---
title: Auxiliary-Receiver Outer Bound
url: https://www.emergentmind.com/topics/auxiliary-receiver-outer-bound
type: topic
---

# Auxiliary-Receiver Outer Bound

Searching arXiv for the cited paper and closely related auxiliary-receiver / broadcast-style outer-bound work.
The **Auxiliary-Receiver Outer Bound** is a converse technique for the discrete memoryless cognitive interference channel (DM-CIFC) in which the proof introduces a fictitious receiver output rather than conventional auxiliary random variables. In the formulation given in “New inner and outer bounds for the discrete memoryless cognitive interference channel and some capacity results” [1003.4328], the method is explicitly motivated by Sato’s idea for the broadcast channel and exploits the fact that, because the receivers do not cooperate, achievability depends only on the single-receiver marginals \(P_{Y_1|X_1,X_2}\) and \(P_{Y_2|X_1,X_2}\), not on the full joint law of \((Y_1,Y_2)\). This permits replacing the original channel by another with the same receiver marginals but a modified joint coupling, yielding a computable outer bound that avoids unresolved cardinality issues associated with earlier auxiliary-variable converses [1003.4328].

## 1. Channel model and converse setting

The bound is defined for the standard two-user discrete memoryless cognitive interference channel. There are two transmitters and two receivers. Transmitter \(i\) wishes to send message \(W_i\) to receiver \(i\), \(i\in\{1,2\}\), with independent uniformly distributed messages
\[
W_i \sim \text{Unif}[1,\ldots,2^{N R_i}], \qquad i\in\{1,2\}.
\]
The key side-information asymmetry is unilateral non-causal message knowledge: transmitter 1, the cognitive transmitter, knows both \(W_1\) and \(W_2\), whereas transmitter 2, the primary transmitter, knows only \(W_2\). The encoders are
\[
X_1^N=X_1^N(W_1,W_2),\qquad X_2^N=X_2^N(W_2),
\]
and the decoders satisfy
\[
\hat W_i=\hat W_i(Y_i^N),\qquad i\in\{1,2\}.
\]
The channel has finite alphabets and memoryless transition law \(p_{Y_1,Y_2|X_1,X_2}\), and the capacity region is the closure of all achievable rate pairs \((R_1,R_2)\) [1003.4328].

This setting lies between an interference channel and a broadcast channel. Earlier work on the cognitive interference channel already used broadcast-channel-style converse ideas, including receiver-observation auxiliaries such as \(V_i=(Y_1^{i-1},Y_{2,i+1}^N)\) in a Nair–El Gamal-type outer bound [0710.3375]. The auxiliary-receiver outer bound of [1003.4328] departs from that line by replacing auxiliary random variables with an auxiliary output.

## 2. From auxiliary random variables to an auxiliary receiver

Before introducing the new converse, [1003.4328] recalls two older general outer bounds. The first is the one-auxiliary-RV bound of Wu et al.,
\[
R_1 \le I(X_1;Y_1|X_2),
\]
\[
R_2 \le I(X_2,U;Y_2),
\]
\[
R_1+R_2 \le I(X_2,U;Y_2)+I(X_1;Y_1|X_2,U),
\]
over distributions \(p_{U,X_1,X_2}p_{Y_1,Y_2|X_1,X_2}\). The second is the broadcast-inspired outer bound of Maric et al., involving auxiliaries \(V,U_1,U_2\). The paper emphasizes that these bounds cannot be evaluated in general because no cardinality bounds are known for the auxiliaries [1003.4328].

That computability issue parallels a well-known theme in broadcast-channel outer bounds. For the 2-receiver DM-BC, the Nair–El Gamal outer bound admits equivalent auxiliary-variable parameterizations and can be made computable only after additional cardinality analysis; in particular, the seemingly stronger \((U,V,W)\) formulation is identical to a correlated-\((U,V)\) formulation, and an equivalent computable form satisfies \(|\mathcal U|,|\mathcal V|\le |\mathcal X|+1\) [0804.3825]. In the DM-CIFC, however, the older auxiliary-variable bounds recalled in [1003.4328] do not come with such cardinality reductions. The auxiliary-receiver construction is introduced precisely to avoid that obstacle.

The resulting tradeoff is explicit. The new outer bound is “looser in general” than the older one-auxiliary-RV bound, but it is explicitly evaluable because it contains no conventional auxiliary random variables. This is the defining feature of the auxiliary-receiver viewpoint: the converse is strengthened by adding a synthetic output variable, not by postulating a latent \(U\)-type code descriptor [1003.4328].

