Electromagnetic compatibility is the ability of electrical equipment to operate as intended in its electromagnetic environment without causing unacceptable disturbance to other equipment. Emission and immunity are complementary questions: how strongly does a device disturb its surroundings, and how well does it tolerate disturbances arriving from them?

Interference requires a source, a coupling path and a susceptible receiver. A switching power converter may couple a disturbance into a sensor through shared supply impedance, capacitive coupling or radiated fields. In an illustrative circuit, a $0.2$ A switching-current change through a shared $0.1\Omega$ return impedance produces a $20$ mV reference shift, potentially much larger than the sensor signal. A balanced connection and finite common-mode rejection ratio help only to the extent that the disturbance remains common to the two inputs.

A useful test compares the sensor output with the converter switched on and off, then deliberately changes the coupling path or applies a calibrated disturbance. Antenna patterns, cable routing, grounding and frequency all enter the inference; passing one test configuration does not imply immunity to every disturbance.^[ref]

Pulse-code modulation represents a waveform by sampling its amplitude at specified times, assigning each sample to one of finitely many reconstruction levels, and coding the chosen level as a sequence of symbols. Sampling discretizes time; quantization discretizes amplitude; coding represents the selected level. These operations answer different questions about preserving a signal.

For an original example, divide the input range $[-1,1]$ volts into eight equal cells, each $\Delta =0.25$ volts wide, and reconstruct each cell by its midpoint. Label the cells in increasing order with three-bit binary words from 000 to 111. An input electric potential difference of $0.31$ volts lies in $[0.25,0.50)$, so its word is 101 and its reconstructed value is $0.375$ volts. The error is $0.065$ volts. Inside the specified range, the error magnitude is at most $\Delta /2=0.125$ volts; an out-of-range input instead introduces clipping.

At $8000$ samples per second, three bits per sample give a raw rate of $24000$ bits per second for one channel, before framing or other overhead. Increasing the sampling rate and increasing the number of quantization levels address different errors: more time samples alone do not remove amplitude rounding.

W. M. Goodall's 1947 paper described an experimental speech encoder and decoder, with binary words identifying sampled amplitude levels. Its listening comparisons examined the effect of code length on speech quality. The reported quality thresholds concerned that apparatus and signal conditions, rather than a universal bit depth for every recording.^[ref]

Encoding standard symbols also makes regeneration useful: a receiver can restore a perturbed pulse to a standard symbol if it still identifies the symbol correctly. This does not recover amplitude information discarded during quantization, and sufficiently large channel disturbances can still cause decoding errors.

Current sensing determines an electric current from a measurable response to that current. A shunt resistor converts it into an electric potential difference; magnetic methods can instead respond to the field associated with the current. The sensor, amplifier, reference, digitizer and conversion model jointly determine the inferred current.

A basic resistive signal chain iswhere $G$ is amplifier gain and $b$ is an output offset. Pete Semig's single-supply low-side reference design demonstrates current measurement with a sensing resistor and an instrumentation-amplifier readout.^[ref]

For an original example, choose $R_s=20\mathrm{m\Omega }$, $G=50$, $b=0.10\mathrm{V}$ and measured output1.10volts. The estimate is1ampere. Ignoring the offset would instead give1.10amperes. A1percent error in the product $G{R}_s$ produces a corresponding scale error, even if the displayed voltage is perfectly repeatable.

In low-side current sensing, the shunt sits between the load and its return reference; its burden can raise the load's local reference potential. In high-side current sensing, it sits toward the supply, so the readout must handle the relevant common-mode voltage. A small differential signal does not imply a small voltage of either input relative to the amplifier's own supply. The LTC6115 is an example of an amplifier designed for high-side current sensing.^[ref]

An analog-to-digital converter can record the amplified signal, but digitization does not remove sensor loading, gain error or aliasing. A useful measurement procedure records zero-current output, several reference currents, connection points and temperature, with enough bandwidth for the waveform of interest. Four-terminal sensing can reduce connection errors for a shunt; separate gain/offset calibration addresses different error sources. The desired time resolution, electrical loading and uncertainty must all be included in the current estimate.

An electric machine uses electromagnetism to transfer energy between electrical and mechanical forms, or between electrical circuits. An electric motor converts electrical input into mechanical output; an electrical generator converts mechanical input into electrical output. The wider machine category also includes the stationary electrical transformer. A rotating electric machine combines force on current-carrying conductors with electromagnetic induction; the same rotating conductors that develop torque can develop an opposing induced electric potential difference.^[ref]

For an original idealized direct current motor example, let $e=k_e\omega$, $\tau =k_{t}I$, and $V=e+IR$, where $e$ is the induced opposing electric potential difference, $\omega$ the angular velocity, $\tau$ the torque, and $R$ the winding electrical resistance. In coherent SI units, assume $k_e=k_t=0.1$, $\omega =100\mathrm{rad}{\mathrm{s}}^{-1}$, $I=2\mathrm{A}$, and $R=1\Omega$. Thenand the electromechanical conversion power balance isThe $20/24$ conversion fraction is about $83.3\%$ before friction and other losses. At rest the induced term vanishes, so applying the same $12\mathrm{V}$ gives $12\mathrm{A}$ in this simplified model. This makes the opposing induced electric potential difference experimentally visible: at fixed applied electric potential difference, a shaft-speed measurement and an electric current measurement distinguish a turning motor from its stalled state. Actual starting control and additional losses change the idealized numbers, not the need to balance electrical input, mechanical output, and heating.

An electric power system is an interconnected arrangement that produces, transfers, controls, and consumes electrical energy. Electricity generation supplies electrical power, electric power transmission carries it between regions, and electric power distribution delivers it to electrical loads. Electrical generators, electrical transformers, and transmission lines perform different parts of this task. The one-line apparatus diagram in the DOE handbook makes these connections concrete: it follows supplies through switching and protection equipment to consuming circuits.^[ref]

A useful calculation separates delivered power from conductor heating. Consider an idealized two-conductor direct current link delivering $P=10\mathrm{MW}$ at load electric potential difference $V=10\mathrm{kV}$, with total outgoing-plus-return electrical resistance $R=0.1\Omega$. Its electric current and resistive loss areDelivering the same power at $100\mathrm{kV}$ reduces the electric current to $100\mathrm{A}$ and the loss to $1\mathrm{kW}$. This is a fixed-load, fixed-resistance comparison; it omits converter loss, insulation constraints, and changes in conductor temperature. In an alternating current network, an electrical transformer changes the operating electric potential difference without itself generating the delivered energy.^[ref]

At each instant, generation, imports, and storage discharge must supply electrical load, exports, losses, and storage charging. A sudden imbalance can initially draw on stored energy; it cannot persist indefinitely without changing the operating state. This electric power balance during a demand step connects network calculations to control and to the smaller, deliberately bounded microgrid.

A power inverter converts electrical power from a direct current input to an alternating current output. Its switches reverse or modulate the connection to the DC supply, while its control and filtering determine the output waveform. It transfers energy from its input; the switching operation itself is not an energy source. The DOE overview explains the switching-and-filtering principle and distinguishes equipment designed for independent operation from equipment relying on a grid reference.^[ref]

An ideal H-bridge uses four controlled switches around a two-terminal load. One diagonal pair applies $+V_d$ across the load, and the opposite pair applies $-V_d$. In a stipulated square-wave mode with equal half-periods,Its root mean square magnitude is $V_d$, but it is not sinusoidal. Its Fourier series isThe fundamental RMS magnitude is $2\sqrt{2}{V}_d/\pi$. Thus a $48\mathrm{V}$ input gives a $48\mathrm{V}$ RMS ideal square-wave output, but only about $43.22\mathrm{V}$ RMS in the fundamental. Equal total RMS does not imply equal waveform or equal response of every electrical load.

For a $12\Omega$ purely resistive load directly across this ideal square-wave output, mean active power is $V_d^2/R=192\mathrm{W}$. The fundamental contributes $8V_d^2/({\pi }^{2}R)\approx 155.63\mathrm{W}$ and the other harmonic components supply the remainder. The harmonic active power sum accounts for that difference. Adding a filter changes the circuit; the directly connected resistor calculation should not then be reused unchanged.

Rapid switching also permits averaged control. If a short switching interval applies $+V_d$ for fraction $d$ and $-V_d$ for the rest, its average output is $(2d-1)V_d$. Choosing $d(t)=(1+m\sin \omega t)/2$, with $0\le m\le 1$, makes that local average sinusoidal. The actual waveform still contains switching components that the load and filter must handle.

