Teban54

Just another transit enthusiast

[Idea] Routing the Green Line Reconfiguration (Part 1): A Theoretical Framework of Capacity and Service Patterns

—

by

Teban54
in

I could not have thought that my 3rd blog post would be 1.5 years after the website’s creation, but here we are.

TL;DR

  • The Green Line Reconfiguration (GLR) system likely has a theoretical maximum capacity of 80 TPH, up from today’s 32; but hitting this maximum is non-trivial.
  • Most prior GLR proposals, namely the “Tripod” and the (naive) “Criss-Cross” models, only hit 64 TPH instead.
  • Adding short turns terminating at Park to Criss-Cross gets you 80 TPH. It comes at the drawback of interlining — but as it turns out, interlining is inevitable for maximum capacity.
  • There are (at least) 5 different service patterns that hit 80 TPH, not just limited to short-turning Criss-Cross.
  • Evaluating them will be left for the future — but share your thoughts now!

Background and Principles of Green Line Reconfiguration

I had written about the Green Line Reconfiguration (GLR) proposal in 2023, and RailsRoadsRiverside included more details on routing and engineering in his blog (where he calls the idea “the Gold Line”). Even more discussions of the topic can be found in the archBoston thread.

But one thing went under the hood: There has never been a settled consensus on how exactly the individual routes under GLR would look like.

In archBoston discussions, most users agree that GLR should involve three “outbound” trunks, where each of them may involve multiple branches further out:

  • Kenmore: The existing Central Subway under Boylston St.
  • Huntington Ave: Today’s E branch; to be improved by extending the Huntington Ave subway to Brookline Village under the GLR proposal, where it joins the D branch.
  • Nubian: New route(s) under the GLR proposal, that run along Washington St and/or parallel streets, feeding into downtown in the vicinity of Bay Village.

We also have three “downtown” trunks:

  • Government Center (GC): The existing Tremont St subway, from Park St to North Station and continuing to the Green Line Extension (GLX). To the south, it connects to the unused outer tracks at Boylston, which continue south to the former Pleasant St portal. (For the purpose of this article, the Green Line branches north of GC, such as GLX, are considered “downtown” trunks.)
  • Park St Inner (PI): Existing inner tracks at Park St, which form a loop for trains that short turn at the station.
  • Seaport: The Seaport Transitway that’s used by Silver Line Waterfront today; to be extended west under the GLR proposal.

The GC and PI trunks are connected via switches between Park St and Boylston stations, which (almost) every Green Line train uses today. This will be important later on.

There’s also general agreement of the following principles for the GLR system:

  • Increased service for the Huntington Ave subway
    • Huntington Ave routes would serve Longwood Medical Area (LMA), the biggest employment center outside of downtown Boston. Not to be overlooked are Prudential Center, Museum of Fine Arts and various academic institutions on the corridor.
  • More grade separation at crucial junctions, so that trains don’t cross unless necessary
    • Today, there’s a flat junction west of Copley, where outbound E trains interfere with inbound B/C/D trains and cause delays.
    • Such junctures, where two or more trunks intersect or merge, should ideally be replaced with flying junctions. Only trains traveling in the same direction would then be required to merge.
  • Increased service and route extension for Seaport
    • Seaport is very underserved today by bus rapid transit (BRT), which has lower capacity than light rail transit (LRT). The current route also only connects with the Red Line within downtown.
    • Most GLR proposals involve extending the Silver Line Transitway westward from South Station and connecting it to one of the westside GLR trunks. But the exact routing remains debatable.

Yet, there has never been a clear answer for how the three trunks on each side should be connected. It’s not just the engineering that’s missing; conceptually, questions like “how many Huntington trains should be sent to Seaport vs. GC” don’t have a clear answer. This is before we even get down to the individual branches, such as “where should Needham trains go”.

In most previous writeups of GLR, we typically present one proposal for how to pair up the six trunks. Putting something out there is necessary for the audience to understand the concept… But are such pairings really optimal?

