# BCFW recursion
## Refs
- original
- [[2004#Britto, Cachazo, Feng]]
- [[2005#Britto, Cachazo, Feng, Witten]]
- proof
- [[2008#Arkani-Hamed, Kaplan]] proof using background field perturbation theory
- review/learning
- [[Rsc0021 Nathaniel Craig 229A Gauge theory Winter 2021]]
- [[Rsc0003 ElvangHuang Scattering amplitudes]]
- [Truijen notes](https://web.archive.org/web/20140221230929/http://testweb.science.uu.nl/ITF/teaching/2012/Brecht%20Truijen.pdf)
- for general field theories
- [[CheungShenTrnka2015]][](https://arxiv.org/pdf/1502.05057.pdf)
- celestial BCFW
- [[2022#Hu, Pasterski (Aug)]]
## Assumptions
- usually tree-level (but can consider loops)
- massless particles
- locality (local Lagrangian)
## Steps
1. define shifted momenta
2. notice $A_{n}=\hat{A}_{n}(z=0)$ is just the residue of the pole at origin for $\frac{\hat{A}_{n}(z)}{z}$
3. use Cauchy's theorem: $A_{n}=-\sum_{z_{I}} \operatorname{Res}_{z=z_{I}} \frac{\hat{A}_{n}(z)}{z}+B_{n}$
4. notice that the poles are whenever the propagator goes on-shell
- $\operatorname{Res}_{z=z_{I}} \frac{\hat{A}_{n}(z)}{z}=-\hat{A}_{\mathrm{L}}\left(z_{I}\right) \frac{1}{P_{I}^{2}} \hat{A}_{\mathrm{R}}\left(z_{I}\right)$
- used $\hat{P}_{I}^{2}=-\frac{P_{I}^{2}}{z_{I}}\left(z-z_{I}\right)$
5. check residue at infinity
## Rules
- two shifted momenta appear on two sides
- otherwise the propagator will not depend on $z$
- preserve cyclic ordering
## Properties
- at every stage it involves only *on-shell* particles, as opposed to virtual particles that propagate inside conventional Feynman diagrams. (the internal propagator is on-shell, and the sub-diagrams are on-shell)
- nice for ==tree-level== amplitudes
- no branch cuts
- only simple poles (follows from locality of Feynman diagrams)
## Relation to CCFT
- there is a natural notion of factorisation in CCFT, which is a nice surprise
- Basically, just like a 2-gluon OPE, one could construct a multi-gluon OPE and factorize a celestial amplitude on it. It is to be expected that factorization on OPE is in 1:1 correspondence with BCFW factorization channels
- see [[2014#Goncalves, Penedones, Trevisani]]
- see also [[2022#Hu, Pasterski (Aug)]]
## Factorisation v.s. collinear
- collinear limits are only a subset, because of signs
- $k_{L}^{2}=\sum_{i \neq j \in L} k_{i} \cdot k_{j}=\sum_{i \neq j \in L} \eta_{i} \eta_{j} \omega_{i} \omega_{j} z_{i j} \bar{z}_{i j}$
- i.e. even non-collinear setup can give a pole for $1/k_L$