## 3. The Sato-style outer bound

The main no-auxiliary outer bound in [1003.4328] is Theorem \(\ref{thm: outer bound CIFC}\). If \((R_1,R_2)\) is in the capacity region of the DM-CIFC, then
\[
R_1 \le I(Y_1;X_1|X_2),
\]
\[
R_2 \le I(X_1,X_2;Y_2),
\]
\[
R_1+R_2 \le I(X_1,X_2;Y_2)+I(Y_1;X_1|Y_2',X_2),
\]
over the union of all input distributions \(p_{X_1,X_2}\) and all conditional distributions \(p_{Y_1,Y_2'|X_1,X_2}\) such that
\[
p_{Y_2'|X_1,X_2}=p_{Y_2|X_1,X_2}.
\]
There is no explicit time-sharing variable in the theorem statement [1003.4328].

The variable \(Y_2'\) is the source of the name **auxiliary receiver**. It is not an auxiliary random variable in the standard converse sense. Rather, it is a fictitious receiver output with the same marginal conditional law as \(Y_2\) but otherwise arbitrary correlation with \(Y_1\) given \((X_1,X_2)\). The converse therefore augments the channel with a synthetic second-receiver output and optimizes over all admissible couplings that preserve the \(Y_2\) marginal.

The conceptual basis is Sato’s broadcast-channel principle. Because decoder 1 observes only \(Y_1\) and decoder 2 observes only \(Y_2\), changing the joint dependence between outputs cannot affect achievability so long as the individual marginals are preserved. This is exactly the non-cooperating-receivers premise that makes the construction valid. The essential assumptions are: the channel is memoryless, the receivers do not cooperate, transmitter 1 knows \(W_2\) non-causally, and the modified channel preserves \(P_{Y_1|X_1,X_2}\) and \(P_{Y_2|X_1,X_2}\) [1003.4328].

## 4. Derivation and single-letterization

The converse begins from Fano’s inequality,
\[
H(W_i|Y_i^N)\le N\epsilon_N,\qquad \epsilon_N\to 0,\quad i\in\{1,2\}.
\]
For \(R_1\), the proof uses the cognitive side-information structure and the fact that \(X_2^N\) is a function of \(W_2\):
\[
N(R_1-\epsilon_N)\le I(W_1;Y_1^N)
\le I(W_1;Y_1^N|W_2)
\]
\[
= I(W_1,X_1^N(W_1,W_2);Y_1^N|W_2,X_2^N(W_2))
\le H(Y_1^N|X_2^N)-H(Y_1^N|X_1^N,X_2^N).
\]
By memorylessness and single-letterization,
\[
R_1\le I(Y_1;X_1|X_2).
\]

For \(R_2\), the proof gives
\[
N(R_2-\epsilon_N)\le I(Y_2^N;W_2)
\le I(Y_2^N;W_2,W_1)
\]
\[
= H(Y_2^N)-H(Y_2^N|X_1^N,X_2^N)
\le \sum_{i=1}^N \big[H(Y_{2i})-H(Y_{2i}|X_{1i},X_{2i})\big],
\]
hence
\[
R_2\le I(X_1,X_2;Y_2).
\]
This term is weaker than the earlier \(R_2\le I(X_2,U;Y_2)\), but it avoids auxiliaries [1003.4328].

The sum-rate proof is where the auxiliary receiver enters. Let \(Y_2'\) satisfy
\[
P_{Y_2'|X_1,X_2}=P_{Y_2|X_1,X_2}.
\]
Then
\[
N(R_1+R_2-2\epsilon_N)\le I(W_1;Y_1^N|W_2)+I(W_2;Y_2^N)
\]
\[
\le I(W_1;Y_1^N,Y_2'^N|W_2)+I(W_2;Y_2^N)
\]
\[
= I(W_2;Y_2^N)+I(W_1;Y_2'^N|W_2)+I(W_1;Y_1^N|Y_2'^N,W_2).
\]
Using the fact that \(Y_2'^N\) and \(Y_2^N\) have the same conditional law given \((X_1^N,X_2^N)\), the proof converts the resulting entropies to expressions involving \(Y_2^N\), inserts the codewords, and obtains
\[
\le I(Y_2^N;X_1^N,X_2^N)+\sum_{i=1}^N \Big(H(Y_{1i}|X_{2i},Y_{2i}')-H(Y_{1i}|X_{1i},X_{2i},Y_{2i}')\Big).
\]
Single-letterization yields
\[
R_1+R_2 \le I(X_1,X_2;Y_2)+I(Y_1;X_1|Y_2',X_2).
\]
The proof introduces a time index during single-letterization, but not a theorem-level time-sharing variable [1003.4328].

## 5. Relation to earlier and later outer bounds

The new outer bound is explicitly compared with the older one-auxiliary-RV converse in [1003.4328]. For fixed \(p_{X_1,X_2}\),
\[
R_1\text{-bound:}\qquad I(X_1;Y_1|X_2)=I(Y_1;X_1|X_2),
\]
so the \(R_1\) inequalities coincide. For \(R_2\),
\[
I(Y_2;X_2,U)\le I(Y_2;X_1,X_2),
\]
using the Markov chain
\[
U-(X_1,X_2)-(Y_1,Y_2),
\]
hence the new \(R_2\) bound is weaker. The paper also shows that the old sum-rate bound is contained in the new one after suitable choice of \(Y_2'\). Accordingly, the new outer region is a superset of the older one and is therefore weaker in general [1003.4328].

