Repository navigation
Releases
MathCAS v2 nightly (75b96ad)
The latest build of main (75b96ad, 2026-11-19). Not a release: for a numbered version see the
releases tagged v*. Windows: the installer (.exe) or the zip; Linux: the tarball, the AppImage or the Flatpak bundle; macOS: the disk image for Apple Silicon (arm64) or Intel (x64), see its README; Android: the .apk.
MathCAS v2 0.17.0
Notebook. A space that multiplies is shown as a space, as in written mathematics (m c^2, 2 pi r); the
multiplication sign that is not typed appears only between two numbers (2 3, which would read as 23) and
where a product runs across a line end. A letter typed right after a number gets a space before it (3pi
becomes 3 pi, so 3 pi^2 cannot be read as the square of 3 pi); 2e5, 3im, 0x1F and 0b101 stay as
typed. On Windows the text pointer over a page is the window's own, in the theme's colours, so it no longer
turns white on a light page on systems where the system's pointer loses its colour.
Linux. The AppImage carries update information, and each release has its .zsync file: AppImageUpdate and
similar tools can bring an installed AppImage to the newest release.
Printed forms that changed. max, min, gcd, lcm, bit_and, bit_or, bit_xor and xor flatten
and sort their arguments (lcm(x, y, 2) is lcm(2, x, y)); indexed(x, 1) is indexed(x, [1]); a
piecewise drops cases that cannot be reached and joins neighbours with one value; unions and differences of
segments of the line are disjoint segments. together and cancel keep a denominator factored as written,
apart writes each fraction as a rational times a primitive numerator, expand_all expands denominators
too, trig_reduce and simplify put the rational content outside ((1/2)*(1 - cos(2*x))), factor reads
powers with one symbolic exponent as a variable, trig_factor merges Pythagorean pairs. piecewise_expand
orders its pieces by value, joins pieces with one value and drops a piece equal to the default; exp_to_trig
writes real exponentials with hyperbolic functions (hyperbolic: false keeps them); trig_to_exp keeps tan
and cot as one fraction; factor_terms pulls out Gaussian content; apart works in one kernel such as
exp(x). The derivative of max(x, 2) is if x <= 2 then 0 else 1. Unit names are written in full
("Meters"). memory_constrained answers failure(limit, memory) also when a builder declines beforehand.
Each docs page states the form its function guarantees.
Corrected answers. spherical_bessel_j(0, x) is sinc(x) (defined at 0);
reduce(1 + product_log(x) == 0, x, reals) gives x == -1/e; solve(y^3 - x*y == 2, y, reals) gives real
roots only; lommel_s1 and lommel_t1 at 0; q_factorial, q_gamma and q_binomial at q = 1;
logical_expand(not unequal(x, y, z)); replace_at inside quote does not evaluate; region_member of a
polygon with symbolic edges stays unevaluated. numerical_sort, sort_by, reverse_sort and ordering
follow one documented order. integrate(1, element(x, ball(3))) is (4/3)*pi;
minimize(sqrt(x^2 + y^2), x + y <= 1) is 0; asymptotic_d_solve_value no longer takes an infinite limit for
a regular singular point; jacobi_epsilon and jacobi_zn at K + iK'; coulomb_f of integer order has no
singularities; function_poles(zeta(x), x) is [[1, 1]].
New. reduce over the complexes with order relations, re, im_part and conjugate; eliminate with
radicals and invertible functions; find_root at a working precision with inexact coefficients;
dirichlet_l and prime_zeta_p of intervals; property sets (domains, periods, singularities) of theta,
Hankel, q-hypergeometric, Weierstrass and modular functions; exact ranges of Bessel, Airy and Struve functions
at certified roots of the derivative; monomial orders in coefficient_rules and monomial_list;
distance_matrix with nominal features; ball(n) is the unit ball of n dimensions.
dirichlet_character(k, j, n) with Conrey labels; elliptic_e(phi, m) and elliptic_pi(n, m) beyond m = 1;
polygamma of non-integer and complex order, polylog of complex order beyond the unit disk, appell_f1
outside the unit bidisk, mittag_leffler_e at machine reals, Kelvin functions of half-integer order;
differential_root and difference_root in one written form with derivatives and series; series at
infinity of erfi and dawson_f and of zeta at its pole; n_dsolve for partial differential equations in t
and x and for vector unknowns; asymptotic_solve with Puiseux branches; optimizers with regions and cases as
constraints; region primitives without arguments take their documented default placement;
centered_interval of exact numbers; clebsch_gordan of symbols. Lists, tables and arrays that would not
fit in memory are declined before they are built.