All Possible Pairings, as a Graph

This blog post, and future posts in this series, will primarily focus on how to “pair up” the three outbound trunks and the three downtown trunks. But before that, we first need a framework to analyze these pairings.

Below, we visualize all potential track connections that either exist today, or can be built with reasonable effort:

Figure 1. All realistic track connections between the three outbound trunks and the three downtown trunks

Clarification: This is a purely abstract visualization using the “graph” structure. The placements of nodes and edges have no geographical implications, and neither do their intersections.

In computer science and discrete math, a graph is an abstract structure that specifies a number of vertices — the circles in Figure 1 — and edges that describe how vertices are connected. For example, K-PI and H-SP are two of the many edges in Figure 1.

Importantly, even though graphs are typically drawn on a two-dimensional surface for visualization, the vertices and edges are unordered, and their spatial placements have no meaning. In other words, both the vertices and the edges can be moved or reshaped arbitrarily.

  • In Figure 1, Seaport being placed above PI and GC doesn’t mean that the former is further north.
  • Likewise, the fact that K-PI and H-SP “intersect” in the figure doesn’t imply that their real-life tracks must form a (grade or flying) junction.

If it helps, think of any intersections between edges as a grade-separated crossing with no track connections.

Figure 1 shows a key assumption on track connections:

  • Both Kenmore and Huntington can connect to all three downtown trunks, possibly with new construction.
  • However, the only realistic connection for Nubian is Government Center (i.e., Boylston and Park St’s outer tracks).

Below, I mention how each of these connections can be enabled. However, the focus of this post is not on their alignment and engineering. These topics will be explored in a future post in this series.

Track connections: Kenmore

Track connections: Huntington

Track connections: Nubian

Capacity Assumptions — and the Goal of 80 TPH

In addition to whether you can physically route a train from X to Y, another important aspect is how many trains can go down such routes.

I impose the following key assumptions on capacity, which will be treated as axioms for this series of posts:

  • Each letter bullet (such as B, C, D, E) runs 8 trains per hour (TPH), or every 7.5 minutes.
  • Any track has a maximum capacity no more than 32 TPH (every 1.875 minutes). This means it can support 4 bullets.
  • Park St Inner has a maximum capacity of 16 TPH (2 bullets, every 3.75 minutes). The lower capacity is due to additional complications of terminating trains at the loop.

These are our best guesses at capacity constraints. As of April 2025, they have not yet been rigorously tested, and some may even be challenged. I recommend treating these TPH numbers as a rule of thumb, not the ground truth.

Here are some analyses and arguments both supporting and challenging the assumptions:

8 TPH and 32 TPH: Analyzing today’s Green Line schedule

16 TPH at Park St Inner… Or not?

As seen from Figure 1, these assumptions imply that we can get at most 80 TPH out of the entire GLR system. Beyond that, you simply won’t have enough capacity at the downtown trunks. We will thus treat 80 TPH as our ultimate goal.

Tripod vs. Criss-Cross

Now let’s see how some existing, concrete GLR proposals fit into our capacity framework.

Tripod (The “Traditional” GLR Plan)

First, consider the “Tripod” model. This was the most standard GLR design until mid-2024: it is the service plan I presented in my own GLR map, as well as RailsRoadsRiverside’s extensive writeups and maps on GLR.

(While neither links above called it a “tripod”, the name arose from visualizing the routes on a map: a triangle formed by Huntington-Park, Nubian-Park and Huntington-Seaport.)

Figure 2. The “Tripod” model. Bullets and colors largely correspond to my GLR map in December 2023, with the exception of the M line being renamed the J.

Despite all the benefits I explained in my GLR map, Figure 2 immediately shows an issue: At a “mere” 64 TPH, Tripod doesn’t unlock the full potential of 80 TPH.

