Information-Friction Loss in Communication

• Information-friction loss is defined as the energy cost to move information, measured as the product of bits and distance (bit-meters) in physical systems.
• In communication circuits, the model uses geometric and combinatorial methods to derive lower bounds on decoding/encoding energy, revealing divergence as error probability decreases.
• In surface science, the concept describes friction-induced information erasure at interfaces, linking thermodynamic stability to state transitions and practical implications for unconventional logic.

Searching arXiv for relevant papers on information-friction and related formulations. Information-friction loss denotes the energy cost associated with moving information across a physical substrate, modeled by analogy with mechanical friction as proportional to the product of information volume and transport distance, measured in bit-meters. In communication circuitry, this notion provides a lower-bound framework for encoding and decoding energy, distinct from transmit energy, and yields the conclusion that total energy per bit must diverge as target error probability tends to zero (Grover, 2014). In a separate tribological and logical interpretation, information-friction loss refers to dissipation-driven erasure of interface information during frictional and wetting dynamics, where instabilities and attractors eliminate the dynamical relevance of detailed initial microstates (Nosonovsky, 2017). The term therefore spans two related but non-identical research programs: one treats information transport as an energetic primitive in computation and communication, and the other treats frictional dissipation as a mechanism of information loss at interfaces.

1. Formal model of information-friction in communication circuits

In the communication-theoretic formulation, information-friction is a physically motivated model for the energy cost of moving information across a computational substrate such as a VLSI chip, a network of processors, or a system of repeaters or wireless relays (Grover, 2014). Its central metric is the bit-meter, defined as the cumulative product of the number of bits transported and the Euclidean distance over which they are transported.

If a unidirectional link carries $b$ bits over distance $d$, the bit-meters consumed by that link are $bd$. For an entire computation, the total bit-meters are

$\mathrm{bm} \;=\; \sum_i b_i d_i,$

equivalently,

$$\mathrm{bm} \;=\; \sum_{\ell \in \text{links}} \left( \sum_{\text{uses of }\ell} b_{\ell} \right) d_{\ell}.$$

Energy is then modeled as

$E \;=\; \mu\,\mathrm{bm},$

where $\mu > 0$ is the coefficient of information-friction (Grover, 2014). In many realistic technologies, $\mu\,\mathrm{bm}$ also serves as a lower bound to actual link energy.

The circuit model is defined on a square substrate $\mathrm{Sq}(l) \subset \mathbb{R}^2$ populated by computational nodes on $\mathrm{Grid}(\lambda)$, where $d$0 is the minimum allowable spacing between nodes. A circuit $d$1 consists of the substrate and a node set $d$2, including input nodes, output bit-nodes, and helper nodes. Links are noiseless, unidirectional connections between ordered node pairs. The model distinguishes fixed-message-length computation, in which both message order and length are predetermined and input-independent, from flexible-message-length computation, in which order remains predetermined but message lengths may depend on the input, subject to a minimum length of one bit (Grover, 2014).

This construction makes information movement, rather than merely arithmetic or logic operations, the primitive object of analysis. A plausible implication is that it shifts lower-bound arguments from computation counts toward communication geometry.

2. Channel assumptions, partitioning method, and lower-bound technique

The principal communication setting is a binary-input AWGN channel with hard decision at the receiver, thereby inducing an effective BSC with crossover probability

$d$3

where $d$4 is transmit power, $d$5 is path-loss, and $d$6 is the noise variance entering the hard decision. The induced channel capacity is

$d$7

For a code of blocklength $d$8, information bits $d$9, and rate $bd$0, the block error probability is

$bd$1

The lower-bound proofs are built around a geometric decomposition called stencil partitioning. The substrate is partitioned into disjoint subcircuits using a stencil of outer squares of side $bd$2, each containing a concentric inner square of side $bd$3, where $bd$4. The annulus width is therefore $bd$5. If $bd$6 denotes the number of bit-nodes in the inner square of subcircuit $bd$7, then by suitable placement of the stencil origin,

$bd$8

The key geometric lemma is the annulus communication bound: to communicate $bd$9 independent bits from outside the outer square to inside the inner square of a subcircuit,

$\mathrm{bm} \;=\; \sum_i b_i d_i,$0

under fixed message lengths, and under flexible message lengths the average bit-meters satisfy the same bound. An analogous statement holds for communication in the reverse direction (Grover, 2014).