Faster. Elliptic K and E above 1 on enclosures (inputs with elliptic_k in reduce and
function_domain went from 20 s and more to under 1 s). cellular_automaton with numbered rules (130 ms to
a few ms), dawson_f at large arguments, extrema of Airy and Bessel expressions, is_prime_power, fourier
of matrices, squares_r, contour plots of interpolations, Stirling number tables.
MathCAS v2 0.16.3
Linux. The AppImage carries the keyboard libraries the window needs under X11 (libxkbcommon-x11 and
what it uses), so it starts on a system that does not have them; the package build opens the window on a bare
display to check it.
MathCAS v2 0.16.2
Linux. The tarball and the AppImage are built against an older C library (glibc 2.35), so they run on
Ubuntu 22.04, Debian 12 and other systems of that age; before, they needed the newest one.
A Flatpak bundle (MathCAS-v2-<version>-linux-x64.flatpak) is built with each release:
flatpak install --user it and run io.github.rajatmo.MathCASv2.
MathCAS v2 0.16.1
Packages for more stores. A Flatpak manifest with its metadata (packaging/flatpak/); the texts and
pictures of the Android app for a store (fastlane/metadata/android/) and a build recipe for F-Droid
(packaging/fdroid/); UNSIGNED=1 packaging/android/build-apk.sh leaves the Android package unsigned for a
store that signs with its own key.
MathCAS v2 0.16.0
macOS on Intel. The disk image is also built for Macs with an Intel processor
(MathCAS-v2-<version>-macos-x64.dmg), beside the one for Apple Silicon.
Changes that old input will notice. They settle the language before 1.0.
Names. Predicates are is_*: the 27 that ended in _q are renamed (interval_member_q is
is_interval_member). Option values and mode arguments are lower-case names, not strings: method: newton,
direction: from_above, axis_object([horizontal, 0]). A string stays only where its spelling is fixed from
outside (a file format, a colour scheme, a plot theme). In notebooks the labels are in N and out N and cell
style names are lower case; a notebook written with the old names is converted when it is opened.
Failures and loops. A failure that is called is that failure: f(1) where f is a failure gives the
failure instead of a call that stays as written. break() and continue() are read by map, table,
select, fold and the other iterating functions, in their parallel forms too. table() without a body stays
unevaluated.
Monitors are functions. evaluation_monitor and step_monitor take a function of the current values;
there are no held option values. Anything else is failure(not_a_function, ...), and a function that has no
monitor says unknown_option (the function plots). find_fit and the Taylor method of n_dsolve now call
monitors.
Only proved answers. The property functions (function_injective, function_sign,
function_monotonicity, function_convexity and the others) answer from enclosures, exact values, composition
rules and limits. Where a sampled answer was given before and no proof is found, the call stays unevaluated.
Functions of several intervals use the corner rule only when every derivative sign is proved over the box.
Work, not seconds. The budgets of integration, limits, d_solve, simplify, sums and the decision service
count operations instead of time, so an answer does not depend on the speed of the machine or on what else it is
doing.
Plots. Plots draw no grid lines unless grid_lines asks for them; grid_lines: automatic gives the grid
they drew before (at the ticks, on logarithmic axes at the powers of ten). show of several plots ticks the
combined picture afresh for the range that holds them all (or the plot_range given to show), and pointing at
it in the notebook snaps to the curves and points of every plot, as on a single plot.
Interrupts. Drawing a picture honours an interrupt within a quarter of a second (sampling, meshing and
writing the picture), and so does prime_pi of a large argument.
