Hartley's Law¶

Fix the maximum information a noiseless bandwidth-limited channel can carry as H = 2WT·log₂(M) — the product of a linear sample-rate factor (bandwidth) and a logarithmic resolution factor (distinguishable signal levels).

Core Idea¶

Hartley's law (Ralph Hartley, 1928) states that the maximum amount of information transmissible through a noiseless bandwidth-limited channel in time T is H = 2WT · log₂(M) bits, where W is the channel's bandwidth in hertz and M is the number of distinguishable signal levels the channel can support. The result rests on the Nyquist sampling rate: a bandwidth-W channel can carry 2W independent samples per second, and if each sample selects among M discrete levels, each sample encodes log₂(M) bits, so capacity scales as their product. The two structural commitments are that information capacity is bounded by a hard quantitative ceiling rather than being open-ended, and that the ceiling has two independent factors — the rate at which samples can be transmitted (set by bandwidth, a spectral resource) and the resolution with which each sample can be specified (set by the number of distinguishable levels) — with the resolution factor entering logarithmically, meaning doubling M adds only one bit per sample while doubling W doubles the sample rate entirely.

Hartley's law is the noiseless precursor and historical foundation of the Shannon-Hartley theorem: Shannon's 1948 extension replaces the deterministic M-levels with a noise-conditioned expression, substituting log₂(1 + S/N) for log₂(M) to handle additive Gaussian noise, but inherits the bandwidth-times-log structure intact. The law is still pedagogically central as the intuition-builder for channel capacity — it shows that any modulation-order increase (BPSK to QPSK to 16-QAM to 256-QAM) buys capacity logarithmically, while bandwidth widening buys it linearly, a structural asymmetry with direct consequences for every generation of wireless and wireline communication standard.

Structural Signature¶

Sig role-phrases:

the noiseless bandwidth-limited channel — the carrier the result describes, with bandwidth W (a spectral resource) in hertz
the signalling alphabet — M distinguishable signal levels each sample can select among
the Nyquist sample rate — 2W independent samples per second, the rate factor bandwidth sets
the per-sample information — log₂(M) bits, the resolution factor the level count sets
the capacity formula — H = 2WT·log₂(M) bits in time T, the product of rate and resolution
the bounded-ceiling guarantee — a hard upper bound read as an impossibility result: no coding scheme whatever can exceed it, so beat-the-capacity proposals are rejected on sight
the trade-off surface — at fixed throughput, capacity bought by widening bandwidth or raising modulation order, any point a design option
the linear-versus-logarithmic asymmetry — the load-bearing fact: doubling bandwidth doubles capacity while doubling the constellation adds only one bit per sample, settling which lever is binding
the noise omission and Shannon extension — the deterministic skeleton that the noisy result clothes by substituting log₂(1 + S/N) for log₂(M), keeping the bandwidth-times-log structure intact
the formula-not-mechanism status — exact only where its precondition holds; the substrate-spanning lesson is the general rate × log-resolution capacity pattern, not Hartley's channel formula

What It Is Not¶

Not the capacity of a real (noisy) channel. Hartley's law is exact only for a noiseless bandwidth-limited channel with a well-defined level count M; it deliberately omits noise. Reading H = 2WT·log₂(M) as the achievable rate of an actual link ignores the very thing it leaves out — the honest move is to read the noiseless ceiling for structural intuition, then switch to the Shannon form C = W·log₂(1 + S/N) where noise binds.
Not a law of physics or a causal mechanism. It is a derived capacity formula — a hard upper bound following from Nyquist's sample rate and the bit-count per M-level sample — not a force the channel exerts. It states a ceiling on what is transmissible, not a process that drives transmission.
Not a claim that bandwidth and signalling levels help equally. The two factors enter with different functional forms: bandwidth linearly (doubling W doubles capacity) but M only logarithmically (doubling the constellation adds one bit per sample). Treating them as symmetric levers misses the law's load-bearing asymmetry — a jump from 16-QAM to 256-QAM, sixteen times the levels, buys only a doubling of bits per sample.
Not the Shannon-Hartley theorem. Hartley's law is the noiseless precursor; Shannon's 1948 result extends it to noisy channels by substituting log₂(1 + S/N) for log₂(M) while inheriting the bandwidth-times-log skeleton. They share structure but the noisy capacity is the later, more general statement, not this one.
Not the Nyquist sampling theorem. Nyquist supplies the rate result Hartley builds on — that a bandwidth-W channel carries 2W independent samples per second. Hartley combines that sample rate with M-level resolution to get capacity; the sampling theorem is an input to the formula, not the formula itself.
Not applicable wherever there is a "channel" and "levels." The law is exact only where its precondition genuinely holds — a noiseless bandwidth-limited channel with a well-defined number of distinguishable levels. Where "distinguishable levels" is not actually well-defined, plugging a number into M yields arithmetic without meaning; the construct must fit its precondition, not be stretched to any information path.
Not the general rate × log-resolution capacity pattern. The substrate-spanning lesson — a system's information capacity is a bounded product of how fast it samples and how finely each sample resolves — recurs in working memory, psychophysics, and measurement, and is carried by that general pattern (a candidate bounded_capacity_with_rate_resolution_tradeoff). Hartley's law is the analogue-channel instance with its bandwidth-in-hertz, Nyquist 2W, and modulation-order cargo; the portable content is the general pattern, not this communications formula.

Scope of Application¶

Hartley's law lives across the channel-capacity and signalling subfields of information theory and communications engineering; it is exact wherever its precondition holds — a noiseless bandwidth-limited channel carrying information through M distinguishable levels — and its reach stops at that domain, the general capacity pattern belonging to its parents.

Telecommunications capacity-budget design. Capacity analysis for copper, coaxial, fibre, and wireless links, where the choice of modulation order (BPSK, QPSK, 16-QAM, 64-QAM, 256-QAM) is a direct application of the M-levels factor.
Modem and Wi-Fi standards. Each generation raises throughput by some mix of bandwidth widening (channel bonding) and constellation-density increase (higher-order QAM) — both Hartley ingredients, with the linear-versus-logarithmic asymmetry settling which lever pays.
Spectrum-allocation policy. The bandwidth dependence underwrites regulatory valuation of frequency bands: sub-6 GHz spectrum is prized for its propagation-and-bandwidth combination.
Source-coding and discrete-alphabet intuition. The log-of-distinguishable-levels factor supplies the binary bit-counting intuition for any discrete signalling alphabet.
Information-theory pedagogy. As the first capacity result and the noiseless skeleton, it is the intuition-builder taught ahead of the noisy-channel (Shannon–Hartley) extension that clothes it.
Channel-based substrates generally. The formula instantiates identically across wireline (telephone, DSL, cable, fibre), wireless (Wi-Fi, cellular, satellite), optical (free-space, fibre, deep-space), and storage channels (the channel running between write and read) — content areas of one substrate, not distinct domains.

Clarity¶

Hartley's law made information-carrying capacity a bounded quantity rather than an open-ended one. Before it, "send more information" read as a matter of cleverer coding or stronger signals; the law fixes a hard ceiling set by physical resources, so the question shifts from "how do we push more through?" to "what is this channel's capacity, and which factor is binding?" It does this by decomposing the ceiling into two independent contributions that practitioners had no reason to hold apart: the sample rate, fixed by bandwidth through Nyquist, and the resolution per sample, fixed by the number of distinguishable signal levels M. Naming both, and the formula that multiplies them, turns a vague sense that "wider channels and finer signalling both help" into a precise budget in which each factor's contribution can be read off and traded against the other.

The sharper distinction the law delivers is the asymmetry between those two factors: bandwidth enters linearly while M enters only logarithmically. That single structural fact reorganizes design thinking — doubling bandwidth doubles capacity, but doubling the constellation size adds just one bit per sample, so a jump from 16-QAM to 256-QAM buys far less than its fourfold increase in levels suggests. The practitioner can now ask the productive question directly: for a required throughput, is it cheaper to widen the channel (a spectral cost) or to raise the modulation order (paid for in signal-to-noise headroom)? The law also makes visible exactly what it leaves out — noise — which is why it stands as the noiseless skeleton that the Shannon-Hartley theorem later clothes by substituting log₂(1 + S/N) for log₂(M) while keeping the bandwidth-times-log structure intact.