A node-count bound is also used:

$\mathrm{bm} \;=\; \sum_i b_i d_i,$1

These ingredients are combined with Fano-type arguments under an erasure reduction from BSC to BEC. The methodological structure is therefore geometric, combinatorial, and information-theoretic: one localizes decoding or encoding tasks to subcircuits, forces a minimum amount of communication across annular cuts, and then converts insufficient communication into a contradiction with the target error probability (Grover, 2014).

3. Encoding and decoding lower bounds

For decoding with fixed message lengths, the main theorem states that any decoder implemented in the specified model and achieving block error $\mathrm{bm} \;=\; \sum_i b_i d_i,$2 must satisfy

$\mathrm{bm} \;=\; \sum_i b_i d_i,$3

provided that

$\mathrm{bm} \;=\; \sum_i b_i d_i,$4

The corresponding energy bound is

$\mathrm{bm} \;=\; \sum_i b_i d_i,$5

(Grover, 2014).

For encoding, the fixed-length assumption is relaxed. The theorem for encoding states that, whether message lengths are fixed or flexible, the average bit-meters satisfy

$\mathrm{bm} \;=\; \sum_i b_i d_i,$6

under the same condition on $\mathrm{bm} \;=\; \sum_i b_i d_i,$7 and $\mathrm{bm} \;=\; \sum_i b_i d_i,$8 (Grover, 2014).

If the condition

$\mathrm{bm} \;=\; \sum_i b_i d_i,$9

fails, then the framework implies that transmit power must grow at least logarithmically:

$$\mathrm{bm} \;=\; \sum_{\ell \in \text{links}} \left( \sum_{\text{uses of }\ell} b_{\ell} \right) d_{\ell}.$$0

This is significant because the model does not merely produce a decoder-energy bound under favorable asymptotics; it bifurcates the argument. Either communication within the circuit already requires substantial bit-meters, or transmit power itself must increase rapidly (Grover, 2014).

The proofs rely on the fact that if a subcircuit’s bit-meters fall below the amount needed to communicate about one third of its inner bits across the annulus, then with positive probability all relevant inputs inside the outer square are erased after the BSC-to-BEC reduction, so the inner bits cannot be recovered with sufficiently small error. This yields the lower bound after summing across subcircuits.

4. Reliability, near-capacity operation, and divergence of energy per bit

A sharper asymptotic statement concerns blocklength dependence. For fixed-message-length decoding,

$$\mathrm{bm} \;=\; \sum_{\ell \in \text{links}} \left( \sum_{\text{uses of }\ell} b_{\ell} \right) d_{\ell}.$$1

as long as

$$\mathrm{bm} \;=\; \sum_{\ell \in \text{links}} \left( \sum_{\text{uses of }\ell} b_{\ell} \right) d_{\ell}.$$2

When $$\mathrm{bm} \;=\; \sum_{\ell \in \text{links}} \left( \sum_{\text{uses of }\ell} b_{\ell} \right) d_{\ell}.$$3 is bounded, as in near-capacity operation at fixed rate, $$\mathrm{bm} \;=\; \sum_{\ell \in \text{links}} \left( \sum_{\text{uses of }\ell} b_{\ell} \right) d_{\ell}.$$4 is bounded, so $$\mathrm{bm} \;=\; \sum_{\ell \in \text{links}} \left( \sum_{\text{uses of }\ell} b_{\ell} \right) d_{\ell}.$$5 diverges with $$\mathrm{bm} \;=\; \sum_{\ell \in \text{links}} \left( \sum_{\text{uses of }\ell} b_{\ell} \right) d_{\ell}.$$6 (Grover, 2014).

Using a standard finite-blocklength relation near capacity,

$$\mathrm{bm} \;=\; \sum_{\ell \in \text{links}} \left( \sum_{\text{uses of }\ell} b_{\ell} \right) d_{\ell}.$$7

for some $$\mathrm{bm} \;=\; \sum_{\ell \in \text{links}} \left( \sum_{\text{uses of }\ell} b_{\ell} \right) d_{\ell}.$$8, one obtains

$$\mathrm{bm} \;=\; \sum_{\ell \in \text{links}} \left( \sum_{\text{uses of }\ell} b_{\ell} \right) d_{\ell}.$$9

This exhibits accelerated divergence as $E \;=\; \mu\,\mathrm{bm},$0 for small $E \;=\; \mu\,\mathrm{bm},$1 (Grover, 2014).

