, \tfrac, and \cfrac as fractions: These variants now parse the same as \frac.

Fractals​

  • New Mandelbrot and Julia functions: Added built-in escape-time fractal operators. Mandelbrot(c, maxIter) and Julia(z, c, maxIter) return a smooth, normalized value in [0, 1] (1 for interior points, fractional for escaping points via log₂(log₂(|z|²)) smoothing). Both evaluate in JavaScript and compile to GLSL/WGSL.

0.52.1 2026-02-19​

Expressions​

  • Exact number literal check: Use isNumber(expr) && expr.isExact to test for exact numeric literals.

  • raw form preserves subtraction: x-1 now parses as ["Subtract", "x", "1"] (instead of ["Add", "x", -1]) when using raw form.

Parsing and Blocks​

  • Fix ;\; parsing in semicolon blocks: Spacing commands after semicolons (\;, \,, \quad, etc.) no longer create spurious Nothing operands.

  • Fix \text{if} parsing with \; spacing: \text{if}\;...\;\text{then}\;...\;\text{else}\;... now parses correctly as If.

  • Block serializer now uses ; : Serialization emits ; (not ;\; ) to avoid reintroducing spacing-related parse issues on round-trip.

  • Block compiler filters Nothing operands: The Block compiler now removes Nothing symbols and empty compile results before generating code.

  • Subscripted variable names in blocks: Names like r_1 are treated as compound symbols (not Subscript) when the base is not a known collection.

  • Non-strict parser supports exponents on bare functions: In strict: false mode, forms like sin^2(x) and cos^{10}(x) now parse correctly as powers.

  • Unicode superscript/subscript digits supported: Superscript and subscript Unicode digits now normalize to ^{...} / _{...} in parsing.

Compilation​

  • Selective GLSL interval preamble: interval-glsl now emits only used helper functions (plus dependencies), typically reducing preamble size by 60-80%.

  • Selective WGSL interval preamble: interval-wgsl now applies the same used-only preamble strategy.

  • Fix recursive GLSL gamma helper: Replaced recursive _gpu_gamma() reflection logic (illegal in GLSL) with a non-recursive implementation.

Equality​

  • .is() now works with assigned variables: Numeric fallback now applies to expressions with no free variables, including variables with assigned values.

  • .is() now accepts an optional tolerance: A per-call tolerance can override engine.tolerance for numeric comparison.

0.52.0 2026-02-18​

Features​

  • Smart .is() / exact .isSame() separation: The .is() and .isSame() methods on expressions now have distinct roles:

    • .isSame(v) — Fast exact structural check. No evaluation, no tolerance. Now accepts primitives (number, bigint, boolean, string) in addition to Expression. This is the method used internally throughout the engine.

    • .is(v) — Smart check with numeric evaluation fallback. Tries .isSame() first; if that fails and the expression is constant (no free variables), evaluates numerically and compares within engine.tolerance. For literal numbers, behaves identically to .isSame() — tolerance only applies to expressions that require evaluation.

    This resolves a common pain point where ce.parse('\\cos(\\pi/2)').is(0) returned false because .is() was purely structural. Now it returns true:

    ce.parse('\\sin(\\pi)').is(0);            // true  (evaluates, within tolerance)
    ce.parse('\\cos(\\frac{\\pi}{2})').is(0); // true
    ce.number(1e-17).is(0);                   // false (literal number, no tolerance)
    ce.parse('x + 1').is(1);                  // false (not constant, no fallback)

  • numericValue() convenience helper: New standalone function that combines the isNumber() guard with .numericValue access. Returns the numeric value if the expression is a number literal, or undefined otherwise. Useful for safely extracting numeric values without verbose ternary patterns:

    import { numericValue } from '@cortex-js/compute-engine';

    // Before
    const val = isNumber(expr) ? expr.numericValue : undefined;

    // After
    const val = numericValue(expr);

  • Stochastic equality check for expressions with unknowns: expr.isEqual() now uses a stochastic fallback when symbolic methods (expand + simplify) can't prove equality. Both expressions are evaluated at 50 sample points (9 well-known values + 41 random) and compared with relative+absolute tolerance. This detects equivalences like sin²(x) + cos²(x) = 1, (x+y)² = x²+2xy+y², and sin(2x) = 2sin(x)cos(x) that were previously returned as undefined. Singularities (NaN at a sample point) are skipped rather than treated as disagreements. The check also works when the two expressions have different unknowns (e.g. x - x + y vs y).

  • expr.freeVariables property: New property on BoxedExpression that returns the free variables of an expression — symbols that are not constants, not operators, not bound to a value, and not locally scoped by constructs like Sum or Product. Semantically identical to expr.unknowns.

  • New interval-js compilation functions: Added Binomial, GCD, LCM, Chop, Erf, Erfc, Exp2, Arctan2, and Hypot to the interval-js compilation target, with corresponding interval arithmetic implementations.

  • GLSL/WGSL variable exponent support: The interval GLSL and WGSL targets now support Power with variable exponents (e.g. (-1)^k, x^n). Previously these threw at compile time. Added ia_pow_interval() to both GPU library preambles using four-corner exp(exp * ln(base)) evaluation with special cases for point-integer exponents and (-1)^n.

  • Factorial, Gamma, GammaLn for GLSL/WGSL interval targets: Added ia_factorial (via ia_gamma(x+1)) to both GPU targets. Added ia_gamma (Lanczos approximation) and ia_gammaln (Stirling asymptotic) to the WGSL target, matching existing GLSL implementations.

Bug Fixes​

  • parse() with form: 'structural' ignored the structural flag: The structural option from formToInternal() was dropped in parseLatexEntrypoint(), making ce.parse(s, { form: 'structural' }) behave identically to { form: 'raw' } (unbound, unsorted). Now correctly produces a bound, structural expression.

  • Partial canonicalization with 'Flatten' form folded numerics: Using ce.parse(s, { form: ['Flatten', 'Order'] }) unexpectedly evaluated numeric operands (e.g. 3×2+1 became 7) because flattenForm() used ce.function() which defaults to full canonical mode. Now uses ce._fn() to preserve operand structure. This enables structural comparison of expressions modulo commutativity and associativity without numeric evaluation — useful for checking the method used to solve a problem rather than just the numeric result:

    const a = ce.parse('3\\times2+1', { form: ['Flatten', 'Order'] });
    const b = ce.parse('1+2\\times3', { form: ['Flatten', 'Order'] });
    a.isSame(b);  // ➔ true  (same structure, different order)

    const c = ce.parse('7', { form: ['Flatten', 'Order'] });
    a.isSame(c);  // ➔ false (different structure)

  • Sum/Product with symbolic bounds compiled incorrectly: Expressions like \sum_{k=0}^{n} f(k, x) where the upper bound is a variable produced loops that iterated 10001 times instead of using the variable n. The compilation extracted bounds via normalizeIndexingSet() which converted symbolic bounds to NaN and fell back to a hardcoded limit. Now bounds are extracted as expressions and compiled to code (e.g. Math.floor(_.n) for JS, Math.floor((_.n).hi) for interval-js). This fixes Taylor series patterns like \sum_{k=0}^{n} \frac{(-1)^k x^{2k+1}}{(2k+1)!} for both JS and interval-js targets.

  • Interval (-1)^k returned empty instead of correct value: The powInterval() function required positive bases for variable exponents, causing (-1)^k patterns in summations (e.g. Taylor series) to fail at runtime. Now correctly delegates to intPow() when the exponent is a point interval with an integer value, preserving even/odd parity. Also handles the case where base is -1 and the exponent spans multiple integers by returning the conservative interval [-1, 1].

  • Factorial missing from interval-js compilation target: Expressions containing n! (e.g. \frac{(-1)^k x^{2k+1}}{(2k+1)!}) failed interval-js compilation with success: false. Added Factorial and Factorial2 interval functions and compilation handlers.

  • expr.unknowns included bound variables: Scoped constructs like Sum, Product, Integrate, and Block bind index variables in a local scope, but expr.unknowns was reporting them as free unknowns. For example, \sum_{k=0}^{10} k \cdot x returned ["k", "x"] instead of ["x"]. Now correctly excludes locally bound variables from the result.

  • Symbolic upper bounds missing from expr.unknowns: In expressions like \sum_{k=0}^{M} k \cdot x, the symbolic upper bound M was incorrectly excluded from unknowns because the scope's bindings map captured all symbols referenced during canonicalization. Now extracts bound variables structurally from Limits/Element/Assign/Declare expressions, so only true bound variables are excluded. This also fixes Block expressions where locally assigned variables (via Assign or Declare) were reported as unknowns.

  • Integrate with symbolic bounds compiled incorrectly: Same issue as Sum/Product — compileIntegrate() used normalizeIndexingSet() which converted symbolic bounds to NaN. Now uses extractLimits() and compiles bounds as expressions.

  • Interval piecewise test fix: Fixed test that incorrectly accessed result.lo directly instead of unwrapping the IntervalResult envelope (result.value.lo). The piecewise() function correctly returns IntervalResult objects.

0.51.1 2026-02-15​

Features​

  • #172 Degrees-Minutes-Seconds (DMS) notation: Parse and serialize geographic angle notation such as 9°30'15". The LaTeX parser now recognizes arc-minute (', \prime) and arc-second (", \doubleprime) symbols when they follow a degree symbol, producing Add(Quantity(…, deg), Quantity(…, arcmin), …) expressions that evaluate and simplify through the existing unit system. Negative angles (e.g. -45°30') are fully supported for latitude/longitude coordinates.
  • dmsFormat serialization option: Set dmsFormat: true in SerializeLatexOptions to serialize angle quantities as DMS notation (e.g. Quantity(9.5, deg) → 9°30').
  • angleNormalization serialization option: Normalize angles during serialization with '0...360' (useful for bearings) or '-180...180' (useful for longitude). Default is 'none'.
  • realOnly compilation option: Pass { realOnly: true } to compile() to automatically convert complex { re, im } results to real numbers — returns re when im === 0, NaN otherwise. Useful for plotting and other contexts that only need real-valued output.
  • Sinc function: Unnormalized cardinal sine sinc(x) = sin(x)/x with sinc(0) = 1. Includes LaTeX parsing via \operatorname{sinc}, JavaScript and interval-arithmetic compilation targets.
  • Fresnel integrals (FresnelS, FresnelC): Numeric evaluation using Cephes rational Chebyshev approximation, LaTeX parsing via \operatorname{FresnelS} / \operatorname{FresnelC}, JavaScript and interval-arithmetic compilation targets.
  • Heaviside step function: H(x) = 0 for x < 0, 1/2 for x = 0, 1 for x > 0. LaTeX parsing via \operatorname{Heaviside}, JavaScript and interval-arithmetic compilation with singularity detection at zero.

LaTeX Syntax​

  • Which compilation: \begin{cases} expressions now compile to JavaScript and interval-js targets as chained ternary operators with NaN fallback when no condition matches.
  • Sum/Product compilation: \sum_{k=a}^{b} and \prod_{k=a}^{b} expressions with numeric bounds now compile to JavaScript loops with accumulator variables, including complex number support.
  • Loop compilation: Loop, Break, Continue, and Return operators compile to JavaScript for loops wrapped in IIFEs with standard control flow keywords.
  • Inline If syntax: Parse \text{if } C \text{ then } A \text{ else } B (or \operatorname{if}) to ["If", C, A, B] expressions.
  • where syntax: Parse E \text{ where } x \coloneq V to Block expressions with implicit variable declarations.
  • Semicolon block syntax: Semicolons (;, \;) act as statement separators, building Block expressions with auto-declared variables when assignments are present.
  • for loop syntax: Parse \text{for } i \text{ from } a \text{ to } b \text{ do } body to ["Loop", body, ["Element", "i", ["Range", a, b]]].

Bug Fixes​

  • Interval-JS compilation for Gamma functions: Added missing gamma and gammaln exports and implementations in the interval-arithmetic library.
  • Interval-JS graceful fallback: The interval-js target no longer throws when encountering unsupported functions. Unsupported operators now produce { success: false } at compile time, and runtime errors return { kind: "entire" } instead of propagating.
  • CompilationResult.run type signature: The TypeScript type for run now correctly reflects the actual calling convention ((...args: unknown[])) instead of the previous misleading (...args: (number | {re, im})[]).
  • Loop compilation for interval-js target: Loop counter now uses raw numbers (not _IA.point()) for the for statement, with loop index references properly wrapped in the body. Conditions in if/break/ continue statements inside loops use scalar comparisons instead of interval comparison functions.

Other Changes​

  • Updated color palettes
  • Deduplicated runtime helper object (SYS_HELPERS) shared between ComputeEngineFunction and ComputeEngineFunctionLiteral in compilation target
  • Centralized sinc implementation in numerics/special-functions.ts (shared by library evaluation and JS compilation runtime)
  • Removed dead args === null checks in compilation base class

0.51.0 2026-02-14​

Colors​

  • New colors library: Four MathJSON operators for color manipulation and color space conversion, available as the "colors" library category.
  • Color: Parse a color string (hex 3/6/8-digit, rgb(), hsl(), named CSS color, transparent) into a canonical sRGB Tuple with components normalized to 0-1. Alpha is included as a fourth component when not equal to 1.
  • Colormap: Sample named visualization palettes. Three variants: no second argument returns the full palette as a List; integer n >= 2 resamples to n evenly spaced colors; real t in [0, 1] interpolates at position t using OKLCh color space with shorter-arc hue interpolation. Includes 8 sequential palettes (viridis, inferno, magma, plasma, cividis, turbo, rocket, mako), 6 categorical palettes (graph6, spectrum6, spectrum12, tableau10, tycho11, kelly22), and 12 diverging palettes (roma, vik, broc, rdbu, coolwarm, ocean-balance, plus reversed variants).
  • ColorToColorspace: Convert an sRGB color (string or Tuple) to components in "rgb", "hsl", "oklch", or "oklab" (alias "lab"). Preserves alpha when present.
  • ColorFromColorspace: Convert color space components back to a canonical sRGB Tuple. Accepts the same color space names as ColorToColorspace.
  • ColorToString: Convert a color (string or sRGB Tuple) to a formatted string. Supports optional format argument: "hex" (default), "rgb", "hsl", or "oklch" for CSS-style output. Alpha is included when not equal to 1.
  • ColorMix: Blend two colors in OKLCh space with an optional ratio (default 0.5). Accepts color strings or sRGB Tuple values. Interpolates lightness and chroma linearly, hue with shorter-arc interpolation.
  • ColorContrast: Compute the APCA contrast ratio between a background and foreground color. Returns a positive value for dark-on-light and negative for light-on-dark.
  • ContrastingColor: Choose the foreground color with better APCA contrast against a background. With one argument, picks between white and black. With three arguments, picks the better of two foreground candidates.
  • LaTeX color support: \textcolor{color}{body}, \colorbox{color}{body}, and \boxed{body} now roundtrip through Annotated expressions. Parsing and serialization are handled in the core Annotated infrastructure.
  • LaTeX font annotations: \textbf, \textit, \texttt, \textsf, \textup now serialize correctly from Annotated expressions via fontWeight, fontStyle, and fontFamily dict keys.
  • JavaScript compilation: All color operators (Color, ColorToString, ColorMix, ColorContrast, ContrastingColor, ColorToColorspace, ColorFromColorspace, Colormap) now compile to JavaScript.
  • oklab() CSS parsing: parseColor() now accepts oklab(L a b) and oklab(L a b / alpha) syntax, matching the existing oklch() support.
  • GPU compilation: ColorMix, ColorContrast, ContrastingColor, ColorToColorspace, and ColorFromColorspace now compile to GLSL and WGSL. Preamble functions provide sRGB ↔ OKLab ↔ OKLCh conversion, color mixing with shorter-arc hue interpolation, and APCA contrast on the GPU.
  • Added rgbToHsl() conversion function. Exported hslToRgb() (previously private).

Bug Fixes​

  • (#290) Derivatives of user-defined functions: \frac{d}{dx} f and f'(x) now correctly evaluate when f is a user-defined function (e.g., f(x) := 2x). Previously \frac{d}{dx} f returned 0 and f'(x) returned a symbolic Apply(Derivative(...)).
  • Cleaner D canonical form: f'(x) now canonicalizes to ["D", ["f", "x"], "x"] instead of the verbose ["D", ["Function", ["Block", ["f", "x"]], "x"], "x"]. Function calls are no longer redundantly wrapped in Function(Block(...)). Similarly, \frac{d}{dx} f where f is a known function symbol canonicalizes to ["D", ["f", "x"], "x"] by applying the function to the differentiation variable.

Free Functions​

  • Free functions (simplify, evaluate, N, expand, expandAll, factor, solve, compile) now accept ExpressionInput in addition to LatexString and Expression. This means you can pass numbers, MathJSON objects, or tuple arrays directly — e.g., evaluate(["Add", 1, 2]) or simplify(["Power", "x", 2]).
  • Added declare() free function to declare symbols without instantiating a ComputeEngine explicitly — e.g., declare('x', 'integer') or declare({ x: 'integer', y: 'real' }).

Units and Quantities​

  • New units library: A comprehensive unit system for physical quantities, available as the "units" library category. Supports SI base units, 18 named derived units, SI prefixes (quetta through quecto), and common non-SI units (imperial, angles, logarithmic).
  • Quantity expression: Pairs a numeric value with a unit: ["Quantity", 9.8, ["Divide", "m", ["Power", "s", 2]]]. Accessors QuantityMagnitude and QuantityUnit extract the parts.
  • Quantity arithmetic: Add, Subtract, Multiply, Divide, and Power are unit-aware. Addition and subtraction automatically convert compatible units and express the result in the unit with the largest scale factor (e.g., 12 cm + 1 m evaluates to 1.12 m). Incompatible dimensions remain unevaluated.
  • Unit conversion: UnitConvert converts between compatible units, including compound units like m/s to km/h. Supports affine temperature conversions (degC, degF, K). Returns an error for incompatible units. UnitSimplify reduces compound units to named derived units when possible (e.g., kg*m/s^2 to N).
  • Dimensional analysis: IsCompatibleUnit tests dimensional compatibility. UnitDimension returns the 7-element SI dimension vector. Both support compound unit expressions.
  • LaTeX parsing: \mathrm{...} and \text{...} containing recognized units produce Quantity expressions when juxtaposed with numbers. Compound units with /, ^, and \cdot are supported (e.g., 5\,\mathrm{m/s^{2}}).
  • siunitx commands: \qty{value}{unit}, \SI{value}{unit}, \unit{unit}, and \si{unit} are parsed.
  • LaTeX serialization: Quantity expressions serialize to value\,\mathrm{unit} notation.
  • DSL string sugar: Compound units can be specified as strings in MathJSON: ["Quantity", 9.8, "m/s^2"] is canonicalized to the structured form. Parentheses are supported for grouping: "kg/(m*s^2)".
  • Temperature units: degC and degF with affine offset conversions.
  • Angular unit unification: Trigonometric functions (Sin, Cos, Tan, etc.) accept Quantity arguments with angular units (deg, rad, grad, arcmin, arcsec) and convert to radians automatically.
  • Physics constants: 11 CODATA 2018 constants defined as Quantity expressions: SpeedOfLight, PlanckConstant, Mu0, StandardGravity, ElementaryCharge, BoltzmannConstant, AvogadroConstant, VacuumPermittivity, GravitationalConstant, StefanBoltzmannConstant, and GasConstant.

Compilation​

  • Tuple and Matrix compilation: Tuple and Matrix expressions can now be compiled across all targets. compile('(\\sin(t), \\cos(t))') produces [Math.sin(t), Math.cos(t)] in JavaScript, vec2(sin(t), cos(t)) in GLSL, vec2f(sin(t), cos(t)) in WGSL, and (np.sin(t), np.cos(t)) in Python.
  • GPU-native matrix types: Square matrices (2x2, 3x3, 4x4) compile to native GPU matrix constructors (mat2/mat3/mat4 in GLSL, mat2x2f/mat3x3f/mat4x4f in WGSL) with proper column-major transposition. Column vectors are flattened to vecN/vecNf instead of nested single-element arrays.
  • Complex number compilation: The JavaScript compilation target now supports complex-valued expressions. The compiler performs static type analysis at compile time to determine whether each subexpression is real or complex, and emits the appropriate code path. Simple arithmetic (Add, Subtract, Multiply, Divide, Negate) uses inline {re, im} field math to avoid allocation. Transcendental functions (Sin, Cos, Exp, Ln, Sqrt, Power, and others) delegate to runtime helpers backed by the complex-esm library. Mixed real/complex operands are promoted inline. ImaginaryUnit compiles to {re: 0, im: 1}. Symbols with unknown type are assumed real. Complex-aware Sum and Product loops emit {re, im} accumulators when the loop body is complex-valued. Reciprocal trig/hyperbolic functions (Cot, Sec, Csc, Coth, Sech, Csch) and their inverses dispatch to complex helpers when operands are complex.
  • Python complex compilation: The Python target now supports complex-valued expressions using Python's native complex() constructor and the cmath module for transcendental functions. Real-valued expressions continue to use NumPy.
  • Gamma function compilation: Gamma and GammaLn can now be compiled to interval-js, glsl, wgsl, and interval-glsl targets. The interval targets include pole detection at non-positive integers and correct monotonicity handling around the minimum at x ≈ 1.46.
  • Special function compilation: 27 additional functions can now be compiled to JavaScript: Erf, Erfc, ErfInv, Beta, Digamma, Trigamma, PolyGamma, Zeta, LambertW, BesselJ, BesselY, BesselI, BesselK, AiryAi, AiryBi, Factorial, Factorial2, Exp2, Log2, Log10, Lg, Arctan2, Hypot, Degrees, Haversine, InverseHaversine, Binomial, and Fibonacci.
  • GPU special functions: Erf, Erfc, ErfInv, Beta, Factorial, Arctan2, Hypot, Haversine, InverseHaversine, Log10, and Lg can now be compiled to GLSL and WGSL targets. Erf/ErfInv use Abramowitz & Stegun polynomial approximations; Beta and Factorial leverage the existing GPU Gamma preamble.

Simplification​

  • Factorial quotient simplification: n!/k! is now simplified to a partial product for both concrete integers (e.g., 10!/7! → 720) and symbolic expressions with small constant difference (e.g., n!/(n-2)! → n(n-1)).
  • Binomial detection: Expressions of the form n!/(k!(n-k)!) are automatically recognized and simplified to Binomial(n, k).
  • Binomial identity simplification: C(n,0) → 1, C(n,1) → n, C(n,n) → 1, C(n,n-1) → n.
  • Factorial sum factoring: Sums and differences of factorials with related arguments are factored out, e.g., n! - (n-1)! → (n-1)! * (n-1), (n+1)! + n! → n! * (n+2).

0.50.2 2026-02-12​

Numerics​

  • Centralized overflow protection: Improved robustness of Rational and ExactNumericValue arithmetic by centralizing overflow checks and automatic promotion to BigInt.
  • [#287](https://github.com/cortex-js/compute-engine/issues/287) Improved precision for large integer products: Multiplications and additions of large integers that would previously lose precision (exceeding Number.MAX_SAFE_INTEGER) are now automatically promoted to BigInt to maintain exact results.

Symbols​

  • #288 Allow reassigning a symbol from operator to value: ce.assign() no longer throws when assigning a plain value to a symbol that was previously declared as a function. Existing expressions using the symbol as a function head will produce a type error at evaluation time if the new value is not callable.

Evaluation​

  • Fixed scope leaks: Ensured that evaluation contexts are correctly popped even when an error or timeout occurs in BoxedFunction.evaluate(), findUnivariateRoots(), and rule-boxing operations.
  • Improved numerical evaluation performance: Sum, Product, Divide, and statistical operators (Mean, Variance, etc.) now correctly propagate the numericApproximation option, significantly speeding up large numerical calculations by avoiding expensive exact arithmetic.

0.50.1 2026-02-11​

Compilation​

  • CompilationResult.preamble for shader targets: compile() with interval-wgsl and interval-glsl targets now returns a preamble field containing the interval arithmetic library (struct definitions, helper functions). Previously, the compiled code referenced functions like ia_div and ia_sin that were not included in the output. Use preamble + code for a self-contained shader, or call compileShaderFunction() on the target directly.

0.50.0 2026-02-11​

Breaking API Changes​

This release includes several breaking changes to the public API.

The most significant is the restructuring of the Expression type hierarchy and the introduction of type-guarded role interfaces, which improves type safety and API ergonomics but requires updates to code that accessed role-specific properties directly on expression instances.

See MIGRATION_GUIDE_0.50.0.md for details.

Naming Alignment: Expression, MathJsonExpression, and ExpressionInput​

  • The compute-engine runtime type is now Expression (preferred name). BoxedExpression is retained as a deprecated alias for migration.
  • The MathJSON type is now MathJsonExpression (the old MathJSON Expression name has been removed from the math-json entrypoint).
  • SemiBoxedExpression is now ExpressionInput (with a deprecated alias for migration).

Role-Specific Properties Moved to Type-Guarded Interfaces​

Properties that were previously on all Expression instances (returning undefined when not applicable) have been moved to role interfaces. They are now only accessible after narrowing with a type guard.

Removed from ExpressionAccess via
.symbolisSymbol(expr) or isSymbol(expr, 'Pi') then expr.symbol
.stringisString(expr) then expr.string
.ops, .nops, .op1/.op2/.op3isFunction(expr) or isFunction(expr, 'Add') then expr.ops etc.
.numericValue, .isNumberLiteralisNumber(expr) then expr.numericValue
.tensorisTensor(expr) then expr.tensor
// Before
if (expr.symbol !== null) console.log(expr.symbol);

// After
import { isSymbol, sym } from '@cortex-js/compute-engine';

if (isSymbol(expr)) console.log(expr.symbol);
// isSymbol() accepts an optional symbol name:
if (isSymbol(expr, 'Pi')) { /* expr is the Pi symbol */ }
// or use the convenience helper:
if (sym(expr) === 'Pi') { /* ... */ }

// isFunction() accepts an optional operator name:
if (isFunction(expr, 'Add')) {
  // expr is narrowed to a function AND has operator 'Add'
  console.log(expr.ops);
}

Properties that remain on Expression: .operator, .re/.im, .shape, all arithmetic methods (.add(), .mul(), etc.), and all numeric predicates (.isPositive, .isInteger, etc.).

Expression Creation: form Replaces canonical/structural​

The canonical (boolean or array) and structural (boolean) options on ce.box(), ce.function(), and ce.parse() have been unified into a single form option.

ce.box(['Add', 1, 'x'], { form: 'canonical' }); // default
ce.box(['Add', 1, 'x'], { form: 'raw' });        // no canonicalization, no binding
ce.function('Add', [1, 'x'], { form: 'structural' }); // bound, not fully canonical
ce.box(['Add', 1, 'x'], { form: ['Number', 'Order'] }); // selective passes

New Free Functions​

Top-level free functions are now available for common operations and use a shared ComputeEngine instance created on first call.