Manages Complexity¶

Without the law, the throughput of a noiseless channel looks like an open-ended engineering problem entangled with the full apparatus of a given link — the modulation scheme, the pulse shapes, the coding, the spectral occupancy, the choice of carrier — and answering "how much can this carry?" or "how do I make it carry more?" means reasoning through that whole apparatus afresh for copper, for fibre, for a radio band, for each modulation order. Hartley's law collapses that sprawl to a single closed form, H = 2WT·log₂(M), governed by exactly two factors: the sample rate set by bandwidth and the resolution set by the number of distinguishable levels. The engineer no longer tracks the channel's full physical detail; he tracks W and M and reads the capacity ceiling straight off their product — and reads, just as directly, that no scheme whatever can beat it, so whole classes of "cleverer coding" proposals are excluded without examination. The two factors define a single trade-off surface: at a fixed required throughput, capacity can be bought by widening the channel or by raising the modulation order, and any point on that line is a design option. The law's structural asymmetry then fixes the branch the analyst reads off — bandwidth enters linearly while M enters only logarithmically, so doubling bandwidth doubles capacity while doubling the constellation adds a single bit per sample. That one fact tells the designer which lever is binding and which is nearly spent: a jump from 16-QAM to 256-QAM, a sixteenfold increase in levels, buys only a doubling of bits per sample, whereas channel bonding buys throughput in direct proportion. So the high-dimensional "design a higher-throughput link" question compresses to: locate W and M, read the ceiling, and decide which of the two factors to spend on — with the linear-versus-logarithmic asymmetry settling the qualitative answer before any detailed modulation work begins. The same two-parameter reduction is what the Shannon-Hartley extension inherits when it replaces M with a noise term, keeping the bandwidth-times-log skeleton and the trade-off it structures.

Abstract Reasoning¶

Hartley's law licenses a set of reasoning moves built on two structural facts: that capacity is a bounded quantity with a hard ceiling, and that the ceiling factors into a linear term (bandwidth, through the sample rate) and a logarithmic term (resolution, through the number of distinguishable levels).

Boundary-drawing — read the ceiling as an impossibility result, and exclude whole classes of proposal without examining them. The characteristic move treats H = 2WT·log₂(M) not as an estimate but as a hard upper bound: from W and M alone, the analyst computes the maximum bits a noiseless channel can carry in time T and infers that no scheme whatever — no cleverer coding, no pulse shape, no modulation trick — can exceed it. So a proposal claiming to beat the Hartley capacity for a given bandwidth and level count is rejected on sight, before its details are read, because it claims the impossible. The reasoning runs from the two physical parameters to a fixed wall, converting an open-ended "how much can we push through?" into a closed "what is the ceiling, and have we hit it?" — and the wall's existence is the first thing the law lets the engineer assert.

Diagnostic — decompose the ceiling into its two factors and identify which one is binding. Having a closed form, the analyst reads capacity as the product of two independent contributions — the sample rate set by bandwidth and the resolution set by M — and diagnoses which factor is limiting a given link. A channel far below its level-supported ceiling is bandwidth-bound (widen it); a channel already at high modulation order is resolution-bound (further levels buy almost nothing). The move is to attribute a throughput shortfall to a specific factor by reading where on the trade-off surface the link sits, rather than treating "low throughput" as an undifferentiated problem requiring a redesign of the whole apparatus.

Interventionist — choose between widening bandwidth and raising modulation order, and predict the gain from the linear-versus-logarithmic asymmetry. The law's sharpest move is quantitative comparison of the two levers, because they enter the formula with different functional forms. Doubling bandwidth doubles capacity (linear); doubling the constellation size adds only one bit per sample (logarithmic). So the analyst predicts, before any detailed modulation work, that channel bonding buys throughput in direct proportion while a jump up the QAM ladder yields diminishing returns — a move from 16-QAM to 256-QAM, a sixteenfold increase in levels, buying only a doubling of bits per sample. The intervention decision (spectral cost of widening versus the cost of finer signalling) is settled by reading which factor is nearly spent: when M is already high, the prediction is that further constellation density is a poor investment and bandwidth or independent streams must supply the throughput.