  • Most notably, Seaport only runs 16 TPH. This doesn’t agree with our GLR principle of increasing service to Seaport as much as possible.
  • The other trunk with underwhelming capacity is Kenmore. While the B/C branches west of Kenmore retain their current frequencies, this is a downgrade for Kenmore through Arlington stations, many of which are major ridership drivers.

Keep in mind, 64 TPH is still double the present-day 32 TPH on the Green Line and a huge improvement. But can we do even better?

Criss-Cross

Recently, RailsRoadsRiverside proposed a different model for routing GLR trains and connecting the outbound and downtown trunks — the “Criss-Cross” model:

Figure 3. The standard “Criss-Cross” model, as described by RailsRoadsRiverside, without any Park St short turns. I have changed the presentation slightly, using outbound-trunk-based colors (see later sections). Bullets are added arbitrarily; this figure also imply a minor change of Needham trains running through Kenmore instead of Huntington.

RailsRoadsRiverside’s initial proposal of Criss-Cross was inspired by certain shortcomings of Tripod, although most of them had nothing to do with capacity. The main idea — and the source of the name “Criss-Cross” — is having two main trunk routes that cross each other in downtown Boston:

  • An entirely isolated Kenmore-Seaport line;
  • A Tremont St line through Government Center, which then branches out to Huntington and Nubian trunks.

RailsRoadsRiverside’s original proposal did not explicitly address any short turns at Park St Inner. If the loop is completely unused, as shown in Figure 3, we still only achieve 64 TPH. From a capacity perspective, the “standard Criss-Cross” solely reallocates capacity across the system, without gaining any additional trains.

  • In particular, one major capacity drawback is with Huntington, again to the contrary of GLR’s principles. I’d argue that Huntington’s capacity is even more important than either Seaport or Arlington-Kenmore (the latter is somewhat more substitutable with Back Bay station and a potential Blue Line extension to Kenmore).

Criss-Cross with Short Turns

Turns out, there’s an “easy” fix to capacity under Criss-Cross: adding trains between Huntington — Park St Inner (H-PI). (RailsRoadsRiverside himself actually implied it in the original proposal, but it was not obvious.)

If you draw the routes on a geographical or schematic maps, it’s easy to see H-PI as a “short turn” of Huntington — Government Center (H-GC) trains. (This is likely why RailsRoadsRiverside didn’t note such short turns explicitly.) But here’s the magic: in our graph representation, the additional service is shown very clearly, and very explicitly, as an additional edge (the N/R edge).

Figure 4. The Criss-Cross model with H-PI short turns

For the first time in this post, we actually achieve the maximum 80 TPH! Woohoo!

But you might be wondering:

  1. Is this the only way to get 80 TPH? Are there any other alternatives?
  2. What drawbacks do the “short turn Criss-Cross” have?

I’ll get to question 1 later in the post, so let’s talk about the drawbacks: there are many (destinations served, ease of transfers, flexibility for passengers, etc.) But I want to highlight a specific “issue” here:

Interlining, and Its Inevitability

Suppose that under the “short turn Criss Cross” in Figure 4, all northbound Huntington trains arrive at Boylston’s inner tracks, and all Nubian trains arrive at the outer tracks. To continue to their downtown trunks, Huntington-PI trains (N, R) can just continue on the inner tracks, while Nubian-GC trains (F, G) stay on the outer tracks.

But Huntington-GC trains (D, E) need to switch from the inner track to the outer track:

Figure 5. Interlining of northbound H-PI, H-GC and N-GC trains in the “short turn Criss Cross” model. For clarity, H-PI trains are labeled Brown, instead of Magenta as in Figure 4.

This is an example of interlining, where we have two largely independent lines (Brown and Gold), but some trains switch from one line to the other (Magenta).

  • I myself call this the “N-merge”, both because the lines form an “N” shape (as seen in Figure 5), and also because one of the most prominent examples is New York City’s Broadway Line, where the N train switches between the local and express tracks. (This video shows delays on the Broadway Line due to the N merge.)
  • Meanwhile, another kind of interlining is what I call the “X-merge”, where a fourth set of trains cross the two lines in the opposite direction, forming a full 2×2 service pattern. (In Figure 5, that would correspond to adding hypothetical Nubian-PI trains.)