FunctionPurpose
getDefaultEngine()Return the shared default ComputeEngine instance.
parse(latex)Parse a LaTeX string into an Expression.
simplify(exprOrLatex)Simplify an expression or LaTeX input.
evaluate(exprOrLatex)Evaluate an expression or LaTeX input symbolically.
N(exprOrLatex)Numerically evaluate an expression or LaTeX input.
assign(id, value) / assign(record)Assign one symbol value or many at once.
expand(exprOrLatex)Expand distributively at the top level (Expression | null).
expandAll(exprOrLatex)Expand distributively recursively (Expression | null).
solve(exprOrLatex, vars?)Solve equations/systems (returns solve result variants).
factor(exprOrLatex)Factor an expression.
compile(exprOrLatex, options?)Compile to a target language with CompilationResult.
import {
  getDefaultEngine,
  parse,
  simplify,
  evaluate,
  N,
  assign,
  expand,
  expandAll,
  solve,
  factor,
  compile,
} from '@cortex-js/compute-engine';

assign('x', 3);

const expr = parse('x^2 - 5x + 6');
solve(expr, 'x');           // [2, 3]
factor('(2x)(4y)');         // 8xy
compile('x^2 + 1').run({ x: 3 }); // 10

Except for parse(), assign(), and getDefaultEngine(), these free functions accept either a LaTeX string or an existing Expression.

Free Function Notes​

  • compile() is now a top-level entry point returning CompilationResult. Custom compilation targets are managed with ce.registerCompilationTarget() and ce.unregisterCompilationTarget().
  • expand() and expandAll() return null when an expression is not expandable.
  • solve() is available as a top-level wrapper over equation/system solving.
  • factor() is the top-level factoring entry point. Specialized helpers such as factorPolynomial() and factorQuadratic() remain expression-only APIs.

trigSimplify() Method Removed​

Use simplify({ strategy: 'fu' }) instead, which is equivalent.

// Before
const result = expr.trigSimplify();

// After
const result = expr.simplify({ strategy: 'fu' });

Library System​

The constructor now accepts a libraries option for controlling which libraries are loaded. Libraries declare their dependencies and are loaded in topological order.

// Load specific standard libraries
const ce = new ComputeEngine({
  libraries: ['core', 'arithmetic', 'trigonometry'],
});

// Add a custom library
const ce = new ComputeEngine({
  libraries: [
    ...ComputeEngine.getStandardLibrary(),
    { name: 'physics', requires: ['arithmetic'], definitions: { /* ... */ } },
  ],
});

User-Extensible Simplification Rules​

ce.simplificationRules is now a public getter/setter. Users can push additional rules or replace the entire rule set.

ce.simplificationRules.push({
  match: ['Power', ['Sin', '_x'], 2],
  replace: ['Subtract', 1, ['Power', ['Cos', '_x'], 2]],
});

Canonicalization​

  • Exact numeric folding during canonicalization: canonicalAdd and canonicalMultiply now fold exact numeric operands at canonicalization time, making behavior consistent with canonicalDivide which already folded coefficients. This means expressions are reduced earlier in the pipeline without waiting for a .simplify() call.

    What gets folded (exact values):

    • Integers: Add(2, x, 5) → Add(x, 7)
    • Rationals: Add(1/3, x, 2/3) → Add(x, 1)
    • Radicals: Add(√2, x, √2) → Add(x, 2√2)
    • Mixed exact: Multiply(2, x, 5) → Multiply(10, x)
    • Full reduction: Add(2, 3) → 5, Multiply(2, 3) → 6
    • Identity elimination: Multiply(1/2, x, 2) → x
    • Complex promotion: Add(1, Complex(0, -1)) → Complex(1, -1)

    What is NOT folded (non-exact values):

    • Machine floats: Add(1.5, x, 0.5) remains Add(x, 0.5, 1.5)
    • Infinity/NaN: Multiply(0, ∞) correctly returns NaN
    • Single numeric: Multiply(5, Pi) is unchanged (nothing to fold)

    The folding uses the existing ExactNumericValue arithmetic, which automatically handles radical grouping (√2 + √2 = 2√2) and rational simplification (1/3 + 2/3 = 1).

  • Exact numeric folding in canonicalPower: Integer powers of numeric literals are now folded during canonicalization when the exponent is an integer with |e| ≤ 64. For machine-number bases, the result must be a safe integer; for exact numeric values (rationals, radicals), NumericValue.pow() is used.

    • Power(2, 3) → 8
    • Power(3, 2) → 9
    • Power(1/2, 2) → 1/4
    • Power(-2, 3) → -8
    • Power(2, 100) remains unevaluated (exponent exceeds limit)

  • Complex promotion handles non-adjacent operands: canonicalAdd now combines a real float with imaginary terms even when they are not adjacent in the operand list. Previously, only a real immediately followed by an imaginary was promoted to a complex number.

Type Inference​

  • Type handlers for 25 operators: Added explicit type handlers to operators that were missing them, enabling the type system to return precise types instead of the broad signature return type.
    • Arithmetic: Factorial, Factorial2, Sign return finite_integer; Ceil and Floor return finite_integer for finite inputs, integer otherwise.
    • Trigonometry: Arctan uses numericTypeHandler (returns finite_real for real inputs, finite_number for complex).
    • Complex: Real, Imaginary, Argument return finite_real.
    • Number theory: Totient, Sigma0, Sigma1, Eulerian, Stirling, NPartition return finite_integer; SigmaMinus1 returns finite_rational.
    • Combinatorics: Choose, Fibonacci, Binomial, Multinomial, Subfactorial, BellNumber return finite_integer.
    • Truncate, GCD, LCM type handlers: Truncate returns finite_integer for finite inputs (matching Ceil/Floor); GCD and LCM always return finite_integer.

Solving​

  • And operator support for systems of equations: solve() now accepts And(Equal(...), Equal(...)) in addition to List(Equal(...), Equal(...)) for representing systems of equations. Both forms route through the same linear, polynomial, and inequality solvers.

  • Parametric solution type filtering: filterSolutionByTypes now uses === false instead of !== true for type predicate checks. This allows underdetermined (parametric) solutions to pass through when type predicates return undefined (unknown) rather than being incorrectly rejected.

  • Or operator support in solve(): Solving Or(Equal(x,1), Equal(x,2)) returns the union of solutions from each branch, with deduplication. Works for both univariate (returns array of values) and multivariate (returns array of records) cases.

  • Mixed equality + inequality systems: solve() now handles systems combining Equal and inequality operators (Less, LessEqual, Greater, GreaterEqual). Equalities are solved first, then solutions are filtered against the inequalities.

  • Parametric solutions omit free variables: Underdetermined linear systems no longer include free variables (self-referential entries) in the result record. Only dependent variables with non-trivial expressions are returned.

Special Functions​

  • Numeric evaluation for Digamma, Trigamma, PolyGamma, Beta, Zeta, LambertW: These six functions now evaluate numerically when .N() is called, at both machine precision and arbitrary precision (bignum). Returns unevaluated without numeric approximation.

    • Digamma/Trigamma: recurrence + asymptotic with Bernoulli numbers
    • PolyGamma: generalized recurrence for arbitrary order n
    • Beta: via gamma, with log-gamma fallback for large arguments
    • Zeta: Cohen-Villegas-Zagier acceleration, functional equation for \operatorname{Re}(s)<0
    • LambertW: Halley's method with branch-point handling

  • Arbitrary-precision (bignum) variants for special functions: When ce.precision > 15, Digamma, Trigamma, PolyGamma, Beta, Zeta, and LambertW now compute results to the requested precision using bignum arithmetic. The asymptotic shift threshold scales with precision to maintain accuracy (e.g., ce.precision = 50 produces 50-digit results for Digamma and Zeta).

  • Numeric evaluation for Bessel functions (BesselJ, BesselY, BesselI, BesselK): Integer-order Bessel functions now evaluate numerically.

    • BesselJ: power series for small |x|, Miller's backward recurrence for intermediate values, Hankel asymptotic expansion for large |x|
    • BesselY: DLMF 10.8.3 series for Y_0/Y_1, forward recurrence for higher orders, shared Hankel asymptotic with BesselJ
    • BesselI: power series + asymptotic expansion
    • BesselK: series for K_0, Wronskian-derived K_1, forward recurrence for higher orders, asymptotic for large x

  • Numeric evaluation for Airy functions (AiryAi, AiryBi): Power series using Maclaurin coefficients for |x| \leq 5, asymptotic expansions (exponential decay for Ai, exponential growth for Bi at positive x, oscillatory for negative x) for large arguments.

Linear Algebra​

(Fix #285)

  • \begin{vmatrix} now parses to Determinant: The vmatrix LaTeX environment now produces ["Determinant", ["Matrix", ...]] instead of ["Matrix", ..., "'||'"]. Serialization round-trips correctly back to \begin{vmatrix}...\end{vmatrix} when the argument is a Matrix expression, and uses \det\left(...\right) for symbol arguments.

  • \begin{Vmatrix} now parses to Norm: The Vmatrix LaTeX environment now produces ["Norm", ["Matrix", ...]] instead of ["Matrix", ..., "'‖‖'"]. Serialization round-trips to \begin{Vmatrix}...\end{Vmatrix} when the argument is a Matrix, and uses \left\Vert...\right\Vert for symbol arguments.

  • A^{-1} produces Inverse for matrix-typed symbols and matrix expressions: When a symbol is declared with type matrix, parsing A^{-1} now returns ["Inverse", "A"] instead of ["Power", "A", -1]. This also works for inline matrix expressions, e.g. \begin{pmatrix}...\end{pmatrix}^{-1}. Undeclared symbols still fall through to the default Power/Divide handling, and function symbols still produce InverseFunction (e.g., \sin^{-1} → Arcsin).

  • Inverse serializes as ^{-1}: ["Inverse", "A"] now serializes to A^{-1} instead of \mathrm{Inverse}(A).

  • Power(A, -1) canonicalizes to Inverse(A) for matrices: When A has a matrix type, ce.box(["Power", "A", -1]) now canonicalizes to ["Inverse", "A"] instead of ["Divide", 1, "A"].

  • \det(A) and \tr(A) now parse correctly: Fixed Determinant and Trace LaTeX dictionary entries to use latexTrigger (\det, \tr) instead of symbolTrigger, which only matches plain identifiers. Both functions also accept plain text forms (det(A), tr(A)).

  • \det A and \tr A work without parentheses: Determinant and Trace now accept implicit arguments, so \det A parses as ["Determinant", "A"] (like \cos x parses as ["Cos", "x"]). Implicit arguments bind at multiplication precedence, so \det 2A + 1 parses as det(2A) + 1.

  • Determinant serialization uses \det A for simple arguments: Symbol arguments serialize as \det A instead of \det\left(A\right). Matrix arguments still serialize as \begin{vmatrix}...\end{vmatrix}.

  • Added standard LaTeX operators \ker, \dim, \deg, \hom: These commands are now in the MathJSON LaTeX dictionary as function entries with implicit arguments, so forms like \ker V, \dim V, \deg p, and \hom(V, W) parse correctly and serialize back to the corresponding standard operator notation. The corresponding function symbols (Kernel, Dimension, Degree, Hom) are also registered in the linear algebra library.

  • Implemented runtime evaluation for Kernel, Dimension, Degree, and Hom:

    • Kernel now computes a numeric null-space basis (for scalar/vector/matrix real inputs) and returns it as a list of basis vectors.
    • Dimension now evaluates finite dimensions for concrete tensors and collections, and computes dim(Hom(V, W)) = dim(V) * dim(W) when both dimensions are inferable.
    • Degree now evaluates polynomial degree for polynomial-form expressions while keeping ambiguous bare symbols (for example Degree(p)) unevaluated.
    • Hom now evaluates/simplifies its arguments while preserving the symbolic Hom(...) form.

LaTeX Parsing​

  • arguments: 'implicit' option for function dictionary entries: Function entries in the LaTeX dictionary can now set arguments: 'implicit' to accept bare arguments without parentheses (e.g., \det A), matching the behavior of trig functions. The default remains 'enclosure' (parentheses required). Applied to \det, \tr, \Re, \Im, \arg, \max, \min, \sup, \inf.

Simplification​

  • Infinity handling for 24+ functions: arctan(∞), arccot(±∞), tanh/coth/sech/csch(±∞), arsinh(-∞), arcosh(-∞), arccoth(±∞), arcsch(±∞), π^∞, ∞^n, (-∞)^{-n}, log_∞(x), log_{0.5}(∞), √∞, ∛∞ now all return correct limits.

  • Root edge cases: Root(x, 0) → NaN, Root(0, n), Root(1, n), Root(+∞, n), and Sqrt(+∞) now handled correctly.

  • Division edge cases: a/a → 1 now works for compound expressions (e.g., (π+1)/(π+1)); 2/0 → ComplexInfinity and 1/(1/0) → 0 propagate correctly.

  • Logarithm edge cases: Fixed infinity detection in simplify-log.ts (was using sym() which fails on BoxedNumber infinity values); added log_∞(∞) → NaN, base-aware log_c(0), guards for log_1(x) and log_c(c^x) evaluation.

  • Absolute value of odd functions: |arcsin(x)|, |sinh(x)|, |arsinh(x)|, |artanh(x)| now simplify to f(|x|).

  • Even function with abs argument: cosh(|x+2|) → cosh(x+2).

  • Trig period shifts: cot(π+x) → cot(x), csc(π+x) → -csc(x).

  • Ln simplification in Add/Multiply operands: ln(x^3) − 3·ln(x) → 0 and ln(x^√2) → √2·ln(x) now work; cost function bypassed for log rules that are mathematically valid but structurally more expensive.

  • Preserved function identity: Removed unconditional expansions of sinh/cosh → exp, arsinh/arcosh/artanh → ln, and arcsin → arctan2 that prevented abs/odd-function rules from firing.

Compilation​

  • WGSL (WebGPU Shading Language) Compilation Target: New built-in WGSL target for compiling mathematical expressions to WebGPU shaders.

    // Via the registry
    const result = compile(expr, { to: 'wgsl' });

    WGSL-specific differences from GLSL:

    • inverseSqrt (camelCase) instead of inversesqrt
    • % operator for mod instead of mod() function
    • vec2f/vec3f/vec4f constructors instead of vec2/vec3/vec4
    • array<f32, n>() instead of float[n]()
    • fn name(x: f32) -> f32 instead of float name(float x)
    • @vertex/@fragment/@compute entry points with struct-based I/O
    • @group/@binding uniform declarations and @workgroup_size for compute

  • Interval WGSL Compilation Target: New interval-wgsl target for interval arithmetic in WebGPU shaders, mirroring the existing interval-glsl target. Since WGSL does not support function overloading, the library uses _v suffixes for internal vec2f-parameter implementations (e.g., ia_add_v), while the public API (ia_add, ia_sin, etc.) takes IntervalResult values.

Bug Fixes​

  • Sequence type inference now returns a proper tuple type: Multi-argument Sequence expressions previously returned 'any' as their inferred type, losing all type information. They now return a tuple<...> type with each element's individual type preserved (e.g., Sequence(1, "a") types as tuple<integer, string>), consistent with the Tuple operator.

  • Subscript parsing now checks for collection type: The LaTeX subscript (_) parser now checks whether the LHS is a collection (symbol declared as indexed_collection, or a list literal) and produces At() directly at parse time, consistent with bracket indexing (x[i]). Multi-index subscripts on collections (A_{k,j}) are now correctly unpacked into separate At arguments instead of being wrapped in a Tuple.

  • NumericValue(0).mul(Infinity) now returns NaN: All three NumericValue subclasses (MachineNumericValue, BigNumericValue, ExactNumericValue) had an early-return if (this.isZero) return this in mul(), which returned 0 without checking if the other operand was infinity. 0 × ±∞ is now correctly indeterminate (NaN), and ±∞ × 0 is handled symmetrically.

  • Power simplification (a^n)^m -> a^{nm} now correctly guarded: The rule was applied unconditionally, which is mathematically incorrect when the base can be negative and exponents are non-integer. The classic counterexample: ((-1)^2)^{1/2} = 1, but (-1)^{2·1/2} = -1. The rule is now only applied when: (1) the base is non-negative, (2) the outer exponent is an integer, or (3) the inner exponent is an odd integer. This fix applies to canonicalization (canonicalPower), the pow() helper, and simplification (simplifyPower). As a result, (x^2)^{1/2} now correctly simplifies to |x| instead of x.

  • Power distribution rules now guarded for non-integer exponents: Three additional power distribution rules in pow() were applied unconditionally, producing wrong results when the exponent is non-integer and operands are negative. (1) (a/b)^c -> a^c / b^c — e.g. ((-2)(-3))^{1/2} = sqrt(6) but distributing gives (-2)^{1/2} * (-3)^{1/2} = -sqrt(6). (2) (a*b)^c -> a^c * b^c — same class of bug. (3) (-x)^n used n % 2 === 0 to test parity, but for non-integer n (e.g. 0.5), 0.5 % 2 = 0.5 falls to the odd branch, giving (-x)^{0.5} -> -(x^{0.5}) which is wrong. All three rules, plus the corresponding canonicalPower() Divide rule, now require integer exponents (or non-negative operands) before distributing.

  • Sqrt/Root exponent rearrangement now guarded: Two more rules in pow() unconditionally rearranged exponents. (1) (√a)^b -> √(a^b) rearranges (a^{1/2})^b to (a^b)^{1/2}, which is wrong for negative a (e.g. (√(-4))^3 = -8i but √((-4)^3) = 8i). Now only applied when a >= 0. The even-integer branches ((√a)^2 -> a, (√a)^{2k} -> a^k) remain unconditional since integer outer exponents are always safe. (2) Root(a,b)^c -> a^{c/b} combined exponents unconditionally. Now guarded with a >= 0 or c is integer. Audit of simplify-power.ts confirmed all rules there are already properly guarded.

  • Relational operators now evaluate: Seven relational operators (TildeFullEqual, TildeEqual, Approx, ApproxEqual, ApproxNotEqual, Precedes, Succeeds) previously had canonical handlers but no evaluate handlers, so expressions like Approx(3.14, 3.14) returned unevaluated. The approximate-equality family (TildeFullEqual, TildeEqual, Approx, ApproxEqual) now checks whether |a - b| <= tolerance via ce.chop(), with support for multi-argument chains. Precedes and Succeeds evaluate as numeric < and > respectively. Negated variants (NotApprox, NotTildeFullEqual, etc.) work automatically through the Not operator.

  • BoxedNumber.operator now returns specific numeric types: The operator property on BoxedNumber instances previously returned the generic 'Number' for all numeric values. It now returns specific types that match the internal type system: 'Integer' for integers, 'Rational' for non-integer rationals, 'Real' for floating-point numbers, 'Complex' for complex numbers with non-zero imaginary part, and 'NaN', 'PositiveInfinity', 'NegativeInfinity' for special values. This improves API consistency with the type property and enables more precise pattern matching and type discrimination in user code. Breaking change: Code that explicitly checks for .operator === 'Number' will need to be updated to check for specific numeric types or use the isNumber() type guard instead.

  • Non-XIDC Unicode characters in symbol names now encoded correctly: When parsing LaTeX symbols containing non-identifier Unicode characters via \unicode{...}, \char, or ^^XX escapes (e.g., figure dash U+2012 in \operatorname{speed\unicode{"2012}of\unicode{"2012}sound}), the characters are now encoded as ____XXXXXX (4 underscores + 6 hex digits) in the symbol name. This encoding is valid per isValidSymbol() and round-trips correctly: the serializer decodes ____XXXXXX back to \unicode{"XXXX"} in LaTeX output. Previously, these characters passed through raw and caused symbol validation to fail.

  • Assign to compound symbol names no longer misinterpreted as sequence definitions (fixes #286): ce.box(["Assign", "t_half", 10]) previously failed because the Assign evaluate handler split any symbol containing _ and treated it as a subscripted sequence definition. User-provided compound symbols like t_half or half_life are now assigned correctly. Sequence definitions via parsed LaTeX (e.g., L_0 := 1) continue to work as before.

0.35.6 2026-02-07​

Bug Fixes​

  • Monte Carlo improper integrals: Fixed two bugs in monteCarloEstimate() that produced incorrect results (typically NaN or Infinity) for improper integrals. The change-of-variables estimator was inverted (f(x) / \mathrm{jacobian} instead of f(x) * \mathrm{jacobian}), and the finite-interval scale factor b - a was applied to transformed domains where it is infinite. Affects NIntegrate and compiled integrate for any integral with infinite bounds.

Compilation​

  • Truncate, Remainder, and Mod for JS/GLSL targets: Added Truncate (Math.trunc / trunc), Remainder, and Mod to the JavaScript and GLSL compilation targets, matching the Python target which already had them.

  • Interval trunc and remainder: Added trunc() and remainder() to the interval arithmetic library. trunc has proper discontinuity detection (behaves like floor for positive, ceil for negative, continuous at zero). remainder(a, b) = a - b * round(a/b) composes existing interval operations with discontinuity detection inherited from round. Added corresponding mappings to both interval JavaScript and interval GLSL targets.

  • Interval Lb, Log, and Root for GLSL: Added ia_log2, ia_log10, and Root to the interval GLSL target for consistency with the interval JavaScript target.

  • Reverse cross-reference test: Added a test that verifies all core CE math functions have compilation support in every target. Currently all 5 targets have full coverage of the 47 compilable math functions.

0.35.5 2026-02-06​

Bug Fixes​

  • Compilation Target Function Name Mismatches: Fixed several function keys in compilation targets that did not match their canonical library operator names, causing silent compilation failures and runtime errors ("Unexpected value"). Affected mappings: Ceiling → Ceil, Sgn → Sign, LogGamma → GammaLn, Arcsinh → Arsinh, Arccosh → Arcosh, Arctanh → Artanh, Re → Real, Im → Imaginary, Arg → Argument across all five compilation targets.

  • Missing Library Operator Definitions: Added library definitions for Exp2, Fract, Log10, Log2, Remainder, and Truncate which were referenced by compilation targets but had no corresponding library entries. Exp2 canonicalizes to Power(2, x), Log10/Log2 canonicalize to Log with the appropriate base, and Fract, Remainder, Truncate have direct numeric evaluation.

  • Derivative Rule for GammaLn: Fixed the derivative table entry that used the non-canonical name LogGamma instead of GammaLn, preventing the derivative d/dx GammaLn(x) = Digamma(x) from being computed.

0.35.4 2026-02-06​

Interval Arithmetic​

  • Discontinuity Continuity Direction: Singular interval results now include an optional continuity field ('left' or 'right') indicating from which side the function is continuous at a jump discontinuity. Floor, Round, Fract, and Mod report 'right' (right-continuous), Ceil reports 'left' (left-continuous). Pole-type singularities (e.g., tan, 1/x) leave the field undefined. This is reflected in both the JavaScript and GLSL interval arithmetic targets (new IA_SINGULAR_RIGHT and IA_SINGULAR_LEFT status constants in GLSL).

0.35.3 2026-02-06​

Compilation​

  • Expanded Function Support Across All Targets: Added comprehensive function mappings to all five compilation targets (JavaScript, GLSL, Interval GLSL, Interval JavaScript, Python): reciprocal trig (Cot, Csc, Sec), inverse reciprocal trig (Arccot, Arccsc, Arcsec), hyperbolic (Sinh, Cosh, Tanh), reciprocal hyperbolic (Coth, Csch, Sech), inverse hyperbolic (Arcosh, Arsinh, Artanh, Arcoth, Arcsch, Arsech), and elementary functions (Sgn, Lb, Log with base, Square, Root, Fract).

  • Interval Discontinuity Detection: Floor, Ceil, Round, Sign, Fract, and Mod now correctly report singularities when an interval spans a discontinuity point, in both the JavaScript and GLSL interval arithmetic targets. Previously these functions returned normal interval bounds even across jump discontinuities, which could cause incorrect connecting lines in plotted curves.

  • New Interval Functions: Added Round, Fract, and Mod to the interval arithmetic targets (both JS and GLSL) with proper discontinuity detection.

0.35.2 2026-02-05​

Bug Fixes​

  • Decimal Number Representation: Numbers written with a decimal point (e.g., 6.02e23) are now correctly treated as approximate decimal values (BigNumericValue) rather than exact integers. Previously, 6.02e23 was incorrectly converted to the exact bigint 602000000000000000000000, which implied false precision and caused memory inefficiency for very large exponents. Numbers without a decimal point (e.g., 602e21) continue to be treated as exact integers when possible. This change aligns with the documented behavior of the parseNumbers: 'auto' option.

  • Scientific Notation Serialization (#284): Fixed toLatex() with scientific and adaptiveScientific notation options to produce properly normalized output. Previously, numbers like 6.02e23 would serialize as 602\cdot10^{21} instead of the expected 6.02\cdot10^{23}. The output now depends only on the numeric value and formatting options, not on the internal representation.

  • Numeric Sum Precision: Fixed precision loss when summing large integers with rational values (e.g., 12345678^3 + 1/3). The ExactNumericValue.sum() method now uses bignumRe instead of re to preserve full precision when handling large integer values from BigNumericValue.

  • Broadcastable Functions with Union/Any Types (#235): Broadcastable (threadable) functions like Multiply and Add no longer reject arguments whose type is a union of numeric and collection types (e.g., number | list) or any. Previously, declaring a symbol as ce.declare('a', 'number | list') and using it in ce.box(['Multiply', 'a', 'b']) would produce an incompatible-type error.

  • Division Canonicalization Over-Simplification (#227): Fixed A/A being incorrectly simplified to 1 during canonicalization for constant expressions that evaluate to infinity or zero, such as tan(π/2)/tan(π/2). This now correctly evaluates to NaN (since ∞/∞ is indeterminate) instead of 1. Expressions with free variables (e.g., x/x, sin(x)/sin(x)) continue to simplify to 1 per standard algebraic convention. Also fixed deferred constant divisions like 0/(1-1) and (1-1)/(1-1) to properly evaluate to NaN instead of remaining as unevaluated expressions.

0.35.1 2026-02-03​

Bug Fixes​

  • Interval Arithmetic (JS/GLSL): Fixed interval evaluation of compound arguments (e.g. sin(2x), sin(x+x), sin(x^2), cos(2x)) by propagating interval results through trig, elementary, and comparison functions in interval-js, and by adding IntervalResult overloads to the GLSL interval library for interval-glsl.

0.35.0 2026-02-02​

Parsing​

  • Large Integer Precision: Fixed precision loss when parsing integers exceeding Number.MAX_SAFE_INTEGER with parseNumbers: 'rational'. Large integers and rational numerators now use BigInt arithmetic to preserve exact values. Fixes #283.

Compilation​

  • Interval Arithmetic Targets: Added two new compilation targets for reliable singularity detection:
    • interval-js - Compiles to JavaScript using interval arithmetic
    • interval-glsl - Compiles to GLSL for GPU-based interval evaluation

0.34.0 2026-02-01​

Parsing​

  • \mathopen and \mathclose: The LaTeX parser supports \mathopen and \mathclose delimiter prefixes for matchfix operators (explicit delimiter spacing control), e.g. \mathopen(a, b\mathclose) and \mathopen{(}a, b\mathclose{)}.

  • Interval Notation Parsing: Added support for parsing mathematical interval notation from LaTeX, including half-open intervals. Addresses #254.