A proposed low-voltage demonstration records DC input electric potential difference and electric current, output waveform and load electric current, comparing average input and output power. Output frequency, total RMS, fundamental RMS and waveform distortion answer different questions. For islanded microgrid operation, control of a usable local supply is an additional capability, not a consequence of DC-to-AC conversion alone.

The Java virtual machine is the abstract execution machine specified for Java class files and bytecode. It defines runtime data areas, stack frames, operand stacks, local variables and instruction behavior, while leaving implementation choices such as interpretation and just-in-time compilation open.^[ref]

For example, two integer loads followed by an integer-add instruction obtain operands from a stack frame and leave their sum on its operand stack. A method call establishes a new stack frame and later returns a result to its caller. These specified effects remain the contract even if a compiler keeps the values in native registers and emits one machine instruction. The virtual-machine interface therefore supports portable compiled code without requiring every implementation to execute bytecode one instruction at a time.

Classical computation processes information described by classical states and transition rules, in contrast with computations that use coherent quantum states. A digital classical model commonly represents data by bits; a Boolean circuit or Turing machine supplies rules for transforming those representations. A randomized algorithm adds random choices, while each particular execution still follows one sequence of classical states.

For example, an AND gate maps two input bits to one output bit, with output one only for input 11. A probabilistic selection between two inputs describes uncertainty about which classical state was used; it is not a coherent quantum superposition of those inputs. Stating the model separates the abstract information-processing rule from its physical implementation: ordinary electronic devices ultimately obey quantum physics even when their exposed computational behavior is classical.

Computational geometry designs and analyzes algorithms for geometric objects represented by coordinates or combinatorial structure. Problems include finding a convex hull, intersecting segments and locating points in a subdivision. The representation and numerical predicates are part of correctness, not merely implementation details.

For points $p,q,r$ in the plane, the orientation predicate uses the sign ofA positive sign places $r$ to the left of the directed line from $p$ to $q$; zero means collinearity. The points $(0,0),(1,0),(0,1)$ give a positive determinant. Near-collinear inputs can make round-off error reverse the sign and corrupt an algorithm's combinatorial decisions. Shewchuk's adaptive predicates increase arithmetic precision only as needed to certify the correct sign.^[ref]

Computational topology develops algorithms for topological questions about finite representations of spaces, such as simplicial complexes. It connects discrete data structures to invariants that ignore irrelevant geometric distortion. connected components and homology groups are examples of computable structural information.^[ref]

For a small simplicial homology calculation over a field, take three vertices joined by three edges in a cycle, without a filled face. The matrix of the first boundary operator has matrix rank two, so this linear map has a kernel of a linear map of dimension one; with no two-dimensional face, that cycle is not a boundary. Thus the first Betti number is one. Adding the filled triangular face makes the same cycle a boundary and reduces the first Betti number to zero. The vertex coordinates need not change: the encoded incidence structure determines whether the hole is present.

Information retrieval finds material relevant to an information need within a collection, commonly documents or passages. A query expresses part of that need; a retrieval system selects or ranks candidate material. Relevance is therefore part of the task, beyond simply checking whether a stored identifier matches.^[ref]

For example, suppose a query returns five documents, three relevant, while the collection contains six relevant documents altogether. Its precision is $3/5$, while its recall is $3/6$. Returning everything increases coverage but can overwhelm the user with irrelevant material. An inverted index accelerates term lookup; scoring and evaluation address which retrieved results are useful. Information retrieval consequently combines data structures, ranking algorithms and evidence about the user's task.

An application binary interface specifies the conventions that separately compiled program components must share to cooperate at the binary level. It includes a calling convention, representations and layout of data, and platform conventions for object files. A common source-language declaration alone does not ensure that the generated machine instructions agree on where arguments and results reside.

For example, Microsoft's x64 calling convention passes the first two integer arguments in the processor registers RCX and RDX, and returns a suitable scalar integer result in RAX.^[ref] A caller placing 7 and 5 in different processor registers from those expected by the called function can obtain the wrong result even if both sides describe an addition function. Matching the application binary interface allows a compiler and hand-written assembly language to cooperate; it does not require them to use the same internal implementation.