An obvious benefit of interlining is convenience, as (some) riders get service to two destinations from the same origin. But it comes at a pretty significant cost: Reliability. If a (delayed) Magenta train and a Gold train arrive at the same time, only one of them can go ahead — and no matter how you do it, delays have been propagated from the “Brown Line” to the “Gold Line”.

(In fact, the NYC subway system is infamous for its heavily interlined system, with various major N- and X-merges everywhere. People often think this causes unreliability across the system, and often results in concrete capacity limits. It’s why “deinterlining the NYC subway” proposals are wildly popular acorss the internet.)

Note that interlining is not limited to switches between two adjacent tracks. Imagine if the Magenta routes in Figure 5 took a long-winded detour with multiple stops to go from Brown to Gold. This doesn’t change the “delay potential” fundamentally: a late Magenta train can still affect a Gold train eventually.

Therefore, any “N” shape on the graph means interlining, regardless of the engineering of track connections.

The Inevitability of Interlining in 80-TPH GLR

Unfortunately, in order to achieve the maximum capacity of 80 TPH in our GLR framework (Figure 1), interlining is necessary.

We can actually prove it mathematically:

Figure 6. Explanation of why interlining is necessary to achieve 80 TPH
  1. We can allocate at most 80 TPH to the three outbound branches. The easiest choice for a “downgrade” to 16 TPH, by far, is Nubian. (Kenmore has many branches that can feed into it, while Huntington connects many important destinations for the entire region. In comparison, some people would argue that even 16 TPH for Nubian is overkill.)
  2. Nubian can only feed into Government Center. So our first 2 bullets are locked in: N-GC (F and G trains in Figure 6). This takes up 16 out of 32 TPH at GC.
  3. GC still has 16 TPH to be filled. They have to come from Kenmore, Huntington, or a mix between the two. In Figure 6, we show the scenario where all of the 16 TPH come from Huntington, meaning H-GC gets 16 TPH. (The arguments for Kenmore and mixtures are identical.)
  4. Huntington now needs to send its remaining 16 TPH to either SP, PI, or a mix. Again, assume for simplicity that they all go to PI, so H-PI gets 16 TPH.
  5. Finally, we’ve arrived at an N-merge: H-PI, H-GC, and N-GC.

This is guaranteed to happen no matter how you make the choice in steps 3 and 4.

Note that the first domino that triggers all this is step 2: Nubian can only feed into GC. You can imagine very different outcomes if this was not the case (which I may write about in a future post).

So, we have to settle on interlining to unlock the maximum capacity out of GLR… That’s a big bummer, right? Not necessarily:

  • Psychologically, this “liberates” us to be more accepting of interlining proposals. While a higher degree of interlining can still negatively affect reliability, the choice of whether interlining at all is acceptable has been eliminated for us, as long as your goal is to maximize capacity.
  • The exercise above also gives us a way to generate (essentially) all ways to achieve maximum capacity, which we’ll do immediately below.

“All” Alternatives for 80 TPH

We can enumerate over all possible GLR service patterns that feed 80 TPH into the downtown trunks, by considering all choices we can make in steps 3 and 4 above:

  • Should the remaining 16 TPH from GC go to Huntington, Kenmore, or a mix of them (with mixing proportions of our choice)?
  • Should Seaport become a dedicated trunk with service from only a single outbound trunk (which is necessarily not the one that gets GC service)? Or should Seaport service also be a mix between Huntington and Kenmore?

Here are the 5 representative alternatives. Note that there’s theoretically a large number of “intermediate” mixtures by extrapolating these 5 extremes, but they’re omitted.

Figure 7. Five main alternatives of GLR service patterns that achieve the maximum capacity of 80 TPH

Note: For simplicity, edges and bullets are colored solely based on their outbound trunks: Kenmore bullets Green, Huntington bullets Magenta, and Nubian bullets Gold.