    // Half-open intervals (American notation)
    ce.parse('[3, 4)').json;   // → ["Interval", 3, ["Open", 4]]
    ce.parse('(3, 4]').json;   // → ["Interval", ["Open", 3], 4]

    // Open intervals (ISO/European notation)
    ce.parse(']3, 4[').json;   // → ["Interval", ["Open", 3], ["Open", 4]]

    // LaTeX bracket commands and sizing prefixes
    ce.parse('\\lbrack 3, 4\\rparen').json;  // → ["Interval", 3, ["Open", 4]]
    ce.parse('\\left[ 3, 4 \\right)').json;  // → ["Interval", 3, ["Open", 4]]
    ce.parse('\\bigl( 3, 4 \\bigr]').json;   // → ["Interval", ["Open", 3], 4]

    Contextual Parsing: Lists and tuples are automatically converted to intervals when used in set contexts (Element, Union, Intersection, etc.):

    ce.parse('x \\in [0, 1]').json;
    // → ["Element", "x", ["Interval", 0, 1]]

    ce.parse('[0, 1] \\cup [2, 3]').json;
    // → ["Union", ["Interval", 0, 1], ["Interval", 2, 3]]

    // Standalone notation remains backward compatible
    ce.parse('[0, 1]').json;  // → ["List", 0, 1]
    ce.parse('(0, 1)').json;  // → ["Tuple", 0, 1]

Compilation​

  • Custom Operator Compilation: The compile() method now supports overriding operators to use function calls instead of native operators. This enables compilation of vector/matrix operations and custom domain-specific languages. Addresses #240.

    // Override operators for vector operations
    const expr = ce.parse('v + w');
    const compiled = expr.compile({
      operators: {
        Add: ['add', 11],      // Convert + to add() function
        Multiply: ['mul', 12]  // Convert * to mul() function
      },
      functions: {
        add: (a, b) => a.map((v, i) => v + b[i]),
        mul: (a, b) => a.map((v, i) => v * b[i])
      }
    });

    const result = compiled({ v: [1, 2, 3], w: [4, 5, 6] });
    // → [5, 7, 9]

    Highlights:

    • Map operators via an object or a function
    • Function-name operators compile to calls; symbol operators compile to infix
    • Supports scalar/collection arguments and partial overrides

  • Exported Compilation Interfaces: Advanced users can now create custom compilation targets by using the exported CompileTarget interface, BaseCompiler class, and JavaScriptTarget class.

    import { BaseCompiler, JavaScriptTarget } from '@cortex-js/compute-engine';

    // Create a custom compilation target
    const customTarget = {
      language: 'my-dsl',
      operators: (op) => ({ Add: ['ADD', 11], Multiply: ['MUL', 12] }[op]),
      functions: (id) => id.toUpperCase(),
      var: (id) => `VAR("${id}")`,
      string: (s) => `"${s}"`,
      number: (n) => n.toString(),
      ws: () => ' ',
      preamble: '',
      indent: 0,
    };

    const expr = ce.parse('x + y * 2');
    const code = BaseCompiler.compile(expr, customTarget);
    // → "ADD(VAR("x"), MUL(VAR("y"), 2))"

    Exported building blocks include CompileTarget, LanguageTarget, CompilationOptions, CompiledExecutable, BaseCompiler, JavaScriptTarget, and GLSLTarget (plus helper types like CompiledOperators and CompiledFunctions).

  • Compilation Plugin Architecture: The Compute Engine now supports registering custom compilation targets, allowing you to compile mathematical expressions to any target language beyond the built-in JavaScript and GLSL targets.

    import { ComputeEngine, BaseCompiler } from '@cortex-js/compute-engine';

    const ce = new ComputeEngine();

    // Define a custom Python target
    class PythonTarget {
      // ... implementation (see documentation)
    }

    // Register the custom target
    ce.registerCompilationTarget('python', new PythonTarget());

    // Compile to Python
    const expr = ce.parse('\\sin(x) + \\cos(y)');
    const pythonCode = expr.compile({ to: 'python' });
    console.log(pythonCode.toString());
    // → math.sin(x) + math.cos(y)

    // Switch between targets
    const jsFunc = expr.compile({ to: 'javascript' });
    const glslCode = expr.compile({ to: 'glsl' });

    Notes:

    • Built-in targets: javascript (executable) and glsl (shader code)
    • Add targets via ce.registerCompilationTarget(name, target)
    • Switch targets with compile({ to: ... }) (or override once with target)

  • Python/NumPy Compilation Target: Added a complete Python/NumPy compilation target for scientific computing workflows. The PythonTarget class compiles mathematical expressions to NumPy-compatible Python code.

    import { ComputeEngine, PythonTarget } from '@cortex-js/compute-engine';

    const ce = new ComputeEngine();
    const python = new PythonTarget({ includeImports: true });

    // Register the target
    ce.registerCompilationTarget('python', python);

    // Compile expressions to Python
    const expr = ce.parse('\\sin(x) + \\cos(y)');
    const code = expr.compile({ to: 'python' });
    console.log(code.toString());
    // → import numpy as np
    //
    //   np.sin(x) + np.cos(y)

    // Generate complete Python functions
    const func = python.compileFunction(
      ce.parse('\\sqrt{x^2 + y^2}'),
      'magnitude',
      ['x', 'y'],
      'Calculate vector magnitude'
    );
    // Generates:
    // import numpy as np
    //
    // def magnitude(x, y):
    //     """Calculate vector magnitude"""
    //     return np.sqrt(x ** 2 + y ** 2)

    Highlights:

    • NumPy-compatible output (including arrays)
    • Function mapping for common math + linear algebra
    • Helpers for full functions, lambdas, and vectorized code

    See the Python/NumPy Target Guide for complete documentation and examples.

  • GLSL Compilation Target: New built-in GLSL (OpenGL Shading Language) target for compiling mathematical expressions to WebGL shaders.

    const expr = ce.parse('x^2 + y^2');
    const glslCode = expr.compile({ to: 'glsl' });
    console.log(glslCode.toString());
    // → pow(x, 2.0) + pow(y, 2.0)

    // Generate complete GLSL functions
    import { GLSLTarget } from '@cortex-js/compute-engine';
    const glsl = new GLSLTarget();

    const distExpr = ce.parse('\\sqrt{x^2 + y^2 + z^2}');
    const func = glsl.compileFunction(distExpr, 'distance3D', 'float', [
      ['x', 'float'],
      ['y', 'float'],
      ['z', 'float'],
    ]);
    console.log(func);
    // → float distance3D(float x, float y, float z) {
    //     return sqrt(pow(x, 2.0) + pow(y, 2.0) + pow(z, 2.0));
    //   }

    // Generate complete shaders
    const shader = glsl.compileShader({
      type: 'fragment',
      version: '300 es',
      outputs: [{ name: 'fragColor', type: 'vec4' }],
      body: [
        {
          variable: 'fragColor',
          expression: ce.box(['List', 1, 0, 0, 1]),
        },
      ],
    });

    Highlights:

    • Native vector/matrix operators and constructors
    • Float literal formatting (2.0)
    • Helpers for functions and complete shaders

Algebra​

  • Polynomial Factoring: The Factor function now supports comprehensive polynomial factoring including perfect square trinomials, difference of squares, and quadratic factoring with rational roots. Addresses #180 and #33.

    // Perfect square trinomials
    ce.parse('x^2 + 2x + 1').factor().latex;
    // → "(x+1)^2"

    ce.parse('4x^2 + 12x + 9').factor().latex;
    // → "(2x+3)^2"

    // Difference of squares
    ce.parse('x^2 - 4').factor().latex;
    // → "(x-2)(x+2)"

    // Quadratic with rational roots
    ce.box(['Factor', ['Add', ['Power', 'x', 2], ['Multiply', 5, 'x'], 6], 'x'])
      .evaluate().latex;
    // → "(x+2)(x+3)"

    Automatic Factoring in sqrt Simplification: Square roots now automatically factor their arguments before applying simplification rules, enabling expressions like √(x²+2x+1) to simplify to |x+1|.

    // Issue #180 - Now works!
    ce.parse('\\sqrt{x^2 + 2x + 1}').simplify().latex;
    // → "\\vert x+1\\vert"

    ce.parse('\\sqrt{4x^2 + 12x + 9}').simplify().latex;
    // → "\\vert 2x+3\\vert"

    ce.parse('\\sqrt{a^2 + 2ab + b^2}').simplify().latex;
    // → "\\vert a+b\\vert"

    Includes perfect square trinomials, difference of squares, and quadratics with rational roots. Helper functions are exported for advanced usage (factorPerfectSquare, factorDifferenceOfSquares, factorQuadratic, factorPolynomial).

    MathJSON API:

    ["Factor", expr]              // Auto-detect variable
    ["Factor", expr, variable]    // Explicit variable specification

    The enhanced factoring system works seamlessly with existing polynomial functions like Expand, Together, Cancel, PolynomialGCD, and others.

Simplification​

  • Absolute Value Power Simplification: Fixed simplification of |x^n| expressions with even and rational exponents. Previously, expressions like |x²| and |x^{2/3}| were not simplified. Now they correctly simplify based on the parity of the exponent's numerator. Addresses #181.

    ce.parse('|x^2|').simplify().latex;      // → "x^2" (even exponent)
    ce.parse('|x^3|').simplify().latex;      // → "|x|^3" (odd exponent)
    ce.parse('|x^{2/3}|').simplify().latex;  // → "x^{2/3}" (even numerator)
    ce.parse('|x^{3/2}|').simplify().latex;  // → "|x|^{3/2}" (odd numerator)

  • Assumption-Based Simplification: Simplification rules use assumptions about symbol signs:

    ce.assume(ce.parse('x > 0'));
    ce.parse('\\sqrt{x^2}').simplify().latex;  // → "x" (was "|x|")
    ce.parse('|x|').simplify().latex;          // → "x" (was "|x|")

    ce.assume(ce.parse('y < 0'));
    ce.parse('\\sqrt{y^2}').simplify().latex;  // → "-y"
    ce.parse('|y|').simplify().latex;          // → "-y"

  • Nested Root Simplification: Nested roots simplify to a single root:

    ce.box(['Sqrt', ['Sqrt', 'x']]).simplify()     // → root(4)(x)
    ce.box(['Root', ['Root', 'x', 3], 2]).simplify() // → root(6)(x)
    ce.box(['Sqrt', ['Root', 'x', 3]]).simplify()  // → root(6)(x)

    Applies to all combinations: sqrt(sqrt(x)), root(sqrt(x), n), sqrt(root(x, n)), and root(root(x, m), n).

  • Extended Coefficient Factoring in Power Combination: The power combination rule now handles additional coefficient forms when combining same-base powers in products:

    • Multi-prime coefficients: 12·2ˣ·3ˣ → 2^(x+2)·3^(x+1) (since 12 = 2²·3). All primes in the factorization must have a matching base. Non-matching multi-prime coefficients like 6·2ˣ are left unchanged.
    • Negative coefficients: -4·2ˣ → -2^(x+2), -8·2ˣ → -2^(x+3). The absolute value is factored and the sign is preserved.
    • Rational-radical coefficients: √2·2ˣ → 2^(x+½), 2√2·2ˣ → 2^(x+3/2), (√2/2)·2ˣ → 2^(x-½). Decomposes (num/den)·√radical into prime contributions from all three components (radical primes get half-integer exponents, numerator primes get positive exponents, denominator primes get negative exponents).
    • Rational coefficients: 2ˣ/4 → 2^(x-2), 3ˣ/9 → 3^(x-2). Factors both numerator (positive exponents) and denominator (negative exponents).

  • Improved Cost Function for Negated Powers: Negate(Power(...)) now costs 3 + cost(exponent), consistent with the cost of Multiply(-1, Power(...)). This makes the cost model more accurate when comparing negated power forms.

Assumptions & Types​

  • Improved ask() Queries: ce.ask() now matches patterns with wildcards correctly, can answer common "bound" queries such as ask(["Greater", "x", "_k"]) and ask(["Greater", "_x", "_k"]), normalizes inequality patterns for matching (e.g. ask(["Greater", "_x", 0])), and falls back to verify() for closed predicates when the fact is known but not stored as an explicit assumption.

  • Tri-state verify(): Implemented ce.verify() as a truth query that returns true, false or undefined when a predicate cannot be determined from the current assumptions and declarations. And/Or/Not use 3-valued logic.

  • Element/NotElement Type Membership: Element(x, T) and NotElement(x, T) now support type-style RHS (e.g. real, finite_real, number, any) in addition to set collections (e.g. RealNumbers, Integers).

  • Value Resolution from Equality Assumptions: After ce.assume(['Equal', symbol, value]), the symbol now evaluates to the assumed value:

    ce.assume(ce.box(['Equal', 'one', 1]));
    ce.box('one').evaluate();               // → 1 (was: 'one')
    ce.box(['Equal', 'one', 1]).evaluate(); // → True (was: ['Equal', 'one', 1])
    ce.box(['Equal', 'one', 0]).evaluate(); // → False
    ce.box('one').type.matches('integer');  // → true

    This also fixes comparison evaluation: Equal(symbol, assumed_value) now correctly evaluates to True instead of staying symbolic.

  • Inequality Evaluation Using Assumptions: Inequality comparisons can use transitive bounds extracted from assumptions.

    ce.assume(ce.box(['Greater', 'x', 4]));
    ce.box(['Greater', 'x', 0]).evaluate();  // → True (x > 4 > 0)
    ce.box(['Less', 'x', 0]).evaluate();     // → False
    ce.box('x').isGreater(0);                // → true
    ce.box('x').isPositive;                  // → true

  • Type Inference from Assumptions: Inequalities infer real; equalities infer from the value.

    ce.assume(ce.box(['Greater', 'x', 4]));
    ce.box('x').type.toString();  // → 'real' (was: 'unknown')

    ce.assume(ce.box(['Equal', 'one', 1]));
    ce.box('one').type.toString();  // → 'integer' (was: 'unknown')

  • Tautology and Contradiction Detection: ce.assume() returns 'tautology' for redundant assumptions and 'contradiction' for conflicts.

    ce.assume(ce.box(['Greater', 'x', 4]));

    // Redundant assumption (x > 4 implies x > 0)
    ce.assume(ce.box(['Greater', 'x', 0]));  // → 'tautology' (was: 'ok')

    // Conflicting assumption (x > 4 contradicts x < 0)
    ce.assume(ce.box(['Less', 'x', 0]));     // → 'contradiction'

    // Same assumption repeated
    ce.assume(ce.box(['Equal', 'one', 1]));
    ce.assume(ce.box(['Equal', 'one', 1]));  // → 'tautology'

    // Conflicting equality
    ce.assume(ce.box(['Less', 'one', 0]));   // → 'contradiction'

Solving​

  • Systems of Linear Equations: The solve() method now handles systems of linear equations parsed from LaTeX \begin{cases}...\end{cases} environments. Returns an object mapping variable names to their solutions.

    const e = ce.parse('\\begin{cases}x+y=70\\\\2x-4y=80\\end{cases}');
    const result = e.solve(['x', 'y']);
    console.log(result.x.json);  // 60
    console.log(result.y.json);  // 10

    // 3x3 systems work too
    const e2 = ce.parse('\\begin{cases}x+y+z=6\\\\2x+y-z=1\\\\x-y+2z=5\\end{cases}');
    const result2 = e2.solve(['x', 'y', 'z']);
    // → { x: 1, y: 2, z: 3 }

    Non-linear systems that don't match known patterns and inconsistent systems return null.

  • Non-linear Polynomial Systems: The solve() method now handles certain non-linear polynomial systems with 2 equations and 2 variables:

    • Product + sum pattern: Systems like xy = p, x + y = s are solved by recognizing that x and y are roots of the quadratic t² - st + p = 0.

    • Substitution method: When one equation is linear in one variable, it substitutes into the other equation and solves the resulting univariate equation.

    Returns an array of solution objects (multiple solutions possible):

    // Product + sum pattern
    const e = ce.parse('\\begin{cases}xy=6\\\\x+y=5\\end{cases}');
    const result = e.solve(['x', 'y']);
    // → [{ x: 2, y: 3 }, { x: 3, y: 2 }]

    // Substitution method
    const e2 = ce.parse('\\begin{cases}x+y=5\\\\x^2+y=7\\end{cases}');
    const result2 = e2.solve(['x', 'y']);
    // → [{ x: 2, y: 3 }, { x: -1, y: 6 }]

    Only real solutions are returned; complex solutions are filtered out.

  • Exact Rational Arithmetic in Linear Systems: The linear system solver now uses exact rational arithmetic throughout the Gaussian elimination process. Systems with fractional coefficients produce exact fractional results rather than floating-point approximations.

    const e = ce.parse('\\begin{cases}x+y=1\\\\x-y=1/2\\end{cases}');
    const result = e.solve(['x', 'y']);
    console.log(result.x.json);  // ["Rational", 3, 4]  (exact 3/4)
    console.log(result.y.json);  // ["Rational", 1, 4]  (exact 1/4)

    // Fractional coefficients
    const e2 = ce.parse('\\begin{cases}x/3+y/2=1\\\\x/4+y/5=1\\end{cases}');
    const result2 = e2.solve(['x', 'y']);
    // → { x: 36/7, y: -10/7 }

  • Linear Inequality Systems: The solve() method now handles systems of linear inequalities in 2 variables, returning the vertices of the feasible region (convex polygon). Supports all inequality operators: <, <=, >, >=.

    // Triangle: x >= 0, y >= 0, x + y <= 10
    const e = ce.parse('\\begin{cases}x\\geq 0\\\\y\\geq 0\\\\x+y\\leq 10\\end{cases}');
    const result = e.solve(['x', 'y']);
    // → [{ x: 0, y: 0 }, { x: 10, y: 0 }, { x: 0, y: 10 }]

    // Square: 0 <= x <= 5, 0 <= y <= 5
    const square = ce.parse('\\begin{cases}x\\geq 0\\\\x\\leq 5\\\\y\\geq 0\\\\y\\leq 5\\end{cases}');
    square.solve(['x', 'y']);
    // → [{ x: 0, y: 0 }, { x: 5, y: 0 }, { x: 5, y: 5 }, { x: 0, y: 5 }]

    Vertices are returned in counterclockwise convex hull order. Returns null for infeasible systems or non-linear constraints.

  • Under-determined Systems (Parametric Solutions): The solve() method now returns parametric solutions for under-determined linear systems (fewer equations than variables) instead of returning null. Free variables appear as themselves in the solution, with other variables expressed in terms of them.

    // Single equation with two variables
    const e = ce.parse('\\begin{cases}x+y=5\\end{cases}');
    const result = e.solve(['x', 'y']);
    // → { x: -y + 5, y: y }  (y is a free variable)

    // Two equations with three variables
    const e2 = ce.parse('\\begin{cases}x+y+z=6\\\\x-y=2\\end{cases}');
    const result2 = e2.solve(['x', 'y', 'z']);
    // → { x: -z/2 + 4, y: -z/2 + 2, z: z }  (z is a free variable)

    Inconsistent systems still return null.

  • Extended Sqrt Equation Solving: The equation solver now handles sqrt equations of the form √(f(x)) = g(x) by squaring both sides and solving the resulting polynomial. Extraneous roots are automatically filtered.

    ce.parse('\\sqrt{x+1} = x').solve('x');      // → [1.618...] (golden ratio)
    ce.parse('\\sqrt{2x+3} = x - 1').solve('x'); // → [4.449...]
    ce.parse('\\sqrt{3x-2} = x').solve('x');     // → [1, 2]
    ce.parse('\\sqrt{x} = x').solve('x');        // → [0, 1]

  • Two Sqrt Equation Solving: The equation solver now handles equations with two sqrt terms of the form √(f(x)) + √(g(x)) = e using double squaring. Both addition and subtraction forms are supported, and extraneous roots are automatically filtered.

    ce.parse('\\sqrt{x+1} + \\sqrt{x+4} = 3').solve('x');  // → [0]
    ce.parse('\\sqrt{x} + \\sqrt{x+7} = 7').solve('x');    // → [9]
    ce.parse('\\sqrt{x+5} - \\sqrt{x-3} = 2').solve('x');  // → [4]
    ce.parse('\\sqrt{2x+1} + \\sqrt{x-1} = 4').solve('x'); // → [46 - 8√29] ≈ 2.919

  • Nested Sqrt Equation Solving: The equation solver now handles nested sqrt equations of the form √(x + √x) = a using substitution. These patterns have √x inside the argument of an outer sqrt. The solver uses u = √x substitution, solves the resulting quadratic, and filters negative u values.

    ce.parse('\\sqrt{x + 2\\sqrt{x}} = 3').solve('x');  // → [11 - 2√10] ≈ 4.675
    ce.parse('\\sqrt{x + \\sqrt{x}} = 2').solve('x');   // → [9/2 - √17/2] ≈ 2.438
    ce.parse('\\sqrt{x - \\sqrt{x}} = 1').solve('x');   // → [φ²] ≈ 2.618

  • Quadratic Equations Without Constant Term: Added support for solving quadratic equations of the form ax² + bx = 0 (missing constant term). These are solved by factoring: x(ax + b) = 0 → x = 0 or x = -b/a.

    ce.parse('x^2 + 3x = 0').solve('x');  // → [0, -3]
    ce.parse('2x^2 - 4x = 0').solve('x'); // → [0, 2]

Subscripts & Indexing​

  • Subscript Evaluation Handler: Define custom evaluation functions for subscripted symbols like mathematical sequences using subscriptEvaluate:

    // Define a Fibonacci sequence
    ce.declare('F', {
      subscriptEvaluate: (subscript, { engine }) => {
        const n = subscript.re;
        if (!Number.isInteger(n) || n < 0) return undefined;
        // Calculate Fibonacci number...
        return engine.number(fibValue);
      },
    });

    ce.parse('F_{10}').evaluate();  // → 55
    ce.parse('F_5').evaluate();     // → 5
    ce.parse('F_n').evaluate();     // → stays symbolic (handler returns undefined)

    Both simple subscripts (F_5) and complex subscripts (F_{5}) are supported. When the handler returns undefined, the expression stays symbolic. Subscripted expressions with subscriptEvaluate have type number and can be used in arithmetic operations: ce.parse('F_{5} + F_{3}').evaluate() works correctly.

  • Type-Aware Subscript Handling: Subscripts on symbols declared as collection types (list, tuple, matrix, etc.) now automatically convert to At() indexing operations:

    ce.declare('v', 'list<number>');
    ce.parse('v_n');      // → At(v, n)
    ce.parse('v_{n+1}');  // → At(v, n+1)
    ce.parse('v_{i,j}');  // → At(v, Tuple(i, j))

    This works for both simple subscripts (v_n) and complex subscripts (v_{n+1}). The type of the At() expression is correctly inferred from the collection's element type, allowing subscripted collection elements to be used in arithmetic.

  • Complex Subscripts in Arithmetic (Issue #273): Subscript expressions like a_{n+1} can now be used in arithmetic operations without type errors:

    ce.parse('a_{n+1} + 1');     // → Add(Subscript(a, n+1), 1)
    ce.parse('2 * a_{n+1}');     // → Multiply(2, Subscript(a, n+1))
    ce.parse('a_{n+1}^2');       // → Power(Subscript(a, n+1), 2)

    Previously, complex subscripts would fail with "incompatible-type" errors when used in arithmetic contexts.

  • Multi-Index At() Support: The At function now supports multiple indices for accessing nested collections (e.g., matrices):

    const matrix = ce.box(['List', ['List', 2, 3, 4], ['List', 6, 7, 9]]);
    ce.box(['At', matrix, 1, 2]).evaluate();  // → 3 (row 1, column 2)

    The signature was updated from single index to variadic: (value: indexed_collection, index: (number|string)+) -> unknown

  • Text Subscripts: Added support for \text{} in subscripts, allowing descriptive subscript names:

    ce.parse('x_{\\text{max}}');  // → symbol "x_max"
    ce.parse('v_{\\text{initial}}');  // → symbol "v_initial"

Sequences​

  • Declarative Sequence Definitions: Define mathematical sequences using recurrence relations with the new declareSequence() method:

    // Fibonacci sequence
    ce.declareSequence('F', {
      base: { 0: 0, 1: 1 },
      recurrence: 'F_{n-1} + F_{n-2}',
    });
    ce.parse('F_{10}').evaluate();  // → 55
    ce.parse('F_{20}').evaluate();  // → 6765

    // Arithmetic sequence: a_n = a_{n-1} + 2, a_0 = 1
    ce.declareSequence('A', {
      base: { 0: 1 },
      recurrence: 'A_{n-1} + 2',
    });
    ce.parse('A_{5}').evaluate();  // → 11

    // Factorial via recurrence
    ce.declareSequence('H', {
      base: { 0: 1 },
      recurrence: 'n \\cdot H_{n-1}',
    });
    ce.parse('H_{5}').evaluate();  // → 120

    Features:

    • Base cases as index → value mapping
    • Recurrence relation as LaTeX string or BoxedExpression
    • Automatic memoization for efficient evaluation (configurable)
    • Custom index variable name (default: n)
    • Domain constraints (min/max valid indices)
    • Symbolic subscripts stay symbolic (e.g., F_k remains unevaluated)

    Alternatively, sequences can be defined using natural LaTeX assignment notation:

    // Arithmetic sequence via LaTeX
    ce.parse('L_0 := 1').evaluate();
    ce.parse('L_n := L_{n-1} + 2').evaluate();
    ce.parse('L_{5}').evaluate();  // → 11

    // Fibonacci via LaTeX
    ce.parse('F_0 := 0').evaluate();
    ce.parse('F_1 := 1').evaluate();
    ce.parse('F_n := F_{n-1} + F_{n-2}').evaluate();
    ce.parse('F_{10}').evaluate();  // → 55

    Base cases and recurrence can be defined in any order. The sequence is finalized when both are present.

  • Sequence Status API: Query the status of sequence definitions with getSequenceStatus():

    ce.parse('F_0 := 0').evaluate();
    ce.getSequenceStatus('F');
    // → { status: 'pending', hasBase: true, hasRecurrence: false, baseIndices: [0] }

    ce.parse('F_n := F_{n-1} + F_{n-2}').evaluate();
    ce.getSequenceStatus('F');
    // → { status: 'complete', hasBase: true, hasRecurrence: true, baseIndices: [0] }

    ce.getSequenceStatus('x');
    // → { status: 'not-a-sequence', hasBase: false, hasRecurrence: false }

  • Sequence Introspection API: Inspect and manage defined sequences:

    // Get sequence information
    ce.getSequence('F');
    // → { name: 'F', variable: 'n', baseIndices: [0, 1], memoize: true, cacheSize: 5 }

    // List all defined sequences
    ce.listSequences();  // → ['F', 'A', 'H']

    // Check if a symbol is a sequence
    ce.isSequence('F');  // → true
    ce.isSequence('x');  // → false

    // Manage memoization cache
    ce.getSequenceCache('F');  // → Map { 2 => 1, 3 => 2, ... }
    ce.clearSequenceCache('F');  // Clear cache for specific sequence
    ce.clearSequenceCache();     // Clear all sequence caches

  • Generate Sequence Terms: Generate a list of sequence terms with getSequenceTerms():

    ce.declareSequence('F', {
      base: { 0: 0, 1: 1 },
      recurrence: 'F_{n-1} + F_{n-2}',
    });

    ce.getSequenceTerms('F', 0, 10);
    // → [0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55]

    // With step parameter (every other term)
    ce.getSequenceTerms('F', 0, 10, 2);
    // → [0, 1, 3, 8, 21, 55]

  • Sum and Product over Sequences: Sum and Product now work seamlessly with user-defined sequences:

    ce.declareSequence('F', {
      base: { 0: 0, 1: 1 },
      recurrence: 'F_{n-1} + F_{n-2}',
    });

    ce.parse('\\sum_{k=0}^{10} F_k').evaluate();  // → 143
    ce.parse('\\prod_{k=1}^{5} A_k').evaluate();  // Works with any defined sequence

  • OEIS Integration: Look up sequences in the Online Encyclopedia of Integer Sequences (OEIS) and verify your sequences against known mathematical sequences:

    // Look up a sequence by its terms
    const results = await ce.lookupOEIS([0, 1, 1, 2, 3, 5, 8, 13]);
    // → [{ id: 'A000045', name: 'Fibonacci numbers', terms: [...], url: '...' }]

    // Check if your sequence matches a known OEIS sequence
    ce.declareSequence('F', {
      base: { 0: 0, 1: 1 },
      recurrence: 'F_{n-1} + F_{n-2}',
    });

    const result = await ce.checkSequenceOEIS('F', 10);
    // → { matches: [{ id: 'A000045', name: 'Fibonacci numbers', ... }], terms: [...] }

    Note: OEIS lookups require network access to oeis.org.