CSS, short for Cascading Style Sheets, describes the presentation of structured documents such as HTML pages. A CSS selector identifies elements to which a rule applies; a CSS declaration assigns a property and value. The CSS cascade resolves competing declarations, while some properties can inherit values from an element's parent.^[ref]

For example, consider the rules p { color: blue; } and .warning { color: red; }. A paragraph with class="warning" matches both CSS selectors. For these ordinary rules in the same styling context, the class selector has greater CSS specificity than the element selector, so the paragraph is red even if the blue rule is written later. A CSS cascade therefore does more than choose whichever declaration was encountered last.^[ref]

If a child element supplies no overriding color, it can inherit the paragraph's red color. That does not mean every property is inherited: properties define their own inheritance behavior. A width declaration such as width: 200px also concerns layout in CSS pixels; it does not promise a width of 200 physical display pixels.^[ref]

Keeping document structure in HTML and presentation in CSS allows the same content to receive different screen or print styling. The rules describe the desired presentation; the browser determines the layout and rendered image from the document, styles and output device.

HTML, short for HyperText Markup Language, is the document language used to structure web pages and applications. An HTML element expresses a role such as a paragraph, heading or emphasized passage. An HTML attribute supplies additional information about an element. A web browser parses the source into a Document Object Model that represents the resulting document structure.^[ref]

Consider <p class="warning">Save <em>your work</em>.</p>. The p element contains text and an em element; class="warning" is an attribute on the paragraph. The two elements form a nested structure. A CSS selector such as .warning can match that paragraph, while a CSS declaration chooses a presentation such as a red foreground color.

This distinction preserves useful meaning across presentations. A heading or emphasized passage remains structurally identified even if a CSS rule changes its size, color or typeface. The browser's default styling is one rendering of that structure, not the definition of its semantics.^[ref]

The serialized source and the live document also need not stay identical after parsing or later changes. In particular, writing text that looks like a start tag inside a quoted source-code example is different from creating an element in the parsed tree. The HTML parser has specified error recovery for malformed input, but that recovery does not make every apparent nesting in a source file a valid document structure.^[ref]

The Document Object Model represents a document as a DOM tree of DOM nodes and defines operations for examining and changing that tree. In an HTML document, elements and text are different node kinds. The model gives programs access to document structure independently of how a source file happened to spell that structure.^[ref]

For example, <p>Hello <em>world</em>.</p> creates a paragraph element with a text child, an emphasis element and another text child. The emphasis element itself contains the text world. Looking only for element children would miss the text nodes between and inside those elements.

A short JavaScript example constructs a paragraph directly:
const p = document.createElement("p"); p.textContent = "Hello"; document.body.appendChild(p);
The first operation creates an element, assigning textContent supplies its text, and appending it inserts it into the document. Creating an element alone does not place it on the page.^[ref]

The Document Object Model is also different from the geometry computed for display. A CSS declaration can change line wrapping or hide an element without deleting that element from the tree. Conversely, inserting a DOM node can alter the tree before or without making a visible change. Separating document structure from rendered boxes helps explain both programmatic editing and browser layout.

A web browser retrieves and presents web resources while providing an environment for interaction with them. For an HTML page, this includes parsing source into a Document Object Model, applying CSS, laying out the result and rendering it. Programs such as JavaScript can then change the document or respond to user actions.^[ref]

Consider a page containing <p id="notice">Waiting</p>. Its visible word begins as Waiting. A program can obtain that element and change its text to Ready; the Document Object Model and displayed page then change without requiring the original response bytes to change. The browser maintains a live document and rendering state, rather than repeatedly displaying an immutable picture of the source.

The visible area also has several coordinate scales. The layout viewport constrains layout, while the visual viewport describes the portion being viewed. Page zoom, pinch zoom and scrolling can change those relationships. This matters when a program places a label relative to a document element or interprets a pointer position.^[ref]

A URL identifies a resource to navigate to; a web search engine helps discover resources. A browser can present search results as a page, but navigating directly to a known address does not require a search query. The retrieval service and the software interpreting its pages remain separate parts of the system.

A markup language describes annotations embedded in or associated with document content. Its syntax distinguishes ordinary content from instructions identifying structure, meaning or presentation. HTML is a concrete example: a paragraph and an emphasized phrase can be represented explicitly instead of inferred from their appearance.^[ref]

For example, <p>Save <em>your work</em>.</p> identifies a paragraph containing an emphasized phrase. The angle-bracketed HTML tags mark elements; the words between them supply content. A web browser can build a Document Object Model from this source, then use CSS to choose its presentation. Changing the emphasis styling need not change which phrase is structurally emphasized.