If you want a text walk-through:

Alternative 1: K-SP isolated (Criss-Cross)

Alternative 2: All destinations, K turns at Park

Alternative 3: H-SP isolated

Alternative 4: All destinations, H turns at Park

Alternative 5: All destinations evenly mixed

It should be emphasized that these five alternatives describe service patterns at an abstract level. They do not (yet) consider the routings for each connection within downtown and the associated engineering challenges. Nor has serious consideration been placed on what each letter bullet means — right now, treat them as arbitrary symbols.

That said, I did have a set of preliminary branches in mind, and tried to ensure consistency in how bullets correspond to branches:

  • A: Either Harvard (via West Station and Lower Allston) or Watertown (via Brighton)
  • B: Boston College (via Commonwealth Ave)
  • C: Cleveland Circle (via Beacon St)
  • D: Riverside
  • E: Heath St / Hyde Sq / Forest Hills, via Huntington Ave
  • F: Nubian, via Washington St (surface)
  • G: Nubian / MLK Blvd / Franklin Park / Four Corners, via grade-separated route north of Nubian (e.g. via I-93 El) and surface south of Nubian
  • J: Heath St / Hyde Sq / Forest Hills, via Kenmore
    • Uses the Highland Branch (Fenway and Longwood) to maintain service to these two stations. New surface tracks connect Brookline Village and Riverway (separate from the grade-separated connector from Brookline Village to Huntington Ave subway).
  • N: Needham Junction / Hersey
  • R: Reservoir, as additional Huntington “short turns” to serve the busiest segment east of Reservoir
    • Can also extend to Boston College via the Chestnut Hill Ave surface tracks
    • Kind of a flexible “throwaway” route
    • If you’re really creative, can even run surface light rail across Longwood Medical Area (e.g. via Longwood Ave – Longwood station – Kenmore)

Conclusion and Next Steps

This post is only the first part of (hopefully) a series of writeups on this topic. The main purpose of this post is solely to describe the framework and alternatives, not to evaluate them or get down to their details. These will be topics for future discussions.

That said, we can already draw one conclusion: Tripod and Criss-Cross are far from the only two service patterns under GLR — in fact, there are at least five of them! Hopefully, this fact has been made clearer now that we have abstracted away the operational and engineering details. It is worth considering each one’s pros and cons; and don’t forget, there may be arguments for not aiming for the maximum 80 TPH in the first place.

Below is a list of natural extensions that I hope to write about in future blog posts, in no particular order:

  1. Routing and engineering: How can we actually enable each of these connections between trunks? Which alignments should they use? How to build the critical junctions where trains merge and separate?
  2. Evaluating the 5 alternatives: Which one is the “best”? What pros and cons do each one have? Or could it be that forgoing some capacity can give us bigger gains overall?
  3. Branches and bullets: Most of the six trunks — both outbound and downtown ones — can potentially branch out to multiple destinations. Which branches make sense? How should they be combined into letter bullets?
  4. An in-depth look at capacity: This post included some preliminary analysis on how the Green Line operates today, especially regarding TPH. Can we dive deeper? How did MBTA decide each branch’s capacity? Does the system have potential for higher TPH than what we assumed? How reliable is each branch — and how does that affect GLR?
  5. How bad is interlining? Even though interlining is generally regarded as undesirable by transit fans (especially in NYC), I haven’t seen much concrete, quantitative analyses of it. Can we measure its effects more directly? How much should we be concerned about it in the context of GLR?
  6. Revisiting the Nubian-GC assumption: Our assumption that Nubian trains can only feed into Government Center plays an important role. But is it possible to have a cleaner connection to other downtown trunks?

Please, please, let me know which of these topics you’re the most interested in, and/or any comments you have about this post! You can either join the discussion on archBoston (preferred), or write a reply under this post. Any discussions will play a huge role in how I will shape future posts in this series.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *