Yamagishi Complexity Torus LabYamagishi Lab

[Paper 631] Adding a Road Slows Everyone Down Only When the Number of Cars Lies Between 3000 and 9000

Abstract

There are two routes from start to finish. One is a road taking the number of cars $/100$ minutes followed by a $45$-minute road; the other is a $45$-minute road followed by a road taking the number of cars $/100$ minutes. A $0$-minute shortcut joining the middles of the two routes is added. Where each driver picks their own shortest route and no one gains by switching alone (the settled state) is counted as the number of cars changes (Braess’s paradox). No new theorem or law is claimed.

First, with $4000$ cars, adding the shortcut turns $65$ minutes into $80$——before, the cars split in half and take $65$ minutes. After, the settled state, found both by closed form and by a numerical method that does not assume symmetry (the Frank--Wolfe method), has everyone on the shortcut, taking $80$ minutes. The other routes take $85$, so no one switches (Section 2).

Second, an assignment minimising the total gains from the shortcut——sending only $500$ of the $4000$ cars through the shortcut gives an average of $64.687500$ minutes, shorter than the $65$ before. With someone assigning routes, adding a road never makes things worse (Section 2).

Third, and this is the core. Whether it gets slower is decided by the number of cars——checking every number from $1$ to $12000$ cars, the shortcut slows things for $3001$ to $8999$ cars, speeds them for $2999$ or fewer ($50$ minutes become $20$ with $1000$ cars), and goes unused from $9000$ cars (Section 3).

Fourth, the ratio of the settled state to the least total reaches $4/3$ at $4500$ cars——the ratio is largest at $4500$ cars, $90$ minutes against $67.5$, which is $\mathbf{1.333333}$. For roads whose times are linear in the number of cars this ratio is known never to exceed $4/3$, and this network reaches that bound exactly (Section 4).

Keywords: Braess’s paradox, traffic network, Wardrop equilibrium, price of anarchy, selfish routing.

Scope (what is not claimed)

No new theorem or law is claimed——that adding a road can slow the settled state (Braess, $1968$), the condition for the settled state (Wardrop, $1952$), and that with linear times the ratio of the settled state to the least total never exceeds $4/3$ (Roughgarden and Tardos, $2002$) are all known. The bound $4/3$ is not proved——it is only counted that this network reaches it. Real roads and policies are not discussed——the network of $45$ minutes and the number of cars $/100$ minutes is a common textbook example, not the numbers of Braess’s original paper. Relation to earlier papers: Paper 437 showed with a two-player payoff table that each choosing the best lands the total at its lowest, placing the separator on “one at a time or together”——here adding one more road slows everyone, and only when the number of cars lies in a certain range. Paper 462 set aside in its scope note the design of roads, such as the number of lanes and merges——this paper takes up that seat, treating the addition of one road. What is added is finding the settled state by closed form and by a numerical method not assuming symmetry, counting the range of cars for which it gets slower, one car at a time, and placing the separator on the number of cars.

1. Notation and the Question

The network: the road from start to point A takes $x/100$ minutes, where $x$ is the number of cars using it. From A to finish takes $45$ minutes. From start to point B takes $45$ minutes, and from B to finish $x/100$ minutes. The shortcut added is a $0$-minute road from A to B. There are then three routes: start--A--finish, start--B--finish, and start--A--B--finish.

The settled state: every route in use takes the same time, and every unused route takes at least as long. No one gains by switching alone (Wardrop’s condition).

Question: does adding the shortcut make the settled state faster or slower, and what decides which?

2. With 4000 Cars

Computation 2.1: before the shortcut, the cars split $2000$ and $2000$, and both routes take $20+45=65$ minutes.

Computation 2.2: the settled state after adding the shortcut was found by closed form and by a numerical method starting with everyone on one route (the Frank--Wolfe method, not assuming symmetry). Both put all $4000$ cars on start--A--B--finish, taking $40+0+40=80$ minutes. The other two routes take $40+45=85$ minutes.

Observation 2.3 to each driver the shortcut always looks better. A road taking the number of cars $/100$ minutes takes $40$ minutes even with everyone on it, shorter than the $45$-minute road. So everyone chooses both such roads and joins them by the shortcut——and as a result all go from $65$ to $80$ minutes, $15$ minutes slower.

Computation 2.4: the assignment minimising the total of everyone’s time was also found by the same two routes. It is $1750$, $1750$ and $500$ on the shortcut, with an average of $64.687500$ minutes.

Observation 2.5 with someone assigning routes, adding a road never makes things worse. An assignment not using the shortcut can always be chosen, so the least total cannot rise when a road is added——what got slower is the settled state, where each chooses their own shortest.

3. Changing the Number of Cars

Computation 3.1: for different numbers of cars, the time before the shortcut, the settled state after it, and the average of the least total were set side by side (for the nine representative numbers, the settled state was found by both routes and agreed within $0.001$ minutes).

carsbeforesettled afterleast-total averageratio
$1000$$50$$20$$20$$1.000000$
$2000$$55$$40$$40$$1.000000$
$3000$$60$$60$$56.25$$1.066667$
$4000$$65$$80$$64.6875$$1.236715$
$4500$$67.5$$90$$67.5$$1.333333$
$6000$$75$$90$$75$$1.200000$
$8000$$85$$90$$85$$1.058824$
$9000$$90$$90$$90$$1.000000$
$10000$$95$$95$$95$$1.000000$

Checking every number from $1$ to $12000$ cars, the shortcut slows things for $3001$ to $8999$ cars, $5999$ cases, and speeds them for $1$ to $2999$ cars. The shortcut goes unused from $9000$ cars.