Wrong answers corrected. Found by testing exact and proved answers against high-precision numbers:
acot, fractional_part and mod of an interval across a jump gave too small an image (they stay unevaluated);
a limit could take a constant that is zero for its leading term; series of sawtooth_wave and square_wave
at a jump gave one side only; function_monotonicity and function_convexity answered across the gaps of a
domain of several pieces (tan on x > 0); hypergeometric_1f1(-2, -2, z); legendre_p of integer order at
high precision; Laguerre, Jacobi, Gegenbauer and Hermite polynomials at a degree given as a machine number;
mittag_leffler_e with integer parameters; the digits claimed by a numerical derivative. In the language: a
failure assigned to a part of a list is held by the list, an assignment never stores break(), match reads a
failure as data, and break()/continue() work in map over an association and in array and outer.
More of the same kind, from checking numerical and exact answers independently: n_integrate lost the mass
at a singular end that is not 0 ((1 - x)^(-2/3) on 0 to 1 gave 2.9999879 as converged); find_root at high
precision could claim digits it did not have; sum(x^k, [k, -inf, inf]) gave 0; simplify gave 0 for
gamma(x + 1)/gamma(x) - sqrt(x^2); the series of asinh(1/x) for negative x; series of the Hankel
functions at 0. Assignments to a span, to several positions or to a column (l[2:4] = 0, m[all, 2] = v) did
nothing; they set every selected part, and a position outside the list is a failure.
Property answers are exact at their boundaries: function_bijective(x^3 + a*x^2 + b*x + c, x) is
a^2 - 3*b <= 0 (the boundary was left out); function_injective no longer uses one sign of the derivative
over separate intervals; convexity with a parameter sees the poles; function_analytic(sqrt(x^2), x) is false
and abs(x^2) is analytic; strict monotonicity of x^3 is answered.
Faster. A small input takes about a tenth of the time it took (parsing no longer starts a thread per
input); a plot draws its PDF and print versions when an export first asks for them; plots of large rasters no
longer print them; sign of a complex constant, root counts with gamma and matrix_exp of a matrix logarithm
are much quicker. A kernel on a busy machine answers a small input as fast as on an idle one; table over
ranges of machine numbers compiles its body (a million elements in a tenth of a second); large 3D pictures
are read and written in about half the time.
Sums carry their conditions. An infinite sum or product with a parameter gives a conditional_expression
with its condition of convergence (sum(p^k/k, [k, 1, inf]) holds for abs(p) < 1 and at p == -1);
generate_conditions: false gives the value alone, and a condition that the assumptions imply is dropped.
Notebook format version. The first line of a .mcn file names the format version. A file of a newer minor
version opens with a warning; one of a newer major version is refused with a message that says so.
Limits of memory and time. simplify, factor and the other functions that walk a term visit a shared
subterm once, so a deeply nested value with shared parts no longer takes exponential time. to_string of a value too large to print gives failure(limit, memory);
polynomial_gcd, together, cancel and expand of very large inputs are faster and honour an interrupt;
n_dsolve's Taylor method stops at the end of its range. Function plots call an evaluation_monitor.
More answers. Poles of a function composed with a polynomial (gamma(x^3 - x/3) in a disc); root counts of
functions with poles and of Bessel, Airy and gamma functions; products that are q-Pochhammer symbols; expansions
at infinity of the error, Fresnel, sine and cosine integral, Hankel, Whittaker and polylogarithm functions; many
closed forms in function_expand; elliptic_f at a complex amplitude to any precision; bessel_j with an
interval order; number theory under assumptions in refine (distinct primes are coprime, Fermat, Wilson);
piecewise_expand of ramp, unit_box, unit_triangle, floor, mod and the other step functions; areas of
Boolean combinations of plane regions; symbolic Cholesky factors; facts about primes, algebraic and complex
numbers in conditions. Heun functions at singular points; Ferrers functions of the second kind with an order;
Zernike functions and spherical harmonics off the integers; centered intervals with complex parts for the
Fresnel, Hankel, gamma and zeta functions; Bessel functions of half-integer order in elementary functions;
recurrences solved by products over several steps and by the recurrences of orthogonal polynomials and Bessel
functions; sums of two, four and eight squares as divisor sums; images of the principal branches of the inverse
functions; series at the singular points of acsch, asech and acoth. Proved complex boxes of the
Weierstrass and modular functions; alternating series of rational functions; sums of Bell and Stirling numbers;
general series coefficients; sums of squares above degree 2 with an exact certificate; eigenvectors and singular
value decompositions of high-precision matrices at their precision; refine joins facts about several symbols
(a > 0 from a > b and a > -b); find_instance proves an empty answer over the integers.
