STRATA
Skip to study
XLVII

Connections between
fields of research

Compare shared sources, mathematical structures, and the measurements needed to test a proposed connection.

Find a connection by name
B01

Renormalization-group flow

When fine-scale degrees of freedom are averaged out, which parameters still predict the same observable? Compare the error at successive scales and identify the range over which the effective model holds.

Sources for this comparison

Formal method with a domain-specific model

Renormalization compares descriptions made at different resolutions while holding a chosen observable fixed. The transferable operation is to integrate out or average over details, then change the effective parameters so the retained description remains useful. A successful transformation in a magnetic model does not automatically supply the correct transformation for a biological population or the early Universe. Each application needs its own variables, dynamics and error estimate.

Averaging can preserve a mean while destroying a correlation.

Use the constructed sequence 1, −1, 1, −1. Its mean is zero, and averaging adjacent pairs produces the shorter sequence 0, 0 with the same mean. However, the nearest-neighbor products in the original sequence are all −1, whereas the coarse sequence has product zero. A model selected only to reproduce the mean would pass; a model asked to reproduce the short-distance correlation would fail. Repeat the transformation at several block sizes and plot the error for each declared observable.

AspectFirst domainSecond domain
Retained quantityFine description: individual values and their spatial relationships.Coarse description: block variables and selected observables.
Parameter meaningMicroscopic parameters describe the original interactions.Effective parameters depend on resolution and the transformation.
Acceptance testMeasure the target observable in the fine model.Compare the same observable, units and uncertainty after coarse-graining.

Take the comparison further.

Which observable must survive the first reduction in resolution, and what maximum error will make that reduction useful?

  1. The Renormalization Group and Critical Phenomena

    Kenneth G. Wilson / Nobel Prize · Original Nobel lecture on scale transformations and critical phenomena.

  2. Entanglement renormalization

    Guifré Vidal · Original proposal to reduce short-range entanglement before multiscale truncation.

B02

Holography and AdS/CFT

Which boundary observable reproduces a bulk prediction, and in what limit?

Sources for this comparison

Conjectured duality with calculable consistency checks

AdS/CFT relates specified gravitational theories in an anti-de Sitter bulk to specified quantum theories on their boundary. It is more demanding than a resemblance between pictures: a dictionary must identify corresponding states, parameters and observables. Maldacena’s proposal and later entanglement calculations provide concrete checks in controlled limits. Extending those results to our cosmological spacetime is a separate problem.

An interval-entropy change can be computed on both sides.

For the vacuum interval in a two-dimensional conformal field theory, doubling the interval length changes the universal entropy term by (c/3)·ln 2. In the corresponding semiclassical AdS3 calculation, the geodesic-length change is 2R·ln 2 and the area rule gives ΔS = R·ln 2/(2G). With the dictionary c = 3R/(2G), the expressions agree. Choosing c = 300 for illustration gives 100·ln 2, about 69.3 nats. The check requires the specified state, regulator treatment and classical bulk limit; it is not a measurement of the entropy of the observable Universe.

AspectFirst domainSecond domain
QuantityBoundary: a change in interval entanglement entropy.Bulk: a change in the associated extremal-geodesic length.
Parameter dictionaryThe central charge c specifies the boundary normalization.The ratio of curvature radius R to Newton constant G fixes the matching normalization.
Scope of agreementThe selected conformal theory and state are explicit.The gravitational description uses its corresponding limit and geometry.

Take the comparison further.

Which correction appears first when the classical bulk approximation is relaxed, and can the same correction be identified in the boundary calculation?

  1. The Large N Limit of Superconformal Field Theories and Supergravity

    Juan M. Maldacena · Original large-N duality proposal for specified theories and spacetime backgrounds.

  2. Holographic Derivation of Entanglement Entropy from AdS/CFT

    Shinsei Ryu and Tadashi Takayanagi · Original minimal-area proposal and its AdS3/CFT2 interval-entropy check in the stated limit.

B03

Entropy and information

What is being counted in each definition of entropy? Follow the change from messages to physical states and identify the assumptions behind each expression.

Sources for this comparison

Formal relation with different physical interpretations

Shannon entropy measures uncertainty in a specified distribution of messages or outcomes. Thermodynamic entropy requires a physical state description and a choice of macroscopic constraints. Their related mathematical forms become useful when the probabilities, units and physical operations are stated. A numerical resemblance alone does not turn the meaning of a sentence into heat, nor establish a general exchange rate between knowledge and energy.

One binary distribution supports two different questions.

An unbiased bit has entropy 1 bit; a bit with probabilities 0.9 and 0.1 has entropy about 0.469 bits. Those values describe the chosen distribution. If an initially unknown, unbiased physical bit is erased in an ideal isothermal setting, the Landauer bound is kB·T·ln 2. At 300 K this is about 2.87 × 10⁻²¹ joules per erased bit. The bound concerns the physical reset operation and its assumptions. It is not the measured energy cost of a laptop instruction, and a reversible logical operation is a different case.

AspectFirst domainSecond domain
Object being describedCommunication: a probability distribution over messages.Thermodynamics: physical states and the constraints that define an ensemble.
UnitsBase-two information entropy is measured in bits.Physical entropy has units of joules per kelvin; heat has units of joules.
Useful testEstimate the source probabilities and achieved code length.Specify the physical cycle and measure its energy and entropy transfers.

Take the comparison further.

What is being reset in the proposed experiment, and which part of its measured dissipation can be separated from the ideal erasure bound?

  1. A Mathematical Theory of Communication

    Claude E. Shannon / Bell Labs · 1948 communication model, entropy and coding framework.

  2. Notes on Landauer’s principle, reversible computation, and Maxwell’s Demon

    Charles H. Bennett / IBM Research · Author’s analysis of logically irreversible erasure and compensating entropy.

B04

Fluctuation to structure

How accurately can present-day structure constrain the fluctuations from which it developed?

Sources for this comparison

Empirical inference through a dynamical model

Cosmological structure connects observations at different times through a model of growth, matter content and measurement. A present-day map is not a direct photograph of an initial fluctuation. Planck’s parameter estimates illustrate how detailed data constrain a specified cosmological model; changing the model can change the inferred initial conditions. The bridge is an inverse problem whose uncertainty includes both measurement and evolution.

The same observed amplitude can imply different initial amplitudes.

In a deliberately simplified linear model, let the observed amplitude be δobs = D·δinitial, where D is a known growth factor. An observation of 0.6 implies an initial amplitude of 0.3 if D = 2, but 0.2 if D = 3. Measuring the final value more precisely cannot by itself decide which growth history was correct. In an actual cosmological analysis, scale-dependent transfer functions, selection effects and nonlinear evolution add further structure. Use the toy example to display the degeneracy, then identify which additional observation constrains D.

AspectFirst domainSecond domain
Starting informationInitial-condition model: a distribution of fluctuations.Observation: a sky map or catalog filtered by an instrument and a selection process.
Connecting operationForward evolution predicts a later statistical observable.Inference searches for initial conditions and parameters compatible with that observable.
UncertaintyAssumptions about growth and the underlying cosmology matter.Noise, foregrounds, incomplete coverage and selection also matter.