Observation 3.2 this is the core: whether adding a road slows things is decided by the number of cars. With $N$ cars, the time before is $N/200+45$ minutes, and with everyone on the shortcut it is $N/50$. Below $3000$ cars the shortcut’s $N/50$ is shorter, and adding it speeds things. Above $4500$ cars some drivers avoid the shortcut, and the settled state stays at $90$ minutes——at $9000$ cars, where the $N/200+45$ minutes before catch up with $90$, the shortcut goes unused and the difference vanishes.

4. The Largest Ratio

Computation 4.1: the settled time divided by the least-total average is largest at $4500$ cars, $\mathbf{1.333333}$ ($90$ minutes against $67.5$).

Observation 4.2 for roads whose times are linear in the number of cars, this ratio is known never to exceed $4/3$. This network reaches that bound exactly at $4500$ cars——for networks of this form, this is as far as individual choices can drift from the least total.

5. The Separator

Observation 5.1 the separator is the number of cars. Adding the same shortcut to the same network speeds things for $2999$ cars or fewer, slows them for $3001$ to $8999$, and changes nothing from $9000$——“adding a road slows things” is a property not of the network alone but of the network together with the number of cars.

Observation 5.2 beside Paper 437 and Paper 462. In Paper 437, two players each choosing the best landed the total at its lowest, and the separator was “one at a time or together”. Here too the least total (together) gains from the shortcut and the settled state (one at a time) loses——on top of that, it was counted that the loss happens only when the number of cars lies in a certain range. Of the design of roads that Paper 462 set aside in its scope note, only the addition of one road is treated here.

Observation 5.3 what is not said. Whether a real shortcut should be closed is not discussed, and the bound $4/3$ is not proved——what is shown is the settled state and least total for one network, and the range of cars for which it gets slower.

6. Roll Call and Closing

  1. $65$ minutes become $80$ with $4000$ cars (Computation 2.2): two routes agree.
  2. the other routes take $85$ (Computation 2.2): no one switches.
  3. the shortcut always looks better (Observation 2.3): $40$ minutes even with everyone.
  4. $64.687500$ minutes for the least total (Computation 2.4): only $500$ on the shortcut.
  5. an assigner never loses (Observation 2.5): the settled state is what got slower.
  6. slower for $3001$ to $8999$ cars (Computation 3.1): counted one car at a time.
  7. $50$ minutes become $20$ with $1000$ cars (Computation 3.1): faster when few.
  8. decided by the number of cars (Observation 3.2): the core.
  9. ratio $\mathbf{1.333333}$ at $4500$ cars (Computation 4.1): reaches the bound $4/3$.
  10. one at a time loses, together gains (Observation 5.2): Paper 437 and Paper 462.

Closing: Adding a $0$-minute shortcut to two routes joining a $45$-minute road and a road taking the number of cars $/100$ minutes turned the settled state for $4000$ cars, where each picks their own shortest, from $65$ minutes into $80$. Assigning routes to minimise the total shrinks it to $64.687500$ minutes with the shortcut, so what got slower is the side of individual choice. Whether it gets slower is decided by the number of cars——slower for $3001$ to $8999$ cars, faster for $2999$ or fewer, and the shortcut unused from $9000$. The ratio of the settled state to the least total reaches $4/3$ at $4500$ cars. The separator is the number of cars. Placed among the earlier papers——to “each one’s best is everyone’s worst” of Paper 437, the direction of adding a road and the range of cars are added. To be honest——neither real roads nor the proof of the bound $4/3$ are entered.

References

  1. Braess, D. (1968). Über ein Paradoxon aus der Verkehrsplanung. Unternehmensforschung.
  2. Roughgarden, T., & Tardos, É. (2002). How bad is selfish routing? Journal of the ACM.
  3. Wardrop, J. G. (1952). Some theoretical aspects of road traffic research. Proceedings of the Institution of Civil Engineers.
  4. Yamagishi, Y. (2026). Each Choosing Best Lands Both on the Worst [Paper 437]. Zenodo. 10.5281/zenodo.22709472
  5. Yamagishi, Y. (2026). Two Different States Carry Exactly the Same Load [Paper 462]. Zenodo. 10.5281/zenodo.22733178

Reference

Yamagishi, Y. (2026). Adding a Road Slows Everyone Down Only When the Number of Cars Lies Between 3000 and 9000 [Paper 631]. YOI: https://www.yamagishiyuuki.com/yoi/631
DOI: 10.5281/zenodo.23014556
ORCID: 0009-0006-8587-3450

PDF (English) · PDF (Japanese) · Date proof of the English PDF (OpenTimestamps .ots) · SHA256SUMS

The record page for this paper (in Japanese, with the PDFs in each language and the date proof): YOI 631. This paper is a preprint and has not been peer reviewed.