  • Multi-Index Sequences: Define sequences with multiple indices like Pascal's triangle P_{n,k} or grid-based recurrences:

    // Pascal's Triangle: P_{n,k} = P_{n-1,k-1} + P_{n-1,k}
    ce.declareSequence('P', {
      variables: ['n', 'k'],
      base: { 'n,0': 1, 'n,n': 1 },  // Pattern-based base cases
      recurrence: 'P_{n-1,k-1} + P_{n-1,k}',
      domain: { n: { min: 0 }, k: { min: 0 } },
      constraints: 'k <= n',  // k must not exceed n
    });

    ce.parse('P_{5,2}').evaluate();  // → 10
    ce.parse('P_{10,5}').evaluate(); // → 252

    Features:

    • Multiple index variables with variables: ['n', 'k']
    • Pattern-based base cases: 'n,0' matches any (n, 0), 'n,n' matches diagonal
    • Per-variable domain constraints
    • Constraint expressions (e.g., 'k <= n')
    • Composite key memoization (e.g., '5,2')
    • Full introspection support with isMultiIndex flag

    Pattern matching for base cases:

    • Exact values: '0,0' matches only (0, 0)
    • Wildcards: 'n,0' matches any value for n with k=0
    • Equality: 'n,n' matches when both indices are equal
    • Priority: exact matches are checked before patterns

Special Functions​

  • Special Function Definitions: Added type signatures for special mathematical functions, enabling them to be used in expressions without type errors:

    • Zeta - Riemann zeta function \zeta(s)
    • Beta - Euler beta function B(a,b) = \Gamma(a)\Gamma(b)/\Gamma(a+b)
    • LambertW - Lambert W function (product logarithm)
    • BesselJ, BesselY, BesselI, BesselK - Bessel functions of first/second kind
    • AiryAi, AiryBi - Airy functions

    These functions now have proper signatures and can be composed with other expressions: ce.box(['Add', 1, ['LambertW', 'x']]) works correctly.

  • Special Function LaTeX Parsing: Added LaTeX parsing support for special functions: \zeta(s), \Beta(a,b), \operatorname{W}(x), Bessel functions via \operatorname{J}, \operatorname{Y}, etc., and Airy functions via \operatorname{Ai}, \operatorname{Bi}.

Calculus​

  • LambertW Derivative: Added derivative rule for the Lambert W function: d/dx W(x) = W(x)/(x·(1+W(x)))

  • Bessel Function Derivatives: Added derivative support for all four Bessel function types using order-dependent recurrence relations:

    ce.box(['D', ['BesselJ', 'n', 'x'], 'x']).evaluate();
    // → 1/2 * BesselJ(n-1, x) - 1/2 * BesselJ(n+1, x)

    ce.box(['D', ['BesselI', 'n', 'x'], 'x']).evaluate();
    // → 1/2 * BesselI(n-1, x) + 1/2 * BesselI(n+1, x)

    ce.box(['D', ['BesselK', 'n', 'x'], 'x']).evaluate();
    // → -1/2 * BesselK(n-1, x) - 1/2 * BesselK(n+1, x)

    Chain rule is automatically applied for composite arguments.

  • Multi-Argument Function Derivatives: Added derivative support for:

    • Log(x, base) - Logarithm with custom base:

      ce.box(['D', ['Log', 'x', 2], 'x']).evaluate();  // → 1/(x·ln(2))
      ce.box(['D', ['Log', 'x', 'a'], 'x']).evaluate(); // → 1/(x·ln(a))

      Also handles cases where both x and base depend on the variable by applying the quotient rule to ln(x)/ln(base).

    • Discrete functions (Mod, GCD, LCM) - Return 0 as these are step functions with derivative 0 almost everywhere:

      ce.box(['D', ['Mod', 'x', 5], 'x']).evaluate();  // → 0
      ce.box(['D', ['GCD', 'x', 6], 'x']).evaluate();  // → 0

  • Integration of 1/(x·ln(x)) Pattern: Added support for integrating expressions where the denominator is a product and one factor is the derivative of another:

    ce.parse('\\int \\frac{1}{x\\ln x} dx').evaluate();  // → ln(|ln(x)|)
    ce.parse('\\int \\frac{3}{x\\ln x} dx').evaluate();  // → 3·ln(|ln(x)|)

    This uses u-substitution: since 1/x = d/dx(ln(x)), the integral becomes ∫ h'(x)/h(x) dx = ln|h(x)|.

  • Cyclic Integration for e^x with Trigonometric Functions: Added support for integrating products of exponentials and trigonometric functions that require the "solve for the integral" technique:

    ce.parse('\\int e^x \\sin x dx').evaluate();
    // → -1/2·cos(x)·e^x + 1/2·sin(x)·e^x

    ce.parse('\\int e^x \\cos x dx').evaluate();
    // → 1/2·sin(x)·e^x + 1/2·cos(x)·e^x

    // Also works with linear arguments:
    ce.parse('\\int e^x \\sin(2x) dx').evaluate();
    // → -2/5·cos(2x)·e^x + 1/5·sin(2x)·e^x

    ce.parse('\\int e^x \\cos(2x) dx').evaluate();
    // → 1/5·cos(2x)·e^x + 2/5·sin(2x)·e^x

    These patterns cannot be solved by standard integration by parts (which would lead to infinite recursion) and instead use direct formulas:

    • ∫ e^x·sin(ax+b) dx = (e^x/(a²+1))·(sin(ax+b) - a·cos(ax+b))
    • ∫ e^x·cos(ax+b) dx = (e^x/(a²+1))·(a·sin(ax+b) + cos(ax+b))

  • Derivative Recursion Safety: Added recursion protection to differentiate() with a depth limit (MAX_DIFFERENTIATION_DEPTH), returning undefined when the limit is exceeded.

  • Equation Equivalence in isEqual() (Issue #275): Two equations are now recognized as equivalent if they have the same solution set:

    ce.parse('2x+1=0').isEqual(ce.parse('x=-1/2'));   // → true
    ce.parse('3x+1=0').isEqual(ce.parse('6x+2=0'));   // → true

    Uses sampling to check whether (LHS₁-RHS₁)/(LHS₂-RHS₂) is a non-zero constant.

Logic​

  • Boolean Simplification Rules: Added absorption laws and improved boolean expression simplification:

    • Absorption: A ∧ (A ∨ B) → A and A ∨ (A ∧ B) → A
    • Idempotence: A ∧ A → A and A ∨ A → A
    • Complementation: A ∧ ¬A → False and A ∨ ¬A → True
    • Identity: A ∧ True → A and A ∨ False → A
    • Domination: A ∧ False → False and A ∨ True → True
    • Double negation: ¬¬A → A

    These rules are applied automatically during simplification:

    ce.box(['And', 'A', ['Or', 'A', 'B']]).simplify();  // → A
    ce.box(['Or', 'A', ['And', 'A', 'B']]).simplify();  // → A

  • Prime Implicants and Minimal Normal Forms: Added Quine-McCluskey algorithm for finding prime implicants/implicates and computing minimal CNF/DNF:

    • PrimeImplicants(expr) - Find all prime implicants (minimal product terms)
    • PrimeImplicates(expr) - Find all prime implicates (minimal sum clauses)
    • MinimalDNF(expr) - Convert to minimal DNF using prime implicant cover
    • MinimalCNF(expr) - Convert to minimal CNF using prime implicate cover
    // Find prime implicants (terms that can't be further simplified)
    ce.box(['PrimeImplicants', ['Or', ['And', 'A', 'B'], ['And', 'A', ['Not', 'B']]]]).evaluate();
    // → [A] (AB and A¬B combine to just A)

    // Compute minimal DNF
    ce.box(['MinimalDNF', ['Or',
      ['And', 'A', 'B'],
      ['And', 'A', ['Not', 'B']],
      ['And', ['Not', 'A'], 'B']
    ]]).evaluate();
    // → A ∨ B (simplified from 3 terms to 2)

    Limited to 12 variables to prevent exponential blowup; larger expressions return unevaluated.

Linear Algebra​

  • Matrix Decompositions: Added four matrix decomposition functions for numerical linear algebra:

    • LUDecomposition(A) → [P, L, U] - LU factorization with partial pivoting
    • QRDecomposition(A) → [Q, R] - QR factorization using Householder reflections
    • CholeskyDecomposition(A) → L - Cholesky factorization for positive definite matrices
    • SVD(A) → [U, Σ, V] - Singular Value Decomposition
    ce.box(['LUDecomposition', [[4, 3], [6, 3]]]).evaluate();
    // → [P, L, U] where PA = LU

    ce.box(['QRDecomposition', [[1, 2], [3, 4]]]).evaluate();
    // → [Q, R] where A = QR, Q orthogonal, R upper triangular

    ce.box(['CholeskyDecomposition', [[4, 2], [2, 2]]]).evaluate();
    // → L where A = LL^T

    ce.box(['SVD', [[1, 2], [3, 4]]]).evaluate();
    // → [U, Σ, V] where A = UΣV^T

Fixed​

  • replace() Literal Matching in Object Rules: .replace({ match: 'a', replace: 2 }) no longer treats 'a' as a wildcard (string rules like "a*x -> 2*x" still auto-wildcard).

    const expr = ce.box(['Add', ['Multiply', 'a', 'x'], 'b']);
    expr.replace({match: 'a', replace: 2}, {recursive: true});
    // → 2x + b (was: 2 - incorrectly matched entire expression)

  • forget() Clears Assumed Values: ce.forget() now clears values set by equality assumptions across all evaluation context frames.

    ce.assume(ce.box(['Equal', 'x', 5]));
    ce.box('x').evaluate();  // → 5
    ce.forget('x');
    ce.box('x').evaluate();  // → 'x' (was: 5)

  • Scoped Assumptions Clean Up on popScope(): Assumptions made inside a scope no longer leak after popScope().

    ce.pushScope();
    ce.assume(ce.box(['Equal', 'y', 10]));
    ce.box('y').evaluate();  // → 10
    ce.popScope();
    ce.box('y').evaluate();  // → 'y' (was: 10)

  • Extraneous Root Filtering for Sqrt Equations: Candidate solutions are now validated against the original expression (before clearing denominators / harmonization) to filter extraneous roots.

    Examples of equations that now correctly filter extraneous roots:

    • √x = x - 2 → returns [4] (filters out x=1)
    • √x + x - 2 = 0 → returns [1] (filters out x=4)
    • √x - x + 2 = 0 → returns [4] (filters out x=1)
    • x - 2√x - 3 = 0 → returns [9] (filters out x=1)
    • 2x + 3√x - 2 = 0 → returns [1/4] (filters out x=4)

  • Simplification (#178):

    • Safer division canonicalization for denominators that may simplify to 0
    • Implicit multiplication powers: xx → x^2
    • Targeted exp/log rewriting for \exp(\log(x)±y)

0.33.0 2026-01-30​

Bug Fixes​

Arithmetic and Infinity​

  • Division by Zero: Improved handling of division by zero:

    • 0/0 returns NaN (indeterminate form)
    • a/0 where a ≠ 0 returns ComplexInfinity (~∞) as a "better NaN" that indicates an infinite result with unknown sign
    • This applies to all forms including 1/0, x/0, and rational literals

  • Infinity Sign Propagation: Fixed infinity multiplication not propagating signs correctly. Now ∞ * (-2) = -∞ and -∞ * 2 = -∞ as expected.

  • Infinity Division: Fixed ∞/∞ incorrectly returning 1. Now correctly returns NaN (indeterminate form). The a/a → 1 simplification rule now excludes infinity values.

Trigonometry​

  • Trigonometric Period Identities: Fixed incorrect sign handling for csc(π+x) and cot(π+x):

    • csc(π+x) now correctly simplifies to -csc(x) (was incorrectly csc(x))
    • cot(π+x) now correctly simplifies to cot(x) (was incorrectly -cot(x), cotangent has period π)

  • Trigonometric Co-function Identities: Fixed co-function identities not applying to canonical form expressions. Now correctly simplifies:

    • sin(π/2 - x) → cos(x)
    • cos(π/2 - x) → sin(x)
    • tan(π/2 - x) → cot(x)
    • cot(π/2 - x) → tan(x)
    • sec(π/2 - x) → csc(x)
    • csc(π/2 - x) → sec(x)

  • Double Angle with Coefficient: Fixed 2sin(x)cos(x) not simplifying to sin(2x). The product-to-sum identity now handles coefficients:

    • 2sin(x)cos(x) → sin(2x)
    • c·sin(x)cos(x) → c·sin(2x)/2 for any coefficient c

  • Trigonometric Product Identities: Improved handling of trig products in simplification. The Multiply rule now correctly defers to trig-specific rules for patterns like sin(x)*cos(x) and tan(x)*cot(x), ensuring these are simplified to sin(2x)/2 and 1 respectively.

Logarithms and Exponentials​

  • Logarithm-Exponential Composition: Fixed log(exp(x)) incorrectly simplifying to x. Now correctly returns x/ln(10) ≈ 0.434x since log₁₀(eˣ) = x·log₁₀(e) = x/ln(10). The identity log(exp(x)) = x only holds for natural logarithm.

  • Logarithm of e: Added simplification for log(e) → 1/ln(10) ≈ 0.434 and log_c(e) → 1/ln(c) for any base c.

  • Logarithm Combination Base Preservation: Fixed log(x) + log(y) (base 10) incorrectly becoming ln(xy). Now correctly produces log(xy) preserving the original base.

  • Logarithm Quotient Rule: Added expansion rule for logarithm of quotients. ln(x/y) now simplifies to ln(x) - ln(y) when x and y are known positive. Similarly for any base: log_c(x/y) → log_c(x) - log_c(y).

  • Exponential-Logarithm Composition: Added simplification for exp(log(x)) where log has a different base than e. Now e^log(x) → x^{1/ln(10)} and more generally e^log_c(x) → x^{1/ln(c)} for any base c.

Powers and Exponents​

  • Zero Power with Symbolic Exponent: Fixed 0^π and similar expressions with positive symbolic exponents not simplifying. Now 0^x → 0 when x is known to be positive (including π, e, etc.).

  • Exponent Evaluation in Products: Fixed (x³)² · (y²)² not simplifying to x⁶y⁴. Numeric subexpressions in exponents (like 2×3 in x^{2×3}) are now evaluated when the expression is part of a product.

  • Negative Exponents on Fractions: Fixed (a/b)^{-n} not simplifying properly. Now (x³/y²)^{-2} correctly simplifies to y⁴/x⁶ during canonicalization by distributing the negative exponent.

  • Negative Base with Fractional Exponent: Fixed (-ax)^{p/q} returning complex results when p and q are both odd. Now correctly factors out the negative sign: (-2x)^{3/5} → -(2x)^{3/5} = -2^{3/5}·x^{3/5}, giving real results. This affects products like (-2x)^{3/5}·x which now correctly simplify to -2^{3/5}·x^{8/5} instead of returning an imaginary value.

Radicals​

  • Radical Perfect Square Factoring: Fixed √(x²y) not simplifying to |x|√y. Adjusted cost function to penalize radicals containing perfect squares, enabling the simplification rule to apply.

  • Generalized Root Extraction: Added comprehensive root simplification rules:

    • √[n]{x^m} → x^{m/n} for odd roots (always valid)
    • √[n]{x^m} → |x|^{m/n} for even roots with integer result
    • √{x^{odd}} → |x|^n · √x factoring (e.g., √{x⁵} → |x|²√x)
    • Handles all combinations: √[4]{x⁶} → |x|^{3/2}, √[3]{x⁶} → x²

  • Symbolic Radicals Preservation: Fixed numeric radicals (√2, ∛5, 2^{3/5}) being evaluated to floating-point approximations during multiplication. Now x * √2 stays as √2 · x instead of 1.414... · x, and x * 2^{1/3} stays as x · ∛2 instead of 1.259... · x. This preserves exact irrational values and allows proper algebraic manipulation. Use .N() to get numeric approximations when needed.

LaTeX Parsing​

  • LaTeX \exp() Juxtaposition: Fixed adjacent \exp() calls not parsing as multiplication. Now \exp(x)\exp(2) correctly parses as e^x · e^2 instead of producing a parse error. The expression then simplifies to e^{x+2} as expected.

Features​

Trigonometry​

  • Fu Algorithm for Trigonometric Simplification: Implemented the Fu algorithm based on Fu, Zhong, and Zeng's paper "Automated and readable simplification of trigonometric expressions" (2006). This provides systematic, high-quality trigonometric simplification through:

    • Transformation Rules (TR1-TR22): Comprehensive set of rewrite rules including reciprocal conversions (sec→1/cos), ratio forms (tan→sin/cos), Pythagorean substitutions (sin²+cos²=1), power reductions, product-to-sum, sum-to-product, angle expansion/contraction, and Morrie's law for cosine product chains.

    • Rule Lists (RL1, RL2): Organized application sequences for tan/cot expressions and sin/cos expressions respectively, with greedy selection of optimal results.

    • Cost Function: Minimizes trigonometric function count as primary metric, with leaf count as secondary, to find the most readable form.

    Usage:

    // Option 1: Use strategy option with simplify()
    const result = expr.simplify({ strategy: 'fu' });

    // Option 2: Dedicated trigSimplify() method
    const result = expr.trigSimplify();

    Examples:

    • sin(x)⁴ - cos(x)⁴ → -cos(2x)
    • tan(x)·cot(x) → 1
    • sin²(x) + cos²(x) → 1
    • 2sin(x)cos(x) → sin(2x)
    • cos(x)·cos(2x)·cos(4x) → sin(8x)/(8sin(x)) (Morrie's law)

    Enhanced Transformations:

    • TRmorrie with Rational Coefficients: Morrie's law now handles angles that are rational multiples of π, such as cos(π/9)·cos(2π/9)·cos(4π/9) → 1/8. The algorithm detects maximal geometric sequences and handles cases where the sine terms cancel to produce pure fractions.

    • TR12i Tangent Sum Identity: Recognizes the pattern tan(A) + tan(B) - k·tan(A)·tan(B) and simplifies to -tan(C) when A + B + C = π and k = tan(C). Works with standard angles (π/6, π/4, π/3, etc.) and handles sign variations.

    • TRpythagorean for Compound Expressions: Detects sin²(x) + cos²(x) pairs within larger Add expressions and simplifies them to 1, e.g., sin²(x) + cos²(x) + 2 → 3.

    • Early TR9 Sum-to-Product: Applies sum-to-product transformation before angle expansion to catch patterns like sin(x+h) + sin(x-h) → 2sin(x)cos(h) that would otherwise be expanded and lose their simplified form.

    • Dual Strategy Approach: The Fu strategy now tries both "Fu first" and "simplify first" approaches and picks the best result. This handles both Morrie-like patterns (which need Fu before evaluation) and period reduction patterns (which need simplification first for angle contraction).

  • Trigonometric Periodicity Reduction: Trigonometric functions now simplify arguments containing integer multiples of π:

    • sin(5π + k) → -sin(k) (period 2π, with sign change for odd multiples)
    • cos(4π + k) → cos(k) (period 2π)
    • tan(3π + k) → tan(k) (period π)
    • Works for all six trig functions: sin, cos, tan, cot, sec, csc
    • Handles both positive and negative multiples of π

  • Pythagorean Trigonometric Identities: Added simplification rules for all Pythagorean identities:

    • sin²(x) + cos²(x) → 1
    • 1 - sin²(x) → cos²(x) and 1 - cos²(x) → sin²(x)
    • sin²(x) - 1 → -cos²(x) and cos²(x) - 1 → -sin²(x)
    • tan²(x) + 1 → sec²(x) and sec²(x) - 1 → tan²(x)
    • 1 + cot²(x) → csc²(x) and csc²(x) - 1 → cot²(x)
    • a·sin²(x) + a·cos²(x) → a (with coefficient)

  • Trigonometric Equation Solving: The solve() method now handles basic trigonometric equations:

    • sin(x) = a → x = arcsin(a) and x = π - arcsin(a) (two solutions)
    • cos(x) = a → x = arccos(a) and x = -arccos(a) (two solutions)
    • tan(x) = a → x = arctan(a) (one solution per period)
    • cot(x) = a → x = arccot(a)
    • Supports coefficient form: a·sin(x) + b = 0
    • Domain validation: returns no solutions when |a| > 1 for sin/cos
    • Automatic deduplication of equivalent solutions (e.g., cos(x) = 1 → single solution 0)

Calculus​

  • (#163) Additional Derivative Notations: Added support for parsing multiple derivative notations beyond Leibniz notation:

    • Newton's dot notation for time derivatives: \dot{x} → ["D", "x", "t"], \ddot{x} for second derivative, \dddot{x} and \ddddot{x} for higher orders. The time variable is configurable via the new timeDerivativeVariable parser option (default: "t").

    • Lagrange prime notation with arguments: f'(x) now parses to ["D", ["f", "x"], "x"], inferring the differentiation variable from the function argument. Works for f''(x), f'''(x), etc. for higher derivatives.

    • Euler's subscript notation: D_x f → ["D", "f", "x"] and D^2_x f or D_x^2 f for second derivatives.

    • Derivative serialization: D expressions now serialize to Leibniz notation (\frac{\mathrm{d}}{\mathrm{d}x}f) for consistent round-trip parsing.

  • Derivative Rules for Special Functions: Added derivative formulas for:

    • d/dx Digamma(x) = Trigamma(x)
    • d/dx Erf(x), d/dx Erfc(x), d/dx Erfi(x)
    • d/dx FresnelS(x), d/dx FresnelC(x)
    • d/dx LogGamma(x) = Digamma(x)

Special Functions​

  • Special Function Definitions: Added type signatures for Digamma, Trigamma, and PolyGamma functions to the library:
    • Digamma(x) - The digamma function ψ(x), logarithmic derivative of Gamma
    • Trigamma(x) - The trigamma function ψ₁(x), derivative of digamma
    • PolyGamma(n, x) - The polygamma function ψₙ(x), nth derivative of digamma

Logarithms and Exponentials​

  • Logarithm Combination Rules: Added simplification rules that combine logarithms with the same base:

    • ln(x) + ln(y) → ln(xy) (addition combines via multiplication)
    • ln(x) - ln(y) → ln(x/y) (subtraction combines via division)
    • log_c(x) + log_c(y) → log_c(xy) (works with any base)
    • log_c(x) - log_c(y) → log_c(x/y)
    • Handles multiple terms: ln(a) + ln(b) - ln(c) → ln(ab/c)

  • Exponential e Simplification: Added rules for combining powers of e:

    • eˣ · eʸ → e^(x+y) (same-base multiplication)
    • eˣ / eʸ → e^(x-y) (same-base division)
    • eˣ · e → e^(x+1) and eˣ / e → e^(x-1)
    • Preserves symbolic form instead of evaluating e^n numerically

Powers and Exponents​

  • Negative Base Power Simplification: Added rules to simplify powers with negated bases:

    • (-x)^n → x^n when n is even (e.g., (-x)^4 → x^4)
    • (-x)^n → -x^n when n is odd (e.g., (-x)^3 → -x^3)
    • (-x)^{n/m} → x^{n/m} when n is even and m is odd
    • (-x)^{n/m} → -x^{n/m} when both n and m are odd
    • (-1)^{p/q} → -1 when both p and q are odd (real odd root)

  • Power Distribution: Added rule to distribute integer exponents over products:

    • (ab)^n → a^n · b^n when n is an integer
    • Example: (x³y²)² → x⁶y⁴
    • Example: (-2x)² → 4x²

  • Same-Base Power Combination: Improved power combination for products with 3+ terms:

    • a³ · a · a² → a⁶ (combines all same-base terms)
    • Works with unknown symbols when sum of exponents is positive
    • Handles mixed products: b³c²dx⁷ya⁵gb²x⁵(3b) → 3dgyx¹²b⁶a⁵c²

Sum and Product​

  • (#133) Element-based Indexing Sets for Sum/Product: Added support for \in notation in summation and product subscripts:

    • Parsing: \sum_{n \in \{1,2,3\}} n now correctly parses to ["Sum", "n", ["Element", "n", ["Set", 1, 2, 3]]] instead of silently dropping the constraint.