Roots of any function. root([f, x0]) with a function that is not a polynomial is an exact number: the
root near x0 is found and proved on a box, real or complex, n gives it to any precision, and it is refused
at a double root, without a root nearby or on a branch cut. solve, reduce and n_solve give such roots
for one analytic equation in a bounded region, and arg_max, arg_min, max_value, min_value, maximize
and minimize on a closed interval compare the proved values at the ends and at the roots of the derivative.
Also: proved ranges of oscillating functions (bessel_j(0, x), sinc); sequences reduced to one
difference_root; systems of recurrences with constant coefficients; asymptotic relations with parameters give
their conditions; Fibonacci and Lucas numbers of real index; Hankel transforms of more functions.
Also new. Proved images of more special functions on wide intervals; prime_pi up to 2^60;
hypergeometric functions (p+1)Fp beyond the unit circle; Lommel and Weber functions at exceptional orders;
eigenvectors of matrices with parameters; membership of a symbolic point in convex regions, cylinders, cones
and tori; list plots of several data sets; reduce in one integer variable; balanced Hankel transforms; double
sums along diagonals; Mathieu characteristic values at complex q.
Limits onto branch cuts and other corrections. A limit taken onto a branch cut of elliptic_e, elliptic_k,
bessel_y, product_log, gamma_regularized and nineteen more functions is the value on the side the limit
comes from. function_domain knows that heaviside_theta, round, floor, ceiling and the real-only
functions (clip, ramp, unit_step) are not defined or not continuous everywhere; piecewise_expand of
clip with symbolic bounds; sum_convergence on the boundary circle and at poles ...
MathCAS v2 0.15.0
Answers that are proved, and more of them: intervals, properties with parameters, limits on branch cuts.
Intervals are enclosures. The image of an interval or centered_interval under a function is now proved:
monotone pieces and extrema come from enclosures of the derivatives, and an interval across a pole gives the whole
line (gamma, tan and zeta across a pole no longer give a bounded interval). Complex centered intervals of the
elementary functions and their inverses evaluate by box arithmetic. Where no image can be proved the call stays
unevaluated: this now happens for some functions on wide intervals (binomial, catalan_number,
harmonic_number, whittaker_w and a few more) that gave a sampled answer before.
Properties and conditions. function_analytic, function_meromorphic and the other property functions
answer false only when it holds for every value of the parameters; otherwise they give a condition on the
parameters or stay unevaluated. reduce and solve over the integers, rationals, primes and algebraic numbers
keep a membership that a free parameter leaves open as a condition, and take more argument forms; step functions
(floor, ceiling, round, mod) are solved on unbounded lines. find_instance searches one integer or real
variable.
Limits and calculus. One-sided limits onto the branch cuts of the inverse circular and hyperbolic functions
take the side the path comes from; direction: complexes gives the value where the function is continuous and
no limit where two directions differ. Equations between powers of positive bases; systems in recurrence_table;
nested harmonic sums as multiple harmonic numbers; Puiseux branches in asymptotic_solve; certified counts of the
roots of non-polynomial functions; Mathieu's equation in d_solve.
Algebra. Gröbner bases over the integers and over rational functions of parameters; sums over roots by their
structure; eliminate with inequations; interpolating polynomials in several variables; generating functions of
sequences with mod, quotient and floor; apart of an expression in several variables.
Special functions and numbers. Modular functions (klein_invariant_j, modular_lambda, dedekind_eta) at
every point of the upper half plane, at machine and arbitrary precision, real on the lines where they are real.
Mathieu functions by Fourier sums (plots are fast) and at high precision. inverse_gudermannian keeps one branch.
Inverse functions of direct functions reduce at numbers (acoth(coth(5))). Norms, singular values and eigenvalues
of arbitrary-precision matrices. Lommel functions of non-integer order; Appell functions as sums.