Order-of-events and extension — read the noiseless skeleton as the foundation the noisy result inherits, and know exactly what the law omits. The law makes visible precisely what it leaves out — noise — and the reasoning move is to treat its deterministic M-levels as the idealized skeleton that the noisy-channel extension clothes by substituting log₂(1 + S/N) for log₂(M) while keeping the bandwidth-times-log structure intact. So the analyst reasons in two stages: read the noiseless ceiling from W and M to get the structural intuition and the trade-off shape, then, where noise matters, replace the resolution term with the signal-to-noise expression to get the achievable capacity — recognizing that the form of the trade-off (linear bandwidth, logarithmic resolution) survives the substitution. The move predicts that any design conclusion drawn from the bandwidth-versus-resolution asymmetry in the noiseless case carries over to the noisy case, because the extension inherits the same two-factor skeleton.

Knowledge Transfer¶

Hartley's law is a quantitative result — a capacity formula and a hard upper bound — rather than a causal mechanism, so it transfers the way a theorem transfers: the construct applies literally wherever its precondition holds, namely a bandwidth-limited channel carrying information through M distinguishable signal levels. Within information theory and communications engineering that precondition is met across every channel-based substrate, so H = 2WT·log₂(M) and its trade-off structure apply identically — and not by analogy but by direct instantiation — to wireline (telephone, DSL, cable, fibre), wireless (Wi-Fi, cellular, satellite), optical (free-space, fibre, deep-space), and storage channels (where the "channel" runs between write and read, in time or across a platter). The full apparatus carries: the bounded-ceiling impossibility argument (no coding scheme beats the capacity), the two-factor decomposition (sample rate set by bandwidth via Nyquist; resolution set by M), and the load-bearing linear-versus-logarithmic asymmetry (doubling bandwidth doubles capacity; doubling the constellation adds one bit per sample) that settles the bandwidth-bonding-versus-higher-QAM design choice. The Shannon–Hartley extension inherits all of it by substituting log₂(1 + S/N) for log₂(M) to handle noise while keeping the bandwidth-times-log skeleton intact, so the noisy result is the same construct with the resolution term re-expressed. These are content areas of one substrate, and the formula is exact in each.

Because the construct is a formula, the boundary to mark is precondition-fit versus over-reading. The law is exact on a noiseless bandwidth-limited channel with a well-defined level count M; reading it as the achievable rate of a real link silently ignores the very thing it omits — noise — which is why the honest two-stage move is to read the noiseless ceiling for the structural intuition and then switch to the Shannon form where noise binds. And where "distinguishable levels" is not actually well-defined, plugging a number into M produces arithmetic without meaning. So the construct should be applied where its precondition genuinely holds and not stretched to channels it does not describe.

The genuinely cross-domain structural lesson is a different matter, and it belongs not to "Hartley's law" but to the more general patterns it instantiates — which do recur across information-bearing systems, under other names, as co-instances. Capacity as a bounded resource that factors into rate × log-resolution recurs in working memory (chunks × bits per chunk), perceptual discrimination (samples per second × just-noticeable-differences), measurement instruments (sample rate × dynamic range / quantization), and even transaction-throughput models (rate × value resolution) — pointing at a candidate parent prime one might call bounded_capacity_with_rate_resolution_tradeoff, of which Hartley's law would be the analogue-channel instance (a common structural parent to the domain-specific capacity primes attentional_capacity, absorptive_capacity, adaptive_capacity). And distinguishability as an information bottleneck — the logarithmic relation between the number of distinguishable states and information content — recurs across psychophysics (Weber–Fechner), measurement (ADC bits), and discrete-alphabet coding. Where the lesson needed elsewhere is "a system's information capacity is a bounded product of how fast it can sample and how finely each sample resolves," the construct to carry is that general rate × log-resolution pattern, not Hartley's channel-capacity formula, whose distinctive cargo (bandwidth W in hertz, Nyquist's 2W, modulation order, spectrum) is communications-specific. Hartley's law is the historically foundational, substrate-specific instance; the substrate-spanning content is the general capacity pattern it exemplifies — exactly the split drawn in Structural Core vs. Domain Accent.

Examples¶

Canonical¶

Take a noiseless telephone-grade channel of bandwidth W = 3000 Hz, let each transmitted sample select among M = 16 distinguishable amplitude levels, over T = 1 second. Nyquist gives 2W = 6000 independent samples per second, and each sample carries log₂(16) = 4 bits, so Hartley's law yields H = 2WT·log₂(M) = 6000 × 1 × 4 = 24,000 bits — a hard ceiling of 24 kbit/s. Now test the asymmetry. Doubling the alphabet to M = 32 levels raises each sample to log₂(32) = 5 bits, giving 30 kbit/s — just one extra bit per sample despite twice the levels. Doubling the bandwidth to 6000 Hz instead gives 12,000 samples per second at 4 bits each, or 48 kbit/s: the sample-rate lever doubles capacity where the resolution lever adds only a sliver.