    • Evaluation: Sums and products over finite sets, lists, and ranges are now evaluated correctly:

      • \sum_{n \in \{1,2,3\}} n → 6
      • \sum_{n \in \{1,2,3\}} n^2 → 14
      • \prod_{k \in \{1,2,3,4\}} k → 24

    • Serialization: Element-based indexing sets serialize back to LaTeX with proper \in notation: \sum_{n\in \{1, 2, 3\}}n

    • Range support: Works with Range expressions via ce.box(): ["Sum", "n", ["Element", "n", ["Range", 1, 5]]] → 15

    • Bracket notation as Range: Two-element integer lists in bracket notation [a,b] are now treated as Range(a,b) when used in Element context:

      • \sum_{n \in [1,5]} n → 15 (iterates 1, 2, 3, 4, 5)
      • Previously returned 6 (treated as List with just elements 1 and 5)

    • Interval support: Interval expressions work with Element-based indexing, including support for Open and Closed boundary markers:

      • ["Interval", 1, 5] → iterates integers 1, 2, 3, 4, 5 (closed bounds)
      • ["Interval", ["Open", 0], 5] → iterates 1, 2, 3, 4, 5 (excludes 0)
      • ["Interval", 1, ["Open", 6]] → iterates 1, 2, 3, 4, 5 (excludes 6)

    • Infinite series with Element notation: Known infinite integer sets are converted to their equivalent Limits form and iterated (capped at 1,000,000):

      • NonNegativeIntegers (ℕ₀) → iterates from 0, like \sum_{n=0}^{\infty}
      • PositiveIntegers (ℤ⁺) → iterates from 1, like \sum_{n=1}^{\infty}
      • Convergent series produce numeric approximations: \sum_{n \in \Z^+} \frac{1}{n^2} → ≈1.6449 (close to π²/6)

    • Non-enumerable domains stay symbolic: When the domain cannot be enumerated (unknown symbol, non-iterable infinite set, or symbolic bounds), the expression stays symbolic instead of returning NaN:

      • \sum_{n \in S} n with unknown S → stays as ["Sum", "n", ["Element", "n", "S"]]
      • \sum_{n \in \Z} n → stays symbolic (bidirectional, can't forward iterate)
      • \sum_{x \in \R} f(x) → stays symbolic (non-countable)
      • \sum_{n \in [1,a]} n with symbolic bound → stays symbolic
      • Previously these would all return NaN with no explanation

    • Multiple Element indexing sets: Comma-separated Element expressions now parse and evaluate correctly:

      • \sum_{n \in A, m \in B} (n+m) → ["Sum", ..., ["Element", "n", "A"], ["Element", "m", "B"]]
      • Nested sums like \sum_{i \in A}\sum_{j \in B} i \cdot j evaluate correctly
      • Mixed indexing sets (Element + Limits) work together

    • Condition/filter support in Element expressions: Conditions can be attached to Element expressions to filter values from the set:

      • \sum_{n \in S, n > 0} n → sums only positive values from S
      • \sum_{n \in S, n \ge 2} n → sums values ≥ 2 from S
      • \prod_{k \in S, k < 0} k → multiplies only negative values from S
      • Supported operators: >, >=, <, <=, !=
      • Conditions are attached as the 4th operand of Element: ["Element", "n", "S", ["Greater", "n", 0]]

Linear Algebra​

  • Matrix Multiplication: Added MatrixMultiply function supporting:

    • Matrix × Matrix: A (m×n) × B (n×p) → result (m×p)
    • Matrix × Vector: A (m×n) × v (n) → result (m)
    • Vector × Matrix: v (m) × B (m×n) → result (n)
    • Vector × Vector (dot product): v1 (n) · v2 (n) → scalar
    • Proper dimension validation with incompatible-dimensions errors
    • LaTeX serialization using \cdot notation

  • Matrix Addition and Scalar Broadcasting: Add now supports element-wise operations on tensors (matrices and vectors):

    • Matrix + Matrix: Element-wise addition (shapes must match)
    • Scalar + Matrix: Broadcasts scalar to all elements
    • Vector + Vector: Element-wise addition
    • Scalar + Vector: Broadcasts scalar to all elements
    • Symbolic support: [[a,b],[c,d]] + [[1,2],[3,4]] evaluates correctly
    • Proper dimension validation with incompatible-dimensions errors

  • Matrix Construction Functions: Added convenience functions for creating common matrices:

    • IdentityMatrix(n): Creates an n×n identity matrix
    • ZeroMatrix(m, n?): Creates an m×n matrix of zeros (square if n omitted)
    • OnesMatrix(m, n?): Creates an m×n matrix of ones (square if n omitted)

  • Matrix and Vector Norms: Added Norm function for computing various norms:

    • Vector norms: L1 (sum of absolute values), L2 (Euclidean, default), L-infinity (max absolute value), and general Lp norms
    • Matrix norms: Frobenius (default, sqrt of sum of squared elements), L1 (max column sum), L-infinity (max row sum)
    • Scalar norms return the absolute value

  • Eigenvalues and Eigenvectors: Added functions for eigenvalue decomposition:

    • Eigenvalues(matrix): Returns list of eigenvalues (2×2: symbolic via characteristic polynomial; 3×3: Cardano's formula; larger: numeric QR)
    • Eigenvectors(matrix): Returns list of corresponding eigenvectors using null space computation via Gaussian elimination
    • Eigen(matrix): Returns tuple of (eigenvalues, eigenvectors)

  • Diagonal Function: Now fully implemented with bidirectional behavior:

    • Vector → Matrix: Creates a diagonal matrix from a vector (Diagonal([1,2,3]) → 3×3 diagonal matrix)
    • Matrix → Vector: Extracts the diagonal as a vector (Diagonal([[1,2],[3,4]]) → [1,4])

  • Higher-Rank Tensor Operations: Extended Transpose, ConjugateTranspose, and Trace to work with rank > 2 tensors:

    • Transpose: Swaps last two axes by default (batch transpose), or specify explicit axes with ['Transpose', T, axis1, axis2]
    • ConjugateTranspose: Same axis behavior as Transpose, plus element-wise complex conjugation
    • Trace (batch trace): Returns a tensor of traces over the last two axes. For a [2,2,2] tensor, returns [trace of T[0], trace of T[1]]. Optional axis parameters: ['Trace', T, axis1, axis2]

  • Reshape Cycling: Implements APL-style ravel cycling. When reshaping to a larger shape, elements cycle from the beginning: Reshape([1,2,3], (2,2)) → [[1,2],[3,1]]

  • Scalar Handling: Most linear algebra functions now handle scalar inputs:

    • Flatten(42) → [42] (single-element list)
    • Transpose(42) → 42 (identity)
    • Determinant(42) → 42 (1×1 matrix determinant)
    • Trace(42) → 42 (1×1 matrix trace)
    • Inverse(42) → 1/42 (scalar reciprocal)
    • ConjugateTranspose(42) → 42 (conjugate of real is itself)
    • Reshape(42, (2,2)) → [[42,42],[42,42]] (scalar replication)

  • Improved Error Messages: Operations requiring square matrices (Determinant, Trace, Inverse) now return expected-square-matrix error for vectors and tensors (rank > 2).

Performance​

  • Pattern Matching Optimization: Significantly improved performance of commutative pattern matching by adding early rejection guards:
    • Arity Guard: Patterns without sequence wildcards (__/___) now immediately reject expressions with mismatched operand counts instead of attempting factorial permutations
    • Anchor Fingerprint: Patterns with literal or symbolic anchors verify anchor presence before attempting permutation matching, eliminating impossible matches in O(n) time
    • Universal Anchoring: Extended the efficient anchor-based backtracking algorithm to all patterns with anchors, not just those with sequence wildcards
    • Hash Bucketing: For patterns with many anchors (4+) against large expressions (6+ operands), uses hash-based indexing to reduce anchor lookup from O(n×m) to O(n+m) average case
    • Example: Matching a + b + c + 1 against x + y + z now rejects immediately (arity mismatch: 4 vs 3) instead of trying 24 permutations

Bug Fixes​

Arithmetic​

  • Indeterminate Form Handling: Fixed incorrect results for mathematical indeterminate forms:

    • 0 * ∞ now correctly returns NaN (previously returned ∞)
    • ∞ / ∞ now correctly returns NaN (previously returned 1)
    • ∞^0 now correctly returns NaN (was already correct)
    • All combinations (0 * (-∞), (-∞) / ∞, etc.) are handled correctly

  • (#176) Power Combination Simplification: Fixed simplification failing to combine powers with the same base when one factor has an implicit exponent or when there are 3+ operands. Previously, expressions like 2 * 2^x, e * e^x * e^{-x}, and x^2 * x would not simplify. Now correctly simplifies to 2^(x+1), e, and x^3 respectively. The fix includes:

    • Extended power combination rules to support numeric literal bases
    • Added functional rule to handle n-ary Multiply expressions (3+ operands)
    • Adjusted simplification cost threshold from 1.2 to 1.3 to accept mathematically valid simplifications where exponents become slightly more complex (e.g., 2 * 2^x → 2^(x+1))

  • Symbolic Factorial: Fixed (n-1)! incorrectly evaluating to NaN instead of staying symbolic. The factorial evaluate function was attempting numeric computation on symbolic arguments. Now correctly returns undefined (keeping the expression symbolic) when the argument is not a number literal.

Linear Algebra​

  • Matrix Operations Type Validation: Fixed matrix operations (Shape, Rank, Flatten, Transpose, Determinant, Inverse, Trace, etc.) returning incorrect results or failing with type errors. The root cause was a type mismatch: function signatures expected matrix type (a 2D list with dimensions), but BoxedTensor.type returned list<number> without dimensions. Now BoxedTensor, BoxedFunction, and BoxedSymbol correctly derive shape and rank from their type's dimensions. Additionally, linear algebra functions now properly evaluate their operands before checking if they are tensors.

Calculus​

  • Numerical Integration: Fixed \int_0^1 \sin(x) dx returning NaN when evaluated numerically with .N(). The integrand was already wrapped in a Function expression by the canonical form, but the numerical evaluation code was wrapping it again, creating a nested function that returned a function instead of a number. Now correctly checks if the integrand is already a Function before wrapping.

LaTeX Parsing and Serialization​

  • Subscript Function Calls: Fixed parsing of function calls with subscripted names like f_\text{a}(5). Previously, this was incorrectly parsed as a Tuple instead of a function call because Subscript expressions weren't being canonicalized before the function call check. Now correctly recognizes that f_a(5) is a function call when the subscript canonicalizes to a symbol.

  • (#130) Prefix/Postfix Operator LaTeX Serialization: Fixed incorrect LaTeX output for prefix operators (like Negate) and postfix operators (like Factorial) when applied to expressions with lower precedence. Previously, Negate(Add(a, b)) incorrectly serialized as -a+b instead of -(a+b), causing round-trip failures where parsing the output produced a mathematically different expression. Similarly, Factorial(Add(a, b)) now correctly serializes as (a+b)! instead of a+b!. The fix ensures operands are wrapped in parentheses when their precedence is lower than the operator's precedence.

  • (#156) Logical Operator Precedence: Fixed parsing of logical operators \vee (Or) and \wedge (And) with relational operators. Previously, expressions like 3=4\vee 7=8 were incorrectly parsed with the wrong precedence. Now correctly parses as ["Or", ["Equal", 3, 4], ["Equal", 7, 8]]. Logical operators have lower precedence (230-235) than comparison operators (245) and set relations (240), so compound propositions parse correctly without requiring parentheses.

  • (#156) Logical Connective Arrows: Added support for additional arrow notation in logical expressions:

    • \rightarrow now parses as Implies (previously parsed as To for set/function mapping)
    • \leftrightarrow now parses as Equivalent (previously produced an "unexpected-command" error)
    • Long arrow variants now supported: \Longrightarrow, \longrightarrow → Implies; \Longleftrightarrow, \longleftrightarrow → Equivalent
    • The existing variants \Rightarrow, \Leftrightarrow, \implies, \iff continue to work
    • \to remains available for function/set mapping notation (e.g., f: A \to B)

Simplification​

  • Rules Cache Isolation: Fixed rules cache building failing with "Invalid rule" errors when user expressions had previously polluted the global scope. For example, parsing x(y+z) would add x as a symbol with function type to the global scope. Later, when the simplification rules cache was built, rule parsing would fail because wildcards like _x in rules would be type-checked against the polluted scope where x had incompatible type. The fix ensures rule parsing uses a clean scope that inherits only from the system scope (containing built-in definitions), not from user-polluted scopes.

  • Simplification Rules: Added and fixed several simplification rules:

    • x + x now correctly simplifies to 2x (term combination)
    • e^x * e^{-x} now correctly simplifies to 1 (exponential inverse)
    • sin(∞) and cos(∞) now correctly evaluate to NaN
    • tanh(∞) now correctly evaluates to 1, tanh(-∞) to -1
    • log_b(x^n) now correctly simplifies to n * log_b(x) (log power rule)
    • Improved cost function to prefer n * ln(x) form over ln(x^n)
    • Trigonometric functions now reduce arguments by their period (e.g., cos(5π + k) simplifies using cos(π + k) = -cos(k))

  • (#178) Non-Canonical Expression Simplification: Fixed .simplify() not working on expressions parsed with { canonical: false }. Previously, ce.parse('x+x', { canonical: false }).simplify() would return x+x instead of 2x. The bug was in the simplification loop detection: when canonicalizing before simplification, the non-canonical form was recorded in the "seen" set, and since isSame() considers non-canonical and canonical forms equivalent, the canonical form was incorrectly detected as already processed. Now the simplification correctly starts fresh when canonicalizing, allowing full simplification to proceed.

0.32.0 2026-01-28​

Bug Fixes​

Calculus​

  • (#230) Root Derivatives: Fixed the D operator not differentiating expressions containing the Root operator (n-th roots). Previously, D(Root(x, 3), x) (derivative of ∛x) would return an unevaluated derivative expression instead of computing the result. Now correctly returns 1/(3x^(2/3)), equivalent to the expected (1/3)·x^(-2/3). The fix adds a special case in the differentiate function to handle Root(base, n) by applying the power rule with exponent 1/n.

  • Abs Derivative: Fixed d/dx |x| returning an error when evaluated with a variable that has an assigned value. The derivative formula now uses Sign(x) instead of a complex Which expression that couldn't be evaluated symbolically.

  • Step Function Derivatives: Fixed D(floor(x), x), D(ceil(x), x), and D(round(x), x) causing infinite recursion. These step functions now correctly return 0 (the derivative is 0 almost everywhere). Also fixed a bug where derivative formulas that evaluate to 0 weren't recognized due to a falsy check.

  • Inverse Trig Integrals: Fixed incorrect integration formulas for arcsin, arccos, and arctan. The previous formulas were completely wrong. Correct:

    • ∫ arcsin(x) dx = x·arcsin(x) + √(1-x²)
    • ∫ arccos(x) dx = x·arccos(x) - √(1-x²)
    • ∫ arctan(x) dx = x·arctan(x) - (1/2)·ln(1+x²)

  • Erfc Derivative: Fixed incorrect derivative formula for erfc(x). Now correctly returns -2/√π · e^(-x²) (the negative of the erf derivative).

  • LogGamma Derivative: Added derivative rule for LogGamma(x) which returns Digamma(x) (the digamma/psi function).

  • Special Function Derivatives: Fixed derivative formulas for several special functions and removed incorrect ones:

    • Fixed d/dx erfi(x) = (2/√π)·e^(x²) (imaginary error function)
    • Fixed d/dx S(x) = sin(πx²/2) (Fresnel sine integral)
    • Fixed d/dx C(x) = cos(πx²/2) (Fresnel cosine integral)
    • Removed incorrect derivative formulas for Zeta, Digamma, PolyGamma, Beta, LambertW, Bessel functions, and Airy functions (these now return symbolic derivatives like Digamma'(x) instead of wrong numeric results)

  • Symbolic Derivative Evaluation: Fixed derivatives of unknown functions returning 0 instead of symbolic derivatives. For example, D(Digamma(x), x) now correctly returns Digamma'(x) (as Apply(Derivative(Digamma, 1), x)) instead of incorrectly returning 0.

LaTeX Parsing and Serialization​

  • (#256) Subscript Symbol Parsing: Fixed parsing of single-letter symbols with subscripts. Previously, i_A was incorrectly parsed as ["Subscript", ["Complex", 0, 1], "A"] because i was recognized as the imaginary unit before the subscript was processed. Now i_A correctly parses as the symbol i_A. This applies to all single-letter symbols including constants like e and i. Complex subscripts containing operators (n+1), commas (n,m), or parentheses ((n+1)) still produce Subscript expressions.

  • LaTeX Serialization: Fixed TypeScript error in power serialization where denom (a number | null) was incorrectly passed where an Expression was expected. Now correctly uses operand(exp, 2) to get the expression form.

  • (#168) Absolute Value: Fixed parsing of nested absolute value expressions that start with a double bar (e.g. ||3-5|-4|), which previously produced an invalid structure instead of evaluating correctly.

  • (#244) Serialization: Fixed LaTeX and ASCIIMath serialization ambiguity for negative bases and negated powers. Powers now render (-2)^2 (instead of -2^2) when the base is negative, and negated powers now render as -(2^2) rather than -2^2.

  • (#243) LaTeX Parsing: Fixed logic operator precedence causing expressions like x = 1 \vee x = 2 to be parsed incorrectly as x = (1 ∨ x) = 2 instead of (x = 1) ∨ (x = 2). Comparison operators (=, <, >, etc.) now correctly bind tighter than logic operators (\land, \lor, \veebar, etc.).

  • (#264) Serialization: Fixed LaTeX serialization of quantified expressions (ForAll, Exists, ExistsUnique, NotForAll, NotExists). Previously, only the quantifier symbol was output (e.g., \forall x instead of \forall x, x>y). The body of the quantified expression is now correctly serialized.

  • (#257) LaTeX Parsing: Fixed \gcd command not parsing function arguments correctly. Previously \gcd\left(24,37\right) would parse as ["Tuple", "GCD", ["Tuple", 24, 37]] instead of the expected ["GCD", 24, 37]. The \operatorname{gcd} form was unaffected. Also added support for \lcm as a LaTeX command (in addition to the existing \operatorname{lcm}).

  • (#223) Serialization: Fixed scientific/engineering LaTeX serialization dropping the leading coefficient for exact powers of ten. For example, 1000 now serializes to 1\cdot10^{3} (or 1\times10^{3} depending on exponentProduct) instead of 10^{3}.

  • LaTeX Parsing: Fixed \cosh incorrectly mapping to Csch instead of Cosh.

  • (#255) LaTeX Parsing: Fixed multi-letter subscripts like A_{CD} causing "incompatible-type" errors in arithmetic operations. Multi-letter subscripts without parentheses are now interpreted as compound symbol names (e.g., A_{CD} → A_CD, x_{ij} → x_ij, T_{max} → T_max). Use parentheses for expression subscripts: A_{(CD)} creates a Subscript expression where CD represents implicit multiplication. The Delimiter wrapper is now stripped from subscript expressions for cleaner output.

First-Order Logic​

  • (#263) Quantifier Scope: Fixed quantifier scope in First-Order Logic expressions. Previously, \forall x.P(x)\rightarrow Q(x) was parsed with the implication inside the quantifier scope: ["ForAll", "x", ["To", P(x), Q(x)]]. Now it correctly follows standard FOL conventions where the quantifier binds only the immediately following formula: ["To", ["ForAll", "x", P(x)], Q(x)]. This applies to all quantifiers (ForAll, Exists, ExistsUnique, NotForAll, NotExists) and all logical connectives (\rightarrow, \to, \implies, \land, \lor, \iff). Use explicit parentheses for wider scope: \forall x.(P(x)\rightarrow Q(x)). Also fixed quantifier type signatures to properly return boolean, enabling correct type checking when quantified expressions are used as arguments to logical operators.

Simplification​

  • Sign Simplification: Fixed Sign(x).simplify() returning 1 instead of -1 when x is negative. The simplification rule incorrectly returned ce.One for both positive and negative cases.

Type System​

  • Ceil Type Signature: Fixed Ceil function signature from (real) -> integer to (number) -> integer to match Floor. This resolves "incompatible-type" errors when computing derivatives of ceiling expressions or using Ceil in contexts expecting a general number type.

Polynomials​

  • Polynomial Degree Detection: Fixed polynomialDegree() returning 0 for expressions like e^x or e^(-x^2) when it should return -1 (not a polynomial). When the base of a power is constant but the exponent depends on the variable, this is not a polynomial. This bug caused infinite recursion in simplification when simplifying expressions containing exponentials, such as the derivative of erf(x) which is (2/√π)·e^(-x²).

Pattern Matching​

  • (#258) Pattern Matching: Fixed BoxedExpression.match() returning null when matching patterns against canonicalized expressions. Several cases are now handled:
    • Rational patterns now match expressions like ['Rational', 'x', 2] which are canonicalized to ['Multiply', ['Rational', 1, 2], 'x']
    • Power patterns now match ['Power', 'x', -1] which is canonicalized to ['Divide', 1, 'x'], returning {_base: x, _exp: -1}
    • Power patterns now match ['Root', 'x', 3] (cube root), returning {_base: x, _exp: ['Divide', 1, 3]}

Sum and Product​

  • (#252) Sum/Product: Fixed Sum and Product returning NaN when the body contains free variables (variables not bound by the index). For example, \sum_{n=1}^{10}(x) now correctly evaluates to 10x instead of NaN, and \prod_{n=1}^{5}(x) evaluates to x^5. Mixed expressions like \sum_{n=1}^{10}(n \cdot x) now return 55x. Also fixed toString() for Sum and Product expressions with non-trivial bodies (e.g., Multiply) which were incorrectly displayed as int().

Equation Solving​

  • (#242) Solve: Fixed solve() returning an empty array for equations with variables in fractions. For example, F = 3g/h solved for g now correctly returns Fh/3 instead of an empty array. The solver now clears denominators before applying solve rules, enabling it to handle expressions like a + bx/c = 0. Also added support for solving equations where the variable is in the denominator (e.g., a/x = b now returns x = a/b).

  • (#220) Solve: Fixed solve() returning an empty array for equations involving square roots of the unknown, e.g. 2x = \sqrt{5x}. The solver now handles equations of the form ax + b√x + c = 0 using quadratic substitution. Also added support for solving logarithmic equations like a·ln(x) + b = 0 which returns x = e^(-b/a).

Improvements​

First-Order Logic​

  • (#263) First-Order Logic: Added several improvements for working with First-Order Logic expressions:
    • Configurable quantifier scope: New quantifierScope parsing option controls how quantifier scope is determined. Use "tight" (default) for standard FOL conventions where quantifiers bind only the immediately following formula, or "loose" for scope extending to the end of the expression. 
      ce.parse('\\forall x. P(x)', { quantifierScope: 'tight' })  // default
      ce.parse('\\forall x. P(x)', { quantifierScope: 'loose' })
    • Automatic predicate inference: Single uppercase letters followed by parentheses (e.g., P(x), Q(a,b)) are now automatically recognized as predicate/function applications without requiring explicit declaration. This enables natural FOL syntax like \forall x. P(x) \rightarrow Q(x) to work out of the box.
    • Quantifier evaluation over finite domains: Quantifiers (ForAll, Exists, ExistsUnique, NotForAll, NotExists) now evaluate to boolean values when the bound variable is constrained to a finite set. For example: 
      ce.box(['ForAll', ['Element', 'x', ['Set', 1, 2, 3]], ['Greater', 'x', 0]]).evaluate()
      // Returns True (all values in {1,2,3} are > 0)
      ce.box(['Exists', ['Element', 'x', ['Set', 1, 2, 3]], ['Greater', 'x', 2]]).evaluate()
      // Returns True (3 > 2)
      ce.box(['ExistsUnique', ['Element', 'x', ['Set', 1, 2, 3]], ['Equal', 'x', 2]]).evaluate()
      // Returns True (only one element equals 2)
      Supports Set, List, Range, and integer Interval domains up to 1000 elements. Nested quantifiers are evaluated over the Cartesian product of their domains.
    • Symbolic simplification for quantifiers: Quantifiers now simplify automatically in special cases:
      • ∀x. True → True, ∀x. False → False
      • ∃x. True → True, ∃x. False → False
      • ∀x. P → P (when P doesn't contain x)
      • ∃x. P → P (when P doesn't contain x)
    • CNF/DNF conversion: New ToCNF and ToDNF functions convert boolean expressions to Conjunctive Normal Form and Disjunctive Normal Form respectively: 
      ce.box(['ToCNF', ['Or', ['And', 'A', 'B'], 'C']]).evaluate()
      // Returns (A ∨ C) ∧ (B ∨ C)
      ce.box(['ToDNF', ['And', ['Or', 'A', 'B'], 'C']]).evaluate()
      // Returns (A ∧ C) ∨ (B ∧ C)
      Handles And, Or, Not, Implies, Equivalent, Xor, Nand, and Nor operators using De Morgan's laws and distribution.
    • Boolean operator evaluation: Added evaluation support for Xor, Nand, and Nor operators with True/False arguments: 
      ce.box(['Xor', 'True', 'False']).evaluate()   // Returns True
      ce.box(['Nand', 'True', 'True']).evaluate()   // Returns False
      ce.box(['Nor', 'False', 'False']).evaluate()  // Returns True
    • N-ary boolean operators: Xor, Nand, and Nor now support any number of arguments:
      • Xor(a, b, c, ...) returns true when an odd number of arguments are true
      • Nand(a, b, c, ...) returns the negation of And(a, b, c, ...)
      • Nor(a, b, c, ...) returns the negation of Or(a, b, c, ...)
    • Satisfiability checking: New IsSatisfiable function checks if a boolean expression can be made true with some assignment of variables: 
      ce.box(['IsSatisfiable', ['And', 'A', ['Not', 'A']]]).evaluate()  // False
      ce.box(['IsSatisfiable', ['Or', 'A', 'B']]).evaluate()            // True
    • Tautology checking: New IsTautology function checks if a boolean expression is true for all possible variable assignments: 
      ce.box(['IsTautology', ['Or', 'A', ['Not', 'A']]]).evaluate()     // True
      ce.box(['IsTautology', ['And', 'A', 'B']]).evaluate()             // False
    • Truth table generation: New TruthTable function generates a complete truth table for a boolean expression: 
      ce.box(['TruthTable', ['And', 'A', 'B']]).evaluate()
      // Returns [["A","B","Result"],["False","False","False"],...]
    • Explicit Predicate function: Added a new Predicate function to explicitly represent predicate applications in First-Order Logic. Inside quantifier scopes (\forall, \exists, etc.), single uppercase letters followed by parentheses are now parsed as ["Predicate", "P", "x"] instead of ["P", "x"]. This distinguishes predicates from regular function applications and avoids naming conflicts with library functions. 
      ce.parse('\\forall x. P(x)').json
      // Returns ["ForAll", "x", ["Predicate", "P", "x"]]
      Outside quantifier scopes, P(x) is still parsed as ["P", "x"] to maintain backward compatibility with function definitions like Q(x) := ....
    • D(f, x) no longer maps to derivative: The LaTeX notation D(f, x) is not standard mathematical notation for derivatives and previously caused confusion with the D derivative function in MathJSON. Now D(f, x) in LaTeX parses as ["Predicate", "D", "f", "x"] instead of the derivative. Use Leibniz notation (\frac{d}{dx}f) for derivatives in LaTeX, or construct the derivative directly in MathJSON: ["D", expr, "x"].
    • N(x) no longer maps to numeric evaluation: Similarly, N(x) in LaTeX is CAS-specific notation, not standard math notation. Now N(x) parses as ["Predicate", "N", "x"] instead of the numeric evaluation function. This allows N to be used as a variable (e.g., "for all N in Naturals"). Use the .N() method for numeric evaluation, or construct it directly in MathJSON: ["N", expr].

Polynomials​

  • Polynomial Simplification: The simplify() function now automatically cancels common polynomial factors in univariate rational expressions. For example, (x² - 1)/(x - 1) simplifies to x + 1, (x³ - x)/(x² - 1) simplifies to x, and (x + 1)/(x² + 3x + 2) simplifies to 1/(x + 2). Previously, this required explicitly calling the Cancel function with a variable argument.