Lists and data. normalized; knapsack_solve with bounds; bin_counts and bin_lists in several
dimensions; the least-area oriented rectangle; more hash methods, and byte arrays hashed by their bytes;
interpolation from derivative data. A new memory_available(): builders of large matrices and grids refuse a
result that does not fit instead of ending the kernel.
Language. The name of a pattern variable is held, so x_ works when x has a value. A new key assigned
inside a local-variable block is kept.
Packages. The Android package and the Linux AppImage are built and published with the other packages.
MathCAS v2 0.14.0
A first Android app, and more answers in calculus, special functions and data.
Android (phones and tablets). The notebook with its kernel as an Android package, with a layout of its own
in Material Design: a start screen with the notebooks kept in the app, "New calculation" and "Open a file" (the
system file picker); a notebook screen where a row of keys with Run sits on the on-screen keyboard and completions
show as chips; Copy and Share for a selected result or plot (a plot is shared as a PNG through the system share
sheet). Changes are saved by themselves. Light and dark follow the system. The package is built with
packaging/android/build-apk.sh (Android SDK and NDK, no Gradle) and tested on an emulator with
packaging/android/smoke.sh; it is not among the published packages yet and has not been tried on many devices.
Calculus. Series of sign, abs and arg of complex arguments; limits of sequences; fundamental solutions
of the heat and wave operators; derivatives of held transforms; radial transforms; unit waves. Hypergeometric
closed forms with integer parameter differences. cos and sin of atan simplify.
Numbers. Logarithms to a base are correctly rounded (log(2.0, 8.0) is 3.0). Newton steps in find_root
are held to a trust radius; roots of imaginary-valued functions are found. Uncertain results keep real centres,
and complex uncertain numbers are supported. Expansions in non-integer bases; Gaussian number theory; linear algebra
modulo a prime. A fast machine-precision Barnes G.
Data and pictures. Charts of data containers; more picture primitives; list argument forms; alphabetical
orders; symbolic cumulants; region measures, and areas from vertex differences. A plot is still drawn when one
epilog primitive has a non-numeric coordinate (that primitive is left out, with a message).
Fixed. A crash in ball comparisons. A numeric routine whose compilation fails is interpreted instead of
failing. is_even and is_odd answer false for what is not an integer.
Desktop notebook. In a narrow window the menus and the toolbar scroll sideways, the status bar's items end at
the file name, and dialogs take the window's width.
MathCAS v2 0.13.0
MathCAS runs on macOS, about sixty new functions, and more answers where there were none.
macOS (Apple Silicon). A disk image with the application MathCAS Notebook.app: the notebook, the
documentation window, the REPL, the kernel and the language server, with the .mcn document type. It is built and
tested with the other packages on every change. The application is not notarized: the first launch needs Open
from the context menu (see the README in the image). Intel Macs are not built yet.
New functions.
- Wavelets (a new library):
discrete_wavelet_transform,stationary_wavelet_transform,
continuous_wavelet_transform, the packet transform and the inverses, wavelet plots and scalograms, and the
Haar, Daubechies, Mexican hat, Morlet and other families. The discrete transforms reconstruct their data exactly. - Series:
inverse_seriesandcompose_series, also for series in fractional powers. - Algebra and numbers:
solve_always,find_linear_recurrence,mathieu_characteristic_b,pauli_matrix,
permute,polynomial_q, bit operations (bit_set,bit_clear,bit_shift_left), distance functions. - Lists and arrays:
array_filter,array_reduce,subset_count,find_shortest_tour(exact, up to 20
points),unitize,interval_intersection,find_divisions. - Language:
break()andcontinue()inwhileandforloops;symbol,symbol_name,with_cleanup; the
pattern headsverbatimandhold_pattern. - Uncertain values:
vector_around,around_replace.
More answers.
- Transforms: Hankel and inverse Hankel transforms through Mellin products; Mellin convolution; Radon
transforms of disks and point masses with their inverses. - Sums:
generate_conditions: truegives the convergence condition (sum(1/n^a, [n, inf])iszeta(a)when
re(a) > 1); Bertrand-type series; divisor sums with a symbolic argument; Dirichlet series and Euler products. - Limits and values at infinity of the special functions;
to_radicalsoptions; Pell equation families. - Lists and numerics: vector constraints in
n_minimizeand its siblings;heads: truein the mapping
functions; ordering functions withup_to; permutedtranspose; gradient, divergence, curl and Laplacian in
polar, cylindrical and spherical coordinates;log(b, x). - Algebra:
trig:andmodulus:options oftogether,cancelandexpand; bitwise laws on symbols;
Gaussian gcd and lcm of rationals; inverse Jacobi function properties form > 1. abs,re,im_part,arg,conjugateandsignall take numeric factors out (abs(-2 k) = 2 abs(k)).