The paper also identifies a regime where the new bound collapses to the strong-interference converse. If
\[
I(X_1;Y_1|X_2)\le I(X_1;Y_2|X_2)\qquad \forall p_{X_1,X_2},
\]
then
\[
I(Y_1;X_1|Y_2',X_2)\le I(Y_2;X_1|Y_2',X_2),
\]
and setting \(Y_2'=Y_2\) gives
\[
I(Y_1;X_1|Y_2,X_2)=0.
\]
The sum-rate inequality reduces to
\[
R_1+R_2\le I(X_1,X_2;Y_2),
\]
together with
\[
R_1\le I(Y_1;X_1|X_2),
\]
which is exactly the strong-interference outer bound [1003.4328].

The method later became useful beyond the original DM-CIFC context. In the sum-broadcast-channel setting, an auxiliary-receiver outer bound with synthetic receivers \(G\) and \(K\) was used to obtain converse decompositions that the standard UVW outer bound could not provide; for broad classes of component channels, that outer bound was shown to match Marton’s inner bound [2606.12839]. This suggests that the original CIFC construction is best understood not as an isolated trick, but as an instance of a broader converse principle: replace difficult auxiliary-variable structure by carefully chosen artificial receiver outputs when marginal preservation or channel composition makes that possible.

## 6. Tightness, special cases, and limitations

The most important exact specialization in [1003.4328] is the deterministic cognitive interference channel, where
\[
Y_1=Y_1(X_1,X_2),\qquad Y_2=Y_2(X_1,X_2).
\]
Taking \(Y_2'=Y_2\) in the outer bound gives
\[
R_1 \le H(Y_1|X_2),
\]
\[
R_2 \le H(Y_2),
\]
\[
R_1+R_2 \le H(Y_2)+H(Y_1|Y_2,X_2),
\]
over all \(p_{X_1,X_2}\). The paper proves achievability of exactly this region, so the auxiliary-receiver outer bound is capacity-achieving for this class [1003.4328].

By contrast, for the semi-deterministic CIFC, where only
\[
Y_1=f_1(X_1,X_2),
\]
the paper does not use the new bound as the sharp converse. Instead it uses the older Wu auxiliary-variable outer bound to obtain
\[
R_1 \le H(Y_1|X_2),
\]
\[
R_2 \le I(Y_2;U,X_2),
\]
\[
R_1+R_2 \le I(Y_2;U,X_2)+H(Y_1|U,X_2),
\]
which yields the capacity region for that regime. Likewise, capacity in the “better cognitive decoding” regime is established with the older one-auxiliary-RV outer bound under the condition
\[
I(Y_1;X_2,U)\ge I(Y_2;X_2,U)\qquad \forall p_{X_1,X_2,U}.
\]
These results make the paper’s own assessment clear: the auxiliary-receiver outer bound is a general-purpose computable converse, but not always the sharpest available one [1003.4328].

A common misconception is that eliminating auxiliary random variables necessarily strengthens a converse by simplifying it. The paper shows the opposite tradeoff. The gain is explicit evaluability and freedom from unresolved cardinality bounds; the price is weaker general tightness. A plausible implication is that the method is most valuable when computability is itself essential, or when the channel has structural properties—such as determinism or strong interference—that force the synthetic-output coupling to become exact.

## 7. Conceptual significance

The auxiliary-receiver outer bound is best viewed as a Sato-style marginal-preserving converse for channels with non-cooperating receivers. Its novelty is not an additional coding idea but a different kind of converse object. Instead of introducing a latent auxiliary that summarizes codebook structure, it introduces an alternative receiver output that preserves the observable marginals while altering the joint law in a converse-friendly way. In the DM-CIFC, this yields the computable region
\[
R_1 \le I(Y_1;X_1|X_2),\qquad
R_2 \le I(X_1,X_2;Y_2),\qquad
R_1+R_2 \le I(X_1,X_2;Y_2)+I(Y_1;X_1|Y_2',X_2),
\]
with \(Y_2'\) constrained only by
\[
p_{Y_2'|X_1,X_2}=p_{Y_2|X_1,X_2}.
\]

In that precise sense, the term **auxiliary-receiver outer bound** is apt. The auxiliary object is a receiver output, not a conventional random variable. The method inherits the broadcast-channel intuition that only receiver marginals matter when decoders are separate, but adapts it to the cognitive interference setting where one encoder knows both messages non-causally. Its enduring significance lies in demonstrating that computable converses can sometimes be obtained by augmenting the output side of the model rather than the latent-variable side, a design choice later reused in broader broadcast-channel constructions [1003.4328; 2606.12839].

Source: https://www.emergentmind.com/topics/auxiliary-receiver-outer-bound