Take the comparison further.

Which second observable would distinguish two growth histories that fit the same final amplitude?

  1. Planck 2018 results. VI. Cosmological parameters

    Planck Collaboration · Final-mission cosmological inference under stated model assumptions.

  2. Evaluation of measurement data—Guide to the expression of uncertainty in measurement

    Joint Committee for Guides in Metrology / BIPM · JCGM 100:2008, reporting a result with a model and evaluated uncertainty.

B05

Decoherence and classical descriptions

Which features of a system’s interaction with its environment determine how quickly measurable interference fades?

Sources for this comparison

Empirical mechanism with a specified open-system model

Decoherence describes how interaction with an environment suppresses observable interference in a reduced description of a system. It explains why some superpositions become difficult to observe without making every philosophical interpretation equivalent. A proposed connection to consciousness or anomalous cognition would need its own measurable coupling and evidence. The presence of those topics in the same research map supplies a question, not support for that connection.

Interference can fade while the outcome probabilities stay fixed.

Take a two-state toy density matrix with diagonal entries 1/2 and off-diagonal entries (1/2)·exp(−t/τ). The populations remain equal, while the coherence relative to its initial value falls to about 0.368 at t = τ and 0.050 at t = 3τ. An interference measurement changes even though a measurement that reads only the two populations does not. The exponential decay is an assumed model; a real experiment must estimate its rate and test whether that form fits.

AspectFirst domainSecond domain
What is observedPopulation measurement: the probabilities of the selected basis outcomes.Interference measurement: sensitivity to relative coherence.
Model requirementA closed-system model can preserve phase relationships.An open-system description includes an environment and a coupling.
Evidence boundaryA fitted decoherence rate describes a tested physical setting.A claim about cognition requires independent operational definitions and observations.

Take the comparison further.

Which measurement distinguishes loss of coherence from an ordinary change in population, and what environment variable can be varied independently?

  1. Decoherence and the transition from quantum to classical—REVISITED

    Wojciech H. Zurek · Author’s account of environment-induced decoherence; not evidence for anomalous cognition.

B06

Tunneling across applications

How do barrier height, width and particle mass enter the tunneling rate in each application?

Sources for this comparison

Shared quantum mechanism with application-specific evidence

Tunneling is a quantum mechanism that can enter different physical and chemical settings, but the rate in a working system depends on its potential, temperature, geometry and competing pathways. An enzyme isotope experiment can support a particular hydrogen-transfer mechanism. It does not establish that every mutation, biochemical effect or condensed-matter observation has the same rate-limiting cause.

A barrier model reveals why width and mass matter differently.

For an illustrative rectangular barrier, use T ≈ exp(−2a·sqrt(2m(V−E))/ℏ). If the exponent is 4, the transmission estimate is exp(−4), about 0.0183. Doubling the barrier width changes it to exp(−8), about 0.000335. Doubling the mass while keeping the other quantities fixed gives exp(−4√2), about 0.00349. This approximation is a teaching model. Interpreting an enzyme’s isotope effect requires additional evidence about geometry, temperature dependence and other changes introduced by isotope substitution.

AspectFirst domainSecond domain
Common mathematical featureSimple barrier: transmission depends exponentially on an action-like quantity.Chemical system: quantum nuclear motion can contribute to a reaction pathway.
What must be measuredPotential shape, energy and incident particles define the model.Reaction rates, isotope effects and structural controls constrain the mechanism.
Transfer boundaryA rectangular barrier isolates a dependency.A biological rate combines several coupled steps and environmental effects.

Take the comparison further.

Which competing mechanism predicts a different temperature or isotope dependence, and what experiment would separate the predictions?

  1. Nature of hydrogen transfer in soybean lipoxygenase 1: separation of primary and secondary isotope effects

    K. W. Rickert and J. P. Klinman / Biochemistry · Original enzyme isotope-effect experiment; evidence is specific to the tested mechanism.

B07

Universality and critical phenomena

Which systems share a critical exponent after the measurements are scaled to comparable units?

Sources for this comparison

Formal universality tested by empirical scaling

Universality concerns shared large-scale behavior near a critical point under specified conditions, even when microscopic details differ. It is assessed through exponents, scaling functions and their uncertainty. A network picture or a branching texture can resemble a critical system without belonging to the same universality class. Dimensionality, symmetry, interaction range and the measured order parameter must remain visible.

A predicted exponent produces a numerical ratio to test.

Suppose an order parameter follows M = A·|t|^β near a transition, where t is a dimensionless distance from the critical point. If β = 1/8, reducing |t| from 0.01 to 0.0001 predicts M2/M1 = (0.01)^(1/8), about 0.562. This constructed scaling check cancels the unknown amplitude A. It is useful only inside the scaling regime; finite sample size, an incorrectly located critical point or corrections to scaling can change the apparent exponent.

AspectFirst domainSecond domain
Microscopic comparisonSystem A has its own particles, lattice or local rules.System B can have different microscopic details.
Required matchEstimate the exponent and scaling regime independently.Use comparable definitions, dimensions and control parameters.
Failure testA fitted straight line may reflect a short data range.Collapse and exponent estimates should survive changes in size and fitting range.

Take the comparison further.

Does the apparent shared exponent persist when the system size doubles and the fitting interval is narrowed toward the critical point?

  1. The Renormalization Group and Critical Phenomena

    Kenneth G. Wilson / Nobel Prize · Original Nobel lecture on scale transformations and critical phenomena.

B08

Computation as common denominator

What would a physical implementation need to reproduce beyond the output of its simulation?

Sources for this comparison

Formal computation with an additional physical implementation question

A computational model specifies which transformations a procedure performs. A physical implementation must also provide states, timing, energy, error control and a way to read the result. Turing’s formal machine and analyses of reversible computation make different parts of that problem explicit. Demonstrating that a model can simulate an observed sequence does not establish that nature uses that implementation or that all of its hidden states are physically real.

The same truth table can have different physical costs.

Consider a one-bit operation. NOT maps 0 to 1 and 1 to 0, so the input can be recovered from the output. RESET maps both 0 and 1 to 0, so the output alone cannot identify the input. A program can express both in one line, yet their logical reversibility differs. A physical experiment must specify how the bit is represented and what happens to the discarded information. Counting source-code lines would miss the distinction entirely.

AspectFirst domainSecond domain
Formal specificationA state-transition rule defines the input–output relation.A device must realize distinguishable states and transitions.
CorrectnessProof or exhaustive checking can establish a finite rule’s behavior.Measurement must establish that the device follows the rule within an error tolerance.
Resource accountingAbstract steps describe one computational cost.Time, memory, dissipation and readout describe implementation costs.

Take the comparison further.

Which physical observation could distinguish the proposed implementation from another machine that produces the same output sequence?

  1. On Computable Numbers, with an Application to the Entscheidungsproblem

    Alan M. Turing / original paper hosted by University of Southern California · Original 1936 computation paper; no inference that every physical question is computable.

  2. Notes on Landauer’s principle, reversible computation, and Maxwell’s Demon

    Charles H. Bennett / IBM Research · Author’s analysis of logically irreversible erasure and compensating entropy.

