Beautiful Math

Numbers begin as a way to remember how much. Place value and zero turn that memory into a language compact enough to build everything else.

Addition, subtraction, multiplication, and division are not rituals. They are transformations - ways of changing a quantity - and some of them can be undone.

A fraction is a number; a ratio is a comparison. Both speak the language of relative size - the same language probability and many ML metrics will speak later.

A variable is a name for a quantity that is not fixed yet. Naming the unknown is how relationships become thinkable - in equations, in science, and in machine learning features and parameters.

Decimals extend place value past the units place. Powers compress repeated multiplication and reveal how length, area, and volume scale. Estimation keeps both honest when exact digits are not the point.

Geometry begins when numbers gain place: a point has position, a line has direction, and an angle measures a turn. That is how quantity becomes something you can see.

Triangles are the simplest rigid polygon. Congruence says two shapes match by a rigid motion. That is the geometric version of equality.

Similar figures share shape under scaling. On a right triangle, that scaling intuition becomes the Pythagorean relation: a² + b² = c².

A circle is constant distance from a center. Area measures how much region a shape occupies. Together they show how continuous magnitude can still be compared and computed.

Rigid motions move figures without stretching them. Symmetry is invariance under a chosen motion. Together they turn congruence into verbs: slide, turn, flip.

Polygon area comes from decomposing into triangles. Prisms and cylinders extend area into volume by stacking. Surface measure asks how much skin wraps the solid.

An equation is not a puzzle with a secret answer key. It is a claim that two expressions name the same amount, and solving means finding when that claim is true.

Two unknowns need two independent relationships. A system is a set of truth claims that must hold together.

A function is a reliable machine: each allowed input gets exactly one output. That single rule is how models, formulas, and graphs stay coherent.

A graph turns a function into a picture: every input-output pair becomes a point. Descartes welded algebra to geometry so eyes could help algebra think.

An inequality is a claim about an ordered region, not a single number. Multiplying by a negative flips the order; intervals name the solution set.

A quadratic is a parabola in disguise: completing the square, the formula, and factoring are three views of the same structure. Polynomials extend the pattern.

Exponential growth multiplies; a logarithm undoes that multiplication into addition. Napier and Briggs built tables so hard products became easier sums.

Sequences list terms by a rule; series sum them. Linear and quadratic patterns jump off tables, and least squares fits a simple model to noisy data - algebra meeting measurement.

Trigonometry starts with measuring turns. Degrees and radians are two rulers for the same idea: how far you have rotated.

Sine, cosine, and tangent begin as ratios of sides in a right triangle. Same angle, same ratios - even when the triangle scales.

On a circle of radius 1, cosine and sine become the x and y coordinates of a point. That single picture extends trigonometry to every angle.

Arcsin, arccos, and arctan pick principal angles from ratios. The laws of sines and cosines extend triangle solving beyond right angles - astronomy's old craft in modern form.

Sine and cosine are periodic waves with amplitude, frequency, and phase. Identities are compressed geometry. Cosine similarity is a later preview once vectors arrive.

Calculus begins with a simple obsession: how fast is something changing right now? Slope of a curve is the geometric answer.

The derivative is a new function that reports the slope of the old function at each point. Leibniz's notation makes the ratio of tiny changes memorable.

If derivatives measure change, integrals accumulate it. Area under a rate curve recovers net change.

Limits ask what a function approaches. Continuity asks whether the approach matches the value. Critical points and linear approximations turn those ideas into search and error control.

In many variables, the gradient packs all partial slopes into one vector that points toward steepest increase. Learning often walks the opposite way.

A vector is magnitude with direction: an arrow in space, or an ordered list of numbers. That dual view is why linear algebra powers both geometry and data.

A matrix packages a linear transformation: multiply to stretch, rotate, shear, or project space in a uniform way.

The dot product measures alignment. From it you get lengths, angles, and projections - the geometry behind similarity.

A basis is a minimal coordinate system for a space. Dimension counts how many independent directions you need. Structure first, coordinates second.

Gaussian elimination is organized matrix moves that simplify linear systems. Overdetermined least squares is a projection story: the best fit is the shadow in the column space.

Ax = λx picks directions that a linear map only stretches or flips. Those eigen-directions are the map's simplest skeleton.

SVD factors any matrix as rotation, scale, rotation. PCA chooses variance-maximizing orthogonal directions - often computed via SVD of centered data.

Probability assigns weight to outcomes. In the finite equal-outcome world it is favorable over total. Deeper theories extend that idea to richer spaces.

A random variable names a numerical outcome of chance. Its distribution (pmf for discrete cases) says how probability mass sits on those numbers - Bernoulli, binomial, and uniform as first citizens.

New information reshapes probability. Conditional probability and Bayes' rule are the mathematics of updating.

Expectation is the probability-weighted average. Loss functions turn prediction errors into numbers; risk is expected loss.

Covariance and correlation measure linear co-movement. Sampling adds uncertainty about estimates. Calibration asks whether a stated probability matches long-run frequency.

Information measures surprise. Entropy is expected surprise under a distribution. Cross-entropy ties probability to the losses classifiers use.