Fixed.
- A freeze: plotting the solution of a differential equation that had been solved again after its session
ended could stop the kernel for good, and the interrupt did nothing. Fixed (a lock was taken twice). box_matrix(0, [n, n])and its siblings divided by zero.- Compiled numerics on 64-bit ARM processors.
Notebook.
- 3D plots in the PDF export are laid out at the width of the text column, as 2D plots are.
- Typing an opening bracket at the top of a very long cell no longer lays the whole cell out again (about six
times faster in a 2,000-line cell). - A closing bracket inserted automatically is kept when the caret moves inside it (a fraction filled with the
arrow keys no longer ends in))); completion does not open on2a.
Under the hood. Output typesetting has its own crate (mc-typeset); the compiler version is pinned in
rust-toolchain.toml; the speed limit of the documentation examples can be set per machine
(MC_DOCS_SLOW_SECS).
MathCAS v2 0.12.0
Answers where there were none: this release works on inputs MathCAS left unevaluated, failed on or timed out on.
Every fix carries tests on the input and on variations of it, each checked by an independent route.
Function properties know the special functions. function_domain, function_singularities,
function_analytic, function_range, function_sign, function_monotonicity, function_convexity,
function_injective, function_surjective and their siblings read one table of what each special function is
(about 160 entries with their sources: domain, singular points, range, bounds, sign, monotonicity, convexity,
periodicity), and combine it through sums, products, powers and compositions. Many inputs that stayed
unevaluated now answer.
Calculus. Hilbert and inverse Hilbert transforms by a general method (exponentials times rational functions,
multiple poles, compact piecewise polynomials); asymptotic relations in several variables; optimisation of periodic
objectives and of vector variables; fourier_sequence_transform of exponentials and periodic sequences; series of
the inverse trigonometric and hyperbolic functions at infinity and at branch points; integrals over regions given
by boole; coefficient of monomials and with modulus:. A limit that oscillates is failure(no_limit).
Algebra and number theory. modulus: in the expand and polynomial-division families; gaussian_integers: in
the factoring and divisor functions; groebner_basis with weight-matrix, graded and elimination orders and over
finite fields; cellular_automaton with rule lists, function rules and ranges; find_sequence_function recognises
the library's integer functions.
Special functions. Derivatives of functions without a derivative rule are computed numerically with a
convergence check (for find_root and plots), or stay unevaluated; Weierstrass functions in closed form on the
degenerate lattice and at machine precision; the hypergeometric function at complex machine arguments outside the
unit disk; Lommel functions; mathieu_c_prime (new).
Lists, numerics, linear algebra. List operations on expressions with any head; n with real precisions and
accuracy goals; solve inverts csch, sech, gudermannian and logistic_sigmoid; eigenvalues and
characteristic polynomials of matrix pencils; rank and row_reduce with modulus:; alphabetical order for
Slovak, Spanish, Swedish and Serbian.
Wrong values fixed. struve_h and struve_l at half-integer order and negative argument; Kelvin functions at
and near 0 (kelvin_ker(2, 0.) is 1/2; values near 0 had lost twelve digits); hyperfactorial and barnes_g at
negative arguments; inverse Jacobi functions for m > 1 are real where the real inverse exists;
hypergeometric_u(a, b, 0); series dropped a term at negative order; options given to several library
functions were ignored when read from an options(...) argument. scorer_hi at large arguments and
alternating_factorial end; a library matrix of more than 2^26 entries declines at once.
Notebook. Plots in the notebook's PDF export are laid out for the page: their numbers print at the size of the
code beside them (the print layout is stored with the plot, about 5 KB each; old notebooks export as before). A
list of plots exports to a PDF with a print page each. Typing in a very long cell is about three times faster. The
language server knows every library's functions.