B09

Prediction, coding, and learning

When does shorter code length predict better classification on new examples?

Sources for this comparison

Formal coding relation with an empirical learning test

A probabilistic predictor assigns a code length through the negative logarithm of its assigned probability. This creates a precise connection between prediction and compression. Classification is an additional task: a model that compresses familiar training data well may exploit identifiers, repeated passages or leakage rather than features that generalize. Historical compression classifiers provide an experiment to revisit, not a promise that the smallest file always identifies the correct class.

A short code is useful only when it predicts unseen symbols.

For a constructed ten-symbol sequence containing nine zeros and one one, a fixed predictor with p(0) = 0.9 assigns about 4.69 bits of ideal sequence code length. A p(0) = 0.5 predictor assigns 10 bits. If the biased probability was chosen after seeing the test sequence, the comparison is misleading. Estimate the predictor on a separate training set, include any model-description overhead relevant to the comparison, and evaluate new sequences. For classification, compare class-conditional code lengths together with class priors.

AspectFirst domainSecond domain
Optimization targetCompression minimizes a description length under a coding scheme.Classification minimizes a declared predictive loss on unseen labels.
Shared quantityA probability model yields −log2 p(x) bits in an ideal code.The same probability contributes to a predictive score.
Necessary controlInclude model and metadata costs when comparing representations.Separate training and test data, including near-duplicate documents.

Take the comparison further.

Does the code-length advantage survive a test set separated by source and date, after the model’s storage overhead is included?

  1. A Mathematical Theory of Communication

    Claude E. Shannon / Bell Labs · 1948 communication model, entropy and coding framework.

  2. Data Compression Using Dynamic Markov Modelling

    G. V. Cormack and R. N. S. Horspool / The Computer Journal · Original December 1987 paper abstract and bibliographic record; adaptive Markov modelling with arithmetic coding.

  3. Spam Filtering Using Statistical Data Compression Models

    Andrej Bratko and colleagues / Journal of Machine Learning Research · Original 2006 compression-based classification study; historical benchmark, not a current performance guarantee.

B10

Bayesian updating and information gain

Which measurement changes the posterior most under each plausible prior?

Sources for this comparison

Formal probability update with assumption-sensitive evidence

Bayesian updating combines a prior with a likelihood and normalizes the result. The rule transfers across disciplines, while the credibility of its inputs does not transfer automatically. Experimental design adds another question: which available measurement is expected to reduce uncertainty, averaged over the outcomes it may produce? A dramatic result after the fact is not necessarily the experiment with the greatest expected information gain.

A positive signal means different things at different base rates.

In a constructed quality-control test, 1 percent of items are defective. The test flags 90 percent of defects and incorrectly flags 5 percent of good items. Among 10,000 items, the expected counts are 90 true flags and 495 false flags. The probability of a defect given a flag is therefore 90/585, about 15.4 percent. If the defect rate were 10 percent with the same test behavior, that posterior would be about 66.7 percent. These are assumed rates for arithmetic, not measured performance.

AspectFirst domainSecond domain
PriorOne population begins with a 1 percent defect rate.Another begins with a 10 percent defect rate.
LikelihoodThe example holds sensitivity and false-positive rate fixed.A real transfer must check whether those rates remain valid.
DecisionA posterior describes uncertainty about an item.A release or review decision also requires costs and consequences.

Take the comparison further.

Which of the assumed rates is least well measured, and would improving that estimate change the next review decision?

  1. An Essay towards Solving a Problem in the Doctrine of Chances

    Thomas Bayes, communicated by Richard Price / Royal Society archive · Original 1763 article; worked diagnostic arithmetic in this seed is newly constructed.

  2. On Information and Sufficiency

    Solomon Kullback and Richard A. Leibler / Annals of Mathematical Statistics · Original divergence paper. DOI and bibliographic metadata verified; publisher text access was limited.

B11

Feedback, control, and agency

Which variable does a feedback loop regulate, and how does the response change when information arrives late?

Sources for this comparison

Formal control method with implementation-specific dynamics

Feedback compares a measured state with a target and uses that difference to change an action. The useful bridge among machines, organisms and software agents is a defined loop: state, sensor, controller, actuator and delay. The word agency adds questions of who sets the target and can revise it. Quantum control also has measurement back-action to account for; a familiar block diagram does not remove that physical difference.

A one-step delay can turn immediate correction into oscillation.

In a toy discrete controller with no delay, let error evolve as e(t+1) = (1−k)e(t). Setting k = 1 removes the error in one step. With a one-step delay, use e(t+1) = e(t)−k·e(t−1). With k = 1 and starting errors 1, 1, the sequence becomes 1, 1, 0, −1, −1, 0, 1 and repeats. The same nominal gain now produces persistent oscillation. This is a defined mathematical example, not a controller setting for physical equipment.

AspectFirst domainSecond domain
Observed variableA sensor provides a selected measurement of the system.The actual state may contain unmeasured variables.
TimeAn immediate correction uses the current error in the toy model.A delayed correction can act on a state that has already changed.
AuthorityA controller follows a stated target and update rule.A responsible operator decides when the target or rule must change.

Take the comparison further.

How does the response change when the measured delay is included, and which gain range remains stable under that delay?

  1. Feedback Systems: An Introduction for Scientists and Engineers

    Karl J. Åström and Richard M. Murray / Caltech author archive · Author-published control-system exposition, modelling assumptions and stability.

B12

Coarse-graining and emergence

Which fine-scale details can be discarded while preserving the behavior measured at a larger scale?

Sources for this comparison

Method transfer constrained by the retained observables

Coarse-graining replaces many details with fewer variables. Emergence describes behavior that becomes meaningful or tractable at that larger scale. The connection is useful when a reduced model predicts a specified observation and its error can be checked. It is incomplete when a visually persuasive aggregate hides correlations or nonlinear effects needed for the prediction. Different fields can share the reduction strategy while requiring different closure assumptions.

Averaging before a nonlinear reaction can change the predicted rate.

Take two equal-volume cells with concentrations 0 and 2 in arbitrary units. Suppose the local reaction rate is k·c². The actual average rate is (0 + 4k)/2 = 2k. A model that replaces both concentrations with their mean, c = 1, predicts only k. Retaining the variance repairs this particular loss because E[c²] = E[c]² + Var(c). The example shows exactly which missing statistic matters for the assumed reaction; it does not prove that one extra statistic closes every biological or physical model.

AspectFirst domainSecond domain
Fine modelLocal concentrations determine local nonlinear rates.The aggregate model initially retains only their mean.
Lost informationSpatial variation contributes to the average of c².The square of the mean does not contain that contribution.
RepairCompute the omitted variance from the fine description.Carry it as an additional variable or bound the resulting error.

Take the comparison further.

Which discarded moment or correlation makes the largest contribution to the reduced model’s prediction error?

  1. The Renormalization Group and Critical Phenomena

    Kenneth G. Wilson / Nobel Prize · Original Nobel lecture on scale transformations and critical phenomena.

  2. The Chemical Basis of Morphogenesis

    Alan M. Turing / Royal Society paper hosted by MIT · Original reaction–diffusion instability model.