Mapped back: The 3000 Hz link is the noiseless bandwidth-limited channel, M = 16 is the signalling alphabet, 2W = 6000 is the Nyquist sample rate, and log₂(16) = 4 bits is the per-sample information; their product is the capacity formula giving a bounded-ceiling guarantee of 24 kbit/s. The contrast between doubling M (+1 bit) and doubling W (×2) is the linear-versus-logarithmic asymmetry made numerical.

Applied / In Practice¶

Successive Wi-Fi standards are a live demonstration of Hartley's two levers. 802.11n used channels up to 40 MHz wide with 64-QAM (6 bits per symbol); 802.11ac widened channels to 80 and 160 MHz and pushed modulation to 256-QAM (8 bits/symbol); 802.11ax (Wi-Fi 6) reached 1024-QAM (10 bits/symbol). The asymmetry shows plainly: climbing the QAM ladder from 64 to 256 to 1024 — a sixteenfold rise in levels across two steps — added only 4 bits per symbol, while doubling channel width from 80 to 160 MHz roughly doubled the sample rate and hence throughput. This is why standards bodies lean hardest on bandwidth — channel bonding and the move into the 6 GHz band — for headline throughput, treating denser constellations as the secondary, SNR-hungry lever.

Mapped back: Channel width sets the Nyquist sample rate and the QAM order sets the signalling alphabet M, so the two knobs are exactly the trade-off surface. That wider channels double throughput while each QAM step adds a fixed few bits is the linear-versus-logarithmic asymmetry driving real engineering choices. And denser constellations being "SNR-hungry" is the noise omission the noiseless law brackets and the Shannon extension reintroduces.

Structural Tensions¶

T1: Hard ceiling versus noiseless idealization (a wall that is exact and fictional at once). The law's chief power is that H = 2WT·log₂(M) is a bound, not an estimate — it licenses rejecting any beat-the-capacity proposal on sight, without reading its details. The tension is that this impossibility force holds only on a channel that does not exist: a truly noiseless bandwidth-limited link. On any real link, noise makes the operative ceiling the Shannon value, so the very exactness that makes Hartley's wall so decisive is bought by omitting the one factor (noise) that sets a real channel's capacity. The bound is rigorous and inapplicable to actual hardware in the same breath — read as the achievable rate of a real modem it overstates capacity, yet read as a structural ceiling it is exactly right. The engineer must hold both: an impossibility result to reason with and an idealization never to compute a real budget from. Diagnostic: Is the Hartley ceiling being used as a structural upper bound and intuition, or wrongly as the achievable rate of a noisy real link that only Shannon's form describes?

T2: Two independent factors versus M set by noise (the independence the extension collapses). The formula's clarity comes from decomposing capacity into two factors practitioners had no reason to hold apart — sample rate (bandwidth) and resolution (the level count M) — presented as independent knobs on a trade-off surface. The tension is that this independence is an artifact of the noiseless idealization: on a real channel the number of distinguishable levels M is not a free design choice but is itself capped by the signal-to-noise ratio, which is exactly why Shannon replaces log₂(M) with log₂(1 + S/N). So the resolution lever a designer thinks she is turning freely is, physically, governed by noise power and transmit power, and pushing M up demands SNR headroom rather than being independent of it. Treating M as an autonomous knob invites constellation choices the channel's noise cannot actually support. Diagnostic: Is M being treated as a freely-chosen level count, or recognized as bounded by the SNR that the noiseless formula hides?

T3: The linear-bandwidth advantage versus the noise wider bandwidth admits (the asymmetry that erodes under noise). Hartley's load-bearing conclusion is that bandwidth enters linearly while resolution enters only logarithmically, so widening the channel doubles capacity where climbing the QAM ladder yields diminishing returns — which is why standards bodies lean hardest on channel bonding. The tension is that this clean asymmetry is a noiseless result: on a real channel a wider bandwidth also admits proportionally more noise power, lowering the SNR available per hertz, so the linear bandwidth gain is not the free lunch the noiseless law advertises. The design intuition that "bandwidth always beats modulation order" is directionally sound but quantitatively overstated once noise scales with W, and a naive reading can push spectral spending past where the extra noise cancels the extra samples. The asymmetry survives Shannon in form but with the bandwidth term's advantage partly eaten by the noise it lets in. Diagnostic: Is the linear bandwidth advantage being applied with the noise that wider bandwidth admits accounted for, or as the unqualified noiseless proportionality?