Sum and Product​

  • Sum/Product Simplification: Added simplification rules for Sum and Product expressions with symbolic bounds:
    • Constant body: \sum_{n=1}^{b}(x) simplifies to b * x
    • Triangular numbers (general bounds): \sum_{n=a}^{b}(n) simplifies to (b(b+1) - a(a-1))/2
    • Sum of squares: \sum_{n=1}^{b}(n^2) simplifies to b(b+1)(2b+1)/6
    • Sum of cubes: \sum_{n=1}^{b}(n^3) simplifies to [b(b+1)/2]^2
    • Geometric series: \sum_{n=0}^{b}(r^n) simplifies to (1-r^(b+1))/(1-r)
    • Alternating unit series: \sum_{n=0}^{b}((-1)^n) simplifies to (1+(-1)^b)/2
    • Alternating linear series: \sum_{n=0}^{b}((-1)^n * n) simplifies to (-1)^b * floor((b+1)/2)
    • Arithmetic progression: \sum_{n=0}^{b}(a + d*n) simplifies to (b+1)(a + db/2)
    • Sum of binomial coefficients: \sum_{k=0}^{n}C(n,k) simplifies to 2^n
    • Alternating binomial sum: \sum_{k=0}^{n}((-1)^k * C(n,k)) simplifies to 0
    • Weighted binomial sum: \sum_{k=0}^{n}(k * C(n,k)) simplifies to n * 2^(n-1)
    • Partial fractions (telescoping): \sum_{k=1}^{n}(1/(k(k+1))) simplifies to n/(n+1)
    • Partial fractions (telescoping): \sum_{k=2}^{n}(1/(k(k-1))) simplifies to (n-1)/n
    • Weighted squared binomial sum: \sum_{k=0}^{n}(k^2 * C(n,k)) simplifies to n(n+1) * 2^(n-2)
    • Weighted cubed binomial sum: \sum_{k=0}^{n}(k^3 * C(n,k)) simplifies to n²(n+3) * 2^(n-3)
    • Alternating weighted binomial sum: \sum_{k=0}^{n}((-1)^k * k * C(n,k)) simplifies to 0 (n ≥ 2)
    • Sum of binomial squares: \sum_{k=0}^{n}(C(n,k)^2) simplifies to C(2n, n)
    • Sum of consecutive products: \sum_{k=1}^{n}(k(k+1)) simplifies to n(n+1)(n+2)/3
    • Arithmetic progression (general bounds): \sum_{n=m}^{b}(a + d*n) simplifies to (b-m+1)(a + d(m+b)/2)
    • Product of constant: \prod_{n=1}^{b}(x) simplifies to x^b
    • Factorial: \prod_{n=1}^{b}(n) simplifies to b!
    • Shifted factorial: \prod_{n=1}^{b}(n+c) simplifies to (b+c)!/c!
    • Odd double factorial: \prod_{n=1}^{b}(2n-1) simplifies to (2b-1)!!
    • Even double factorial: \prod_{n=1}^{b}(2n) simplifies to 2^b * b!
    • Rising factorial (Pochhammer): \prod_{k=0}^{n-1}(x+k) simplifies to (x)_n
    • Falling factorial: \prod_{k=0}^{n-1}(x-k) simplifies to x!/(x-n)!
    • Telescoping product: \prod_{k=1}^{n}((k+1)/k) simplifies to n+1
    • Wallis-like product: \prod_{k=2}^{n}(1 - 1/k^2) simplifies to (n+1)/(2n)
    • Factor out constants: \sum_{n=1}^{b}(c \cdot f(n)) simplifies to c \cdot \sum_{n=1}^{b}(f(n)), and similarly for products where the constant is raised to the power of the iteration count
    • Nested sums/products: inner sums/products are simplified first, enabling cascading simplification
    • Edge cases: empty ranges (upper < lower) return identity elements (0 for Sum, 1 for Product), and single-iteration ranges substitute the bound value

0.31.0 2026-01-27​

Breaking Changes​

  • The [Length] function has been renamed to [Count].
  • The xsize property of collections has been renamed to count.
  • The xcontains() method of collections has been renamed to contains().
  • Handling of dictionaries (["Dictionary"] expressions and \{dict:...\} shorthand) has been improved.
  • Inverse hyperbolic functions have been renamed to follow the ISO 80000-2 standard: Arcsinh → Arsinh, Arccosh → Arcosh, Arctanh → Artanh, Arccoth → Arcoth, Arcsech → Arsech, Arccsch → Arcsch. The "ar" prefix (for "area") is mathematically correct since these functions relate to areas on a hyperbola, not arc lengths. Both LaTeX spellings (\arsinh and \arcsinh) are accepted as input (Postel's law).

Bug Fixes​

LaTeX Parsing​

  • Metadata Preservation: Fixed verbatimLatex not being preserved when parsing with preserveLatex: true. The original LaTeX source is now correctly stored on parsed expressions (when using non-canonical mode). Also fixed metadata (latex, wikidata) being lost when boxing MathJSON objects that contain these attributes.

  • String Parsing: Fixed parsing of \text{...} with preserveLatex: true which was incorrectly returning an "invalid-symbol" error instead of a string expression.

Calculus​

  • Derivatives: d/dx e^x now correctly simplifies to e^x instead of ln(e) * e^x. The hasSymbolicTranscendental() function now recognizes that transcendentals which simplify to exact rational values (like ln(e) = 1) should not be preserved symbolically.

  • Derivatives: d/dx log(x) now returns 1 / (x * ln(10)) symbolically instead of evaluating to 0.434... / x. Fixed by using substitution instead of function application when applying derivative formulas, which preserves symbolic transcendental constants.

Arithmetic​

  • Rationals: Fixed reducedRational() to properly normalize negative denominators before the early return check. Previously 1/-2 would not canonicalize to -1/2.

  • Arithmetic: Fixed .mul() to preserve logarithms symbolically. Previously multiplying expressions containing Ln or Log would evaluate the logarithm to its numeric value.

Serialization​

  • Serialization: Fixed case inconsistency in toString() output for trigonometric functions. Some functions like Cot were being serialized with capital letters while others like csc were lowercase. All trig functions now consistently serialize in lowercase (e.g., cot(x) instead of Cot(x)).

  • Serialization: Improved display of inverse trig derivatives and similar expressions:

    • Negative exponents like x^(-1/2) now display as 1/sqrt(x) in both LaTeX and ASCII-math output
    • When a sum starts with a negative term and contains a positive constant, the constant is moved to the front (e.g., -x^2 + 1 displays as 1 - x^2) while preserving polynomial ordering (e.g., x^2 - x + 3 stays unchanged)
    • d/dx arcsin(x) now displays as 1/sqrt(1-x^2) instead of (-x^2+1)^(-1/2)

  • Scientific Notation: Fixed normalization of scientific notation for fractional values (e.g., numbers less than 1).

Sum and Product​

  • Compilation: Fixed compilation of Sum and Product expressions.

  • Sum/Product: Fixed sum and prod library functions to correctly handle substitution of index variables.

New Features and Improvements​

Serialization​

  • Number Serialization: Added adaptiveScientific notation mode. When serializing numbers to LaTeX, this mode uses scientific notation but avoids exponents within a configurable range (controlled by avoidExponentsInRange). This provides a balance between readability and precision for numbers across different orders of magnitude.

Type System​

  • Refactored the type parser to use a modular architecture. This allows for better extensibility and maintainability of the type system.

Pattern Matching​

  • Pattern Matching: The validatePattern() function is now exported from the public API. Use it to check patterns for invalid combinations like consecutive sequence wildcards before using them.

Polynomials​

  • Polynomial Arithmetic: Added new library functions for polynomial operations:
    • PolynomialDegree(expr, var) - Get the degree of a polynomial
    • CoefficientList(expr, var) - Get the list of coefficients
    • PolynomialQuotient(dividend, divisor, var) - Polynomial division quotient
    • PolynomialRemainder(dividend, divisor, var) - Polynomial division remainder
    • PolynomialGCD(a, b, var) - Greatest common divisor of polynomials
    • Cancel(expr, var) - Cancel common factors in rational expressions

Calculus​

  • Integration: Significantly expanded symbolic integration capabilities:
    • Polynomial division: Integrals like ∫ x²/(x²+1) dx now correctly divide first, yielding x - arctan(x)
    • Repeated linear roots: ∫ 1/(x-1)² dx = -1/(x-1) and higher powers
    • Derivative pattern recognition: ∫ f'(x)/f(x) dx = ln|f(x)| is now recognized automatically
    • Completing the square: Irreducible quadratics like ∫ 1/(x²+2x+2) dx now yield arctan(x+1)
    • Reduction formulas: ∫ 1/(x²+1)² dx now works using reduction formulas
    • Mixed partial fractions: ∫ 1/((x-1)(x²+1)) dx now decomposes correctly
    • Factor cancellation: ∫ (x+1)/(x²+3x+2) dx simplifies before integrating
    • Inverse hyperbolic: Added ∫ 1/√(x²+1) dx = arcsinh(x) and ∫ 1/√(x²-1) dx = arccosh(x)
    • Arcsec pattern: Added ∫ 1/(x·√(x²-1)) dx = arcsec(x)
    • Trigonometric substitution: Added support for ∫√(a²-x²) dx, ∫√(x²+a²) dx, and ∫√(x²-a²) dx using trig/hyperbolic substitution

0.30.2 2025-07-15​

Breaking Changes​

  • The expr.value property reflects the value of the expression if it is a number literal or a symbol with a literal value. If you previously used the expr.value property to get the value of an expression, you should now use the expr.N().valueOf() method instead. The valueOf() method is suitable for interoperability with JavaScript, but it may result in a loss of precision for numbers with more than 15 digits.

  • BoxedExpr.sgn now returns undefined for complex numbers, or symbols with a complex-number value.

  • The ce.assign() method previously accepted ce.assign("f(x, y)", ce.parse("x+y")). This is now deprecated. Use ce.assign("f", ce.parse("(x, y) \\mapsto x+y") instead.

  • It was previously possible to invoke expr.evaluate() or expr.N() on a non-canonical expression. This will now return the expression itself.

    To evaluate a non-canonical expression, use expr.canonical.evaluate() or expr.canonical.N().

    That's also the case for the methods numeratorDenominator(), numerator(), and denominator().

    In addition, invoking the methods inv(), abs(), add(), mul(), div(), pow(), root(), ln() will throw an error if the expression is not canonical.

New Features and Improvements​

  • Collections now support lazy materialization. This means that the elements of some collection are not computed until they are needed. This can significantly improve performance when working with large collections, and allow working with infinite collections. For example:

    ce.box(['Map', 'Integers', 'Square']).evaluate().print();
    // -> [0, 1, 4, 9, 16, ...]

    Materialization can be controlled with the materialization option of the evaluate() method. Lazy collections are materialized by default when converted to a string or LaTeX, or when assigned to a variable.

  • The bindings of symbols and function expressions is now consistently done during canonicalization.

  • It was previously not possible to change the type of an identifier from a function to a value or vice versa. This is now possible.

  • Antiderivatives are now computed symbolically:

ce.parse(`\\int_0^1 \\sin(\\pi x) dx`).evaluate().print();
// -> 2 / pi
ce.parse(`\\int \\sin(\\pi x) dx`).evaluate().print();
// -> -cos(pi * x) / pi

Requesting a numeric approximation of the integral will use a Monte Carlo method:

ce.parse(`\\int_0^1 \\sin(\\pi x) dx`).N().print();
// -> 0.6366

  • Numeric approximations of integrals is several order of magnitude faster.

  • Added Number Theory functions: Totient, Sigma0, Sigma1, SigmaMinus1, IsPerfect, Eulerian, Stirling, NPartition, IsTriangular, IsSquare, IsOctahedral, IsCenteredSquare, IsHappy, IsAbundant.

  • Added Combinatorics functions: Choose, Fibonacci, Binomial, CartesianProduct, PowerSet, Permutations, Combinations, Multinomial, Subfactorial and BellNumber.

  • The symbol type can be refined to match a specific symbol. For example symbol<True>. The type expression can be refined to match expressions with a specific operator, for example expression<Add> is a type that matches expressions with the Add operator. The numeric types can be refined with a lower and upper bound. For example integer<0..10> is a type that matches integers between 0 and 10. The type real<1..> matches real numbers greater than 1 and rational<..0> matches non-positive rational numbers.

  • Numeric types can now be constrained with a lower and upper bound. For example, real<0..10> is a type that matches real numbers between 0 and 10. The type integer<1..> matches integers greater than or equal to 1.

  • Collections that can be indexed (list, tuple) are now a subtype of indexed_collection.

  • The map type has been replaced with dictionary for collections of arbitrary key-value pairs and record for collections of structured key-value pairs.

  • Support for structural typing has been added. To define a structural type, use ce.declareType() with the alias flag, for example:

    ce.declareType(
      "point", "tuple<x: integer, y: integer>",
      { alias: true }
    );

  • Recursive types are now supported by using the type keyword to forward reference types. For example, to define a type for a binary tree:

    ce.declareType(
      "binary_tree",
      "tuple<value: integer, left: type binary_tree?, right: type binary_tree?>",
    );

  • The syntax for variadic arguments has changeed. To indicate a variadic argument, use a + or * after the type, for example:

    ce.declare('f', '(number+) -> number');

    Use + for a non-empty list of arguments and * for a possibly empty list.

  • Added a rule to solve the equation a^x + b = 0

  • The LaTeX parser now supports the \placeholder[]{}, \phantom{}, \hphantom{}, \vphantom{}, \mathstrut, \strut and \smash{} commands.

  • The range of recognized sign values, i.e. as returned from BoxedExpression.sgn has been simplified (e.g. '...-infinity' and 'nan' have been removed)

  • The Power canonical-form is less aggressive - only carrying-out ops. as listed in doc. - is much more careful in its consideration of operand types & values... (for example, typically, exponents are required to be numbers: e.g. x^1 will simplify, but x^y (where y===0), or x^{1+0}, will not)

Issues Resolved​

  • Ensure expression LaTeX serialization is based on MathJSON generated with matching "pretty" formatting (or not), therefore resulting in LaTeX with less prettification, where prettify === false (#daef87f)

  • Symbols declare with a constant flag are now not marked as "inferred"

  • Some BoxedSymbols properties now more consistently return undefined, instead of a boolean (i.e. because the symbol is non-bound)

  • Some expr.root() computations

  • Canonical-forms

    • Fixes the Number form
    • Forms (at least, Number, Power) do not mistakenly fully canonicalize operands
    • This (partial canonicalization) now substitutes symbols (constants) with a holdUntil value of "never" during/prior-to canonicalization (i.e. just like for full canonicalization)

0.29.1 2025-03-31​

  • #231 During evaluation, some numbers, for example 10e-15 were incorrectly rounded to 0.

0.28.0 2025-02-06​

Issues Resolved​

  • #211 More consistent canonicalization and serialization of exact numeric values of the form (a√b)/c.

  • #219 The invisibleOperator canonicalization previously also canonicalized some multiplication.

  • #218 Improved performance of parsing invisible operators, including fixing some cases where the parsing was incorrect.

  • #216 Correctly parse subscripts with a single character, for example x_1.

  • #216 Parse some non-standard integral signs, for example \int x \cdot \differentialD x (both the \cdot and the \differentialD are non-standard).

  • #210 Numeric approximation of odd nth roots of negative numbers evaluate correctly.

  • #153 Correctly parse integrals with \limits, e.g. \int\limits_0^1 x^2 \mathrm{d} x.

  • Correctly serialize to ASCIIMath Delimiter expressions.

  • When inferring the type of numeric values do not constrain them to be real. As a result:

    ce.assign('a', ce.parse('i'));
    ce.parse('a+1').evaluate().print();

    now returns 1 + i instead of throwing a type error.

  • Correctly parse and evaluate unary and binary \pm and \mp operators.

New Features and Improvements​

  • expr.isEqual() will now return true/false if the expressions include the same unknowns and are structurally equal after expansion and simplifications. For example:

    console.info(ce.parse('(x+1)^2').isEqual(ce.parse('x^2+2x+1')));
    // -> true

Asynchronous Operations​

Some computations can be time-consuming, for example, computing a very large factorial. To prevent the browser from freezing, the Compute Engine can now perform some operations asynchronously.

To perform an asynchronous operation, use the expr.evaluateAsync method. For example:

try {
  const fact = ce.parse('(70!)!');
  const factResult = await fact.evaluateAsync();
  factResult.print();
} catch (e) {
  console.error(e);
}

It is also possible to interrupt an operation, for example by providing a pause/cancel button that the user can press. To do so, use an AbortController object and a signal. For example:

const abort = new AbortController();
const signal = abort.signal;
setTimeout(() => abort.abort(), 500);
try {
  const fact = ce.parse('(70!)!');
  const factResult = await fact.evaluateAsync({ signal });
  factResult.print();
} catch (e) {
  console.error(e);
}

In the example above, we trigger an abort after 500ms.

It is also possible to control how long an operation can run by setting the ce.timeLimit property with a value in milliseconds. For example:

ce.timeLimit = 1000;
try {
  const fact = ce.parse('(70!)!');
  fact.evaluate().print();
} catch (e) {
  console.error(e);
}

The time limit applies to either the synchronous or asynchronous evaluation.

The default time limit is 2,000ms (2 seconds).

When an operation is canceled either because of a timeout or an abort, a CancellationError is thrown.

0.27.0 2024-12-02​

  • #217 Correctly parse LaTeX expressions that include a command followed by a * such as \\pi*2.

  • #217 Correctly calculate the angle of trigonometric expressions with an expression containing a reference to Pi, for example \\sin(\\pi^2).

  • The Factorial function will now time out if the argument is too large. The timeout is signaled by throwing a CancellationError.

  • When specifying exp.toMathJSON({shorthands:[]}), i.e., not to use shorthands in the MathJSON, actually avoid using shorthands.

  • Correctly use custom multiply, plus, etc. for LaTeX serialization.

  • When comparing two numeric values, the tolerance is now used to determine if the values are equal. The tolerance can be set with the ce.tolerance property.

  • When comparing two expressions with isEqual() the values are compared structurally when necessary, or with a stochastic test when the expressions are too complex to compare structurally.

  • Correctly serialize nested superscripts, e.g. x^{y^z}.

  • The result of evaluating a Hold expression is now the expression itself.

  • To prevent evaluation of an expression temporarily, use the Unevaluated function. The result of evaluating an Unevaluated expression is its argument.

  • The type of a Hold expression was incorrectly returned as string. It now returns the type of its argument.

  • The statistics function (Mean, Median, Variance, StandardDeviation, Kurtosis, Skewness, Mode, Quartiles and InterQuartileRange) now accept as argument either a collection or a sequence of values.

    ce.parse("\\mathrm{Mean}([7, 2, 11])").evaluate().print();
    // -> 20/3
    ce.parse("\\mathrm{Mean}(7, 2, 11)").evaluate().print();
    // -> 20/3

  • The Variance and StandardDeviation functions now have variants for population statistics, PopulationVariance and PopulationStandardDeviation. The default is to use sample statistics.

    ce.parse("\\mathrm{PopulationVariance}([7, 2, 11])").evaluate().print();
    // -> 13.555
    ce.parse("\\mathrm{Variance}([7, 2, 11])").evaluate().print();
    // -> 20.333

  • The statistics function can now be compiled to JavaScript:

    const code = ce.parse("\\mathrm{Mean}(7, 2, 11)").compile();
    console.log(code());
    // -> 13.555

  • The statistics function calculate either using machine numbers or bignums depending on the precision. The precision can be set with the precision property of the Compute Engine.

  • The argument of compiled function is now optional.

  • Compiled expressions can now reference external JavaScript functions. For example:

    ce.defineFunction('Foo', {
      signature: 'number -> number',
      evaluate: ([x]) => ce.box(['Add', x, 1]),
    });

    const fn = ce.box(['Foo', 3]).compile({
      functions: { Foo: (x) => x + 1 },
    })!;

    console.info(fn());
    // -> 4
    ce.defineFunction('Foo', {
      signature: 'number -> number',
      evaluate: ([x]) => ce.box(['Add', x, 1]),
    });

    function foo(x) {
      return x + 1;
    }

    const fn = ce.box(['Foo', 3]).compile({
      functions: { Foo: foo },
    })!;

    console.info(fn());
    // -> 4

    Additionally, functions can be implicitly imported (in case they are needed by other JavaScript functions):

    ce.defineFunction('Foo', {
      signature: 'number -> number',
      evaluate: ([x]) => ce.box(['Add', x, 1]),
    });

    function bar(x, y) {
      return x + y;
    }

    function foo(x) {
      return bar(x, 1);
    }


    const fn = ce.box(['Foo', 3]).compile({
      functions: { Foo: 'foo' },
      imports: [foo, bar],
    })!;

    console.info(fn());
    // -> 4

  • Compiled expression can now include an arbitrary preamble (JavaScript source) that is executed before the compiled function is executed. This can be used to define additional functions or constants.

    ce.defineFunction('Foo', {
      signature: 'number -> number',
      evaluate: ([x]) => ce.box(['Add', x, 1]),
    });

    const code = ce.box(['Foo', 3]).compile({
      preamble: "function Foo(x) { return x + 1};",
    });

  • The hold function definition flag has been renamed to lazy

0.26.4 2024-10-17​

  • #201 Identifiers of the form A_\text{1} were not parsed correctly.
  • #202 Fixed serialization of integrals and bigops.

0.26.3 2024-10-17​

  • Correctly account for fractionalDigits when formatting numbers.
  • #191 Correctly handle \\lnot\\forall and \\lnot\\exists.
  • #206 The square root of 1000000 was canonicalized to 0.
  • #207 When a square root with a literal base greater than 1e6 was preceded by a non-integer literal number, the literal number was ignored during canonicalization.
  • #208 #204 Correctly evaluate numeric approximation of roots, e.g. \\sqrt[3]{125}.
  • #205 1/ln(0) was incorrectly evaluated to 1. It now returns 0.

0.26.1 2024-10-04​

Issues Resolved​

  • #194 Correctly handle the precedence of unary negate, for example in -5^{\frac12} or -5!.
  • When using a function definition with ce.declare(), do not generate a runtime error.

New Features and Improvements​

  • Added .expand() method to boxed expression. This method expands the expression, for example ce.parse("(x+1)^2").expand() will return x^2 + 2x + 1.

0.26.0 2024-10-01​

Breaking Changes​

  • The property expr.head has been deprecated. Use expr.operator instead. expr.head is still supported in this version but will be removed in a future update.

  • The MathJSON utility functions head() and op() have been renamed to operator() and operand() respectively.

  • The methods for algebraic operations (add, div, mul, etc...) have been moved from the Compute Engine to the Boxed Expression class. Instead of calling ce.add(a, b), call a.add(b).

    Those methods also behave more consistently: they apply some additional simplication rules over canonicalization. For example, while ce.parse('1 + 2') return ["Add", 1, 2], ce.box(1).add(2) will return 3.

  • The ce.numericMode option has been removed. Instead, set the ce.precision property to the desired precision. Set the precision to "machine" for machine precision calculations (about 15 digits). Set it to "auto" for a default of 21 digits. Set it to a number for a greater fixed precision.

  • The MathJSON Dictionary element has been deprecated. Use a Dictionary expression instead.

  • The ExtendedRealNumbers, ExtendedComplexNumbers domains have been deprecated. Use the RealNumbers and ComplexNumbers domains instead.

  • The "Domain" expression has been deprecated. Use types instead (see below).

  • Some BoxedExpression properties have been removed:

    • Instead of expr.isZero, use expr.is(0).
    • Instead of expr.isNotZero, use !expr.is(0).
    • Instead of expr.isOne, use expr.is(1).
    • Instead of expr.isNegativeOne, use expr.is(-1).

  • The signature of ce.declare() has changed. In particular, the N handler has been replaced with evaluate.

// Before
ce.declare('Mean', {
  N: (ce: IComputeEngine): BoxedExpression => {
    return ce.number(1);
  },
});

// Now
ce.declare('Mean', { evaluate: (ops, { engine }) => ce.number(1) });

New Features and Improvements​

  • New Simplification Engine

    The way expressions are simplified has been completely rewritten. The new engine is more powerful and more flexible.

    The core API remains the same: to simplify an expression, use expr.simplify().

    To use a custom set of rules, pass the rules as an argument to simplify():

    expr.simplify({rules: [
      "|x:<0| -> -x",
      "|x:>=0| -> x",
    ]});

    There are a few changes to the way rules are represented. The priority property has been removed. Instead, rules are applied in the order in which they are defined.

    A rule can also now be a function that takes an expression and returns a new expression. For example:

    expr.simplify({rules: [
      (expr) => {
        if (expr.operator !== 'Abs') return undefined;
        const x = expr.args[0];
        return x.isNegative ? x.negate() : expr;
      }
    ]});

    This can be used to perform more complex transformations at the cost of more verbose JavaScript code.

    The algorithm for simplification has been simplified. It attempts to apply each rule in the rule set in turn, then restarts the process until no more rules can be applied or the result of applying a rule returns a previously seen expression.

    Function definitions previously included a simplify handler that could be used to perform simplifications specific to this function. This has been removed. Instead, use a rule that matches the function and returns the simplified expression.

  • Types

    Previously, an expression was associated with a domain such as RealNumbers or ComplexNumbers. This has been replaced with a more flexible system of types.

    A type is a set of values that an expression can take. For example, the type real is the set of real numbers, the type integer is the set of integers,

    The type of an expression can be set with the type property. For example:

    const expr = ce.parse('\\sqrt{-1}');
    console.info(expr.type); // -> imaginary

    The type of a symbol can be set when declaring the symbol. For example:

    ce.declare('x', 'imaginary');

    In addition to primitive types, the type system supports more complex types such union types, intersection types, and function types.

    For example, the type real|imaginary is the union of the real and imaginary numbers.

    When declaring a function, the type of the arguments and the return value can be specified. For example, to declare a function f that takes two integers and returns a real number:

    ce.declare('f', '(integer, integer) -> real');

    The sets of numbers are defined as follows:

    • number - any number, real or complex, including NaN and infinity
    • non_finite_number - NaN or infinity
    • real
    • finite_real - finite real numbers (exclude NaN and infinity)
    • imaginary - imaginary numbers (complex numbers with a real part of 0)
    • finite_imaginary
    • complex - complex numbers with a real and imaginary part not equal to 0
    • finite_complex
    • rational
    • finite_rational
    • integer
    • finite_integer

    To check the type of an expression, use the isSubtypeOf() method. For example:

    let expr = ce.parse('5');
    console.info(expr.type.isSubtypeOf('rational')); // -> true
    console.info(expr.type.isSubtypeOf('integer')); // -> true

    expr = ce.parse('\\frac{1}{2}');
    console.info(expr.type.isSubtypeOf('rational')); // -> true
    console.info(expr.type.isSubtypeOf('integer')); // -> false

    As a shortcut, the properties isReal, isRational, isInteger are available on boxed expressions. For example:

    let expr = ce.parse('5');
    console.info(expr.isInteger); // -> true
    console.info(expr.isRational); // -> true

    They are equivalent to expr.type.isSubtypeOf('integer') and expr.type.isSubtypeOf('rational') respectively.

    To check if a number has a non-zero imaginary part, use:

    let expr = ce.parse('5i');
    console.info(expr.isNumber && expr.isReal === false); // -> true

  • Collections

    Support for collections has been improved. Collections include List, Set, Tuple, Range, Interval, Linspace and Dictionary.

    It is now possible to check if an element is contained in a collection using an Element expression. For example:

    let expr = ce.parse('[1, 2, 3]');
    ce.box(['Element', 3, expr]).print(); // -> True
    ce.box(['Element', 5, expr]).print(); // -> False

    To check if a collection is a subset of another collection, use the Subset expression. For example:

    ce.box(['Subset', 'Integers', 'RealNumbers']).print(); // -> True

    Collections can also be compared for equality. For example:

    let set1 = ce.parse('\\lbrace 1, 2, 3 \\rbrace');
    let set2 = ce.parse('\\lbrace 3, 2, 1 \\rbrace');
    console.info(set1.isEqual(set2)); // -> true

    There are also additional convenience methods on boxed expressions:

    • expr.isCollection
    • expr.contains(element)
    • expr.size
    • expr.isSubsetOf(other)
    • expr.indexOf(element)
    • expr.at(index)
    • expr.each()
    • expr.get(key)

  • Exact calculations

    The Compute Engine has a new backed for numerical calculations. The new backed can handle arbitrary precision calculations, including real and complex numbers. It can also handle exact calculations, preserving calculations with rationals and radicals (square root of integers). For example 1/2 + 1/3 is evaluated to 5/6 instead of 0.8(3).

    To get an approximate result, use the N() method, for example ce.parse("\\frac12 + \\frac13").N().

    Previously the result of calculations was not always an exact number but returned a numerical approximation instead.

    This has now been improved by introducing a NumericValue type that encapsulates exact numbers and by doing all calculations in this type. Previously the calculations were handled manually in the various evaluation functions. This made the code complicated and error prone.

    A NumericValue is made of:

    • an imaginary part, represented as a fixed-precision number
    • a real part, represented either as a fixed or arbitrary precision number or as the product of a rational number and the square root of an integer.

    For example:

    • 234.567
    • 1/2
    • 3√5
    • √7/3
    • 4-3i

    While this is a significant change internally, the external API remains the same. The result of calculations should be more predictable and more accurate.

    One change to the public API is that the expr.numericValue property is now either a machine precision number or a NumericValue object.

  • Rule Wildcards

    When defining a rule as a LaTeX expression, single character identifiers are interpreted as wildcards. For example, the rule x + x -> 2x will match any expression with two identical terms. The wildcard corresponding to x is _x.

    It is now possible to define sequence wildcards and optional sequence wildcards. Sequence wildcards match 1 or more expressions, while optional sequence wildcards match 0 or more expressions.

    They are indicated in LaTeX as ...x and ...x? respectively. For example:

    expr.simplify("x + ...y -> 2x");

    If expr is a + b + c the rule will match and return 2a

    expr.simplify("x + ...y? -> 3x");

    If expr is a + b + c the rule will match and return 3a. If expr is a the rule will match and return 3a.

  • Conditional Rules

    Rules can now include conditions that are evaluated at runtime. If the condition is not satisfied, the rules does not apply.

    For example, to simplify the expression |x|:

    expr.simplify({rules: [
      "|x_{>=0}| -> x",
      "|x_{<0}| -> -x",
    ]});

    The condition is indicated as a subscript of the wildcard. The condition can be one of:

    • boolean - a boolean value, True or False

    • string - a string of characters

    • number - a number literal

    • symbol

    • expression

    • numeric - an expression that has a numeric value, i.e. 2√3, 1/2, 3.14

    • integer - an integer value, -2, -1, 0, 1, 2, 3, ...

    • natural - a natural number, 0, 1, 2, 3, ...

    • real - real numbers, including integers

    • imaginary - imaginary numbers, i.e. 2i, 3√-1 (not including real numbers)

    • complex - complex numbers, including real and imaginary

    • rational - rational numbers, 1/2, 3/4, 5/6, ...

    • irrational - irrational numbers, √2, √3, π, ...

    • algebraic - algebraic numbers, rational and irrational

    • transcendental - transcendental numbers, π, e, ...

    • positive - positive real numbers, > 0

    • negative - negative real numbers, < 0

    • nonnegative - nonnegative real numbers, >= 0

    • nonpositive - nonpositive real numbers, <= 0

    • even - even integers, 0, 2, 4, 6, ...

    • odd - odd integers, 1, 3, 5, 7, ...

    • prime :A000040 - prime numbers, 2, 3, 5, 7, 11, ...

    • composite :A002808 - composite numbers, 4, 6, 8, 9, 10, ...

    • notzero - a value that is not zero

    • notone - a value that is not one

    • finite - a finite value, not infinite

    • infinite

    • constant

    • variable

    • function

    • operator

    • relation - an equation or inequality

    • equation

    • inequality

    • vector - a tensor of rank 1

    • matrix - a tensor of rank 2

    • list - a collection of values

    • set - a collection of unique values

    • tuple - a fixed length list

    • single - a tuple of length 1

    • pair - a tuple of length 2

    • triple - a tuple of length 3

    • collection - a list, set, or tuple

    • tensor - a nested list of values of the same type

    • scalar - not a tensor or list

    or one of the following expressions:

    • >0' -> positive,
    • \gt0' -> positive,
    • <0' -> negative,
    • \lt0' -> negative,
    • >=0' -> nonnegative,
    • \geq0' -> nonnegative,
    • <=0' -> nonpositive,
    • \leq0' -> nonpositive,
    • !=0' -> notzero,
    • \neq0' -> notzero,
    • !=1' -> notone,
    • \neq1' -> notone,
    • \in\Z' -> integer,
    • \in\mathbb{Z}' -> integer,
    • \in\N' -> natural,
    • \in\mathbb{N}' -> natural,
    • \in\R' -> real,
    • \in\mathbb{R}' -> real,
    • \in\C' -> complex,
    • \in\mathbb{C}' -> complex,
    • \in\Q' -> rational,
    • \in\mathbb{Q}' -> rational,
    • \in\Z^+' -> integer,positive,
    • \in\Z^-' -> intger,negative,
    • \in\Z^*' -> nonzero,
    • \in\R^+' -> positive,
    • \in\R^-' -> negative,
    • \in\R^*' -> real,nonzero,
    • \in\N^*' -> integer,positive,
    • \in\N_0' -> integer,nonnegative,
    • \in\R\backslash\Q' -> irrational,

    More complex conditions can be specified following a semi-colon, for example:

    expr.simplify({x -> 2x; x < 10});

    Note that this syntax complements the existing rule syntax, and can be used together with the existing, more verbose, rule syntax.

    expr.simplify({rules: [
      {match: "x + x", replace: "2x", condition: "x < 10"}
    ]});

    This advanced syntax can specify more complex conditions, for example above the rule will only apply if x is less than 10.

  • Improved results for Expand. In some cases the expression was not fully expanded. For example, 4x(3x+2)-5(5x-4) now returns 12x^2 - 17x + 20. Previously it returned 4x(3x+2)+25x-20.

  • AsciiMath serialization The expr.toString() method now returns a serialization of the expression using the AsciiMath format.

    The serialization to AsciiMath can be customized using the toAsciiMath() method. For example:

    console.log(ce.box(['Sigma', 2]).toAsciiMath({functions: {Sigma: 'sigma'}}));
    // -> sigma(2)

  • The tolerance can now be specified with a value of "auto" which will use the precision to determine a reasonable tolerance. The tolerance is used when comparing two numbers for equality. The tolerance can be specified with the ce.tolerance property or in the Compute Engine constructor.

  • Boxed expressions have some additional properties:

    • expr.isNumberLiteral - true if the expression is a number literal.This is equivalent to checking if expr.numericValue is not null.
    • expr.re - the real part of the expression, if it is a number literal, undefined if not a number literal.
    • expr.im - the imaginary part of the expression, if it is a number literal, undefined if not a number literal.
    • expr.bignumRe - the real part of the expression as a bignum, if it is a number literal, undefined if not a number literal or a bignum representation is not available.
    • expr.bignumIm - the imaginary part of the expression as a bignum, if it is a number literal, undefined if not a number literal or if a bignum representation is not available.
    • expr.root() to get the root of the expression. For example, expr.root(3) will return the cube root of the expression.
    • Additionally, the relational operators (expr.isLess(), expr.isEqual(), etc...) now accept a number argument. For example, expr.isGreater(1) will return true if the expression is greater than 1.

  • Added LaTeX syntax to index collections. If a is a collection:

    • a[i] is parsed as ["At", "a", "i"].
    • a[i,j] is parsed as ["At", "a", "i", "j"].
    • a_i is parsed as ["At", "a", "i"].
    • a_{i,j} is parsed as ["At", "a", "i", "j"].

  • Added support for Kronecker delta notation, i.e. \delta_{ij}, which is parsed as ["KroneckerDelta", "i", "j"] and is equal to 1 if i = j and 0 otherwise.

    When a single index is provided the value of the function is 1 if the index is 0 and 0 otherwise

    When multiple index are provided, the value of the function is 1 if all the indexes are equal and 0 otherwise.

  • Added support for Iverson Bracket notation, i.e. [a = b], which is parsed as ["Boole", ["Equal", "a", "b"]] and is equal to 1 if its argument is true and 0 otherwise. The argument is expected to be a relational expression.

  • Implemented Unique and Tally on collections. Unique returns a collection with only the unique elements of the input collection, and Tally returns a collection with the count of each unique element.

    console.log(ce.box(['Unique', ['List', 1, 2, 3, 1, 2, 3, 4, 5]]).value);
    // -> [1, 2, 3, 4, 5]

    console.log(ce.box(['Tally', ['List', 1, 2, 3, 1, 2, 3, 4, 5]]).value);
    // -> [['List', 1, 2, 3, 4, 5], ['List', 2, 2, 2, 1, 1]]

  • Implemented the Map, Filter and Tabulate functions. These functions can be used to transform collections, for example:

    // Using LaTeX
    console.log(ce.parse('\\mathrm{Map}([3, 5, 7], x \\mapsto x^2)').toString());
    // -> [9, 25, 49]

    // Using boxed expressions
    console.log(
      ce.box(['Map', ['List', 3, 5, 7], ['Square', '_']]).value
    );
    // -> [9, 25, 49]

    console.log(ce.box(['Tabulate',['Square', '_'], 5]).value);
    // -> [1, 4, 9, 16, 25]

    Tabulate can be used with multiple indexes. For example, to generate a 4x4 unit matrix:

    console.log(ce.box(['Tabulate', ['If', ['Equal', '_1', '_2'], 1, 0]], 4, 4).value);
    // -> [[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]]

    // Using the Kronecker delta notation:
    console.log(ce.parse('\\mathrm{Tabulate}(i, j \\mapsto \\delta_{ij}, 4, 4)').value);
    // -> [[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]]

  • Added Random function. ["Random"] returns a real pseudo-random number betwen 0 and 1. ["Random", 10] returns an integer between 0 and 9, ["Random", 5, 10] returns an integer between 5 and 10.

  • Extended the definition of expr.isConstant. Previously, it only applied to symbols, e.g. Pi. Now it apply to all expressions. expr.isConstant is true if the expression is a number literal, a symbol with a constant value, or a pure function with constant arguments.

  • The boxed expression properties isPositive, isNegative, isNonNegative, isNonPositive, isZero, isNotZero now return a useful value for most function expressions. For example, ce.parse('|x + 1|').isPositive is true.

    If the value cannot be determined, the property will return undefined. For example, ce.parse('|x + 1|').isZero is undefined.

    If the expression is not a real number, the property will return NaN. For example, ce.parse('i').isPositive is NaN.

  • Added Choose function to compute binomial coefficients, i.e. Choose(5, 2) is equal to 10.

  • The fallback for non-constructible complex values of trigonometric functions is now implemented via rules.

  • The canonical order of the arguments has changed and should be more consistent and predictable. In particular, for polynomials, the monomial order is now degrevlex.

  • Canonical expressions can now include a Root expression. For example, the canonical form of \\sqrt[3]{5} is ["Root", 5, 3]. Previously, these were represented as ["Power", 5, ["Divide", 1, 3]].

  • The function definitions no longer have a N handler. Instead the evaluate handler has an optional {numericApproximation} argument.

Issues Resolved​

  • #188 Throw an error when invalid expressions are boxed, for example ce.box(["Add", ["3"]]).

  • Some LaTeX renderer can't render \/, so use / instead.

  • When definitions are added to the LaTeX dictionary, they now take precedence over the built-in definitions. This allows users to override the built-in definitions.

  • Improved parsing of functions, including when a mixture of named and positional arguments are used.

  • #175 Matching some patterns when the target had not enough operands would result in a runtime error.

0.25.1 2024-06-27​

Issues Resolved​

  • #174 Fixed some simplifications, such as \frac{a^n}{a^m} = a^{n-m)

New Features​

  • Rules can be defined using a new shorthand syntax, where each rule is a string of LaTeX:

    expr.simplify(["\\frac{x}{x} -> 1", "x + x -> 2x"]);

Single letter variables are assumed to be wildcards, so x is interpreted as the wildcard _x.

Additionally, the expanded form can also include LaTeX strings. The previous syntax using expressions can still be used, and the new and old syntax can be mixed.

For example:

expr.simplify([
  {
    match: "\\frac{x}{x}",
    replace: "1"
  },
  {
    match: ["Add", "x", "x"],
    replace: "2x"
  }
]);

The condition function can also be expressed as a LaTeX string.

  expr.simplify([ { match: "\\frac{x}{x}", replace: 1, condition: "x != 0" }, ]);

The shorthand syntax can be used any where a ruleset is expected, including with the ce.rule() function.

  • A new ce.getRuleSet() method gives access to the built-in rules.
  • #171 The Subtract and Divide function can now accept an arbitrary number of arguments. For example, ["Subtract", 1, 2, 3] is equivalent to ["Subtract", ["Subtract", 1, 2], 3].

0.25.0 2024-06-25​

Breaking Changes​

  • The canonical form of expressions has changed. It is now more consistent and simpler and should produce more predictable results.

    For example, previously ce.parse("1-x^2") would produce ["Subtract", 1, ["Square", "x"]].

    While this is a readable form, it introduces some complications when manipulating the expression: both the Subtract and Square functions have to be handled, in addition to Add and Power.

    The new canonical form of this expression is ["Add", 1, ["Negate", ["Power", "x", 2]]]. It is a bit more verbose, but it is simpler to manipulate.

  • The ce.serialize() method has been replaced with expr.toLatex() and expr.toMathJson(). The ce.latexOptions and ce.jsonSerializationOptions properties have been removed. Instead, pass the formating options directly to the toLatex() and toMathJson() methods. The ce.parse() method now takes an optional argument to specify the format of the input string.

  • The default JSON serialization of an expression has changed.

    Previously, the default JSON serialization, accessed via the .json property, had some transformations applied to it (sugaring) to make the JSON more human readable.

    For example, ce.parse("\frac12").json would return the symbol "Half" instead of ["Divide", 1, 2].

    However, this could lead to some confusion when manipulating the JSON directly. Since the JSON is intended to be used by machine more than humans, these additional transformations have been removed.

    The expr.json property now returns the JSON representing the expression, without any transformations.

    To get a version of JSON with some transformations applied use the ce.toMathJson() function.

    expr = ce.box(["Subtract", 1, ["Square", "x"]]);
    console.log(expr.json);
    // -> ["Add", 1, ["Negate", ["Power", "x", 2]]]
    expr.toMathJson()
    // -> ["Subtract", 1, ["Square", "x"]]
    expr.toMathJson({exclude: "Square"})
    // -> ["Subtract", 1, ["Power", "x", 2]]

    In practice, the impact of both of these changes should be minimal. If you were manipulating expressions using BoxedExpression, the new canonical form should make it easier to manipulate expressions. You can potentially simplify your code by removing special cases for functions such as Square and Subtract.

    If you were using the JSON serialization directly, you may also be able to simplify you code since the default output from expr.json is now more consistent and simpler.

  • The name of some number formatting options has changed. The number formatting options are an optional argument of ce.parse() and ce.toLatex(). See the NumberFormat and NumberSerializationFormat types.

  • The values +infinity, -infinity and NaN are now represented preferably with the symbols PositiveInfinity, NegativeInfinity and NaN respectively. Previously they were represented with numeric values, i.e. {num: "+Infinity"}, {num: "-Infinity"} and {num: "NaN"}. The numeric values are still supported, but the symbols are preferred.

  • The method expr.isNothing has been removed. Instead, use expr.symbol === "Nothing".

New Features​

  • When serializing to LaTeX, the output can be "prettified". This involves modifying the LaTeX output to make it more pleasant to read, for example:

    • a+\\frac{-b}{c} -> a-\\frac{b}{c}
    • a\\times b^{-1} -> \\frac{a}{b}
    • \\frac{a}{b}\\frac{c}{d} -> \\frac{a\\cdot c}{b\\cdot d}
    • --2 -> 2

    This is on by default and can be turned off by setting the prettify option to false. For example:

    ce.parse("a+\\frac{-b}{c}").toLatex({prettify: true})
    // -> "a-\\frac{b}{c}"
    ce.parse("a+\\frac{-b}{c}").toLatex({prettify: false})
    // -> "a+\\frac{-b}{c}"

  • Numbers can have a different digit group length for the whole and fractional part of a number. For example, ce.toLatex(ce.parse("1234.5678"), {digitGroup: [3, 0]}) will return 1\,234.5678.

  • Numbers can now be formatted using South-East Asian Numbering System, i.e. lakh and crore. For example:

    ce.toLatex(ce.parse("12345678"), {digitGroup: "lakh"})
    // -> "1,23,45,678"

  • Expressions with Integrate functions can now be compiled to JavaScript. The compiled function can be used to evaluate the integral numerically. For example:

    const f = ce.parse("\\int_0^1 x^2 dx");
    const compiled = f.compile();
    console.log(compiled()); // -> 0.33232945619482307

  • #82 Support for angular units. The default is radians, but degrees can be used by setting ce.angularUnit = "deg". Other possible values are "grad" and "turn". This affects how unitless numbers with a trigonometric function are interpreted. For example, sin(90) will return 1 when ce.angularUnit is "deg", 0.8939966636005579 when ce.angularUnit is "grad" and 0 when ce.angularUnit is "turn".

  • Added expr.map(fn) method to apply a function to each subexpression of an expression. This can be useful to apply custom canonical forms and compare two expressions.

  • An optional canonical form can now be specified with the ce.function().

Issues Resolved​

  • #173 Parsing 1++2 would result in an expression with a PreIncrement function. It is now correctly parsed as ["Add", 1, 2].
  • #161 Power expressions would not be processed when their argument was a Divide expression.
  • #165 More aggressive simplification of expressions with exponent greater than 3.
  • #169 Calculating a constant integral (and integral that did not depend on the variable) would result in a runtime error.
  • #164 Negative mixed fractions (e.g. -1\frac23) are now parsed correctly.
  • #162 Numeric evaluation of expressions with large exponents could result in machine precision numbers instead of bignum numbers.
  • #155 The expression ["Subtract", ["Multiply", 0.5, "x"], ["Divide", "x", 2]] will now evaluate to 0.
  • #154 In some cases, parsing implicit argument of trig function return more natural results, for example \cos a \sin b is now parsed as (\cos a)(\sin b) and not \cos (a \sin b).
  • #147 The associativity of some operators, including / was not applied correctly, resulting in unexpected results. For example, 1/2/3 would be parsed as ["Divide", 1, ["Divide", 2, 3]] instead of ["Divide", ["Divide", 1, 2], 3].
  • #146 When parsing an expression like x(x+1) where x is an undeclared symbol, do not infer that x is a function. Instead, infer that x is a variable and that the expression is a product.
  • #145 The expression ["Or", "False", "False"], that is when all the arguments are False, is now evaluates to False.
  • Fixed canonical form of e^x^2, and more generally apply power rule in more cases.
  • Added missing "Sech" and "Csch" functions.
  • The digit grouping serializing would place the separator in the wrong place for some numbers.
  • The avoidExponentsInRange formating option would not always avoid exponents in the specified range.

0.24.0 2024-02-23​

Issues Resolved​

  • Fix parsing of very deeply nested expressions.
  • Correctly apply rules to deeply nested expressions.
  • expr.print() now correctly prints the expression when using the minified version of the library.
  • expr.isEqual() now correctly compares equalities and inequalities.
  • expr.match() has been improved and works correctly in more cases. The signature of the match function has been changed so that the pattern is the first argument, i.e. instead of pattern.match(expr) use expr.match(pattern).
  • Fix expr.print() when using the minified version of the library.
  • #142 Accept complex expressions as the subcript of \ln and \log in LaTeX.
  • #139 Parse quantifiers \forall and \exists in LaTeX.

0.23.1 2024-01-27​

Issues Resolved​

  • Using a custom canonical order of "Multiply" would not distribute the Negate function.
  • #141 The canonical form "Order" was applied to non-commutative functions.

0.23.0 2024-01-01​

New Features​

  • Added ExpandAll function to expand an expression recursively.
  • Added Factor function to factor an expression.
  • Added Together function to combine rational expressions into a single fraction.

Issues Resolved​

  • The expression \frac5 7 is now parsed correctly as \frac{5}{7} instead of \frac{5}{}7.
  • Do not sugar non-canonical expression. Previously, ce.parse('\\frac{1}{2}', {canonical: false}) would return Half instead of ['Divide', '1', '2'].
  • #132 Attempting to set a value to 0 with ce.defineSymbol("count", {value: 0}) would fail: the symbol would be undefined.
  • Correctly evaluate power expressions in some cases, for example (\sqrt2 + \sqrt2)^2.
  • Comparison of expressions containing non-exact numbers could fail. For example: 2(13.1+3.1x) and 26.2+6.2x would not be considered equal.

Improvements​

  • Significant improvements to symbolic computation. Now, boxing, canonicalization and evaluation are more consistent and produce more predictable results.
  • Adedd the \neg command, synonym for \lnot -> Not.
  • Relational expressions (inequalities, etc...) are now properly factored.
  • Integers are now factored when simplifying, i.e. 2x = 4x -> x = 2x.

0.22.0 2023-11-13​

Breaking Changes​

  • Rule Syntax

    The syntax to describe rules has changed. The syntax for a rule was previously a tuple [lhs, rhs, {condition} ]. The new syntax is an object with the properties match, replace and condition. For example:

    • previous syntax: [["Add", "_x", "_x"], ["Multiply", 2, "_x"]]
    • new syntax: {match: ["Add", "_x", "_x"], replace: ["Multiply", 2, "_x"]}

    The condition property is optional, and is either a boxed function or a JavaScript function. For example, to add a condition that checks that _x is a number literal:

    {
      match: ["Add", "_x", "_x"],
      replace: ["Multiply", 2, "_x"],
      condition: ({_x}) => _x.isNumberLiteral
    }

  • CanonicalForm

    The CanonicalOrder function has been replaced by the more flexible CanonicalForm function. The CanonicalForm function takes an expression and a list of transformations to apply. To apply the same transformations as CanonicalOrder, use:

    ['CanonicalForm', expr, 'Order']

    These canonical forms can also be specified with box() and parse() options:

    ce.box(expr, { canonical: "Order" });
    ce.parse("x^2 + 2x + 1", { canonical: "Order" });

Work In Progress​

  • Linear algebra functions: Rank, Shape,Reshape, Flatten, Determinant, Trace, Transpose, ConjugateTranspose, Inverse. See the Linear Algebra reference guide. Some of these function may not yet return correct result in all cases.

New Features​

  • Added a expr.print() method as a synonym for console.log(expr.toString()).
  • Added an exact option (false by default) to the expr.match() pattern matching method. When true some additional patterns are automatically recognized, for example, x will match ["Multiply", '_a', 'x'] when exact is false, but not when exact is true.

Improvements​

  • The equation solver used by expr.solve() has been improved and can now solve more equations.
  • The pattern matching engine has been improved and can now match more expressions, including sequences for commutative functions.

0.21.0 2023-11-02​

New Features​

  • #125 Parse and serialize environemnts, i.e. \begin{matrix} 1 & 2 \\ 3 & 4 \end{matrix} will be parsed as ["Matrix", ["List", ["List", 1, 2], ["List", 3, 4]]].

    A new section on Linear Algebra has some details on the supported formats.

    The linear algebra operations are limited at the moment, but will be expanded in the future.

  • Added IsSame function, which is the function expression corresponding to expr.isSame().

  • Added CanonicalOrder function, which sorts the arguments of commutative functions into canonical order. This is useful to compare two non-canonical expressions for equality. 
ce.box(["CanonicalOrder", ["Add", 1, "x"]]).isSame(
  ce.box(["CanonicalOrder", ["Add", "x", 1]])
);
// -> true

Issue Resolved​

  • When evaluating a sum (\sum) with a bound that is not a number, return the sum expression instead of an error.

0.20.2 2023-10-31​

Issues Resolved​

  • Fixed numerical evaluation of integrals and limits when parsed from LaTeX.
console.info(ce.parse("\\lim_{x \\to 0} \\frac{\\sin(x)}{x}").value);
// -> 1

console.info(ce.parse("\\int_{0}^{2} x^2 dx").value);
// -> 2.6666666666666665

0.20.1 2023-10-31​

Issues Resolved​

  • Fixed evaluation of functions with multiple arguments
  • Fixed compilation of some function assignments
  • Improved serialization of function assignment

0.20.0 2023-10-30​

Breaking Changes​

  • Architectural changes: the invisible operator is used to represent the multiplication of two adjacent symbols, i.e. 2x. It was previously handled during parsing, but it is now handled during canonicalization. This allows more complex syntactic structures to be handled correctly, for example f(x) := 2x: previously, the left-hand-side argument would have been parsed as a function application, while in this case it should be interpreted as a function definition.

    A new InvisibleOperator function has been added to support this.

    The applyInvisibleOperator parsing option has been removed. To support custom invisible operators, use the InvisibleOperator function.

Issues Resolved​

  • #25 Correctly parse chained relational operators, i.e. a < b <= c
  • #126 Logic operators only accepted up to two arguments.
  • #127 Correctly compile Log with bases other than 10.
  • Correctly parse numbers with repeating patterns but no fractional digits, i.e. 0.(1234)
  • Correctly parse |1+|a|+2|

New Features and Improvements​

  • Function assignment can now be done with this syntax: f(x) := 2x+1. This syntax is equivalent to f := x -> 2x+1.
  • Implement the Mod and Congruent function.
  • Correctly parse 11 \bmod 5 (Mod) and 26\equiv 11 \pmod5 (Congruent)
  • Better handle empty argument lists, i.e. f()
  • When a function is used before being declared, infer that the symbol is a function, e.g. f(12) will infer that f is a function (and not a variable f multiplied by 12)
  • When a constant is followed by some parentheses, don't assume this is a function application, e.g. \pi(3+n) is now parsed as ["Multiply", "Pi", ["Add", 3, "n"]] instead of ["Pi", ["Add", 3, "n"]]
  • Improved parsing of nested lists, sequences and sets.
  • Improved error messages when syntax errors are encountered during LaTeX parsing.
  • When parsing with the canonical option set to false, preserve more closely the original LaTeX syntax.
  • When parsing text strings, convert some LaTeX commands to Unicode, including spacing commands. As a result, ce.parse("\\text{dead\;beef}_{16}") correctly gets evaluated to 3,735,928,559.

0.19.1 2023-10-26​

Issues Resolved​

  • Assigning a function to an indentifier works correctly now, i.e.
ce.parse("\\operatorname{f} := x \\mapsto 2x").evaluate();

0.19.0 2023-10-25​

Breaking Changes​

  • The domain property of the function definition signature is deprecated and replaced with the params, optParams, restParam and result properties instead. The domain property is still supported for backward compatibility, but will be removed in a future version.

Issues Resolved​

  • When invoking a declared function in a numeric operation, correctly infer the result type.
["Assign", "f", ["Add", "_", 1]]
["Add", ["f", 1], 1]
// -> 3

Previously a domain error was returned, now f is inferred to have a numeric return type.

  • Fixed a runtime error when inverting a fraction, i.e. \frac{3}{4}^{-1}
  • The tangent of π/2 now correctly returns ComplexInfinity.
  • The exact values of some constructible trigonometric operations (e.g. \tan 18\degree = \frac{\sqrt{25-10\sqrt5}}{5}) returned incorrect results. The unit test case was incorrect and did not detect the problem. The unit test case has been fixed and the returned values are now correct.

New Features​

  • Implemented Union and Intersection of collections, for example:
["Intersection", ["List", 3, 5, 7], ["List", 2, 5, 9]]
// -> ["Set", 5]

["Union", ["List", 3, 5, 7], ["List", 2, 5, 9]]
// -> ["Set", 3, 5, 7, 2, 9]

  • Parse ranges, for example 1..5 or 1, 3..10. Ranges are collections and can be used anywhere collections can be used.

  • The functions Sum, Product, Min, Max, and the statistics functions (Mean, Median, Variance, etc...) now handle collection arguments: collections:

    • ["Range"], ["Interval"], ["Linspace"] expressions
    • ["List"] or ["Set"] expressions
    • ["Tuple"], ["Pair"], ["Pair"], ["Triple"] expressions
    • ["Sequence"] expressions

  • Most mathematical functions are now threadable, that is their arguments can be collections, for example:

["Sin", ["List", 0, 1, 5]]
// -> ["List", 0, 0.8414709848078965, -0.9589242746631385]

["Add", ["List", 1, 2], ["List", 3, 4]]
// -> ["List", 4, 6]
  • Added GCD and LCM functions
["GCD", 10, 5, 15]
// -> 5

["LCM", 10, 5, 15]
// -> 30

  • Added Numerator, Denominator, NumeratorDenominator functions. These functions can be used on non-canonical expressions.

  • Added Head and Tail functions which can be used on non-canonical expressions.

  • Added display-quotient and inline-quotient style for formatting of division expressions in LaTeX.

Improvements​

  • Improved parsing of \degree command
ce.parse("30\\degree)
// -> ["Divide", "Pi", 6]
  • Improved interoperability with JavaScript: expr.value will return a JavaScript primitive (number, boolean, string, etc...) when possible. This is a more succinct version of expr.N().valueOf().

0.18.1 2023-10-16​

Issues Resolved​

  • Parsing of whole numbers while in rational mode would return incorrect results.
  • The ND function to evaluate derivatives numerically now return correct values.
ce.parse("\\mathrm{ND}(x \\mapsto 3x^2+5x+7, 2)").N();
// -> 17.000000000001

Improvements​

  • Speed up NIntegrate by temporarily switching the numeric mode to machine while computing the Monte Carlo approximation.

0.18.0 2023-10-16​

New Features​

  • Expanded LaTeX dictionary with \max, \min, \sup, \inf and \lim functions
  • Added Supremum and Infimum functions
  • Compilation of Block expressions, local variables, return statements and conditionals If.
  • Added numerical evaluation of limits with Limit functions and NLimit functions, using a Richardson Extrapolation.
console.info(ce.parse("\\lim_{x\\to0} \\frac{\\sin x}{x}").N().json);
// -> 1

console.info(
  ce.box(["NLimit", ["Divide", ["Sin", "_"], "_"], 0]).evaluate().json
);
// -> 1

console.info(ce.parse("\\lim_{x\\to \\infty} \\cos \\frac{1}{x}").N().json);
// -> 1

  • Added Assign and Declare functions to assign values to symbols and declare symbols with a domain.

  • Block evaluations with local variables work now. For example:

ce.box(["Block", ["Assign", "c", 5], ["Multiply", "c", 2]]).evaluate().json;
// -> 10

  • When decimal numbers are parsed they are interpreted as inexact numbers by default, i.e. "1.2" -> {num: "1.2"}. To force the number to be interpreted as a rational number, set ce.latexOptions.parseNumbers = "rational". In that case, "1.2" -> ["Rational", 12, 10], an exact number.

    While regular decimals are considered "inexact" numbers (i.e. they are assumed to be an approximation), rationals are assumed to be exact. In most cases, the safest thing to do is to consider decimal numbers as inexact to avoid introducing errors in calculations. If you know that the decimal numbers you parse are exact, you can use this option to consider them as exact numbers.

Improvements​

  • LaTeX parser: empty superscripts are now ignored, e.g. 4^{} is interpreted as 4.

0.17.0 2023-10-12​

Breaking Changes​

  • The Nothing domain has been renamed to NothingDomain
  • The Functions, Maybe, Sequence, Dictionary, List and Tuple domain constructors have been renamed to FunctionOf, OptArg, VarArg, DictionaryOf, ListOf and TupleOf, respectively.
  • Domains no longer require a ["Domain"] expression wrapper, so for example ce.box("Pi").domain returns "TranscendentalNumbers" instead of ["Domain", "TranscendentalNumbers"].
  • The VarArg domain constructor now indicates the presence of 0 or more arguments, instead of 1 or more arguments.
  • The MaybeBooleans domain has been dropped. Use ["Union", "Booleans", "NothingDomain"] instead.
  • The ce.defaultDomain has been dropped. The domain of a symbol is now determined by the context in which it is used, or by the ce.assume() method. In some circumstances, the domain of a symbol can be undefined.

New Features​

  • Symbolic derivatives of expressions can be calculated using the D function. For example, ce.box(["D", ce.parse("x^2 + 3x + 1"), "x"]).evaluate().latex returns "2x + 3".

Improvements​

  • Some frequently used expressions are now available as predefined constants, for example ce.Pi, ce.True and ce.Numbers.
  • Improved type checking and inference, especially for functions with complicated or non-numeric signatures.

Bugs Fixed​

  • Invoking a function repeatedly would invoke the function in the original scope rather than using a new scope for each invocation.

0.16.0 2023-09-29​

Breaking Changes​

  • The methods ce.let() and ce.set() have been renamed to ce.declare() and ce.assign() respectively.
  • The method ce.assume() requires a predicate.
  • The signatures of ce.assume() and ce.ask() have been simplified.
  • The signature of ce.pushScope() has been simplified.
  • The expr.freeVars property has been renamed to expr.unknowns. It returns the identifiers used in the expression that do not have a value associated with them. The expr.freeVariables property now return the identifiers used in the expression that are defined outside of the local scope and are not arguments of the function, if a function.

New Features​

  • Domain Inference when the domain of a symbol is not set explicitly (for example with ce.declare()), the domain is inferred from the value of the symbol or from the context of its usage.

  • Added Assume, Identity, Which, Parse, N, Evaluate, Simplify, Domain.

  • Assignments in LaTeX: x \\coloneq 42 produce ["Assign", "x", 42]

  • Added ErfInv (inverse error function)

  • Added Factorial2 (double factorial)

Functions​

  • Functions can now be defined:

    • using ce.assign() or ce.declare()
    • evaluating LaTeX: (x, y) \mapsto x^2 + y^2
    • evaluating MathJSON: ["Function", ["Add", ["Power", "x", 2], ["Power", "y", 2]]], "x", "y"]

  • Function can be applied using \operatorname{apply} or the operators \rhd and \lhd:

    • \operatorname{apply}(f, x)
    • f \rhd x
    • x \lhd f

See Adding New Definitions and Functions.

Control Structures​

  • Added FixedPoint, Block, If, Loop
  • Added Break, Continue and Return statements

See Control Structures

Calculus​

  • Added numeric approximation of derivatives, using an 8-th order centered difference approximation, with the ND function.
  • Added numeric approximation of integrals, using a Monte Carlo method with rebasing for improper integrals, with the NIntegrate function
  • Added symbolic calculation of derivatives with the D function.

Collections​

Added support for collections such as lists, tuples, ranges, etc...

See Collections

Collections can be used to represent various data structures, such as lists, vectors, matrixes and more.

They can be iterated, sliced, filtered, mapped, etc...

["Length", ["List", 19, 23, 5]]
// -> 3

["IsEmpty", ["Range", 1, 10]]
// -> "False"

["Take", ["Linspace", 0, 100, 50], 4]
// -> ["List", 0, 2, 4, 6]

["Map", ["List", 1, 2, 3], ["Function", "x", ["Power", "x", 2]]]
// -> ["List", 1, 4, 9]

["Exclude", ["List", 33, 45, 12, 89, 65], -2, 2]
// -> ["List", 33, 12, 65]


["First", ["List", 33, 45, 12, 89, 65]]
// -> 33

Improvements​

  • The documentation has been significantly rewritten with help from an AI-powered writing assistant.

Issues Resolved​

  • The LaTeX string returned in ["Error"] expression was incorrectly tagged as Latex instead of LatexString.

0.15.0 2023-09-14​

Improvements​

  • The ce.serialize() function now takes an optional canonical argument. Set it to false to prevent some transformations that are done to produce more readable LaTeX, but that may not match exactly the MathJSON. For example, by default ce.serialize(["Power", "x", -1]) returns \frac{1}{x} while ce.serialize(["Power", "x", -1], {canonical: false}) returns x^{-1}.
  • Improved parsing of delimiters, i.e. \left(, \right], etc...
  • Added complex functions Real, Imaginary, Arg, Conjugate, AbsArg. See Complex
  • Added parsing and evaluation of \Re, \Im, \arg, ^\star (Conjugate).
  • #104 Added the ["ComplexRoots", x, n] function which returns the nthroot of x.
  • Added parsing and evaluation of statistics functions Mean, Median, StandardDeviation, Variance, Skewness, Kurtosis, Quantile, Quartiles, InterquartileRange, Mode, Count, Erf, Erfc. See Statistics

0.14.0 2023-09-13​

Breaking Changes​

  • The entries in the LaTeX syntax dictionary can now have LaTeX triggers (latexTrigger) or triggers based on identifiers (symbolTrigger). The former replaces the trigger property. The latter is new. An entry with a triggerIdentifier of average will match \operatorname{average}, \mathrm{average} and other variants.
  • The ce.latexOptions and ce.jsonSerializationOptions properties are more robust. They can be modified directly or one of their properties can be modified.

Improvements​

  • Added more functions and symbols supported by expr.compile():

    • Factorial postfix operator 5!
    • Gamma function \Gamma(2)
    • LogGamma function \operatorname{LogGamma}(2)
    • Gcd function \operatorname{gcd}(20, 5)
    • Lcm function \operatorname{lcm}(20, 5)
    • Chop function \operatorname{chop}(0.00000000001)
    • Half constant \frac{1}{2}
    • 'MachineEpsilon' constant
    • GoldenRatio constant
    • CatalanConstant constant
    • EulerGamma constant \gamma
    • Max function \operatorname{max}(1, 2, 3)
    • Min function \operatorname{min}(13, 5, 7)
    • Relational operators: Less, Greater, LessEqual, GreaterEqual, 'Equal', 'NotEqual'
    • Some logical operators and constants: And, Or, Not, True, False

  • More complex identifiers syntax are recognized, including \mathbin{}, \mathord{}, etc... \operatorname{} is the recommended syntax, though: it will display the identifier in upright font and with the propert spacing, and is properly enclosing. Some commands, such as \mathrm{} are not properly enclosing: two adjacent \mathrm{} command could be merged into one.

  • Environments are now parsed and serialized correctly.

  • When parsing LaTeX, function application is properly handled in more cases, including custom functions, e.g. f(x)

  • When parsing LaTeX, multiple arguments are properly handled, e.g. f(x, y)

  • Add LaTeX syntax for logical operators:

    • And: \land, \operatorname{and} (infix or function)
    • Or: \lor, \operatorname{or} (infix or function)
    • Not: \lnot, \operatorname{not} (prefix or function)
    • Xor: \veebar (infix)
    • Nand: \barwedge (infix)
    • Nor: ^^^^22BD (infix)
    • Implies: \implies (infix)
    • Equivalent: \iff (infix)

  • When a postfix operator is defined in the LaTeX syntax dictionary of the form ^ plus a single token, a definition with braces is added automatically so that both forms will be recognized.

  • Extended the LaTeX dictionary with:

    • floor
    • ceil
    • round
    • sgn
    • exp
    • abs
    • gcd
    • lcm
    • apply

  • Properly handle inverse and derivate notations, e.g. \sin^{-1}(x), \sin'(x), \cos''(x), \cos^{(4)}(x) or even \sin^{-1}''(x)

0.13.0 2023-09-09​

New Features​

  • Compilation Some expressions can be compiled to Javascript. This is useful to evaluate an expression many times, for example in a loop. The compiled expression is faster to evaluate than the original expression. To get the compiled expression, use expr.compile(). Read more at Compiling

Issues Resolved and Improvements​

  • Fixed parsing and serialization of extended LaTeX synonyms for e and i.
  • Fixed serialization of Half.
  • Fixed serialization of Which
  • Improved serialization of ["Delimiter"] expressions.

0.12.7 2023-09-08​

Improvements​

  • Made customization of the LaTeX dictionary simpler. The ce.latexDictionary property can be used to access and modify the dictionary. The documentation has been updated.

0.12.6 2023-09-08​

Breaking Changes​

  • New API for the Parser class.

Improvements and Bux Fixes​

  • The ComputeEngine now exports the bignum() and complex() methods that can be used to create bignum and complex numbers from strings or numbers. The methods isBigNum() and isComplex() have also been added to check if a value is a bignum (Decimal) or complex (Complex) number, for example as returned by expr.numericValue.
  • #69 \leq was incorrectly parsed as Equals instead of LessEqual
  • #94 The \exp command was not parsed correctly.
  • Handle PlusMinus in infix and prefix position, i.e. a\pm b and \pm a.
  • Improved parsing, serialization
  • Improved simplification
  • Improved evaluation of Sum and Product
  • Support complex identifiers (i.e. non-latin scripts, emojis).
  • Fixed serialization of mixed numbers.

0.12.1 2022-12-01​

Work around unpckg.com issue with libraries using BigInt.

0.12.0 2022-11-27​

Breaking Changes​

  • The expr.symbols property return an array of string. Previously it returned an array of BoxedExpression.

Improvements​

  • Rewrote the rational computation engine to use JavaScript bigint instead of Decimal instances. Performance improvements of up to 100x.
  • expr.freeVars provides the free variables in an expression.
  • Improved performance of prime factorization of big num by x100.
  • Added ["RandomExpression"]
  • Improved accuracy of some operations, for example expr.parse("1e999 + 1").simplify()

Issues Resolved​

  • When ce.numericMode === "auto", square roots of negative numbers would return an expression instead of a complex number.
  • The formatting of LaTeX numbers when using ce.latexOptions.notation = "engineering" or "scientific" was incorrect.
  • The trig functions no longer "simplify" to the less simple exponential formulas.
  • The canonical order of polynomials now orders non-lexicographic terms of degree 1 last, i.e. "ax^2+ bx+ c" instead of "x + ax^2 + bx".
  • Fixed evaluation of inverse functions
  • Fixed expr.isLess, expr.isGreater, expr.isLessEqual, expr.isGreaterEqual and ["Min"], ["Max"]

0.11.0 2022-11-18​

Breaking Changes​

  • The signature of ce.defineSymbol(), ce.defineFunction() and ce.pushScope() have changed

Improvements​

  • When a constant should be held or substituted with its value can now be more precisely controlled. The hold symbol attribute is now holdUntil and can specify at which stage the substitution should take place.

Issues Resolved​

  • Some constants would return a value as bignum or complex even when the numericMode did not allow it.
  • Changing the value or domain of a symbol is now correctly taken into account. Changes can be made with ce.assume(), ce.set() or expr.value.
  • When a symbol does not have a value associated with it, assumptions about it (e.g. "x > 0") are now correctly tracked and reflected.

0.10.0 2022-11-17​

Breaking Changes​

  • expr.isLiteral has been removed. Use expr.numericValue !== null and expr.string !== null instead.

Issues Resolved​

  • Calling ce.forget() would not affect expressions that previously referenced the symbol.

Improvements​

  • More accurate calculations of some trig functions when using bignums.
  • Improved performance when changing a value with ce.set(). Up to 10x faster when evaluating a simple polynomial in a loop.
  • ce.strict can be set to false to bypass some domain and validity checks.

0.9.0 2022-11-15​

Breaking Changes​

  • The head of a number expression is always Number. Use expr.domain to be get more specific info about what kind of number this is.
  • By default, ce.box() and ce.parse() return a canonical expression. A flag can be used if a non-canonical expression is desired.
  • The API surface of BoxedExpression has been reduced. The properties machineValue, bignumValue, asFloat, asSmallInteger, asRational etc... have been replaced with a single numericValue property.
  • parseUnknownSymbol is now parseUnknownIdentifier

Improvements​

  • Support angles in degrees with 30\degree, 30^\circ and \ang{30}.

  • More accurate error expressions, for example if there is a missing closing delimiter an ["Error", ["ErrorCode", "'expected-closing-delimiter'", "')'"]] is produced.

  • ["Expand"] handles more cases

  • The trig functions can now have a regular exponent, i.e.\cos^2(x) in addition to -1 for inverse, and a combination of \prime, \doubleprime and ' for derivatives.

  • ce.assume() handle more expressions and can be used to define new symbols by domain or value.

  • Better error message when parsing, e.g. \sqrt(2) (instead of \sqrt{2})

  • Better simplification for square root expressions:

    • \sqrt{25x^2} -> 5x

  • Improved evaluation of ["Power"] expressions, including for negative arguments and non-integer exponents and complex arguments and exponents.

  • Added Arccot, Arcoth, Arcsch, Arcscc, Arsech and Arccsc

  • expr.solve() returns result for polynomials of order up to 2.

  • The pattern.match() function now work correctly for commutative functions, i.e. ce.pattern(['Add', '_a', 'x']).match(ce.parse('x+y')) -> {"_a": "y"}

  • Added ce.let() and ce.set() to declare and assign values to identifiers.

  • Preserve exact calculations involving rationals or square root of rationals.

    • \sqrt{\frac{49}{25}} -> \frac{7}{5}

  • Addition and multiplication provide more consistent results for evaluate() and N(). Evaluate returns an exact result when possible.

    • EXACT
      • 2 + 5 -> 7
      • 2 + 5/7 -> 19/7
      • 2 + √2 -> 2 + √2
      • 2 + √(5/7) -> 2 + √(5/7)
      • 5/7 + 9/11 -> 118/77
      • 5/7 + √2 -> 5/7 + √2
      • 10/14 + √(18/9) -> 5/7 + √2
      • √2 + √5 -> √2 + √5
      • √2 + √2 -> 2√2
      • sin(2) -> sin(2)
      • sin(π/3) -> √3/2
    • APPROXIMATE
      • 2 + 2.1 -> 4.1
      • 2 + √2.1 -> 3.44914
      • 5/7 + √2.1 -> 2.16342
      • sin(2) + √2.1 -> 2.35844

  • More consistent behavior of the auto numeric mode: calculations are done with bignum and complex in most cases.

  • JsonSerializationOptions has a new option to specify the numeric precision in the MathJSON serialization.

  • Shorthand numbers can now be strings if they do not fit in a float-64:

// Before
["Rational", { "num": "1234567890123456789"}, { "num": "2345678901234567889"}]

// Now
["Rational", "1234567890123456789", "2345678901234567889"]
  • \sum is now correctly parsed and evaluated. This includes creating a local scope with the index and expression value of the sum.

Bugs Fixed​

  • The parsing and evaluation of log functions could produce unexpected results
  • The \gamma command now correctly maps to ["Gamma"]
  • Fixed numeric evaluation of the ["Gamma"] function when using bignum
  • #57 Substituting 0 (i.e. with expr.subs({})) did not work.
  • #60 Correctly parse multi-char symbols with underscore, i.e. \mathrm{V_a}
  • Parsing a number with repeating decimals and an exponent would drop the exponent.
  • Correct calculation of complex square roots
    • \sqrt{-49} -> 7i
  • Calculations were not always performed as bignum in "auto" numeric mode if the precision was less than 15. Now, if the numeric mode is "auto", calculations are done as bignum or complex numbers.
  • If an identifier contained multiple strings of digits, it would not be rendered to LaTeX correctly, e.g. V20_20.
  • Correctly return isReal for real numbers

0.8.0 2022-10-02​

Breaking Changes​

  • Corrected the implementation of expr.toJSON(), expr.valueOf() and added the esoteric [Symbol.toPrimitive]() method. These are used by JavaScript when interacting with other primitive types. A major change is that expr.toJSON() now returns an Expression as an object literal, and not a string serialization of the Expression.

  • Changed from "decimal" to "bignum". "Decimal" is a confusing name, since it is used to represent both integers and floating point numbers. Its key characteristic is that it is an arbitrary precision number, aka "bignum". This affects ce.numericMode which now uses bignum instead of decimal, expr.decimalValue->expr.bignumValue, decimalValue()->bignumValue()

Bugs Fixed​

  • Numerical evaluation of expressions containing complex numbers when in decimal or auto mode produced incorrect results. Example: e^{i\\pi}

0.7.0 2022-09-30​

Breaking Changes​

  • The ce.latexOptions.preserveLatex default value is now false
  • The first argument of the ["Error"] expression (default value) has been dropped. The first argument is now an error code, either as a string or an ["ErrorCode"] expression.

Features​

  • Much improved LaTeX parser, in particular when parsing invalid LaTeX. The parser now avoids throwing, but will return a partial expression with ["Error"] subexpressions indicating where the problems were.
  • Implemented new domain computation system (similar to type systems in programming languages)
  • Added support for multiple signatures per function (ad-hoc polymorphism)
  • Added FixedPoint, Loop, Product, Sum, Break, Continue, Block, If, Let, Set, Function, Apply, Return
  • Added Min, Max, Clamp
  • Parsing of \sum, \prod, \int.
  • Added parsing of log functions, \lb, \ln, \ln_{10}, \ln_2, etc...
  • Added expr.subexpressions, expr.getSubexpressions(), expr.errors, expr.symbols, expr.isValid.
  • Symbols can now be used to represent functions, i.e. ce.box('Sin').domain correctly returns ["Domain", "Function"].
  • Correctly handle rational numbers with a numerator or denominator outside the range of a 64-bit float.
  • Instead of a Missing symbol an ["Error", "'missing'"] expression is used.
  • Name binding is now done lazily
  • Correctly handle MathJSON numbers with repeating decimals, e.g. 1.(3).
  • Correctly evaluate inverse functions, e.g. ce.parse('\\sin^{-1}(.5)).N()
  • Fixed some LaTeX serialization issues

Read more at Core Reference and [Arithmetic Reference] (https://cortexjs.io/compute-engine/reference/arithmetic/)

Bugs Fixed​

  • #43 If the input of ce.parse() is an empty string, return an empty string for expr.latex or expr.json.latex: that is, ensure verbatim LaTeX round-tripping
  • Evaluating some functions, such as \arccos would result in a crash
  • Correctly handle parsing of multi-token decimal markers, e.g. {,}

0.6.0 2022-04-18​

Improvements​

  • Parse more cases of tabular environments
  • Handle simplify and evaluate of inert functions by default
  • Avoid unnecessary wrapping of functions when serializing LaTeX
  • Parse arguments of LaTeX commands (e.g. \vec{})
  • #42 Export static ComputeEngine.getLatexDictionary
  • Parse multi-character constants and variables, e.g. \mathit{speed} and \mathrm{radius}
  • Parse/serialize some LaTeX styling commands: \displaystyle, \tiny and more

0.5.0 2022-04-05​

Improvements​

  • Correctly parse tabular content (for example in \begin{pmatrix}...\end{pmatrix}
  • Correctly parse LaTeX groups, i.e. {...}
  • Ensure constructible trigonometric values are canonical
  • Correct and simplify evaluation loop for simplify(), evaluate() and N().
  • #41 Preserve the parsed LaTeX verbatim for top-level expressions
  • #40 Correctly calculate the synthetic LaTeX metadata for numbers
  • Only require Node LTS (16.14.2)
  • Improved documentation, including Dark Mode support

0.4.4​

Release Date: 2022-03-27

Improvements​

  • Added option to specify custom LaTeX dictionaries in ComputeEngine constructor
  • expr.valueOf returns rational numbers as [number, number] when applicable
  • The non-ESM builds (compute-engine.min.js) now targets vintage JavaScript for improved compatibility with outdated toolchains (e.g. Webpack 4) and environments. The ESM build (compute-engine.min.esm.js) targets evergreen JavaScript (currently ECMAScript 2020).

0.4.3​

Release Date: 2022-03-21

Transition Guide from 0.4.2​

The API has changed substantially between 0.4.2 and 0.4.3, however adapting code to the new API is very straightforward.

The two major changes are the introduction of the BoxedExpression class and the removal of top level functions.

Boxed Expression​

The BoxedExpression class is a immutable box (wrapper) that encapsulates a MathJSON Expression. It provides some member functions that can be used to manipulate the expression, for example expr.simplify() or expr.evaluate().

The boxed expresson itself is immutable. For example, calling expr.simplify() will return a new, simplified, expression, without modifying expr.

To create a "boxed" expression from a "raw" MathJSON expression, use ce.box(). To create a boxed expression from a LaTeX string, use ce.parse().

To access the "raw" MathJSON expression, use the expr.json property. To serialize the expression to LaTeX, use the expr.latex property.

The top level functions such as parse() and evaluate() are now member functions of the ComputeEngine class or the BoxedExpression class.

There are additional member functions to examine the content of a boxed expression. For example, expr.symbol will return null if the expression is not a MathJSON symbol, otherwise it will return the name of the symbol as a string. Similarly, expr.ops return the arguments (operands) of a function, expr.asFloat return null if the expression does not have a numeric value that can be represented by a float, a number otherwise, etc...

Canonical Form​

Use expr.canonical to obtain the canonical form of an expression rather than the ce.format() method.

The canonical form is less aggressive in its attempt to simplify than what was performed by ce.format().

The canonical form still accounts for distributive and associative functions, and will collapse some integer constants. However, in some cases it may be necessary to invoke expr.simplify() in order to get the same results as ce.format(expr).

Rational and Division​

In addition to machine floating points, arbitrary precision numbers and complex numbers, the Compute Engine now also recognize and process rational numbers.

This is mostly an implementation detail, although you may see ["Rational", 3, 4], for example, in the value of a expr.json property.

If you do not want rational numbers represented in the value of the .json property, you can exclude the Rational function from the serialization of JSON (see below) in which case Divide will be used instead.

Note also that internally (as a result of boxing), Divide is represented as a product of a power with a negative exponent. This makes some pattern detection and simplifications easier. However, when the .json property is accessed, product of powers with a negative exponents are converted to a Divide, unless you have included Divide as an excluded function for serialization.

Similarly, Subtract is converted internally to Add, but may be serialized unless excluded.

Parsing and Serialization Customization​

Rather than using a separate instance of the LatexSyntax class to customize the parsing or serialization, use a ComputeEngine instance and its ce.parse() method and the expr.latex property.

Custom dictionaries (to parse/serialize custom LaTeX syntax) can be passed as an argument to the ComputeEngine constructor.

For more advanced customizations, use ce.latexOptions = {...}. For example, to change the formatting options of numbers, how the invisible operator is interpreted, how unknown commands and symbols are interpreted, etc...

Note that there are also now options available for the "serialization" to MathJSON, i.e. when the expr.json property is used. It is possible to control for example if metadata should be included, if shorthand forms are allowed, or whether some functions should be avoided (Divide, Sqrt, Subtract, etc...). These options can be set using ce.jsonSerializationOptions = {...}.

Comparing Expressions​

There are more options to compare two expressions.

Previously, match() could be used to check if one expression matched another as a pattern.

If match() returned null, the first expression could not be matched to the second. If it returned an object literal, the two expressions matched.

The top-level match() function is replaced by the expr.match() method. However, there are two other options that may offer better results:

  • expr.isSame(otherExpr) return true if expr and otherExpr are structurally identical. Structural identity is closely related to the concept of pattern matching, that is ["Add", 1, "x"] and ["Add", "x", 1] are not the same, since the order of the arguments is different. It is useful for example to compare some input to an answer that is expected to have a specific form.
  • expr.isEqual(otherExpr) return true if expr and otherExpr are mathematically identical. For example ce.parse("1+1").isEqual(ce.parse("2")) will return true. This is useful if the specific structure of the expression is not important.

It is also possible to evaluate a boolean expression with a relational operator, such as Equal:

console.log(ce.box(["Equal", expr, 2]).evaluate().symbol);
// -> "True"

console.log(expr.isEqual(ce.box(2)));
// -> true

Before / After​

BeforeAfter
expr = ["Add", 1, 2]expr = ce.box(["Add", 1, 2])
expr = ce.evaluate(expr)expr = expr.evaluate()
console.log(expr)console.log(expr.json)
expr = new LatexSyntax().parse("x^2+1")expr = ce.parse("x^2+1")
new LatexSyntax().serialize(expr)expr.latex
ce.simplify(expr)expr.simplify()
await ce.evaluate(expr)expr.evaluate()
ce.N(expr)expr.N()
ce.domain(expr)expr.domain
ce.format(expr...)expr.canonical
expr.simplify()

0.3.0​

Release Date: 2021-06-18

Improvements​

  • In LaTeX, parse \operatorname{foo} as the MathJSON symbol "foo".