B13

Symmetry and conservation

Which continuous symmetry gives rise to the conserved quantity in each system?

Sources for this comparison

Formal theorem under explicit symmetry assumptions

Noether’s variational theorem connects continuous symmetries of an action with conservation laws under its stated conditions. The connection is exact enough to calculate, while its application still requires the correct action, boundary conditions and definition of the conserved quantity. In general relativity, a local conservation statement does not by itself provide a single globally conserved energy for every spacetime. The geometry and its symmetries matter.

A fixed spring conserves energy while changing momentum.

For a toy oscillator with mass m = 1 and spring constant k = 4, the energy is E = v²/2 + 2x². Starting at x = 1 with v = 0 gives E = 2. At x = 0, conservation requires |v| = 2. The fixed potential has no explicit time dependence, but it is not invariant under arbitrary spatial translation; momentum changes under the spring force. If the spring constant is deliberately varied with time, an external agent can exchange energy with the oscillator.

AspectFirst domainSecond domain
TransformationTime translation tests whether the action changes when the time origin shifts.Spatial translation tests whether moving the configuration changes the action.
Conserved quantityTime symmetry supports the corresponding energy conservation law.Spatial symmetry supports the corresponding momentum conservation law.
Boundary of useSpecify the full closed system or the external driving.Include geometry and boundaries before making a global claim.

Take the comparison further.

Which symmetry is actually present in the proposed system, and what term in its model breaks that symmetry?

  1. Invariant Variation Problems

    Emmy Noether / M. A. Tavel’s English translation · Original 1918 variational theorems, preserved in translation.

  2. Feedback Systems: An Introduction for Scientists and Engineers

    Karl J. Åström and Richard M. Murray / Caltech author archive · Author-published control-system exposition, modelling assumptions and stability.

B14

Statistical physics and learning

Under what assumptions can a learning problem be expressed as a statistical-mechanical model?

Sources for this comparison

Formal model correspondence with restricted assumptions

Some neural-network models admit an energy function, making statistical-physics tools directly useful for analyzing attractors and collective behavior. Hopfield’s construction is a concrete example. This relationship does not make every training loss a physical energy or every modern learning system an equilibrium model. Symmetric couplings, update rules, stochastic assumptions and the distinction between learning weights and evolving states must be specified.

A network update can be checked against an explicit energy.

Take three binary units s = ±1, no self-couplings, and equal symmetric coupling 1 between each pair. Define E = −(s1s2 + s1s3 + s2s3). The state (1,1,−1) has E = 1. Updating only the third unit to agree with its positive local field gives (1,1,1), whose energy is −3. This one asynchronous update lowers E by 4. The calculation demonstrates the chosen dynamics; a different update schedule or asymmetric weights require a new argument.

AspectFirst domainSecond domain
State dynamicsA specified network evolves among configurations.A statistical model assigns an energy or probability to configurations.
LearningTraining changes parameters using data and an objective.Relaxation changes the state under fixed parameters.
ValidationCheck task performance and generalization on unseen examples.Check whether the claimed energy property holds under the actual update rule.

Take the comparison further.

Does the energy still decrease under the implemented update schedule, and does reaching a low-energy state solve the intended task?

  1. Neural networks and physical systems with emergent collective computational abilities

    John J. Hopfield / Caltech author archive · Original 1982 network model; energy-function conclusions depend on its update and coupling assumptions.

  2. Neuromorphic electronic systems

    Carver Mead / Caltech author archive · Original 1990 account of using electronic device physics to implement neural principles.

B15

Error correction in communications, computation, and biology

Which errors can each encoding recover, and what physical resources does recovery consume?

Sources for this comparison

Shared error-reduction goal with distinct mechanisms

Communication codes, quantum codes and biochemical proofreading all spend resources to reduce specified errors. Their mechanisms differ. A classical repetition code uses redundancy and a decoding rule; a quantum code protects encoded states through an appropriate code space and error syndromes; kinetic proofreading uses driven reaction pathways. The comparison becomes useful when error models, resource costs and failure conditions are stated for each.

Redundancy improves a particular error model and can fail under another.

Encode a classical 1 as 111 and decode by majority. With independent bit-flip probability p = 0.1, decoding fails when two or three bits flip: 3p²(1−p) + p³ = 0.028, compared with 0.1 for one unprotected bit. If a common disturbance flips all three together with probability 0.1, majority voting offers no improvement. This example is not a quantum cloning procedure or a model of DNA repair. Those systems need their own error processes and accounting.

AspectFirst domainSecond domain
Classical channelRedundant symbols and a decoder correct a declared set of symbol errors.Correlated errors can defeat a code designed for independent noise.
Quantum memoryAn encoded subspace and syndrome extraction protect quantum information.Copying an arbitrary unknown quantum state is not the mechanism.
Biochemical pathwayDriven reaction steps can increase molecular discrimination.Energy use, reaction rates and rejected substrates must be measured.

Take the comparison further.

Which correlation in the real error process most strongly violates the independence assumption used to predict the improvement?

  1. Scheme for reducing decoherence in quantum computer memory

    Peter W. Shor / Physical Review A · Original quantum error-correcting construction under independent-noise assumptions.

  2. Kinetic proofreading: A new mechanism for reducing errors in biosynthetic processes requiring high specificity

    John J. Hopfield / Proceedings of the National Academy of Sciences · Original proposed biochemical proofreading mechanism and its kinetic and energetic conditions.

B16

Networks and collective behavior

Can systems with identical connectivity exhibit different collective behavior under different local rules?

Sources for this comparison

Method transfer requiring both topology and dynamics

A graph records which entities can interact. Collective behavior also depends on what those interactions do, when they occur and how states change. The small-world framework makes certain structural properties explicit, but equal path lengths or clustering coefficients do not imply equal learning, contagion or coordination. A useful cross-domain comparison holds the graph fixed while varying the local rule.

The same ring supports either spreading or containment.

Construct a ring of four nodes and activate one node initially. In rule A, an inactive node becomes active when at least one neighbor is active; active nodes remain active. With synchronous updates, all four are active after two rounds. In rule B, activation requires both neighbors to be active. The same starting condition never spreads beyond the original node. Connectivity, degree and graph size are identical; the different outcome comes from the update rule.

AspectFirst domainSecond domain
TopologyBoth examples use the same four-node ring.Structural graph statistics therefore cannot distinguish the outcomes.
DynamicsRule A uses a one-neighbor activation threshold.Rule B requires two active neighbors.
EvidenceA network map identifies possible interactions.Time-resolved observations are needed to infer the actual state-transition rule.

Take the comparison further.

Can the observed dynamics be predicted after the initial activation pattern changes, while the network itself remains fixed?

  1. Collective dynamics of ‘small-world’ networks

    Duncan J. Watts and Steven H. Strogatz / Nature · Original small-world network study; topology and dynamical rules remain separate.

  2. Neural networks and physical systems with emergent collective computational abilities

    John J. Hopfield / Caltech author archive · Original 1982 network model; energy-function conclusions depend on its update and coupling assumptions.

B17

Causal inference across models and experiments

Under what conditions does an effect measured in one population transfer to another?