T4: Exact formula versus precondition-fit (arithmetic that can be meaningless). As a theorem, Hartley's law transfers by direct instantiation wherever its precondition holds — a bandwidth-limited channel with a well-defined level count M — and is exact there. The tension is that the formula is always computable even when its precondition fails: one can always plug a bandwidth and a level count into 2WT·log₂(M) and get a number, but where "distinguishable levels" is not genuinely well-defined, that number is arithmetic without meaning. The construct's formula-hood is thus a double edge — it invites application to any information path with a nameable "channel" and "levels," precisely because the calculation never errors out. The discipline the law demands is to verify precondition-fit before computing, an act the mechanical availability of the formula constantly tempts one to skip. Diagnostic: Does this channel genuinely have a bandwidth W in hertz and a well-defined count of distinguishable levels M, or is a number being plugged into M to manufacture a capacity figure with no referent?

T5: Foundational skeleton versus superseded formula (a law you reason with but rarely compute from). Hartley's law is historically foundational and pedagogically central — the intuition-builder taught ahead of the noisy-channel result — and its two-factor, bandwidth-times-log skeleton survives intact into Shannon–Hartley. The tension is that this enduring value is structural, not operational: essentially no real capacity budget is computed from the noiseless formula, because Shannon's C = W·log₂(1 + S/N) is the operative one for any actual link. So the law is simultaneously indispensable (its skeleton and asymmetry frame the whole problem) and superseded (its number is not the one you use). Its worth is as the scaffolding whose form the real result inherits, which means its lasting contribution is a way of thinking rather than a formula to plug into — a status easy to misread in either direction, over-computing with it or dismissing the foundation the operative result is built on. Diagnostic: Is the law being used for the structural intuition its skeleton supplies, or mistakenly as the computational tool the Shannon form has replaced?

T6: Autonomy versus reduction (a named channel formula or an instance of rate × log-resolution capacity). Hartley's law is a canonically foundational, substrate-specific result with proprietary cargo — bandwidth W in hertz, Nyquist's 2W, modulation order, spectrum, the Shannon extension. Yet its portable structure is not proprietary: a system's information capacity is a bounded product of how fast it samples and how finely each sample resolves is the general bounded_capacity_with_rate_resolution_tradeoff pattern (with the log-of-distinguishable-states relation a parent of its own), recurring as co-instances in working memory (chunks × bits per chunk), psychophysics (Weber–Fechner), and measurement (sample rate × ADC bits). The tension is between a standalone communications formula that earns its own impossibility argument and design trade-offs, and the recognition that everything travelling beyond channels belongs to that general rate × log-resolution parent. Calling a working-memory limit a "Hartley's law" borrows the channel formula for a capacity structure that owns itself. Diagnostic: Resolve toward the parent (bounded_capacity_with_rate_resolution_tradeoff) when the lesson is "capacity is rate times log-resolution" in any non-channel substrate; toward the named law when computing or bounding an actual bandwidth-limited channel with a defined level count.

Structural–Framed Character¶

Hartley's law sits toward the structural end of the spectrum — best read as mixed-structural, in the same family as the halting problem and the Hardy-Weinberg principle: an evaluatively neutral formal result whose portable skeleton travels as genuine cross-substrate co-instances, kept off the pole only by irreducibly communications-specific vocabulary.