The syntax needs a way to represent its own reserved characters as content. In HTML, < can represent a literal less-than character in text rather than starting a tag. A document explaining markup therefore needs to distinguish a displayed example from active markup that the parser interprets as document structure.^[ref]

A markup language is useful when documents must be transformed, searched or rendered according to their structure. It does not follow that every source annotation creates a visible box, or that a visible line break implies a particular structural element. Source syntax, the parsed document and the final rendering are related representations with different jobs.

JavaScript is a general-purpose programming language widely used to implement behavior in web applications. Its core language is standardized by ECMAScript; environments such as a web browser supply additional facilities, including access to the Document Object Model. A language expression and a browser operation consequently need not come from the same specification.^[ref]

For example, const squares = [1, 2, 3].map(x => x * x); produces an array containing 1, 4, 9. The expression supplies a function that the array's map operation applies to each entry. This computation does not require a page or a document tree. A browser expression such as document.createElement("p") instead uses the host's document interface.^[ref][ref]

Values and conversion rules matter. 1 + 2 evaluates to the number three, while "1" + 2 evaluates to the text "12" because that addition operation performs string concatenation. Reading the exact operand types explains the result more reliably than treating every plus sign as numerical addition.^[ref]

Lexical scope determines which binding an expression refers to from the surrounding program structure. In an interactive page, the language can use those bindings while reading or changing DOM nodes. Distinguishing the core language, its environment and the program's own data makes examples easier to transfer between browser and nonbrowser implementations.

An ablation study compares a computational system with variants in which a component, input feature or training choice is removed or replaced. In machine learning, the purpose is to investigate which parts contribute to the measured behavior under a specified training and evaluation procedure. A useful comparison names the change, the metric and the conditions held fixed.

A measured example appears in Attention Is All You Need by Ashish Vaswani and others. Its transformer model comparisons include removing label smoothing: the base system's development-set BLEU is $25.8$, versus $25.3$ without label smoothing. Meanwhile, wordpiece perplexity changes from $4.92$ to $4.67$, an improvement for that metric. The same removal can therefore improve one measurement and worsen another. These are English-to-German machine translation results on the named development set, not universal effects of label smoothing.^[ref]

A synthetic example shows why effects need not add. Suppose accuracy is $70\%$ with neither component A nor B, $80\%$ with A alone, $80\%$ with B alone, and $80\%$ with both. Removing either from the full system changes accuracy by zero percentage points. Adding either to the system with neither improves accuracy by ten points. The components provide redundant benefits: a zero removal effect does not imply that a component could never be useful. The statistical interaction is visible in $80-80-80+70=-10$ percentage points.

For an interpretable ablation study, record the data split, metric, training budget and any settings changed to keep the variant viable. Retraining a reduced artificial neural network and suppressing a component only during evaluation answer different questions. Repeated training runs help assess variability; one difference alone does not establish statistical reliability. Broader claims require further comparisons.

Machine translation uses a computational system to produce text in one human language from text in another. It must choose an output sequence that preserves relevant meaning while fitting the target language, rather than merely replacing each word independently. A transformer model is one architecture used for this task.^[ref]

A probabilistic formulation assigns conditional probabilities $p(y\mid x)$ to candidate output sequences $y$ for input $x$. A decoder searches for a suitable sequence, often using model probability together with additional search or length rules. For a deliberately simplified finite example, suppose three complete candidates have probabilities $0.55$, $0.30$ and $0.15$. Choosing the most probable gives the first candidate, but this is a model decision rather than a proof that its meaning is correct.

Context explains why word-by-word replacement is insufficient. The intended translation of an ambiguous word can depend on the surrounding sentence, and target word order can differ from source word order. A system can therefore produce a fluent sentence that omits or alters information; fluency and preservation of meaning require separate attention.

Training and evaluation also answer different questions. A neural language model can supply probabilities for sequences, while cross-entropy measures how much probability is assigned to observed text. BLEU instead compares output sequences with reference translations through matching n-grams and a length penalty. Reference overlap is useful for repeated comparisons, but alternative valid phrasing and context limit its interpretation as an assessment of any single sentence.^[ref]