Sources for this comparison

Formal causal inference with empirical transport assumptions

An effect measured under intervention can transfer to another population only under assumptions about which causal mechanisms remain stable and which population features change. Transportability makes those assumptions explicit. A similar correlation or a similar demographic label is insufficient. The analysis must distinguish who received the intervention, what was measured and which variables modify its effect.

A changed mixture can change the average effect without changing either subgroup.

In a hypothetical randomized training study, beginners gain eight completed tasks per period and experienced participants gain two. If the study population is 80 percent beginners, the average effect is 0.8·8 + 0.2·2 = 6.8 tasks. In a target population that is only 20 percent beginners, reweighting gives 0.2·8 + 0.8·2 = 3.2 tasks. This transfer is justified only if the subgroup effects remain applicable, the subgroup definitions match and important effect modifiers have not been omitted.

AspectFirst domainSecond domain
Within-study questionRandomization supports a causal effect for the studied setting.It does not automatically establish the target population’s effect.
Population changeThe source has one distribution of effect-modifying variables.The target can have a different distribution.
Transport testEstimate conditional effects with adequate observations.Check mechanism stability, overlap and the target distribution before reweighting.

Take the comparison further.

Which plausible difference between the source and target could change the subgroup effects themselves rather than only their mixture?

  1. External Validity: From Do-Calculus to Transportability Across Populations

    Judea Pearl and Elias Bareinboim · Formal transportability analysis under explicit causal assumptions.

B18

Measurement and inverse problems

Which distinct source configurations remain compatible with the calibrated measurements?

Sources for this comparison

Shared inverse-problem structure with instrument-specific evidence

An inverse problem asks which hidden states could have produced a measurement. Identifiability concerns whether different states can be distinguished in principle; uncertainty concerns how precisely the available data constrain them. Cosmology, sensing and laboratory reconstruction can share this structure without sharing the same instrument or prior. A sharper image can still conceal a non-unique inverse solution.

A second kind of measurement can resolve an ambiguity that repetition cannot.

Suppose an instrument measures only y = x1 + x2 = 10. The hidden pairs (3,7), (6,4) and infinitely many others fit exactly. Repeating the same noiseless sum does not choose among them. Add an independently calibrated measurement z = x1 − x2 = −4, and the equations give x1 = 3 and x2 = 7. With noisy measurements, carry their uncertainty into both estimates. The example shows why experimental design should target the unresolved direction rather than simply collect more of the same data.

AspectFirst domainSecond domain
Forward modelA proposed hidden state predicts an instrument response.Different hidden states may predict the same response.
New evidenceMore repetitions can reduce random uncertainty.A different measurement can remove a structural ambiguity.
ReportingGive the fitted state and its residuals.Also show alternative states, calibration assumptions and uncertainty.

Take the comparison further.

Which unmeasured combination of variables remains invisible to the present instrument, and what additional observation would make it visible?

  1. Evaluation of measurement data—Guide to the expression of uncertainty in measurement

    Joint Committee for Guides in Metrology / BIPM · JCGM 100:2008, reporting a result with a model and evaluated uncertainty.

  2. Planck 2018 results. VI. Cosmological parameters

    Planck Collaboration · Final-mission cosmological inference under stated model assumptions.

B19

Sensing across engineered and living systems

What is the smallest change each sensor resolves at the same noise level and observation time?

Sources for this comparison

Empirical comparison using a common measurement definition

Engineered and living sensors can be compared through a defined stimulus, response, noise level and observation time. The mechanism still requires independent evidence. Magnetic sensitivity observed in isolated cryptochrome establishes a molecular result under experimental conditions; connecting that result to navigation requires evidence across additional biological scales. A sensitivity number without bandwidth, averaging time or detection criterion is incomplete.

Longer averaging helps only under a stated noise model.

Suppose an illustrative sensor has an uncertainty of 10 nanotesla after one second, with independent stationary noise that averages as 1/√T. It would reach 1 nanotesla after 100 seconds under that model. A second sensor reported at 1 nanotesla after one second is not equivalent to the first at one second. If slow drift dominates, the predicted 100-second improvement may fail. Compare both sensors using the same stimulus protocol and false-alarm criterion rather than their smallest advertised numbers.

AspectFirst domainSecond domain
StimulusAn engineered sensor can be calibrated against a controlled field.A living response must be linked to a controlled field and relevant behavior or chemistry.
SensitivityReport units, bandwidth, integration time and noise model.Use the same definitions before making a quantitative comparison.
MechanismInstrument characterization identifies a transduction pathway.Molecular sensitivity does not alone identify an organism’s complete sensing pathway.

Take the comparison further.

Does the claimed sensitivity survive matched observation time, blinded stimulus variation and a predeclared false-alarm rate?

  1. Magnetic sensitivity of cryptochrome 4 from a migratory songbird

    Jingjing Xu and colleagues / Nature · Original molecular experiment on isolated cryptochrome; not by itself an in-vivo navigation demonstration.

  2. Evaluation of measurement data—Guide to the expression of uncertainty in measurement

    Joint Committee for Guides in Metrology / BIPM · JCGM 100:2008, reporting a result with a model and evaluated uncertainty.

B20

Quantum chemistry to biological mechanism

Which measurements connect a molecular calculation to a change in biological function?

Empirical hypothesis spanning distinct biological scales

Quantum chemistry supplies models of molecular states and transitions. Establishing a biological mechanism requires connecting those models to measurements, then to a change in function under relevant conditions. A spectroscopic coherence, an isotope-sensitive reaction rate and a whole-organism advantage are different observations. The bridge becomes stronger when an intervention changes the proposed mechanism and a predicted functional outcome together.

Improving a fast step may barely change the whole process.

In a constructed two-step pathway, the mean completion time is 1/k1 + 1/k2. Let k1 = 1,000 per second and k2 = 1 per second. The effective completion rate is about 0.9990 per second. Doubling the first rate to 2,000 changes the total rate to about 0.9995, an increase of roughly 0.05 percent. Even a large molecular improvement can therefore have little system-level effect when another step is limiting. Real pathways may branch or feed back, requiring a more complete model.

AspectFirst domainSecond domain
Molecular evidenceSpectroscopy or a reaction experiment constrains states and transition rates.A calculation must predict an observable under the same conditions.
Functional evidenceA pathway’s output depends on all relevant steps.A change in one fast step may not control the final output.
Causal testPerturb the proposed mechanism and measure the local effect.Also measure the predicted downstream consequence and competing explanations.

Take the comparison further.

Which intervention would alter the proposed quantum contribution while leaving the main alternative explanation sufficiently controlled?

  1. Evidence for wavelike energy transfer through quantum coherence in photosynthetic systems

    Gregory S. Engel and colleagues / Nature · Original spectroscopic experiment; functional biological benefit requires additional evidence.

  2. Nature of hydrogen transfer in soybean lipoxygenase 1: separation of primary and secondary isotope effects

    K. W. Rickert and J. P. Klinman / Biochemistry · Original enzyme isotope-effect experiment; evidence is specific to the tested mechanism.

B21

Nonequilibrium thermodynamics and life

How does an open system’s measured energy budget constrain the work it can perform?

Sources for this comparison

Shared conservation laws with nonequilibrium constraints

Living systems maintain organization while exchanging matter and energy with their surroundings. A useful thermodynamic description defines the system boundary, the flows across it and the timescale. Energy conservation constrains the account, while entropy production and free-energy availability constrain what transformations can occur. An organized pattern is not evidence that the second law has been suspended; the environment belongs in the accounting.

An open-system ledger can expose an omitted flow.

For an illustrative one-minute interval, let 10 joules enter a small system. Suppose 3 joules leave as work, 4 as heat and 2 as energy carried by outgoing material. The remaining 1 joule is stored internally. A claim of 4 joules stored with those same flows would leave 3 joules unexplained. This is a first-law check only. Determining whether the process is physically feasible also requires temperatures, chemical potentials and the associated entropy balance.

AspectFirst domainSecond domain
BoundaryA closed description can hide the system’s exchanges.An open description lists heat, work and material flows.
First-law testEnergy input and output must match the change in stored energy.An unexplained residual identifies a missing flow or measurement problem.
Second-law testEnergy quantity alone does not establish available work.Entropy production and free-energy changes constrain the process.

Take the comparison further.

Which boundary flow is least well measured, and can the energy residual be bounded before interpreting the system’s efficiency?

  1. Stochastic thermodynamics, fluctuation theorems, and molecular machines

    Udo Seifert · Author’s formal treatment of fluctuating work, heat and entropy production for specified open systems.

  2. Kinetic proofreading: A new mechanism for reducing errors in biosynthetic processes requiring high specificity

    John J. Hopfield / Proceedings of the National Academy of Sciences · Original proposed biochemical proofreading mechanism and its kinetic and energetic conditions.

B22

Evolution, search, and optimization

How does the distribution of variation affect the outcomes of selection and of computational search?

Sources for this comparison

Algorithmic analogy with a limited formal comparison

Evolution and computational search both allow variation and selection to change a population of candidates. The analogy becomes precise only after defining inheritance, variation, selection and the environment or objective. A designed optimizer usually receives an explicit score; biological reproductive success emerges from a changing context and has no foresight. Shared vocabulary should not conceal these differences.

The allowed variation can determine whether a better state is reachable.

Construct a two-bit search problem with scores f(00)=2, f(01)=1, f(10)=1 and f(11)=3. A greedy search starting at 00 and accepting only single-bit improvements cannot move. If each bit mutates independently with probability 0.1 in one proposed offspring, the chance of proposing 11 directly is 0.01. If rejected offspring leave the parent unchanged, the mean waiting time for that direct proposal is 100 attempts. The example isolates mutation reachability; it is not a model of a real organism’s adaptive history.

AspectFirst domainSecond domain
ObjectiveThe constructed optimizer uses a fixed, externally assigned score.Biological success depends on reproduction in an environment that can change.
VariationThe algorithm designer specifies which moves are permitted.Mutation, recombination and developmental constraints shape biological variation.
InterpretationA successful search demonstrates performance on its defined landscape.A historical adaptation requires evidence about inheritance and selection in context.

Take the comparison further.

Does the apparent advantage come from the selection rule or from giving one method access to a more useful set of variations?

  1. On the Origin of Species, first edition

    Charles Darwin / Darwin Online primary-text archive · Original account of variation and natural selection; not a modern quantitative population-genetics model.

B23

Development, pattern formation, and engineered tissues

Which features of a tissue pattern does a mathematical model predict across changes of scale?

Sources for this comparison

Formal pattern-forming mechanism requiring biological validation

Reaction–diffusion models show how interacting substances can destabilize a uniform state and generate spatial structure under particular conditions. Turing’s mechanism is a mathematical possibility with experimentally testable implications. A striped image alone does not identify it, because growth, mechanics, cell movement and boundary conditions can generate similar patterns. Engineered tissues require the relevant variables and interventions to be measured.

Diffusion can destabilize a locally stable toy system.

Take the local linear reaction matrix J = [[1,−2],[2,−3]]. Its trace is −2 and determinant is 1, so the uniform reaction dynamics are stable. Add diffusion coefficients 1 and 10. For a spatial mode with squared wave number 0.3, the mode matrix becomes [[0.7,−2],[2,−6]], whose determinant is −0.2. One eigenvalue is therefore positive, so that mode grows in this linearized toy model. Nonlinear saturation and boundaries determine what happens later; the calculation alone does not establish a tissue pattern.

AspectFirst domainSecond domain
Without spatial transportLocal perturbations decay under the chosen reaction matrix.Uniform stability says nothing yet about every spatial mode.
With diffusionEach spatial wavelength changes the effective linear dynamics.Unequal diffusion can create a band of growing modes.
Biological testMeasure candidate chemical or cellular variables.Perturb transport or reaction parameters and predict how the pattern changes.

Take the comparison further.

Does changing a measured transport coefficient shift the dominant pattern wavelength in the direction and amount predicted by the model?

  1. The Chemical Basis of Morphogenesis

    Alan M. Turing / Royal Society paper hosted by MIT · Original reaction–diffusion instability model.

B24

Brains, artificial networks, and neuromorphic devices

Which feature of a neural circuit produces a measurable benefit when implemented in hardware?

Sources for this comparison

Engineering transfer with an empirical hardware benchmark

Neuromorphic engineering borrows selected principles from neural systems and implements them using available device physics. The transferred feature might be event-driven communication, local memory or analog dynamics. An artificial network does not thereby become a complete model of a brain. A credible benefit is measured for a task, a workload and a full system boundary, including sensing, conversion, memory and idle power.

Sparse events can help while a dense stream removes the advantage.

In a constructed energy account, a sampled design processes 1,000 samples per second at one unit of energy each. An event-driven design uses one unit per event plus 200 units per second of fixed overhead. At 10 events per second it uses 210 units, compared with 1,000. At 1,000 events it uses 1,200. The same architecture changes from lower to higher energy use as activity rises. These are hypothetical accounting units, not a benchmark of a commercial chip or biological neuron.

AspectFirst domainSecond domain
Borrowed principleNeural inspiration identifies a selected operating feature.The engineered implementation uses its own devices and constraints.
WorkloadSparse inputs can reduce unnecessary activity.Dense inputs and overhead can remove the expected savings.
Success criterionMeasure task accuracy and response time.Measure total energy under the same input stream and quality requirement.

Take the comparison further.

At what event rate does the proposed system lose its energy advantage after interface and idle costs are included?

  1. Neuromorphic electronic systems

    Carver Mead / Caltech author archive · Original 1990 account of using electronic device physics to implement neural principles.

  2. Neural networks and physical systems with emergent collective computational abilities

    John J. Hopfield / Caltech author archive · Original 1982 network model; energy-function conclusions depend on its update and coupling assumptions.

B25

Information geometry and statistical inference

Which distance between probability distributions does a model use, and how does that choice affect inference?

Sources for this comparison

Formal statistical geometry with a declared divergence

Information geometry studies families of probability distributions using structures such as Fisher information. A divergence can quantify the cost of using one distribution in place of another, but it need not behave like ordinary Euclidean distance. Its direction and the choice of parameterization matter. A graphical separation between points is meaningful only after the represented distributions and the mathematical quantity are defined.

Reversing the comparison changes the divergence.

For Bernoulli distributions with p = 0.8 and q = 0.5, DKL(p‖q) is about 0.1927 nats, while DKL(q‖p) is about 0.2231 nats. The asymmetry follows from which distribution supplies the expectation. Locally, Fisher information for a Bernoulli parameter is 1/[p(1−p)]; it equals 4 at p = 0.5 and about 11.11 at p = 0.9. The same small parameter displacement therefore need not represent the same local statistical distinguishability.

AspectFirst domainSecond domain
Ordinary metricA metric is symmetric and satisfies a triangle inequality.KL divergence is directional and is not generally a metric.
Reference distributionD(p‖q) averages the log ratio under p.D(q‖p) averages a different log ratio under q.
Local geometryA raw change in a parameter depends on its coordinates.Fisher information weights that change by statistical sensitivity.

Take the comparison further.

Which distribution represents the observations, which represents the approximation, and is the direction of the divergence consistent with that role?

  1. On Information and Sufficiency

    Solomon Kullback and Richard A. Leibler / Annals of Mathematical Statistics · Original divergence paper. DOI and bibliographic metadata verified; publisher text access was limited.

B26

Tensor networks and multiscale representations

Which correlations survive when a tensor-network representation is truncated?

Sources for this comparison

Formal representation method with observable-specific error

Tensor networks represent high-dimensional objects through linked factors. Truncation reduces the resources required to store or contract the representation, while discarding some components. Vidal’s entanglement-renormalization construction makes the treatment of short-range entanglement part of that reduction. Connections to holography or learning are useful when a specific tensor structure and observable are matched; the network’s appearance alone supplies no equivalence.

Retaining most of a state’s norm does not preserve every observable equally.

Consider a normalized bipartite pure state whose squared Schmidt coefficients are 0.80, 0.15 and 0.05. Keeping the first two terms and renormalizing discards weight 0.05. The squared overlap with the original state is 0.95. However, an observable designed to detect only the discarded third component changes its expectation from 0.05 to zero. The state approximation has a clear global fidelity, while that particular observable has a 100 percent relative error. Both statements can be true.

AspectFirst domainSecond domain
Representation costThe full factorization retains every component.The truncated one stores fewer components.
Global errorDiscarded weight or overlap describes one overall loss.It does not assign the same percentage error to each observable.
Physical testChoose correlations or measurements relevant to the question.Check those quantities as the retained bond dimension increases.

Take the comparison further.

Which target correlation converges slowest as the representation is enlarged, and does it control the scientific conclusion?

  1. Entanglement renormalization

    Guifré Vidal · Original proposal to reduce short-range entanglement before multiscale truncation.

  2. Holographic Derivation of Entanglement Entropy from AdS/CFT

    Shinsei Ryu and Tadashi Takayanagi · Original minimal-area proposal and its AdS3/CFT2 interval-entropy check in the stated limit.

B27

Quantum advantage and classical alternatives

Does a computational advantage persist after preparation, sampling and error correction are counted?

Empirical resource comparison informed by formal complexity

A quantum algorithm can offer an advantage in a defined computational model, while a practical comparison must count data access, preparation, repetition, error handling and readout. Classical alternatives must receive comparable input access and accuracy requirements. Shor’s algorithms and Tang’s recommendation-system result illustrate why the exact problem and access assumptions matter more than the presence of the word quantum.

A fast kernel can lose the first query and win an amortized workload.

Use hypothetical timings: a classical method takes 100 seconds per query. A quantum workflow takes 120 seconds to prepare reusable data, then 2 seconds for its kernel and 20 seconds for readout per query. The first query costs 142 seconds, so it is slower. Two queries cost 164 seconds versus 200 classically, provided preparation is truly reusable and both methods meet the same quality target. Noise correction and repeated sampling must be added if they are not already included. These numbers describe an accounting example, not demonstrated hardware.

AspectFirst domainSecond domain
Theoretical comparisonSpecify the problem, input model and scaling parameter.Prove or analyze the resource bound within that model.
Practical comparisonInclude the full classical pipeline and a strong baseline.Include quantum preparation, repetitions, correction and output extraction.
ReuseState which classical preprocessing is amortized.Grant or deny reusable quantum preparation under the same workload assumptions.

Take the comparison further.

Does the advantage remain when the input access model and output accuracy are made identical for the best available classical baseline?

  1. Polynomial-Time Algorithms for Prime Factorization and Discrete Logarithms on a Quantum Computer

    Peter W. Shor · Original quantum algorithms in the stated computational model.

  2. A quantum-inspired classical algorithm for recommendation systems

    Ewin Tang · Original classical algorithm using comparable sampling access assumptions.

B28

Scientific machine learning and symbolic discovery

Can a recovered equation predict a regime absent from its training data?

Sources for this comparison

Empirical model discovery with an out-of-range test

Scientific machine learning can search for compact equations that explain measured dynamics. Sparse identification makes a concrete assumption: useful dynamics can be expressed with relatively few terms in a chosen candidate library. The result depends on that library, the observations and how derivatives and noise are handled. Recovering a compact equation is the beginning of a model test, not evidence that its range of validity is unlimited.

A missing nonlinear term can hide inside a narrow training range.

Construct a system with dx/dt = −x − x³. Near x = 0.1, the cubic term is 0.001 and the linear term is 0.1, so the nonlinearity contributes only about one percent of the linear magnitude. A model fitted to small noisy excursions may appear well described by dx/dt = −x. At x = 1, that fitted equation predicts −1 while the constructed system gives −2. A held-out large-amplitude trajectory reveals the missing behavior that random train–test splits inside the narrow range may miss.

AspectFirst domainSecond domain
Candidate libraryA restricted library can force a simple expression.A richer library permits more mechanisms but raises selection and overfitting questions.
Fit qualityLow error inside the observed range demonstrates interpolation.New regimes test whether the recovered mechanism transfers.
Experimental designRepeated small excursions improve precision locally.A deliberate larger excursion can reveal a structurally missing term.

Take the comparison further.

Which safe, unobserved regime would make the leading candidate equations disagree most clearly?

  1. Discovering governing equations from data by sparse identification of nonlinear dynamical systems

    Steven L. Brunton, Joshua L. Proctor and J. Nathan Kutz / PNAS · Original sparse-regression discovery method and its candidate-library assumptions.

B29

Compression, memory, and low-resource intelligence

At the same accuracy, how much memory and energy does each representation require?

Sources for this comparison

Empirical resource comparison built on coding methods

Compression can reduce a stored representation while introducing a model, a working state and decoding costs. Low-resource intelligence requires the whole account: task quality, peak memory, latency and energy. Dynamic Markov Compression and later compression-based classifiers provide documented methods to test. A historical name, including an uncertain attribution, should be resolved through the publication record before it becomes a technical claim.

The smaller payload can require the larger complete package.

Compare two hypothetical predictors at the same declared classification quality. Method A stores a 2 MB compressed payload and a 12 MB model, totaling 14 MB. Method B stores a 5 MB payload and a 1 MB model, totaling 6 MB. A payload-only chart favors A; the distributable package favors B. Neither total tells us the peak working memory or joules per prediction. Measure those separately on the same hardware and input stream, including startup and decoding.

AspectFirst domainSecond domain
Stored sizePayload size describes the encoded data alone.Complete size also includes the model, dictionary and required metadata.
Runtime costPeak memory and latency depend on execution behavior.They cannot be inferred directly from the compressed file size.
Task qualityHold the error metric and test set constant.Check whether savings persist when the acceptable error threshold changes.

Take the comparison further.

Which cost dominates the intended device: stored bytes, peak working memory, latency or energy per successful prediction?

  1. Data Compression Using Dynamic Markov Modelling

    G. V. Cormack and R. N. S. Horspool / The Computer Journal · Original December 1987 paper abstract and bibliographic record; adaptive Markov modelling with arithmetic coding.

  2. Spam Filtering Using Statistical Data Compression Models

    Andrej Bratko and colleagues / Journal of Machine Learning Research · Original 2006 compression-based classification study; historical benchmark, not a current performance guarantee.

  3. Neuromorphic electronic systems

    Carver Mead / Caltech author archive · Original 1990 account of using electronic device physics to implement neural principles.

B30

Lost methods and changed engineering conditions

Which constraint ended the original experiment, and when did that constraint change?

Sources for this comparison

Historical transmission followed by a new engineering test

A method can become newly useful when hardware, data access, measurement tools or task requirements change. That possibility is different from claiming that older work was forgotten, suppressed or already solved a present problem. The 1987 Dynamic Markov Compression paper and the 2006 compression-classification study provide a dated route through related methods. An archive should preserve what each paper actually attempted and which constraints its evidence addressed.

A removed memory limit is a reason to retest, not a result.

Imagine an archived algorithm whose reproducible implementation needs 64 MB of working memory. A hypothetical target device limited to 8 MB cannot run it. A later device with 256 MB removes that obstacle, but the algorithm may still miss the latency or energy budget. Reproduce the original result first, then rerun the same task under the new resource limit and compare a contemporary baseline. The memory capacities are constructed for this example; they are not claims about the devices used in the DMC papers.

AspectFirst domainSecond domain
Historical questionIdentify the original task, date, authors and measured constraint.Distinguish a documented limitation from a later story about abandonment.
Changed conditionThe earlier environment restricts one or more resources.The new environment changes a named constraint that can be measured.
ReassessmentReproduce the original protocol where possible.Evaluate current usefulness with a matched modern baseline and resource budget.

Take the comparison further.

Which specific constraint has changed since the original work, and what unchanged constraint could still prevent the proposed application?

  1. Data Compression Using Dynamic Markov Modelling

    G. V. Cormack and R. N. S. Horspool / The Computer Journal · Original December 1987 paper abstract and bibliographic record; adaptive Markov modelling with arithmetic coding.

  2. Spam Filtering Using Statistical Data Compression Models

    Andrej Bratko and colleagues / Journal of Machine Learning Research · Original 2006 compression-based classification study; historical benchmark, not a current performance guarantee.

B31

Formal verification, provenance, and reproducible studies

Which parts of a published calculation can a reader verify directly from its inputs and executable procedure?

Sources for this comparison

Complementary formal and empirical verification methods

Formal verification checks a stated property of a model or program under assumptions. Computational reproducibility checks whether specified inputs and procedures yield the reported result. Empirical replication checks whether observations support the claim again. Provenance connects these activities to their records and responsible agents. None substitutes for all the others; a perfectly repeatable calculation can faithfully propagate a mistaken unit or an invalid physical assumption.

A repeatable unit error can survive a correct arithmetic routine.

Suppose a source table records 120, 130 and 140 centimeters. A correct mean routine returns 130. If the output labels that value as meters, every rerun can reproduce the same wrong physical statement. Preserving the source unit and an explicit conversion yields 1.3 meters. A proof that the routine computes the arithmetic mean would not detect a false assumption about its input unit unless units were included in the specification. The provenance record should identify the original table, transformation, code version and resulting output.

AspectFirst domainSecond domain
Formal propertyA proof can establish behavior under a written specification.The specification’s connection to the measured world still needs justification.
ReproducibilityA fixed procedure can reproduce a numerical output.Repeatability does not show that the input interpretation is correct.
ProvenancePreserve entities, activities and responsibility through each transformation.Use that record to locate where an assumption or error entered.

Take the comparison further.

Which assumption is required for the calculation to describe the real measurement, and where is that assumption recorded and tested?

  1. PROV-DM: The PROV Data Model

    W3C · 30 April 2013 Recommendation: entities, activities, agents and derivation.

  2. Evaluation of measurement data—Guide to the expression of uncertainty in measurement

    Joint Committee for Guides in Metrology / BIPM · JCGM 100:2008, reporting a result with a model and evaluated uncertainty.

  3. On Computable Numbers, with an Application to the Entscheidungsproblem

    Alan M. Turing / original paper hosted by University of Southern California · Original 1936 computation paper; no inference that every physical question is computable.

B32

Historical interpretation and recursive reference

Where do two accounts cite a common source, and where does a later reader introduce the resemblance?

Historical and textual inference with a separate interpretive claim

Recursive reference can be studied as a network of quotations, allusions, translations and later interpretations. An explicit citation is stronger evidence of textual dependence than a resemblance introduced by a modern reader. Manuscripts, dates and editions help locate that dependence, while theological meaning remains a further interpretive question. The same discipline helps trace alchemical, natural-philosophical and scientific texts without assuming that shared words always name the same mechanism.

An explicit scriptural reference can be separated from its later interpretation.

Matthew 1:22–23 explicitly presents a passage associated with Isaiah 7:14 as fulfilled in the narrative. A research record can therefore connect those passages, preserve their surrounding chapters and identify the translation used. The next layer records how an interpreter understands fulfillment and how that reading handles differences of language and historical setting. A resemblance between an unrelated modern theory and either passage would be a different kind of edge. These distinctions allow the recursive chain to grow without making every new connection equally established.

AspectFirst domainSecond domain
Documented dependenceA later text explicitly quotes or identifies an earlier passage.The evidence supports a traceable textual relationship.
InterpretationA theological or philosophical reading explains the relationship’s significance.Its premises and tradition should be named so another reader can assess them.
AnalogyA modern comparison can generate a useful question.Similarity alone does not establish historical transmission or a shared physical law.

Take the comparison further.

For the next proposed connection, what is the earliest accessible textual witness, and does it document a quotation, an allusion or a later reader’s analogy?

  1. Matthew, Chapter 1

    United States Conference of Catholic Bishops / New American Bible · Primary translated text at Matthew 1:22–23, with editorial notes identified as notes.

  2. Isaiah, Chapter 7

    United States Conference of Catholic Bishops / New American Bible · Primary translated text at Isaiah 7:14 in its chapter context; translation is a specified edition, not every manuscript tradition.

  3. Codex Sinaiticus: See the Manuscript

    Codex Sinaiticus Project / participating manuscript-holding institutions · Digitized manuscript and transcription; manuscript evidence is distinct from a theological interpretation.

  4. Newton Papers

    Cambridge University Library · Primary manuscript collection, including early notebooks and the Waste Book.