On evaluative_weight it is at the structural extreme: H = 2WT·log₂(M) is a capacity ceiling, praising and blaming nothing — even its impossibility force ("no scheme beats this") is a neutral bound, not a censure. On human_practice_bound it is not practice-constituted: the law is a derived theorem (Nyquist's sample rate combined with per-sample bit-counting), a mathematical necessity that holds regardless of who states it and does not dissolve when the engineers leave — though, like the halting problem, it is a fact about an abstract construct (a channel) rather than a physical process nature simply runs, so it occupies the mathematician's substrate-independence rather than isostasy's literal observer-free-in-nature sense. On institutional_origin it is none: Hartley in 1928 derived the bound, he did not legislate it; the communications-engineering apparatus (modulation ladders, spectrum policy) is practice built around the theorem, not its source. On vocab_travels it fails in the domain-specific direction — bandwidth in hertz, the Nyquist 2W sample rate, modulation order, spectrum, the S/N extension are all communications furniture with no referent off a channel — which is the decisive framed-ward pull, and where "distinguishable levels" is not genuinely defined, plugging a number into M yields arithmetic without meaning. But on import_vs_recognize it patterns strongly structural: within channel-based substrates the formula applies by direct instantiation (wireline, wireless, optical, storage — exact, not analogy), and beyond channels the underlying rate × log-resolution capacity pattern recurs as genuine co-instances — working memory (chunks × bits per chunk), psychophysics (Weber–Fechner), measurement (sample rate × ADC bits) — recognized as the same structure rather than borrowed by metaphor.

The portable structural skeleton is bounded capacity as rate × log-resolution: a system's information capacity is a hard-bounded product of how fast it can sample and how finely each sample resolves, with the resolution factor entering logarithmically. That skeleton is substrate-portable and recurs as recognized co-instances across information-bearing systems, and it is precisely what Hartley's law instantiates from its umbrella (the candidate bounded_capacity_with_rate_resolution_tradeoff prime, with the log-of-distinguishable-states relation as a parent in its own right), not what makes "Hartley's law" itself travel: the cross-domain reach belongs to that general capacity pattern, of which the channel formula is the analogue-communications instance, while the bandwidth-in-hertz, Nyquist-2W, and modulation-order cargo stay home. Its character: an evaluatively neutral, non-institutional formal capacity result whose rate × log-resolution skeleton is a genuinely cross-substrate pattern it instantiates from its umbrella, kept mixed-structural rather than a free-floating prime by the communications-engineering vocabulary that pins the named formula — not the pattern beneath it — to its home domain.

Structural Core vs. Domain Accent¶

This section decides why Hartley's law is a domain-specific abstraction and not a prime, and it carries the case for its domain-specificity — there is no separate section for that.

What is skeletal (could lift toward a cross-domain prime). Strip the communications and a thin relational structure survives: a system's information capacity is a hard-bounded product of how fast it can sample and how finely each sample resolves, with the resolution factor entering logarithmically. The pieces that travel are abstract: a rate factor (samples per unit time), a resolution factor (distinguishable states per sample), a multiplicative capacity ceiling, and the load-bearing asymmetry that the rate enters linearly while resolution enters only as a logarithm. That skeleton — bounded capacity as rate × log-resolution — is genuinely substrate-portable, which is exactly why the entry names the candidate bounded_capacity_with_rate_resolution_tradeoff parent (with the log-of-distinguishable-states relation a parent in its own right, and the domain-specific capacity primes attentional_capacity, absorptive_capacity, and adaptive_capacity as siblings under it). It recurs as recognized co-instances — working memory (chunks × bits per chunk), psychophysics (Weber–Fechner), measurement (sample rate × ADC bits). But it is the core Hartley's law shares, not what makes it distinctive.

What is domain-bound. Almost everything that makes it Hartley's law in particular is communications-engineering furniture: the bandwidth W in hertz as the spectral resource; the Nyquist 2W sample-rate result it builds on; the signalling alphabet of M distinguishable levels and the modulation-order ladder (BPSK, QPSK, 16-QAM, 256-QAM); the spectrum whose regulatory value the bandwidth dependence underwrites; and the Shannon extension that clothes the noiseless skeleton by substituting log₂(1 + S/N) for log₂(M). The decisive test: remove the bandwidth-limited channel with its well-defined level count and there is no Hartley's law — where "distinguishable levels" is not genuinely well-defined, plugging a number into M yields arithmetic without meaning, computable but empty. The bandwidth-in-hertz and Nyquist-2W cargo, the part that makes it this formula, has no referent off a channel substrate.

Why this does not clear the prime bar. A prime is a relational structure whose vocabulary travels and whose transfer is recognition of the same mechanism, not analogy. Hartley's law's transfer is bimodal, with an instrument-flavored near side. Within channel-based substrates it applies by direct instantiation and exactly — wireline, wireless, optical, and storage channels are content areas of one substrate, so the bounded-ceiling impossibility argument, the two-factor decomposition, and the linear-versus-logarithmic asymmetry are recognized, not re-derived, and even the Shannon–Hartley extension is the same construct with the resolution term re-expressed. Beyond channels the named formula does not travel: calling a working-memory limit "a Hartley's law" borrows the channel formula for a capacity structure that owns itself. What genuinely recurs there is the rate × log-resolution pattern, carried as co-instances by the parent. So when the bare structural lesson — capacity is a bounded product of sampling rate and log-resolution — is needed cross-domain, it is already supplied, in more general form, by the parent bounded_capacity_with_rate_resolution_tradeoff (and the log-of-distinguishable-states relation beneath it). The cross-domain reach belongs to that parent, of which Hartley's law is the analogue-communications instance; "Hartley's law," as named, carries channel baggage — bandwidth in hertz, Nyquist's 2W, modulation order, spectrum — that does not and should not travel.

Relationships to Other Abstractions¶

Current abstraction Hartley's Law Domain-specific

Parents (1) — more general patterns this builds on

Hartley's Law is a kind of Channel Capacity Prime

Hartley's Law is Channel Capacity specialized to a noiseless bandwidth-limited channel whose ceiling is the product of signaling opportunities and the logarithm of distinguishable levels.

Hierarchy path (1) — routes to 1 parentless root

Hartley's Law → Channel Capacity

Not to Be Confused With¶

Shannon–Hartley theorem. The noisy-channel successor, C = W·log₂(1 + S/N), which extends Hartley by replacing the deterministic level count M with a noise-conditioned term to give the capacity of a real (additive-Gaussian-noise) channel. Hartley's law is the noiseless precursor; Shannon–Hartley is the operative formula for any actual link. They share the bandwidth-times-log skeleton but differ in whether noise is present. Tell: is the resolution term a free level count log₂(M) on a noiseless channel (Hartley's law), or log₂(1 + S/N) set by the noise floor of a real channel (Shannon–Hartley)?
Nyquist sampling theorem. The result that a bandwidth-W channel carries 2W independent samples per second (and that a signal must be sampled at twice its highest frequency to be reconstructed). Nyquist supplies the rate factor Hartley's law builds on; it is an input to the capacity formula, not the formula itself. Nyquist says how many samples; Hartley multiplies that by the per-sample resolution to get capacity. Tell: is the claim about how many independent samples a bandwidth admits (Nyquist), or about total information capacity combining that sample rate with M-level resolution (Hartley)?
Shannon's source coding theorem (entropy limit). The source-side result that a source's output can be compressed to, but not below, its entropy in bits per symbol. This bounds how few bits are needed to represent a source; Hartley's law bounds how many bits a channel can carry. One is compression, the other transmission — the two ends of a communication system. Tell: is the limit about the minimum bits to encode a source losslessly (source coding / entropy), or the maximum bits a channel can transmit (Hartley's law)?
Hartley function / Hartley entropy / the hartley (unit). Ralph Hartley's other namesakes: the Hartley entropy H₀ = log|X| (the log of the number of possible states, a max-entropy measure), and the hartley, the unit of information for base-10 logarithms. The per-sample log₂(M) in the law is Hartley's information measure applied to a channel alphabet, but "Hartley's law" is the full channel-capacity result (rate × resolution × time), not the bare log-of-states measure or the unit. Tell: are you naming a measure of information content as the log of a state count, or the unit for it (Hartley function / hartley), versus the channel-capacity formula H = 2WT·log₂(M) (Hartley's law)?
Bounded capacity as rate × log-resolution (umbrella). The substrate-neutral pattern Hartley's law instantiates — a system's information capacity is a hard-bounded product of how fast it samples and how finely each sample resolves, resolution entering logarithmically — recurring as co-instances in working memory (chunks × bits per chunk), psychophysics (Weber–Fechner), and measurement (sample rate × ADC bits). The umbrella carries the cross-substrate lesson; Hartley's law adds the bandwidth-in-hertz, Nyquist-2W, and modulation-order cargo that stay home. Tell: strip away the channel, bandwidth, and level count and what remains is "capacity is rate times log-resolution" — the parent pattern, not Hartley's channel formula. (Treated fully in a later section.)

Neighborhood in Abstraction Space¶

Hartley's Law sits in a sparse region of the domain-specific corpus (90^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (309 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12