recip(x) Is 1 Node — ELSb Closes the Gap
Result: R16-C1 · Tag: THEOREM
The Finding
The reciprocal function $1/x$ is computable in exactly one node over the 16-operator family $\mathcal{F}_{16}$, using the operator $\operatorname{ELSb}(a, b) = \exp(a - \ln b) = e^a / b$:
$$\operatorname{ELSb}(0,, x) = \exp(0 - \ln x) = \exp(-\ln x) = \frac{1}{x}$$
That is all. Set the first argument to $0$, the second argument to $x$, and $e^0 / x = 1/x$ exactly.
The previous best was 2 nodes — a construction via EDL that first converted $x$ to $e^x$ and then divided through a logarithm. R16-C1 saves that intermediate step entirely.
Why It Was Missed
The natural operator family for reciprocals was EDL: $\operatorname{EDL}(a,b) = e^a / \ln b$.
To compute $1/x$ via EDL, you need the denominator to equal $\ln b = x$, which means $b = e^x$. That requires an inner node to produce $e^x$. Two nodes total.
The ELSb operator was studied for division: $\operatorname{ELSb}(\ln x, y) = e^{\ln x}/y = x/y$ (a 2-node construction needing an inner $\ln x$ node). Division was the headline use case. Nobody checked what happened when the first argument is the constant $0$ instead of $\ln x$.
$\operatorname{ELSb}(0, x) = e^0/x = 1/x$.
One node. The operator already knew how to do reciprocal. We just had to ask.
This follows the same structural pattern as the earlier reductions:
| Result | Operator | Key move | Savings |
|---|---|---|---|
| T10u: mul $3\to2$ | ELAd | $\operatorname{ELAd}(\ln x, y) = xy$ | 1 node |
| T33: sub $3\to2$ | LEdiv | $\operatorname{LEdiv}(x, e^y) = x-y$ | 1 node |
| R16-C1: recip $2\to1$ | ELSb | $\operatorname{ELSb}(0, x) = 1/x$ | 1 node |
In each case: take an operator in $\mathcal{F}_{16} \setminus \mathcal{F}_6$ that accepts one argument directly (no $\exp$ or $\ln$ wrapper), set the free argument to a constant, and the formula collapses to the target function.
SuperBEST v4: 18 Nodes, 75.3% Savings
The updated table:
| Operation | v3 (F16) | v4 (F16) | Operator |
|---|---|---|---|
| exp(x) | 1 | 1 | EML(x,1) |
| ln(x) | 1 | 1 | EXL(0,x) |
| neg(x) | 2 | 2 | EXL(0, DEML(x,1)) |
| recip(x) | 2 | 1 | ELSb(0,x) |
| mul(x,y) | 2 | 2 | ELAd(EXL(0,x),y) |
| sub(x,y) | 2 | 2 | LEdiv(x, EML(y,1)) |
| add(x,y) | 3 | 3 | EAL(EXL(0,x), EXL(0,y)) |
| pow(x,n) | 3 | 3 | EML(EXL(0,x)·n, 1) |
| Total | 19 | 18 |
Savings: $1 - 18/73 = 55/73 \approx \mathbf{75.3%}$ vs naive EML.
The naive-EML baseline is fixed at 73 nodes. Every node we remove is a permanent reduction in the compression floor.
Note: recip in $\mathcal{F}_6$ still requires 2 nodes. ELSb is not in the six-operator library. Every $\mathcal{F}_6$ operator applies $\ln$ to its second argument, so $x$ cannot appear directly in the denominator without an intermediate $e^x$ conversion step.
Ripple: Which Downstream Results Update
Most results are unaffected. The reason: in practice, standalone $\recip(x)$ as an explicit unary operation is rare. Most formulas use division ($x/y$), which was already 1 node via EDL, not 2.
The affected cases are formulas where $1/x$ appears as a true unary reciprocal:
Geometry catalog — Gaussian curvature:
The entry K(z = ln r) — 7 nodes — mul+recip+neg drops to 6 nodes.
The recip sub-tree was contributing 2n; it now contributes 1n.
SuperBEST table itself: 19n $\to$ 18n. This is the primary update.
Unaffected:
- Taylor series: coefficients $1/n!$ are constants (folded to 0 nodes). No recip node.
- Partition function $Z = \mathrm{Tr}[e^{-\beta H}]$: no recip.
- Density matrix $\rho = e^{-\beta H}/Z$: the $1/Z$ is a division by a scalar matrix — handled by EDL, already 1 node.
- Free energy $F = -\ln Z / \beta$: no recip.
- 295+-equation catalog: the few equations using explicit $1/x$ drop 1 node each; the catalog total decreases by an estimated $\leq 10$ nodes across all 295+ equations.
Connection to the Structural Audit
R16-C1 sits on the T08 $\to$ THEOREM path of the structural audit. The audit identified that the SuperBEST table should be verified by exhaustive search over $\mathcal{F}_{16}$, not just $\mathcal{F}_6$. T10u, T33, and R16-C1 are the three results found during that systematic re-check.
The search principle:
For every operator $\mathrm{op} \in \mathcal{F}_{16} \setminus \mathcal{F}_6$, enumerate all constant-argument specializations from ${0, 1}$. Many simplify to known functions.
This is a finite search (16 operators $\times$ 4 terminal pairs = 64 candidates for 1-node trees). R16-C1 fell out of that enumeration.
Formal Reference
Theorem paper: D:/monogate/python/paper/theorems/recip_One_Node.tex
Result identifier: R16-C1 (Census item 16, Construction 1)
Proof: construction by explicit formula + lower bound by non-constancy argument. No exhaustive search required for the lower bound — it follows immediately that no constant equals $1/x$ for all $x > 0$.
Monogate Research (2026). “R16-C1: recip(x) Is 1 Node — ELSb Closes the Gap.” monogate research blog. Session: 1-Node Integration. https://monogate.org/blog/recip-one-node