For total energy per bit, with transmit energy included and decoding omitted only because it has the same order, the lower bound is

$E \;=\; \mu\,\mathrm{bm},$2

Using the hard-decision BSC approximation $E \;=\; \mu\,\mathrm{bm},$3, this becomes

$E \;=\; \mu\,\mathrm{bm},$4

for some $E \;=\; \mu\,\mathrm{bm},$5. Minimizing over $E \;=\; \mu\,\mathrm{bm},$6 yields

$E \;=\; \mu\,\mathrm{bm},$7

Hence,

$E \;=\; \mu\,\mathrm{bm},$8

under the information-friction model (Grover, 2014).

A common misconception is that only transmit energy matters in the low-error regime. The results above show that, under the stated implementation assumptions, encoding and decoding communication costs impose a separate divergence, and near-capacity operation can make the divergence faster because bounded $E \;=\; \mu\,\mathrm{bm},$9 forces blocklength growth.

5. Relation to other physical and computational energy models

The bit-meters framework is explicitly contrasted with Landauer’s principle and Thompson’s VLSI area-time model. Landauer’s principle concerns the energy cost of erasure in thermodynamic equilibrium and allows arbitrarily small communication energy only under vanishing friction or noise at zero speed. By contrast, information-friction models finite-time, finite-friction movement of information across a substrate, including wired, wireless, and transport-based communication (Grover, 2014).

Thompson’s VLSI model approximates energy through wiring infrastructure and time. Bit-meters differ in three stated respects: they account for the amount and distance of actual information movement rather than installed wire length alone; they apply to non-metal interconnects, high-degree connectivity, and input-dependent message lengths; and they provide input-sensitive lower bounds through message-length dependence and cutset-like arguments (Grover, 2014).

Prior decoding-complexity bounds often measured operations or area. Bit-meters instead supply a direct lower bound on energy due to interconnect communication. This suggests a complementary perspective rather than a replacement: node computation costs are not eliminated from consideration, but the framework isolates a physically interpretable communication term that can dominate in some regimes.

The paper also notes several limitations. Broadcast or multicast may lead to energy scaling sublinearly with cumulative distance, making bit-meters bounds loose. Timing-based communication at extremely low speeds could reduce the effective coefficient $\mu > 0$0, though synchronization overheads may offset such gains. Optical links, wireless on-chip relays, and fluidic computing may alter the effective friction coefficient or distance scaling. The model also ignores noise in computation nodes and only prices link movement (Grover, 2014).

6. Practical implications for code and architecture design

The communication-theoretic treatment derives explicit design guidance. To reduce information-friction loss, one should minimize interconnect lengths, place decoding nodes near the data they need, and use short, local routes. Localized or hierarchical decoding is recommended, with many small subcircuits and local message-passing to reduce long-haul communication. Code constructions and graph realizations that promote locality, such as LDPC realizations with local neighborhoods on the substrate, are favored over architectures requiring global message broadcasts. Pipelining and caching are also recommended to limit long-distance memory access (Grover, 2014).

For short-range communication, the framework suggests preferring simpler codes with smaller wiring and messaging overhead, and approaching capacity only when communication distance is large enough that transmit energy dominates (Grover, 2014). This is not stated as an absolute rule; rather, it is a design consequence of the transmit-versus-decoding trade-off.

The paper gives an illustrative quantitative example for on-chip metal wires. With capacitance per unit length around $\mu > 0$1 pF/mm, a charge-discharge at $\mu > 0$2 V costs approximately $\mu > 0$3 pJ/mm per bit transition, giving a plausible coefficient $\mu > 0$4 pJ/(bit·mm). For $\mu > 0$5 mm, $\mu > 0$6, $\mu > 0$7, and $\mu > 0$8, the decoding lower bound evaluates to approximately $\mu > 0$9 bit·mm, corresponding to about $\mu\,\mathrm{bm}$0 pJ per codeword due to decoding communication alone (Grover, 2014). Larger $\mu\,\mathrm{bm}$1 or tighter $\mu\,\mathrm{bm}$2 increase this value.

Open questions identified in the work include the role of feedback, especially noisy feedback; trade-offs in multiuser or interference-limited systems; extensions to noisy computation and memory; constructive architectures that might match the lower bounds; and refinements of the distance metric to cover Manhattan distance, nonuniform substrates, or dynamic interconnect reconfiguration (Grover, 2014).

7. Frictional interfaces, information erasure, and unconventional logic

A distinct use of the term appears in the surface-science literature, where information-friction loss refers to the erasure of interface information by dissipation-driven instability and self-organization (Nosonovsky, 2017). Here, “information” means the distinguishability of microscopic or mesoscopic interface states, such as roughness distributions, microstructure, or instantaneous dynamic state. When frictional sliding evolves through instabilities toward a robust steady regime or limit cycle, details of the initial interface state cease to matter dynamically; the interface is said to “forget” its initial microstate.

To formalize this, the paper introduces a ternary logic for the state predicate $\mu\,\mathrm{bm}$3 of a contact:

• $\mu\,\mathrm{bm}$4: motion,
• $\mu\,\mathrm{bm}$5: rest,
• $\mu\,\mathrm{bm}$6: paradoxical or dynamically unstable state.

This is used to interpret Painlevé-type paradoxes in rigid-body mechanics with Coulomb friction. In the illustrative two-slider model, paradoxes arise when $\mu\,\mathrm{bm}$7, and the binary logical law $\mu\,\mathrm{bm}$8 fails. The paper resolves this by assigning the third logical value:

$\mu\,\mathrm{bm}$9

When elastic compliance $\mathrm{Sq}(l) \subset \mathbb{R}^2$0 is introduced, the paradoxical rigid-body state becomes a dynamically unstable one, summarized by

$\mathrm{Sq}(l) \subset \mathbb{R}^2$1

and in the rigid limit,

$\mathrm{Sq}(l) \subset \mathbb{R}^2$2

Thus, as $\mathrm{Sq}(l) \subset \mathbb{R}^2$3,

$\mathrm{Sq}(l) \subset \mathbb{R}^2$4

(Nosonovsky, 2017).

The same work formulates a thermodynamic stability criterion via the second variation of entropy production:

$\mathrm{Sq}(l) \subset \mathbb{R}^2$5

For purely mechanical sliding at temperature $\mathrm{Sq}(l) \subset \mathbb{R}^2$6, with normal load $\mathrm{Sq}(l) \subset \mathbb{R}^2$7, Coulomb friction $\mathrm{Sq}(l) \subset \mathbb{R}^2$8, and velocity $\mathrm{Sq}(l) \subset \mathbb{R}^2$9,

$\mathrm{Grid}(\lambda)$0

Varying $\mathrm{Grid}(\lambda)$1 and $\mathrm{Grid}(\lambda)$2 together yields

$\mathrm{Grid}(\lambda)$3

so stability requires

$\mathrm{Grid}(\lambda)$4

whereas $\mathrm{Grid}(\lambda)$5 produces instability and stick-slip-type behavior (Nosonovsky, 2017).

The paper also introduces a Shannon entropy for surface roughness during run-in:

$\mathrm{Grid}(\lambda)$6

and proposes the minimization principle

$\mathrm{Grid}(\lambda)$7

During run-in, $\mathrm{Grid}(\lambda)$8 decreases as asperities deform and fracture and the surface reorganizes toward a more ordered steady-sliding configuration (Nosonovsky, 2017). The paper does not invoke formal Landauer bounds, although it notes the general statement that erasing one bit costs at least $\mathrm{Grid}(\lambda)$9.

This surface-science formulation extends further to wetting, capillarity, and unconventional computation. Wetting is treated as a binary logic with Wenzel and Cassie–Baxter states governed by Young–Dupré, Wenzel, and Cassie–Baxter relations. Bubble and droplet logic are discussed as physical logic primitives in which gate operations dissipate viscous and capillary energy while collapsing continuous hydraulic states to discrete outcomes, thereby erasing information at the interface (Nosonovsky, 2017).

The two meanings of information-friction loss are therefore related by a common theme—energy expenditure tied to information movement or information erasure—but they operate at different levels. In communication circuits, the term is a quantitative lower-bound model based on bit-meters. In surface science, it is a logical and thermodynamic interpretation of how frictional or capillary dynamics remove microstate information. A plausible implication is that the shared terminology reflects a broader attempt to unify energetic dissipation and information loss across physical systems, though the two formalisms are not interchangeable.